diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc64')
244 files changed, 10208 insertions, 2394 deletions
diff --git a/sysdeps/powerpc/powerpc64/__longjmp-common.S b/sysdeps/powerpc/powerpc64/__longjmp-common.S index 70c370439f..3c792b495e 100644 --- a/sysdeps/powerpc/powerpc64/__longjmp-common.S +++ b/sysdeps/powerpc/powerpc64/__longjmp-common.S @@ -17,6 +17,7 @@ <http://www.gnu.org/licenses/>. */ #include <sysdep.h> +#include <stap-probe.h> #define _ASM #define _SETJMP_H #ifdef __NO_VMX__ @@ -57,7 +58,7 @@ ENTRY (__longjmp) beq L(no_vmx) la r5,((JB_VRS)*8)(3) andi. r6,r5,0xf - lwz r0,((JB_VRSAVE)*8)(3) + lwz r0,((JB_VRSAVE)*8)(3) /* 32-bit VRSAVE. */ mtspr VRSAVE,r0 beq+ L(aligned_restore_vmx) addi r6,r5,16 @@ -130,7 +131,7 @@ L(no_vmx): ld r14,((JB_GPRS+0)*8)(r3) lfd fp14,((JB_FPRS+0)*8)(r3) #if defined SHARED && !defined IS_IN_rtld - std r2,40(r1) /* Restore the callers TOC save area. */ + std r2,FRAME_TOC_SAVE(r1) /* Restore the callers TOC save area. */ #endif ld r15,((JB_GPRS+1)*8)(r3) lfd fp15,((JB_FPRS+1)*8)(r3) @@ -147,20 +148,23 @@ L(no_vmx): #ifdef PTR_DEMANGLE PTR_DEMANGLE2 (r0, r25) #endif + /* longjmp/longjmp_target probe expects longjmp first argument (8@3), + second argument (-4@4), and target address (8@0), respectively. */ + LIBC_PROBE (longjmp, 3, 8@3, -4@4, 8@0) mtlr r0 -/* std r2,40(r1) Restore the TOC save area. */ +/* std r2,FRAME_TOC_SAVE(r1) Restore the TOC save area. */ ld r21,((JB_GPRS+7)*8)(r3) lfd fp21,((JB_FPRS+7)*8)(r3) ld r22,((JB_GPRS+8)*8)(r3) lfd fp22,((JB_FPRS+8)*8)(r3) - ld r0,(JB_CR*8)(r3) + lwz r5,((JB_CR*8)+4)(r3) /* 32-bit CR. */ ld r23,((JB_GPRS+9)*8)(r3) lfd fp23,((JB_FPRS+9)*8)(r3) ld r24,((JB_GPRS+10)*8)(r3) lfd fp24,((JB_FPRS+10)*8)(r3) ld r25,((JB_GPRS+11)*8)(r3) lfd fp25,((JB_FPRS+11)*8)(r3) - mtcrf 0xFF,r0 + mtcrf 0xFF,r5 ld r26,((JB_GPRS+12)*8)(r3) lfd fp26,((JB_FPRS+12)*8)(r3) ld r27,((JB_GPRS+13)*8)(r3) @@ -173,6 +177,7 @@ L(no_vmx): lfd fp30,((JB_FPRS+16)*8)(r3) ld r31,((JB_GPRS+17)*8)(r3) lfd fp31,((JB_FPRS+17)*8)(r3) + LIBC_PROBE (longjmp_target, 3, 8@3, -4@4, 8@0) mr r3,r4 blr END (__longjmp) diff --git a/sysdeps/powerpc/powerpc64/addmul_1.S b/sysdeps/powerpc/powerpc64/addmul_1.S new file mode 100644 index 0000000000..f256506304 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/addmul_1.S @@ -0,0 +1,208 @@ +/* PowerPC64 __mpn_addmul_1 -- Multiply a limb vector with a limb and add + the result to a second limb vector. + Copyright (C) 1999-2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#ifdef USE_AS_SUBMUL +# define FUNC __mpn_submul_1 +# define ADDSUBC subfe +# define ADDSUB subfc +#else +# define FUNC __mpn_addmul_1 +# define ADDSUBC adde +# define ADDSUB addc +#endif + +#define RP r3 +#define UP r4 +#define N r5 +#define VL r6 + +EALIGN(FUNC, 5, 0) + std r31, -8(r1) + rldicl. r0, N, 0, 62 + std r30, -16(r1) + cmpdi VL, r0, 2 + std r29, -24(r1) + addi N, N, 3 + std r28, -32(r1) + srdi N, N, 2 + std r27, -40(r1) + mtctr N + beq cr0, L(b00) + blt cr6, L(b01) + beq cr6, L(b10) + +L(b11): ld r9, 0(UP) + ld r28, 0(RP) + mulld r0, r9, VL + mulhdu r12, r9, VL + ADDSUB r0, r0, r28 + std r0, 0(RP) + addi RP, RP, 8 + ld r9, 8(UP) + ld r27, 16(UP) + addi UP, UP, 24 +#ifdef USE_AS_SUBMUL + subfe r11, r11, r11 +#endif + b L(bot) + + .align 4 +L(b00): ld r9, 0(UP) + ld r27, 8(UP) + ld r28, 0(RP) + ld r29, 8(RP) + mulld r0, r9, VL + mulhdu N, r9, VL + mulld r7, r27, VL + mulhdu r8, r27, VL + addc r7, r7, N + addze r12, r8 + ADDSUB r0, r0, r28 + std r0, 0(RP) + ADDSUBC r7, r7, r29 + std r7, 8(RP) + addi RP, RP, 16 + ld r9, 16(UP) + ld r27, 24(UP) + addi UP, UP, 32 +#ifdef USE_AS_SUBMUL + subfe r11, r11, r11 +#endif + b L(bot) + + .align 4 +L(b01): bdnz L(gt1) + ld r9, 0(UP) + ld r11, 0(RP) + mulld r0, r9, VL + mulhdu r8, r9, VL + ADDSUB r0, r0, r11 + std r0, 0(RP) +#ifdef USE_AS_SUBMUL + subfe r11, r11, r11 + addic r11, r11, 1 +#endif + addze RP, r8 + blr + +L(gt1): ld r9, 0(UP) + ld r27, 8(UP) + mulld r0, r9, VL + mulhdu N, r9, VL + mulld r7, r27, VL + mulhdu r8, r27, VL + ld r9, 16(UP) + ld r28, 0(RP) + ld r29, 8(RP) + ld r30, 16(RP) + mulld r11, r9, VL + mulhdu r10, r9, VL + addc r7, r7, N + adde r11, r11, r8 + addze r12, r10 + ADDSUB r0, r0, r28 + std r0, 0(RP) + ADDSUBC r7, r7, r29 + std r7, 8(RP) + ADDSUBC r11, r11, r30 + std r11, 16(RP) + addi RP, RP, 24 + ld r9, 24(UP) + ld r27, 32(UP) + addi UP, UP, 40 +#ifdef USE_AS_SUBMUL + subfe r11, r11, r11 +#endif + b L(bot) + +L(b10): addic r0, r0, r0 + li r12, 0 + ld r9, 0(UP) + ld r27, 8(UP) + bdz L(end) + addi UP, UP, 16 + + .align 4 +L(top): mulld r0, r9, VL + mulhdu N, r9, VL + mulld r7, r27, VL + mulhdu r8, r27, VL + ld r9, 0(UP) + ld r28, 0(RP) + ld r27, 8(UP) + ld r29, 8(RP) + adde r0, r0, r12 + adde r7, r7, N + mulld N, r9, VL + mulhdu r10, r9, VL + mulld r11, r27, VL + mulhdu r12, r27, VL + ld r9, 16(UP) + ld r30, 16(RP) + ld r27, 24(UP) + ld r31, 24(RP) + adde N, N, r8 + adde r11, r11, r10 + addze r12, r12 + ADDSUB r0, r0, r28 + std r0, 0(RP) + ADDSUBC r7, r7, r29 + std r7, 8(RP) + ADDSUBC N, N, r30 + std N, 16(RP) + ADDSUBC r11, r11, r31 + std r11, 24(RP) + addi UP, UP, 32 +#ifdef USE_AS_SUBMUL + subfe r11, r11, r11 +#endif + addi RP, RP, 32 +L(bot): +#ifdef USE_AS_SUBMUL + addic r11, r11, 1 +#endif + bdnz L(top) + +L(end): mulld r0, r9, VL + mulhdu N, r9, VL + mulld r7, r27, VL + mulhdu r8, r27, VL + ld r28, 0(RP) + ld r29, 8(RP) + adde r0, r0, r12 + adde r7, r7, N + addze r8, r8 + ADDSUB r0, r0, r28 + std r0, 0(RP) + ADDSUBC r7, r7, r29 + std r7, 8(RP) +#ifdef USE_AS_SUBMUL + subfe r11, r11, r11 + addic r11, r11, 1 +#endif + addze RP, r8 + ld r31, -8(r1) + ld r30, -16(r1) + ld r29, -24(r1) + ld r28, -32(r1) + ld r27, -40(r1) + blr +END(FUNC) diff --git a/sysdeps/powerpc/powerpc64/configure b/sysdeps/powerpc/powerpc64/configure index 5ddac55922..92ec607e01 100644 --- a/sysdeps/powerpc/powerpc64/configure +++ b/sysdeps/powerpc/powerpc64/configure @@ -1,4 +1,4 @@ -# This file is generated from configure.in by Autoconf. DO NOT EDIT! +# This file is generated from configure.ac by Autoconf. DO NOT EDIT! # Local configure fragment for sysdeps/powerpc/powerpc64. # The Aix ld uses global .symbol_names instead of symbol_names diff --git a/sysdeps/powerpc/powerpc64/configure.in b/sysdeps/powerpc/powerpc64/configure.ac index 67aac663d8..67aac663d8 100644 --- a/sysdeps/powerpc/powerpc64/configure.in +++ b/sysdeps/powerpc/powerpc64/configure.ac diff --git a/sysdeps/powerpc/powerpc64/crti.S b/sysdeps/powerpc/powerpc64/crti.S index 967dc669bb..6e1ece8d66 100644 --- a/sysdeps/powerpc/powerpc64/crti.S +++ b/sysdeps/powerpc/powerpc64/crti.S @@ -60,23 +60,14 @@ .LC0: .tc PREINIT_FUNCTION[TC], PREINIT_FUNCTION #endif - .type BODY_LABEL (_init), @function - .globl _init - .section ".opd", "aw" - .align 3 -_init: OPD_ENT (_init) -#ifdef HAVE_ASM_GLOBAL_DOT_NAME - .globl BODY_LABEL (_init) - .size _init, 24 -#else - .type _init, @function -#endif .section ".init", "ax", @progbits + ENTRY_2(_init) .align ALIGNARG (2) BODY_LABEL (_init): + LOCALENTRY(_init) mflr 0 - std 0, 16(r1) - stdu r1, -112(r1) + std 0, FRAME_LR_SAVE(r1) + stdu r1, -FRAME_MIN_SIZE_PARM(r1) #if PREINIT_FUNCTION_WEAK addis r9, r2, .LC0@toc@ha ld r0, .LC0@toc@l(r9) @@ -87,20 +78,11 @@ BODY_LABEL (_init): nop 1: - .type BODY_LABEL (_fini), @function - .globl _fini - .section ".opd", "aw" - .align 3 -_fini: OPD_ENT (_fini) -#ifdef HAVE_ASM_GLOBAL_DOT_NAME - .globl BODY_LABEL (_fini) - .size _fini, 24 -#else - .type _fini, @function -#endif .section ".fini", "ax", @progbits + ENTRY_2(_fini) .align ALIGNARG (2) BODY_LABEL (_fini): + LOCALENTRY(_fini) mflr 0 - std 0, 16(r1) - stdu r1, -112(r1) + std 0, FRAME_LR_SAVE(r1) + stdu r1, -FRAME_MIN_SIZE_PARM(r1) diff --git a/sysdeps/powerpc/powerpc64/crtn.S b/sysdeps/powerpc/powerpc64/crtn.S index 364e53aae7..cdd3b0f340 100644 --- a/sysdeps/powerpc/powerpc64/crtn.S +++ b/sysdeps/powerpc/powerpc64/crtn.S @@ -39,13 +39,13 @@ #include <sysdep.h> .section .init,"ax",@progbits - addi r1, r1, 112 - ld r0, 16(r1) + addi r1, r1, FRAME_MIN_SIZE_PARM + ld r0, FRAME_LR_SAVE(r1) mtlr r0 blr .section .fini,"ax",@progbits - addi r1, r1, 112 - ld r0, 16(r1) + addi r1, r1, FRAME_MIN_SIZE_PARM + ld r0, FRAME_LR_SAVE(r1) mtlr r0 blr diff --git a/sysdeps/powerpc/powerpc64/dl-irel.h b/sysdeps/powerpc/powerpc64/dl-irel.h index d85c6148ac..a500aa6f4c 100644 --- a/sysdeps/powerpc/powerpc64/dl-irel.h +++ b/sysdeps/powerpc/powerpc64/dl-irel.h @@ -50,7 +50,11 @@ elf_irela (const Elf64_Rela *reloc) { Elf64_Addr *const reloc_addr = (void *) reloc->r_offset; Elf64_Addr value = elf_ifunc_invoke(reloc->r_addend); +#if _CALL_ELF != 2 *(Elf64_FuncDesc *) reloc_addr = *(Elf64_FuncDesc *) value; +#else + *reloc_addr = value; +#endif } else __libc_fatal ("unexpected reloc type in static binary"); diff --git a/sysdeps/powerpc/powerpc64/dl-machine.h b/sysdeps/powerpc/powerpc64/dl-machine.h index 059fdafd53..36f3916999 100644 --- a/sysdeps/powerpc/powerpc64/dl-machine.h +++ b/sysdeps/powerpc/powerpc64/dl-machine.h @@ -31,6 +31,7 @@ in l_info array. */ #define DT_PPC64(x) (DT_PPC64_##x - DT_LOPROC + DT_NUM) +#if _CALL_ELF != 2 /* A PowerPC64 function descriptor. The .plt (procedure linkage table) and .opd (official procedure descriptor) sections are arrays of these. */ @@ -40,6 +41,7 @@ typedef struct Elf64_Addr fd_toc; Elf64_Addr fd_aux; } Elf64_FuncDesc; +#endif #define ELF_MULT_MACHINES_SUPPORTED @@ -47,6 +49,18 @@ typedef struct static inline int elf_machine_matches_host (const Elf64_Ehdr *ehdr) { + /* Verify that the binary matches our ABI version. */ + if ((ehdr->e_flags & EF_PPC64_ABI) != 0) + { +#if _CALL_ELF != 2 + if ((ehdr->e_flags & EF_PPC64_ABI) != 1) + return 0; +#else + if ((ehdr->e_flags & EF_PPC64_ABI) != 2) + return 0; +#endif + } + return ehdr->e_machine == EM_PPC64; } @@ -122,15 +136,9 @@ elf_machine_dynamic (void) #define RTLD_START \ asm (".pushsection \".text\"\n" \ " .align 2\n" \ -" .type " BODY_PREFIX "_start,@function\n" \ -" .pushsection \".opd\",\"aw\"\n" \ -" .align 3\n" \ -" .globl _start\n" \ " " ENTRY_2(_start) "\n" \ -"_start:\n" \ -" " OPD_ENT(_start) "\n" \ -" .popsection\n" \ BODY_PREFIX "_start:\n" \ +" " LOCALENTRY(_start) "\n" \ /* We start with the following on the stack, from top: \ argc (4 bytes); \ arguments for program (terminated by NULL); \ @@ -154,11 +162,6 @@ BODY_PREFIX "_start:\n" \ ".LT__start_name_end:\n" \ " .align 2\n" \ " " END_2(_start) "\n" \ -" .globl _dl_start_user\n" \ -" .pushsection \".opd\",\"aw\"\n" \ -"_dl_start_user:\n" \ -" " OPD_ENT(_dl_start_user) "\n" \ -" .popsection\n" \ " .pushsection \".toc\",\"aw\"\n" \ DL_STARTING_UP_DEF \ ".LC__rtld_local:\n" \ @@ -170,7 +173,6 @@ DL_STARTING_UP_DEF \ ".LC__dl_fini:\n" \ " .tc _dl_fini[TC],_dl_fini\n" \ " .popsection\n" \ -" .type " BODY_PREFIX "_dl_start_user,@function\n" \ " " ENTRY_2(_dl_start_user) "\n" \ /* Now, we do our main work of calling initialisation procedures. \ The ELF ABI doesn't say anything about parameters for these, \ @@ -178,6 +180,7 @@ DL_STARTING_UP_DEF \ Changing these is strongly discouraged (not least because argc is \ passed by value!). */ \ BODY_PREFIX "_dl_start_user:\n" \ +" " LOCALENTRY(_dl_start_user) "\n" \ /* the address of _start in r30. */ \ " mr 30,3\n" \ /* &_dl_argc in 29, &_dl_argv in 27, and _dl_loaded in 28. */ \ @@ -228,10 +231,7 @@ BODY_PREFIX "_dl_start_user:\n" \ /* Now, call the start function descriptor at r30... */ \ " .globl ._dl_main_dispatch\n" \ "._dl_main_dispatch:\n" \ -" ld 0,0(30)\n" \ -" ld 2,8(30)\n" \ -" mtctr 0\n" \ -" ld 11,16(30)\n" \ +" " PPC64_LOAD_FUNCPTR(30) "\n" \ " bctr\n" \ ".LT__dl_start_user:\n" \ " .long 0\n" \ @@ -272,8 +272,22 @@ BODY_PREFIX "_dl_start_user:\n" \ relocations behave "normally", ie. always use the real address like PLT relocations. So always set ELF_RTYPE_CLASS_PLT. */ +#if _CALL_ELF != 2 #define elf_machine_type_class(type) \ (ELF_RTYPE_CLASS_PLT | (((type) == R_PPC64_COPY) * ELF_RTYPE_CLASS_COPY)) +#else +/* And now that you have read that large comment, you can disregard it + all for ELFv2. ELFv2 does need the special SHN_UNDEF treatment. */ +#define IS_PPC64_TLS_RELOC(R) \ + (((R) >= R_PPC64_TLS && (R) <= R_PPC64_DTPREL16_HIGHESTA) \ + || ((R) >= R_PPC64_TPREL16_HIGH && (R) <= R_PPC64_DTPREL16_HIGHA)) + +#define elf_machine_type_class(type) \ + ((((type) == R_PPC64_JMP_SLOT \ + || (type) == R_PPC64_ADDR24 \ + || IS_PPC64_TLS_RELOC (type)) * ELF_RTYPE_CLASS_PLT) \ + | (((type) == R_PPC64_COPY) * ELF_RTYPE_CLASS_COPY)) +#endif /* A reloc type used for ld.so cmdline arg lookups to reject PLT entries. */ #define ELF_MACHINE_JMP_SLOT R_PPC64_JMP_SLOT @@ -282,8 +296,19 @@ BODY_PREFIX "_dl_start_user:\n" \ #define ELF_MACHINE_NO_REL 1 /* Stuff for the PLT. */ +#if _CALL_ELF != 2 #define PLT_INITIAL_ENTRY_WORDS 3 +#define PLT_ENTRY_WORDS 3 +#define GLINK_INITIAL_ENTRY_WORDS 8 +/* The first 32k entries of glink can set an index and branch using two + instructions; past that point, glink uses three instructions. */ +#define GLINK_ENTRY_WORDS(I) (((I) < 0x8000)? 2 : 3) +#else +#define PLT_INITIAL_ENTRY_WORDS 2 +#define PLT_ENTRY_WORDS 1 #define GLINK_INITIAL_ENTRY_WORDS 8 +#define GLINK_ENTRY_WORDS(I) 1 +#endif #define PPC_DCBST(where) asm volatile ("dcbst 0,%0" : : "r"(where) : "memory") #define PPC_DCBT(where) asm volatile ("dcbt 0,%0" : : "r"(where) : "memory") @@ -328,38 +353,45 @@ elf_machine_runtime_setup (struct link_map *map, int lazy, int profile) if (lazy) { - /* The function descriptor of the appropriate trampoline - routine is used to set the 1st and 2nd doubleword of the - plt_reserve. */ - Elf64_FuncDesc *resolve_fd; Elf64_Word glink_offset; - /* the plt_reserve area is the 1st 3 doublewords of the PLT */ - Elf64_FuncDesc *plt_reserve = (Elf64_FuncDesc *) plt; Elf64_Word offset; + Elf64_Addr dlrr; - resolve_fd = (Elf64_FuncDesc *) (profile ? _dl_profile_resolve - : _dl_runtime_resolve); + dlrr = (Elf64_Addr) (profile ? _dl_profile_resolve + : _dl_runtime_resolve); if (profile && GLRO(dl_profile) != NULL && _dl_name_match_p (GLRO(dl_profile), map)) /* This is the object we are looking for. Say that we really want profiling and the timers are started. */ GL(dl_profile_map) = map; - +#if _CALL_ELF != 2 /* We need to stuff the address/TOC of _dl_runtime_resolve into doublewords 0 and 1 of plt_reserve. Then we need to stuff the map address into doubleword 2 of plt_reserve. This allows the GLINK0 code to transfer control to the correct trampoline which will transfer control to fixup in dl-machine.c. */ - plt_reserve->fd_func = resolve_fd->fd_func; - plt_reserve->fd_toc = resolve_fd->fd_toc; - plt_reserve->fd_aux = (Elf64_Addr) map; + { + /* The plt_reserve area is the 1st 3 doublewords of the PLT. */ + Elf64_FuncDesc *plt_reserve = (Elf64_FuncDesc *) plt; + Elf64_FuncDesc *resolve_fd = (Elf64_FuncDesc *) dlrr; + plt_reserve->fd_func = resolve_fd->fd_func; + plt_reserve->fd_toc = resolve_fd->fd_toc; + plt_reserve->fd_aux = (Elf64_Addr) map; #ifdef RTLD_BOOTSTRAP - /* When we're bootstrapping, the opd entry will not have - been relocated yet. */ - plt_reserve->fd_func += l_addr; - plt_reserve->fd_toc += l_addr; + /* When we're bootstrapping, the opd entry will not have + been relocated yet. */ + plt_reserve->fd_func += l_addr; + plt_reserve->fd_toc += l_addr; +#endif + } +#else + /* When we don't have function descriptors, the first doubleword + of the PLT holds the address of _dl_runtime_resolve, and the + second doubleword holds the map address. */ + plt[0] = dlrr; + plt[1] = (Elf64_Addr) map; #endif /* Set up the lazy PLT entries. */ @@ -370,14 +402,8 @@ elf_machine_runtime_setup (struct link_map *map, int lazy, int profile) { plt[offset] = (Elf64_Xword) &glink[glink_offset]; - offset += 3; - /* The first 32k entries of glink can set an index and - branch using two instructions; Past that point, - glink uses three instructions. */ - if (i < 0x8000) - glink_offset += 2; - else - glink_offset += 3; + offset += PLT_ENTRY_WORDS; + glink_offset += GLINK_ENTRY_WORDS (i); } /* Now, we've modified data. We need to write the changes from @@ -398,6 +424,42 @@ elf_machine_runtime_setup (struct link_map *map, int lazy, int profile) return lazy; } +#if _CALL_ELF == 2 +/* If the PLT entry whose reloc is 'reloc' resolves to a function in + the same object, return the target function's local entry point + offset if usable. */ +static inline Elf64_Addr __attribute__ ((always_inline)) +ppc64_local_entry_offset (struct link_map *map, lookup_t sym_map, + const Elf64_Rela *reloc) +{ + const Elf64_Sym *symtab; + const Elf64_Sym *sym; + + /* If the target function is in a different object, we cannot + use the local entry point. */ + if (sym_map != map) + return 0; + + /* If the linker inserted multiple TOCs, we cannot use the + local entry point. */ + if (map->l_info[DT_PPC64(OPT)] + && (map->l_info[DT_PPC64(OPT)]->d_un.d_val & PPC64_OPT_MULTI_TOC)) + return 0; + + /* Otherwise, we can use the local entry point. Retrieve its offset + from the symbol's ELF st_other field. */ + symtab = (const void *) D_PTR (map, l_info[DT_SYMTAB]); + sym = &symtab[ELFW(R_SYM) (reloc->r_info)]; + + /* If the target function is an ifunc then the local entry offset is + for the resolver, not the final destination. */ + if (__builtin_expect (ELFW(ST_TYPE) (sym->st_info) == STT_GNU_IFUNC, 0)) + return 0; + + return PPC64_LOCAL_ENTRY_OFFSET (sym->st_other); +} +#endif + /* Change the PLT entry whose reloc is 'reloc' to call the actual routine. */ static inline Elf64_Addr __attribute__ ((always_inline)) @@ -405,6 +467,7 @@ elf_machine_fixup_plt (struct link_map *map, lookup_t sym_map, const Elf64_Rela *reloc, Elf64_Addr *reloc_addr, Elf64_Addr finaladdr) { +#if _CALL_ELF != 2 Elf64_FuncDesc *plt = (Elf64_FuncDesc *) reloc_addr; Elf64_FuncDesc *rel = (Elf64_FuncDesc *) finaladdr; Elf64_Addr offset = 0; @@ -442,13 +505,20 @@ elf_machine_fixup_plt (struct link_map *map, lookup_t sym_map, plt->fd_func = rel->fd_func + offset; PPC_DCBST (&plt->fd_func); PPC_ISYNC; +#else + finaladdr += ppc64_local_entry_offset (map, sym_map, reloc); + *reloc_addr = finaladdr; +#endif return finaladdr; } static inline void __attribute__ ((always_inline)) -elf_machine_plt_conflict (Elf64_Addr *reloc_addr, Elf64_Addr finaladdr) +elf_machine_plt_conflict (struct link_map *map, lookup_t sym_map, + const Elf64_Rela *reloc, + Elf64_Addr *reloc_addr, Elf64_Addr finaladdr) { +#if _CALL_ELF != 2 Elf64_FuncDesc *plt = (Elf64_FuncDesc *) reloc_addr; Elf64_FuncDesc *rel = (Elf64_FuncDesc *) finaladdr; @@ -459,6 +529,10 @@ elf_machine_plt_conflict (Elf64_Addr *reloc_addr, Elf64_Addr finaladdr) PPC_DCBST (&plt->fd_aux); PPC_DCBST (&plt->fd_toc); PPC_SYNC; +#else + finaladdr += ppc64_local_entry_offset (map, sym_map, reloc); + *reloc_addr = finaladdr; +#endif } /* Return the final value of a plt relocation. */ @@ -471,8 +545,13 @@ elf_machine_plt_value (struct link_map *map, const Elf64_Rela *reloc, /* Names of the architecture-specific auditing callback functions. */ +#if _CALL_ELF != 2 #define ARCH_LA_PLTENTER ppc64_gnu_pltenter #define ARCH_LA_PLTEXIT ppc64_gnu_pltexit +#else +#define ARCH_LA_PLTENTER ppc64v2_gnu_pltenter +#define ARCH_LA_PLTEXIT ppc64v2_gnu_pltexit +#endif #endif /* dl_machine_h */ @@ -528,6 +607,7 @@ auto inline Elf64_Addr __attribute__ ((always_inline)) resolve_ifunc (Elf64_Addr value, const struct link_map *map, const struct link_map *sym_map) { +#if _CALL_ELF != 2 #ifndef RESOLVE_CONFLICT_FIND_MAP /* The function we are calling may not yet have its opd entry relocated. */ Elf64_FuncDesc opd; @@ -545,6 +625,7 @@ resolve_ifunc (Elf64_Addr value, value = (Elf64_Addr) &opd; } #endif +#endif return ((Elf64_Addr (*) (unsigned long int)) value) (GLRO(dl_hwcap)); } @@ -561,6 +642,12 @@ elf_machine_rela (struct link_map *map, Elf64_Addr *const reloc_addr = reloc_addr_arg; const int r_type = ELF64_R_TYPE (reloc->r_info); const Elf64_Sym *const refsym = sym; + union unaligned + { + uint16_t u2; + uint32_t u4; + uint64_t u8; + } __attribute__ ((__packed__)); if (r_type == R_PPC64_RELATIVE) { @@ -604,7 +691,7 @@ elf_machine_rela (struct link_map *map, /* Fall thru */ case R_PPC64_JMP_SLOT: #ifdef RESOLVE_CONFLICT_FIND_MAP - elf_machine_plt_conflict (reloc_addr, value); + elf_machine_plt_conflict (map, sym_map, reloc, reloc_addr, value); #else elf_machine_fixup_plt (map, sym_map, reloc, reloc_addr, value); #endif @@ -663,11 +750,25 @@ elf_machine_rela (struct link_map *map, case R_PPC64_TPREL16_HI: value = elf_machine_tprel (map, sym_map, sym, reloc); + if (dont_expect (value + 0x80000000 >= 0x100000000LL)) + _dl_reloc_overflow (map, "R_PPC64_TPREL16_HI", reloc_addr, refsym); + *(Elf64_Half *) reloc_addr = PPC_HI (value); + break; + + case R_PPC64_TPREL16_HIGH: + value = elf_machine_tprel (map, sym_map, sym, reloc); *(Elf64_Half *) reloc_addr = PPC_HI (value); break; case R_PPC64_TPREL16_HA: value = elf_machine_tprel (map, sym_map, sym, reloc); + if (dont_expect (value + 0x80008000 >= 0x100000000LL)) + _dl_reloc_overflow (map, "R_PPC64_TPREL16_HA", reloc_addr, refsym); + *(Elf64_Half *) reloc_addr = PPC_HA (value); + break; + + case R_PPC64_TPREL16_HIGHA: + value = elf_machine_tprel (map, sym_map, sym, reloc); *(Elf64_Half *) reloc_addr = PPC_HA (value); break; @@ -703,17 +804,23 @@ elf_machine_rela (struct link_map *map, break; case R_PPC64_ADDR16_HI: + if (dont_expect (value + 0x80000000 >= 0x100000000LL)) + _dl_reloc_overflow (map, "R_PPC64_ADDR16_HI", reloc_addr, refsym); + case R_PPC64_ADDR16_HIGH: *(Elf64_Half *) reloc_addr = PPC_HI (value); break; case R_PPC64_ADDR16_HA: + if (dont_expect (value + 0x80008000 >= 0x100000000LL)) + _dl_reloc_overflow (map, "R_PPC64_ADDR16_HA", reloc_addr, refsym); + case R_PPC64_ADDR16_HIGHA: *(Elf64_Half *) reloc_addr = PPC_HA (value); break; case R_PPC64_ADDR30: { Elf64_Addr delta = value - (Elf64_Xword) reloc_addr; - if (dont_expect ((delta + 0x80000000) >= 0x10000000 + if (dont_expect ((delta + 0x80000000) >= 0x100000000LL || (delta & 3) != 0)) _dl_reloc_overflow (map, "R_PPC64_ADDR30", reloc_addr, refsym); BIT_INSERT (*(Elf64_Word *) reloc_addr, delta, 0xfffffffc); @@ -741,27 +848,15 @@ elf_machine_rela (struct link_map *map, return; case R_PPC64_UADDR64: - /* We are big-endian. */ - ((char *) reloc_addr_arg)[0] = (value >> 56) & 0xff; - ((char *) reloc_addr_arg)[1] = (value >> 48) & 0xff; - ((char *) reloc_addr_arg)[2] = (value >> 40) & 0xff; - ((char *) reloc_addr_arg)[3] = (value >> 32) & 0xff; - ((char *) reloc_addr_arg)[4] = (value >> 24) & 0xff; - ((char *) reloc_addr_arg)[5] = (value >> 16) & 0xff; - ((char *) reloc_addr_arg)[6] = (value >> 8) & 0xff; - ((char *) reloc_addr_arg)[7] = (value >> 0) & 0xff; + ((union unaligned *) reloc_addr)->u8 = value; return; case R_PPC64_UADDR32: - /* We are big-endian. */ - ((char *) reloc_addr_arg)[0] = (value >> 24) & 0xff; - ((char *) reloc_addr_arg)[1] = (value >> 16) & 0xff; - ((char *) reloc_addr_arg)[2] = (value >> 8) & 0xff; - ((char *) reloc_addr_arg)[3] = (value >> 0) & 0xff; + ((union unaligned *) reloc_addr)->u4 = value; return; case R_PPC64_ADDR32: - if (dont_expect ((value + 0x80000000) >= 0x10000000)) + if (dont_expect ((value + 0x80000000) >= 0x100000000LL)) _dl_reloc_overflow (map, "R_PPC64_ADDR32", reloc_addr, refsym); *(Elf64_Word *) reloc_addr = value; return; @@ -781,10 +876,8 @@ elf_machine_rela (struct link_map *map, case R_PPC64_UADDR16: if (dont_expect ((value + 0x8000) >= 0x10000)) _dl_reloc_overflow (map, "R_PPC64_UADDR16", reloc_addr, refsym); - /* We are big-endian. */ - ((char *) reloc_addr_arg)[0] = (value >> 8) & 0xff; - ((char *) reloc_addr_arg)[1] = (value >> 0) & 0xff; - break; + ((union unaligned *) reloc_addr)->u2 = value; + return; case R_PPC64_ADDR16_DS: if (dont_expect ((value + 0x8000) >= 0x10000 || (value & 3) != 0)) diff --git a/sysdeps/powerpc/powerpc64/dl-trampoline.S b/sysdeps/powerpc/powerpc64/dl-trampoline.S index 4dde2763a3..69ce523e90 100644 --- a/sysdeps/powerpc/powerpc64/dl-trampoline.S +++ b/sysdeps/powerpc/powerpc64/dl-trampoline.S @@ -26,13 +26,13 @@ parm1 (r3) and the index (r0) need to be converted to an offset (index * 24) in parm2 (r4). */ -#define FRAME_SIZE 176 +#define FRAME_SIZE (FRAME_MIN_SIZE+64) /* We need to save the registers used to pass parameters, ie. r3 thru r10; Use local var space rather than the parameter save area, because gcc as of 2010/05 doesn't allocate a proper stack frame for a function that makes no calls except for __tls_get_addr and we might be here resolving the __tls_get_addr call. */ -#define INT_PARMS 112 +#define INT_PARMS FRAME_MIN_SIZE EALIGN(_dl_runtime_resolve, 4, 0) stdu r1,-FRAME_SIZE(r1) cfi_adjust_cfa_offset (FRAME_SIZE) @@ -48,36 +48,31 @@ EALIGN(_dl_runtime_resolve, 4, 0) mflr r0 std r8,INT_PARMS+40(r1) /* Store the LR in the LR Save area. */ - std r0,FRAME_SIZE+16(r1) - cfi_offset (lr, 16) - mfcr r0 + std r0,FRAME_SIZE+FRAME_LR_SAVE(r1) + cfi_offset (lr, FRAME_LR_SAVE) std r9,INT_PARMS+48(r1) std r10,INT_PARMS+56(r1) -/* I'm almost certain we don't have to save cr... be safe. */ - std r0,FRAME_SIZE+8(r1) bl JUMPTARGET(_dl_fixup) #ifndef SHARED nop #endif /* Put the registers back. */ - ld r0,FRAME_SIZE+16(r1) + ld r0,FRAME_SIZE+FRAME_LR_SAVE(r1) ld r10,INT_PARMS+56(r1) ld r9,INT_PARMS+48(r1) ld r8,INT_PARMS+40(r1) ld r7,INT_PARMS+32(r1) mtlr r0 - ld r0,FRAME_SIZE+8(r1) ld r6,INT_PARMS+24(r1) ld r5,INT_PARMS+16(r1) ld r4,INT_PARMS+8(r1) - mtcrf 0xFF,r0 -/* Load the target address, toc and static chain reg from the function - descriptor returned by fixup. */ - ld r0,0(r3) - ld r2,8(r3) - mtctr r0 - ld r11,16(r3) +/* Prepare for calling the function returned by fixup. */ + PPC64_LOAD_FUNCPTR r3 ld r3,INT_PARMS+0(r1) +#if _CALL_ELF == 2 +/* Restore the caller's TOC in case we jump to a local entry point. */ + ld r2,FRAME_SIZE+FRAME_TOC_SAVE(r1) +#endif /* Unwind the stack frame, and jump. */ addi r1,r1,FRAME_SIZE bctr @@ -85,17 +80,30 @@ END(_dl_runtime_resolve) #undef FRAME_SIZE #undef INT_PARMS - /* Stack layout: - +592 previous backchain - +584 spill_r31 - +576 spill_r30 - +560 v1 - +552 fp4 - +544 fp3 - +536 fp2 - +528 fp1 - +520 r4 - +512 r3 + /* Stack layout: ELFv2 ABI. + +752 previous backchain + +744 spill_r31 + +736 spill_r30 + +720 v8 + +704 v7 + +688 v6 + +672 v5 + +656 v4 + +640 v3 + +624 v2 + +608 v1 + +600 fp10 + ELFv1 ABI +592 fp9 + +592 previous backchain +584 fp8 + +584 spill_r31 +576 fp7 + +576 spill_r30 +568 fp6 + +560 v1 +560 fp5 + +552 fp4 +552 fp4 + +544 fp3 +544 fp3 + +536 fp2 +536 fp2 + +528 fp1 +528 fp1 + +520 r4 +520 r4 + +512 r3 +512 r3 return values +504 free +496 stackframe @@ -147,18 +155,24 @@ END(_dl_runtime_resolve) +64 parm3 +56 parm2 +48 parm1 - * Parameter save area, Allocated by the call, at least 8 double words - +40 TOC save area - +32 Reserved for linker - +24 Reserved for compiler + * Parameter save area + * (v1 ABI: Allocated by the call, at least 8 double words) + +40 v1 ABI: TOC save area + +32 v1 ABI: Reserved for linker + +24 v1 ABI: Reserved for compiler / v2 ABI: TOC save area +16 LR save area +8 CR save area r1+0 stack back chain */ -#define FRAME_SIZE 592 +#if _CALL_ELF == 2 +# define FRAME_SIZE 752 +# define VR_RTN 608 +#else +# define FRAME_SIZE 592 +# define VR_RTN 560 +#endif #define INT_RTN 512 #define FPR_RTN 528 -#define VR_RTN 560 #define STACK_FRAME 496 #define CALLING_LR 488 #define CALLING_SP 480 @@ -203,18 +217,14 @@ EALIGN(_dl_profile_resolve, 4, 0) mflr r5 std r7,INT_PARMS+32(r1) std r8,INT_PARMS+40(r1) -/* Store the LR in the LR Save area of the previous frame. */ -/* XXX Do we have to do this? */ +/* Store the LR in the LR Save area. */ la r8,FRAME_SIZE(r1) - std r5,FRAME_SIZE+16(r1) - cfi_offset (lr, 16) + std r5,FRAME_SIZE+FRAME_LR_SAVE(r1) + cfi_offset (lr, FRAME_LR_SAVE) std r5,CALLING_LR(r1) - mfcr r0 std r9,INT_PARMS+48(r1) std r10,INT_PARMS+56(r1) std r8,CALLING_SP(r1) -/* I'm almost certain we don't have to save cr... be safe. */ - std r0,FRAME_SIZE+8(r1) ld r12,.LC__dl_hwcap@toc(r2) #ifdef SHARED /* Load _rtld_local_ro._dl_hwcap. */ @@ -311,24 +321,22 @@ L(saveFP): lvx v12,r11,r10 lvx v13,r11,r9 L(restoreFXR): - ld r0,FRAME_SIZE+16(r1) + ld r0,FRAME_SIZE+FRAME_LR_SAVE(r1) ld r10,INT_PARMS+56(r1) ld r9,INT_PARMS+48(r1) ld r8,INT_PARMS+40(r1) ld r7,INT_PARMS+32(r1) mtlr r0 - ld r0,FRAME_SIZE+8(r1) ld r6,INT_PARMS+24(r1) ld r5,INT_PARMS+16(r1) ld r4,INT_PARMS+8(r1) - mtcrf 0xFF,r0 -/* Load the target address, toc and static chain reg from the function - descriptor returned by fixup. */ - ld r0,0(r3) - ld r2,8(r3) - ld r11,16(r3) +/* Prepare for calling the function returned by fixup. */ + PPC64_LOAD_FUNCPTR r3 ld r3,INT_PARMS+0(r1) - mtctr r0 +#if _CALL_ELF == 2 +/* Restore the caller's TOC in case we jump to a local entry point. */ + ld r2,FRAME_SIZE+FRAME_TOC_SAVE(r1) +#endif /* Load the floating point registers. */ lfd fp1,FPR_PARMS+0(r1) lfd fp2,FPR_PARMS+8(r1) @@ -344,10 +352,11 @@ L(restoreFXR): lfd fp12,FPR_PARMS+88(r1) lfd fp13,FPR_PARMS+96(r1) /* Unwind the stack frame, and jump. */ - ld r31,584(r1) - ld r30,576(r1) + ld r31,FRAME_SIZE-8(r1) + ld r30,FRAME_SIZE-16(r1) addi r1,r1,FRAME_SIZE bctr + L(do_pltexit): la r10,(VR_PARMS+0)(r1) la r9,(VR_PARMS+16)(r1) @@ -375,25 +384,19 @@ L(do_pltexit): lvx v12,r11,r10 lvx v13,r11,r9 L(restoreFXR2): - ld r0,FRAME_SIZE+16(r1) + ld r0,FRAME_SIZE+FRAME_LR_SAVE(r1) ld r10,INT_PARMS+56(r1) ld r9,INT_PARMS+48(r1) ld r8,INT_PARMS+40(r1) ld r7,INT_PARMS+32(r1) mtlr r0 - ld r0,FRAME_SIZE+8(r1) ld r6,INT_PARMS+24(r1) ld r5,INT_PARMS+16(r1) ld r4,INT_PARMS+8(r1) - mtcrf 0xFF,r0 -/* Load the target address, toc and static chain reg from the function - descriptor returned by fixup. */ - ld r0,0(r3) - std r2,40(r1) - ld r2,8(r3) - ld r11,16(r3) +/* Prepare for calling the function returned by fixup. */ + std r2,FRAME_TOC_SAVE(r1) + PPC64_LOAD_FUNCPTR r3 ld r3,INT_PARMS+0(r1) - mtctr r0 /* Load the floating point registers. */ lfd fp1,FPR_PARMS+0(r1) lfd fp2,FPR_PARMS+8(r1) @@ -410,21 +413,42 @@ L(restoreFXR2): lfd fp13,FPR_PARMS+96(r1) /* Call the target function. */ bctrl - ld r2,40(r1) + ld r2,FRAME_TOC_SAVE(r1) lwz r12,VR_VRSAVE(r1) /* But return here and store the return values. */ std r3,INT_RTN(r1) std r4,INT_RTN+8(r1) - stfd fp1,FPR_PARMS+0(r1) - stfd fp2,FPR_PARMS+8(r1) + stfd fp1,FPR_RTN+0(r1) + stfd fp2,FPR_RTN+8(r1) cmpdi cr0,r12,0 la r10,VR_RTN(r1) - stfd fp3,FPR_PARMS+16(r1) - stfd fp4,FPR_PARMS+24(r1) + stfd fp3,FPR_RTN+16(r1) + stfd fp4,FPR_RTN+24(r1) +#if _CALL_ELF == 2 + la r12,VR_RTN+16(r1) + stfd fp5,FPR_RTN+32(r1) + stfd fp6,FPR_RTN+40(r1) + li r5,32 + li r6,64 + stfd fp7,FPR_RTN+48(r1) + stfd fp8,FPR_RTN+56(r1) + stfd fp9,FPR_RTN+64(r1) + stfd fp10,FPR_RTN+72(r1) +#endif mr r3,r31 mr r4,r30 beq L(callpltexit) stvx v2,0,r10 +#if _CALL_ELF == 2 + stvx v3,0,r12 + stvx v4,r5,r10 + stvx v5,r5,r12 + addi r5,r5,64 + stvx v6,r6,r10 + stvx v7,r6,r12 + stvx v8,r5,r10 + stvx v9,r5,r12 +#endif L(callpltexit): addi r5,r1,INT_PARMS addi r6,r1,INT_RTN @@ -436,18 +460,39 @@ L(callpltexit): lwz r12,VR_VRSAVE(r1) ld r3,INT_RTN(r1) ld r4,INT_RTN+8(r1) - lfd fp1,FPR_PARMS+0(r1) - lfd fp2,FPR_PARMS+8(r1) + lfd fp1,FPR_RTN+0(r1) + lfd fp2,FPR_RTN+8(r1) cmpdi cr0,r12,0 - la r10,VR_RTN(r1) - lfd fp3,FPR_PARMS+16(r1) - lfd fp4,FPR_PARMS+24(r1) + la r11,VR_RTN(r1) + lfd fp3,FPR_RTN+16(r1) + lfd fp4,FPR_RTN+24(r1) +#if _CALL_ELF == 2 + la r12,VR_RTN+16(r1) + lfd fp5,FPR_RTN+32(r1) + lfd fp6,FPR_RTN+40(r1) + li r30,32 + li r31,64 + lfd fp7,FPR_RTN+48(r1) + lfd fp8,FPR_RTN+56(r1) + lfd fp9,FPR_RTN+64(r1) + lfd fp10,FPR_RTN+72(r1) +#endif beq L(pltexitreturn) - lvx v2,0,r10 + lvx v2,0,r11 +#if _CALL_ELF == 2 + lvx v3,0,r12 + lvx v4,r30,r11 + lvx v5,r30,r12 + addi r30,r30,64 + lvx v6,r31,r11 + lvx v7,r31,r12 + lvx v8,r30,r11 + lvx v9,r30,r12 +#endif L(pltexitreturn): - ld r0,FRAME_SIZE+16(r1) - ld r31,584(r1) - ld r30,576(r1) + ld r0,FRAME_SIZE+FRAME_LR_SAVE(r1) + ld r31,FRAME_SIZE-8(r1) + ld r30,FRAME_SIZE-16(r1) mtlr r0 ld r1,0(r1) blr diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile new file mode 100644 index 0000000000..ebf957e40b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile @@ -0,0 +1,33 @@ +ifeq ($(subdir),math) +sysdep_routines += s_isnan-power7 s_isnan-power6x s_isnan-power6 \ + s_isnan-power5 s_isnan-ppc64 s_copysign-power6 \ + s_copysign-ppc64 s_finite-power7 s_finite-ppc64 \ + s_finitef-ppc64 s_isinff-ppc64 s_isinf-power7 \ + s_isinf-ppc64 s_modf-power5+ s_modf-ppc64 \ + s_modff-power5+ s_modff-ppc64 + +libm-sysdep_routines += s_isnan-power7 s_isnan-power6x s_isnan-power6 \ + s_isnan-power5 s_isnan-ppc64 s_llround-power6x \ + s_llround-power5+ s_llround-ppc64 s_ceil-power5+ \ + s_ceil-ppc64 s_ceilf-power5+ s_ceilf-ppc64 \ + s_floor-power5+ s_floor-ppc64 s_floorf-power5+ \ + s_floorf-ppc64 s_round-power5+ s_round-ppc64 \ + s_roundf-power5+ s_roundf-ppc64 s_trunc-power5+ \ + s_trunc-ppc64 s_truncf-power5+ s_truncf-ppc64 \ + s_copysign-power6 s_copysign-ppc64 s_llrint-power6x \ + s_llrint-ppc64 s_finite-power7 s_finite-ppc64 \ + s_finitef-ppc64 s_isinff-ppc64 s_isinf-power7 \ + s_isinf-ppc64 s_logb-power7 s_logbf-power7 \ + s_logbl-power7 s_logb-ppc64 s_logbf-ppc64 \ + s_logbl-ppc64 s_modf-power5+ s_modf-ppc64 \ + s_modff-power5+ s_modff-ppc64 e_hypot-ppc64 \ + e_hypot-power7 e_hypotf-ppc64 e_hypotf-power7 + +CFLAGS-s_logbf-power7.c = -mcpu=power7 +CFLAGS-s_logbl-power7.c = -mcpu=power7 +CFLAGS-s_logb-power7.c = -mcpu=power7 +CFLAGS-s_modf-power5+.c = -mcpu=power5+ +CFLAGS-s_modff-power5+.c = -mcpu=power5+ +CFLAGS-e_hypot-power7.c = -mcpu=power7 +CFLAGS-e_hypotf-power7.c = -mcpu=power7 +endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypof.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypof.c new file mode 100644 index 0000000000..526130223b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypof.c @@ -0,0 +1,32 @@ +/* Multiple versions of ieee754_hypot. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__ieee754_hypotf) __ieee754_hypotf_ppc64 attribute_hidden; +extern __typeof (__ieee754_hypotf) __ieee754_hypotf_power7 attribute_hidden; + +libc_ifunc (__ieee754_hypotf, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __ieee754_hypotf_power7 + : __ieee754_hypotf_ppc64); + +strong_alias (__ieee754_hypotf, __hypotf_finite) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot-power7.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot-power7.c new file mode 100644 index 0000000000..7380c3d395 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot-power7.c @@ -0,0 +1,19 @@ +/* __ieee_hypot() POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/fpu/multiarch/e_hypot-power7.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot-ppc64.c new file mode 100644 index 0000000000..9b0c1b0d22 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot-ppc64.c @@ -0,0 +1,26 @@ +/* __ieee_hypot() PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef strong_alias +#define strong_alias(a, b) + +#define __ieee754_hypot __ieee754_hypot_ppc64 + +#include <sysdeps/powerpc/fpu/e_hypot.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot.c new file mode 100644 index 0000000000..faaf55032a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypot.c @@ -0,0 +1,32 @@ +/* Multiple versions of ieee754_hypot. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__ieee754_hypot) __ieee754_hypot_ppc64 attribute_hidden; +extern __typeof (__ieee754_hypot) __ieee754_hypot_power7 attribute_hidden; + +libc_ifunc (__ieee754_hypot, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __ieee754_hypot_power7 + : __ieee754_hypot_ppc64); + +strong_alias (__ieee754_hypot, __hypot_finite) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypotf-power7.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypotf-power7.c new file mode 100644 index 0000000000..9ee841ba3d --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypotf-power7.c @@ -0,0 +1,19 @@ +/* __ieee_hypotf() POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/fpu/multiarch/e_hypotf-power7.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypotf-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypotf-ppc64.c new file mode 100644 index 0000000000..734cfe2836 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/e_hypotf-ppc64.c @@ -0,0 +1,26 @@ +/* __ieee_hypot() PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef strong_alias +#define strong_alias(a, b) + +#define __ieee754_hypotf __ieee754_hypotf_ppc64 + +#include <sysdeps/powerpc/fpu/e_hypotf.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S new file mode 100644 index 0000000000..cbfe96e614 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S @@ -0,0 +1,31 @@ +/* ceil function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __ceil __ceil_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_ceil.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S new file mode 100644 index 0000000000..4539c17c1f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S @@ -0,0 +1,31 @@ +/* ceil function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __ceil __ceil_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_ceil.S> diff --git a/sysdeps/powerpc/powerpc64/power4/fpu/w_sqrt.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil.c index 78bba57a28..587c0dc000 100644 --- a/sysdeps/powerpc/powerpc64/power4/fpu/w_sqrt.c +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil.c @@ -1,5 +1,5 @@ -/* Double-precision floating point square root wrapper. - Copyright (C) 2004-2013 Free Software Foundation, Inc. +/* Multiple versions of ceil. + Copyright (C) 2013 Free Software Foundation, Inc. This file is part of the GNU C Library. The GNU C Library is free software; you can redistribute it and/or @@ -16,40 +16,25 @@ License along with the GNU C Library; if not, see <http://www.gnu.org/licenses/>. */ -#include <math_ldbl_opt.h> #include <math.h> -#include <math_private.h> -#include <fenv_libc.h> - -double -__sqrt (double x) /* wrapper sqrt */ -{ - double z; -/* Power4 (ISA V2.0) and above implement sqrt in hardware. */ - __asm __volatile ( - " fsqrt %0,%1\n" - : "=f" (z) - : "f" (x)); -#ifdef _IEEE_LIBM - return z; -#else - if (__builtin_expect (_LIB_VERSION == _IEEE_, 0)) - return z; - - if (__builtin_expect (x != x, 0)) - return z; - - if (__builtin_expect (x < 0.0, 0)) - return __kernel_standard (x, x, 26); /* sqrt(negative) */ - else - return z; -#endif -} +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__ceil) __ceil_ppc64 attribute_hidden; +extern __typeof (__ceil) __ceil_power5plus attribute_hidden; + +libc_ifunc (__ceil, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __ceil_power5plus + : __ceil_ppc64); + +weak_alias (__ceil, ceil) -weak_alias (__sqrt, sqrt) #ifdef NO_LONG_DOUBLE - strong_alias (__sqrt, __sqrtl) weak_alias (__sqrt, sqrtl) +strong_alias (__ceil, __ceill) +weak_alias (__ceil, ceill) #endif #if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0) -compat_symbol (libm, __sqrt, sqrtl, GLIBC_2_0); +compat_symbol (libm, __ceil, ceill, GLIBC_2_0); #endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S new file mode 100644 index 0000000000..7790843f90 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S @@ -0,0 +1,26 @@ +/* ceilf function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __ceilf __ceilf_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_ceilf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S new file mode 100644 index 0000000000..e9036224ad --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S @@ -0,0 +1,26 @@ +/* ceilf function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __ceilf __ceilf_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_ceilf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf.c new file mode 100644 index 0000000000..a65174580b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf.c @@ -0,0 +1,32 @@ +/* Multiple versions of ceilf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__ceilf) __ceilf_ppc64 attribute_hidden; +extern __typeof (__ceilf) __ceilf_power5plus attribute_hidden; + +libc_ifunc (__ceilf, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __ceilf_power5plus + : __ceilf_ppc64); + +weak_alias (__ceilf, ceilf) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign-power6.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign-power6.S new file mode 100644 index 0000000000..595d318a43 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign-power6.S @@ -0,0 +1,33 @@ +/* copysign(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a, b, c, d) +#undef hidden_def +#define hidden_def(name) + +#define __copysign __copysign_power6 + +#include <sysdeps/powerpc/powerpc64/fpu/s_copysign.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign-ppc64.S new file mode 100644 index 0000000000..022ea06320 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign-ppc64.S @@ -0,0 +1,35 @@ +/* copysign(). PowerPC64 default version. + Copyright (C) 2010-2013 Free Software Foundation, Inc. + Contributed by Luis Machado <luisgpm@br.ibm.com>. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a, b, c, d) + +#define __copysign __copysign_ppc64 +#undef hidden_def +#define hidden_def(name) \ + strong_alias (__copysign_ppc64, __GI___copysign) + +#include <sysdeps/powerpc/powerpc64/fpu/s_copysign.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign.c new file mode 100644 index 0000000000..ee7ce0a886 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysign.c @@ -0,0 +1,51 @@ +/* Multiple versions of copysign. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Redefine copysign so that the compiler won't complain about the type + mismatch with the IFUNC selector in strong_alias below. */ +#undef __copysign +#define __copysign __redirect_copysign +#include <math.h> +#include <math_ldbl_opt.h> +#undef __copysign +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__redirect_copysign) __copysign_ppc64 attribute_hidden; +extern __typeof (__redirect_copysign) __copysign_power6 attribute_hidden; + +extern __typeof (__redirect_copysign) __libm_copysign; +libc_ifunc (__libm_copysign, + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __copysign_power6 + : __copysign_ppc64); + +strong_alias (__libm_copysign, __copysign) +weak_alias (__copysign, copysign) + +#ifdef NO_LONG_DOUBLE +weak_alias (__copysign,copysignl) +strong_alias(__copysign,__copysignl) +#endif +#ifdef IS_IN_libm +# if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0) +compat_symbol (libm, __copysign, copysignl, GLIBC_2_0); +# endif +#elif LONG_DOUBLE_COMPAT(libc, GLIBC_2_0) +compat_symbol (libc, __copysign, copysignl, GLIBC_2_0); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysignf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysignf.c new file mode 100644 index 0000000000..89c3c069a0 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_copysignf.c @@ -0,0 +1,32 @@ +/* Multiple versions of copysignf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <shlib-compat.h> +#include "init-arch.h" + +/* It's safe to use double-precision implementation for single-precision. */ +extern __typeof (__copysignf) __copysign_ppc64 attribute_hidden; +extern __typeof (__copysignf) __copysign_power6 attribute_hidden; + +libc_ifunc (__copysignf, + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __copysign_power6 + : __copysign_ppc64); + +weak_alias (__copysignf, copysignf) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-power7.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-power7.S new file mode 100644 index 0000000000..67f4f46429 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-power7.S @@ -0,0 +1,33 @@ +/* isnan(). PowerPC64/POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef hidden_def +#define hidden_def(name) +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(lib, name, symbol, ver) + +#define __finite __finite_power7 + +#include <sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-ppc64.c new file mode 100644 index 0000000000..b5848fa53f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-ppc64.c @@ -0,0 +1,34 @@ +/* finite(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + Contributed by Luis Machado <luisgpm@br.ibm.com>. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef weak_alias +#define weak_alias(a, b) +#undef strong_alias +#define strong_alias(a, b) + +#define FINITE __finite_ppc64 +#ifdef SHARED +# undef hidden_def +# define hidden_def(a) \ + __hidden_ver1 (__finite_ppc64, __GI___finite, __finite_ppc64); +#endif + +#include <sysdeps/ieee754/dbl-64/s_finite.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c new file mode 100644 index 0000000000..66cc437423 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c @@ -0,0 +1,51 @@ +/* Multiple versions of finite. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__finite) __finite_ppc64 attribute_hidden; +extern __typeof (__finite) __finite_power7 attribute_hidden; + +libc_ifunc (__finite, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __finite_power7 + : __finite_ppc64); + +weak_alias (__finite, finite) + +#ifdef NO_LONG_DOUBLE +strong_alias (__finite, __finitel) +weak_alias (__finite, finitel) +#endif + +#ifdef IS_IN_libm +# if LONG_DOUBLE_COMPAT (libm, GLIBC_2_0) +compat_symbol (libm, finite, finitel, GLIBC_2_0); +# endif +# if LONG_DOUBLE_COMPAT (libm, GLIBC_2_1) +compat_symbol (libm, __finite, __finitel, GLIBC_2_1); +# endif +#else +# if LONG_DOUBLE_COMPAT (libc, GLIBC_2_0) +compat_symbol (libc, __finite, __finitel, GLIBC_2_0); +compat_symbol (libc, finite, finitel, GLIBC_2_0); +# endif +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef-ppc64.c new file mode 100644 index 0000000000..1f2f8acff5 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef-ppc64.c @@ -0,0 +1,32 @@ +/* finitef(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + Contributed by Luis Machado <luisgpm@br.ibm.com>. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef weak_alias +#define weak_alias(a, b) + +#define FINITEF __finitef_ppc64 +#ifdef SHARED +# undef hidden_def +# define hidden_def(a) \ + __hidden_ver1 (__finitef_ppc64, __GI___finitef, __finitef_ppc64); +#endif + +#include <sysdeps/ieee754/flt-32/s_finitef.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef.c new file mode 100644 index 0000000000..25fe1a8705 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef.c @@ -0,0 +1,32 @@ +/* Multiple versions of finitef. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__finitef) __finitef_ppc64 attribute_hidden; +/* The double-precision version also works for single-precision. */ +extern __typeof (__finitef) __finite_power7 attribute_hidden; + +libc_ifunc (__finitef, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __finite_power7 + : __finitef_ppc64); + +weak_alias (__finitef, finitef) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor-power5+.S new file mode 100644 index 0000000000..fc329ca021 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor-power5+.S @@ -0,0 +1,31 @@ +/* floor function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __floor __floor_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_floor.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor-ppc64.S new file mode 100644 index 0000000000..50db6d8571 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor-ppc64.S @@ -0,0 +1,31 @@ +/* floor function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __floor __floor_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_floor.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor.c new file mode 100644 index 0000000000..12a1baee86 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floor.c @@ -0,0 +1,40 @@ +/* Multiple versions of floor. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__floor) __floor_ppc64 attribute_hidden; +extern __typeof (__floor) __floor_power5plus attribute_hidden; + +libc_ifunc (__floor, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __floor_power5plus + : __floor_ppc64); + +weak_alias (__floor, floor) + +#ifdef NO_LONG_DOUBLE +strong_alias (__floor, __floorl) +weak_alias (__floor, floorl) +#endif +#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0) +compat_symbol (libm, __floor, floorl, GLIBC_2_0); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf-power5+.S new file mode 100644 index 0000000000..1824bc13f5 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf-power5+.S @@ -0,0 +1,26 @@ +/* floorf function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __floorf __floorf_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_floorf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf-ppc64.S new file mode 100644 index 0000000000..7c8b99cfed --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf-ppc64.S @@ -0,0 +1,27 @@ +/* floorf function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __floorf __floorf_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_floorf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf.c new file mode 100644 index 0000000000..c7b753f8b0 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_floorf.c @@ -0,0 +1,32 @@ +/* Multiple versions of floorf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__floorf) __floorf_ppc64 attribute_hidden; +extern __typeof (__floorf) __floorf_power5plus attribute_hidden; + +libc_ifunc (__floorf, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __floorf_power5plus + : __floorf_ppc64); + +weak_alias (__floorf, floorf) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-power7.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-power7.S new file mode 100644 index 0000000000..e4e9430b00 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-power7.S @@ -0,0 +1,33 @@ +/* isinf(). PowerPC64/POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef hidden_def +#define hidden_def(name) +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(lib, name, alias, ver) + +#define __isinf __isinf_power7 + +#include <sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-ppc64.c new file mode 100644 index 0000000000..d1691286a5 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-ppc64.c @@ -0,0 +1,33 @@ +/* isinf(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef weak_alias +#define weak_alias(a, b) +#undef strong_alias +#define strong_alias(a, b) + +#define __isinf __isinf_ppc64 +#ifdef SHARED +# undef hidden_def +# define hidden_def(a) \ + __hidden_ver1 (__isinf_ppc64, __GI___isinf, __isinf_ppc64); +#endif + +#include <sysdeps/ieee754/dbl-64/s_isinf.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c new file mode 100644 index 0000000000..994f9773c4 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c @@ -0,0 +1,44 @@ +/* Multiple versions of isinf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__isinf) __isinf_ppc64 attribute_hidden; +extern __typeof (__isinf) __isinf_power7 attribute_hidden; + +libc_ifunc (__isinf, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __isinf_power7 + : __isinf_ppc64); + +weak_alias (__isinf, isinf) + +#ifdef NO_LONG_DOUBLE +strong_alias (__isinf, __isinfl) +weak_alias (__isinf, isinfl) +#endif + +#ifndef IS_IN_libm +# if LONG_DOUBLE_COMPAT (libc, GLIBC_2_0) +compat_symbol (libc, __isinf, __isinfl, GLIBC_2_0); +compat_symbol (libc, isinf, isinfl, GLIBC_2_0); +# endif +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff-ppc64.c new file mode 100644 index 0000000000..65a28924d6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff-ppc64.c @@ -0,0 +1,31 @@ +/* isinff(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef weak_alias +#define weak_alias(a, b) + +#define __isinff __isinff_ppc64 +#ifdef SHARED +# undef hidden_def +# define hidden_def(a) \ + __hidden_ver1 (__isinff_ppc64, __GI___isinff, __isinff_ppc64); +#endif + +#include <sysdeps/ieee754/flt-32/s_isinff.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff.c new file mode 100644 index 0000000000..44961f29ec --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff.c @@ -0,0 +1,33 @@ +/* Multiple versions of isinf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__isinff) __isinff_ppc64 attribute_hidden; +/* The double-precision version also works for single-precision. */ +extern __typeof (__isinff) __isinf_power7 attribute_hidden; + +libc_ifunc (__isinff, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __isinf_power7 + : __isinff_ppc64); + +weak_alias (__isinff, isinff) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power5.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power5.S new file mode 100644 index 0000000000..757e580fb3 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power5.S @@ -0,0 +1,33 @@ +/* isnan(). PowerPC64/POWER5 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef hidden_def +#define hidden_def(name) +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(lib, name, symbol, ver) + +#define __isnan __isnan_power5 + +#include <sysdeps/powerpc/powerpc64/power5/fpu/s_isnan.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6.S new file mode 100644 index 0000000000..e0d8010276 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6.S @@ -0,0 +1,33 @@ +/* isnan(). PowerPC64/POWER6 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef hidden_def +#define hidden_def(name) +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(lib, name, symbol, ver) + +#define __isnan __isnan_power6 + +#include <sysdeps/powerpc/powerpc64/power6/fpu/s_isnan.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6x.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6x.S new file mode 100644 index 0000000000..84ce8e5548 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6x.S @@ -0,0 +1,33 @@ +/* isnan(). PowerPC64/POWER6X version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef hidden_def +#define hidden_def(name) +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(lib, name, symbol, ver) + +#define __isnan __isnan_power6x + +#include <sysdeps/powerpc/powerpc64/power6x/fpu/s_isnan.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power7.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power7.S new file mode 100644 index 0000000000..f3f5a67fef --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power7.S @@ -0,0 +1,33 @@ +/* isnan(). PowerPC64/POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef hidden_def +#define hidden_def(name) +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(lib, name, symbol, ver) + +#define __isnan __isnan_power7 + +#include <sysdeps/powerpc/powerpc64/power7/fpu/s_isnan.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-ppc64.S new file mode 100644 index 0000000000..5ce6f52cf5 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-ppc64.S @@ -0,0 +1,32 @@ +/* isnan(). PowerPC32 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) + +#define __isnan __isnan_ppc64 +#undef hidden_def +#define hidden_def(name) \ + .globl __GI___isnan ; .set __GI___isnan,__isnan_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_isnan.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c new file mode 100644 index 0000000000..70353df91a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c @@ -0,0 +1,53 @@ +/* Multiple versions of isnan. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__isnan) __isnan_ppc64 attribute_hidden; +extern __typeof (__isnan) __isnan_power5 attribute_hidden; +extern __typeof (__isnan) __isnan_power6 attribute_hidden; +extern __typeof (__isnan) __isnan_power6x attribute_hidden; +extern __typeof (__isnan) __isnan_power7 attribute_hidden; + +libc_ifunc (__isnan, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __isnan_power7 : + (hwcap & PPC_FEATURE_POWER6_EXT) + ? __isnan_power6x : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __isnan_power6 : + (hwcap & PPC_FEATURE_POWER5) + ? __isnan_power5 + : __isnan_ppc64); + +weak_alias (__isnan, isnan) + +#ifdef NO_LONG_DOUBLE +strong_alias (__isnan, __isnanl) +weak_alias (__isnan, isnanl) +#endif + +#ifndef IS_IN_libm +# if LONG_DOUBLE_COMPAT(libc, GLIBC_2_0) +compat_symbol (libc, __isnan, __isnanl, GLIBC_2_0); +compat_symbol (libc, isnan, isnanl, GLIBC_2_0); +# endif +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c new file mode 100644 index 0000000000..3e80b69b6c --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c @@ -0,0 +1,40 @@ +/* Multiple versions of isnan. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include "init-arch.h" + +/* The double-precision implementation also works for the single one. */ +extern __typeof (__isnanf) __isnan_ppc64 attribute_hidden; +extern __typeof (__isnanf) __isnan_power5 attribute_hidden; +extern __typeof (__isnanf) __isnan_power6 attribute_hidden; +extern __typeof (__isnanf) __isnan_power6x attribute_hidden; +extern __typeof (__isnanf) __isnan_power7 attribute_hidden; + +libc_ifunc (__isnanf, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __isnan_power7 : + (hwcap & PPC_FEATURE_POWER6_EXT) + ? __isnan_power6x : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __isnan_power6 : + (hwcap & PPC_FEATURE_POWER5) + ? __isnan_power5 + : __isnan_ppc64); + +weak_alias (__isnanf, isnanf) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-power6x.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-power6x.S new file mode 100644 index 0000000000..36b1546292 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-power6x.S @@ -0,0 +1,31 @@ +/* Round double to long int. PowerPC64/POWER6X default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __llrint __llrint_power6x + +#include <sysdeps/powerpc/powerpc64/power6x/fpu/s_llrint.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-ppc64.S new file mode 100644 index 0000000000..22fbe60592 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-ppc64.S @@ -0,0 +1,31 @@ +/* Round double to long int. PowerPC32 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __llrint __llrint_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_llrint.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint.c new file mode 100644 index 0000000000..dfc6ddfbdb --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint.c @@ -0,0 +1,57 @@ +/* Multiple versions of llrint. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Redefine lrint/__lrint so that the compiler won't complain about the type + mismatch with the IFUNC selector in strong_alias below. */ +#define lrint __hidden_lrint +#define __lrint __hidden___lrint + +#include <math.h> +#include <math_ldbl_opt.h> +#undef lrint +#undef __lrint +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__llrint) __llrint_ppc64 attribute_hidden; +extern __typeof (__llrint) __llrint_power6x attribute_hidden; + +libc_ifunc (__llrint, + (hwcap & PPC_FEATURE_POWER6_EXT) + ? __llrint_power6x + : __llrint_ppc64); + +weak_alias (__llrint, llrint) +#ifdef NO_LONG_DOUBLE +strong_alias (__llrint, __llrintl) +weak_alias (__llrint, llrintl) +#endif +#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1) +compat_symbol (libm, __llrint, llrintl, GLIBC_2_1); +#endif + +/* long has the same width as long long on PowerPC64. */ +strong_alias (__llrint, __lrint) +weak_alias (__lrint, lrint) +#ifdef NO_LONG_DOUBLE +strong_alias (__lrint, __lrintl) +weak_alias (__lrint, lrintl) +#endif +#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1) +compat_symbol (libm, __lrint, lrintl, GLIBC_2_1); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-power5+.S new file mode 100644 index 0000000000..01d86e700c --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-power5+.S @@ -0,0 +1,32 @@ +/* llround(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __llround __llround_power5plus +#define __lround __lround_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_llround.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-power6x.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-power6x.S new file mode 100644 index 0000000000..22fa2cd767 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-power6x.S @@ -0,0 +1,32 @@ +/* llround(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(name, alias) +#undef strong_alias +#define strong_alias(name, alias) +#undef compat_symbol +#define compat_symbol(lib, name, alias, ver) + +#define __llround __llround_power6x +#define __lround __lround_power6x + +#include <sysdeps/powerpc/powerpc64/power6x/fpu/s_llround.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-ppc64.S new file mode 100644 index 0000000000..38f1eed3d3 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround-ppc64.S @@ -0,0 +1,28 @@ +/* llround(). PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __llround __llround_ppc64 +#define __lround __lround_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_llround.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround.c new file mode 100644 index 0000000000..3559ce4934 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_llround.c @@ -0,0 +1,60 @@ +/* Multiple versions of llround. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#define lround __hidden_lround +#define __lround __hidden___lround + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__llround) __llround_ppc64 attribute_hidden; +extern __typeof (__llround) __llround_power5plus attribute_hidden; +extern __typeof (__llround) __llround_power6x attribute_hidden; + +libc_ifunc (__llround, + (hwcap & PPC_FEATURE_POWER6_EXT) + ? __llround_power6x : + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __llround_power5plus + : __llround_ppc64); + +weak_alias (__llround, llround) + +#ifdef NO_LONG_DOUBLE +weak_alias (__llround, llroundl) +strong_alias (__llround, __llroundl) +#endif +#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1) +compat_symbol (libm, __llround, llroundl, GLIBC_2_1); +compat_symbol (libm, llround, lroundl, GLIBC_2_1); +#endif + +/* long has the same width as long long on PPC64. */ +#undef lround +#undef __lround +strong_alias (__llround, __lround) +weak_alias (__llround, lround) +#ifdef NO_LONG_DOUBLE +strong_alias (__llround, __llroundl) +weak_alias (__llround, llroundl) +#endif +#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1) +compat_symbol (libm, __lround, lroundl, GLIBC_2_1); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb-power7.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb-power7.c new file mode 100644 index 0000000000..6a50eb43ce --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb-power7.c @@ -0,0 +1,19 @@ +/* logb(). PowerPC64/POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_logb-power7.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb-ppc64.c new file mode 100644 index 0000000000..6e42fcb03b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb-ppc64.c @@ -0,0 +1,28 @@ +/* logb(). PowerPC32/POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef weak_alias +#define weak_alias(a, b) +#undef strong_alias +#define strong_alias(a, b) + +#define __logb __logb_ppc64 + +#include <sysdeps/ieee754/dbl-64/s_logb.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb.c new file mode 100644 index 0000000000..e14bfa2ca3 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logb.c @@ -0,0 +1,41 @@ +/* Multiple versions of logb. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__logb) __logb_ppc64 attribute_hidden; +extern __typeof (__logb) __logb_power7 attribute_hidden; + +libc_ifunc (__logb, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __logb_power7 + : __logb_ppc64); + +weak_alias (__logb, logb) + +#ifdef NO_LONG_DOUBLE +strong_alias (__logb, __logbl) +weak_alias (__logb, logbl) +#endif + +#if LONG_DOUBLE_COMPAT (libm, GLIBC_2_0) +compat_symbol (libm, logb, logbl, GLIBC_2_0); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf-power7.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf-power7.c new file mode 100644 index 0000000000..cfd2902563 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf-power7.c @@ -0,0 +1,19 @@ +/* logb(). PowerPC64/POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_logbf-power7.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf-ppc64.c new file mode 100644 index 0000000000..8a699d1059 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf-ppc64.c @@ -0,0 +1,26 @@ +/* logbf(). PowerPC64 default implementation. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef weak_alias +#define weak_alias(a, b) + +#define __logbf __logbf_ppc64 + +#include <sysdeps/ieee754/flt-32/s_logbf.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf.c new file mode 100644 index 0000000000..a300125dc2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbf.c @@ -0,0 +1,32 @@ +/* Multiple versions of logbf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__logbf) __logbf_ppc64 attribute_hidden; +extern __typeof (__logbf) __logbf_power7 attribute_hidden; + +libc_ifunc (__logbf, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __logbf_power7 + : __logbf_ppc64); + +weak_alias (__logbf, logbf) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl-power7.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl-power7.c new file mode 100644 index 0000000000..4fe43638aa --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl-power7.c @@ -0,0 +1,19 @@ +/* logb(). PowerPC64/POWER7 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_logbl-power7.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl-ppc64.c new file mode 100644 index 0000000000..8bb0f18308 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl-ppc64.c @@ -0,0 +1,21 @@ +/* logbl(). PowerPC64/POWER7 version. + Copyright (C) 2012-2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#define __logbl __logbl_ppc64 + +#include <sysdeps/ieee754/ldbl-128ibm/s_logbl.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl.c new file mode 100644 index 0000000000..21af5711f2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_logbl.c @@ -0,0 +1,32 @@ +/* Multiple versions of logbl. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__logbl) __logbl_ppc64 attribute_hidden; +extern __typeof (__logbl) __logbl_power7 attribute_hidden; + +libc_ifunc (__logbl, + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __logbl_power7 + : __logbl_ppc64); + +long_double_symbol (libm, __logbl, logbl); diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_lrint.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_lrint.c new file mode 100644 index 0000000000..d09286267b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_lrint.c @@ -0,0 +1 @@ + /* __lrint is in s_llrint.c */ diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_lround.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_lround.c new file mode 100644 index 0000000000..0dab5443e2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_lround.c @@ -0,0 +1 @@ +/* __lround is in s_llround.c */ diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf-power5+.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf-power5+.c new file mode 100644 index 0000000000..04ec0eb686 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf-power5+.c @@ -0,0 +1,19 @@ +/* PowerPC/POWER5+ implementation for modf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_modf-power5+.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf-ppc64.c new file mode 100644 index 0000000000..30216872cd --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf-ppc64.c @@ -0,0 +1,29 @@ +/* PowerPC64 default implementation for modf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) + +#define __modf __modf_ppc64 + +#include <sysdeps/ieee754/dbl-64/s_modf.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf.c new file mode 100644 index 0000000000..931d0c375a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modf.c @@ -0,0 +1,44 @@ +/* Multiple versions of modf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__modf) __modf_ppc64 attribute_hidden; +extern __typeof (__modf) __modf_power5plus attribute_hidden; + +libc_ifunc (__modf, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __modf_power5plus + : __modf_ppc64); + +weak_alias (__modf, modf) + +#ifdef NO_LONG_DOUBLE +strong_alias (__modf, __modfl) +weak_alias (__modf, modfl) +#endif +#ifdef IS_IN_libm +# if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0) +compat_symbol (libm, __modf, modfl, GLIBC_2_0); +# endif +#elif LONG_DOUBLE_COMPAT(libc, GLIBC_2_0) +compat_symbol (libc, __modf, modfl, GLIBC_2_0); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff-power5+.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff-power5+.c new file mode 100644 index 0000000000..b3d7a0ad60 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff-power5+.c @@ -0,0 +1,19 @@ +/* PowerPC/POWER5+ implementation for modff. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_modff-power5+.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff-ppc64.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff-ppc64.c new file mode 100644 index 0000000000..3a98d0958f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff-ppc64.c @@ -0,0 +1,26 @@ +/* PowerPC64 default implementation for modff. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __modff __modff_ppc64 + +#include <sysdeps/ieee754/flt-32/s_modff.c> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff.c new file mode 100644 index 0000000000..0629339b39 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_modff.c @@ -0,0 +1,30 @@ +/* Multiple versions of modff. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include "init-arch.h" + +extern __typeof (__modff) __modff_ppc64 attribute_hidden; +extern __typeof (__modff) __modff_power5plus attribute_hidden; + +libc_ifunc (__modff, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __modff_power5plus + : __modff_ppc64); + +weak_alias (__modff, modff) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round-power5+.S new file mode 100644 index 0000000000..e30568db49 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round-power5+.S @@ -0,0 +1,31 @@ +/* round function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __round __round_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_round.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round-ppc64.S new file mode 100644 index 0000000000..ed208ddec6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round-ppc64.S @@ -0,0 +1,31 @@ +/* round function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __round __round_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_round.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round.c new file mode 100644 index 0000000000..cfeac1f661 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_round.c @@ -0,0 +1,40 @@ +/* Multiple versions of round. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__round) __round_ppc64 attribute_hidden; +extern __typeof (__round) __round_power5plus attribute_hidden; + +libc_ifunc (__round, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __round_power5plus + : __round_ppc64); + +weak_alias (__round, round) + +#ifdef NO_LONG_DOUBLE +strong_alias (__round, __roundl) +weak_alias (__round, roundl) +#endif +#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0) +compat_symbol (libm, __round, roundl, GLIBC_2_0); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf-power5+.S new file mode 100644 index 0000000000..8c1684ca64 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf-power5+.S @@ -0,0 +1,26 @@ +/* roundf function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __roundf __roundf_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_roundf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf-ppc64.S new file mode 100644 index 0000000000..a9d7981232 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf-ppc64.S @@ -0,0 +1,26 @@ +/* roundf function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __roundf __roundf_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_roundf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf.c new file mode 100644 index 0000000000..6377354365 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_roundf.c @@ -0,0 +1,32 @@ +/* Multiple versions of roundf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__roundf) __roundf_ppc64 attribute_hidden; +extern __typeof (__roundf) __roundf_power5plus attribute_hidden; + +libc_ifunc (__roundf, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __roundf_power5plus + : __roundf_ppc64); + +weak_alias (__roundf, roundf) diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc-power5+.S new file mode 100644 index 0000000000..27519649f8 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc-power5+.S @@ -0,0 +1,31 @@ +/* trunc function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __trunc __trunc_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_trunc.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc-ppc64.S new file mode 100644 index 0000000000..5282609a8f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc-ppc64.S @@ -0,0 +1,31 @@ +/* trunc function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include <math_ldbl_opt.h> + +#undef weak_alias +#define weak_alias(a,b) +#undef strong_alias +#define strong_alias(a,b) +#undef compat_symbol +#define compat_symbol(a,b,c,d) + +#define __trunc __trunc_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_trunc.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc.c new file mode 100644 index 0000000000..a42687ce43 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_trunc.c @@ -0,0 +1,40 @@ +/* Multiple versions of trunc. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__trunc) __trunc_ppc64 attribute_hidden; +extern __typeof (__trunc) __trunc_power5plus attribute_hidden; + +libc_ifunc (__trunc, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __trunc_power5plus + : __trunc_ppc64); + +weak_alias (__trunc, trunc) + +#ifdef NO_LONG_DOUBLE +strong_alias (__trunc, __truncl) +weak_alias (__trunc, truncl) +#endif +#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_0) +compat_symbol (libm, __trunc, truncl, GLIBC_2_0); +#endif diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf-power5+.S new file mode 100644 index 0000000000..bb1008fa10 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf-power5+.S @@ -0,0 +1,26 @@ +/* truncf function. PowerPC64/power5+ version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __truncf __truncf_power5plus + +#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_truncf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf-ppc64.S new file mode 100644 index 0000000000..6358c61860 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf-ppc64.S @@ -0,0 +1,26 @@ +/* truncf function. PowerPC64 default version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef weak_alias +#define weak_alias(a,b) + +#define __truncf __truncf_ppc64 + +#include <sysdeps/powerpc/powerpc64/fpu/s_truncf.S> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf.c new file mode 100644 index 0000000000..4602583606 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_truncf.c @@ -0,0 +1,32 @@ +/* Multiple versions of truncf. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <math.h> +#include <math_ldbl_opt.h> +#include <shlib-compat.h> +#include "init-arch.h" + +extern __typeof (__truncf) __truncf_ppc64 attribute_hidden; +extern __typeof (__truncf) __truncf_power5plus attribute_hidden; + +libc_ifunc (__truncf, + (hwcap & PPC_FEATURE_POWER5_PLUS) + ? __truncf_power5plus + : __truncf_ppc64); + +weak_alias (__truncf, truncf) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S index 801af5d0aa..45f71d7ac5 100644 --- a/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S +++ b/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S @@ -19,8 +19,10 @@ #include <sysdep.h> .section ".toc","aw" + .p2align 3 .LC0: /* 2**23 */ - .tc FD_4b000000_0[TC],0x4b00000000000000 + .long 0x4b000000 + .long 0x0 .section ".text" EALIGN (__ceilf, 4, 0) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_floorf.S b/sysdeps/powerpc/powerpc64/fpu/s_floorf.S index a0a22e7eb9..e85b820b21 100644 --- a/sysdeps/powerpc/powerpc64/fpu/s_floorf.S +++ b/sysdeps/powerpc/powerpc64/fpu/s_floorf.S @@ -19,8 +19,10 @@ #include <sysdep.h> .section ".toc","aw" + .p2align 3 .LC0: /* 2**23 */ - .tc FD_4b000000_0[TC],0x4b00000000000000 + .long 0x4b000000 + .long 0x0 .section ".text" EALIGN (__floorf, 4, 0) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_nearbyintf.S b/sysdeps/powerpc/powerpc64/fpu/s_nearbyintf.S index 876707c766..b1a2b8cd65 100644 --- a/sysdeps/powerpc/powerpc64/fpu/s_nearbyintf.S +++ b/sysdeps/powerpc/powerpc64/fpu/s_nearbyintf.S @@ -26,8 +26,10 @@ /* float [fp1] nearbyintf(float [fp1]) */ .section ".toc","aw" + .p2align 3 .LC0: /* 2**23 */ - .tc FD_4b000000_0[TC],0x4b00000000000000 + .long 0x4b000000 + .long 0x0 .section ".text" EALIGN (__nearbyintf, 4, 0) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_rintf.S b/sysdeps/powerpc/powerpc64/fpu/s_rintf.S index cb28ec748d..1887717420 100644 --- a/sysdeps/powerpc/powerpc64/fpu/s_rintf.S +++ b/sysdeps/powerpc/powerpc64/fpu/s_rintf.S @@ -19,8 +19,10 @@ #include <sysdep.h> .section ".toc","aw" + .p2align 3 .LC0: /* 2**23 */ - .tc FD_4b000000_0[TC],0x4b00000000000000 + .long 0x4b000000 + .long 0x0 .section ".text" EALIGN (__rintf, 4, 0) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_roundf.S b/sysdeps/powerpc/powerpc64/fpu/s_roundf.S index 980a77bde0..4f2c851631 100644 --- a/sysdeps/powerpc/powerpc64/fpu/s_roundf.S +++ b/sysdeps/powerpc/powerpc64/fpu/s_roundf.S @@ -19,10 +19,12 @@ #include <sysdep.h> .section ".toc","aw" + .p2align 3 .LC0: /* 2**23 */ - .tc FD_4b000000_0[TC],0x4b00000000000000 + .long 0x4b000000 .LC1: /* 0.5 */ - .tc FD_3f000000_0[TC],0x3f00000000000000 + .long 0x3f000000 + .section ".text" /* float [fp1] roundf (float x [fp1]) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_truncf.S b/sysdeps/powerpc/powerpc64/fpu/s_truncf.S index 5ea5f3d04a..b8fd050319 100644 --- a/sysdeps/powerpc/powerpc64/fpu/s_truncf.S +++ b/sysdeps/powerpc/powerpc64/fpu/s_truncf.S @@ -19,8 +19,10 @@ #include <sysdep.h> .section ".toc","aw" + .p2align 3 .LC0: /* 2**23 */ - .tc FD_4b000000_0[TC],0x4b00000000000000 + .long 0x4b000000 + .long 0x0 .section ".text" /* float [fp1] truncf (float x [fp1]) diff --git a/sysdeps/powerpc/powerpc64/lshift.S b/sysdeps/powerpc/powerpc64/lshift.S new file mode 100644 index 0000000000..a997451c45 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/lshift.S @@ -0,0 +1,177 @@ +/* PowerPC64 mpn_lshift -- rp[] = up[] << cnt + Copyright (C) 2003-2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#define RP r3 +#define UP r4 +#define N r5 +#define CNT r6 + +#define TNC r0 +#define U0 r30 +#define U1 r31 +#define RETVAL r5 + +EALIGN(__mpn_lshift, 5, 0) + std U1, -8(r1) + std U0, -16(r1) + subfic TNC, CNT, 64 + sldi r7, N, RP + add UP, UP, r7 + add RP, RP, r7 + rldicl. U0, N, 0, 62 + cmpdi CNT, U0, 2 + addi U1, N, RP + ld r10, -8(UP) + srd RETVAL, r10, TNC + + srdi U1, U1, 2 + mtctr U1 + beq cr0, L(b00) + blt cr6, L(b01) + ld r11, -16(UP) + beq cr6, L(b10) + + .align 4 +L(b11): sld r8, r10, CNT + srd r9, r11, TNC + ld U1, -24(UP) + addi UP, UP, -24 + sld r12, r11, CNT + srd r7, U1, TNC + addi RP, RP, 16 + bdnz L(gt3) + + or r11, r8, r9 + sld r8, U1, CNT + b L(cj3) + + .align 4 +L(gt3): ld U0, -8(UP) + or r11, r8, r9 + sld r8, U1, CNT + srd r9, U0, TNC + ld U1, -16(UP) + or r10, r12, r7 + b L(L11) + + .align 5 +L(b10): sld r12, r10, CNT + addi RP, RP, 24 + srd r7, r11, TNC + bdnz L(gt2) + + sld r8, r11, CNT + or r10, r12, r7 + b L(cj2) + +L(gt2): ld U0, -24(UP) + sld r8, r11, CNT + srd r9, U0, TNC + ld U1, -32(UP) + or r10, r12, r7 + sld r12, U0, CNT + srd r7, U1, 0 + ld U0, -40(UP) + or r11, r8, r9 + addi UP, UP, -16 + b L(L10) + + .align 4 +L(b00): ld U1, -16(UP) + sld r12, r10, CNT + srd r7, U1, TNC + ld U0, -24(UP) + sld r8, U1, CNT + srd r9, U0, TNC + ld U1, -32(UP) + or r10, r12, r7 + sld r12, U0, CNT + srd r7, U1, TNC + addi RP, RP, r8 + bdz L(cj4) + +L(gt4): addi UP, UP, -32 + ld U0, -8(UP) + or r11, r8, r9 + b L(L00) + + .align 4 +L(b01): bdnz L(gt1) + sld r8, r10, CNT + std r8, -8(RP) + b L(ret) + +L(gt1): ld U0, -16(UP) + sld r8, r10, CNT + srd r9, U0, TNC + ld U1, -24(UP) + sld r12, U0, CNT + srd r7, U1, TNC + ld U0, -32(UP) + or r11, r8, r9 + sld r8, U1, CNT + srd r9, U0, TNC + ld U1, -40(UP) + addi UP, UP, -40 + or r10, r12, r7 + bdz L(end) + + .align 5 +L(top): sld r12, U0, CNT + srd r7, U1, TNC + ld U0, -8(UP) + std r11, -8(RP) + or r11, r8, r9 +L(L00): sld r8, U1, CNT + srd r9, U0, TNC + ld U1, -16(UP) + std r10, -16(RP) + or r10, r12, r7 +L(L11): sld r12, U0, CNT + srd r7, U1, TNC + ld U0, -24(UP) + std r11, -24(RP) + or r11, r8, r9 +L(L10): sld r8, U1, CNT + srd r9, U0, TNC + ld U1, -32(UP) + addi UP, UP, -32 + std r10, -32(RP) + addi RP, RP, -32 + or r10, r12, r7 + bdnz L(top) + + .align 5 +L(end): sld r12, U0, CNT + srd r7, U1, TNC + std r11, -8(RP) +L(cj4): or r11, r8, r9 + sld r8, U1, CNT + std r10, -16(RP) +L(cj3): or r10, r12, r7 + std r11, -24(RP) +L(cj2): std r10, -32(RP) + std r8, -40(RP) + +L(ret): ld U1, -8(r1) + ld U0, -16(r1) + mr RP, RETVAL + blr +END(__mpn_lshift) diff --git a/sysdeps/powerpc/powerpc64/memcpy.S b/sysdeps/powerpc/powerpc64/memcpy.S index b8c4cc8b10..5fc7401c99 100644 --- a/sysdeps/powerpc/powerpc64/memcpy.S +++ b/sysdeps/powerpc/powerpc64/memcpy.S @@ -212,15 +212,28 @@ EALIGN (memcpy, 5, 0) blt cr6,5f srdi 7,6,16 bgt cr6,3f +#ifdef __LITTLE_ENDIAN__ + sth 7,0(3) +#else sth 6,0(3) +#endif b 7f .align 4 3: +#ifdef __LITTLE_ENDIAN__ + rotlwi 6,6,24 + stb 6,0(3) + sth 7,1(3) +#else stb 7,0(3) sth 6,1(3) +#endif b 7f .align 4 5: +#ifdef __LITTLE_ENDIAN__ + rotlwi 6,6,8 +#endif stb 6,0(3) 7: cmpldi cr1,10,16 @@ -328,7 +341,11 @@ EALIGN (memcpy, 5, 0) ld 7,8(5) subfic 9,10,64 beq 2f +#ifdef __LITTLE_ENDIAN__ + srd 0,6,10 +#else sld 0,6,10 +#endif cmpldi 11,1 mr 6,7 addi 4,4,-8 @@ -336,15 +353,25 @@ EALIGN (memcpy, 5, 0) b 1f 2: addi 5,5,8 .align 4 +#ifdef __LITTLE_ENDIAN__ +0: srd 0,6,10 + sld 8,7,9 +#else 0: sld 0,6,10 srd 8,7,9 +#endif cmpldi 11,2 ld 6,8(5) or 0,0,8 addi 11,11,-2 std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srd 0,7,10 +1: sld 8,6,9 +#else sld 0,7,10 1: srd 8,6,9 +#endif or 0,0,8 beq 8f ld 7,16(5) diff --git a/sysdeps/powerpc/powerpc64/memset.S b/sysdeps/powerpc/powerpc64/memset.S index 6acf149c8a..22c7d4e29d 100644 --- a/sysdeps/powerpc/powerpc64/memset.S +++ b/sysdeps/powerpc/powerpc64/memset.S @@ -55,14 +55,14 @@ L(_memset): /* Align to doubleword boundary. */ cmpldi cr5, rLEN, 31 - rlwimi rCHR, rCHR, 8, 16, 23 /* Replicate byte to halfword. */ + insrdi rCHR, rCHR, 8, 48 /* Replicate byte to halfword. */ beq+ L(aligned2) mtcrf 0x01, rMEMP0 subfic rALIGN, rALIGN, 8 cror 28,30,31 /* Detect odd word aligned. */ add rMEMP, rMEMP, rALIGN sub rLEN, rLEN, rALIGN - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ bt 29, L(g4) /* Process the even word of doubleword. */ bf+ 31, L(g2) @@ -84,14 +84,14 @@ L(g0): /* Handle the case of size < 31. */ L(aligned2): - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ L(aligned): mtcrf 0x01, rLEN ble cr5, L(medium) /* Align to 32-byte boundary. */ andi. rALIGN, rMEMP, 0x18 subfic rALIGN, rALIGN, 0x20 - insrdi rCHR,rCHR,32,0 /* Replicate word to double word. */ + insrdi rCHR, rCHR, 32, 0 /* Replicate word to double word. */ beq L(caligned) mtcrf 0x01, rALIGN add rMEMP, rMEMP, rALIGN @@ -212,7 +212,7 @@ L(le4): /* Memset of 0-31 bytes. */ .align 5 L(medium): - insrdi rCHR,rCHR,32,0 /* Replicate word to double word. */ + insrdi rCHR, rCHR, 32, 0 /* Replicate word to double word. */ cmpldi cr1, rLEN, 16 L(medium_tail2): add rMEMP, rMEMP, rLEN @@ -247,6 +247,7 @@ L(medium_28t): END_GEN_TB (memset,TB_TOCLESS) libc_hidden_builtin_def (memset) +#ifndef NO_BZERO_IMPL /* Copied from bzero.S to prevent the linker from inserting a stub between bzero and memset. */ ENTRY (__bzero) @@ -257,3 +258,4 @@ ENTRY (__bzero) END_GEN_TB (__bzero,TB_TOCLESS) weak_alias (__bzero, bzero) +#endif diff --git a/sysdeps/powerpc/powerpc64/mul_1.S b/sysdeps/powerpc/powerpc64/mul_1.S new file mode 100644 index 0000000000..68a1646462 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/mul_1.S @@ -0,0 +1,135 @@ +/* PowerPC64 __mpn_mul_1 -- Multiply a limb vector with a limb and store + the result in a second limb vector. + Copyright (C) 1999-2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#define RP r3 +#define UP r4 +#define N r5 +#define VL r6 + +EALIGN(__mpn_mul_1, 5, 0) + std r27, -40(r1) + std r26, -48(r1) + li r12, 0 + ld r26, 0(UP) + + rldicl. r0, N, 0, 62 + cmpdi VL, r0, 2 + addic N, N, RP + srdi N, N, 2 + mtctr N + beq cr0, L(b00) + blt cr6, L(b01) + beq cr6, L(b10) + +L(b11): mr cr7, r12 + mulld cr0, r26, VL + mulhdu r12, r26, VL + addi UP, UP, 8 + addc r0, r0, r7 + std r0, 0(RP) + addi RP, RP, 8 + b L(fic) + +L(b00): ld r27, r8(UP) + addi UP, UP, 16 + mulld r0, r26, VL + mulhdu N, r26, VL + mulld r7, r27, VL + mulhdu r8, r27, VL + addc r0, r0, r12 + adde r7, r7, N + addze r12, r8 + std r0, 0(RP) + std r7, 8(RP) + addi RP, RP, 16 + b L(fic) + + nop +L(b01): bdnz L(gt1) + mulld r0, r26, VL + mulhdu r8, r26, VL + addc r0, r0, r12 + std r0, 0(RP) + b L(ret) +L(gt1): ld r27, 8(UP) + nop + mulld r0, r26, VL + mulhdu N, r26, VL + ld r26, 16(UP) + mulld r7, r27, VL + mulhdu r8, r27, VL + mulld r9, r26, VL + mulhdu r10, r26, VL + addc r0, r0, r12 + adde r7, r7, N + adde r9, r9, r8 + addze r12, r10 + std r0, 0(RP) + std r7, 8(RP) + std r9, 16(RP) + addi UP, UP, 24 + addi RP, RP, 24 + b L(fic) + + nop +L(fic): ld r26, 0(UP) +L(b10): ld r27, 8(UP) + addi UP, UP, 16 + bdz L(end) + +L(top): mulld r0, r26, VL + mulhdu N, r26, VL + mulld r7, r27, VL + mulhdu r8, r27, VL + ld r26, 0(UP) + ld r27, 8(UP) + adde r0, r0, r12 + adde r7, r7, N + mulld r9, r26, VL + mulhdu r10, r26, VL + mulld r11, r27, VL + mulhdu r12, r27, VL + ld r26, 16(UP) + ld r27, 24(UP) + std r0, 0(RP) + adde r9, r9, r8 + std r7, 8(RP) + adde r11, r11, r10 + std r9, 16(RP) + addi UP, UP, 32 + std r11, 24(RP) + + addi RP, RP, 32 + bdnz L(top) + +L(end): mulld r0, r26, VL + mulhdu N, r26, VL + mulld r7, r27, VL + mulhdu r8, r27, VL + adde r0, r0, r12 + adde r7, r7, N + std r0, 0(RP) + std r7, 8(RP) +L(ret): addze RP, r8 + ld r27, -40(r1) + ld r26, -48(r1) + blr +END(__mpn_mul_1) diff --git a/sysdeps/powerpc/powerpc64/multiarch/Makefile b/sysdeps/powerpc/powerpc64/multiarch/Makefile new file mode 100644 index 0000000000..3c47316bda --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/Makefile @@ -0,0 +1,28 @@ +ifeq ($(subdir),string) +sysdep_routines += memcpy-power7 memcpy-a2 memcpy-power6 memcpy-cell \ + memcpy-power4 memcpy-ppc64 memcmp-power7 memcmp-power4 \ + memcmp-ppc64 memset-power7 memset-power6 memset-power4 \ + memset-ppc64 bzero-power4 bzero-power6 bzero-power7 \ + mempcpy-power7 mempcpy-ppc64 memchr-power7 memchr-ppc64 \ + memrchr-power7 memrchr-ppc64 rawmemchr-power7 \ + rawmemchr-ppc64 strlen-power7 strlen-ppc64 strnlen-power7 \ + strnlen-ppc64 strcasecmp-power7 strcasecmp_l-power7 \ + strncase-power7 strncase_l-power7 strncmp-power7 \ + strncmp-power4 strncmp-ppc64 strchr-power7 strchr-ppc64 \ + strchrnul-power7 strchrnul-ppc64 wcschr-power7 \ + wcschr-power6 wcschr-ppc64 wcsrchr-power7 wcsrchr-power6 \ + wcsrchr-ppc64 wcscpy-power7 wcscpy-power6 wcscpy-ppc64 \ + wordcopy-power7 wordcopy-power6 wordcopy-ppc64 \ + strcpy-power7 strcpy-ppc64 stpcpy-power7 stpcpy-ppc64 + +CFLAGS-strncase-power7.c += -mcpu=power7 -funroll-loops +CFLAGS-strncase_l-power7.c += -mcpu=power7 -funroll-loops +CFLAGS-wcschr-power7.c += -mcpu=power7 +CFLAGS-wcschr-power6.c += -mcpu=power6 +CFLAGS-wcsrchr-power7.c += -mcpu=power7 +CFLAGS-wcsrchr-power6.c += -mcpu=power6 +CFLAGS-wcscpy-power7.c += -mcpu=power7 +CFLAGS-wcscpy-power6.c += -mcpu=power6 +CFLAGS-wordcopy-power7.c += -mcpu=power7 +CFLAGS-wordcopy-power6.c += -mcpu=power6 +endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/bzero-power4.S b/sysdeps/powerpc/powerpc64/multiarch/bzero-power4.S new file mode 100644 index 0000000000..cf58fc5d7b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/bzero-power4.S @@ -0,0 +1,26 @@ +/* Optimized bzero implementation for PowerPC64/POWER4. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +ENTRY (__bzero_power4) + CALL_MCOUNT 3 + mr r5,r4 + li r4,0 + b __memset_power4 +END_GEN_TB (__bzero_power4,TB_TOCLESS) diff --git a/sysdeps/powerpc/powerpc64/multiarch/bzero-power6.S b/sysdeps/powerpc/powerpc64/multiarch/bzero-power6.S new file mode 100644 index 0000000000..ee3b919c7e --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/bzero-power6.S @@ -0,0 +1,26 @@ +/* Optimized bzero implementation for PowerPC64/POWER6. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +ENTRY (__bzero_power6) + CALL_MCOUNT 3 + mr r5,r4 + li r4,0 + b __memset_power6 +END_GEN_TB (__bzero_power6,TB_TOCLESS) diff --git a/sysdeps/powerpc/powerpc64/multiarch/bzero-power7.S b/sysdeps/powerpc/powerpc64/multiarch/bzero-power7.S new file mode 100644 index 0000000000..7abc6db849 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/bzero-power7.S @@ -0,0 +1,26 @@ +/* Optimized bzero implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +ENTRY (__bzero_power7) + CALL_MCOUNT 3 + mr r5,r4 + li r4,0 + b __memset_power7 +END_GEN_TB (__bzero_power7,TB_TOCLESS) diff --git a/sysdeps/powerpc/powerpc64/multiarch/bzero.c b/sysdeps/powerpc/powerpc64/multiarch/bzero.c new file mode 100644 index 0000000000..d7b3aaa877 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/bzero.c @@ -0,0 +1,40 @@ +/* Multiple versions of bzero. PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Define multiple versions only for definition in libc. */ +#ifndef NOT_IN_libc +# include <string.h> +# include <strings.h> +# include "init-arch.h" + +extern __typeof (bzero) __bzero_ppc attribute_hidden; +extern __typeof (bzero) __bzero_power4 attribute_hidden; +extern __typeof (bzero) __bzero_power6 attribute_hidden; +extern __typeof (bzero) __bzero_power7 attribute_hidden; + +libc_ifunc (__bzero, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __bzero_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __bzero_power6 : + (hwcap & PPC_FEATURE_POWER4) + ? __bzero_power4 + : __bzero_ppc); + +weak_alias (__bzero, bzero) +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c new file mode 100644 index 0000000000..33fc29e7f0 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c @@ -0,0 +1,242 @@ +/* Enumerate available IFUNC implementations of a function. PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <assert.h> +#include <string.h> +#include <wchar.h> +#include <ldsodefs.h> +#include <ifunc-impl-list.h> + +/* Maximum number of IFUNC implementations. */ +#define MAX_IFUNC 6 + +size_t +__libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, + size_t max) +{ + assert (max >= MAX_IFUNC); + + size_t i = 0; + + unsigned long int hwcap = GLRO(dl_hwcap); + /* hwcap contains only the latest supported ISA, the code checks which is + and fills the previous supported ones. */ + if (hwcap & PPC_FEATURE_ARCH_2_06) + hwcap |= PPC_FEATURE_ARCH_2_05 | PPC_FEATURE_POWER5_PLUS | + PPC_FEATURE_POWER5 | PPC_FEATURE_POWER4; + else if (hwcap & PPC_FEATURE_ARCH_2_05) + hwcap |= PPC_FEATURE_POWER5_PLUS | PPC_FEATURE_POWER5 | PPC_FEATURE_POWER4; + else if (hwcap & PPC_FEATURE_POWER5_PLUS) + hwcap |= PPC_FEATURE_POWER5 | PPC_FEATURE_POWER4; + else if (hwcap & PPC_FEATURE_POWER5) + hwcap |= PPC_FEATURE_POWER4; + +#ifdef SHARED + /* Support sysdeps/powerpc/powerpc64/multiarch/memcpy.c. */ + IFUNC_IMPL (i, name, memcpy, + IFUNC_IMPL_ADD (array, i, memcpy, hwcap & PPC_FEATURE_HAS_VSX, + __memcpy_power7) + IFUNC_IMPL_ADD (array, i, memcpy, hwcap & PPC_FEATURE_ARCH_2_06, + __memcpy_a2) + IFUNC_IMPL_ADD (array, i, memcpy, hwcap & PPC_FEATURE_ARCH_2_05, + __memcpy_power6) + IFUNC_IMPL_ADD (array, i, memcpy, hwcap & PPC_FEATURE_CELL_BE, + __memcpy_cell) + IFUNC_IMPL_ADD (array, i, memcpy, hwcap & PPC_FEATURE_POWER4, + __memcpy_power4) + IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/memset.c. */ + IFUNC_IMPL (i, name, memset, + IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_HAS_VSX, + __memset_power7) + IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_ARCH_2_05, + __memset_power6) + IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_POWER4, + __memset_power4) + IFUNC_IMPL_ADD (array, i, memset, 1, __memset_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strcpy.c. */ + IFUNC_IMPL (i, name, strcpy, + IFUNC_IMPL_ADD (array, i, strcpy, hwcap & PPC_FEATURE_HAS_VSX, + __strcpy_power7) + IFUNC_IMPL_ADD (array, i, strcpy, 1, + __strcpy_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/stpcpy.c. */ + IFUNC_IMPL (i, name, stpcpy, + IFUNC_IMPL_ADD (array, i, stpcpy, hwcap & PPC_FEATURE_HAS_VSX, + __stpcpy_power7) + IFUNC_IMPL_ADD (array, i, stpcpy, 1, + __stpcpy_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strlen.c. */ + IFUNC_IMPL (i, name, strlen, + IFUNC_IMPL_ADD (array, i, strlen, hwcap & PPC_FEATURE_HAS_VSX, + __strlen_power7) + IFUNC_IMPL_ADD (array, i, strlen, 1, + __strlen_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strncmp.c. */ + IFUNC_IMPL (i, name, strncmp, + IFUNC_IMPL_ADD (array, i, strncmp, hwcap & PPC_FEATURE_HAS_VSX, + __strncmp_power7) + IFUNC_IMPL_ADD (array, i, strncmp, hwcap & PPC_FEATURE_POWER4, + __strncmp_power4) + IFUNC_IMPL_ADD (array, i, strncmp, 1, + __strncmp_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strchr.c. */ + IFUNC_IMPL (i, name, strchr, + IFUNC_IMPL_ADD (array, i, strchr, + hwcap & PPC_FEATURE_HAS_VSX, + __strchr_power7) + IFUNC_IMPL_ADD (array, i, strchr, 1, + __strchr_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strchrnul.c. */ + IFUNC_IMPL (i, name, strchrnul, + IFUNC_IMPL_ADD (array, i, strchrnul, + hwcap & PPC_FEATURE_HAS_VSX, + __strchrnul_power7) + IFUNC_IMPL_ADD (array, i, strchrnul, 1, + __strchrnul_ppc)) +#endif + + /* Support sysdeps/powerpc/powerpc64/multiarch/memcmp.c. */ + IFUNC_IMPL (i, name, memcmp, + IFUNC_IMPL_ADD (array, i, memcmp, hwcap & PPC_FEATURE_HAS_VSX, + __memcmp_power7) + IFUNC_IMPL_ADD (array, i, memcmp, hwcap & PPC_FEATURE_POWER4, + __memcmp_power4) + IFUNC_IMPL_ADD (array, i, memcmp, 1, __memcmp_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/bzero.c. */ + IFUNC_IMPL (i, name, bzero, + IFUNC_IMPL_ADD (array, i, bzero, hwcap & PPC_FEATURE_HAS_VSX, + __bzero_power7) + IFUNC_IMPL_ADD (array, i, bzero, hwcap & PPC_FEATURE_ARCH_2_05, + __bzero_power6) + IFUNC_IMPL_ADD (array, i, bzero, hwcap & PPC_FEATURE_POWER4, + __bzero_power4) + IFUNC_IMPL_ADD (array, i, bzero, 1, __bzero_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/mempcpy.c. */ + IFUNC_IMPL (i, name, mempcpy, + IFUNC_IMPL_ADD (array, i, mempcpy, + hwcap & PPC_FEATURE_HAS_VSX, + __mempcpy_power7) + IFUNC_IMPL_ADD (array, i, mempcpy, 1, + __mempcpy_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/memchr.c. */ + IFUNC_IMPL (i, name, memchr, + IFUNC_IMPL_ADD (array, i, memchr, + hwcap & PPC_FEATURE_HAS_VSX, + __memchr_power7) + IFUNC_IMPL_ADD (array, i, memchr, 1, + __memchr_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/memrchr.c. */ + IFUNC_IMPL (i, name, memrchr, + IFUNC_IMPL_ADD (array, i, memrchr, + hwcap & PPC_FEATURE_HAS_VSX, + __memrchr_power7) + IFUNC_IMPL_ADD (array, i, memrchr, 1, + __memrchr_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c. */ + IFUNC_IMPL (i, name, rawmemchr, + IFUNC_IMPL_ADD (array, i, rawmemchr, + hwcap & PPC_FEATURE_HAS_VSX, + __rawmemchr_power7) + IFUNC_IMPL_ADD (array, i, rawmemchr, 1, + __rawmemchr_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strnlen.c. */ + IFUNC_IMPL (i, name, strnlen, + IFUNC_IMPL_ADD (array, i, strnlen, hwcap & PPC_FEATURE_HAS_VSX, + __strnlen_power7) + IFUNC_IMPL_ADD (array, i, strnlen, 1, + __strnlen_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c. */ + IFUNC_IMPL (i, name, strcasecmp, + IFUNC_IMPL_ADD (array, i, strcasecmp, + hwcap & PPC_FEATURE_HAS_VSX, + __strcasecmp_power7) + IFUNC_IMPL_ADD (array, i, strcasecmp, 1, __strcasecmp_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c. */ + IFUNC_IMPL (i, name, strcasecmp_l, + IFUNC_IMPL_ADD (array, i, strcasecmp_l, + hwcap & PPC_FEATURE_HAS_VSX, + __strcasecmp_l_power7) + IFUNC_IMPL_ADD (array, i, strcasecmp_l, 1, + __strcasecmp_l_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strncase.c. */ + IFUNC_IMPL (i, name, strncasecmp, + IFUNC_IMPL_ADD (array, i, strncasecmp, + hwcap & PPC_FEATURE_HAS_VSX, + __strncasecmp_power7) + IFUNC_IMPL_ADD (array, i, strncasecmp, 1, __strncasecmp_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/strncase_l.c. */ + IFUNC_IMPL (i, name, strncasecmp_l, + IFUNC_IMPL_ADD (array, i, strncasecmp_l, + hwcap & PPC_FEATURE_HAS_VSX, + __strncasecmp_l_power7) + IFUNC_IMPL_ADD (array, i, strncasecmp_l, 1, + __strncasecmp_l_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/wcschr.c. */ + IFUNC_IMPL (i, name, wcschr, + IFUNC_IMPL_ADD (array, i, wcschr, + hwcap & PPC_FEATURE_HAS_VSX, + __wcschr_power7) + IFUNC_IMPL_ADD (array, i, wcschr, + hwcap & PPC_FEATURE_ARCH_2_05, + __wcschr_power6) + IFUNC_IMPL_ADD (array, i, wcschr, 1, + __wcschr_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/wcschr.c. */ + IFUNC_IMPL (i, name, wcsrchr, + IFUNC_IMPL_ADD (array, i, wcsrchr, + hwcap & PPC_FEATURE_HAS_VSX, + __wcsrchr_power7) + IFUNC_IMPL_ADD (array, i, wcsrchr, + hwcap & PPC_FEATURE_ARCH_2_05, + __wcsrchr_power6) + IFUNC_IMPL_ADD (array, i, wcsrchr, 1, + __wcsrchr_ppc)) + + /* Support sysdeps/powerpc/powerpc64/multiarch/wcscpy.c. */ + IFUNC_IMPL (i, name, wcscpy, + IFUNC_IMPL_ADD (array, i, wcscpy, + hwcap & PPC_FEATURE_HAS_VSX, + __wcscpy_power7) + IFUNC_IMPL_ADD (array, i, wcscpy, + hwcap & PPC_FEATURE_ARCH_2_05, + __wcscpy_power6) + IFUNC_IMPL_ADD (array, i, wcscpy, 1, + __wcscpy_ppc)) + + return i; +} diff --git a/sysdeps/powerpc/powerpc64/multiarch/init-arch.h b/sysdeps/powerpc/powerpc64/multiarch/init-arch.h new file mode 100644 index 0000000000..b7d238cc08 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/init-arch.h @@ -0,0 +1,18 @@ +/* This file is part of the GNU C Library. + Copyright (C) 2013 Free Software Foundation, Inc. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/init-arch.h> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memchr-power7.S b/sysdeps/powerpc/powerpc64/multiarch/memchr-power7.S new file mode 100644 index 0000000000..094e2cfdab --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memchr-power7.S @@ -0,0 +1,41 @@ +/* Optimized memchr implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__memchr_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__memchr_power7): \ + cfi_startproc; \ + LOCALENTRY(__memchr_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__memchr_power7) \ + END_2(__memchr_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) +#undef weak_alias +#define weak_alias(name,alias) + +#include <sysdeps/powerpc/powerpc64/power7/memchr.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memchr-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/memchr-ppc64.c new file mode 100644 index 0000000000..8ec14c6a68 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memchr-ppc64.c @@ -0,0 +1,19 @@ +/* PowerPC64 default implementation of memchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/memchr-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memchr.c b/sysdeps/powerpc/powerpc64/multiarch/memchr.c new file mode 100644 index 0000000000..a4237f3967 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memchr.c @@ -0,0 +1,38 @@ +/* Multiple versions of memchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__memchr) __memchr_ppc attribute_hidden; +extern __typeof (__memchr) __memchr_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (__memchr, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __memchr_power7 + : __memchr_ppc); + +weak_alias (__memchr, memchr) +libc_hidden_builtin_def (memchr) +#else +#include <string/memchr.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcmp-power4.S b/sysdeps/powerpc/powerpc64/multiarch/memcmp-power4.S new file mode 100644 index 0000000000..d98857d06b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcmp-power4.S @@ -0,0 +1,42 @@ +/* Optimized memcmp implementation for PowerPC64/POWER4. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcmp_power4) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcmp_power4): \ + cfi_startproc; \ + LOCALENTRY(__memcmp_power4) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__memcmp_power4) \ + END_2(__memcmp_power4) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) +#undef weak_alias +#define weak_alias(name,alias) + +#include <sysdeps/powerpc/powerpc64/power4/memcmp.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcmp-power7.S b/sysdeps/powerpc/powerpc64/multiarch/memcmp-power7.S new file mode 100644 index 0000000000..9e027d47dc --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcmp-power7.S @@ -0,0 +1,42 @@ +/* Optimized memcmp implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcmp_power7) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcmp_power7): \ + cfi_startproc; \ + LOCALENTRY(__memcmp_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__memcmp_power7) \ + END_2(__memcmp_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) +#undef weak_alias +#define weak_alias(name,alias) + +#include <sysdeps/powerpc/powerpc64/power7/memcmp.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcmp-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/memcmp-ppc64.c new file mode 100644 index 0000000000..685e530bc6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcmp-ppc64.c @@ -0,0 +1,33 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <string.h> + +#define MEMCMP __memcmp_ppc +#undef weak_alias +#define weak_alias(name, aliasname) \ + extern __typeof (__memcmp_ppc) aliasname \ + __attribute__ ((weak, alias ("__memcmp_ppc"))); +#if !defined(NOT_IN_libc) && defined(SHARED) +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + __hidden_ver1(__memcmp_ppc, __GI_memcmp, __memcmp_ppc); +#endif + +extern __typeof (memcmp) __memcmp_ppc attribute_hidden; + +#include <string/memcmp.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcmp.c b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c new file mode 100644 index 0000000000..b63c8e13fb --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c @@ -0,0 +1,39 @@ +/* Multiple versions of memcmp. PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Define multiple versions only for definition in libc. */ +#ifndef NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (memcmp) __memcmp_ppc attribute_hidden; +extern __typeof (memcmp) __memcmp_power4 attribute_hidden; +extern __typeof (memcmp) __memcmp_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (memcmp, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __memcmp_power7 : + (hwcap & PPC_FEATURE_POWER4) + ? __memcmp_power4 + : __memcmp_ppc); +#else +#include <string/memcmp.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcpy-a2.S b/sysdeps/powerpc/powerpc64/multiarch/memcpy-a2.S new file mode 100644 index 0000000000..79796739b5 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy-a2.S @@ -0,0 +1,40 @@ +/* Optimized memcpy implementation for PowerPC A2. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcpy_a2) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcpy_a2): \ + cfi_startproc; \ + LOCALENTRY(__memcpy_a2) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memcpy_a2,mask) \ + END_2(__memcpy_a2) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/a2/memcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcpy-cell.S b/sysdeps/powerpc/powerpc64/multiarch/memcpy-cell.S new file mode 100644 index 0000000000..7b3c102400 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy-cell.S @@ -0,0 +1,40 @@ +/* Optimized memcpy implementation for PowerPC/CELL. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcpy_cell) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcpy_cell): \ + cfi_startproc; \ + LOCALENTRY(__memcpy_cell) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memcpy_cell,mask) \ + END_2(__memcpy_cell) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/cell/memcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcpy-power4.S b/sysdeps/powerpc/powerpc64/multiarch/memcpy-power4.S new file mode 100644 index 0000000000..295a1f204a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy-power4.S @@ -0,0 +1,40 @@ +/* Optimized memcpy implementation for PowerPC64/POWER4. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcpy_power4) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcpy_power4): \ + cfi_startproc; \ + LOCALENTRY(__memcpy_power4) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memcpy_power4,mask) \ + END_2(__memcpy_power4) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power4/memcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcpy-power6.S b/sysdeps/powerpc/powerpc64/multiarch/memcpy-power6.S new file mode 100644 index 0000000000..35fe8875c3 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy-power6.S @@ -0,0 +1,40 @@ +/* Optimized memcpy implementation for PowerPC/POWER6. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcpy_power6) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcpy_power6): \ + cfi_startproc; \ + LOCALENTRY(__memcpy_power6) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memcpy_power6,mask) \ + END_2(__memcpy_power6) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power6/memcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcpy-power7.S b/sysdeps/powerpc/powerpc64/multiarch/memcpy-power7.S new file mode 100644 index 0000000000..aa725d2f26 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy-power7.S @@ -0,0 +1,40 @@ +/* Optimized memcpy implementation for PowerPC/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcpy_power7) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcpy_power7): \ + cfi_startproc; \ + LOCALENTRY(__memcpy_power7) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memcpy_power7,mask) \ + END_2(__memcpy_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/memcpy.S> diff --git a/sysdeps/powerpc/powerpc64/power4/fpu/w_sqrtf.c b/sysdeps/powerpc/powerpc64/multiarch/memcpy-ppc64.S index 12d9f6273d..c8b70a0590 100644 --- a/sysdeps/powerpc/powerpc64/power4/fpu/w_sqrtf.c +++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy-ppc64.S @@ -1,5 +1,5 @@ -/* Single-precision floating point square root wrapper. - Copyright (C) 2004-2013 Free Software Foundation, Inc. +/* Default memcpy implementation for PowerPC64. + Copyright (C) 2013 Free Software Foundation, Inc. This file is part of the GNU C Library. The GNU C Library is free software; you can redistribute it and/or @@ -16,38 +16,28 @@ License along with the GNU C Library; if not, see <http://www.gnu.org/licenses/>. */ -#include <math.h> -#include <math_private.h> -#include <fenv_libc.h> - #include <sysdep.h> -#include <ldsodefs.h> - -float -__sqrtf (float x) /* wrapper sqrtf */ -{ -#ifdef _IEEE_LIBM - return __ieee754_sqrtf (x); -#else - float z; -/* Power4 (ISA V2.0) and above implement sqrtf in hardware. */ - __asm __volatile ( - " fsqrts %0,%1\n" - : "=f" (z) - : "f" (x)); - - if (__builtin_expect (_LIB_VERSION == _IEEE_, 0)) - return z; - - if (__builtin_expect (x != x, 0)) - return z; - - if (__builtin_expect (x < 0.0, 0)) - /* sqrtf(negative) */ - return (float) __kernel_standard ((double) x, (double) x, 126); - else - return z; + +#if defined SHARED && !defined NOT_IN_libc +# undef EALIGN +# define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memcpy_ppc) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memcpy_ppc): \ + cfi_startproc; \ + LOCALENTRY(__memcpy_ppc) + +# undef END_GEN_TB +# define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memcpy_ppc,mask) \ + END_2(__memcpy_ppc) + +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + .globl __GI_memcpy; __GI_memcpy = __memcpy_ppc #endif -} -weak_alias (__sqrtf, sqrtf) +#include <sysdeps/powerpc/powerpc64/memcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcpy.c b/sysdeps/powerpc/powerpc64/multiarch/memcpy.c new file mode 100644 index 0000000000..b580ef7e5c --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy.c @@ -0,0 +1,48 @@ +/* Multiple versions of memcpy. PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Define multiple versions only for the definition in lib and for + DSO. In static binaries we need memcpy before the initialization + happened. */ +#if defined SHARED && !defined NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (memcpy) __memcpy_ppc attribute_hidden; +extern __typeof (memcpy) __memcpy_power4 attribute_hidden; +extern __typeof (memcpy) __memcpy_cell attribute_hidden; +extern __typeof (memcpy) __memcpy_power6 attribute_hidden; +extern __typeof (memcpy) __memcpy_a2 attribute_hidden; +extern __typeof (memcpy) __memcpy_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (memcpy, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __memcpy_power7 : + (hwcap & PPC_FEATURE_ARCH_2_06) + ? __memcpy_a2 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __memcpy_power6 : + (hwcap & PPC_FEATURE_CELL_BE) + ? __memcpy_cell : + (hwcap & PPC_FEATURE_POWER4) + ? __memcpy_power4 + : __memcpy_ppc); +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/mempcpy-power7.S b/sysdeps/powerpc/powerpc64/multiarch/mempcpy-power7.S new file mode 100644 index 0000000000..6d7f002d68 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/mempcpy-power7.S @@ -0,0 +1,42 @@ +/* Optimized mempcpy implementation for PowerPC/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__mempcpy_power7) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__mempcpy_power7): \ + cfi_startproc; \ + LOCALENTRY(__mempcpy_power7) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__mempcpy_power7,mask) \ + END_2(__mempcpy_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) +#undef weak_alias +#define weak_alias(name, alias) + +#include <sysdeps/powerpc/powerpc64/power7/mempcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/mempcpy-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/mempcpy-ppc64.c new file mode 100644 index 0000000000..40f6c42432 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/mempcpy-ppc64.c @@ -0,0 +1,19 @@ +/* PowerPC64 default implementation of mempcpy. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/mempcpy-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/mempcpy.c b/sysdeps/powerpc/powerpc64/multiarch/mempcpy.c new file mode 100644 index 0000000000..7100eb513a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/mempcpy.c @@ -0,0 +1,38 @@ +/* Multiple versions of mempcpy. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__mempcpy) __mempcpy_ppc attribute_hidden; +extern __typeof (__mempcpy) __mempcpy_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (__mempcpy, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __mempcpy_power7 + : __mempcpy_ppc); + +weak_alias (__mempcpy, mempcpy) +libc_hidden_def (mempcpy) +#else +# include <string/mempcpy.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/memrchr-power7.S b/sysdeps/powerpc/powerpc64/multiarch/memrchr-power7.S new file mode 100644 index 0000000000..a6f825c0f1 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memrchr-power7.S @@ -0,0 +1,41 @@ +/* Optimized memrchr implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__memrchr_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__memrchr_power7): \ + cfi_startproc; \ + LOCALENTRY(__memrchr_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__memrchr_power7) \ + END_2(__memrchr_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) +#undef weak_alias +#define weak_alias(name,alias) + +#include <sysdeps/powerpc/powerpc64/power7/memrchr.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memrchr-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/memrchr-ppc64.c new file mode 100644 index 0000000000..8c291e6cfb --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memrchr-ppc64.c @@ -0,0 +1,19 @@ +/* PowerPC64 default implementation of memrchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/memrchr-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memrchr.c b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c new file mode 100644 index 0000000000..cc362bac28 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c @@ -0,0 +1,37 @@ +/* Multiple versions of memrchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__memrchr) __memrchr_ppc attribute_hidden; +extern __typeof (__memrchr) __memrchr_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (__memrchr, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __memrchr_power7 + : __memrchr_ppc); + +weak_alias (__memrchr, memrchr) +#else +#include <string/memrchr.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset-power4.S b/sysdeps/powerpc/powerpc64/multiarch/memset-power4.S new file mode 100644 index 0000000000..aac7fb1295 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memset-power4.S @@ -0,0 +1,41 @@ +/* Optimized memset implementation for PowerPC64/POWER4. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memset_power4) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memset_power4): \ + cfi_startproc; \ + LOCALENTRY(__memset_power4) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memset_power4,mask) \ + END_2(__memset_power4) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#define NO_BZERO_IMPL +#include <sysdeps/powerpc/powerpc64/power4/memset.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset-power6.S b/sysdeps/powerpc/powerpc64/multiarch/memset-power6.S new file mode 100644 index 0000000000..3144c699d5 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memset-power6.S @@ -0,0 +1,41 @@ +/* Optimized memset implementation for PowerPC64/POWER6. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memset_power6) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memset_power6): \ + cfi_startproc; \ + LOCALENTRY(__memset_power6) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memset_power6,mask) \ + END_2(__memset_power6) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#define NO_BZERO_IMPL +#include <sysdeps/powerpc/powerpc64/power6/memset.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset-power7.S b/sysdeps/powerpc/powerpc64/multiarch/memset-power7.S new file mode 100644 index 0000000000..ecfa963bd7 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memset-power7.S @@ -0,0 +1,41 @@ +/* Optimized memset implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memset_power7) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memset_power7): \ + cfi_startproc; \ + LOCALENTRY(__memset_power7) + +#undef END_GEN_TB +#define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memset_power7,mask) \ + END_2(__memset_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#define NO_BZERO_IMPL +#include <sysdeps/powerpc/powerpc64/power7/memset.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/memset-ppc64.S new file mode 100644 index 0000000000..95b98f8834 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memset-ppc64.S @@ -0,0 +1,56 @@ +/* Default memset/bzero implementation for PowerPC64. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#if defined SHARED && !defined NOT_IN_libc + +/* Copied from bzero.S to prevent the linker from inserting a stub + between bzero and memset. */ +ENTRY (__bzero_ppc) + CALL_MCOUNT 3 + mr r5,r4 + li r4,0 + b L(_memset) +END_GEN_TB (__bzero_ppc,TB_TOCLESS) + +# undef EALIGN +# define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__memset_ppc) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__memset_ppc): \ + cfi_startproc; \ + LOCALENTRY(__memset_ppc) + +# undef END_GEN_TB +# define END_GEN_TB(name, mask) \ + cfi_endproc; \ + TRACEBACK_MASK(__memset_ppc,mask) \ + END_2(__memset_ppc) + +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + .globl __GI_memset; __GI_memset = __memset_ppc + +/* Do not implement __bzero at powerpc64/memset.S. */ +# define NO_BZERO_IMPL +#endif + +#include <sysdeps/powerpc/powerpc64/memset.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset.c b/sysdeps/powerpc/powerpc64/multiarch/memset.c new file mode 100644 index 0000000000..226799e64d --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/memset.c @@ -0,0 +1,40 @@ +/* Multiple versions of memset. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Define multiple versions only for definition in libc. */ +#if defined SHARED && !defined NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (memset) __memset_ppc attribute_hidden; +extern __typeof (memset) __memset_power4 attribute_hidden; +extern __typeof (memset) __memset_power6 attribute_hidden; +extern __typeof (memset) __memset_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (memset, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __memset_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __memset_power6 : + (hwcap & PPC_FEATURE_POWER4) + ? __memset_power4 + : __memset_ppc); +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-power7.S b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-power7.S new file mode 100644 index 0000000000..d3ab4f1512 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-power7.S @@ -0,0 +1,36 @@ +/* Optimized rawmemchr implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__rawmemchr_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__rawmemchr_power7): \ + cfi_startproc; \ + LOCALENTRY(__rawmemchr_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__rawmemchr_power7) \ + END_2(__rawmemchr_power7) + +#include <sysdeps/powerpc/powerpc64/power7/rawmemchr.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-ppc64.c new file mode 100644 index 0000000000..c69b213b45 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-ppc64.c @@ -0,0 +1,19 @@ +/* PowerPC64 default implementation of rawmemchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/rawmemchr-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c new file mode 100644 index 0000000000..c083490f3f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c @@ -0,0 +1,37 @@ +/* Multiple versions of rawmemchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__rawmemchr) __rawmemchr_ppc attribute_hidden; +extern __typeof (__rawmemchr) __rawmemchr_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (__rawmemchr, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __rawmemchr_power7 + : __rawmemchr_ppc); + +weak_alias (__rawmemchr, rawmemchr) +#else +#include <string/rawmemchr.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/rtld-memset.c b/sysdeps/powerpc/powerpc64/multiarch/rtld-memset.c new file mode 100644 index 0000000000..c74a770528 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/rtld-memset.c @@ -0,0 +1,18 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc64/rtld-memset.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/rtld-strchr.S b/sysdeps/powerpc/powerpc64/multiarch/rtld-strchr.S new file mode 100644 index 0000000000..93f0eb49ee --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/rtld-strchr.S @@ -0,0 +1,18 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc64/strchr.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.S b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.S new file mode 100644 index 0000000000..1b75dd18e9 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.S @@ -0,0 +1,40 @@ +/* Optimized stpcpy implementation for POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__stpcpy_power7) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__stpcpy_power7): \ + cfi_startproc; \ + LOCALENTRY(__stpcpy_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__stpcpy_power7) \ + END_2(__stpcpy_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/stpcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.S new file mode 100644 index 0000000000..874d8aa8f3 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.S @@ -0,0 +1,48 @@ +/* Default stpcpy implementation for PowerPC64. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#if defined SHARED && !defined NOT_IN_libc +# undef EALIGN +# define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__stpcpy_ppc) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__stpcpy_ppc): \ + cfi_startproc; \ + LOCALENTRY(__stpcpy_ppc) + +# undef END +# define END(name) \ + cfi_endproc; \ + TRACEBACK(__stpcpy_ppc) \ + END_2(__stpcpy_ppc) + +# undef weak_alias +# define weak_alias(name, alias) +# undef libc_hidden_def +# define libc_hidden_def(name) + +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + .globl __GI___stpcpy; __GI___stpcpy = __stpcpy_ppc +#endif + +#include <sysdeps/powerpc/powerpc64/stpcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c b/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c new file mode 100644 index 0000000000..4df555f204 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c @@ -0,0 +1,34 @@ +/* Multiple versions of stpcpy. PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#if defined SHARED && !defined NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__stpcpy) __stpcpy_ppc attribute_hidden; +extern __typeof (__stpcpy) __stpcpy_power7 attribute_hidden; + +libc_ifunc (__stpcpy, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __stpcpy_power7 + : __stpcpy_ppc); + +weak_alias (__stpcpy, stpcpy) +libc_hidden_def (stpcpy) +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp-power7.S new file mode 100644 index 0000000000..20d072219b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp-power7.S @@ -0,0 +1,42 @@ +/* Optimized strcasecmp implementation foOWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strcasecmp_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strcasecmp_power7): \ + cfi_startproc; \ + LOCALENTRY(__strcasecmp_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strcasecmp_power7) \ + END_2(__strcasecmp_power7) + +#undef weak_alias(name, alias) +#define weak_alias(name, alias) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/strcasecmp.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c new file mode 100644 index 0000000000..21d80d3254 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c @@ -0,0 +1,40 @@ +/* Multiple versions of strcasecmp. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# define strcasecmp __strcasecmp_ppc +extern __typeof (__strcasecmp) __strcasecmp_ppc attribute_hidden; +extern __typeof (__strcasecmp) __strcasecmp_power7 attribute_hidden; +#endif + +#include <string/strcasecmp.c> +#undef strcasecmp + +#ifndef NOT_IN_libc +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__strcasecmp) __libc_strcasecmp; +libc_ifunc (__libc_strcasecmp, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strcasecmp_power7 + : __strcasecmp_ppc); + +weak_alias (__libc_strcasecmp, strcasecmp) +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l-power7.S new file mode 100644 index 0000000000..3bf2ba6207 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l-power7.S @@ -0,0 +1,44 @@ +/* Optimized strcasecmp_l implementation for POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strcasecmp_l_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strcasecmp_l_power7): \ + cfi_startproc; \ + LOCALENTRY(__strcasecmp_l_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strcasecmp_l_power7) \ + END_2(__strcasecmp_l_power7) + +#undef weak_alias(name, alias) +#define weak_alias(name, alias) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#define USE_IN_EXTENDED_LOCALE_MODEL + +#include <sysdeps/powerpc/powerpc64/power7/strcasecmp.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c new file mode 100644 index 0000000000..975dcef17c --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c @@ -0,0 +1,40 @@ +/* Multiple versions of strcasecmp_l. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# define strcasecmp_l __strcasecmp_l_ppc +extern __typeof (__strcasecmp_l) __strcasecmp_l_ppc attribute_hidden; +extern __typeof (__strcasecmp_l) __strcasecmp_l_power7 attribute_hidden; +#endif + +#include <string/strcasecmp_l.c> +#undef strcasecmp_l + +#ifndef NOT_IN_libc +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__strcasecmp_l) __libc_strcasecmp_l; +libc_ifunc (__libc_strcasecmp_l, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strcasecmp_l_power7 + : __strcasecmp_l_ppc); + +weak_alias (__libc_strcasecmp_l, strcasecmp_l) +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchr-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strchr-power7.S new file mode 100644 index 0000000000..b1cf338e47 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strchr-power7.S @@ -0,0 +1,39 @@ +/* Optimized strchr implementation for POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strchr_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strchr_power7): \ + cfi_startproc; \ + LOCALENTRY(__strchr_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strchr_power7) \ + END_2(__strchr_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/strchr.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchr-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/strchr-ppc64.S new file mode 100644 index 0000000000..814ed87d01 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strchr-ppc64.S @@ -0,0 +1,42 @@ +/* PowerPC64 default implementation of strchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#ifdef SHARED +# undef ENTRY +# define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strchr_ppc) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strchr_ppc): \ + cfi_startproc; \ + LOCALENTRY(__strchr_ppc) + +# undef END +# define END(name) \ + cfi_endproc; \ + TRACEBACK(__strchr_ppc) \ + END_2(__strchr_ppc) + +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + .globl __GI_strchr; __GI_strchr = __strchr_ppc +#endif + +#include <sysdeps/powerpc/powerpc64/strchr.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchr.c b/sysdeps/powerpc/powerpc64/multiarch/strchr.c new file mode 100644 index 0000000000..8a7dc74833 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strchr.c @@ -0,0 +1,35 @@ +/* Multiple versions of strchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Define multiple versions only for definition in libc. */ +#if defined SHARED && !defined NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (strchr) __strchr_ppc attribute_hidden; +extern __typeof (strchr) __strchr_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (strchr, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strchr_power7 + : __strchr_ppc); +weak_alias (strchr, index) +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchrnul-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strchrnul-power7.S new file mode 100644 index 0000000000..9454b0ca29 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strchrnul-power7.S @@ -0,0 +1,39 @@ +/* Optimized strchrnul implementation for POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strchrnul_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strchrnul_power7): \ + cfi_startproc; \ + LOCALENTRY(__strchrnul_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strchrnul_power7) \ + END_2(__strchrnul_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/strchrnul.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchrnul-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/strchrnul-ppc64.c new file mode 100644 index 0000000000..7172ddf04a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strchrnul-ppc64.c @@ -0,0 +1,19 @@ +/* PowerPC64 default implementation of strchrnul. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/strchrnul-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c new file mode 100644 index 0000000000..95138580d8 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c @@ -0,0 +1,37 @@ +/* Multiple versions of strchrnul. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__strchrnul) __strchrnul_ppc attribute_hidden; +extern __typeof (__strchrnul) __strchrnul_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (__strchrnul, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strchrnul_power7 + : __strchrnul_ppc); + +weak_alias (__strchrnul, strchrnul) +#else +#include <string/strchrnul.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.S new file mode 100644 index 0000000000..127422f439 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.S @@ -0,0 +1,40 @@ +/* Optimized strcpy implementation for POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__strcpy_power7) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__strcpy_power7): \ + cfi_startproc; \ + LOCALENTRY(__strcpy_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strcpy_power7) \ + END_2(__strcpy_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/strcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.S new file mode 100644 index 0000000000..255fc3f1fb --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.S @@ -0,0 +1,43 @@ +/* Default strcpy implementation for PowerPC64. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#if defined SHARED && !defined NOT_IN_libc +# undef EALIGN +# define EALIGN(name, alignt, words) \ + .section ".text"; \ + ENTRY_2(__strcpy_ppc) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__strcpy_ppc): \ + cfi_startproc; \ + LOCALENTRY(__strcpy_ppc) + +# undef END +# define END(name) \ + cfi_endproc; \ + TRACEBACK(__strcpy_ppc) \ + END_2(__strcpy_ppc) + +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + .globl __GI_strcpy; __GI_strcpy = __strcpy_ppc +#endif + +#include <sysdeps/powerpc/powerpc64/strcpy.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy.c b/sysdeps/powerpc/powerpc64/multiarch/strcpy.c new file mode 100644 index 0000000000..8d555c9417 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy.c @@ -0,0 +1,31 @@ +/* Multiple versions of strcpy. PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#if defined SHARED && !defined NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (strcpy) __strcpy_ppc attribute_hidden; +extern __typeof (strcpy) __strcpy_power7 attribute_hidden; + +libc_ifunc (strcpy, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strcpy_power7 + : __strcpy_ppc); +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strlen-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strlen-power7.S new file mode 100644 index 0000000000..1027d7d076 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strlen-power7.S @@ -0,0 +1,39 @@ +/* Optimized strlen implementation for POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strlen_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strlen_power7): \ + cfi_startproc; \ + LOCALENTRY(__strlen_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strlen_power7) \ + END_2(__strlen_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/strlen.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strlen-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/strlen-ppc64.S new file mode 100644 index 0000000000..a78460efbd --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strlen-ppc64.S @@ -0,0 +1,42 @@ +/* Default strlen implementation for PowerPC64. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#if defined SHARED && !defined NOT_IN_libc +# undef ENTRY +# define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strlen_ppc) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strlen_ppc): \ + cfi_startproc; \ + LOCALENTRY(__strlen_ppc) + +# undef END +# define END(name) \ + cfi_endproc; \ + TRACEBACK(__strlen_ppc) \ + END_2(__strlen_ppc) + +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + .globl __GI_strlen; __GI_strlen = __strlen_ppc +#endif + +#include <sysdeps/powerpc/powerpc64/strlen.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strlen.c b/sysdeps/powerpc/powerpc64/multiarch/strlen.c new file mode 100644 index 0000000000..21b9294707 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strlen.c @@ -0,0 +1,31 @@ +/* Multiple versions of strlen. PowerPC64 version. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#if defined SHARED && !defined NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (strlen) __strlen_ppc attribute_hidden; +extern __typeof (strlen) __strlen_power7 attribute_hidden; + +libc_ifunc (strlen, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strlen_power7 + : __strlen_ppc); +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase-power7.c b/sysdeps/powerpc/powerpc64/multiarch/strncase-power7.c new file mode 100644 index 0000000000..de6ac409e9 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncase-power7.c @@ -0,0 +1,24 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <string.h> + +#define __strncasecmp __strncasecmp_power7 + +extern __typeof (strncasecmp) __strncasecmp_power7 attribute_hidden; + +#include <string/strncase.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase.c b/sysdeps/powerpc/powerpc64/multiarch/strncase.c new file mode 100644 index 0000000000..7fea0e3cf9 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncase.c @@ -0,0 +1,41 @@ +/* Multiple versions of strncasecmp + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# define strncasecmp __strncasecmp_ppc +extern __typeof (__strncasecmp) __strncasecmp_ppc attribute_hidden; +extern __typeof (__strncasecmp) __strncasecmp_power7 attribute_hidden; +#endif + +#include <string/strncase.c> +#undef strncasecmp + +#ifndef NOT_IN_libc +# include <shlib-compat.h> +# include "init-arch.h" + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +extern __typeof (__strncasecmp) __libc_strncasecmp; +libc_ifunc (__libc_strncasecmp, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strncasecmp_power7 + : __strncasecmp_ppc); +weak_alias (__libc_strncasecmp, strncasecmp) +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase_l-power7.c b/sysdeps/powerpc/powerpc64/multiarch/strncase_l-power7.c new file mode 100644 index 0000000000..10a49fc9ec --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncase_l-power7.c @@ -0,0 +1,25 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <string.h> + +#define __strncasecmp_l __strncasecmp_l_power7 +#define USE_IN_EXTENDED_LOCALE_MODEL 1 + +extern __typeof (strncasecmp_l) __strncasecmp_l_power7 attribute_hidden; + +#include <string/strncase.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c b/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c new file mode 100644 index 0000000000..4f3fb91909 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c @@ -0,0 +1,42 @@ +/* Multiple versions of strncasecmp_l + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# define strncasecmp_l __strncasecmp_l_ppc +extern __typeof (__strncasecmp_l) __strncasecmp_l_ppc attribute_hidden; +extern __typeof (__strncasecmp_l) __strncasecmp_l_power7 attribute_hidden; +#endif + +#include <string/strncase_l.c> +#undef strncasecmp_l + +#ifndef NOT_IN_libc +# include <shlib-compat.h> +# include "init-arch.h" + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +extern __typeof (__strncasecmp_l) __libc_strncasecmp_l; +libc_ifunc (__libc_strncasecmp_l, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strncasecmp_l_power7 + : __strncasecmp_l_ppc); + +weak_alias (__libc_strncasecmp_l, strncasecmp_l) +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp-power4.S b/sysdeps/powerpc/powerpc64/multiarch/strncmp-power4.S new file mode 100644 index 0000000000..ec786fd6b9 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp-power4.S @@ -0,0 +1,39 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name,alignt,words) \ + .section ".text"; \ + ENTRY_2(__strncmp_power4) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__strncmp_power4): \ + cfi_startproc; \ + LOCALENTRY(__strncmp_power4) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strncmp_power4) \ + END_2(__strncmp_power4) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power4/strncmp.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strncmp-power7.S new file mode 100644 index 0000000000..3293728370 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp-power7.S @@ -0,0 +1,40 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef EALIGN +#define EALIGN(name,alignt,words) \ + .section ".text"; \ + ENTRY_2(__strncmp_power7) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__strncmp_power7): \ + cfi_startproc; \ + LOCALENTRY(__strncmp_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strncmp_power7) \ + END_2(__strncmp_power7) + + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) + +#include <sysdeps/powerpc/powerpc64/power7/strncmp.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/strncmp-ppc64.S new file mode 100644 index 0000000000..fff07df4fa --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp-ppc64.S @@ -0,0 +1,42 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#if defined SHARED && !defined NOT_IN_libc +#undef EALIGN +#define EALIGN(name,alignt,words) \ + .section ".text"; \ + ENTRY_2(__strncmp_ppc) \ + .align ALIGNARG(alignt); \ + EALIGN_W_##words; \ + BODY_LABEL(__strncmp_ppc): \ + cfi_startproc; \ + LOCALENTRY(__strncmp_ppc) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strncmp_ppc) \ + END_2(__strncmp_ppc) + +# undef libc_hidden_builtin_def +# define libc_hidden_builtin_def(name) \ + .globl __GI_strncmp; __GI_strncmp = __strncmp_ppc +#endif + +#include <sysdeps/powerpc/powerpc64/strncmp.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c new file mode 100644 index 0000000000..09db0be3bb --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c @@ -0,0 +1,37 @@ +/* Multiple versions of strncmp. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +/* Define multiple versions only for definition in libc. */ +#if defined SHARED && !defined NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (strncmp) __strncmp_ppc attribute_hidden; +extern __typeof (strncmp) __strncmp_power4 attribute_hidden; +extern __typeof (strncmp) __strncmp_power7 attribute_hidden; + +/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle + ifunc symbol properly. */ +libc_ifunc (strncmp, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strncmp_power7 : + (hwcap & PPC_FEATURE_POWER4) + ? __strncmp_power4 + : __strncmp_ppc); +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/strnlen-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strnlen-power7.S new file mode 100644 index 0000000000..69a3683a0b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strnlen-power7.S @@ -0,0 +1,41 @@ +/* Optimized strnlen version for POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#undef ENTRY +#define ENTRY(name) \ + .section ".text"; \ + ENTRY_2(__strnlen_power7) \ + .align ALIGNARG(2); \ + BODY_LABEL(__strnlen_power7): \ + cfi_startproc; \ + LOCALENTRY(__strnlen_power7) + +#undef END +#define END(name) \ + cfi_endproc; \ + TRACEBACK(__strnlen_power7) \ + END_2(__strnlen_power7) + +#undef libc_hidden_builtin_def +#define libc_hidden_builtin_def(name) +#undef weak_alias(name, alias) +#define weak_alias(name, alias) + +#include <sysdeps/powerpc/powerpc64/power7/strnlen.S> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strnlen-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/strnlen-ppc64.c new file mode 100644 index 0000000000..bc4d8ae3cb --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strnlen-ppc64.c @@ -0,0 +1,18 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/strnlen-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strnlen.c b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c new file mode 100644 index 0000000000..00806779c2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c @@ -0,0 +1,36 @@ +/* Multiple versions of strnlen. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <string.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (__strnlen) __strnlen_ppc attribute_hidden; +extern __typeof (__strnlen) __strnlen_power7 attribute_hidden; + +libc_ifunc (__strnlen, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __strnlen_power7 + : __strnlen_ppc); +weak_alias (__strnlen, strnlen) +libc_hidden_def (strnlen) + +#else +#include <string/strnlen.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcschr-power6.c b/sysdeps/powerpc/powerpc64/multiarch/wcschr-power6.c new file mode 100644 index 0000000000..b03004e81b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcschr-power6.c @@ -0,0 +1,19 @@ +/* wcschr.c - Wide Character Search for powerpc64/power6. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcschr-power6.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcschr-power7.c b/sysdeps/powerpc/powerpc64/multiarch/wcschr-power7.c new file mode 100644 index 0000000000..fa0a696c7e --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcschr-power7.c @@ -0,0 +1,19 @@ +/* wcschr.c - Wide Character Search for powerpc64/power7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcschr-power7.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcschr-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/wcschr-ppc64.c new file mode 100644 index 0000000000..b155dc929b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcschr-ppc64.c @@ -0,0 +1,18 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcschr-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcschr.c b/sysdeps/powerpc/powerpc64/multiarch/wcschr.c new file mode 100644 index 0000000000..51fbefea2b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcschr.c @@ -0,0 +1,38 @@ +/* Multiple versions of wcschr + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <wchar.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (wcschr) __wcschr_ppc attribute_hidden; +extern __typeof (wcschr) __wcschr_power6 attribute_hidden; +extern __typeof (wcschr) __wcschr_power7 attribute_hidden; + +libc_ifunc (wcschr, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __wcschr_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __wcschr_power6 + : __wcschr_ppc); +#else +#undef libc_hidden_def +#define libc_hidden_def(a) +#include <wcsmbs/wcschr.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcscpy-power6.c b/sysdeps/powerpc/powerpc64/multiarch/wcscpy-power6.c new file mode 100644 index 0000000000..10cf9c7ee2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcscpy-power6.c @@ -0,0 +1,19 @@ +/* wcscpy.c - Wide Character Search for powerpc64/power6. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-power6.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcscpy-power7.c b/sysdeps/powerpc/powerpc64/multiarch/wcscpy-power7.c new file mode 100644 index 0000000000..cac2e3550f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcscpy-power7.c @@ -0,0 +1,19 @@ +/* wcscpy.c - Wide Character Search for powerpc64/power7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-power7.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcscpy-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/wcscpy-ppc64.c new file mode 100644 index 0000000000..a2d0f233ee --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcscpy-ppc64.c @@ -0,0 +1,18 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcscpy.c b/sysdeps/powerpc/powerpc64/multiarch/wcscpy.c new file mode 100644 index 0000000000..27715b3174 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcscpy.c @@ -0,0 +1,36 @@ +/* Multiple versions of wcscpy. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <wchar.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (wcscpy) __wcscpy_ppc attribute_hidden; +extern __typeof (wcscpy) __wcscpy_power6 attribute_hidden; +extern __typeof (wcscpy) __wcscpy_power7 attribute_hidden; + +libc_ifunc (wcscpy, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __wcscpy_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __wcscpy_power6 + : __wcscpy_ppc); +#else +#include <wcsmbs/wcscpy.c> +#endif: diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-power6.c b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-power6.c new file mode 100644 index 0000000000..d1b29b1fb2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-power6.c @@ -0,0 +1,19 @@ +/* wcsrchr.c - Wide Character Search for powerpc64/power6. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcsrchr-power6.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-power7.c b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-power7.c new file mode 100644 index 0000000000..a2717bc49a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-power7.c @@ -0,0 +1,19 @@ +/* wcsrchr.c - Wide Character Search for powerpc64/power7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcsrchr-power7.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-ppc64.c new file mode 100644 index 0000000000..3e698a56b4 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr-ppc64.c @@ -0,0 +1,18 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wcsrchr-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcsrchr.c b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr.c new file mode 100644 index 0000000000..01470c9a5b --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wcsrchr.c @@ -0,0 +1,36 @@ +/* Multiple versions of wcsrchr. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <wchar.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (wcsrchr) __wcsrchr_ppc attribute_hidden; +extern __typeof (wcsrchr) __wcsrchr_power6 attribute_hidden; +extern __typeof (wcsrchr) __wcsrchr_power7 attribute_hidden; + +libc_ifunc (wcsrchr, + (hwcap & PPC_FEATURE_HAS_VSX) + ? __wcsrchr_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? __wcsrchr_power6 + : __wcsrchr_ppc); +#else +#include <wcsmbs/wcsrchr.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/wordcopy-power6.c b/sysdeps/powerpc/powerpc64/multiarch/wordcopy-power6.c new file mode 100644 index 0000000000..671d6d9564 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wordcopy-power6.c @@ -0,0 +1,19 @@ +/* wordcopy routines for powerpc64/power6. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wordcopy-power6.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wordcopy-power7.c b/sysdeps/powerpc/powerpc64/multiarch/wordcopy-power7.c new file mode 100644 index 0000000000..fb690503c7 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wordcopy-power7.c @@ -0,0 +1,19 @@ +/* wordcopy routines for powerpc64/power7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, see <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wordcopy-power7.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wordcopy-ppc64.c b/sysdeps/powerpc/powerpc64/multiarch/wordcopy-ppc64.c new file mode 100644 index 0000000000..c5830a63f1 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wordcopy-ppc64.c @@ -0,0 +1,18 @@ +/* Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdeps/powerpc/powerpc32/power4/multiarch/wordcopy-ppc32.c> diff --git a/sysdeps/powerpc/powerpc64/multiarch/wordcopy.c b/sysdeps/powerpc/powerpc64/multiarch/wordcopy.c new file mode 100644 index 0000000000..78233dce66 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/wordcopy.c @@ -0,0 +1,86 @@ +/* Multiple versions of wordcopy functions. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef NOT_IN_libc +# include <stddef.h> +# include <memcopy.h> +# include <shlib-compat.h> +# include "init-arch.h" + +extern __typeof (_wordcopy_fwd_aligned) _wordcopy_fwd_aligned_ppc +attribute_hidden; +extern __typeof (_wordcopy_fwd_aligned) _wordcopy_fwd_aligned_power6 +attribute_hidden; +extern __typeof (_wordcopy_fwd_aligned) _wordcopy_fwd_aligned_power7 +attribute_hidden; + +libc_ifunc (_wordcopy_fwd_aligned, + (hwcap & PPC_FEATURE_HAS_VSX) + ? _wordcopy_fwd_aligned_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? _wordcopy_fwd_aligned_power6 + : _wordcopy_fwd_aligned_ppc); + + +extern __typeof (_wordcopy_fwd_dest_aligned) _wordcopy_fwd_dest_aligned_ppc +attribute_hidden; +extern __typeof (_wordcopy_fwd_dest_aligned) _wordcopy_fwd_dest_aligned_power6 +attribute_hidden; +extern __typeof (_wordcopy_fwd_dest_aligned) _wordcopy_fwd_dest_aligned_power7 +attribute_hidden; + +libc_ifunc (_wordcopy_fwd_dest_aligned, + (hwcap & PPC_FEATURE_HAS_VSX) + ? _wordcopy_fwd_dest_aligned_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? _wordcopy_fwd_dest_aligned_power6 + : _wordcopy_fwd_dest_aligned_ppc); + + +extern __typeof (_wordcopy_bwd_aligned) _wordcopy_bwd_aligned_ppc +attribute_hidden; +extern __typeof (_wordcopy_bwd_aligned) _wordcopy_bwd_aligned_power6 +attribute_hidden; +extern __typeof (_wordcopy_bwd_aligned) _wordcopy_bwd_aligned_power7 +attribute_hidden; + +libc_ifunc (_wordcopy_bwd_aligned, + (hwcap & PPC_FEATURE_HAS_VSX) + ? _wordcopy_bwd_aligned_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? _wordcopy_bwd_aligned_power6 + : _wordcopy_bwd_aligned_ppc); + + +extern __typeof (_wordcopy_bwd_dest_aligned) _wordcopy_bwd_dest_aligned_ppc +attribute_hidden; +extern __typeof (_wordcopy_bwd_dest_aligned) _wordcopy_bwd_dest_aligned_power6 +attribute_hidden; +extern __typeof (_wordcopy_bwd_dest_aligned) _wordcopy_bwd_dest_aligned_power7 +attribute_hidden; + +libc_ifunc (_wordcopy_bwd_dest_aligned, + (hwcap & PPC_FEATURE_HAS_VSX) + ? _wordcopy_bwd_dest_aligned_power7 : + (hwcap & PPC_FEATURE_ARCH_2_05) + ? _wordcopy_bwd_dest_aligned_power6 + : _wordcopy_bwd_dest_aligned_ppc); + +#else +#include <sysdeps/powerpc/power4/wordcopy.c> +#endif diff --git a/sysdeps/powerpc/powerpc64/power4/fpu/Implies b/sysdeps/powerpc/powerpc64/power4/fpu/Implies new file mode 100644 index 0000000000..c1f617b7da --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power4/fpu/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/fpu diff --git a/sysdeps/powerpc/powerpc64/power4/fpu/multiarch/Implies b/sysdeps/powerpc/powerpc64/power4/fpu/multiarch/Implies new file mode 100644 index 0000000000..8d6531a174 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power4/fpu/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power4/memcmp.S b/sysdeps/powerpc/powerpc64/power4/memcmp.S index 69caedc9ff..80d67c9aaa 100644 --- a/sysdeps/powerpc/powerpc64/power4/memcmp.S +++ b/sysdeps/powerpc/powerpc64/power4/memcmp.S @@ -1,4 +1,4 @@ -/* Optimized strcmp implementation for PowerPC64. +/* Optimized memcmp implementation for PowerPC64. Copyright (C) 2003-2013 Free Software Foundation, Inc. This file is part of the GNU C Library. @@ -18,13 +18,14 @@ #include <sysdep.h> -/* int [r3] memcmp (const char *s1 [r3], const char *s2 [r4], size_t size [r5]) */ +/* int [r3] memcmp (const char *s1 [r3], + const char *s2 [r4], + size_t size [r5]) */ .machine power4 EALIGN (memcmp, 4, 0) CALL_MCOUNT 3 -#define rTMP r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -35,107 +36,127 @@ EALIGN (memcmp, 4, 0) #define rWORD4 r9 /* next word in s2 */ #define rWORD5 r10 /* next word in s1 */ #define rWORD6 r11 /* next word in s2 */ -#define rBITDIF r12 /* bits that differ in s1 & s2 words */ #define rWORD7 r30 /* next word in s1 */ #define rWORD8 r31 /* next word in s2 */ - xor rTMP, rSTR2, rSTR1 + xor r0, rSTR2, rSTR1 cmpldi cr6, rN, 0 cmpldi cr1, rN, 12 - clrldi. rTMP, rTMP, 61 - clrldi rBITDIF, rSTR1, 61 - cmpldi cr5, rBITDIF, 0 + clrldi. r0, r0, 61 + clrldi r12, rSTR1, 61 + cmpldi cr5, r12, 0 beq- cr6, L(zeroLength) - dcbt 0,rSTR1 - dcbt 0,rSTR2 + dcbt 0, rSTR1 + dcbt 0, rSTR2 /* If less than 8 bytes or not aligned, use the unaligned byte loop. */ blt cr1, L(bytealigned) - std rWORD8,-8(r1) - cfi_offset(rWORD8,-8) - std rWORD7,-16(r1) - cfi_offset(rWORD7,-16) + std rWORD8, -8(r1) + cfi_offset(rWORD8, -8) + std rWORD7, -16(r1) + cfi_offset(rWORD7, -16) bne L(unaligned) /* At this point we know both strings have the same alignment and the - compare length is at least 8 bytes. rBITDIF contains the low order + compare length is at least 8 bytes. r12 contains the low order 3 bits of rSTR1 and cr5 contains the result of the logical compare - of rBITDIF to 0. If rBITDIF == 0 then we are already double word - aligned and can perform the DWaligned loop. + of r12 to 0. If r12 == 0 then we are already double word + aligned and can perform the DW aligned loop. Otherwise we know the two strings have the same alignment (but not - yet DW). So we can force the string addresses to the next lower DW - boundary and special case this first DW word using shift left to + yet DW). So we force the string addresses to the next lower DW + boundary and special case this first DW using shift left to eliminate bits preceding the first byte. Since we want to join the - normal (DWaligned) compare loop, starting at the second double word, + normal (DW aligned) compare loop, starting at the second double word, we need to adjust the length (rN) and special case the loop - versioning for the first DW. This insures that the loop count is - correct and the first DW (shifted) is in the expected resister pair. */ - .align 4 + versioning for the first DW. This ensures that the loop count is + correct and the first DW (shifted) is in the expected register pair. */ + .align 4 L(samealignment): clrrdi rSTR1, rSTR1, 3 clrrdi rSTR2, rSTR2, 3 beq cr5, L(DWaligned) - add rN, rN, rBITDIF - sldi r11, rBITDIF, 3 - srdi rTMP, rN, 5 /* Divide by 32 */ - andi. rBITDIF, rN, 24 /* Get the DW remainder */ + add rN, rN, r12 + sldi rWORD6, r12, 3 + srdi r0, rN, 5 /* Divide by 32 */ + andi. r12, rN, 24 /* Get the DW remainder */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 0(rSTR1) ld rWORD2, 0(rSTR2) - cmpldi cr1, rBITDIF, 16 +#endif + cmpldi cr1, r12, 16 cmpldi cr7, rN, 32 clrldi rN, rN, 61 beq L(dPs4) - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ bgt cr1, L(dPs3) beq cr1, L(dPs2) /* Remainder is 8 */ - .align 3 + .align 3 L(dsP1): - sld rWORD5, rWORD1, r11 - sld rWORD6, rWORD2, r11 + sld rWORD5, rWORD1, rWORD6 + sld rWORD6, rWORD2, rWORD6 cmpld cr5, rWORD5, rWORD6 blt cr7, L(dP1x) /* Do something useful in this cycle since we have to branch anyway. */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 8(rSTR1) ld rWORD2, 8(rSTR2) - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 b L(dP1e) /* Remainder is 16 */ - .align 4 + .align 4 L(dPs2): - sld rWORD5, rWORD1, r11 - sld rWORD6, rWORD2, r11 + sld rWORD5, rWORD1, rWORD6 + sld rWORD6, rWORD2, rWORD6 cmpld cr6, rWORD5, rWORD6 blt cr7, L(dP2x) /* Do something useful in this cycle since we have to branch anyway. */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD7, 8(rSTR1) ld rWORD8, 8(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 b L(dP2e) /* Remainder is 24 */ - .align 4 + .align 4 L(dPs3): - sld rWORD3, rWORD1, r11 - sld rWORD4, rWORD2, r11 + sld rWORD3, rWORD1, rWORD6 + sld rWORD4, rWORD2, rWORD6 cmpld cr1, rWORD3, rWORD4 b L(dP3e) /* Count is a multiple of 32, remainder is 0 */ - .align 4 + .align 4 L(dPs4): - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ - sld rWORD1, rWORD1, r11 - sld rWORD2, rWORD2, r11 - cmpld cr0, rWORD1, rWORD2 + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ + sld rWORD1, rWORD1, rWORD6 + sld rWORD2, rWORD2, rWORD6 + cmpld cr7, rWORD1, rWORD2 b L(dP4e) /* At this point we know both strings are double word aligned and the compare length is at least 8 bytes. */ - .align 4 + .align 4 L(DWaligned): - andi. rBITDIF, rN, 24 /* Get the DW remainder */ - srdi rTMP, rN, 5 /* Divide by 32 */ - cmpldi cr1, rBITDIF, 16 + andi. r12, rN, 24 /* Get the DW remainder */ + srdi r0, rN, 5 /* Divide by 32 */ + cmpldi cr1, r12, 16 cmpldi cr7, rN, 32 clrldi rN, rN, 61 beq L(dP4) @@ -143,174 +164,343 @@ L(DWaligned): beq cr1, L(dP2) /* Remainder is 8 */ - .align 4 + .align 4 L(dP1): - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ /* Normally we'd use rWORD7/rWORD8 here, but since we might exit early (8-15 byte compare), we want to use only volatile registers. This means we can avoid restoring non-volatile registers since we did not change any on the early exit path. The key here is the non-early exit path only cares about the condition code (cr5), not about which register pair was used. */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 0(rSTR1) ld rWORD6, 0(rSTR2) +#endif cmpld cr5, rWORD5, rWORD6 blt cr7, L(dP1x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 8(rSTR1) ld rWORD2, 8(rSTR2) - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 L(dP1e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 16(rSTR1) ld rWORD4, 16(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 24(rSTR1) ld rWORD6, 24(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 - bne cr5, L(dLcr5) - bne cr0, L(dLcr0) + bne cr5, L(dLcr5x) + bne cr7, L(dLcr7x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ldu rWORD7, 32(rSTR1) ldu rWORD8, 32(rSTR2) +#endif bne cr1, L(dLcr1) cmpld cr5, rWORD7, rWORD8 bdnz L(dLoop) bne cr6, L(dLcr6) - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - .align 3 + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) + .align 3 L(dP1x): sldi. r12, rN, 3 - bne cr5, L(dLcr5) + bne cr5, L(dLcr5x) subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ bne L(d00) li rRTN, 0 blr /* Remainder is 16 */ - .align 4 + .align 4 L(dP2): - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 0(rSTR1) ld rWORD6, 0(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 blt cr7, L(dP2x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD7, 8(rSTR1) ld rWORD8, 8(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 L(dP2e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 16(rSTR1) ld rWORD2, 16(rSTR2) - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 24(rSTR1) ld rWORD4, 24(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 8 addi rSTR2, rSTR2, 8 +#endif bne cr6, L(dLcr6) bne cr5, L(dLcr5) b L(dLoop2) /* Again we are on a early exit path (16-23 byte compare), we want to only use volatile registers and avoid restoring non-volatile registers. */ - .align 4 + .align 4 L(dP2x): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 8(rSTR1) ld rWORD4, 8(rSTR2) - cmpld cr5, rWORD3, rWORD4 +#endif + cmpld cr1, rWORD3, rWORD4 sldi. r12, rN, 3 - bne cr6, L(dLcr6) + bne cr6, L(dLcr6x) +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 8 addi rSTR2, rSTR2, 8 - bne cr5, L(dLcr5) +#endif + bne cr1, L(dLcr1x) subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ bne L(d00) li rRTN, 0 blr /* Remainder is 24 */ - .align 4 + .align 4 L(dP3): - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 0(rSTR1) ld rWORD4, 0(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 L(dP3e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 8(rSTR1) ld rWORD6, 8(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 blt cr7, L(dP3x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD7, 16(rSTR1) ld rWORD8, 16(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 24(rSTR1) ld rWORD2, 24(rSTR2) - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 16 addi rSTR2, rSTR2, 16 +#endif bne cr1, L(dLcr1) bne cr6, L(dLcr6) b L(dLoop1) /* Again we are on a early exit path (24-31 byte compare), we want to only use volatile registers and avoid restoring non-volatile registers. */ - .align 4 + .align 4 L(dP3x): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 16(rSTR1) ld rWORD2, 16(rSTR2) - cmpld cr5, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 sldi. r12, rN, 3 - bne cr1, L(dLcr1) + bne cr1, L(dLcr1x) +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 16 addi rSTR2, rSTR2, 16 - bne cr6, L(dLcr6) +#endif + bne cr6, L(dLcr6x) subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ - bne cr5, L(dLcr5) + bne cr7, L(dLcr7x) bne L(d00) li rRTN, 0 blr /* Count is a multiple of 32, remainder is 0 */ - .align 4 + .align 4 L(dP4): - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 0(rSTR1) ld rWORD2, 0(rSTR2) - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 L(dP4e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 8(rSTR1) ld rWORD4, 8(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 16(rSTR1) ld rWORD6, 16(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ldu rWORD7, 24(rSTR1) ldu rWORD8, 24(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 - bne cr0, L(dLcr0) + bne cr7, L(dLcr7) bne cr1, L(dLcr1) bdz- L(d24) /* Adjust CTR as we start with +4 */ /* This is the primary loop */ - .align 4 + .align 4 L(dLoop): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 8(rSTR1) ld rWORD2, 8(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 bne cr6, L(dLcr6) L(dLoop1): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 16(rSTR1) ld rWORD4, 16(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 bne cr5, L(dLcr5) L(dLoop2): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 24(rSTR1) ld rWORD6, 24(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 - bne cr0, L(dLcr0) + bne cr7, L(dLcr7) L(dLoop3): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ldu rWORD7, 32(rSTR1) ldu rWORD8, 32(rSTR2) +#endif bne- cr1, L(dLcr1) - cmpld cr0, rWORD1, rWORD2 + cmpld cr7, rWORD1, rWORD2 bdnz+ L(dLoop) L(dL4): @@ -320,7 +510,7 @@ L(dL4): bne cr5, L(dLcr5) cmpld cr5, rWORD7, rWORD8 L(d44): - bne cr0, L(dLcr0) + bne cr7, L(dLcr7) L(d34): bne cr1, L(dLcr1) L(d24): @@ -329,60 +519,74 @@ L(d14): sldi. r12, rN, 3 bne cr5, L(dLcr5) L(d04): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ beq L(zeroLength) /* At this point we have a remainder of 1 to 7 bytes to compare. Since we are aligned it is safe to load the whole double word, and use shift right double to eliminate bits beyond the compare length. */ L(d00): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 8(rSTR1) ld rWORD2, 8(rSTR2) +#endif srd rWORD1, rWORD1, rN srd rWORD2, rWORD2, rN - cmpld cr5, rWORD1, rWORD2 - bne cr5, L(dLcr5x) + cmpld cr7, rWORD1, rWORD2 + bne cr7, L(dLcr7x) li rRTN, 0 blr - .align 4 -L(dLcr0): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + + .align 4 +L(dLcr7): + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) +L(dLcr7x): li rRTN, 1 - bgtlr cr0 + bgtlr cr7 li rRTN, -1 blr - .align 4 + .align 4 L(dLcr1): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) +L(dLcr1x): li rRTN, 1 bgtlr cr1 li rRTN, -1 blr - .align 4 + .align 4 L(dLcr6): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) +L(dLcr6x): li rRTN, 1 bgtlr cr6 li rRTN, -1 blr - .align 4 + .align 4 L(dLcr5): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) L(dLcr5x): li rRTN, 1 bgtlr cr5 li rRTN, -1 blr - .align 4 + .align 4 L(bytealigned): - mtctr rN /* Power4 wants mtctr 1st in dispatch group */ + mtctr rN /* Power4 wants mtctr 1st in dispatch group */ +#if 0 +/* Huh? We've already branched on cr6! */ beq- cr6, L(zeroLength) +#endif /* We need to prime this loop. This loop is swing modulo scheduled to avoid pipe delays. The dependent instruction latencies (load to @@ -397,7 +601,7 @@ L(bytealigned): lbz rWORD1, 0(rSTR1) lbz rWORD2, 0(rSTR2) bdz- L(b11) - cmpld cr0, rWORD1, rWORD2 + cmpld cr7, rWORD1, rWORD2 lbz rWORD3, 1(rSTR1) lbz rWORD4, 1(rSTR2) bdz- L(b12) @@ -405,11 +609,11 @@ L(bytealigned): lbzu rWORD5, 2(rSTR1) lbzu rWORD6, 2(rSTR2) bdz- L(b13) - .align 4 + .align 4 L(bLoop): lbzu rWORD1, 1(rSTR1) lbzu rWORD2, 1(rSTR2) - bne- cr0, L(bLcr0) + bne- cr7, L(bLcr7) cmpld cr6, rWORD5, rWORD6 bdz- L(b3i) @@ -418,7 +622,7 @@ L(bLoop): lbzu rWORD4, 1(rSTR2) bne- cr1, L(bLcr1) - cmpld cr0, rWORD1, rWORD2 + cmpld cr7, rWORD1, rWORD2 bdz- L(b2i) lbzu rWORD5, 1(rSTR1) @@ -435,23 +639,23 @@ L(bLoop): tested. In this case we must complete the pending operations before returning. */ L(b1i): - bne- cr0, L(bLcr0) + bne- cr7, L(bLcr7) bne- cr1, L(bLcr1) b L(bx56) - .align 4 + .align 4 L(b2i): bne- cr6, L(bLcr6) - bne- cr0, L(bLcr0) + bne- cr7, L(bLcr7) b L(bx34) - .align 4 + .align 4 L(b3i): bne- cr1, L(bLcr1) bne- cr6, L(bLcr6) b L(bx12) - .align 4 -L(bLcr0): + .align 4 +L(bLcr7): li rRTN, 1 - bgtlr cr0 + bgtlr cr7 li rRTN, -1 blr L(bLcr1): @@ -466,14 +670,14 @@ L(bLcr6): blr L(b13): - bne- cr0, L(bx12) + bne- cr7, L(bx12) bne- cr1, L(bx34) L(bx56): sub rRTN, rWORD5, rWORD6 blr nop L(b12): - bne- cr0, L(bx12) + bne- cr7, L(bx12) L(bx34): sub rRTN, rWORD3, rWORD4 blr @@ -481,101 +685,106 @@ L(b11): L(bx12): sub rRTN, rWORD1, rWORD2 blr - .align 4 -L(zeroLengthReturn): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + .align 4 L(zeroLength): li rRTN, 0 blr - .align 4 + .align 4 /* At this point we know the strings have different alignment and the - compare length is at least 8 bytes. rBITDIF contains the low order + compare length is at least 8 bytes. r12 contains the low order 3 bits of rSTR1 and cr5 contains the result of the logical compare - of rBITDIF to 0. If rBITDIF == 0 then rStr1 is double word + of r12 to 0. If r12 == 0 then rStr1 is double word aligned and can perform the DWunaligned loop. Otherwise we know that rSTR1 is not already DW aligned yet. So we can force the string addresses to the next lower DW - boundary and special case this first DW word using shift left to + boundary and special case this first DW using shift left to eliminate bits preceding the first byte. Since we want to join the normal (DWaligned) compare loop, starting at the second double word, we need to adjust the length (rN) and special case the loop - versioning for the first DW. This insures that the loop count is + versioning for the first DW. This ensures that the loop count is correct and the first DW (shifted) is in the expected resister pair. */ -#define rSHL r29 /* Unaligned shift left count. */ -#define rSHR r28 /* Unaligned shift right count. */ -#define rB r27 /* Left rotation temp for rWORD2. */ -#define rD r26 /* Left rotation temp for rWORD4. */ -#define rF r25 /* Left rotation temp for rWORD6. */ -#define rH r24 /* Left rotation temp for rWORD8. */ -#define rA r0 /* Right rotation temp for rWORD2. */ -#define rC r12 /* Right rotation temp for rWORD4. */ -#define rE r0 /* Right rotation temp for rWORD6. */ -#define rG r12 /* Right rotation temp for rWORD8. */ +#define rSHL r29 /* Unaligned shift left count. */ +#define rSHR r28 /* Unaligned shift right count. */ +#define rWORD8_SHIFT r27 /* Left rotation temp for rWORD2. */ +#define rWORD2_SHIFT r26 /* Left rotation temp for rWORD4. */ +#define rWORD4_SHIFT r25 /* Left rotation temp for rWORD6. */ +#define rWORD6_SHIFT r24 /* Left rotation temp for rWORD8. */ L(unaligned): - std r29,-24(r1) - cfi_offset(r29,-24) + std rSHL, -24(r1) + cfi_offset(rSHL, -24) clrldi rSHL, rSTR2, 61 beq- cr6, L(duzeroLength) - std r28,-32(r1) - cfi_offset(r28,-32) + std rSHR, -32(r1) + cfi_offset(rSHR, -32) beq cr5, L(DWunaligned) - std r27,-40(r1) - cfi_offset(r27,-40) -/* Adjust the logical start of rSTR2 ro compensate for the extra bits + std rWORD8_SHIFT, -40(r1) + cfi_offset(rWORD8_SHIFT, -40) +/* Adjust the logical start of rSTR2 to compensate for the extra bits in the 1st rSTR1 DW. */ - sub r27, rSTR2, rBITDIF + sub rWORD8_SHIFT, rSTR2, r12 /* But do not attempt to address the DW before that DW that contains the actual start of rSTR2. */ clrrdi rSTR2, rSTR2, 3 - std r26,-48(r1) - cfi_offset(r26,-48) -/* Compute the left/right shift counts for the unalign rSTR2, + std rWORD2_SHIFT, -48(r1) + cfi_offset(rWORD2_SHIFT, -48) +/* Compute the left/right shift counts for the unaligned rSTR2, compensating for the logical (DW aligned) start of rSTR1. */ - clrldi rSHL, r27, 61 + clrldi rSHL, rWORD8_SHIFT, 61 clrrdi rSTR1, rSTR1, 3 - std r25,-56(r1) - cfi_offset(r25,-56) + std rWORD4_SHIFT, -56(r1) + cfi_offset(rWORD4_SHIFT, -56) sldi rSHL, rSHL, 3 - cmpld cr5, r27, rSTR2 - add rN, rN, rBITDIF - sldi r11, rBITDIF, 3 - std r24,-64(r1) - cfi_offset(r24,-64) + cmpld cr5, rWORD8_SHIFT, rSTR2 + add rN, rN, r12 + sldi rWORD6, r12, 3 + std rWORD6_SHIFT, -64(r1) + cfi_offset(rWORD6_SHIFT, -64) subfic rSHR, rSHL, 64 - srdi rTMP, rN, 5 /* Divide by 32 */ - andi. rBITDIF, rN, 24 /* Get the DW remainder */ + srdi r0, rN, 5 /* Divide by 32 */ + andi. r12, rN, 24 /* Get the DW remainder */ /* We normally need to load 2 DWs to start the unaligned rSTR2, but in this special case those bits may be discarded anyway. Also we must avoid loading a DW where none of the bits are part of rSTR2 as this may cross a page boundary and cause a page fault. */ li rWORD8, 0 blt cr5, L(dus0) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD8, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else ld rWORD8, 0(rSTR2) - la rSTR2, 8(rSTR2) + addi rSTR2, rSTR2, 8 +#endif sld rWORD8, rWORD8, rSHL L(dus0): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 0(rSTR1) ld rWORD2, 0(rSTR2) - cmpldi cr1, rBITDIF, 16 +#endif + cmpldi cr1, r12, 16 cmpldi cr7, rN, 32 - srd rG, rWORD2, rSHR + srd r12, rWORD2, rSHR clrldi rN, rN, 61 beq L(duPs4) - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ - or rWORD8, rG, rWORD8 + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ + or rWORD8, r12, rWORD8 bgt cr1, L(duPs3) beq cr1, L(duPs2) /* Remainder is 8 */ - .align 4 + .align 4 L(dusP1): - sld rB, rWORD2, rSHL - sld rWORD7, rWORD1, r11 - sld rWORD8, rWORD8, r11 + sld rWORD8_SHIFT, rWORD2, rSHL + sld rWORD7, rWORD1, rWORD6 + sld rWORD8, rWORD8, rWORD6 bge cr7, L(duP1e) /* At this point we exit early with the first double word compare complete and remainder of 0 to 7 bytes. See L(du14) for details on @@ -585,95 +794,133 @@ L(dusP1): bne cr5, L(duLcr5) cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA, 0 + li r0, 0 ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else ld rWORD2, 8(rSTR2) - srd rA, rWORD2, rSHR +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Remainder is 16 */ - .align 4 + .align 4 L(duPs2): - sld rH, rWORD2, rSHL - sld rWORD5, rWORD1, r11 - sld rWORD6, rWORD8, r11 + sld rWORD6_SHIFT, rWORD2, rSHL + sld rWORD5, rWORD1, rWORD6 + sld rWORD6, rWORD8, rWORD6 b L(duP2e) /* Remainder is 24 */ - .align 4 + .align 4 L(duPs3): - sld rF, rWORD2, rSHL - sld rWORD3, rWORD1, r11 - sld rWORD4, rWORD8, r11 + sld rWORD4_SHIFT, rWORD2, rSHL + sld rWORD3, rWORD1, rWORD6 + sld rWORD4, rWORD8, rWORD6 b L(duP3e) /* Count is a multiple of 32, remainder is 0 */ - .align 4 + .align 4 L(duPs4): - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ - or rWORD8, rG, rWORD8 - sld rD, rWORD2, rSHL - sld rWORD1, rWORD1, r11 - sld rWORD2, rWORD8, r11 + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ + or rWORD8, r12, rWORD8 + sld rWORD2_SHIFT, rWORD2, rSHL + sld rWORD1, rWORD1, rWORD6 + sld rWORD2, rWORD8, rWORD6 b L(duP4e) /* At this point we know rSTR1 is double word aligned and the compare length is at least 8 bytes. */ - .align 4 + .align 4 L(DWunaligned): - std r27,-40(r1) - cfi_offset(r27,-40) + std rWORD8_SHIFT, -40(r1) + cfi_offset(rWORD8_SHIFT, -40) clrrdi rSTR2, rSTR2, 3 - std r26,-48(r1) - cfi_offset(r26,-48) - srdi rTMP, rN, 5 /* Divide by 32 */ - std r25,-56(r1) - cfi_offset(r25,-56) - andi. rBITDIF, rN, 24 /* Get the DW remainder */ - std r24,-64(r1) - cfi_offset(r24,-64) + std rWORD2_SHIFT, -48(r1) + cfi_offset(rWORD2_SHIFT, -48) + srdi r0, rN, 5 /* Divide by 32 */ + std rWORD4_SHIFT, -56(r1) + cfi_offset(rWORD4_SHIFT, -56) + andi. r12, rN, 24 /* Get the DW remainder */ + std rWORD6_SHIFT, -64(r1) + cfi_offset(rWORD6_SHIFT, -64) sldi rSHL, rSHL, 3 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD6, 0, rSTR2 + addi rSTR2, rSTR2, 8 + ldbrx rWORD8, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else ld rWORD6, 0(rSTR2) ldu rWORD8, 8(rSTR2) - cmpldi cr1, rBITDIF, 16 +#endif + cmpldi cr1, r12, 16 cmpldi cr7, rN, 32 clrldi rN, rN, 61 subfic rSHR, rSHL, 64 - sld rH, rWORD6, rSHL + sld rWORD6_SHIFT, rWORD6, rSHL beq L(duP4) - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ bgt cr1, L(duP3) beq cr1, L(duP2) /* Remainder is 8 */ - .align 4 + .align 4 L(duP1): - srd rG, rWORD8, rSHR + srd r12, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else ld rWORD7, 0(rSTR1) - sld rB, rWORD8, rSHL - or rWORD8, rG, rH +#endif + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT blt cr7, L(duP1x) L(duP1e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 8(rSTR1) ld rWORD2, 8(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 - srd rA, rWORD2, rSHR - sld rD, rWORD2, rSHL - or rWORD2, rA, rB + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 16(rSTR1) ld rWORD4, 16(rSTR2) - cmpld cr0, rWORD1, rWORD2 - srd rC, rWORD4, rSHR - sld rF, rWORD4, rSHL +#endif + cmpld cr7, rWORD1, rWORD2 + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL bne cr5, L(duLcr5) - or rWORD4, rC, rD + or rWORD4, r12, rWORD2_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 24(rSTR1) ld rWORD6, 24(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 - srd rE, rWORD6, rSHR - sld rH, rWORD6, rSHL - bne cr0, L(duLcr0) - or rWORD6, rE, rF + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + bne cr7, L(duLcr7) + or rWORD6, r0, rWORD4_SHIFT cmpld cr6, rWORD5, rWORD6 b L(duLoop3) - .align 4 + .align 4 /* At this point we exit early with the first double word compare complete and remainder of 0 to 7 bytes. See L(du14) for details on how we handle the remaining bytes. */ @@ -683,186 +930,321 @@ L(duP1x): bne cr5, L(duLcr5) cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA, 0 + li r0, 0 ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else ld rWORD2, 8(rSTR2) - srd rA, rWORD2, rSHR +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Remainder is 16 */ - .align 4 + .align 4 L(duP2): - srd rE, rWORD8, rSHR + srd r0, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else ld rWORD5, 0(rSTR1) - or rWORD6, rE, rH - sld rH, rWORD8, rSHL +#endif + or rWORD6, r0, rWORD6_SHIFT + sld rWORD6_SHIFT, rWORD8, rSHL L(duP2e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD7, 8(rSTR1) ld rWORD8, 8(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 - srd rG, rWORD8, rSHR - sld rB, rWORD8, rSHL - or rWORD8, rG, rH + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT blt cr7, L(duP2x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 16(rSTR1) ld rWORD2, 16(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 bne cr6, L(duLcr6) - srd rA, rWORD2, rSHR - sld rD, rWORD2, rSHL - or rWORD2, rA, rB + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 24(rSTR1) ld rWORD4, 24(rSTR2) - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 bne cr5, L(duLcr5) - srd rC, rWORD4, rSHR - sld rF, rWORD4, rSHL - or rWORD4, rC, rD + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL + or rWORD4, r12, rWORD2_SHIFT +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 8 addi rSTR2, rSTR2, 8 +#endif cmpld cr1, rWORD3, rWORD4 b L(duLoop2) - .align 4 + .align 4 L(duP2x): cmpld cr5, rWORD7, rWORD8 +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 8 addi rSTR2, rSTR2, 8 +#endif bne cr6, L(duLcr6) sldi. rN, rN, 3 bne cr5, L(duLcr5) cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA, 0 + li r0, 0 ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else ld rWORD2, 8(rSTR2) - srd rA, rWORD2, rSHR +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Remainder is 24 */ - .align 4 + .align 4 L(duP3): - srd rC, rWORD8, rSHR + srd r12, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else ld rWORD3, 0(rSTR1) - sld rF, rWORD8, rSHL - or rWORD4, rC, rH +#endif + sld rWORD4_SHIFT, rWORD8, rSHL + or rWORD4, r12, rWORD6_SHIFT L(duP3e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 8(rSTR1) ld rWORD6, 8(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 - srd rE, rWORD6, rSHR - sld rH, rWORD6, rSHL - or rWORD6, rE, rF + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + or rWORD6, r0, rWORD4_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD7, 16(rSTR1) ld rWORD8, 16(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 bne cr1, L(duLcr1) - srd rG, rWORD8, rSHR - sld rB, rWORD8, rSHL - or rWORD8, rG, rH + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT blt cr7, L(duP3x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 24(rSTR1) ld rWORD2, 24(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 bne cr6, L(duLcr6) - srd rA, rWORD2, rSHR - sld rD, rWORD2, rSHL - or rWORD2, rA, rB + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 16 addi rSTR2, rSTR2, 16 - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 b L(duLoop1) - .align 4 + .align 4 L(duP3x): +#ifndef __LITTLE_ENDIAN__ addi rSTR1, rSTR1, 16 addi rSTR2, rSTR2, 16 +#endif +#if 0 +/* Huh? We've already branched on cr1! */ bne cr1, L(duLcr1) +#endif cmpld cr5, rWORD7, rWORD8 bne cr6, L(duLcr6) sldi. rN, rN, 3 bne cr5, L(duLcr5) cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA, 0 + li r0, 0 ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else ld rWORD2, 8(rSTR2) - srd rA, rWORD2, rSHR +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Count is a multiple of 32, remainder is 0 */ - .align 4 + .align 4 L(duP4): - mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */ - srd rA, rWORD8, rSHR + mtctr r0 /* Power4 wants mtctr 1st in dispatch group */ + srd r0, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else ld rWORD1, 0(rSTR1) - sld rD, rWORD8, rSHL - or rWORD2, rA, rH +#endif + sld rWORD2_SHIFT, rWORD8, rSHL + or rWORD2, r0, rWORD6_SHIFT L(duP4e): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 8(rSTR1) ld rWORD4, 8(rSTR2) - cmpld cr0, rWORD1, rWORD2 - srd rC, rWORD4, rSHR - sld rF, rWORD4, rSHL - or rWORD4, rC, rD +#endif + cmpld cr7, rWORD1, rWORD2 + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL + or rWORD4, r12, rWORD2_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 16(rSTR1) ld rWORD6, 16(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 - bne cr0, L(duLcr0) - srd rE, rWORD6, rSHR - sld rH, rWORD6, rSHL - or rWORD6, rE, rF + bne cr7, L(duLcr7) + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + or rWORD6, r0, rWORD4_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ldu rWORD7, 24(rSTR1) ldu rWORD8, 24(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 bne cr1, L(duLcr1) - srd rG, rWORD8, rSHR - sld rB, rWORD8, rSHL - or rWORD8, rG, rH + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT cmpld cr5, rWORD7, rWORD8 bdz- L(du24) /* Adjust CTR as we start with +4 */ /* This is the primary loop */ - .align 4 + .align 4 L(duLoop): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD1, 8(rSTR1) ld rWORD2, 8(rSTR2) +#endif cmpld cr1, rWORD3, rWORD4 bne cr6, L(duLcr6) - srd rA, rWORD2, rSHR - sld rD, rWORD2, rSHL - or rWORD2, rA, rB + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT L(duLoop1): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD3, 16(rSTR1) ld rWORD4, 16(rSTR2) +#endif cmpld cr6, rWORD5, rWORD6 bne cr5, L(duLcr5) - srd rC, rWORD4, rSHR - sld rF, rWORD4, rSHL - or rWORD4, rC, rD + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL + or rWORD4, r12, rWORD2_SHIFT L(duLoop2): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ld rWORD5, 24(rSTR1) ld rWORD6, 24(rSTR2) +#endif cmpld cr5, rWORD7, rWORD8 - bne cr0, L(duLcr0) - srd rE, rWORD6, rSHR - sld rH, rWORD6, rSHL - or rWORD6, rE, rF + bne cr7, L(duLcr7) + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + or rWORD6, r0, rWORD4_SHIFT L(duLoop3): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else ldu rWORD7, 32(rSTR1) ldu rWORD8, 32(rSTR2) - cmpld cr0, rWORD1, rWORD2 +#endif + cmpld cr7, rWORD1, rWORD2 bne- cr1, L(duLcr1) - srd rG, rWORD8, rSHR - sld rB, rWORD8, rSHL - or rWORD8, rG, rH + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT bdnz+ L(duLoop) L(duL4): +#if 0 +/* Huh? We've already branched on cr1! */ bne cr1, L(duLcr1) +#endif cmpld cr1, rWORD3, rWORD4 bne cr6, L(duLcr6) cmpld cr6, rWORD5, rWORD6 bne cr5, L(duLcr5) cmpld cr5, rWORD7, rWORD8 L(du44): - bne cr0, L(duLcr0) + bne cr7, L(duLcr7) L(du34): bne cr1, L(duLcr1) L(du24): @@ -872,103 +1254,110 @@ L(du14): bne cr5, L(duLcr5) /* At this point we have a remainder of 1 to 7 bytes to compare. We use shift right double to eliminate bits beyond the compare length. - This allows the use of double word subtract to compute the final - result. However it may not be safe to load rWORD2 which may be beyond the string length. So we compare the bit length of the remainder to the right shift count (rSHR). If the bit count is less than or equal we do not need to load rWORD2 (all significant bits are already in - rB). */ + rWORD8_SHIFT). */ cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA, 0 + li r0, 0 ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else ld rWORD2, 8(rSTR2) - srd rA, rWORD2, rSHR - .align 4 +#endif + srd r0, rWORD2, rSHR + .align 4 L(dutrim): +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 +#else ld rWORD1, 8(rSTR1) - ld rWORD8,-8(r1) +#endif + ld rWORD8, -8(r1) subfic rN, rN, 64 /* Shift count is 64 - (rN * 8). */ - or rWORD2, rA, rB - ld rWORD7,-16(r1) - ld r29,-24(r1) + or rWORD2, r0, rWORD8_SHIFT + ld rWORD7, -16(r1) + ld rSHL, -24(r1) srd rWORD1, rWORD1, rN srd rWORD2, rWORD2, rN - ld r28,-32(r1) - ld r27,-40(r1) + ld rSHR, -32(r1) + ld rWORD8_SHIFT, -40(r1) li rRTN, 0 - cmpld cr0, rWORD1, rWORD2 - ld r26,-48(r1) - ld r25,-56(r1) - beq cr0, L(dureturn24) + cmpld cr7, rWORD1, rWORD2 + ld rWORD2_SHIFT, -48(r1) + ld rWORD4_SHIFT, -56(r1) + beq cr7, L(dureturn24) li rRTN, 1 - ld r24,-64(r1) - bgtlr cr0 + ld rWORD6_SHIFT, -64(r1) + bgtlr cr7 li rRTN, -1 blr - .align 4 -L(duLcr0): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + .align 4 +L(duLcr7): + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) li rRTN, 1 - bgt cr0, L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) + bgt cr7, L(dureturn29) + ld rSHL, -24(r1) + ld rSHR, -32(r1) li rRTN, -1 b L(dureturn27) - .align 4 + .align 4 L(duLcr1): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) li rRTN, 1 bgt cr1, L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) + ld rSHL, -24(r1) + ld rSHR, -32(r1) li rRTN, -1 b L(dureturn27) - .align 4 + .align 4 L(duLcr6): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) li rRTN, 1 bgt cr6, L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) + ld rSHL, -24(r1) + ld rSHR, -32(r1) li rRTN, -1 b L(dureturn27) - .align 4 + .align 4 L(duLcr5): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) li rRTN, 1 bgt cr5, L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) + ld rSHL, -24(r1) + ld rSHR, -32(r1) li rRTN, -1 b L(dureturn27) .align 3 L(duZeroReturn): - li rRTN,0 + li rRTN, 0 .align 4 L(dureturn): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) L(dureturn29): - ld r29,-24(r1) - ld r28,-32(r1) + ld rSHL, -24(r1) + ld rSHR, -32(r1) L(dureturn27): - ld r27,-40(r1) + ld rWORD8_SHIFT, -40(r1) L(dureturn26): - ld r26,-48(r1) + ld rWORD2_SHIFT, -48(r1) L(dureturn25): - ld r25,-56(r1) + ld rWORD4_SHIFT, -56(r1) L(dureturn24): - ld r24,-64(r1) + ld rWORD6_SHIFT, -64(r1) blr L(duzeroLength): - li rRTN,0 + li rRTN, 0 blr END (memcmp) diff --git a/sysdeps/powerpc/powerpc64/power4/memcpy.S b/sysdeps/powerpc/powerpc64/power4/memcpy.S index 4317c7e786..f9a7260dcb 100644 --- a/sysdeps/powerpc/powerpc64/power4/memcpy.S +++ b/sysdeps/powerpc/powerpc64/power4/memcpy.S @@ -214,15 +214,28 @@ EALIGN (memcpy, 5, 0) blt cr6,5f srdi 7,6,16 bgt cr6,3f +#ifdef __LITTLE_ENDIAN__ + sth 7,0(3) +#else sth 6,0(3) +#endif b 7f .align 4 3: +#ifdef __LITTLE_ENDIAN__ + rotlwi 6,6,24 + stb 6,0(3) + sth 7,1(3) +#else stb 7,0(3) sth 6,1(3) +#endif b 7f .align 4 5: +#ifdef __LITTLE_ENDIAN__ + rotlwi 6,6,8 +#endif stb 6,0(3) 7: cmpldi cr1,10,16 @@ -334,13 +347,23 @@ EALIGN (memcpy, 5, 0) bf 30,1f /* there are at least two DWs to copy */ +#ifdef __LITTLE_ENDIAN__ + srd 0,6,10 + sld 8,7,9 +#else sld 0,6,10 srd 8,7,9 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srd 0,7,10 + sld 8,6,9 +#else sld 0,7,10 srd 8,6,9 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -349,8 +372,13 @@ EALIGN (memcpy, 5, 0) blt cr6,8f /* if total DWs = 3, then bypass loop */ bf 31,4f /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srd 0,6,10 + sld 8,7,9 +#else sld 0,6,10 srd 8,7,9 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -361,8 +389,13 @@ EALIGN (memcpy, 5, 0) b 4f .align 4 1: +#ifdef __LITTLE_ENDIAN__ + srd 0,6,10 + sld 8,7,9 +#else sld 0,6,10 srd 8,7,9 +#endif addi 5,5,16 or 0,0,8 bf 31,4f @@ -373,23 +406,44 @@ EALIGN (memcpy, 5, 0) addi 4,4,8 .align 4 /* copy 32 bytes at a time */ -4: sld 0,6,10 +4: +#ifdef __LITTLE_ENDIAN__ + srd 0,6,10 + sld 8,7,9 +#else + sld 0,6,10 srd 8,7,9 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srd 0,7,10 + sld 8,6,9 +#else sld 0,7,10 srd 8,6,9 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srd 0,6,10 + sld 8,7,9 +#else sld 0,6,10 srd 8,7,9 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srd 0,7,10 + sld 8,6,9 +#else sld 0,7,10 srd 8,6,9 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -399,8 +453,13 @@ EALIGN (memcpy, 5, 0) .align 4 8: /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srd 0,6,10 + sld 8,7,9 +#else sld 0,6,10 srd 8,7,9 +#endif or 0,0,8 std 0,0(4) 3: diff --git a/sysdeps/powerpc/powerpc64/power4/memset.S b/sysdeps/powerpc/powerpc64/power4/memset.S index dbecee8b97..9d6fef569a 100644 --- a/sysdeps/powerpc/powerpc64/power4/memset.S +++ b/sysdeps/powerpc/powerpc64/power4/memset.S @@ -50,14 +50,14 @@ L(_memset): /* Align to doubleword boundary. */ cmpldi cr5, rLEN, 31 - rlwimi rCHR, rCHR, 8, 16, 23 /* Replicate byte to halfword. */ + insrdi rCHR, rCHR, 8, 48 /* Replicate byte to halfword. */ beq+ L(aligned2) mtcrf 0x01, rMEMP0 subfic rALIGN, rALIGN, 8 cror 28,30,31 /* Detect odd word aligned. */ add rMEMP, rMEMP, rALIGN sub rLEN, rLEN, rALIGN - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ bt 29, L(g4) /* Process the even word of doubleword. */ bf+ 31, L(g2) @@ -79,14 +79,14 @@ L(g0): /* Handle the case of size < 31. */ L(aligned2): - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ L(aligned): mtcrf 0x01, rLEN ble cr5, L(medium) /* Align to 32-byte boundary. */ andi. rALIGN, rMEMP, 0x18 subfic rALIGN, rALIGN, 0x20 - insrdi rCHR,rCHR,32,0 /* Replicate word to double word. */ + insrdi rCHR, rCHR, 32, 0 /* Replicate word to double word. */ beq L(caligned) mtcrf 0x01, rALIGN add rMEMP, rMEMP, rALIGN @@ -146,24 +146,24 @@ L(zloopstart): L(getCacheAligned): cmpldi cr1,rLEN,32 andi. rTMP,rMEMP,127 - blt cr1,L(handletail32) - beq L(cacheAligned) + blt cr1,L(handletail32) + beq L(cacheAligned) addi rMEMP,rMEMP,32 addi rLEN,rLEN,-32 - std rCHR,-32(rMEMP) - std rCHR,-24(rMEMP) - std rCHR,-16(rMEMP) - std rCHR,-8(rMEMP) - b L(getCacheAligned) + std rCHR,-32(rMEMP) + std rCHR,-24(rMEMP) + std rCHR,-16(rMEMP) + std rCHR,-8(rMEMP) + b L(getCacheAligned) /* Now we are aligned to the cache line and can use dcbz. */ L(cacheAligned): cmpld cr1,rLEN,rCLS - blt cr1,L(handletail32) + blt cr1,L(handletail32) dcbz 0,rMEMP subf rLEN,rCLS,rLEN - add rMEMP,rMEMP,rCLS - b L(cacheAligned) + add rMEMP,rMEMP,rCLS + b L(cacheAligned) /* We are here because the cache line size was set and was not 32-bytes and the remainder (rLEN) is less than the actual cache line size. @@ -200,7 +200,7 @@ L(le4): /* Memset of 0-31 bytes. */ .align 5 L(medium): - insrdi rCHR,rCHR,32,0 /* Replicate word to double word. */ + insrdi rCHR, rCHR, 32, 0 /* Replicate word to double word. */ cmpldi cr1, rLEN, 16 L(medium_tail2): add rMEMP, rMEMP, rLEN @@ -235,6 +235,7 @@ L(medium_28t): END_GEN_TB (memset,TB_TOCLESS) libc_hidden_builtin_def (memset) +#ifndef NO_BZERO_IMPL /* Copied from bzero.S to prevent the linker from inserting a stub between bzero and memset. */ ENTRY (__bzero) @@ -245,3 +246,4 @@ ENTRY (__bzero) END_GEN_TB (__bzero,TB_TOCLESS) weak_alias (__bzero, bzero) +#endif diff --git a/sysdeps/powerpc/powerpc64/power4/multiarch/Implies b/sysdeps/powerpc/powerpc64/power4/multiarch/Implies new file mode 100644 index 0000000000..30edcf7f9d --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power4/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/multiarch diff --git a/sysdeps/powerpc/powerpc64/power4/strncmp.S b/sysdeps/powerpc/powerpc64/power4/strncmp.S index 1276e16a59..5d136cfa21 100644 --- a/sysdeps/powerpc/powerpc64/power4/strncmp.S +++ b/sysdeps/powerpc/powerpc64/power4/strncmp.S @@ -25,7 +25,7 @@ EALIGN (strncmp, 4, 0) CALL_MCOUNT 3 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -38,6 +38,7 @@ EALIGN (strncmp, 4, 0) #define r7F7F r9 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 or rTMP, rSTR2, rSTR1 @@ -79,12 +80,59 @@ L(g1): add rTMP, rFEFE, rWORD1 we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzd rBITDIF, rBITDIF @@ -93,7 +141,7 @@ L(endstring): cmpd cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 blt- cr1, L(equal) - sradi rRTN, rRTN, 63 + sradi rRTN, rRTN, 63 /* must return an int. */ ori rRTN, rRTN, 1 blr L(equal): @@ -101,7 +149,7 @@ L(equal): blr L(different): - ldu rWORD1, -8(rSTR1) + ld rWORD1, -8(rSTR1) xor. rBITDIF, rWORD1, rWORD2 sub rRTN, rWORD1, rWORD2 blt- L(highbit) @@ -109,11 +157,10 @@ L(different): ori rRTN, rRTN, 1 blr L(highbit): - srdi rWORD2, rWORD2, 56 - srdi rWORD1, rWORD1, 56 - sub rRTN, rWORD1, rWORD2 + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 diff --git a/sysdeps/powerpc/powerpc64/power4/wordcopy.c b/sysdeps/powerpc/powerpc64/power4/wordcopy.c deleted file mode 100644 index f427b48e7a..0000000000 --- a/sysdeps/powerpc/powerpc64/power4/wordcopy.c +++ /dev/null @@ -1 +0,0 @@ -#include "../../powerpc32/power4/wordcopy.c" diff --git a/sysdeps/powerpc/powerpc64/power5+/fpu/Implies b/sysdeps/powerpc/powerpc64/power5+/fpu/Implies new file mode 100644 index 0000000000..c0e67848e2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power5+/fpu/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power5/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power5+/fpu/multiarch/Implies b/sysdeps/powerpc/powerpc64/power5+/fpu/multiarch/Implies new file mode 100644 index 0000000000..c0e67848e2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power5+/fpu/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power5/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power5+/multiarch/Implies b/sysdeps/powerpc/powerpc64/power5+/multiarch/Implies new file mode 100644 index 0000000000..0851b19fa2 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power5+/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power5/multiarch diff --git a/sysdeps/powerpc/powerpc64/power5/Implies b/sysdeps/powerpc/powerpc64/power5/Implies index b36831e287..bedb20b65c 100644 --- a/sysdeps/powerpc/powerpc64/power5/Implies +++ b/sysdeps/powerpc/powerpc64/power5/Implies @@ -1,4 +1,2 @@ -powerpc/power5/fpu -powerpc/power5 powerpc/powerpc64/power4/fpu powerpc/powerpc64/power4 diff --git a/sysdeps/powerpc/powerpc64/power5/fpu/Implies b/sysdeps/powerpc/powerpc64/power5/fpu/Implies new file mode 100644 index 0000000000..3740d050a6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power5/fpu/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power4/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power5/fpu/multiarch/Implies b/sysdeps/powerpc/powerpc64/power5/fpu/multiarch/Implies new file mode 100644 index 0000000000..3740d050a6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power5/fpu/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power4/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power5/multiarch/Implies b/sysdeps/powerpc/powerpc64/power5/multiarch/Implies new file mode 100644 index 0000000000..9a3cbb0938 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power5/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power4/multiarch diff --git a/sysdeps/powerpc/powerpc64/power6/fpu/Implies b/sysdeps/powerpc/powerpc64/power6/fpu/Implies new file mode 100644 index 0000000000..f09854edb6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power6/fpu/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power5+/fpu diff --git a/sysdeps/powerpc/powerpc64/power6/fpu/multiarch/Implies b/sysdeps/powerpc/powerpc64/power6/fpu/multiarch/Implies new file mode 100644 index 0000000000..fca8a4ef0f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power6/fpu/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power5+/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power6/memcpy.S b/sysdeps/powerpc/powerpc64/power6/memcpy.S index d6d242d293..e3f3d8a303 100644 --- a/sysdeps/powerpc/powerpc64/power6/memcpy.S +++ b/sysdeps/powerpc/powerpc64/power6/memcpy.S @@ -400,15 +400,28 @@ L(das_tail2): blt cr6,5f srdi 7,6,16 bgt cr6,3f +#ifdef __LITTLE_ENDIAN__ + sth 7,0(3) +#else sth 6,0(3) +#endif b 7f .align 4 3: +#ifdef __LITTLE_ENDIAN__ + rotlwi 6,6,24 + stb 6,0(3) + sth 7,1(3) +#else stb 7,0(3) sth 6,1(3) +#endif b 7f .align 4 5: +#ifdef __LITTLE_ENDIAN__ + rotlwi 6,6,8 +#endif stb 6,0(3) 7: cmpldi cr1,10,16 @@ -595,13 +608,24 @@ L(du1_do): bf 30,L(du1_1dw) /* there are at least two DWs to copy */ + /* FIXME: can combine last shift and "or" into "rldimi" */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 8 + sldi 8,7, 64-8 +#else sldi 0,6, 8 srdi 8,7, 64-8 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 8 + sldi 8,6, 64-8 +#else sldi 0,7, 8 srdi 8,6, 64-8 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -610,8 +634,13 @@ L(du1_do): blt cr6,L(du1_fini) /* if total DWs = 3, then bypass loop */ bf 31,L(du1_loop) /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 8 + sldi 8,7, 64-8 +#else sldi 0,6, 8 srdi 8,7, 64-8 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -622,8 +651,13 @@ L(du1_do): b L(du1_loop) .align 4 L(du1_1dw): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 8 + sldi 8,7, 64-8 +#else sldi 0,6, 8 srdi 8,7, 64-8 +#endif addi 5,5,16 or 0,0,8 bf 31,L(du1_loop) @@ -635,23 +669,43 @@ L(du1_1dw): .align 4 /* copy 32 bytes at a time */ L(du1_loop): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 8 + sldi 8,7, 64-8 +#else sldi 0,6, 8 srdi 8,7, 64-8 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 8 + sldi 8,6, 64-8 +#else sldi 0,7, 8 srdi 8,6, 64-8 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 8 + sldi 8,7, 64-8 +#else sldi 0,6, 8 srdi 8,7, 64-8 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 8 + sldi 8,6, 64-8 +#else sldi 0,7, 8 srdi 8,6, 64-8 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -661,8 +715,13 @@ L(du1_loop): .align 4 L(du1_fini): /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 8 + sldi 8,7, 64-8 +#else sldi 0,6, 8 srdi 8,7, 64-8 +#endif or 0,0,8 std 0,0(4) b L(du_done) @@ -672,13 +731,23 @@ L(du2_do): bf 30,L(du2_1dw) /* there are at least two DWs to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 16 + sldi 8,7, 64-16 +#else sldi 0,6, 16 srdi 8,7, 64-16 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 16 + sldi 8,6, 64-16 +#else sldi 0,7, 16 srdi 8,6, 64-16 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -687,8 +756,13 @@ L(du2_do): blt cr6,L(du2_fini) /* if total DWs = 3, then bypass loop */ bf 31,L(du2_loop) /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 16 + sldi 8,7, 64-16 +#else sldi 0,6, 16 srdi 8,7, 64-16 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -699,8 +773,13 @@ L(du2_do): b L(du2_loop) .align 4 L(du2_1dw): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 16 + sldi 8,7, 64-16 +#else sldi 0,6, 16 srdi 8,7, 64-16 +#endif addi 5,5,16 or 0,0,8 bf 31,L(du2_loop) @@ -712,23 +791,43 @@ L(du2_1dw): .align 4 /* copy 32 bytes at a time */ L(du2_loop): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 16 + sldi 8,7, 64-16 +#else sldi 0,6, 16 srdi 8,7, 64-16 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 16 + sldi 8,6, 64-16 +#else sldi 0,7, 16 srdi 8,6, 64-16 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 16 + sldi 8,7, 64-16 +#else sldi 0,6, 16 srdi 8,7, 64-16 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 16 + sldi 8,6, 64-16 +#else sldi 0,7, 16 srdi 8,6, 64-16 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -738,8 +837,13 @@ L(du2_loop): .align 4 L(du2_fini): /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 16 + sldi 8,7, 64-16 +#else sldi 0,6, 16 srdi 8,7, 64-16 +#endif or 0,0,8 std 0,0(4) b L(du_done) @@ -749,13 +853,23 @@ L(du3_do): bf 30,L(du3_1dw) /* there are at least two DWs to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 24 + sldi 8,7, 64-24 +#else sldi 0,6, 24 srdi 8,7, 64-24 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 24 + sldi 8,6, 64-24 +#else sldi 0,7, 24 srdi 8,6, 64-24 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -764,8 +878,13 @@ L(du3_do): blt cr6,L(du3_fini) /* if total DWs = 3, then bypass loop */ bf 31,L(du3_loop) /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 24 + sldi 8,7, 64-24 +#else sldi 0,6, 24 srdi 8,7, 64-24 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -776,8 +895,13 @@ L(du3_do): b L(du3_loop) .align 4 L(du3_1dw): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 24 + sldi 8,7, 64-24 +#else sldi 0,6, 24 srdi 8,7, 64-24 +#endif addi 5,5,16 or 0,0,8 bf 31,L(du3_loop) @@ -789,23 +913,43 @@ L(du3_1dw): .align 4 /* copy 32 bytes at a time */ L(du3_loop): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 24 + sldi 8,7, 64-24 +#else sldi 0,6, 24 srdi 8,7, 64-24 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 24 + sldi 8,6, 64-24 +#else sldi 0,7, 24 srdi 8,6, 64-24 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 24 + sldi 8,7, 64-24 +#else sldi 0,6, 24 srdi 8,7, 64-24 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 24 + sldi 8,6, 64-24 +#else sldi 0,7, 24 srdi 8,6, 64-24 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -815,8 +959,13 @@ L(du3_loop): .align 4 L(du3_fini): /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 24 + sldi 8,7, 64-24 +#else sldi 0,6, 24 srdi 8,7, 64-24 +#endif or 0,0,8 std 0,0(4) b L(du_done) @@ -832,13 +981,23 @@ L(du4_dox): bf 30,L(du4_1dw) /* there are at least two DWs to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 32 + sldi 8,7, 64-32 +#else sldi 0,6, 32 srdi 8,7, 64-32 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 32 + sldi 8,6, 64-32 +#else sldi 0,7, 32 srdi 8,6, 64-32 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -847,8 +1006,13 @@ L(du4_dox): blt cr6,L(du4_fini) /* if total DWs = 3, then bypass loop */ bf 31,L(du4_loop) /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 32 + sldi 8,7, 64-32 +#else sldi 0,6, 32 srdi 8,7, 64-32 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -859,8 +1023,13 @@ L(du4_dox): b L(du4_loop) .align 4 L(du4_1dw): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 32 + sldi 8,7, 64-32 +#else sldi 0,6, 32 srdi 8,7, 64-32 +#endif addi 5,5,16 or 0,0,8 bf 31,L(du4_loop) @@ -872,23 +1041,43 @@ L(du4_1dw): .align 4 /* copy 32 bytes at a time */ L(du4_loop): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 32 + sldi 8,7, 64-32 +#else sldi 0,6, 32 srdi 8,7, 64-32 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 32 + sldi 8,6, 64-32 +#else sldi 0,7, 32 srdi 8,6, 64-32 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 32 + sldi 8,7, 64-32 +#else sldi 0,6, 32 srdi 8,7, 64-32 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 32 + sldi 8,6, 64-32 +#else sldi 0,7, 32 srdi 8,6, 64-32 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -898,8 +1087,13 @@ L(du4_loop): .align 4 L(du4_fini): /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 32 + sldi 8,7, 64-32 +#else sldi 0,6, 32 srdi 8,7, 64-32 +#endif or 0,0,8 std 0,0(4) b L(du_done) @@ -909,13 +1103,23 @@ L(du5_do): bf 30,L(du5_1dw) /* there are at least two DWs to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 40 + sldi 8,7, 64-40 +#else sldi 0,6, 40 srdi 8,7, 64-40 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 40 + sldi 8,6, 64-40 +#else sldi 0,7, 40 srdi 8,6, 64-40 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -924,8 +1128,13 @@ L(du5_do): blt cr6,L(du5_fini) /* if total DWs = 3, then bypass loop */ bf 31,L(du5_loop) /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 40 + sldi 8,7, 64-40 +#else sldi 0,6, 40 srdi 8,7, 64-40 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -936,8 +1145,13 @@ L(du5_do): b L(du5_loop) .align 4 L(du5_1dw): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 40 + sldi 8,7, 64-40 +#else sldi 0,6, 40 srdi 8,7, 64-40 +#endif addi 5,5,16 or 0,0,8 bf 31,L(du5_loop) @@ -949,23 +1163,43 @@ L(du5_1dw): .align 4 /* copy 32 bytes at a time */ L(du5_loop): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 40 + sldi 8,7, 64-40 +#else sldi 0,6, 40 srdi 8,7, 64-40 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 40 + sldi 8,6, 64-40 +#else sldi 0,7, 40 srdi 8,6, 64-40 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 40 + sldi 8,7, 64-40 +#else sldi 0,6, 40 srdi 8,7, 64-40 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 40 + sldi 8,6, 64-40 +#else sldi 0,7, 40 srdi 8,6, 64-40 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -975,8 +1209,13 @@ L(du5_loop): .align 4 L(du5_fini): /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 40 + sldi 8,7, 64-40 +#else sldi 0,6, 40 srdi 8,7, 64-40 +#endif or 0,0,8 std 0,0(4) b L(du_done) @@ -986,13 +1225,23 @@ L(du6_do): bf 30,L(du6_1dw) /* there are at least two DWs to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 48 + sldi 8,7, 64-48 +#else sldi 0,6, 48 srdi 8,7, 64-48 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 48 + sldi 8,6, 64-48 +#else sldi 0,7, 48 srdi 8,6, 64-48 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -1001,8 +1250,13 @@ L(du6_do): blt cr6,L(du6_fini) /* if total DWs = 3, then bypass loop */ bf 31,L(du6_loop) /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 48 + sldi 8,7, 64-48 +#else sldi 0,6, 48 srdi 8,7, 64-48 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -1013,8 +1267,13 @@ L(du6_do): b L(du6_loop) .align 4 L(du6_1dw): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 48 + sldi 8,7, 64-48 +#else sldi 0,6, 48 srdi 8,7, 64-48 +#endif addi 5,5,16 or 0,0,8 bf 31,L(du6_loop) @@ -1026,23 +1285,43 @@ L(du6_1dw): .align 4 /* copy 32 bytes at a time */ L(du6_loop): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 48 + sldi 8,7, 64-48 +#else sldi 0,6, 48 srdi 8,7, 64-48 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 48 + sldi 8,6, 64-48 +#else sldi 0,7, 48 srdi 8,6, 64-48 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 48 + sldi 8,7, 64-48 +#else sldi 0,6, 48 srdi 8,7, 64-48 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 48 + sldi 8,6, 64-48 +#else sldi 0,7, 48 srdi 8,6, 64-48 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -1052,8 +1331,13 @@ L(du6_loop): .align 4 L(du6_fini): /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 48 + sldi 8,7, 64-48 +#else sldi 0,6, 48 srdi 8,7, 64-48 +#endif or 0,0,8 std 0,0(4) b L(du_done) @@ -1063,13 +1347,23 @@ L(du7_do): bf 30,L(du7_1dw) /* there are at least two DWs to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 56 + sldi 8,7, 64-56 +#else sldi 0,6, 56 srdi 8,7, 64-56 +#endif or 0,0,8 ld 6,16(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 56 + sldi 8,6, 64-56 +#else sldi 0,7, 56 srdi 8,6, 64-56 +#endif or 0,0,8 ld 7,24(5) std 0,8(4) @@ -1078,8 +1372,13 @@ L(du7_do): blt cr6,L(du7_fini) /* if total DWs = 3, then bypass loop */ bf 31,L(du7_loop) /* there is a third DW to copy */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 56 + sldi 8,7, 64-56 +#else sldi 0,6, 56 srdi 8,7, 64-56 +#endif or 0,0,8 std 0,0(4) mr 6,7 @@ -1090,8 +1389,13 @@ L(du7_do): b L(du7_loop) .align 4 L(du7_1dw): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 56 + sldi 8,7, 64-56 +#else sldi 0,6, 56 srdi 8,7, 64-56 +#endif addi 5,5,16 or 0,0,8 bf 31,L(du7_loop) @@ -1103,23 +1407,43 @@ L(du7_1dw): .align 4 /* copy 32 bytes at a time */ L(du7_loop): +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 56 + sldi 8,7, 64-56 +#else sldi 0,6, 56 srdi 8,7, 64-56 +#endif or 0,0,8 ld 6,0(5) std 0,0(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 56 + sldi 8,6, 64-56 +#else sldi 0,7, 56 srdi 8,6, 64-56 +#endif or 0,0,8 ld 7,8(5) std 0,8(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 56 + sldi 8,7, 64-56 +#else sldi 0,6, 56 srdi 8,7, 64-56 +#endif or 0,0,8 ld 6,16(5) std 0,16(4) +#ifdef __LITTLE_ENDIAN__ + srdi 0,7, 56 + sldi 8,6, 64-56 +#else sldi 0,7, 56 srdi 8,6, 64-56 +#endif or 0,0,8 ld 7,24(5) std 0,24(4) @@ -1129,8 +1453,13 @@ L(du7_loop): .align 4 L(du7_fini): /* calculate and store the final DW */ +#ifdef __LITTLE_ENDIAN__ + srdi 0,6, 56 + sldi 8,7, 64-56 +#else sldi 0,6, 56 srdi 8,7, 64-56 +#endif or 0,0,8 std 0,0(4) b L(du_done) diff --git a/sysdeps/powerpc/powerpc64/power6/memset.S b/sysdeps/powerpc/powerpc64/power6/memset.S index 3e8ae2d25e..15a83b74fa 100644 --- a/sysdeps/powerpc/powerpc64/power6/memset.S +++ b/sysdeps/powerpc/powerpc64/power6/memset.S @@ -47,14 +47,14 @@ L(_memset): /* Align to doubleword boundary. */ cmpldi cr5, rLEN, 31 - rlwimi rCHR, rCHR, 8, 16, 23 /* Replicate byte to halfword. */ + insrdi rCHR, rCHR, 8, 48 /* Replicate byte to halfword. */ beq+ L(aligned2) mtcrf 0x01, rMEMP0 subfic rALIGN, rALIGN, 8 cror 28,30,31 /* Detect odd word aligned. */ add rMEMP, rMEMP, rALIGN sub rLEN, rLEN, rALIGN - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ bt 29, L(g4) /* Process the even word of doubleword. */ bf+ 31, L(g2) @@ -76,14 +76,14 @@ L(g0): /* Handle the case of size < 31. */ L(aligned2): - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ L(aligned): mtcrf 0x01, rLEN ble cr5, L(medium) /* Align to 32-byte boundary. */ andi. rALIGN, rMEMP, 0x18 subfic rALIGN, rALIGN, 0x20 - insrdi rCHR,rCHR,32,0 /* Replicate word to double word. */ + insrdi rCHR, rCHR, 32, 0 /* Replicate word to double word. */ beq L(caligned) mtcrf 0x01, rALIGN add rMEMP, rMEMP, rALIGN @@ -344,7 +344,7 @@ L(le4): /* Memset of 0-31 bytes. */ .align 5 L(medium): - insrdi rCHR,rCHR,32,0 /* Replicate word to double word. */ + insrdi rCHR, rCHR, 32, 0 /* Replicate word to double word. */ cmpldi cr1, rLEN, 16 L(medium_tail2): add rMEMP, rMEMP, rLEN @@ -379,6 +379,7 @@ L(medium_28t): END_GEN_TB (memset,TB_TOCLESS) libc_hidden_builtin_def (memset) +#ifndef NO_BZERO_IMPL /* Copied from bzero.S to prevent the linker from inserting a stub between bzero and memset. */ ENTRY (__bzero) @@ -389,3 +390,4 @@ ENTRY (__bzero) END_GEN_TB (__bzero,TB_TOCLESS) weak_alias (__bzero, bzero) +#endif diff --git a/sysdeps/powerpc/powerpc64/power6/multiarch/Implies b/sysdeps/powerpc/powerpc64/power6/multiarch/Implies new file mode 100644 index 0000000000..2ebe304fa6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power6/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power5+/multiarch diff --git a/sysdeps/powerpc/powerpc64/power6/wcschr.c b/sysdeps/powerpc/powerpc64/power6/wcschr.c index 9136c02cb5..ae04a130cc 100644 --- a/sysdeps/powerpc/powerpc64/power6/wcschr.c +++ b/sysdeps/powerpc/powerpc64/power6/wcschr.c @@ -1 +1 @@ -#include "../../powerpc32/power6/wcschr.c" +#include <sysdeps/powerpc/power6/wcschr.c> diff --git a/sysdeps/powerpc/powerpc64/power6/wcscpy.c b/sysdeps/powerpc/powerpc64/power6/wcscpy.c index 57b706ab99..722c8f995b 100644 --- a/sysdeps/powerpc/powerpc64/power6/wcscpy.c +++ b/sysdeps/powerpc/powerpc64/power6/wcscpy.c @@ -1 +1 @@ -#include "../../powerpc32/power6/wcscpy.c" +#include <sysdeps/powerpc/power6/wcscpy.c> diff --git a/sysdeps/powerpc/powerpc64/power6/wcsrchr.c b/sysdeps/powerpc/powerpc64/power6/wcsrchr.c index 2327c05b04..b86472d7bd 100644 --- a/sysdeps/powerpc/powerpc64/power6/wcsrchr.c +++ b/sysdeps/powerpc/powerpc64/power6/wcsrchr.c @@ -1 +1 @@ -#include "../../powerpc32/power6/wcsrchr.c" +#include <sysdeps/powerpc/power6/wcsrchr.c> diff --git a/sysdeps/powerpc/powerpc64/power6/wordcopy.c b/sysdeps/powerpc/powerpc64/power6/wordcopy.c deleted file mode 100644 index 751789339d..0000000000 --- a/sysdeps/powerpc/powerpc64/power6/wordcopy.c +++ /dev/null @@ -1,217 +0,0 @@ -/* _memcopy.c -- subroutines for memory copy functions. - Copyright (C) 1991-2013 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Torbjorn Granlund (tege@sics.se). - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library; if not, see - <http://www.gnu.org/licenses/>. */ - -/* BE VERY CAREFUL IF YOU CHANGE THIS CODE...! */ - -#include <stddef.h> -#include <memcopy.h> - -/* _wordcopy_fwd_aligned -- Copy block beginning at SRCP to - block beginning at DSTP with LEN `op_t' words (not LEN bytes!). - Both SRCP and DSTP should be aligned for memory operations on `op_t's. */ - -void -_wordcopy_fwd_aligned (dstp, srcp, len) - long int dstp; - long int srcp; - size_t len; -{ - op_t a0, a1; - - if (len & 1) - { - ((op_t *) dstp)[0] = ((op_t *) srcp)[0]; - - if (len == 1) - return; - srcp += OPSIZ; - dstp += OPSIZ; - len -= 1; - } - - do - { - a0 = ((op_t *) srcp)[0]; - a1 = ((op_t *) srcp)[1]; - ((op_t *) dstp)[0] = a0; - ((op_t *) dstp)[1] = a1; - - srcp += 2 * OPSIZ; - dstp += 2 * OPSIZ; - len -= 2; - } - while (len != 0); -} - -#define fwd_align_merge(align) \ - do \ - { \ - a1 = ((op_t *) srcp)[1]; \ - a2 = ((op_t *) srcp)[2]; \ - ((op_t *) dstp)[0] = MERGE (a0, align*8, a1, (64-align*8)); \ - ((op_t *) dstp)[1] = MERGE (a1, align*8, a2, (64-align*8)); \ - a0 = a2; \ - srcp += 2 * OPSIZ; \ - dstp += 2 * OPSIZ; \ - len -= 2; \ - } \ - while (len != 0) - - -/* _wordcopy_fwd_dest_aligned -- Copy block beginning at SRCP to - block beginning at DSTP with LEN `op_t' words (not LEN bytes!). - DSTP should be aligned for memory operations on `op_t's, but SRCP must - *not* be aligned. */ - -void -_wordcopy_fwd_dest_aligned (dstp, srcp, len) - long int dstp; - long int srcp; - size_t len; -{ - op_t a0, a1, a2; - int sh_1, sh_2; - int align; - - /* Calculate how to shift a word read at the memory operation - aligned srcp to make it aligned for copy. */ - - align = srcp % OPSIZ; - sh_1 = 8 * (srcp % OPSIZ); - sh_2 = 8 * OPSIZ - sh_1; - - /* Make SRCP aligned by rounding it down to the beginning of the `op_t' - it points in the middle of. */ - srcp &= -OPSIZ; - a0 = ((op_t *) srcp)[0]; - - if (len & 1) - { - a1 = ((op_t *) srcp)[1]; - ((op_t *) dstp)[0] = MERGE (a0, sh_1, a1, sh_2); - - if (len == 1) - return; - - a0 = a1; - srcp += OPSIZ; - dstp += OPSIZ; - len -= 1; - } - - fwd_align_merge (align); - -} - -/* _wordcopy_bwd_aligned -- Copy block finishing right before - SRCP to block finishing right before DSTP with LEN `op_t' words - (not LEN bytes!). Both SRCP and DSTP should be aligned for memory - operations on `op_t's. */ - -void -_wordcopy_bwd_aligned (dstp, srcp, len) - long int dstp; - long int srcp; - size_t len; -{ - op_t a0, a1; - - if (len & 1) - { - srcp -= OPSIZ; - dstp -= OPSIZ; - ((op_t *) dstp)[0] = ((op_t *) srcp)[0]; - - if (len == 1) - return; - len -= 1; - } - - do - { - srcp -= 2 * OPSIZ; - dstp -= 2 * OPSIZ; - - a1 = ((op_t *) srcp)[1]; - a0 = ((op_t *) srcp)[0]; - ((op_t *) dstp)[1] = a1; - ((op_t *) dstp)[0] = a0; - - len -= 2; - } - while (len != 0); -} - -#define bwd_align_merge(align) \ - do \ - { \ - srcp -= 2 * OPSIZ; \ - dstp -= 2 * OPSIZ; \ - a1 = ((op_t *) srcp)[1]; \ - a0 = ((op_t *) srcp)[0]; \ - ((op_t *) dstp)[1] = MERGE (a1, align*8, a2, (64-align*8)); \ - ((op_t *) dstp)[0] = MERGE (a0, align*8, a1, (64-align*8)); \ - a2 = a0; \ - len -= 2; \ - } \ - while (len != 0) - -/* _wordcopy_bwd_dest_aligned -- Copy block finishing right - before SRCP to block finishing right before DSTP with LEN `op_t' - words (not LEN bytes!). DSTP should be aligned for memory - operations on `op_t', but SRCP must *not* be aligned. */ - -void -_wordcopy_bwd_dest_aligned (dstp, srcp, len) - long int dstp; - long int srcp; - size_t len; -{ - op_t a0, a1, a2; - int sh_1, sh_2; - int align; - - /* Calculate how to shift a word read at the memory operation - aligned srcp to make it aligned for copy. */ - - align = srcp % OPSIZ; - sh_1 = 8 * (srcp % OPSIZ); - sh_2 = 8 * OPSIZ - sh_1; - - /* Make srcp aligned by rounding it down to the beginning of the op_t - it points in the middle of. */ - srcp &= -OPSIZ; - a2 = ((op_t *) srcp)[0]; - - if (len & 1) - { - srcp -= OPSIZ; - dstp -= OPSIZ; - a1 = ((op_t *) srcp)[0]; - ((op_t *) dstp)[0] = MERGE (a1, sh_1, a2, sh_2); - - if (len == 1) - return; - - a2 = a1; - len -= 1; - } - - bwd_align_merge (align); -} diff --git a/sysdeps/powerpc/powerpc64/power6x/fpu/multiarch/Implies b/sysdeps/powerpc/powerpc64/power6x/fpu/multiarch/Implies new file mode 100644 index 0000000000..f54ff23500 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power6x/fpu/multiarch/Implies @@ -0,0 +1 @@ +sysdeps/powerpc/powerpc64/power6/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power7/add_n.S b/sysdeps/powerpc/powerpc64/power7/add_n.S new file mode 100644 index 0000000000..d90e4fce0d --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/add_n.S @@ -0,0 +1,98 @@ +/* PowerPC64 mpn_lshift -- mpn_add_n/mpn_sub_n -- mpn addition and + subtraction. + Copyright (C) 2003-2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +/* cycles/limb + * POWER7 2.18 + */ + +#ifdef USE_AS_SUB +# define FUNC __mpn_sub_n +# define ADDSUBC subfe +#else +# define FUNC __mpn_add_n +# define ADDSUBC adde +#endif + +#define RP r3 +#define UP r4 +#define VP r5 +#define N r6 + +EALIGN(FUNC, 5, 0) +#ifdef USE_AS_SUB + addic r0, r0, 0 +#else + addic r0, r1, -1 +#endif + andi. r7, N, 1 + beq L(bx0) + + ld r7, 0(UP) + ld r9, r0(VP) + ADDSUBC r11, r9, r7 + std r11, r0(RP) + cmpldi N, N, 1 + beq N, L(end) + addi UP, UP, 8 + addi VP, VP, 8 + addi RP, RP, 8 + +L(bx0): addi r0, N, 2 + srdi r0, r0, 2 + mtctr r0 + + andi. r7, N, 2 + bne L(mid) + + addi UP, UP, 16 + addi VP, VP, 16 + addi RP, RP, 16 + + .align 5 +L(top): ld r6, -16(UP) + ld r7, -8(UP) + ld r8, -16(VP) + ld r9, -8(VP) + ADDSUBC r10, r8, N + ADDSUBC r11, r9, r7 + std r10, -16(RP) + std r11, -8(RP) +L(mid): ld r6, 0(UP) + ld r7, 8(UP) + ld r8, 0(VP) + ld r9, 8(VP) + ADDSUBC r10, r8, N + ADDSUBC r11, r9, r7 + std r10, 0(RP) + std r11, 8(RP) + addi UP, UP, 32 + addi VP, VP, 32 + addi RP, RP, 32 + bdnz L(top) + +L(end): subfe r3, r0, r0 +#ifdef USE_AS_SUB + neg r3, r3 +#else + addi r3, r3, 1 +#endif + blr +END(FUNC) diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/Implies b/sysdeps/powerpc/powerpc64/power7/fpu/Implies new file mode 100644 index 0000000000..410d289a6d --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/fpu/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power6/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/multiarch/Implies b/sysdeps/powerpc/powerpc64/power7/fpu/multiarch/Implies new file mode 100644 index 0000000000..410d289a6d --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/fpu/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power6/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S b/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S index d0071c7658..ebec0e0bad 100644 --- a/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S +++ b/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S @@ -39,10 +39,8 @@ EALIGN (__finite, 4, 0) stfd fp1,-16(r1) /* Transfer FP to GPR's. */ ori 2,2,0 /* Force a new dispatch group. */ - - lhz r4,-16(r1) /* Fetch the upper portion of the high word of - the FP value (where the exponent and sign bits - are). */ + lhz r4,-16+HISHORT(r1) /* Fetch the upper 16 bits of the FP value + (biased exponent and sign bit). */ clrlwi r4,r4,17 /* r4 = abs(r4). */ cmpwi cr7,r4,0x7ff0 /* r4 == 0x7ff0? */ bltlr cr7 /* LT means finite, other non-finite. */ diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S b/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S index 1aea12383f..8d088db5af 100644 --- a/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S +++ b/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S @@ -38,9 +38,8 @@ EALIGN (__isinf, 4, 0) stfd fp1,-16(r1) /* Transfer FP to GPR's. */ ori 2,2,0 /* Force a new dispatch group. */ - lhz r4,-16(r1) /* Fetch the upper portion of the high word of - the FP value (where the exponent and sign bits - are). */ + lhz r4,-16+HISHORT(r1) /* Fetch the upper 16 bits of the FP value + (biased exponent and sign bit). */ cmpwi cr7,r4,0x7ff0 /* r4 == 0x7ff0? */ li r3,1 beqlr cr7 /* EQ means INF, otherwise -INF. */ diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/s_logb.c b/sysdeps/powerpc/powerpc64/power7/fpu/s_logb.c index ff3a9e0c77..2599c771d9 100644 --- a/sysdeps/powerpc/powerpc64/power7/fpu/s_logb.c +++ b/sysdeps/powerpc/powerpc64/power7/fpu/s_logb.c @@ -1 +1 @@ -#include <sysdeps/powerpc/powerpc32/power7/fpu/s_logb.c> +#include <sysdeps/powerpc/power7/fpu/s_logb.c> diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/s_logbf.c b/sysdeps/powerpc/powerpc64/power7/fpu/s_logbf.c index e79a28f775..7a5a8032e0 100644 --- a/sysdeps/powerpc/powerpc64/power7/fpu/s_logbf.c +++ b/sysdeps/powerpc/powerpc64/power7/fpu/s_logbf.c @@ -1 +1 @@ -#include <sysdeps/powerpc/powerpc32/power7/fpu/s_logbf.c> +#include <sysdeps/powerpc/power7/fpu/s_logbf.c> diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/s_logbl.c b/sysdeps/powerpc/powerpc64/power7/fpu/s_logbl.c index 463e411b4e..524ae2c78d 100644 --- a/sysdeps/powerpc/powerpc64/power7/fpu/s_logbl.c +++ b/sysdeps/powerpc/powerpc64/power7/fpu/s_logbl.c @@ -1 +1 @@ -#include <sysdeps/powerpc/powerpc32/power7/fpu/s_logbl.c> +#include <sysdeps/powerpc/power7/fpu/s_logbl.c> diff --git a/sysdeps/powerpc/powerpc64/power7/memchr.S b/sysdeps/powerpc/powerpc64/power7/memchr.S index 3416897f50..421d6d4660 100644 --- a/sysdeps/powerpc/powerpc64/power7/memchr.S +++ b/sysdeps/powerpc/powerpc64/power7/memchr.S @@ -22,112 +22,115 @@ /* int [r3] memchr (char *s [r3], int byte [r4], int size [r5]) */ .machine power7 ENTRY (__memchr) - CALL_MCOUNT 2 + CALL_MCOUNT 3 dcbt 0,r3 clrrdi r8,r3,3 - rlwimi r4,r4,8,16,23 - rlwimi r4,r4,16,0,15 + insrdi r4,r4,8,48 add r7,r3,r5 /* Calculate the last acceptable address. */ + insrdi r4,r4,16,32 cmpldi r5,32 + li r9, -1 + rlwinm r6,r3,3,26,28 /* Calculate padding. */ insrdi r4,r4,32,0 + addi r7,r7,-1 +#ifdef __LITTLE_ENDIAN__ + sld r9,r9,r6 +#else + srd r9,r9,r6 +#endif ble L(small_range) - cmpld cr7,r3,r7 /* Compare the starting address (r3) with the - ending address (r7). If (r3 >= r7), - the size passed in was zero or negative. */ - ble cr7,L(proceed) - - li r7,-1 /* Artificially set our ending address (r7) - such that we will exit early. */ - -L(proceed): - rlwinm r6,r3,3,26,28 /* Calculate padding. */ - cmpldi cr6,r6,0 /* cr6 == Do we have padding? */ ld r12,0(r8) /* Load doubleword from memory. */ - cmpb r10,r12,r4 /* Check for BYTEs in DWORD1. */ - beq cr6,L(proceed_no_padding) - sld r10,r10,r6 - srd r10,r10,r6 -L(proceed_no_padding): - cmpldi cr7,r10,0 /* Does r10 indicate we got a hit? */ + cmpb r3,r12,r4 /* Check for BYTEs in DWORD1. */ + and r3,r3,r9 + clrldi r5,r7,61 /* Byte count - 1 in last dword. */ + clrrdi r7,r7,3 /* Address of last doubleword. */ + cmpldi cr7,r3,0 /* Does r3 indicate we got a hit? */ bne cr7,L(done) - /* See if we are at the last acceptable address yet. */ - addi r9,r8,8 - cmpld cr6,r9,r7 - bge cr6,L(null) - mtcrf 0x01,r8 /* Are we now aligned to a quadword boundary? If so, skip to the main loop. Otherwise, go through the alignment code. */ - bt 28,L(loop_setup) /* Handle DWORD2 of pair. */ ldu r12,8(r8) - cmpb r10,r12,r4 - cmpldi cr7,r10,0 + cmpb r3,r12,r4 + cmpldi cr7,r3,0 bne cr7,L(done) - /* Are we done already? */ - addi r9,r8,8 - cmpld cr6,r9,r7 - bge cr6,L(null) - L(loop_setup): - sub r5,r7,r9 - srdi r6,r5,4 /* Number of loop iterations. */ + /* The last dword we want to read in the loop below is the one + containing the last byte of the string, ie. the dword at + (s + size - 1) & ~7, or r7. The first dword read is at + r8 + 8, we read 2 * cnt dwords, so the last dword read will + be at r8 + 8 + 16 * cnt - 8. Solving for cnt gives + cnt = (r7 - r8) / 16 */ + sub r6,r7,r8 + srdi r6,r6,4 /* Number of loop iterations. */ mtctr r6 /* Setup the counter. */ - b L(loop) - /* Main loop to look for BYTE backwards in the string. Since - it's a small loop (< 8 instructions), align it to 32-bytes. */ - .p2align 5 + + /* Main loop to look for BYTE in the string. Since + it's a small loop (8 instructions), align it to 32-bytes. */ + .align 5 L(loop): /* Load two doublewords, compare and merge in a single register for speed. This is an attempt to speed up the byte-checking process for bigger strings. */ ld r12,8(r8) ldu r11,16(r8) - cmpb r10,r12,r4 + cmpb r3,r12,r4 cmpb r9,r11,r4 - or r5,r9,r10 /* Merge everything in one doubleword. */ - cmpldi cr7,r5,0 + or r6,r9,r3 /* Merge everything in one doubleword. */ + cmpldi cr7,r6,0 bne cr7,L(found) bdnz L(loop) - /* We're here because the counter reached 0, and that means we - didn't have any matches for BYTE in the whole range. */ - subi r11,r7,8 - cmpld cr6,r8,r11 - blt cr6,L(loop_small) - b L(null) + /* We may have one more dword to read. */ + cmpld r8,r7 + beqlr + ldu r12,8(r8) + cmpb r3,r12,r4 + cmpldi cr6,r3,0 + bne cr6,L(done) + blr + + .align 4 +L(found): /* OK, one (or both) of the doublewords contains BYTE. Check the first doubleword and decrement the address in case the first doubleword really contains BYTE. */ - .align 4 -L(found): - cmpldi cr6,r10,0 + cmpldi cr6,r3,0 addi r8,r8,-8 bne cr6,L(done) /* BYTE must be in the second doubleword. Adjust the address - again and move the result of cmpb to r10 so we can calculate the + again and move the result of cmpb to r3 so we can calculate the pointer. */ - mr r10,r9 + mr r3,r9 addi r8,r8,8 - /* r10 has the output of the cmpb instruction, that is, it contains + /* r3 has the output of the cmpb instruction, that is, it contains 0xff in the same position as BYTE in the original doubleword from the string. Use that to calculate the pointer. We need to make sure BYTE is *before* the end of the range. */ L(done): - cntlzd r0,r10 /* Count leading zeroes before the match. */ - srdi r0,r0,3 /* Convert leading zeroes to bytes. */ +#ifdef __LITTLE_ENDIAN__ + addi r0,r3,-1 + andc r0,r0,r3 + popcntd r0,r0 /* Count trailing zeros. */ +#else + cntlzd r0,r3 /* Count leading zeros before the match. */ +#endif + cmpld r8,r7 /* Are we on the last dword? */ + srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */ add r3,r8,r0 - cmpld r3,r7 - bge L(null) + cmpld cr7,r0,r5 /* If on the last dword, check byte offset. */ + bnelr + blelr cr7 + li r3,0 blr .align 4 @@ -139,67 +142,44 @@ L(null): .align 4 L(small_range): cmpldi r5,0 - rlwinm r6,r3,3,26,28 /* Calculate padding. */ - beq L(null) /* This branch is for the cmpldi r5,0 above. */ + beq L(null) ld r12,0(r8) /* Load word from memory. */ - cmpldi cr6,r6,0 /* cr6 == Do we have padding? */ - cmpb r10,r12,r4 /* Check for BYTE in DWORD1. */ - /* If no padding, skip the shifts. */ - beq cr6,L(small_no_padding) - sld r10,r10,r6 - srd r10,r10,r6 -L(small_no_padding): - cmpldi cr7,r10,0 + cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */ + and r3,r3,r9 + cmpldi cr7,r3,0 + clrldi r5,r7,61 /* Byte count - 1 in last dword. */ + clrrdi r7,r7,3 /* Address of last doubleword. */ + cmpld r8,r7 /* Are we done already? */ bne cr7,L(done) + beqlr - /* Are we done already? */ - addi r9,r8,8 - cmpld r9,r7 - bge L(null) - /* If we're not done, drop through into loop_small. */ - -L(loop_small): /* loop_small has been unrolled. */ ldu r12,8(r8) - cmpb r10,r12,r4 - addi r9,r8,8 - cmpldi cr6,r10,0 - cmpld r9,r7 + cmpb r3,r12,r4 + cmpldi cr6,r3,0 + cmpld r8,r7 bne cr6,L(done) /* Found something. */ - bge L(null) /* Hit end of string (length). */ + beqlr /* Hit end of string (length). */ ldu r12,8(r8) - cmpb r10,r12,r4 - addi r9,r8,8 - cmpldi cr6,r10,0 - cmpld r9,r7 - bne cr6,L(done) /* Found something. */ - bge L(null) + cmpb r3,r12,r4 + cmpldi cr6,r3,0 + cmpld r8,r7 + bne cr6,L(done) + beqlr ldu r12,8(r8) - subi r11,r7,8 - cmpb r10,r12,r4 - cmpldi cr6,r10,0 - ori r2,r2,0 /* Force a dispatch group. */ + cmpb r3,r12,r4 + cmpldi cr6,r3,0 + cmpld r8,r7 bne cr6,L(done) + beqlr - cmpld r8,r11 /* At end of range? */ - bge L(null) - - /* For most cases we will never get here. Under some combinations of - padding + length there is a leftover double that still needs to be - checked. */ ldu r12,8(r8) - cmpb r10,r12,r4 - addi r9,r8,8 - cmpldi cr6,r10,0 - cmpld r9,r7 - bne cr6,L(done) /* Found something. */ - - /* Save a branch and exit directly. */ - li r3,0 + cmpb r3,r12,r4 + cmpldi cr6,r3,0 + bne cr6,L(done) blr - END (__memchr) weak_alias (__memchr, memchr) libc_hidden_builtin_def (memchr) diff --git a/sysdeps/powerpc/powerpc64/power7/memcmp.S b/sysdeps/powerpc/powerpc64/power7/memcmp.S index f190c64611..6851cdc75b 100644 --- a/sysdeps/powerpc/powerpc64/power7/memcmp.S +++ b/sysdeps/powerpc/powerpc64/power7/memcmp.S @@ -23,10 +23,9 @@ size_t size [r5]) */ .machine power7 -EALIGN (memcmp,4,0) +EALIGN (memcmp, 4, 0) CALL_MCOUNT 3 -#define rTMP r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -37,354 +36,557 @@ EALIGN (memcmp,4,0) #define rWORD4 r9 /* next word in s2 */ #define rWORD5 r10 /* next word in s1 */ #define rWORD6 r11 /* next word in s2 */ -#define rBITDIF r12 /* bits that differ in s1 & s2 words */ #define rWORD7 r30 /* next word in s1 */ #define rWORD8 r31 /* next word in s2 */ - xor rTMP,rSTR2,rSTR1 - cmpldi cr6,rN,0 - cmpldi cr1,rN,12 - clrldi. rTMP,rTMP,61 - clrldi rBITDIF,rSTR1,61 - cmpldi cr5,rBITDIF,0 - beq- cr6,L(zeroLength) - dcbt 0,rSTR1 - dcbt 0,rSTR2 + xor r0, rSTR2, rSTR1 + cmpldi cr6, rN, 0 + cmpldi cr1, rN, 12 + clrldi. r0, r0, 61 + clrldi r12, rSTR1, 61 + cmpldi cr5, r12, 0 + beq- cr6, L(zeroLength) + dcbt 0, rSTR1 + dcbt 0, rSTR2 /* If less than 8 bytes or not aligned, use the unaligned byte loop. */ - blt cr1,L(bytealigned) - std rWORD8,-8(r1) - cfi_offset(rWORD8,-8) - std rWORD7,-16(r1) - cfi_offset(rWORD7,-16) + blt cr1, L(bytealigned) + std rWORD8, -8(r1) + cfi_offset(rWORD8, -8) + std rWORD7, -16(r1) + cfi_offset(rWORD7, -16) bne L(unaligned) /* At this point we know both strings have the same alignment and the - compare length is at least 8 bytes. rBITDIF contains the low order + compare length is at least 8 bytes. r12 contains the low order 3 bits of rSTR1 and cr5 contains the result of the logical compare - of rBITDIF to 0. If rBITDIF == 0 then we are already double word - aligned and can perform the DWaligned loop. + of r12 to 0. If r12 == 0 then we are already double word + aligned and can perform the DW aligned loop. Otherwise we know the two strings have the same alignment (but not - yet DW). So we can force the string addresses to the next lower DW - boundary and special case this first DW word using shift left to + yet DW). So we force the string addresses to the next lower DW + boundary and special case this first DW using shift left to eliminate bits preceding the first byte. Since we want to join the - normal (DWaligned) compare loop, starting at the second double word, + normal (DW aligned) compare loop, starting at the second double word, we need to adjust the length (rN) and special case the loop - versioning for the first DW. This insures that the loop count is - correct and the first DW (shifted) is in the expected resister pair. */ + versioning for the first DW. This ensures that the loop count is + correct and the first DW (shifted) is in the expected register pair. */ .align 4 L(samealignment): - clrrdi rSTR1,rSTR1,3 - clrrdi rSTR2,rSTR2,3 - beq cr5,L(DWaligned) - add rN,rN,rBITDIF - sldi r11,rBITDIF,3 - srdi rTMP,rN,5 /* Divide by 32 */ - andi. rBITDIF,rN,24 /* Get the DW remainder */ - ld rWORD1,0(rSTR1) - ld rWORD2,0(rSTR2) - cmpldi cr1,rBITDIF,16 - cmpldi cr7,rN,32 - clrldi rN,rN,61 + clrrdi rSTR1, rSTR1, 3 + clrrdi rSTR2, rSTR2, 3 + beq cr5, L(DWaligned) + add rN, rN, r12 + sldi rWORD6, r12, 3 + srdi r0, rN, 5 /* Divide by 32 */ + andi. r12, rN, 24 /* Get the DW remainder */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 0(rSTR1) + ld rWORD2, 0(rSTR2) +#endif + cmpldi cr1, r12, 16 + cmpldi cr7, rN, 32 + clrldi rN, rN, 61 beq L(dPs4) - mtctr rTMP - bgt cr1,L(dPs3) - beq cr1,L(dPs2) + mtctr r0 + bgt cr1, L(dPs3) + beq cr1, L(dPs2) /* Remainder is 8 */ .align 3 L(dsP1): - sld rWORD5,rWORD1,r11 - sld rWORD6,rWORD2,r11 - cmpld cr5,rWORD5,rWORD6 - blt cr7,L(dP1x) + sld rWORD5, rWORD1, rWORD6 + sld rWORD6, rWORD2, rWORD6 + cmpld cr5, rWORD5, rWORD6 + blt cr7, L(dP1x) /* Do something useful in this cycle since we have to branch anyway. */ - ld rWORD1,8(rSTR1) - ld rWORD2,8(rSTR2) - cmpld cr0,rWORD1,rWORD2 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 8(rSTR1) + ld rWORD2, 8(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 b L(dP1e) /* Remainder is 16 */ .align 4 L(dPs2): - sld rWORD5,rWORD1,r11 - sld rWORD6,rWORD2,r11 - cmpld cr6,rWORD5,rWORD6 - blt cr7,L(dP2x) + sld rWORD5, rWORD1, rWORD6 + sld rWORD6, rWORD2, rWORD6 + cmpld cr6, rWORD5, rWORD6 + blt cr7, L(dP2x) /* Do something useful in this cycle since we have to branch anyway. */ - ld rWORD7,8(rSTR1) - ld rWORD8,8(rSTR2) - cmpld cr5,rWORD7,rWORD8 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD7, 8(rSTR1) + ld rWORD8, 8(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 b L(dP2e) /* Remainder is 24 */ .align 4 L(dPs3): - sld rWORD3,rWORD1,r11 - sld rWORD4,rWORD2,r11 - cmpld cr1,rWORD3,rWORD4 + sld rWORD3, rWORD1, rWORD6 + sld rWORD4, rWORD2, rWORD6 + cmpld cr1, rWORD3, rWORD4 b L(dP3e) /* Count is a multiple of 32, remainder is 0 */ .align 4 L(dPs4): - mtctr rTMP - sld rWORD1,rWORD1,r11 - sld rWORD2,rWORD2,r11 - cmpld cr0,rWORD1,rWORD2 + mtctr r0 + sld rWORD1, rWORD1, rWORD6 + sld rWORD2, rWORD2, rWORD6 + cmpld cr7, rWORD1, rWORD2 b L(dP4e) /* At this point we know both strings are double word aligned and the compare length is at least 8 bytes. */ .align 4 L(DWaligned): - andi. rBITDIF,rN,24 /* Get the DW remainder */ - srdi rTMP,rN,5 /* Divide by 32 */ - cmpldi cr1,rBITDIF,16 - cmpldi cr7,rN,32 - clrldi rN,rN,61 + andi. r12, rN, 24 /* Get the DW remainder */ + srdi r0, rN, 5 /* Divide by 32 */ + cmpldi cr1, r12, 16 + cmpldi cr7, rN, 32 + clrldi rN, rN, 61 beq L(dP4) - bgt cr1,L(dP3) - beq cr1,L(dP2) + bgt cr1, L(dP3) + beq cr1, L(dP2) /* Remainder is 8 */ .align 4 L(dP1): - mtctr rTMP + mtctr r0 /* Normally we'd use rWORD7/rWORD8 here, but since we might exit early (8-15 byte compare), we want to use only volatile registers. This means we can avoid restoring non-volatile registers since we did not change any on the early exit path. The key here is the non-early exit path only cares about the condition code (cr5), not about which register pair was used. */ - ld rWORD5,0(rSTR1) - ld rWORD6,0(rSTR2) - cmpld cr5,rWORD5,rWORD6 - blt cr7,L(dP1x) - ld rWORD1,8(rSTR1) - ld rWORD2,8(rSTR2) - cmpld cr0,rWORD1,rWORD2 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 0(rSTR1) + ld rWORD6, 0(rSTR2) +#endif + cmpld cr5, rWORD5, rWORD6 + blt cr7, L(dP1x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 8(rSTR1) + ld rWORD2, 8(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 L(dP1e): - ld rWORD3,16(rSTR1) - ld rWORD4,16(rSTR2) - cmpld cr1,rWORD3,rWORD4 - ld rWORD5,24(rSTR1) - ld rWORD6,24(rSTR2) - cmpld cr6,rWORD5,rWORD6 - bne cr5,L(dLcr5) - bne cr0,L(dLcr0) - - ldu rWORD7,32(rSTR1) - ldu rWORD8,32(rSTR2) - bne cr1,L(dLcr1) - cmpld cr5,rWORD7,rWORD8 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 16(rSTR1) + ld rWORD4, 16(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 24(rSTR1) + ld rWORD6, 24(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + bne cr5, L(dLcr5x) + bne cr7, L(dLcr7x) + +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ldu rWORD7, 32(rSTR1) + ldu rWORD8, 32(rSTR2) +#endif + bne cr1, L(dLcr1) + cmpld cr5, rWORD7, rWORD8 bdnz L(dLoop) - bne cr6,L(dLcr6) - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + bne cr6, L(dLcr6) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) .align 3 L(dP1x): - sldi. r12,rN,3 - bne cr5,L(dLcr5) - subfic rN,r12,64 /* Shift count is 64 - (rN * 8). */ + sldi. r12, rN, 3 + bne cr5, L(dLcr5x) + subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ bne L(d00) - li rRTN,0 + li rRTN, 0 blr /* Remainder is 16 */ .align 4 L(dP2): - mtctr rTMP - ld rWORD5,0(rSTR1) - ld rWORD6,0(rSTR2) - cmpld cr6,rWORD5,rWORD6 - blt cr7,L(dP2x) - ld rWORD7,8(rSTR1) - ld rWORD8,8(rSTR2) - cmpld cr5,rWORD7,rWORD8 + mtctr r0 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 0(rSTR1) + ld rWORD6, 0(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + blt cr7, L(dP2x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD7, 8(rSTR1) + ld rWORD8, 8(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 L(dP2e): - ld rWORD1,16(rSTR1) - ld rWORD2,16(rSTR2) - cmpld cr0,rWORD1,rWORD2 - ld rWORD3,24(rSTR1) - ld rWORD4,24(rSTR2) - cmpld cr1,rWORD3,rWORD4 - addi rSTR1,rSTR1,8 - addi rSTR2,rSTR2,8 - bne cr6,L(dLcr6) - bne cr5,L(dLcr5) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 16(rSTR1) + ld rWORD2, 16(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 24(rSTR1) + ld rWORD4, 24(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#endif + bne cr6, L(dLcr6) + bne cr5, L(dLcr5) b L(dLoop2) /* Again we are on a early exit path (16-23 byte compare), we want to only use volatile registers and avoid restoring non-volatile registers. */ .align 4 L(dP2x): - ld rWORD3,8(rSTR1) - ld rWORD4,8(rSTR2) - cmpld cr5,rWORD3,rWORD4 - sldi. r12,rN,3 - bne cr6,L(dLcr6) - addi rSTR1,rSTR1,8 - addi rSTR2,rSTR2,8 - bne cr5,L(dLcr5) - subfic rN,r12,64 /* Shift count is 64 - (rN * 8). */ +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 8(rSTR1) + ld rWORD4, 8(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 + sldi. r12, rN, 3 + bne cr6, L(dLcr6x) +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#endif + bne cr1, L(dLcr1x) + subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ bne L(d00) - li rRTN,0 + li rRTN, 0 blr /* Remainder is 24 */ .align 4 L(dP3): - mtctr rTMP - ld rWORD3,0(rSTR1) - ld rWORD4,0(rSTR2) - cmpld cr1,rWORD3,rWORD4 + mtctr r0 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 0(rSTR1) + ld rWORD4, 0(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 L(dP3e): - ld rWORD5,8(rSTR1) - ld rWORD6,8(rSTR2) - cmpld cr6,rWORD5,rWORD6 - blt cr7,L(dP3x) - ld rWORD7,16(rSTR1) - ld rWORD8,16(rSTR2) - cmpld cr5,rWORD7,rWORD8 - ld rWORD1,24(rSTR1) - ld rWORD2,24(rSTR2) - cmpld cr0,rWORD1,rWORD2 - addi rSTR1,rSTR1,16 - addi rSTR2,rSTR2,16 - bne cr1,L(dLcr1) - bne cr6,L(dLcr6) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 8(rSTR1) + ld rWORD6, 8(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + blt cr7, L(dP3x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD7, 16(rSTR1) + ld rWORD8, 16(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 24(rSTR1) + ld rWORD2, 24(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 16 + addi rSTR2, rSTR2, 16 +#endif + bne cr1, L(dLcr1) + bne cr6, L(dLcr6) b L(dLoop1) /* Again we are on a early exit path (24-31 byte compare), we want to only use volatile registers and avoid restoring non-volatile registers. */ .align 4 L(dP3x): - ld rWORD1,16(rSTR1) - ld rWORD2,16(rSTR2) - cmpld cr5,rWORD1,rWORD2 - sldi. r12,rN,3 - bne cr1,L(dLcr1) - addi rSTR1,rSTR1,16 - addi rSTR2,rSTR2,16 - bne cr6,L(dLcr6) - subfic rN,r12,64 /* Shift count is 64 - (rN * 8). */ - bne cr5,L(dLcr5) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 16(rSTR1) + ld rWORD2, 16(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 + sldi. r12, rN, 3 + bne cr1, L(dLcr1x) +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 16 + addi rSTR2, rSTR2, 16 +#endif + bne cr6, L(dLcr6x) + subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ + bne cr7, L(dLcr7x) bne L(d00) - li rRTN,0 + li rRTN, 0 blr /* Count is a multiple of 32, remainder is 0 */ .align 4 L(dP4): - mtctr rTMP - ld rWORD1,0(rSTR1) - ld rWORD2,0(rSTR2) - cmpld cr0,rWORD1,rWORD2 + mtctr r0 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 0(rSTR1) + ld rWORD2, 0(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 L(dP4e): - ld rWORD3,8(rSTR1) - ld rWORD4,8(rSTR2) - cmpld cr1,rWORD3,rWORD4 - ld rWORD5,16(rSTR1) - ld rWORD6,16(rSTR2) - cmpld cr6,rWORD5,rWORD6 - ldu rWORD7,24(rSTR1) - ldu rWORD8,24(rSTR2) - cmpld cr5,rWORD7,rWORD8 - bne cr0,L(dLcr0) - bne cr1,L(dLcr1) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 8(rSTR1) + ld rWORD4, 8(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 16(rSTR1) + ld rWORD6, 16(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ldu rWORD7, 24(rSTR1) + ldu rWORD8, 24(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 + bne cr7, L(dLcr7) + bne cr1, L(dLcr1) bdz- L(d24) /* Adjust CTR as we start with +4 */ /* This is the primary loop */ .align 4 L(dLoop): - ld rWORD1,8(rSTR1) - ld rWORD2,8(rSTR2) - cmpld cr1,rWORD3,rWORD4 - bne cr6,L(dLcr6) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 8(rSTR1) + ld rWORD2, 8(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 + bne cr6, L(dLcr6) L(dLoop1): - ld rWORD3,16(rSTR1) - ld rWORD4,16(rSTR2) - cmpld cr6,rWORD5,rWORD6 - bne cr5,L(dLcr5) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 16(rSTR1) + ld rWORD4, 16(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + bne cr5, L(dLcr5) L(dLoop2): - ld rWORD5,24(rSTR1) - ld rWORD6,24(rSTR2) - cmpld cr5,rWORD7,rWORD8 - bne cr0,L(dLcr0) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 24(rSTR1) + ld rWORD6, 24(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 + bne cr7, L(dLcr7) L(dLoop3): - ldu rWORD7,32(rSTR1) - ldu rWORD8,32(rSTR2) - bne cr1,L(dLcr1) - cmpld cr0,rWORD1,rWORD2 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ldu rWORD7, 32(rSTR1) + ldu rWORD8, 32(rSTR2) +#endif + bne cr1, L(dLcr1) + cmpld cr7, rWORD1, rWORD2 bdnz L(dLoop) L(dL4): - cmpld cr1,rWORD3,rWORD4 - bne cr6,L(dLcr6) - cmpld cr6,rWORD5,rWORD6 - bne cr5,L(dLcr5) - cmpld cr5,rWORD7,rWORD8 + cmpld cr1, rWORD3, rWORD4 + bne cr6, L(dLcr6) + cmpld cr6, rWORD5, rWORD6 + bne cr5, L(dLcr5) + cmpld cr5, rWORD7, rWORD8 L(d44): - bne cr0,L(dLcr0) + bne cr7, L(dLcr7) L(d34): - bne cr1,L(dLcr1) + bne cr1, L(dLcr1) L(d24): - bne cr6,L(dLcr6) + bne cr6, L(dLcr6) L(d14): - sldi. r12,rN,3 - bne cr5,L(dLcr5) + sldi. r12, rN, 3 + bne cr5, L(dLcr5) L(d04): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - subfic rN,r12,64 /* Shift count is 64 - (rN * 8). */ + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) + subfic rN, r12, 64 /* Shift count is 64 - (rN * 8). */ beq L(zeroLength) /* At this point we have a remainder of 1 to 7 bytes to compare. Since we are aligned it is safe to load the whole double word, and use shift right double to eliminate bits beyond the compare length. */ L(d00): - ld rWORD1,8(rSTR1) - ld rWORD2,8(rSTR2) - srd rWORD1,rWORD1,rN - srd rWORD2,rWORD2,rN - cmpld cr5,rWORD1,rWORD2 - bne cr5,L(dLcr5x) - li rRTN,0 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 8(rSTR1) + ld rWORD2, 8(rSTR2) +#endif + srd rWORD1, rWORD1, rN + srd rWORD2, rWORD2, rN + cmpld cr7, rWORD1, rWORD2 + bne cr7, L(dLcr7x) + li rRTN, 0 blr + .align 4 -L(dLcr0): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - li rRTN,1 - bgtlr cr0 - li rRTN,-1 +L(dLcr7): + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) +L(dLcr7x): + li rRTN, 1 + bgtlr cr7 + li rRTN, -1 blr .align 4 L(dLcr1): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - li rRTN,1 + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) +L(dLcr1x): + li rRTN, 1 bgtlr cr1 - li rRTN,-1 + li rRTN, -1 blr .align 4 L(dLcr6): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - li rRTN,1 + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) +L(dLcr6x): + li rRTN, 1 bgtlr cr6 - li rRTN,-1 + li rRTN, -1 blr .align 4 L(dLcr5): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) L(dLcr5x): - li rRTN,1 + li rRTN, 1 bgtlr cr5 - li rRTN,-1 + li rRTN, -1 blr .align 4 L(bytealigned): mtctr rN - beq cr6,L(zeroLength) +#if 0 +/* Huh? We've already branched on cr6! */ + beq cr6, L(zeroLength) +#endif /* We need to prime this loop. This loop is swing modulo scheduled to avoid pipe delays. The dependent instruction latencies (load to @@ -396,38 +598,38 @@ L(bytealigned): So we must precondition some registers and condition codes so that we don't exit the loop early on the first iteration. */ - lbz rWORD1,0(rSTR1) - lbz rWORD2,0(rSTR2) + lbz rWORD1, 0(rSTR1) + lbz rWORD2, 0(rSTR2) bdz L(b11) - cmpld cr0,rWORD1,rWORD2 - lbz rWORD3,1(rSTR1) - lbz rWORD4,1(rSTR2) + cmpld cr7, rWORD1, rWORD2 + lbz rWORD3, 1(rSTR1) + lbz rWORD4, 1(rSTR2) bdz L(b12) - cmpld cr1,rWORD3,rWORD4 - lbzu rWORD5,2(rSTR1) - lbzu rWORD6,2(rSTR2) + cmpld cr1, rWORD3, rWORD4 + lbzu rWORD5, 2(rSTR1) + lbzu rWORD6, 2(rSTR2) bdz L(b13) .align 4 L(bLoop): - lbzu rWORD1,1(rSTR1) - lbzu rWORD2,1(rSTR2) - bne cr0,L(bLcr0) + lbzu rWORD1, 1(rSTR1) + lbzu rWORD2, 1(rSTR2) + bne cr7, L(bLcr7) - cmpld cr6,rWORD5,rWORD6 + cmpld cr6, rWORD5, rWORD6 bdz L(b3i) - lbzu rWORD3,1(rSTR1) - lbzu rWORD4,1(rSTR2) - bne cr1,L(bLcr1) + lbzu rWORD3, 1(rSTR1) + lbzu rWORD4, 1(rSTR2) + bne cr1, L(bLcr1) - cmpld cr0,rWORD1,rWORD2 + cmpld cr7, rWORD1, rWORD2 bdz L(b2i) - lbzu rWORD5,1(rSTR1) - lbzu rWORD6,1(rSTR2) - bne cr6,L(bLcr6) + lbzu rWORD5, 1(rSTR1) + lbzu rWORD6, 1(rSTR2) + bne cr6, L(bLcr6) - cmpld cr1,rWORD3,rWORD4 + cmpld cr1, rWORD3, rWORD4 bdnz L(bLoop) /* We speculatively loading bytes before we have tested the previous @@ -437,542 +639,727 @@ L(bLoop): tested. In this case we must complete the pending operations before returning. */ L(b1i): - bne cr0,L(bLcr0) - bne cr1,L(bLcr1) + bne cr7, L(bLcr7) + bne cr1, L(bLcr1) b L(bx56) .align 4 L(b2i): - bne cr6,L(bLcr6) - bne cr0,L(bLcr0) + bne cr6, L(bLcr6) + bne cr7, L(bLcr7) b L(bx34) .align 4 L(b3i): - bne cr1,L(bLcr1) - bne cr6,L(bLcr6) + bne cr1, L(bLcr1) + bne cr6, L(bLcr6) b L(bx12) .align 4 -L(bLcr0): - li rRTN,1 - bgtlr cr0 - li rRTN,-1 +L(bLcr7): + li rRTN, 1 + bgtlr cr7 + li rRTN, -1 blr L(bLcr1): - li rRTN,1 + li rRTN, 1 bgtlr cr1 - li rRTN,-1 + li rRTN, -1 blr L(bLcr6): - li rRTN,1 + li rRTN, 1 bgtlr cr6 - li rRTN,-1 + li rRTN, -1 blr L(b13): - bne cr0,L(bx12) - bne cr1,L(bx34) + bne cr7, L(bx12) + bne cr1, L(bx34) L(bx56): - sub rRTN,rWORD5,rWORD6 + sub rRTN, rWORD5, rWORD6 blr nop L(b12): - bne cr0,L(bx12) + bne cr7, L(bx12) L(bx34): - sub rRTN,rWORD3,rWORD4 + sub rRTN, rWORD3, rWORD4 blr L(b11): L(bx12): - sub rRTN,rWORD1,rWORD2 + sub rRTN, rWORD1, rWORD2 blr .align 4 -L(zeroLengthReturn): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) L(zeroLength): - li rRTN,0 + li rRTN, 0 blr .align 4 /* At this point we know the strings have different alignment and the - compare length is at least 8 bytes. rBITDIF contains the low order + compare length is at least 8 bytes. r12 contains the low order 3 bits of rSTR1 and cr5 contains the result of the logical compare - of rBITDIF to 0. If rBITDIF == 0 then rStr1 is double word + of r12 to 0. If r12 == 0 then rStr1 is double word aligned and can perform the DWunaligned loop. Otherwise we know that rSTR1 is not already DW aligned yet. So we can force the string addresses to the next lower DW - boundary and special case this first DW word using shift left to + boundary and special case this first DW using shift left to eliminate bits preceding the first byte. Since we want to join the normal (DWaligned) compare loop, starting at the second double word, we need to adjust the length (rN) and special case the loop - versioning for the first DW. This insures that the loop count is + versioning for the first DW. This ensures that the loop count is correct and the first DW (shifted) is in the expected resister pair. */ -#define rSHL r29 /* Unaligned shift left count. */ -#define rSHR r28 /* Unaligned shift right count. */ -#define rB r27 /* Left rotation temp for rWORD2. */ -#define rD r26 /* Left rotation temp for rWORD4. */ -#define rF r25 /* Left rotation temp for rWORD6. */ -#define rH r24 /* Left rotation temp for rWORD8. */ -#define rA r0 /* Right rotation temp for rWORD2. */ -#define rC r12 /* Right rotation temp for rWORD4. */ -#define rE r0 /* Right rotation temp for rWORD6. */ -#define rG r12 /* Right rotation temp for rWORD8. */ +#define rSHL r29 /* Unaligned shift left count. */ +#define rSHR r28 /* Unaligned shift right count. */ +#define rWORD8_SHIFT r27 /* Left rotation temp for rWORD2. */ +#define rWORD2_SHIFT r26 /* Left rotation temp for rWORD4. */ +#define rWORD4_SHIFT r25 /* Left rotation temp for rWORD6. */ +#define rWORD6_SHIFT r24 /* Left rotation temp for rWORD8. */ L(unaligned): - std r29,-24(r1) - cfi_offset(r29,-24) - clrldi rSHL,rSTR2,61 - beq cr6,L(duzeroLength) - std r28,-32(r1) - cfi_offset(r28,-32) - beq cr5,L(DWunaligned) - std r27,-40(r1) - cfi_offset(r27,-40) -/* Adjust the logical start of rSTR2 ro compensate for the extra bits + std rSHL, -24(r1) + cfi_offset(rSHL, -24) + clrldi rSHL, rSTR2, 61 + beq cr6, L(duzeroLength) + std rSHR, -32(r1) + cfi_offset(rSHR, -32) + beq cr5, L(DWunaligned) + std rWORD8_SHIFT, -40(r1) + cfi_offset(rWORD8_SHIFT, -40) +/* Adjust the logical start of rSTR2 to compensate for the extra bits in the 1st rSTR1 DW. */ - sub r27,rSTR2,rBITDIF + sub rWORD8_SHIFT, rSTR2, r12 /* But do not attempt to address the DW before that DW that contains the actual start of rSTR2. */ - clrrdi rSTR2,rSTR2,3 - std r26,-48(r1) - cfi_offset(r26,-48) + clrrdi rSTR2, rSTR2, 3 + std rWORD2_SHIFT, -48(r1) + cfi_offset(rWORD2_SHIFT, -48) /* Compute the left/right shift counts for the unaligned rSTR2, compensating for the logical (DW aligned) start of rSTR1. */ - clrldi rSHL,r27,61 - clrrdi rSTR1,rSTR1,3 - std r25,-56(r1) - cfi_offset(r25,-56) - sldi rSHL,rSHL,3 - cmpld cr5,r27,rSTR2 - add rN,rN,rBITDIF - sldi r11,rBITDIF,3 - std r24,-64(r1) - cfi_offset(r24,-64) - subfic rSHR,rSHL,64 - srdi rTMP,rN,5 /* Divide by 32 */ - andi. rBITDIF,rN,24 /* Get the DW remainder */ + clrldi rSHL, rWORD8_SHIFT, 61 + clrrdi rSTR1, rSTR1, 3 + std rWORD4_SHIFT, -56(r1) + cfi_offset(rWORD4_SHIFT, -56) + sldi rSHL, rSHL, 3 + cmpld cr5, rWORD8_SHIFT, rSTR2 + add rN, rN, r12 + sldi rWORD6, r12, 3 + std rWORD6_SHIFT, -64(r1) + cfi_offset(rWORD6_SHIFT, -64) + subfic rSHR, rSHL, 64 + srdi r0, rN, 5 /* Divide by 32 */ + andi. r12, rN, 24 /* Get the DW remainder */ /* We normally need to load 2 DWs to start the unaligned rSTR2, but in this special case those bits may be discarded anyway. Also we must avoid loading a DW where none of the bits are part of rSTR2 as this may cross a page boundary and cause a page fault. */ - li rWORD8,0 - blt cr5,L(dus0) - ld rWORD8,0(rSTR2) - la rSTR2,8(rSTR2) - sld rWORD8,rWORD8,rSHL + li rWORD8, 0 + blt cr5, L(dus0) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD8, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else + ld rWORD8, 0(rSTR2) + addi rSTR2, rSTR2, 8 +#endif + sld rWORD8, rWORD8, rSHL L(dus0): - ld rWORD1,0(rSTR1) - ld rWORD2,0(rSTR2) - cmpldi cr1,rBITDIF,16 - cmpldi cr7,rN,32 - srd rG,rWORD2,rSHR - clrldi rN,rN,61 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 0(rSTR1) + ld rWORD2, 0(rSTR2) +#endif + cmpldi cr1, r12, 16 + cmpldi cr7, rN, 32 + srd r12, rWORD2, rSHR + clrldi rN, rN, 61 beq L(duPs4) - mtctr rTMP - or rWORD8,rG,rWORD8 - bgt cr1,L(duPs3) - beq cr1,L(duPs2) + mtctr r0 + or rWORD8, r12, rWORD8 + bgt cr1, L(duPs3) + beq cr1, L(duPs2) /* Remainder is 8 */ .align 4 L(dusP1): - sld rB,rWORD2,rSHL - sld rWORD7,rWORD1,r11 - sld rWORD8,rWORD8,r11 - bge cr7,L(duP1e) + sld rWORD8_SHIFT, rWORD2, rSHL + sld rWORD7, rWORD1, rWORD6 + sld rWORD8, rWORD8, rWORD6 + bge cr7, L(duP1e) /* At this point we exit early with the first double word compare complete and remainder of 0 to 7 bytes. See L(du14) for details on how we handle the remaining bytes. */ - cmpld cr5,rWORD7,rWORD8 - sldi. rN,rN,3 - bne cr5,L(duLcr5) - cmpld cr7,rN,rSHR + cmpld cr5, rWORD7, rWORD8 + sldi. rN, rN, 3 + bne cr5, L(duLcr5) + cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA,0 - ble cr7,L(dutrim) - ld rWORD2,8(rSTR2) - srd rA,rWORD2,rSHR + li r0, 0 + ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else + ld rWORD2, 8(rSTR2) +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Remainder is 16 */ .align 4 L(duPs2): - sld rH,rWORD2,rSHL - sld rWORD5,rWORD1,r11 - sld rWORD6,rWORD8,r11 + sld rWORD6_SHIFT, rWORD2, rSHL + sld rWORD5, rWORD1, rWORD6 + sld rWORD6, rWORD8, rWORD6 b L(duP2e) /* Remainder is 24 */ .align 4 L(duPs3): - sld rF,rWORD2,rSHL - sld rWORD3,rWORD1,r11 - sld rWORD4,rWORD8,r11 + sld rWORD4_SHIFT, rWORD2, rSHL + sld rWORD3, rWORD1, rWORD6 + sld rWORD4, rWORD8, rWORD6 b L(duP3e) /* Count is a multiple of 32, remainder is 0 */ .align 4 L(duPs4): - mtctr rTMP - or rWORD8,rG,rWORD8 - sld rD,rWORD2,rSHL - sld rWORD1,rWORD1,r11 - sld rWORD2,rWORD8,r11 + mtctr r0 + or rWORD8, r12, rWORD8 + sld rWORD2_SHIFT, rWORD2, rSHL + sld rWORD1, rWORD1, rWORD6 + sld rWORD2, rWORD8, rWORD6 b L(duP4e) /* At this point we know rSTR1 is double word aligned and the compare length is at least 8 bytes. */ .align 4 L(DWunaligned): - std r27,-40(r1) - cfi_offset(r27,-40) - clrrdi rSTR2,rSTR2,3 - std r26,-48(r1) - cfi_offset(r26,-48) - srdi rTMP,rN,5 /* Divide by 32 */ - std r25,-56(r1) - cfi_offset(r25,-56) - andi. rBITDIF,rN,24 /* Get the DW remainder */ - std r24,-64(r1) - cfi_offset(r24,-64) - sldi rSHL,rSHL,3 - ld rWORD6,0(rSTR2) - ldu rWORD8,8(rSTR2) - cmpldi cr1,rBITDIF,16 - cmpldi cr7,rN,32 - clrldi rN,rN,61 - subfic rSHR,rSHL,64 - sld rH,rWORD6,rSHL + std rWORD8_SHIFT, -40(r1) + cfi_offset(rWORD8_SHIFT, -40) + clrrdi rSTR2, rSTR2, 3 + std rWORD2_SHIFT, -48(r1) + cfi_offset(rWORD2_SHIFT, -48) + srdi r0, rN, 5 /* Divide by 32 */ + std rWORD4_SHIFT, -56(r1) + cfi_offset(rWORD4_SHIFT, -56) + andi. r12, rN, 24 /* Get the DW remainder */ + std rWORD6_SHIFT, -64(r1) + cfi_offset(rWORD6_SHIFT, -64) + sldi rSHL, rSHL, 3 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD6, 0, rSTR2 + addi rSTR2, rSTR2, 8 + ldbrx rWORD8, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else + ld rWORD6, 0(rSTR2) + ldu rWORD8, 8(rSTR2) +#endif + cmpldi cr1, r12, 16 + cmpldi cr7, rN, 32 + clrldi rN, rN, 61 + subfic rSHR, rSHL, 64 + sld rWORD6_SHIFT, rWORD6, rSHL beq L(duP4) - mtctr rTMP - bgt cr1,L(duP3) - beq cr1,L(duP2) + mtctr r0 + bgt cr1, L(duP3) + beq cr1, L(duP2) /* Remainder is 8 */ .align 4 L(duP1): - srd rG,rWORD8,rSHR - ld rWORD7,0(rSTR1) - sld rB,rWORD8,rSHL - or rWORD8,rG,rH - blt cr7,L(duP1x) + srd r12, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else + ld rWORD7, 0(rSTR1) +#endif + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT + blt cr7, L(duP1x) L(duP1e): - ld rWORD1,8(rSTR1) - ld rWORD2,8(rSTR2) - cmpld cr5,rWORD7,rWORD8 - srd rA,rWORD2,rSHR - sld rD,rWORD2,rSHL - or rWORD2,rA,rB - ld rWORD3,16(rSTR1) - ld rWORD4,16(rSTR2) - cmpld cr0,rWORD1,rWORD2 - srd rC,rWORD4,rSHR - sld rF,rWORD4,rSHL - bne cr5,L(duLcr5) - or rWORD4,rC,rD - ld rWORD5,24(rSTR1) - ld rWORD6,24(rSTR2) - cmpld cr1,rWORD3,rWORD4 - srd rE,rWORD6,rSHR - sld rH,rWORD6,rSHL - bne cr0,L(duLcr0) - or rWORD6,rE,rF - cmpld cr6,rWORD5,rWORD6 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 8(rSTR1) + ld rWORD2, 8(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 16(rSTR1) + ld rWORD4, 16(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL + bne cr5, L(duLcr5) + or rWORD4, r12, rWORD2_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 24(rSTR1) + ld rWORD6, 24(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + bne cr7, L(duLcr7) + or rWORD6, r0, rWORD4_SHIFT + cmpld cr6, rWORD5, rWORD6 b L(duLoop3) .align 4 /* At this point we exit early with the first double word compare complete and remainder of 0 to 7 bytes. See L(du14) for details on how we handle the remaining bytes. */ L(duP1x): - cmpld cr5,rWORD7,rWORD8 - sldi. rN,rN,3 - bne cr5,L(duLcr5) - cmpld cr7,rN,rSHR + cmpld cr5, rWORD7, rWORD8 + sldi. rN, rN, 3 + bne cr5, L(duLcr5) + cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA,0 - ble cr7,L(dutrim) - ld rWORD2,8(rSTR2) - srd rA,rWORD2,rSHR + li r0, 0 + ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else + ld rWORD2, 8(rSTR2) +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Remainder is 16 */ .align 4 L(duP2): - srd rE,rWORD8,rSHR - ld rWORD5,0(rSTR1) - or rWORD6,rE,rH - sld rH,rWORD8,rSHL + srd r0, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else + ld rWORD5, 0(rSTR1) +#endif + or rWORD6, r0, rWORD6_SHIFT + sld rWORD6_SHIFT, rWORD8, rSHL L(duP2e): - ld rWORD7,8(rSTR1) - ld rWORD8,8(rSTR2) - cmpld cr6,rWORD5,rWORD6 - srd rG,rWORD8,rSHR - sld rB,rWORD8,rSHL - or rWORD8,rG,rH - blt cr7,L(duP2x) - ld rWORD1,16(rSTR1) - ld rWORD2,16(rSTR2) - cmpld cr5,rWORD7,rWORD8 - bne cr6,L(duLcr6) - srd rA,rWORD2,rSHR - sld rD,rWORD2,rSHL - or rWORD2,rA,rB - ld rWORD3,24(rSTR1) - ld rWORD4,24(rSTR2) - cmpld cr0,rWORD1,rWORD2 - bne cr5,L(duLcr5) - srd rC,rWORD4,rSHR - sld rF,rWORD4,rSHL - or rWORD4,rC,rD - addi rSTR1,rSTR1,8 - addi rSTR2,rSTR2,8 - cmpld cr1,rWORD3,rWORD4 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD7, 8(rSTR1) + ld rWORD8, 8(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT + blt cr7, L(duP2x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 16(rSTR1) + ld rWORD2, 16(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 + bne cr6, L(duLcr6) + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 24(rSTR1) + ld rWORD4, 24(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 + bne cr5, L(duLcr5) + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL + or rWORD4, r12, rWORD2_SHIFT +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#endif + cmpld cr1, rWORD3, rWORD4 b L(duLoop2) .align 4 L(duP2x): - cmpld cr5,rWORD7,rWORD8 - addi rSTR1,rSTR1,8 - addi rSTR2,rSTR2,8 - bne cr6,L(duLcr6) - sldi. rN,rN,3 - bne cr5,L(duLcr5) - cmpld cr7,rN,rSHR + cmpld cr5, rWORD7, rWORD8 +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#endif + bne cr6, L(duLcr6) + sldi. rN, rN, 3 + bne cr5, L(duLcr5) + cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA,0 - ble cr7,L(dutrim) - ld rWORD2,8(rSTR2) - srd rA,rWORD2,rSHR + li r0, 0 + ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else + ld rWORD2, 8(rSTR2) +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Remainder is 24 */ .align 4 L(duP3): - srd rC,rWORD8,rSHR - ld rWORD3,0(rSTR1) - sld rF,rWORD8,rSHL - or rWORD4,rC,rH + srd r12, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else + ld rWORD3, 0(rSTR1) +#endif + sld rWORD4_SHIFT, rWORD8, rSHL + or rWORD4, r12, rWORD6_SHIFT L(duP3e): - ld rWORD5,8(rSTR1) - ld rWORD6,8(rSTR2) - cmpld cr1,rWORD3,rWORD4 - srd rE,rWORD6,rSHR - sld rH,rWORD6,rSHL - or rWORD6,rE,rF - ld rWORD7,16(rSTR1) - ld rWORD8,16(rSTR2) - cmpld cr6,rWORD5,rWORD6 - bne cr1,L(duLcr1) - srd rG,rWORD8,rSHR - sld rB,rWORD8,rSHL - or rWORD8,rG,rH - blt cr7,L(duP3x) - ld rWORD1,24(rSTR1) - ld rWORD2,24(rSTR2) - cmpld cr5,rWORD7,rWORD8 - bne cr6,L(duLcr6) - srd rA,rWORD2,rSHR - sld rD,rWORD2,rSHL - or rWORD2,rA,rB - addi rSTR1,rSTR1,16 - addi rSTR2,rSTR2,16 - cmpld cr0,rWORD1,rWORD2 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 8(rSTR1) + ld rWORD6, 8(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + or rWORD6, r0, rWORD4_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD7, 16(rSTR1) + ld rWORD8, 16(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + bne cr1, L(duLcr1) + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT + blt cr7, L(duP3x) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 24(rSTR1) + ld rWORD2, 24(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 + bne cr6, L(duLcr6) + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 16 + addi rSTR2, rSTR2, 16 +#endif + cmpld cr7, rWORD1, rWORD2 b L(duLoop1) .align 4 L(duP3x): - addi rSTR1,rSTR1,16 - addi rSTR2,rSTR2,16 - bne cr1,L(duLcr1) - cmpld cr5,rWORD7,rWORD8 - bne cr6,L(duLcr6) - sldi. rN,rN,3 - bne cr5,L(duLcr5) - cmpld cr7,rN,rSHR +#ifndef __LITTLE_ENDIAN__ + addi rSTR1, rSTR1, 16 + addi rSTR2, rSTR2, 16 +#endif +#if 0 +/* Huh? We've already branched on cr1! */ + bne cr1, L(duLcr1) +#endif + cmpld cr5, rWORD7, rWORD8 + bne cr6, L(duLcr6) + sldi. rN, rN, 3 + bne cr5, L(duLcr5) + cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA,0 - ble cr7,L(dutrim) - ld rWORD2,8(rSTR2) - srd rA,rWORD2,rSHR + li r0, 0 + ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else + ld rWORD2, 8(rSTR2) +#endif + srd r0, rWORD2, rSHR b L(dutrim) /* Count is a multiple of 32, remainder is 0 */ .align 4 L(duP4): - mtctr rTMP - srd rA,rWORD8,rSHR - ld rWORD1,0(rSTR1) - sld rD,rWORD8,rSHL - or rWORD2,rA,rH + mtctr r0 + srd r0, rWORD8, rSHR +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + addi rSTR1, rSTR1, 8 +#else + ld rWORD1, 0(rSTR1) +#endif + sld rWORD2_SHIFT, rWORD8, rSHL + or rWORD2, r0, rWORD6_SHIFT L(duP4e): - ld rWORD3,8(rSTR1) - ld rWORD4,8(rSTR2) - cmpld cr0,rWORD1,rWORD2 - srd rC,rWORD4,rSHR - sld rF,rWORD4,rSHL - or rWORD4,rC,rD - ld rWORD5,16(rSTR1) - ld rWORD6,16(rSTR2) - cmpld cr1,rWORD3,rWORD4 - bne cr0,L(duLcr0) - srd rE,rWORD6,rSHR - sld rH,rWORD6,rSHL - or rWORD6,rE,rF - ldu rWORD7,24(rSTR1) - ldu rWORD8,24(rSTR2) - cmpld cr6,rWORD5,rWORD6 - bne cr1,L(duLcr1) - srd rG,rWORD8,rSHR - sld rB,rWORD8,rSHL - or rWORD8,rG,rH - cmpld cr5,rWORD7,rWORD8 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 8(rSTR1) + ld rWORD4, 8(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL + or rWORD4, r12, rWORD2_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 16(rSTR1) + ld rWORD6, 16(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 + bne cr7, L(duLcr7) + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + or rWORD6, r0, rWORD4_SHIFT +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ldu rWORD7, 24(rSTR1) + ldu rWORD8, 24(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + bne cr1, L(duLcr1) + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT + cmpld cr5, rWORD7, rWORD8 bdz L(du24) /* Adjust CTR as we start with +4 */ /* This is the primary loop */ .align 4 L(duLoop): - ld rWORD1,8(rSTR1) - ld rWORD2,8(rSTR2) - cmpld cr1,rWORD3,rWORD4 - bne cr6,L(duLcr6) - srd rA,rWORD2,rSHR - sld rD,rWORD2,rSHL - or rWORD2,rA,rB +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 + ldbrx rWORD2, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD1, 8(rSTR1) + ld rWORD2, 8(rSTR2) +#endif + cmpld cr1, rWORD3, rWORD4 + bne cr6, L(duLcr6) + srd r0, rWORD2, rSHR + sld rWORD2_SHIFT, rWORD2, rSHL + or rWORD2, r0, rWORD8_SHIFT L(duLoop1): - ld rWORD3,16(rSTR1) - ld rWORD4,16(rSTR2) - cmpld cr6,rWORD5,rWORD6 - bne cr5,L(duLcr5) - srd rC,rWORD4,rSHR - sld rF,rWORD4,rSHL - or rWORD4,rC,rD +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD3, 0, rSTR1 + ldbrx rWORD4, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD3, 16(rSTR1) + ld rWORD4, 16(rSTR2) +#endif + cmpld cr6, rWORD5, rWORD6 + bne cr5, L(duLcr5) + srd r12, rWORD4, rSHR + sld rWORD4_SHIFT, rWORD4, rSHL + or rWORD4, r12, rWORD2_SHIFT L(duLoop2): - ld rWORD5,24(rSTR1) - ld rWORD6,24(rSTR2) - cmpld cr5,rWORD7,rWORD8 - bne cr0,L(duLcr0) - srd rE,rWORD6,rSHR - sld rH,rWORD6,rSHL - or rWORD6,rE,rF +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD5, 0, rSTR1 + ldbrx rWORD6, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ld rWORD5, 24(rSTR1) + ld rWORD6, 24(rSTR2) +#endif + cmpld cr5, rWORD7, rWORD8 + bne cr7, L(duLcr7) + srd r0, rWORD6, rSHR + sld rWORD6_SHIFT, rWORD6, rSHL + or rWORD6, r0, rWORD4_SHIFT L(duLoop3): - ldu rWORD7,32(rSTR1) - ldu rWORD8,32(rSTR2) - cmpld cr0,rWORD1,rWORD2 - bne- cr1,L(duLcr1) - srd rG,rWORD8,rSHR - sld rB,rWORD8,rSHL - or rWORD8,rG,rH +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD7, 0, rSTR1 + ldbrx rWORD8, 0, rSTR2 + addi rSTR1, rSTR1, 8 + addi rSTR2, rSTR2, 8 +#else + ldu rWORD7, 32(rSTR1) + ldu rWORD8, 32(rSTR2) +#endif + cmpld cr7, rWORD1, rWORD2 + bne cr1, L(duLcr1) + srd r12, rWORD8, rSHR + sld rWORD8_SHIFT, rWORD8, rSHL + or rWORD8, r12, rWORD6_SHIFT bdnz L(duLoop) L(duL4): - bne cr1,L(duLcr1) - cmpld cr1,rWORD3,rWORD4 - bne cr6,L(duLcr6) - cmpld cr6,rWORD5,rWORD6 - bne cr5,L(duLcr5) - cmpld cr5,rWORD7,rWORD8 +#if 0 +/* Huh? We've already branched on cr1! */ + bne cr1, L(duLcr1) +#endif + cmpld cr1, rWORD3, rWORD4 + bne cr6, L(duLcr6) + cmpld cr6, rWORD5, rWORD6 + bne cr5, L(duLcr5) + cmpld cr5, rWORD7, rWORD8 L(du44): - bne cr0,L(duLcr0) + bne cr7, L(duLcr7) L(du34): - bne cr1,L(duLcr1) + bne cr1, L(duLcr1) L(du24): - bne cr6,L(duLcr6) + bne cr6, L(duLcr6) L(du14): - sldi. rN,rN,3 - bne cr5,L(duLcr5) + sldi. rN, rN, 3 + bne cr5, L(duLcr5) /* At this point we have a remainder of 1 to 7 bytes to compare. We use shift right double to eliminate bits beyond the compare length. - This allows the use of double word subtract to compute the final - result. However it may not be safe to load rWORD2 which may be beyond the string length. So we compare the bit length of the remainder to the right shift count (rSHR). If the bit count is less than or equal we do not need to load rWORD2 (all significant bits are already in - rB). */ - cmpld cr7,rN,rSHR + rWORD8_SHIFT). */ + cmpld cr7, rN, rSHR beq L(duZeroReturn) - li rA,0 - ble cr7,L(dutrim) - ld rWORD2,8(rSTR2) - srd rA,rWORD2,rSHR + li r0, 0 + ble cr7, L(dutrim) +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD2, 0, rSTR2 + addi rSTR2, rSTR2, 8 +#else + ld rWORD2, 8(rSTR2) +#endif + srd r0, rWORD2, rSHR .align 4 L(dutrim): - ld rWORD1,8(rSTR1) - ld rWORD8,-8(r1) - subfic rN,rN,64 /* Shift count is 64 - (rN * 8). */ - or rWORD2,rA,rB - ld rWORD7,-16(r1) - ld r29,-24(r1) - srd rWORD1,rWORD1,rN - srd rWORD2,rWORD2,rN - ld r28,-32(r1) - ld r27,-40(r1) - li rRTN,0 - cmpld cr0,rWORD1,rWORD2 - ld r26,-48(r1) - ld r25,-56(r1) - beq cr0,L(dureturn24) - li rRTN,1 - ld r24,-64(r1) - bgtlr cr0 - li rRTN,-1 +#ifdef __LITTLE_ENDIAN__ + ldbrx rWORD1, 0, rSTR1 +#else + ld rWORD1, 8(rSTR1) +#endif + ld rWORD8, -8(r1) + subfic rN, rN, 64 /* Shift count is 64 - (rN * 8). */ + or rWORD2, r0, rWORD8_SHIFT + ld rWORD7, -16(r1) + ld rSHL, -24(r1) + srd rWORD1, rWORD1, rN + srd rWORD2, rWORD2, rN + ld rSHR, -32(r1) + ld rWORD8_SHIFT, -40(r1) + li rRTN, 0 + cmpld cr7, rWORD1, rWORD2 + ld rWORD2_SHIFT, -48(r1) + ld rWORD4_SHIFT, -56(r1) + beq cr7, L(dureturn24) + li rRTN, 1 + ld rWORD6_SHIFT, -64(r1) + bgtlr cr7 + li rRTN, -1 blr .align 4 -L(duLcr0): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - li rRTN,1 - bgt cr0,L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) - li rRTN,-1 +L(duLcr7): + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) + li rRTN, 1 + bgt cr7, L(dureturn29) + ld rSHL, -24(r1) + ld rSHR, -32(r1) + li rRTN, -1 b L(dureturn27) .align 4 L(duLcr1): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - li rRTN,1 - bgt cr1,L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) - li rRTN,-1 + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) + li rRTN, 1 + bgt cr1, L(dureturn29) + ld rSHL, -24(r1) + ld rSHR, -32(r1) + li rRTN, -1 b L(dureturn27) .align 4 L(duLcr6): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - li rRTN,1 - bgt cr6,L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) - li rRTN,-1 + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) + li rRTN, 1 + bgt cr6, L(dureturn29) + ld rSHL, -24(r1) + ld rSHR, -32(r1) + li rRTN, -1 b L(dureturn27) .align 4 L(duLcr5): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) - li rRTN,1 - bgt cr5,L(dureturn29) - ld r29,-24(r1) - ld r28,-32(r1) - li rRTN,-1 + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) + li rRTN, 1 + bgt cr5, L(dureturn29) + ld rSHL, -24(r1) + ld rSHR, -32(r1) + li rRTN, -1 b L(dureturn27) .align 3 L(duZeroReturn): - li rRTN,0 + li rRTN, 0 .align 4 L(dureturn): - ld rWORD8,-8(r1) - ld rWORD7,-16(r1) + ld rWORD8, -8(r1) + ld rWORD7, -16(r1) L(dureturn29): - ld r29,-24(r1) - ld r28,-32(r1) + ld rSHL, -24(r1) + ld rSHR, -32(r1) L(dureturn27): - ld r27,-40(r1) + ld rWORD8_SHIFT, -40(r1) L(dureturn26): - ld r26,-48(r1) + ld rWORD2_SHIFT, -48(r1) L(dureturn25): - ld r25,-56(r1) + ld rWORD4_SHIFT, -56(r1) L(dureturn24): - ld r24,-64(r1) + ld rWORD6_SHIFT, -64(r1) blr L(duzeroLength): - li rRTN,0 + li rRTN, 0 blr END (memcmp) libc_hidden_builtin_def (memcmp) -weak_alias (memcmp,bcmp) +weak_alias (memcmp, bcmp) diff --git a/sysdeps/powerpc/powerpc64/power7/memcpy.S b/sysdeps/powerpc/powerpc64/power7/memcpy.S index 800a9f1bb1..e8df75f593 100644 --- a/sysdeps/powerpc/powerpc64/power7/memcpy.S +++ b/sysdeps/powerpc/powerpc64/power7/memcpy.S @@ -23,418 +23,361 @@ /* __ptr_t [r3] memcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]); Returns 'dst'. */ +#define dst 11 /* Use r11 so r3 kept unchanged. */ +#define src 4 +#define cnt 5 + .machine power7 EALIGN (memcpy, 5, 0) CALL_MCOUNT 3 - cmpldi cr1,5,31 + cmpldi cr1,cnt,31 neg 0,3 - std 3,-16(1) - std 31,-8(1) - cfi_offset(31,-8) ble cr1, L(copy_LT_32) /* If move < 32 bytes use short move code. */ - andi. 11,3,7 /* Check alignment of DST. */ - - - clrldi 10,4,61 /* Check alignment of SRC. */ - cmpld cr6,10,11 /* SRC and DST alignments match? */ - mr 12,4 - mr 31,5 +#ifdef __LITTLE_ENDIAN__ +/* In little-endian mode, power7 takes an alignment trap on any lxvd2x + or stxvd2x crossing a 32-byte boundary, so ensure the aligned_copy + loop is only used for quadword aligned copies. */ + andi. 10,3,15 + clrldi 11,4,60 +#else + andi. 10,3,7 /* Check alignment of DST. */ + clrldi 11,4,61 /* Check alignment of SRC. */ +#endif + cmpld cr6,10,11 /* SRC and DST alignments match? */ + + mr dst,3 bne cr6,L(copy_GE_32_unaligned) + beq L(aligned_copy) - srdi 9,5,3 /* Number of full quadwords remaining. */ - - beq L(copy_GE_32_aligned_cont) - - clrldi 0,0,61 - mtcrf 0x01,0 - subf 31,0,5 - - /* Get the SRC aligned to 8 bytes. */ - -1: bf 31,2f - lbz 6,0(12) - addi 12,12,1 - stb 6,0(3) - addi 3,3,1 -2: bf 30,4f - lhz 6,0(12) - addi 12,12,2 - sth 6,0(3) - addi 3,3,2 -4: bf 29,0f - lwz 6,0(12) - addi 12,12,4 - stw 6,0(3) - addi 3,3,4 -0: - clrldi 10,12,61 /* Check alignment of SRC again. */ - srdi 9,31,3 /* Number of full doublewords remaining. */ - -L(copy_GE_32_aligned_cont): - - clrldi 11,31,61 - mtcrf 0x01,9 - - srdi 8,31,5 - cmpldi cr1,9,4 - cmpldi cr6,11,0 - mr 11,12 - - /* Copy 1~3 doublewords so the main loop starts - at a multiple of 32 bytes. */ + mtocrf 0x01,0 +#ifdef __LITTLE_ENDIAN__ + clrldi 0,0,60 +#else + clrldi 0,0,61 +#endif - bf 30,1f - ld 6,0(12) - ld 7,8(12) - addi 11,12,16 - mtctr 8 - std 6,0(3) - std 7,8(3) - addi 10,3,16 - bf 31,4f - ld 0,16(12) - std 0,16(3) - blt cr1,3f - addi 11,12,24 - addi 10,3,24 - b 4f - - .align 4 -1: /* Copy 1 doubleword and set the counter. */ - mr 10,3 - mtctr 8 - bf 31,4f - ld 6,0(12) - addi 11,12,8 - std 6,0(3) - addi 10,3,8 - -L(aligned_copy): - /* Main aligned copy loop. Copies up to 128-bytes at a time. */ - .align 4 +/* Get the DST and SRC aligned to 8 bytes (16 for little-endian). */ +1: + bf 31,2f + lbz 6,0(src) + addi src,src,1 + stb 6,0(dst) + addi dst,dst,1 +2: + bf 30,4f + lhz 6,0(src) + addi src,src,2 + sth 6,0(dst) + addi dst,dst,2 4: - /* check for any 32-byte or 64-byte lumps that are outside of a - nice 128-byte range. R8 contains the number of 32-byte - lumps, so drop this into the CR, and use the SO/EQ bits to help - handle the 32- or 64- byte lumps. Then handle the rest with an - unrolled 128-bytes-at-a-time copy loop. */ - mtocrf 1,8 - li 6,16 # 16() index - li 7,32 # 32() index - li 8,48 # 48() index - -L(aligned_32byte): - /* if the SO bit (indicating a 32-byte lump) is not set, move along. */ - bns cr7,L(aligned_64byte) - lxvd2x 6,0,11 - lxvd2x 7,11,6 - addi 11,11,32 - stxvd2x 6,0,10 - stxvd2x 7,10,6 - addi 10,10,32 - -L(aligned_64byte): - /* if the EQ bit (indicating a 64-byte lump) is not set, move along. */ - bne cr7,L(aligned_128setup) - lxvd2x 6,0,11 - lxvd2x 7,11,6 - lxvd2x 8,11,7 - lxvd2x 9,11,8 - addi 11,11,64 - stxvd2x 6,0,10 - stxvd2x 7,10,6 - stxvd2x 8,10,7 - stxvd2x 9,10,8 - addi 10,10,64 - -L(aligned_128setup): - /* Set up for the 128-byte at a time copy loop. */ - srdi 8,31,7 - cmpdi 8,0 # Any 4x lumps left? - beq 3f # if not, move along. - lxvd2x 6,0,11 - lxvd2x 7,11,6 - mtctr 8 # otherwise, load the ctr and begin. - li 8,48 # 48() index + bf 29,8f + lwz 6,0(src) + addi src,src,4 + stw 6,0(dst) + addi dst,dst,4 +8: +#ifdef __LITTLE_ENDIAN__ + bf 28,16f + ld 6,0(src) + addi src,src,8 + std 6,0(dst) + addi dst,dst,8 +16: +#endif + subf cnt,0,cnt + +/* Main aligned copy loop. Copies 128 bytes at a time. */ +L(aligned_copy): + li 6,16 + li 7,32 + li 8,48 + mtocrf 0x02,cnt + srdi 12,cnt,7 + cmpdi 12,0 + beq L(aligned_tail) + lxvd2x 6,0,src + lxvd2x 7,src,6 + mtctr 12 b L(aligned_128loop) + .align 4 L(aligned_128head): /* for the 2nd + iteration of this loop. */ - lxvd2x 6,0,11 - lxvd2x 7,11,6 + lxvd2x 6,0,src + lxvd2x 7,src,6 L(aligned_128loop): - lxvd2x 8,11,7 - lxvd2x 9,11,8 - stxvd2x 6,0,10 - addi 11,11,64 - stxvd2x 7,10,6 - stxvd2x 8,10,7 - stxvd2x 9,10,8 - lxvd2x 6,0,11 - lxvd2x 7,11,6 - addi 10,10,64 - lxvd2x 8,11,7 - lxvd2x 9,11,8 - addi 11,11,64 - stxvd2x 6,0,10 - stxvd2x 7,10,6 - stxvd2x 8,10,7 - stxvd2x 9,10,8 - addi 10,10,64 + lxvd2x 8,src,7 + lxvd2x 9,src,8 + stxvd2x 6,0,dst + addi src,src,64 + stxvd2x 7,dst,6 + stxvd2x 8,dst,7 + stxvd2x 9,dst,8 + lxvd2x 6,0,src + lxvd2x 7,src,6 + addi dst,dst,64 + lxvd2x 8,src,7 + lxvd2x 9,src,8 + addi src,src,64 + stxvd2x 6,0,dst + stxvd2x 7,dst,6 + stxvd2x 8,dst,7 + stxvd2x 9,dst,8 + addi dst,dst,64 bdnz L(aligned_128head) -3: - /* Check for tail bytes. */ - rldicr 0,31,0,60 - mtcrf 0x01,31 - beq cr6,0f - -.L9: - add 3,3,0 - add 12,12,0 - - /* At this point we have a tail of 0-7 bytes and we know that the - destination is doubleword-aligned. */ -4: /* Copy 4 bytes. */ - bf 29,2f - - lwz 6,0(12) - addi 12,12,4 - stw 6,0(3) - addi 3,3,4 -2: /* Copy 2 bytes. */ - bf 30,1f - - lhz 6,0(12) - addi 12,12,2 - sth 6,0(3) - addi 3,3,2 -1: /* Copy 1 byte. */ - bf 31,0f - - lbz 6,0(12) - stb 6,0(3) -0: /* Return original DST pointer. */ - ld 31,-8(1) - ld 3,-16(1) +L(aligned_tail): + mtocrf 0x01,cnt + bf 25,32f + lxvd2x 6,0,src + lxvd2x 7,src,6 + lxvd2x 8,src,7 + lxvd2x 9,src,8 + addi src,src,64 + stxvd2x 6,0,dst + stxvd2x 7,dst,6 + stxvd2x 8,dst,7 + stxvd2x 9,dst,8 + addi dst,dst,64 +32: + bf 26,16f + lxvd2x 6,0,src + lxvd2x 7,src,6 + addi src,src,32 + stxvd2x 6,0,dst + stxvd2x 7,dst,6 + addi dst,dst,32 +16: + bf 27,8f + lxvd2x 6,0,src + addi src,src,16 + stxvd2x 6,0,dst + addi dst,dst,16 +8: + bf 28,4f + ld 6,0(src) + addi src,src,8 + std 6,0(dst) + addi dst,dst,8 +4: /* Copies 4~7 bytes. */ + bf 29,L(tail2) + lwz 6,0(src) + stw 6,0(dst) + bf 30,L(tail5) + lhz 7,4(src) + sth 7,4(dst) + bflr 31 + lbz 8,6(src) + stb 8,6(dst) + /* Return original DST pointer. */ blr - /* Handle copies of 0~31 bytes. */ - .align 4 + +/* Handle copies of 0~31 bytes. */ + .align 4 L(copy_LT_32): - cmpldi cr6,5,8 - mr 12,4 - mtcrf 0x01,5 + mr dst,3 + cmpldi cr6,cnt,8 + mtocrf 0x01,cnt ble cr6,L(copy_LE_8) /* At least 9 bytes to go. */ neg 8,4 - clrrdi 11,4,2 - andi. 0,8,3 - cmpldi cr1,5,16 - mr 10,5 + andi. 0,8,3 + cmpldi cr1,cnt,16 beq L(copy_LT_32_aligned) - /* Force 4-bytes alignment for SRC. */ - mtocrf 0x01,0 - subf 10,0,5 -2: bf 30,1f - - lhz 6,0(12) - addi 12,12,2 - sth 6,0(3) - addi 3,3,2 -1: bf 31,L(end_4bytes_alignment) - - lbz 6,0(12) - addi 12,12,1 - stb 6,0(3) - addi 3,3,1 - - .align 4 + /* Force 4-byte alignment for SRC. */ + mtocrf 0x01,0 + subf cnt,0,cnt +2: + bf 30,1f + lhz 6,0(src) + addi src,src,2 + sth 6,0(dst) + addi dst,dst,2 +1: + bf 31,L(end_4bytes_alignment) + lbz 6,0(src) + addi src,src,1 + stb 6,0(dst) + addi dst,dst,1 + + .align 4 L(end_4bytes_alignment): - cmpldi cr1,10,16 - mtcrf 0x01,10 + cmpldi cr1,cnt,16 + mtocrf 0x01,cnt L(copy_LT_32_aligned): /* At least 6 bytes to go, and SRC is word-aligned. */ blt cr1,8f /* Copy 16 bytes. */ - lwz 6,0(12) - lwz 7,4(12) - stw 6,0(3) - lwz 8,8(12) - stw 7,4(3) - lwz 6,12(12) - addi 12,12,16 - stw 8,8(3) - stw 6,12(3) - addi 3,3,16 + lwz 6,0(src) + lwz 7,4(src) + stw 6,0(dst) + lwz 8,8(src) + stw 7,4(dst) + lwz 6,12(src) + addi src,src,16 + stw 8,8(dst) + stw 6,12(dst) + addi dst,dst,16 8: /* Copy 8 bytes. */ - bf 28,4f + bf 28,L(tail4) + lwz 6,0(src) + lwz 7,4(src) + addi src,src,8 + stw 6,0(dst) + stw 7,4(dst) + addi dst,dst,8 + + .align 4 +/* Copies 4~7 bytes. */ +L(tail4): + bf 29,L(tail2) + lwz 6,0(src) + stw 6,0(dst) + bf 30,L(tail5) + lhz 7,4(src) + sth 7,4(dst) + bflr 31 + lbz 8,6(src) + stb 8,6(dst) + /* Return original DST pointer. */ + blr - lwz 6,0(12) - lwz 7,4(12) - addi 12,12,8 - stw 6,0(3) - stw 7,4(3) - addi 3,3,8 -4: /* Copy 4 bytes. */ - bf 29,2f - - lwz 6,0(12) - addi 12,12,4 - stw 6,0(3) - addi 3,3,4 -2: /* Copy 2-3 bytes. */ + .align 4 +/* Copies 2~3 bytes. */ +L(tail2): bf 30,1f - - lhz 6,0(12) - sth 6,0(3) - bf 31,0f - lbz 7,2(12) - stb 7,2(3) - ld 3,-16(1) + lhz 6,0(src) + sth 6,0(dst) + bflr 31 + lbz 7,2(src) + stb 7,2(dst) blr - .align 4 -1: /* Copy 1 byte. */ - bf 31,0f + .align 4 +L(tail5): + bflr 31 + lbz 6,4(src) + stb 6,4(dst) + blr - lbz 6,0(12) - stb 6,0(3) -0: /* Return original DST pointer. */ - ld 3,-16(1) + .align 4 +1: + bflr 31 + lbz 6,0(src) + stb 6,0(dst) + /* Return original DST pointer. */ blr - /* Handles copies of 0~8 bytes. */ - .align 4 + +/* Handles copies of 0~8 bytes. */ + .align 4 L(copy_LE_8): - bne cr6,4f + bne cr6,L(tail4) /* Though we could've used ld/std here, they are still slow for unaligned cases. */ - lwz 6,0(4) - lwz 7,4(4) - stw 6,0(3) - stw 7,4(3) - ld 3,-16(1) /* Return original DST pointers. */ + lwz 6,0(src) + lwz 7,4(src) + stw 6,0(dst) + stw 7,4(dst) blr - .align 4 -4: /* Copies 4~7 bytes. */ - bf 29,2b - - lwz 6,0(4) - stw 6,0(3) - bf 30,5f - lhz 7,4(4) - sth 7,4(3) - bf 31,0f - lbz 8,6(4) - stb 8,6(3) - ld 3,-16(1) - blr - - .align 4 -5: /* Copy 1 byte. */ - bf 31,0f - - lbz 6,4(4) - stb 6,4(3) - -0: /* Return original DST pointer. */ - ld 3,-16(1) - blr - /* Handle copies of 32+ bytes where DST is aligned (to quadword) but - SRC is not. Use aligned quadword loads from SRC, shifted to realign - the data, allowing for aligned DST stores. */ - .align 4 +/* Handle copies of 32+ bytes where DST is aligned (to quadword) but + SRC is not. Use aligned quadword loads from SRC, shifted to realign + the data, allowing for aligned DST stores. */ + .align 4 L(copy_GE_32_unaligned): - clrldi 0,0,60 /* Number of bytes until the 1st - quadword. */ - andi. 11,3,15 /* Check alignment of DST (against - quadwords). */ - srdi 9,5,4 /* Number of full quadwords remaining. */ + clrldi 0,0,60 /* Number of bytes until the 1st dst quadword. */ +#ifndef __LITTLE_ENDIAN__ + andi. 10,3,15 /* Check alignment of DST (against quadwords). */ +#endif + srdi 9,cnt,4 /* Number of full quadwords remaining. */ beq L(copy_GE_32_unaligned_cont) - /* SRC is not quadword aligned, get it aligned. */ + /* DST is not quadword aligned, get it aligned. */ - mtcrf 0x01,0 - subf 31,0,5 + mtocrf 0x01,0 + subf cnt,0,cnt /* Vector instructions work best when proper alignment (16-bytes) is present. Move 0~15 bytes as needed to get DST quadword-aligned. */ -1: /* Copy 1 byte. */ +1: bf 31,2f - - lbz 6,0(12) - addi 12,12,1 - stb 6,0(3) - addi 3,3,1 -2: /* Copy 2 bytes. */ + lbz 6,0(src) + addi src,src,1 + stb 6,0(dst) + addi dst,dst,1 +2: bf 30,4f - - lhz 6,0(12) - addi 12,12,2 - sth 6,0(3) - addi 3,3,2 -4: /* Copy 4 bytes. */ + lhz 6,0(src) + addi src,src,2 + sth 6,0(dst) + addi dst,dst,2 +4: bf 29,8f - - lwz 6,0(12) - addi 12,12,4 - stw 6,0(3) - addi 3,3,4 -8: /* Copy 8 bytes. */ + lwz 6,0(src) + addi src,src,4 + stw 6,0(dst) + addi dst,dst,4 +8: bf 28,0f - - ld 6,0(12) - addi 12,12,8 - std 6,0(3) - addi 3,3,8 + ld 6,0(src) + addi src,src,8 + std 6,0(dst) + addi dst,dst,8 0: - clrldi 10,12,60 /* Check alignment of SRC. */ - srdi 9,31,4 /* Number of full quadwords remaining. */ + srdi 9,cnt,4 /* Number of full quadwords remaining. */ /* The proper alignment is present, it is OK to copy the bytes now. */ L(copy_GE_32_unaligned_cont): /* Setup two indexes to speed up the indexed vector operations. */ - clrldi 11,31,60 - li 6,16 /* Index for 16-bytes offsets. */ + clrldi 10,cnt,60 + li 6,16 /* Index for 16-bytes offsets. */ li 7,32 /* Index for 32-bytes offsets. */ - cmpldi cr1,11,0 - srdi 8,31,5 /* Setup the loop counter. */ - mr 10,3 - mr 11,12 - mtcrf 0x01,9 - cmpldi cr6,9,1 - lvsl 5,0,12 - lvx 3,0,12 - bf 31,L(setup_unaligned_loop) - - /* Copy another 16 bytes to align to 32-bytes due to the loop . */ - lvx 4,12,6 - vperm 6,3,4,5 - addi 11,12,16 - addi 10,3,16 - stvx 6,0,3 + cmpldi cr1,10,0 + srdi 8,cnt,5 /* Setup the loop counter. */ + mtocrf 0x01,9 + cmpldi cr6,9,1 +#ifdef __LITTLE_ENDIAN__ + lvsr 5,0,src +#else + lvsl 5,0,src +#endif + lvx 3,0,src + li 0,0 + bf 31,L(setup_unaligned_loop) + + /* Copy another 16 bytes to align to 32-bytes due to the loop. */ + lvx 4,src,6 +#ifdef __LITTLE_ENDIAN__ + vperm 6,4,3,5 +#else + vperm 6,3,4,5 +#endif + addi src,src,16 + stvx 6,0,dst + addi dst,dst,16 vor 3,4,4 + clrrdi 0,src,60 L(setup_unaligned_loop): - mtctr 8 - ble cr6,L(end_unaligned_loop) + mtctr 8 + ble cr6,L(end_unaligned_loop) /* Copy 32 bytes at a time using vector instructions. */ - .align 4 + .align 4 L(unaligned_loop): /* Note: vr6/vr10 may contain data that was already copied, @@ -442,62 +385,55 @@ L(unaligned_loop): some portions again. This is faster than having unaligned vector instructions though. */ - lvx 4,11,6 /* vr4 = r11+16. */ - vperm 6,3,4,5 /* Merge the correctly-aligned portions - of vr3/vr4 into vr6. */ - lvx 3,11,7 /* vr3 = r11+32. */ - vperm 10,4,3,5 /* Merge the correctly-aligned portions - of vr3/vr4 into vr10. */ - addi 11,11,32 - stvx 6,0,10 - stvx 10,10,6 - addi 10,10,32 - + lvx 4,src,6 +#ifdef __LITTLE_ENDIAN__ + vperm 6,4,3,5 +#else + vperm 6,3,4,5 +#endif + lvx 3,src,7 +#ifdef __LITTLE_ENDIAN__ + vperm 10,3,4,5 +#else + vperm 10,4,3,5 +#endif + addi src,src,32 + stvx 6,0,dst + stvx 10,dst,6 + addi dst,dst,32 bdnz L(unaligned_loop) - .align 4 + clrrdi 0,src,60 + + .align 4 L(end_unaligned_loop): /* Check for tail bytes. */ - rldicr 0,31,0,59 - mtcrf 0x01,31 - beq cr1,0f + mtocrf 0x01,cnt + beqlr cr1 - add 3,3,0 - add 12,12,0 + add src,src,0 /* We have 1~15 tail bytes to copy, and DST is quadword aligned. */ -8: /* Copy 8 bytes. */ + /* Copy 8 bytes. */ bf 28,4f - - lwz 6,0(12) - lwz 7,4(12) - addi 12,12,8 - stw 6,0(3) - stw 7,4(3) - addi 3,3,8 -4: /* Copy 4 bytes. */ - bf 29,2f - - lwz 6,0(12) - addi 12,12,4 - stw 6,0(3) - addi 3,3,4 -2: /* Copy 2~3 bytes. */ - bf 30,1f - - lhz 6,0(12) - addi 12,12,2 - sth 6,0(3) - addi 3,3,2 -1: /* Copy 1 byte. */ - bf 31,0f - - lbz 6,0(12) - stb 6,0(3) -0: /* Return original DST pointer. */ - ld 31,-8(1) - ld 3,-16(1) + lwz 6,0(src) + lwz 7,4(src) + addi src,src,8 + stw 6,0(dst) + stw 7,4(dst) + addi dst,dst,8 +4: /* Copy 4~7 bytes. */ + bf 29,L(tail2) + lwz 6,0(src) + stw 6,0(dst) + bf 30,L(tail5) + lhz 7,4(src) + sth 7,4(dst) + bflr 31 + lbz 8,6(src) + stb 8,6(dst) + /* Return original DST pointer. */ blr END_GEN_TB (memcpy,TB_TOCLESS) diff --git a/sysdeps/powerpc/powerpc64/power7/mempcpy.S b/sysdeps/powerpc/powerpc64/power7/mempcpy.S index f20be938d2..b93ab7da52 100644 --- a/sysdeps/powerpc/powerpc64/power7/mempcpy.S +++ b/sysdeps/powerpc/powerpc64/power7/mempcpy.S @@ -365,13 +365,21 @@ L(copy_GE_32_unaligned_cont): mr 11,12 mtcrf 0x01,9 cmpldi cr6,9,1 - lvsl 5,0,12 +#ifdef __LITTLE_ENDIAN__ + lvsr 5,0,12 +#else + lvsl 5,0,12 +#endif lvx 3,0,12 bf 31,L(setup_unaligned_loop) /* Copy another 16 bytes to align to 32-bytes due to the loop . */ lvx 4,12,6 - vperm 6,3,4,5 +#ifdef __LITTLE_ENDIAN__ + vperm 6,4,3,5 +#else + vperm 6,3,4,5 +#endif addi 11,12,16 addi 10,3,16 stvx 6,0,3 @@ -391,11 +399,17 @@ L(unaligned_loop): vector instructions though. */ lvx 4,11,6 /* vr4 = r11+16. */ - vperm 6,3,4,5 /* Merge the correctly-aligned portions - of vr3/vr4 into vr6. */ +#ifdef __LITTLE_ENDIAN__ + vperm 6,4,3,5 +#else + vperm 6,3,4,5 +#endif lvx 3,11,7 /* vr3 = r11+32. */ - vperm 10,4,3,5 /* Merge the correctly-aligned portions - of vr3/vr4 into vr10. */ +#ifdef __LITTLE_ENDIAN__ + vperm 10,3,4,5 +#else + vperm 10,4,3,5 +#endif addi 11,11,32 stvx 6,0,10 stvx 10,10,6 diff --git a/sysdeps/powerpc/powerpc64/power7/memrchr.S b/sysdeps/powerpc/powerpc64/power7/memrchr.S index d24fbbb1b9..683bfed7d8 100644 --- a/sysdeps/powerpc/powerpc64/power7/memrchr.S +++ b/sysdeps/powerpc/powerpc64/power7/memrchr.S @@ -22,119 +22,133 @@ /* int [r3] memrchr (char *s [r3], int byte [r4], int size [r5]) */ .machine power7 ENTRY (__memrchr) - CALL_MCOUNT - dcbt 0,r3 - mr r7,r3 - add r3,r7,r5 /* Calculate the last acceptable address. */ - cmpld cr7,r3,r7 /* Is the address equal or less than r3? */ + CALL_MCOUNT 3 + add r7,r3,r5 /* Calculate the last acceptable address. */ + neg r0,r7 + addi r7,r7,-1 + mr r10,r3 + clrrdi r6,r7,7 + li r9,3<<5 + dcbt r9,r6,16 /* Stream hint, decreasing addresses. */ /* Replicate BYTE to doubleword. */ - rlwimi r4,r4,8,16,23 - rlwimi r4,r4,16,0,15 + insrdi r4,r4,8,48 + insrdi r4,r4,16,32 insrdi r4,r4,32,0 - bge cr7,L(proceed) - - li r3,-1 /* Make r11 the biggest if r4 <= 0. */ -L(proceed): li r6,-8 - addi r9,r3,-1 - clrrdi r8,r9,3 - addi r8,r8,8 - neg r0,r3 + li r9,-1 rlwinm r0,r0,3,26,28 /* Calculate padding. */ - + clrrdi r8,r7,3 + srd r9,r9,r0 cmpldi r5,32 + clrrdi r0,r10,3 ble L(small_range) - ldbrx r12,r8,r6 /* Load reversed doubleword from memory. */ - cmpb r10,r12,r4 /* Check for BYTE in DWORD1. */ - sld r10,r10,r0 - srd r10,r10,r0 - cmpldi cr7,r10,0 /* If r10 == 0, no BYTEs have been found. */ +#ifdef __LITTLE_ENDIAN__ + ldx r12,0,r8 +#else + ldbrx r12,0,r8 /* Load reversed doubleword from memory. */ +#endif + cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */ + and r3,r3,r9 + cmpldi cr7,r3,0 /* If r3 == 0, no BYTEs have been found. */ bne cr7,L(done) - /* Are we done already? */ - addi r9,r8,-8 - cmpld cr6,r9,r7 - ble cr6,L(null) - mtcrf 0x01,r8 - /* Are we now aligned to a doubleword boundary? If so, skip to + /* Are we now aligned to a quadword boundary? If so, skip to the main loop. Otherwise, go through the alignment code. */ - mr r8,r9 - bt 28,L(loop_setup) + bf 28,L(loop_setup) /* Handle DWORD2 of pair. */ +#ifdef __LITTLE_ENDIAN__ + ldx r12,r8,r6 +#else ldbrx r12,r8,r6 - cmpb r10,r12,r4 - cmpldi cr7,r10,0 - bne cr7,L(done) - - /* Are we done already. */ +#endif addi r8,r8,-8 - cmpld cr6,r8,r7 - ble cr6,L(null) + cmpb r3,r12,r4 + cmpldi cr7,r3,0 + bne cr7,L(done) L(loop_setup): - li r0,-16 - sub r5,r8,r7 - srdi r9,r5,4 /* Number of loop iterations. */ + /* The last dword we want to read in the loop below is the one + containing the first byte of the string, ie. the dword at + s & ~7, or r0. The first dword read is at r8 - 8, we + read 2 * cnt dwords, so the last dword read will be at + r8 - 8 - 16 * cnt + 8. Solving for cnt gives + cnt = (r8 - r0) / 16 */ + sub r5,r8,r0 + addi r8,r8,-8 + srdi r9,r5,4 /* Number of loop iterations. */ mtctr r9 /* Setup the counter. */ - b L(loop) - /* Main loop to look for BYTE backwards in the string. Since it's a - small loop (< 8 instructions), align it to 32-bytes. */ - .p2align 5 + + /* Main loop to look for BYTE backwards in the string. + FIXME: Investigate whether 32 byte align helps with this + 9 instruction loop. */ + .align 5 L(loop): /* Load two doublewords, compare and merge in a single register for speed. This is an attempt to speed up the byte-checking process for bigger strings. */ - ldbrx r12,r8,r6 - ldbrx r11,r8,r0 - addi r8,r8,-8 - cmpb r10,r12,r4 +#ifdef __LITTLE_ENDIAN__ + ldx r12,0,r8 + ldx r11,r8,r6 +#else + ldbrx r12,0,r8 + ldbrx r11,r8,r6 +#endif + cmpb r3,r12,r4 cmpb r9,r11,r4 - or r5,r9,r10 /* Merge everything in one doubleword. */ + or r5,r9,r3 /* Merge everything in one doubleword. */ cmpldi cr7,r5,0 bne cr7,L(found) - addi r8,r8,-8 + addi r8,r8,-16 bdnz L(loop) - /* We're here because the counter reached 0, and that means we - didn't have any matches for BYTE in the whole range. Just return - the original range. */ - addi r9,r8,8 - cmpld cr6,r9,r7 - bgt cr6,L(loop_small) - b L(null) - - /* OK, one (or both) of the words contains BYTE. Check - the first word and decrement the address in case the first - word really contains BYTE. */ + + /* We may have one more word to read. */ + cmpld r8,r0 + bnelr + +#ifdef __LITTLE_ENDIAN__ + ldx r12,0,r8 +#else + ldbrx r12,0,r8 +#endif + cmpb r3,r12,r4 + cmpldi cr7,r3,0 + bne cr7,L(done) + blr + .align 4 L(found): - cmpldi cr6,r10,0 - addi r8,r8,8 + /* OK, one (or both) of the dwords contains BYTE. Check + the first dword. */ + cmpldi cr6,r3,0 bne cr6,L(done) /* BYTE must be in the second word. Adjust the address - again and move the result of cmpb to r10 so we can calculate the + again and move the result of cmpb to r3 so we can calculate the pointer. */ - mr r10,r9 + mr r3,r9 addi r8,r8,-8 - /* r10 has the output of the cmpb instruction, that is, it contains - 0xff in the same position as the BYTE in the original + /* r3 has the output of the cmpb instruction, that is, it contains + 0xff in the same position as BYTE in the original word from the string. Use that to calculate the pointer. We need to make sure BYTE is *before* the end of the range. */ L(done): - cntlzd r0,r10 /* Count leading zeroes before the match. */ - srdi r6,r0,3 /* Convert leading zeroes to bytes. */ - addi r0,r6,1 + cntlzd r9,r3 /* Count leading zeros before the match. */ + cmpld r8,r0 /* Are we on the last word? */ + srdi r6,r9,3 /* Convert leading zeros to bytes. */ + addi r0,r6,-7 sub r3,r8,r0 - cmpld r3,r7 - blt L(null) + cmpld cr7,r3,r10 + bnelr + bgelr cr7 + li r3,0 blr .align 4 @@ -148,29 +162,35 @@ L(small_range): cmpldi r5,0 beq L(null) - ldbrx r12,r8,r6 /* Load reversed doubleword from memory. */ - cmpb r10,r12,r4 /* Check for BYTE in DWORD1. */ - sld r10,r10,r0 - srd r10,r10,r0 - cmpldi cr7,r10,0 +#ifdef __LITTLE_ENDIAN__ + ldx r12,0,r8 +#else + ldbrx r12,0,r8 /* Load reversed doubleword from memory. */ +#endif + cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */ + and r3,r3,r9 + cmpldi cr7,r3,0 bne cr7,L(done) /* Are we done already? */ + cmpld r8,r0 addi r8,r8,-8 - cmpld r8,r7 - ble L(null) - b L(loop_small) + beqlr - .p2align 5 + .align 5 L(loop_small): - ldbrx r12,r8,r6 - cmpb r10,r12,r4 - cmpldi cr6,r10,0 - bne cr6,L(done) +#ifdef __LITTLE_ENDIAN__ + ldx r12,0,r8 +#else + ldbrx r12,0,r8 +#endif + cmpb r3,r12,r4 + cmpld r8,r0 + cmpldi cr7,r3,0 + bne cr7,L(done) addi r8,r8,-8 - cmpld r8,r7 - ble L(null) - b L(loop_small) + bne L(loop_small) + blr END (__memrchr) weak_alias (__memrchr, memrchr) diff --git a/sysdeps/powerpc/powerpc64/power7/memset.S b/sysdeps/powerpc/powerpc64/power7/memset.S index b24cfa163a..03a45c23c7 100644 --- a/sysdeps/powerpc/powerpc64/power7/memset.S +++ b/sysdeps/powerpc/powerpc64/power7/memset.S @@ -32,8 +32,8 @@ L(_memset): mr 10,3 /* Replicate byte to word. */ - rlwimi 4,4,8,16,23 - rlwimi 4,4,16,0,15 + insrdi 4,4,8,48 + insrdi 4,4,16,32 ble cr6,L(small) /* If length <= 8, use short copy code. */ neg 0,3 @@ -321,7 +321,7 @@ L(medium): clrldi 0,0,62 beq L(medium_aligned) - /* Force 4-bytes alignment for SRC. */ + /* Force 4-bytes alignment for DST. */ mtocrf 0x01,0 subf 5,0,5 1: /* Copy 1 byte. */ @@ -383,6 +383,7 @@ L(small): END_GEN_TB (memset,TB_TOCLESS) libc_hidden_builtin_def (memset) +#ifndef NO_BZERO_IMPL /* Copied from bzero.S to prevent the linker from inserting a stub between bzero and memset. */ ENTRY (__bzero) @@ -393,3 +394,4 @@ ENTRY (__bzero) END_GEN_TB (__bzero,TB_TOCLESS) weak_alias (__bzero, bzero) +#endif diff --git a/sysdeps/powerpc/powerpc64/power7/multiarch/Implies b/sysdeps/powerpc/powerpc64/power7/multiarch/Implies new file mode 100644 index 0000000000..bf5d6171a5 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power6/multiarch diff --git a/sysdeps/powerpc/powerpc64/power7/rawmemchr.S b/sysdeps/powerpc/powerpc64/power7/rawmemchr.S index 50a33d8fae..547aed771f 100644 --- a/sysdeps/powerpc/powerpc64/power7/rawmemchr.S +++ b/sysdeps/powerpc/powerpc64/power7/rawmemchr.S @@ -27,8 +27,8 @@ ENTRY (__rawmemchr) clrrdi r8,r3,3 /* Align the address to doubleword boundary. */ /* Replicate byte to doubleword. */ - rlwimi r4,r4,8,16,23 - rlwimi r4,r4,16,0,15 + insrdi r4,r4,8,48 + insrdi r4,r4,16,32 insrdi r4,r4,32,0 /* Now r4 has a doubleword of c bytes. */ @@ -36,8 +36,13 @@ ENTRY (__rawmemchr) rlwinm r6,r3,3,26,28 /* Calculate padding. */ ld r12,0(r8) /* Load doubleword from memory. */ cmpb r5,r12,r4 /* Compare each byte against c byte. */ +#ifdef __LITTLE_ENDIAN__ + srd r5,r5,r6 + sld r5,r5,r6 +#else sld r5,r5,r6 /* Move left to discard ignored bits. */ srd r5,r5,r6 /* Bring the bits back as zeros. */ +#endif cmpdi cr7,r5,0 /* If r5 == 0, no c bytes have been found. */ bne cr7,L(done) @@ -91,8 +96,14 @@ L(loop): doubleword from the string. Use that fact to find out what is the position of the byte inside the string. */ L(done): +#ifdef __LITTLE_ENDIAN__ + addi r0,r5,-1 + andc r0,r0,r5 + popcntd r0,r0 /* Count trailing zeros. */ +#else cntlzd r0,r5 /* Count leading zeros before the match. */ - srdi r0,r0,3 /* Convert leading zeroes to bytes. */ +#endif + srdi r0,r0,3 /* Convert leading zeros to bytes. */ add r3,r8,r0 /* Return address of the matching char. */ blr END (__rawmemchr) diff --git a/sysdeps/powerpc/powerpc64/power7/stpcpy.S b/sysdeps/powerpc/powerpc64/power7/stpcpy.S new file mode 100644 index 0000000000..727dd06e74 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/stpcpy.S @@ -0,0 +1,24 @@ +/* Optimized stpcpy implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#define USE_AS_STPCPY +#include <sysdeps/powerpc/powerpc64/power7/strcpy.S> + +weak_alias (__stpcpy, stpcpy) +libc_hidden_def (__stpcpy) +libc_hidden_builtin_def (stpcpy) diff --git a/sysdeps/powerpc/powerpc64/power7/strcasecmp.S b/sysdeps/powerpc/powerpc64/power7/strcasecmp.S index 9eee384692..37785ea423 100644 --- a/sysdeps/powerpc/powerpc64/power7/strcasecmp.S +++ b/sysdeps/powerpc/powerpc64/power7/strcasecmp.S @@ -32,7 +32,11 @@ #endif ENTRY (__STRCMP) +#ifndef USE_IN_EXTENDED_LOCALE_MODEL CALL_MCOUNT 2 +#else + CALL_MCOUNT 3 +#endif #define rRTN r3 /* Return value */ #define rSTR1 r5 /* 1st string */ diff --git a/sysdeps/powerpc/powerpc64/power7/strchr.S b/sysdeps/powerpc/powerpc64/power7/strchr.S index 3ffe7a1887..4679a158f1 100644 --- a/sysdeps/powerpc/powerpc64/power7/strchr.S +++ b/sysdeps/powerpc/powerpc64/power7/strchr.S @@ -35,8 +35,8 @@ ENTRY (strchr) beq cr7,L(null_match) /* Replicate byte to doubleword. */ - rlwimi r4,r4,8,16,23 - rlwimi r4,r4,16,0,15 + insrdi r4,r4,8,48 + insrdi r4,r4,16,32 insrdi r4,r4,32,0 /* Now r4 has a doubleword of c bytes and r0 has @@ -47,11 +47,17 @@ ENTRY (strchr) /* Move the doublewords left and right to discard the bits that are not part of the string and bring them back as zeros. */ - +#ifdef __LITTLE_ENDIAN__ + srd r10,r10,r6 + srd r11,r11,r6 + sld r10,r10,r6 + sld r11,r11,r6 +#else sld r10,r10,r6 sld r11,r11,r6 srd r10,r10,r6 srd r11,r11,r6 +#endif or r5,r10,r11 /* OR the results to speed things up. */ cmpdi cr7,r5,0 /* If r5 == 0, no c or null bytes have been found. */ @@ -108,15 +114,24 @@ L(loop): mr r11,r7 addi r8,r8,8 - /* r5 has the output of the cmpb instruction, that is, it contains + /* r10/r11 have the output of the cmpb instructions, that is, 0xff in the same position as the c/null byte in the original doubleword from the string. Use that to calculate the pointer. */ L(done): - cntlzd r4,r10 /* Count leading zeroes before c matches. */ - cntlzd r0,r11 /* Count leading zeroes before null matches. */ - cmpld cr7,r4,r0 +#ifdef __LITTLE_ENDIAN__ + addi r3,r10,-1 + andc r3,r3,r10 + popcntd r0,r3 + addi r4,r11,-1 + andc r4,r4,r11 + cmpld cr7,r3,r4 bgt cr7,L(no_match) - srdi r0,r4,3 /* Convert leading zeroes to bytes. */ +#else + cntlzd r0,r10 /* Count leading zeros before c matches. */ + cmpld cr7,r11,r10 + bgt cr7,L(no_match) +#endif + srdi r0,r0,3 /* Convert leading zeros to bytes. */ add r3,r8,r0 /* Return address of the matching c byte or null in case c was not found. */ blr @@ -135,9 +150,13 @@ L(null_match): /* Move the doublewords left and right to discard the bits that are not part of the string and bring them back as zeros. */ - +#ifdef __LITTLE_ENDIAN__ + srd r5,r5,r6 + sld r5,r5,r6 +#else sld r5,r5,r6 srd r5,r5,r6 +#endif cmpdi cr7,r5,0 /* If r10 == 0, no c or null bytes have been found. */ bne cr7,L(done_null) @@ -192,7 +211,13 @@ L(loop_null): 0xff in the same position as the null byte in the original doubleword from the string. Use that to calculate the pointer. */ L(done_null): +#ifdef __LITTLE_ENDIAN__ + addi r0,r5,-1 + andc r0,r0,r5 + popcntd r0,r0 +#else cntlzd r0,r5 /* Count leading zeros before the match. */ +#endif srdi r0,r0,3 /* Convert leading zeros to bytes. */ add r3,r8,r0 /* Return address of the matching null byte. */ blr diff --git a/sysdeps/powerpc/powerpc64/power7/strchrnul.S b/sysdeps/powerpc/powerpc64/power7/strchrnul.S index 9dbc51b0d1..df457525e2 100644 --- a/sysdeps/powerpc/powerpc64/power7/strchrnul.S +++ b/sysdeps/powerpc/powerpc64/power7/strchrnul.S @@ -27,8 +27,8 @@ ENTRY (__strchrnul) clrrdi r8,r3,3 /* Align the address to doubleword boundary. */ /* Replicate byte to doubleword. */ - rlwimi r4,r4,8,16,23 - rlwimi r4,r4,16,0,15 + insrdi r4,r4,8,48 + insrdi r4,r4,16,32 insrdi r4,r4,32,0 rlwinm r6,r3,3,26,28 /* Calculate padding. */ @@ -44,10 +44,17 @@ ENTRY (__strchrnul) /* Move the doublewords left and right to discard the bits that are not part of the string and to bring them back as zeros. */ +#ifdef __LITTLE_ENDIAN__ + srd r10,r10,r6 + srd r9,r9,r6 + sld r10,r10,r6 + sld r9,r9,r6 +#else sld r10,r10,r6 sld r9,r9,r6 srd r10,r10,r6 srd r9,r9,r6 +#endif or r5,r9,r10 /* OR the results to speed things up. */ cmpdi cr7,r5,0 /* If r5 == 0, no c or null bytes have been found. */ @@ -97,7 +104,7 @@ L(loop): bne cr6,L(done) /* The c/null byte must be in the second doubleword. Adjust the - address again and move the result of cmpb to r10 so we can calculate + address again and move the result of cmpb to r5 so we can calculate the pointer. */ mr r5,r10 addi r8,r8,8 @@ -106,7 +113,13 @@ L(loop): 0xff in the same position as the c/null byte in the original doubleword from the string. Use that to calculate the pointer. */ L(done): +#ifdef __LITTLE_ENDIAN__ + addi r0,r5,-1 + andc r0,r0,r5 + popcntd r0,r0 +#else cntlzd r0,r5 /* Count leading zeros before the match. */ +#endif srdi r0,r0,3 /* Convert leading zeros to bytes. */ add r3,r8,r0 /* Return address of matching c/null byte. */ blr diff --git a/sysdeps/powerpc/powerpc64/power7/strcpy.S b/sysdeps/powerpc/powerpc64/power7/strcpy.S new file mode 100644 index 0000000000..5c341a1483 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/strcpy.S @@ -0,0 +1,274 @@ +/* Optimized strcpy/stpcpy implementation for PowerPC64/POWER7. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +/* Implements the function + + char * [r3] strcpy (char *dest [r3], const char *src [r4]) + + or + + char * [r3] strcpy (char *dest [r3], const char *src [r4]) + + if USE_AS_STPCPY is defined. It tries to use aligned memory accesses + when possible using the following algorithm: + + if (((((uintptr_t)dst & 0x7UL) == 0) && ((uintptr_t)src & 0x7UL) == 0)) + goto aligned_doubleword_copy; + if (((((uintptr_t)dst & 0x3UL) == 0) && ((uintptr_t)src & 0x3UL) == 0)) + goto aligned_word_copy; + if (((uintptr_t)dst & 0x7UL) == ((uintptr_t)src & 0x7UL)) + goto same_alignment; + goto unaligned; + + The aligned comparison are made using cmpb instructions. */ + +#ifdef USE_AS_STPCPY +# define FUNC_NAME __stpcpy +#else +# define FUNC_NAME strcpy +#endif + + .machine power7 +EALIGN (FUNC_NAME, 4, 0) + CALL_MCOUNT 2 + +#define rTMP r0 +#ifdef USE_AS_STPCPY +#define rRTN r3 /* pointer to previous word/doubleword in dest */ +#else +#define rRTN r12 /* pointer to previous word/doubleword in dest */ +#endif +#define rSRC r4 /* pointer to previous word/doubleword in src */ +#define rMASK r5 /* mask 0xffffffff | 0xffffffffffffffff */ +#define rWORD r6 /* current word from src */ +#define rALT r7 /* alternate word from src */ +#define rRTNAL r8 /* alignment of return pointer */ +#define rSRCAL r9 /* alignment of source pointer */ +#define rALCNT r10 /* bytes to read to reach 8 bytes alignment */ +#define rSUBAL r11 /* doubleword minus unaligned displacement */ + +#ifndef USE_AS_STPCPY +/* Save the dst pointer to use as return value. */ + mr rRTN, r3 +#endif + or rTMP, rSRC, rRTN + clrldi. rTMP, rTMP, 61 + bne L(check_word_alignment) + b L(aligned_doubleword_copy) + +L(same_alignment): +/* Src and dst with same alignment: align both to doubleword. */ + mr rALCNT, rRTN + lbz rWORD, 0(rSRC) + subfic rSUBAL, rRTNAL, 8 + addi rRTN, rRTN, 1 + addi rSRC, rSRC, 1 + cmpdi cr7, rWORD, 0 + stb rWORD, 0(rALCNT) + beq cr7, L(s2) + + add rALCNT, rALCNT, rSUBAL + subf rALCNT, rRTN, rALCNT + addi rALCNT, rALCNT, 1 + mtctr rALCNT + b L(s1) + + .align 4 +L(s0): + addi rSRC, rSRC, 1 + lbz rWORD, -1(rSRC) + cmpdi cr7, rWORD, 0 + stb rWORD, -1(rALCNT) + beqlr cr7 + mr rRTN, rALCNT +L(s1): + addi rALCNT, rRTN,1 + bdnz L(s0) + b L(aligned_doubleword_copy) + .align 4 +L(s2): + mr rRTN, rALCNT + blr + +/* For doubleword aligned memory, operate using doubleword load and stores. */ + .align 4 +L(aligned_doubleword_copy): + li rMASK, 0 + addi rRTN, rRTN, -8 + ld rWORD, 0(rSRC) + b L(g2) + + .align 4 +L(g0): ldu rALT, 8(rSRC) + stdu rWORD, 8(rRTN) + cmpb rTMP, rALT, rMASK + cmpdi rTMP, 0 + bne L(g1) + ldu rWORD, 8(rSRC) + stdu rALT, 8(rRTN) +L(g2): cmpb rTMP, rWORD, rMASK + cmpdi rTMP, 0 /* If rTMP is 0, no null's have been found. */ + beq L(g0) + + mr rALT, rWORD +/* We've hit the end of the string. Do the rest byte-by-byte. */ +L(g1): +#ifdef __LITTLE_ENDIAN__ + extrdi. rTMP, rALT, 8, 56 + stbu rALT, 8(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 48 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 40 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 32 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 24 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 16 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 8 + stbu rTMP, 1(rRTN) + beqlr- + extrdi rTMP, rALT, 8, 0 + stbu rTMP, 1(rRTN) +#else + extrdi. rTMP, rALT, 8, 0 + stbu rTMP, 8(rRTN) + beqlr + extrdi. rTMP, rALT, 8, 8 + stbu rTMP, 1(rRTN) + beqlr + extrdi. rTMP, rALT, 8, 16 + stbu rTMP, 1(rRTN) + beqlr + extrdi. rTMP, rALT, 8, 24 + stbu rTMP, 1(rRTN) + beqlr + extrdi. rTMP, rALT, 8, 32 + stbu rTMP, 1(rRTN) + beqlr + extrdi. rTMP, rALT, 8, 40 + stbu rTMP, 1(rRTN) + beqlr + extrdi. rTMP, rALT, 8, 48 + stbu rTMP, 1(rRTN) + beqlr + stbu rALT, 1(rRTN) +#endif + blr + +L(check_word_alignment): + clrldi. rTMP, rTMP, 62 + beq L(aligned_word_copy) + rldicl rRTNAL, rRTN, 0, 61 + rldicl rSRCAL, rSRC, 0, 61 + cmpld cr7, rSRCAL, rRTNAL + beq cr7, L(same_alignment) + b L(unaligned) + +/* For word aligned memory, operate using word load and stores. */ + .align 4 +L(aligned_word_copy): + li rMASK, 0 + addi rRTN, rRTN, -4 + lwz rWORD, 0(rSRC) + b L(g5) + + .align 4 +L(g3): lwzu rALT, 4(rSRC) + stwu rWORD, 4(rRTN) + cmpb rTMP, rALT, rMASK + cmpwi rTMP, 0 + bne L(g4) + lwzu rWORD, 4(rSRC) + stwu rALT, 4(rRTN) +L(g5): cmpb rTMP, rWORD, rMASK + cmpwi rTMP, 0 /* If rTMP is 0, no null in word. */ + beq L(g3) + + mr rALT, rWORD +/* We've hit the end of the string. Do the rest byte-by-byte. */ +L(g4): +#ifdef __LITTLE_ENDIAN__ + rlwinm. rTMP, rALT, 0, 24, 31 + stbu rALT, 4(rRTN) + beqlr- + rlwinm. rTMP, rALT, 24, 24, 31 + stbu rTMP, 1(rRTN) + beqlr- + rlwinm. rTMP, rALT, 16, 24, 31 + stbu rTMP, 1(rRTN) + beqlr- + rlwinm rTMP, rALT, 8, 24, 31 + stbu rTMP, 1(rRTN) +#else + rlwinm. rTMP, rALT, 8, 24, 31 + stbu rTMP, 4(rRTN) + beqlr + rlwinm. rTMP, rALT, 16, 24, 31 + stbu rTMP, 1(rRTN) + beqlr + rlwinm. rTMP, rALT, 24, 24, 31 + stbu rTMP, 1(rRTN) + beqlr + stbu rALT, 1(rRTN) +#endif + blr + +/* Oh well. In this case, we just do a byte-by-byte copy. */ + .align 4 +L(unaligned): + lbz rWORD, 0(rSRC) + addi rRTN, rRTN, -1 + cmpdi rWORD, 0 + beq L(u2) + + .align 5 +L(u0): lbzu rALT, 1(rSRC) + stbu rWORD, 1(rRTN) + cmpdi rALT, 0 + beq L(u1) + lbzu rWORD, 1(rSRC) + stbu rALT, 1(rRTN) + cmpdi rWORD, 0 + beq L(u2) + lbzu rALT, 1(rSRC) + stbu rWORD, 1(rRTN) + cmpdi rALT, 0 + beq L(u1) + lbzu rWORD, 1(rSRC) + stbu rALT, 1(rRTN) + cmpdi rWORD, 0 + bne L(u0) +L(u2): stbu rWORD, 1(rRTN) + blr +L(u1): stbu rALT, 1(rRTN) + blr +END (FUNC_NAME) + +#ifndef USE_AS_STPCPY +libc_hidden_builtin_def (strcpy) +#endif diff --git a/sysdeps/powerpc/powerpc64/power7/strlen.S b/sysdeps/powerpc/powerpc64/power7/strlen.S index 343216952e..807ef1082e 100644 --- a/sysdeps/powerpc/powerpc64/power7/strlen.S +++ b/sysdeps/powerpc/powerpc64/power7/strlen.S @@ -30,7 +30,11 @@ ENTRY (strlen) with cmpb. */ li r5,-1 /* MASK = 0xffffffffffffffff. */ ld r12,0(r4) /* Load doubleword from memory. */ +#ifdef __LITTLE_ENDIAN__ + sld r5,r5,r6 +#else srd r5,r5,r6 /* MASK = MASK >> padding. */ +#endif orc r9,r12,r5 /* Mask bits that are not part of the string. */ cmpb r10,r9,r0 /* Check for null bytes in DWORD1. */ cmpdi cr7,r10,0 /* If r10 == 0, no null's have been found. */ @@ -48,9 +52,6 @@ ENTRY (strlen) cmpb r10,r12,r0 cmpdi cr7,r10,0 bne cr7,L(done) - b L(loop) /* We branch here (rather than falling through) - to skip the nops due to heavy alignment - of the loop below. */ /* Main loop to look for the end of the string. Since it's a small loop (< 8 instructions), align it to 32-bytes. */ @@ -87,9 +88,15 @@ L(loop): 0xff in the same position as the null byte in the original doubleword from the string. Use that to calculate the length. */ L(done): - cntlzd r0,r10 /* Count leading zeroes before the match. */ +#ifdef __LITTLE_ENDIAN__ + addi r9, r10, -1 /* Form a mask from trailing zeros. */ + andc r9, r9, r10 + popcntd r0, r9 /* Count the bits in the mask. */ +#else + cntlzd r0,r10 /* Count leading zeros before the match. */ +#endif subf r5,r3,r4 - srdi r0,r0,3 /* Convert leading zeroes to bytes. */ + srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */ add r3,r5,r0 /* Compute final length. */ blr END (strlen) diff --git a/sysdeps/powerpc/powerpc64/power7/strncmp.S b/sysdeps/powerpc/powerpc64/power7/strncmp.S index 77ecad5ab1..e618b010bf 100644 --- a/sysdeps/powerpc/powerpc64/power7/strncmp.S +++ b/sysdeps/powerpc/powerpc64/power7/strncmp.S @@ -27,7 +27,7 @@ EALIGN (strncmp,5,0) CALL_MCOUNT 3 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -40,6 +40,7 @@ EALIGN (strncmp,5,0) #define r7F7F r9 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 nop @@ -83,12 +84,57 @@ L(g1): add rTMP,rFEFE,rWORD1 we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + cmpb rBITDIF, rWORD1, rWORD2 /* 0xff on equal bytes. */ + addi rNEG, rBITDIF, 1 + orc rNEG, rNEG, rBITDIF /* 0's below LS differing byte. */ + sldi rNEG, rNEG, 8 /* 1's above LS differing byte. */ + andc rWORD1, rWORD1, rNEG /* mask off MS bytes. */ + andc rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + cmpb rBITDIF, rWORD1, rWORD2 /* 0xff on equal bytes. */ + addi rNEG, rBITDIF, 1 + orc rNEG, rNEG, rBITDIF /* 0's below LS differing byte. */ + sldi rNEG, rNEG, 8 /* 1's above LS differing byte. */ + andc rWORD1, rWORD1, rNEG /* mask off MS bytes. */ + andc rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else L(endstring): and rTMP,r7F7F,rWORD1 beq cr1,L(equal) add rTMP,rTMP,r7F7F xor. rBITDIF,rWORD1,rWORD2 - andc rNEG,rNEG,rTMP blt L(highbit) cntlzd rBITDIF,rBITDIF @@ -97,7 +143,7 @@ L(endstring): cmpd cr1,rNEG,rBITDIF sub rRTN,rWORD1,rWORD2 blt cr1,L(equal) - sradi rRTN,rRTN,63 + sradi rRTN,rRTN,63 /* must return an int. */ ori rRTN,rRTN,1 blr L(equal): @@ -105,7 +151,7 @@ L(equal): blr L(different): - ldu rWORD1,-8(rSTR1) + ld rWORD1,-8(rSTR1) xor. rBITDIF,rWORD1,rWORD2 sub rRTN,rWORD1,rWORD2 blt L(highbit) @@ -113,11 +159,10 @@ L(different): ori rRTN,rRTN,1 blr L(highbit): - srdi rWORD2,rWORD2,56 - srdi rWORD1,rWORD1,56 - sub rRTN,rWORD1,rWORD2 + sradi rRTN,rWORD2,63 + ori rRTN,rRTN,1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 diff --git a/sysdeps/powerpc/powerpc64/power7/strnlen.S b/sysdeps/powerpc/powerpc64/power7/strnlen.S index 37c7dbfe81..51591069df 100644 --- a/sysdeps/powerpc/powerpc64/power7/strnlen.S +++ b/sysdeps/powerpc/powerpc64/power7/strnlen.S @@ -24,33 +24,29 @@ ENTRY (__strnlen) CALL_MCOUNT 2 dcbt 0,r3 - clrrdi r8,r3,3 + clrrdi r8,r3,3 add r7,r3,r4 /* Calculate the last acceptable address. */ cmpldi r4,32 li r0,0 /* Doubleword with null chars. */ + addi r7,r7,-1 + /* If we have less than 33 bytes to search, skip to a faster code. */ ble L(small_range) - cmpld cr7,r3,r7 /* Is the address equal or less than r3? If - it's equal or less, it means size is either 0 - or a negative number. */ - ble cr7,L(proceed) - - li r7,-1 /* Make r11 the biggest if r4 <= 0. */ -L(proceed): rlwinm r6,r3,3,26,28 /* Calculate padding. */ ld r12,0(r8) /* Load doubleword from memory. */ cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */ +#ifdef __LITTLE_ENDIAN__ + srd r10,r10,r6 + sld r10,r10,r6 +#else sld r10,r10,r6 srd r10,r10,r6 +#endif cmpldi cr7,r10,0 /* If r10 == 0, no null's have been found. */ bne cr7,L(done) - /* Are we done already? */ - addi r9,r8,8 - cmpld cr6,r9,r7 - bge cr6,L(end_max) - + clrrdi r7,r7,3 /* Address of last doubleword. */ mtcrf 0x01,r8 /* Are we now aligned to a quadword boundary? If so, skip to the main loop. Otherwise, go through the alignment code. */ @@ -63,17 +59,18 @@ L(proceed): cmpldi cr7,r10,0 bne cr7,L(done) - /* Are we done already? */ - addi r9,r8,8 - cmpld cr6,r9,r7 - bge cr6,L(end_max) - L(loop_setup): - sub r5,r7,r9 + /* The last dword we want to read in the loop below is the one + containing the last byte of the string, ie. the dword at + (s + size - 1) & ~7, or r7. The first dword read is at + r8 + 8, we read 2 * cnt dwords, so the last dword read will + be at r8 + 8 + 16 * cnt - 8. Solving for cnt gives + cnt = (r7 - r8) / 16 */ + sub r5,r7,r8 srdi r6,r5,4 /* Number of loop iterations. */ mtctr r6 /* Setup the counter. */ - b L(loop) - /* Main loop to look for the null byte backwards in the string. Since + + /* Main loop to look for the null byte in the string. Since it's a small loop (< 8 instructions), align it to 32-bytes. */ .p2align 5 L(loop): @@ -89,15 +86,18 @@ L(loop): cmpldi cr7,r5,0 bne cr7,L(found) bdnz L(loop) - /* We're here because the counter reached 0, and that means we - didn't have any matches for null in the whole range. Just return - the original size. */ - addi r9,r8,8 - cmpld cr6,r9,r7 - blt cr6,L(loop_small) + + /* We may have one more dword to read. */ + cmpld cr6,r8,r7 + beq cr6,L(end_max) + + ldu r12,8(r8) + cmpb r10,r12,r0 + cmpldi cr6,r10,0 + bne cr6,L(done) L(end_max): - sub r3,r7,r3 + mr r3,r4 blr /* OK, one (or both) of the doublewords contains a null byte. Check @@ -119,52 +119,59 @@ L(found): /* r10 has the output of the cmpb instruction, that is, it contains 0xff in the same position as the null byte in the original doubleword from the string. Use that to calculate the length. - We need to make sure the null char is *before* the start of the - range (since we're going backwards). */ + We need to make sure the null char is *before* the end of the + range. */ L(done): - cntlzd r0,r10 /* Count leading zeroes before the match. */ - srdi r0,r0,3 /* Convert leading zeroes to bytes. */ - add r9,r8,r0 - sub r6,r9,r3 /* Length until the match. */ - cmpld r9,r7 - bgt L(end_max) - mr r3,r6 - blr - - .align 4 -L(zero): - li r3,0 +#ifdef __LITTLE_ENDIAN__ + addi r0,r10,-1 + andc r0,r0,r10 + popcntd r0,r0 +#else + cntlzd r0,r10 /* Count leading zeros before the match. */ +#endif + sub r3,r8,r3 + srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */ + add r3,r3,r0 /* Length until the match. */ + cmpld r3,r4 + blelr + mr r3,r4 blr /* Deals with size <= 32. */ .align 4 L(small_range): cmpldi r4,0 - beq L(zero) + beq L(end_max) + + clrrdi r7,r7,3 /* Address of last doubleword. */ rlwinm r6,r3,3,26,28 /* Calculate padding. */ - ld r12,0(r8) /* Load word from memory. */ + ld r12,0(r8) /* Load doubleword from memory. */ cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */ +#ifdef __LITTLE_ENDIAN__ + srd r10,r10,r6 + sld r10,r10,r6 +#else sld r10,r10,r6 srd r10,r10,r6 +#endif cmpldi cr7,r10,0 bne cr7,L(done) - addi r9,r8,8 - cmpld r9,r7 - bge L(end_max) - b L(loop_small) + cmpld r8,r7 + beq L(end_max) .p2align 5 L(loop_small): ldu r12,8(r8) cmpb r10,r12,r0 - addi r9,r8,8 cmpldi cr6,r10,0 bne cr6,L(done) - cmpld r9,r7 - bge L(end_max) - b L(loop_small) + cmpld r8,r7 + bne L(loop_small) + mr r3,r4 + blr + END (__strnlen) weak_alias (__strnlen, strnlen) libc_hidden_builtin_def (strnlen) diff --git a/sysdeps/powerpc/powerpc64/power7/sub_n.S b/sysdeps/powerpc/powerpc64/power7/sub_n.S new file mode 100644 index 0000000000..6afb81cf50 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/sub_n.S @@ -0,0 +1,23 @@ +/* PowerPC64 mpn_lshift -- mpn_add_n/mpn_sub_n -- mpn addition and + subtraction. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> + +#define USE_AS_SUB +#include "add_n.S" diff --git a/sysdeps/powerpc/powerpc64/power8/fpu/Implies b/sysdeps/powerpc/powerpc64/power8/fpu/Implies new file mode 100644 index 0000000000..7fd86fdf87 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power8/fpu/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power7/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power8/fpu/multiarch/Implies b/sysdeps/powerpc/powerpc64/power8/fpu/multiarch/Implies new file mode 100644 index 0000000000..7fd86fdf87 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power8/fpu/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power7/fpu/multiarch diff --git a/sysdeps/powerpc/powerpc64/power8/multiarch/Implies b/sysdeps/powerpc/powerpc64/power8/multiarch/Implies new file mode 100644 index 0000000000..1fc7b7cd39 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power8/multiarch/Implies @@ -0,0 +1 @@ +powerpc/powerpc64/power7/multiarch diff --git a/sysdeps/powerpc/powerpc64/ppc-mcount.S b/sysdeps/powerpc/powerpc64/ppc-mcount.S index 3d21a70669..9824a55f5f 100644 --- a/sysdeps/powerpc/powerpc64/ppc-mcount.S +++ b/sysdeps/powerpc/powerpc64/ppc-mcount.S @@ -24,16 +24,16 @@ ENTRY(_mcount) mflr r4 ld r11, 0(r1) - stdu r1,-112(r1) - cfi_adjust_cfa_offset (112) - std r4, 128(r1) - cfi_offset (lr, 16) - ld r3, 16(r11) + stdu r1,-FRAME_MIN_SIZE(r1) + cfi_adjust_cfa_offset (FRAME_MIN_SIZE) + std r4, FRAME_MIN_SIZE+FRAME_LR_SAVE(r1) + cfi_offset (lr, FRAME_LR_SAVE) + ld r3, FRAME_LR_SAVE(r11) bl JUMPTARGET(__mcount_internal) nop - ld r0, 128(r1) + ld r0, FRAME_MIN_SIZE+FRAME_LR_SAVE(r1) mtlr r0 - addi r1,r1,112 + addi r1,r1,FRAME_MIN_SIZE blr END(_mcount) diff --git a/sysdeps/powerpc/powerpc64/setjmp-common.S b/sysdeps/powerpc/powerpc64/setjmp-common.S index 58ec610620..6ab44d63ed 100644 --- a/sysdeps/powerpc/powerpc64/setjmp-common.S +++ b/sysdeps/powerpc/powerpc64/setjmp-common.S @@ -17,6 +17,7 @@ <http://www.gnu.org/licenses/>. */ #include <sysdep.h> +#include <stap-probe.h> #define _ASM #ifdef __NO_VMX__ #include <novmxsetjmp.h> @@ -42,11 +43,11 @@ #endif .machine "altivec" -ENTRY (setjmp) +ENTRY (setjmp_symbol) CALL_MCOUNT 1 li r4,1 /* Set second argument to 1. */ - b JUMPTARGET (GLUE(__sigsetjmp,_ent)) -END (setjmp) + b JUMPTARGET (GLUE(__sigsetjmp_symbol,_ent)) +END (setjmp_symbol) #if defined SHARED && !defined IS_IN_rtld && !defined __NO_VMX__ /* When called from within libc we need a special version of _setjmp @@ -54,22 +55,23 @@ END (setjmp) bugz #269. __GI__setjmp is used in csu/libc-start.c when HAVE_CLEANUP_JMP_BUF is defined. */ ENTRY (__GI__setjmp) - std r2,40(r1) /* Save the callers TOC in the save area. */ - cfi_endproc -END_2 (__GI__setjmp) -/* Fall thru. */ + std r2,FRAME_TOC_SAVE(r1) /* Save the callers TOC in the save area. */ + CALL_MCOUNT 1 + li r4,0 /* Set second argument to 0. */ + b JUMPTARGET (GLUE(__sigsetjmp_symbol,_ent)) +END (__GI__setjmp) #endif -ENTRY (_setjmp) +ENTRY (_setjmp_symbol) CALL_MCOUNT 1 li r4,0 /* Set second argument to 0. */ - b JUMPTARGET (GLUE(__sigsetjmp,_ent)) -END (_setjmp) -libc_hidden_def (_setjmp) + b JUMPTARGET (GLUE(__sigsetjmp_symbol,_ent)) +END (_setjmp_symbol) +libc_hidden_def (_setjmp_symbol) -ENTRY (__sigsetjmp) +ENTRY (__sigsetjmp_symbol) CALL_MCOUNT 2 -JUMPTARGET(GLUE(__sigsetjmp,_ent)): +JUMPTARGET(GLUE(__sigsetjmp_symbol,_ent)): #ifdef PTR_MANGLE mr r5, r1 PTR_MANGLE (r5, r6) @@ -79,11 +81,14 @@ JUMPTARGET(GLUE(__sigsetjmp,_ent)): #endif mflr r0 #if defined SHARED && !defined IS_IN_rtld - ld r5,40(r1) /* Retrieve the callers TOC. */ + ld r5,FRAME_TOC_SAVE(r1) /* Retrieve the callers TOC. */ std r5,(JB_GPR2*8)(3) #else std r2,(JB_GPR2*8)(3) #endif + /* setjmp probe expects longjmp first argument (8@3), second argument + (-4@4), and target address (8@0), respectively. */ + LIBC_PROBE (setjmp, 3, 8@3, -4@4, 8@0) std r14,((JB_GPRS+0)*8)(3) stfd fp14,((JB_FPRS+0)*8)(3) #ifdef PTR_MANGLE @@ -95,7 +100,7 @@ JUMPTARGET(GLUE(__sigsetjmp,_ent)): mfcr r0 std r16,((JB_GPRS+2)*8)(3) stfd fp16,((JB_FPRS+2)*8)(3) - std r0,(JB_CR*8)(3) + stw r0,((JB_CR*8)+4)(3) /* 32-bit CR. */ std r17,((JB_GPRS+3)*8)(3) stfd fp17,((JB_FPRS+3)*8)(3) std r18,((JB_GPRS+4)*8)(3) @@ -139,50 +144,46 @@ JUMPTARGET(GLUE(__sigsetjmp,_ent)): la r5,((JB_VRS)*8)(3) andi. r6,r5,0xf mfspr r0,VRSAVE - stw r0,((JB_VRSAVE)*8)(3) + stw r0,((JB_VRSAVE)*8)(3) /* 32-bit VRSAVE. */ addi r6,r5,16 beq+ L(aligned_save_vmx) - lvsr v0,0,r5 - vspltisb v1,-1 /* set v1 to all 1's */ - vspltisb v2,0 /* set v2 to all 0's */ - vperm v3,v2,v1,v0 /* v3 contains shift mask with num all 1 bytes - on left = misalignment */ + lvsr v0,0,r5 + lvsl v1,0,r5 + addi r6,r5,-16 - /* Special case for v20 we need to preserve what is in save area - below v20 before obliterating it */ - lvx v5,0,r5 - vperm v20,v20,v20,v0 - vsel v5,v5,v20,v3 - vsel v20,v20,v2,v3 - stvx v5,0,r5 +# define save_misaligned_vmx(savevr,prevvr,shiftvr,tmpvr,savegpr,addgpr) \ + addi addgpr,addgpr,32; \ + vperm tmpvr,prevvr,savevr,shiftvr; \ + stvx tmpvr,0,savegpr -# define save_2vmx_partial(savevr,prev_savevr,hivr,shiftvr,maskvr,savegpr,addgpr) \ - addi addgpr,addgpr,32; \ - vperm savevr,savevr,savevr,shiftvr; \ - vsel hivr,prev_savevr,savevr,maskvr; \ - stvx hivr,0,savegpr; + /* + * We have to be careful not to corrupt the data below v20 and + * above v31. To keep things simple we just rotate both ends in + * the opposite direction to our main permute so we can use + * the common macro. + */ - save_2vmx_partial(v21,v20,v5,v0,v3,r6,r5) - save_2vmx_partial(v22,v21,v5,v0,v3,r5,r6) - save_2vmx_partial(v23,v22,v5,v0,v3,r6,r5) - save_2vmx_partial(v24,v23,v5,v0,v3,r5,r6) - save_2vmx_partial(v25,v24,v5,v0,v3,r6,r5) - save_2vmx_partial(v26,v25,v5,v0,v3,r5,r6) - save_2vmx_partial(v27,v26,v5,v0,v3,r6,r5) - save_2vmx_partial(v28,v27,v5,v0,v3,r5,r6) - save_2vmx_partial(v29,v28,v5,v0,v3,r6,r5) - save_2vmx_partial(v30,v29,v5,v0,v3,r5,r6) + /* load and rotate data below v20 */ + lvx v2,0,r5 + vperm v2,v2,v2,v1 + save_misaligned_vmx(v20,v2,v0,v3,r5,r6) + save_misaligned_vmx(v21,v20,v0,v3,r6,r5) + save_misaligned_vmx(v22,v21,v0,v3,r5,r6) + save_misaligned_vmx(v23,v22,v0,v3,r6,r5) + save_misaligned_vmx(v24,v23,v0,v3,r5,r6) + save_misaligned_vmx(v25,v24,v0,v3,r6,r5) + save_misaligned_vmx(v26,v25,v0,v3,r5,r6) + save_misaligned_vmx(v27,v26,v0,v3,r6,r5) + save_misaligned_vmx(v28,v27,v0,v3,r5,r6) + save_misaligned_vmx(v29,v28,v0,v3,r6,r5) + save_misaligned_vmx(v30,v29,v0,v3,r5,r6) + save_misaligned_vmx(v31,v30,v0,v3,r6,r5) + /* load and rotate data above v31 */ + lvx v2,0,r6 + vperm v2,v2,v2,v1 + save_misaligned_vmx(v2,v31,v0,v3,r5,r6) - /* Special case for r31 we need to preserve what is in save area - above v31 before obliterating it */ - addi r5,r5,32 - vperm v31,v31,v31,v0 - lvx v4,0,r5 - vsel v5,v30,v31,v3 - stvx v5,0,r6 - vsel v4,v31,v4,v3 - stvx v4,0,r5 b L(no_vmx) L(aligned_save_vmx): @@ -216,18 +217,18 @@ L(no_vmx): li r3,0 blr #elif defined SHARED - b JUMPTARGET (__sigjmp_save) + b JUMPTARGET (__sigjmp_save_symbol) #else mflr r0 - std r0,16(r1) - stdu r1,-112(r1) - cfi_adjust_cfa_offset(112) - cfi_offset(lr,16) - bl JUMPTARGET (__sigjmp_save) + std r0,FRAME_LR_SAVE(r1) + stdu r1,-FRAME_MIN_SIZE(r1) + cfi_adjust_cfa_offset(FRAME_MIN_SIZE) + cfi_offset(lr,FRAME_LR_SAVE) + bl JUMPTARGET (__sigjmp_save_symbol) nop - ld r0,112+16(r1) - addi r1,r1,112 + ld r0,FRAME_MIN_SIZE+FRAME_LR_SAVE(r1) + addi r1,r1,FRAME_MIN_SIZE mtlr r0 blr #endif -END (__sigsetjmp) +END (__sigsetjmp_symbol) diff --git a/sysdeps/powerpc/powerpc64/setjmp.S b/sysdeps/powerpc/powerpc64/setjmp.S index 667b9d12dd..5c6baf52df 100644 --- a/sysdeps/powerpc/powerpc64/setjmp.S +++ b/sysdeps/powerpc/powerpc64/setjmp.S @@ -22,35 +22,39 @@ #if defined NOT_IN_libc /* Build a non-versioned object for rtld-*. */ +# define setjmp_symbol setjmp +# define _setjmp_symbol _setjmp +# define __sigsetjmp_symbol __sigsetjmp +# define __sigjmp_save_symbol __sigjmp_save # include "setjmp-common.S" #else /* !NOT_IN_libc */ /* Build a versioned object for libc. */ -default_symbol_version (__vmxsetjmp, setjmp, GLIBC_2.3.4) -default_symbol_version (__vmx_setjmp,_setjmp,GLIBC_2.3.4) -default_symbol_version (__vmx__sigsetjmp,__sigsetjmp,GLIBC_2.3.4) -# define setjmp __vmxsetjmp -# define _setjmp __vmx_setjmp -# define __sigsetjmp __vmx__sigsetjmp -# define __sigjmp_save __vmx__sigjmp_save +versioned_symbol (libc, __vmxsetjmp, setjmp, GLIBC_2_3_4) +versioned_symbol (libc, __vmx_setjmp, _setjmp, GLIBC_2_3_4) +versioned_symbol (libc, __vmx__sigsetjmp, __sigsetjmp, GLIBC_2_3_4) +# define setjmp_symbol __vmxsetjmp +# define _setjmp_symbol __vmx_setjmp +# define __sigsetjmp_symbol __vmx__sigsetjmp +# define __sigjmp_save_symbol __vmx__sigjmp_save # include "setjmp-common.S" strong_alias (__vmxsetjmp, __vmx__setjmp) strong_alias (__vmx__sigsetjmp, __setjmp) # if defined SHARED && SHLIB_COMPAT (libc, GLIBC_2_3, GLIBC_2_3_4) -# undef setjmp -# undef _setjmp -# undef __sigsetjmp -# undef __sigjmp_save +# undef setjmp_symbol +# undef _setjmp_symbol +# undef __sigsetjmp_symbol +# undef __sigjmp_save_symbol # undef JB_SIZE # define __NO_VMX__ -symbol_version (__novmxsetjmp, setjmp, GLIBC_2.3) -symbol_version (__novmx_setjmp,_setjmp,GLIBC_2.3); -symbol_version (__novmx__sigsetjmp,__sigsetjmp,GLIBC_2.3) -# define setjmp __novmxsetjmp -# define _setjmp __novmx_setjmp -# define __sigsetjmp __novmx__sigsetjmp -# define __sigjmp_save __novmx__sigjmp_save +compat_symbol (libc, __novmxsetjmp, setjmp, GLIBC_2_3) +compat_symbol (libc, __novmx_setjmp,_setjmp, GLIBC_2_3); +compat_symbol (libc, __novmx__sigsetjmp,__sigsetjmp, GLIBC_2_3) +# define setjmp_symbol __novmxsetjmp +# define _setjmp_symbol __novmx_setjmp +# define __sigsetjmp_symbol __novmx__sigsetjmp +# define __sigjmp_save_symbol __novmx__sigjmp_save # include "setjmp-common.S" strong_alias (__novmxsetjmp, __novmx__setjmp) # endif diff --git a/sysdeps/powerpc/powerpc64/stackguard-macros.h b/sysdeps/powerpc/powerpc64/stackguard-macros.h index 9da879c611..e80a683e64 100644 --- a/sysdeps/powerpc/powerpc64/stackguard-macros.h +++ b/sysdeps/powerpc/powerpc64/stackguard-macros.h @@ -2,3 +2,13 @@ #define STACK_CHK_GUARD \ ({ uintptr_t x; asm ("ld %0,-28688(13)" : "=r" (x)); x; }) + +#define POINTER_CHK_GUARD \ + ({ \ + uintptr_t x; \ + asm ("ld %0,%1(13)" \ + : "=r" (x) \ + : "i" (offsetof (tcbhead_t, pointer_guard) - TLS_TCB_OFFSET - sizeof (tcbhead_t)) \ + ); \ + x; \ + }) diff --git a/sysdeps/powerpc/powerpc64/stpcpy.S b/sysdeps/powerpc/powerpc64/stpcpy.S index 070cd4662f..09aa3be6b5 100644 --- a/sysdeps/powerpc/powerpc64/stpcpy.S +++ b/sysdeps/powerpc/powerpc64/stpcpy.S @@ -16,87 +16,8 @@ License along with the GNU C Library; if not, see <http://www.gnu.org/licenses/>. */ -#include <sysdep.h> - -/* See strlen.s for comments on how the end-of-string testing works. */ - -/* char * [r3] stpcpy (char *dest [r3], const char *src [r4]) */ - -EALIGN (__stpcpy, 4, 0) - CALL_MCOUNT 2 - -#define rTMP r0 -#define rRTN r3 -#define rDEST r3 /* pointer to previous word in dest */ -#define rSRC r4 /* pointer to previous word in src */ -#define rWORD r6 /* current word from src */ -#define rFEFE r7 /* 0xfefefeff */ -#define r7F7F r8 /* 0x7f7f7f7f */ -#define rNEG r9 /* ~(word in src | 0x7f7f7f7f) */ -#define rALT r10 /* alternate word from src */ - - or rTMP, rSRC, rDEST - clrldi. rTMP, rTMP, 62 - addi rDEST, rDEST, -4 - bne L(unaligned) - - lis rFEFE, -0x101 - lis r7F7F, 0x7f7f - lwz rWORD, 0(rSRC) - addi rFEFE, rFEFE, -0x101 - addi r7F7F, r7F7F, 0x7f7f - b L(g2) - -L(g0): lwzu rALT, 4(rSRC) - stwu rWORD, 4(rDEST) - add rTMP, rFEFE, rALT - nor rNEG, r7F7F, rALT - and. rTMP, rTMP, rNEG - bne- L(g1) - lwzu rWORD, 4(rSRC) - stwu rALT, 4(rDEST) -L(g2): add rTMP, rFEFE, rWORD - nor rNEG, r7F7F, rWORD - and. rTMP, rTMP, rNEG - beq+ L(g0) - - mr rALT, rWORD -/* We've hit the end of the string. Do the rest byte-by-byte. */ -L(g1): rlwinm. rTMP, rALT, 8, 24, 31 - stbu rTMP, 4(rDEST) - beqlr- - rlwinm. rTMP, rALT, 16, 24, 31 - stbu rTMP, 1(rDEST) - beqlr- - rlwinm. rTMP, rALT, 24, 24, 31 - stbu rTMP, 1(rDEST) - beqlr- - stbu rALT, 1(rDEST) - blr - -/* Oh well. In this case, we just do a byte-by-byte copy. */ - .align 4 - nop -L(unaligned): - lbz rWORD, 0(rSRC) - addi rDEST, rDEST, 3 - cmpwi rWORD, 0 - beq- L(u2) - -L(u0): lbzu rALT, 1(rSRC) - stbu rWORD, 1(rDEST) - cmpwi rALT, 0 - beq- L(u1) - nop /* Let 601 load start of loop. */ - lbzu rWORD, 1(rSRC) - stbu rALT, 1(rDEST) - cmpwi rWORD, 0 - bne+ L(u0) -L(u2): stbu rWORD, 1(rDEST) - blr -L(u1): stbu rALT, 1(rDEST) - blr -END (__stpcpy) +#define USE_AS_STPCPY +#include <sysdeps/powerpc/powerpc64/strcpy.S> weak_alias (__stpcpy, stpcpy) libc_hidden_def (__stpcpy) diff --git a/sysdeps/powerpc/powerpc64/strchr.S b/sysdeps/powerpc/powerpc64/strchr.S index d2d8cd361a..da707ae587 100644 --- a/sysdeps/powerpc/powerpc64/strchr.S +++ b/sysdeps/powerpc/powerpc64/strchr.S @@ -37,11 +37,13 @@ ENTRY (strchr) #define rIGN r10 /* number of bits we should ignore in the first word */ #define rMASK r11 /* mask with the bits to ignore set to 0 */ #define rTMP3 r12 +#define rTMP4 rIGN +#define rTMP5 rMASK dcbt 0,rRTN - rlwimi rCHR, rCHR, 8, 16, 23 + insrdi rCHR, rCHR, 8, 48 li rMASK, -1 - rlwimi rCHR, rCHR, 16, 0, 15 + insrdi rCHR, rCHR, 16, 32 rlwinm rIGN, rRTN, 3, 26, 28 insrdi rCHR, rCHR, 32, 0 lis rFEFE, -0x101 @@ -54,64 +56,93 @@ ENTRY (strchr) add rFEFE, rFEFE, rTMP1 /* Test the first (partial?) word. */ ld rWORD, 0(rSTR) +#ifdef __LITTLE_ENDIAN__ + sld rMASK, rMASK, rIGN +#else srd rMASK, rMASK, rIGN +#endif orc rWORD, rWORD, rMASK add rTMP1, rFEFE, rWORD nor rTMP2, r7F7F, rWORD - and. rTMP1, rTMP1, rTMP2 + and. rTMP4, rTMP1, rTMP2 xor rTMP3, rCHR, rWORD orc rTMP3, rTMP3, rMASK b L(loopentry) /* The loop. */ -L(loop):ldu rWORD, 8(rSTR) - and. rTMP1, rTMP1, rTMP2 +L(loop): + ldu rWORD, 8(rSTR) + and. rTMP5, rTMP1, rTMP2 /* Test for 0. */ - add rTMP1, rFEFE, rWORD - nor rTMP2, r7F7F, rWORD + add rTMP1, rFEFE, rWORD /* x - 0x01010101. */ + nor rTMP2, r7F7F, rWORD /* ~(x | 0x7f7f7f7f) == ~x & 0x80808080. */ bne L(foundit) - and. rTMP1, rTMP1, rTMP2 + and. rTMP4, rTMP1, rTMP2 /* (x - 0x01010101) & ~x & 0x80808080. */ /* Start test for the bytes we're looking for. */ xor rTMP3, rCHR, rWORD L(loopentry): add rTMP1, rFEFE, rTMP3 nor rTMP2, r7F7F, rTMP3 beq L(loop) + /* There is a zero byte in the word, but may also be a matching byte (either before or after the zero byte). In fact, we may be looking for a - zero byte, in which case we return a match. We guess that this hasn't - happened, though. */ -L(missed): - and. rTMP1, rTMP1, rTMP2 + zero byte, in which case we return a match. */ + and. rTMP5, rTMP1, rTMP2 li rRTN, 0 beqlr -/* It did happen. Decide which one was first... - I'm not sure if this is actually faster than a sequence of - rotates, compares, and branches (we use it anyway because it's shorter). */ +/* At this point: + rTMP5 bytes are 0x80 for each match of c, 0 otherwise. + rTMP4 bytes are 0x80 for each match of 0, 0 otherwise. + But there may be false matches in the next most significant byte from + a true match due to carries. This means we need to recalculate the + matches using a longer method for big-endian. */ +#ifdef __LITTLE_ENDIAN__ + addi rTMP1, rTMP5, -1 + andc rTMP1, rTMP1, rTMP5 + cntlzd rCLZB, rTMP1 + addi rTMP2, rTMP4, -1 + andc rTMP2, rTMP2, rTMP4 + cmpld rTMP1, rTMP2 + bgtlr + subfic rCLZB, rCLZB, 64-7 +#else +/* I think we could reduce this by two instructions by keeping the "nor" + results from the loop for reuse here. See strlen.S tail. Similarly + one instruction could be pruned from L(foundit). */ and rFEFE, r7F7F, rWORD - or rMASK, r7F7F, rWORD + or rTMP5, r7F7F, rWORD and rTMP1, r7F7F, rTMP3 - or rIGN, r7F7F, rTMP3 + or rTMP4, r7F7F, rTMP3 add rFEFE, rFEFE, r7F7F add rTMP1, rTMP1, r7F7F - nor rWORD, rMASK, rFEFE - nor rTMP2, rIGN, rTMP1 + nor rWORD, rTMP5, rFEFE + nor rTMP2, rTMP4, rTMP1 + cntlzd rCLZB, rTMP2 cmpld rWORD, rTMP2 bgtlr - cntlzd rCLZB, rTMP2 +#endif srdi rCLZB, rCLZB, 3 add rRTN, rSTR, rCLZB blr L(foundit): +#ifdef __LITTLE_ENDIAN__ + addi rTMP1, rTMP5, -1 + andc rTMP1, rTMP1, rTMP5 + cntlzd rCLZB, rTMP1 + subfic rCLZB, rCLZB, 64-7-64 + sradi rCLZB, rCLZB, 3 +#else and rTMP1, r7F7F, rTMP3 - or rIGN, r7F7F, rTMP3 + or rTMP4, r7F7F, rTMP3 add rTMP1, rTMP1, r7F7F - nor rTMP2, rIGN, rTMP1 + nor rTMP2, rTMP4, rTMP1 cntlzd rCLZB, rTMP2 subi rSTR, rSTR, 8 srdi rCLZB, rCLZB, 3 +#endif add rRTN, rSTR, rCLZB blr END (strchr) diff --git a/sysdeps/powerpc/powerpc64/strcmp.S b/sysdeps/powerpc/powerpc64/strcmp.S index c9d6dac121..70854689d3 100644 --- a/sysdeps/powerpc/powerpc64/strcmp.S +++ b/sysdeps/powerpc/powerpc64/strcmp.S @@ -25,7 +25,7 @@ EALIGN (strcmp, 4, 0) CALL_MCOUNT 2 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -35,6 +35,7 @@ EALIGN (strcmp, 4, 0) #define r7F7F r8 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r9 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r10 /* bits that differ in s1 & s2 words */ +#define rTMP r11 dcbt 0,rSTR1 or rTMP, rSTR2, rSTR1 @@ -58,19 +59,66 @@ L(g0): ldu rWORD1, 8(rSTR1) ldu rWORD2, 8(rSTR2) L(g1): add rTMP, rFEFE, rWORD1 nor rNEG, r7F7F, rWORD1 - and. rTMP, rTMP, rNEG cmpd cr1, rWORD1, rWORD2 beq+ L(g0) -L(endstring): + /* OK. We've hit the end of the string. We need to be careful that we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else +L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzd rBITDIF, rBITDIF @@ -79,7 +127,7 @@ L(endstring): cmpd cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 blt- cr1, L(equal) - sradi rRTN, rRTN, 63 + sradi rRTN, rRTN, 63 /* must return an int. */ ori rRTN, rRTN, 1 blr L(equal): @@ -95,11 +143,10 @@ L(different): ori rRTN, rRTN, 1 blr L(highbit): - srdi rWORD2, rWORD2, 56 - srdi rWORD1, rWORD1, 56 - sub rRTN, rWORD1, rWORD2 + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 diff --git a/sysdeps/powerpc/powerpc64/strcpy.S b/sysdeps/powerpc/powerpc64/strcpy.S index 4c6fd3f9d7..793325d7be 100644 --- a/sysdeps/powerpc/powerpc64/strcpy.S +++ b/sysdeps/powerpc/powerpc64/strcpy.S @@ -22,25 +22,38 @@ /* char * [r3] strcpy (char *dest [r3], const char *src [r4]) */ -EALIGN (strcpy, 4, 0) +#ifdef USE_AS_STPCPY +# define FUNC_NAME __stpcpy +#else +# define FUNC_NAME strcpy +#endif + +EALIGN (FUNC_NAME, 4, 0) CALL_MCOUNT 2 #define rTMP r0 -#define rRTN r3 /* incoming DEST arg preserved as result */ -#define rSRC r4 /* pointer to previous word in src */ -#define rDEST r5 /* pointer to previous word in dest */ +#ifdef USE_AS_STPCPY +#define rRTN r3 /* pointer to previous word/doubleword in dest */ +#else +#define rRTN r12 /* pointer to previous word/doubleword in dest */ +#endif +#define rSRC r4 /* pointer to previous word/doubleword in src */ #define rWORD r6 /* current word from src */ -#define rFEFE r7 /* constant 0xfefefefefefefeff (-0x0101010101010101) */ -#define r7F7F r8 /* constant 0x7f7f7f7f7f7f7f7f */ -#define rNEG r9 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ +#define rFEFE r7 /* constant 0xfefefeff | 0xfefefefefefefeff */ +#define r7F7F r8 /* constant 0x7f7f7f7f | 0x7f7f7f7f7f7f7f7f */ +#define rNEG r9 /* ~(word in s1 | r7F7F) */ #define rALT r10 /* alternate word from src */ - dcbt 0,rSRC +#ifndef USE_AS_STPCPY +/* Save the dst pointer to use as return value. */ + mr rRTN, r3 +#endif or rTMP, rSRC, rRTN clrldi. rTMP, rTMP, 61 - addi rDEST, rRTN, -8 - dcbtst 0,rRTN - bne L(unaligned) + bne L(check_word_alignment) + +/* For doubleword aligned memory, operate using doubleword load and stores. */ + addi rRTN, rRTN, -8 lis rFEFE, -0x101 lis r7F7F, 0x7f7f @@ -53,13 +66,13 @@ EALIGN (strcpy, 4, 0) b L(g2) L(g0): ldu rALT, 8(rSRC) - stdu rWORD, 8(rDEST) + stdu rWORD, 8(rRTN) add rTMP, rFEFE, rALT nor rNEG, r7F7F, rALT and. rTMP, rTMP, rNEG bne- L(g1) ldu rWORD, 8(rSRC) - stdu rALT, 8(rDEST) + stdu rALT, 8(rRTN) L(g2): add rTMP, rFEFE, rWORD nor rNEG, r7F7F, rWORD and. rTMP, rTMP, rNEG @@ -68,28 +81,110 @@ L(g2): add rTMP, rFEFE, rWORD mr rALT, rWORD /* We've hit the end of the string. Do the rest byte-by-byte. */ L(g1): +#ifdef __LITTLE_ENDIAN__ + extrdi. rTMP, rALT, 8, 56 + stbu rALT, 8(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 48 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 40 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 32 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 24 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 16 + stbu rTMP, 1(rRTN) + beqlr- + extrdi. rTMP, rALT, 8, 8 + stbu rTMP, 1(rRTN) + beqlr- + extrdi rTMP, rALT, 8, 0 + stbu rTMP, 1(rRTN) +#else extrdi. rTMP, rALT, 8, 0 - stb rTMP, 8(rDEST) + stbu rTMP, 8(rRTN) beqlr- extrdi. rTMP, rALT, 8, 8 - stb rTMP, 9(rDEST) + stbu rTMP, 1(rRTN) beqlr- extrdi. rTMP, rALT, 8, 16 - stb rTMP, 10(rDEST) + stbu rTMP, 1(rRTN) beqlr- extrdi. rTMP, rALT, 8, 24 - stb rTMP, 11(rDEST) + stbu rTMP, 1(rRTN) beqlr- extrdi. rTMP, rALT, 8, 32 - stb rTMP, 12(rDEST) - beqlr- + stbu rTMP, 1(rRTN) + beqlr extrdi. rTMP, rALT, 8, 40 - stb rTMP, 13(rDEST) + stbu rTMP, 1(rRTN) beqlr- extrdi. rTMP, rALT, 8, 48 - stb rTMP, 14(rDEST) + stbu rTMP, 1(rRTN) + beqlr- + stbu rALT, 1(rRTN) +#endif + blr + +L(check_word_alignment): + clrldi. rTMP, rTMP, 62 + bne L(unaligned) + +/* For word aligned memory, operate using word load and stores. */ + addi rRTN, rRTN, -4 + + lis rFEFE, -0x101 + lis r7F7F, 0x7f7f + lwz rWORD, 0(rSRC) + addi rFEFE, rFEFE, -0x101 + addi r7F7F, r7F7F, 0x7f7f + b L(g5) + +L(g3): lwzu rALT, 4(rSRC) + stwu rWORD, 4(rRTN) + add rTMP, rFEFE, rALT + nor rNEG, r7F7F, rALT + and. rTMP, rTMP, rNEG + bne- L(g4) + lwzu rWORD, 4(rSRC) + stwu rALT, 4(rRTN) +L(g5): add rTMP, rFEFE, rWORD + nor rNEG, r7F7F, rWORD + and. rTMP, rTMP, rNEG + beq+ L(g3) + + mr rALT, rWORD +/* We've hit the end of the string. Do the rest byte-by-byte. */ +L(g4): +#ifdef __LITTLE_ENDIAN__ + rlwinm. rTMP, rALT, 0, 24, 31 + stbu rALT, 4(rRTN) + beqlr- + rlwinm. rTMP, rALT, 24, 24, 31 + stbu rTMP, 1(rRTN) + beqlr- + rlwinm. rTMP, rALT, 16, 24, 31 + stbu rTMP, 1(rRTN) + beqlr- + rlwinm rTMP, rALT, 8, 24, 31 + stbu rTMP, 1(rRTN) +#else + rlwinm. rTMP, rALT, 8, 24, 31 + stbu rTMP, 4(rRTN) + beqlr- + rlwinm. rTMP, rALT, 16, 24, 31 + stbu rTMP, 1(rRTN) + beqlr- + rlwinm. rTMP, rALT, 24, 24, 31 + stbu rTMP, 1(rRTN) beqlr- - stb rALT, 15(rDEST) + stbu rALT, 1(rRTN) +#endif blr /* Oh well. In this case, we just do a byte-by-byte copy. */ @@ -97,23 +192,25 @@ L(g1): nop L(unaligned): lbz rWORD, 0(rSRC) - addi rDEST, rRTN, -1 + addi rRTN, rRTN, -1 cmpwi rWORD, 0 beq- L(u2) L(u0): lbzu rALT, 1(rSRC) - stbu rWORD, 1(rDEST) + stbu rWORD, 1(rRTN) cmpwi rALT, 0 beq- L(u1) nop /* Let 601 load start of loop. */ lbzu rWORD, 1(rSRC) - stbu rALT, 1(rDEST) + stbu rALT, 1(rRTN) cmpwi rWORD, 0 bne+ L(u0) -L(u2): stb rWORD, 1(rDEST) +L(u2): stbu rWORD, 1(rRTN) blr -L(u1): stb rALT, 1(rDEST) +L(u1): stbu rALT, 1(rRTN) blr +END (FUNC_NAME) -END (strcpy) +#ifndef USE_AS_STPCPY libc_hidden_builtin_def (strcpy) +#endif diff --git a/sysdeps/powerpc/powerpc64/strlen.S b/sysdeps/powerpc/powerpc64/strlen.S index 0f9b5eea9f..4ed1ba3ad1 100644 --- a/sysdeps/powerpc/powerpc64/strlen.S +++ b/sysdeps/powerpc/powerpc64/strlen.S @@ -29,7 +29,12 @@ 1 is subtracted you get a value in the range 0x00-0x7f, none of which have their high bit set. The expression here is (x + 0xfefefeff) & ~(x | 0x7f7f7f7f), which gives 0x00000000 when - there were no 0x00 bytes in the word. + there were no 0x00 bytes in the word. You get 0x80 in bytes that + match, but possibly false 0x80 matches in the next more significant + byte to a true match due to carries. For little-endian this is + of no consequence since the least significant match is the one + we're interested in, but big-endian needs method 2 to find which + byte matches. 2) Given a word 'x', we can test to see _which_ byte was zero by calculating ~(((x & 0x7f7f7f7f) + 0x7f7f7f7f) | x | 0x7f7f7f7f). @@ -62,7 +67,7 @@ Answer: 1) Added a Data Cache Block Touch early to prefetch the first 128 byte cache line. Adding dcbt instructions to the loop would not be - effective since most strings will be shorter than the cache line.*/ + effective since most strings will be shorter than the cache line. */ /* Some notes on register usage: Under the SVR4 ABI, we can use registers 0 and 3 through 12 (so long as we don't call any procedures) without @@ -78,7 +83,7 @@ ENTRY (strlen) CALL_MCOUNT 1 -#define rTMP1 r0 +#define rTMP4 r0 #define rRTN r3 /* incoming STR arg, outgoing result */ #define rSTR r4 /* current string position */ #define rPADN r5 /* number of padding bits we prepend to the @@ -88,9 +93,9 @@ ENTRY (strlen) #define rWORD1 r8 /* current string doubleword */ #define rWORD2 r9 /* next string doubleword */ #define rMASK r9 /* mask for first string doubleword */ -#define rTMP2 r10 -#define rTMP3 r11 -#define rTMP4 r12 +#define rTMP1 r10 +#define rTMP2 r11 +#define rTMP3 r12 dcbt 0,rRTN clrrdi rSTR, rRTN, 3 @@ -100,30 +105,36 @@ ENTRY (strlen) addi r7F7F, r7F7F, 0x7f7f li rMASK, -1 insrdi r7F7F, r7F7F, 32, 0 -/* That's the setup done, now do the first pair of doublewords. - We make an exception and use method (2) on the first two doublewords, - to reduce overhead. */ +/* We use method (2) on the first two doublewords, because rFEFE isn't + required which reduces setup overhead. Also gives a faster return + for small strings on big-endian due to needing to recalculate with + method (2) anyway. */ +#ifdef __LITTLE_ENDIAN__ + sld rMASK, rMASK, rPADN +#else srd rMASK, rMASK, rPADN +#endif and rTMP1, r7F7F, rWORD1 or rTMP2, r7F7F, rWORD1 lis rFEFE, -0x101 add rTMP1, rTMP1, r7F7F addi rFEFE, rFEFE, -0x101 - nor rTMP1, rTMP2, rTMP1 - and. rWORD1, rTMP1, rMASK + nor rTMP3, rTMP2, rTMP1 + and. rTMP3, rTMP3, rMASK mtcrf 0x01, rRTN bne L(done0) - sldi rTMP1, rFEFE, 32 - add rFEFE, rFEFE, rTMP1 + sldi rTMP1, rFEFE, 32 + add rFEFE, rFEFE, rTMP1 /* Are we now aligned to a doubleword boundary? */ bt 28, L(loop) /* Handle second doubleword of pair. */ +/* Perhaps use method (1) here for little-endian, saving one instruction? */ ldu rWORD1, 8(rSTR) and rTMP1, r7F7F, rWORD1 or rTMP2, r7F7F, rWORD1 add rTMP1, rTMP1, r7F7F - nor. rWORD1, rTMP2, rTMP1 + nor. rTMP3, rTMP2, rTMP1 bne L(done0) /* The loop. */ @@ -137,28 +148,52 @@ L(loop): add rTMP3, rFEFE, rWORD2 nor rTMP4, r7F7F, rWORD2 bne L(done1) - and. rTMP1, rTMP3, rTMP4 + and. rTMP3, rTMP3, rTMP4 beq L(loop) +#ifndef __LITTLE_ENDIAN__ and rTMP1, r7F7F, rWORD2 add rTMP1, rTMP1, r7F7F - andc rWORD1, rTMP4, rTMP1 + andc rTMP3, rTMP4, rTMP1 b L(done0) L(done1): and rTMP1, r7F7F, rWORD1 subi rSTR, rSTR, 8 add rTMP1, rTMP1, r7F7F - andc rWORD1, rTMP2, rTMP1 + andc rTMP3, rTMP2, rTMP1 /* When we get to here, rSTR points to the first doubleword in the string that - contains a zero byte, and the most significant set bit in rWORD1 is in that - byte. */ + contains a zero byte, and rTMP3 has 0x80 for bytes that are zero, and 0x00 + otherwise. */ L(done0): - cntlzd rTMP3, rWORD1 + cntlzd rTMP3, rTMP3 subf rTMP1, rRTN, rSTR srdi rTMP3, rTMP3, 3 add rRTN, rTMP1, rTMP3 blr +#else + +L(done0): + addi rTMP1, rTMP3, -1 /* Form a mask from trailing zeros. */ + andc rTMP1, rTMP1, rTMP3 + cntlzd rTMP1, rTMP1 /* Count bits not in the mask. */ + subf rTMP3, rRTN, rSTR + subfic rTMP1, rTMP1, 64-7 + srdi rTMP1, rTMP1, 3 + add rRTN, rTMP1, rTMP3 + blr + +L(done1): + addi rTMP3, rTMP1, -1 + andc rTMP3, rTMP3, rTMP1 + cntlzd rTMP3, rTMP3 + subf rTMP1, rRTN, rSTR + subfic rTMP3, rTMP3, 64-7-64 + sradi rTMP3, rTMP3, 3 + add rRTN, rTMP1, rTMP3 + blr +#endif + END (strlen) libc_hidden_builtin_def (strlen) diff --git a/sysdeps/powerpc/powerpc64/strncmp.S b/sysdeps/powerpc/powerpc64/strncmp.S index 779d9f7f6f..8f842c4b26 100644 --- a/sysdeps/powerpc/powerpc64/strncmp.S +++ b/sysdeps/powerpc/powerpc64/strncmp.S @@ -25,7 +25,7 @@ EALIGN (strncmp, 4, 0) CALL_MCOUNT 3 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -36,6 +36,7 @@ EALIGN (strncmp, 4, 0) #define r7F7F r9 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 or rTMP, rSTR2, rSTR1 @@ -77,12 +78,59 @@ L(g1): add rTMP, rFEFE, rWORD1 we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzd rBITDIF, rBITDIF @@ -91,7 +139,7 @@ L(endstring): cmpd cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 blt- cr1, L(equal) - sradi rRTN, rRTN, 63 + sradi rRTN, rRTN, 63 /* must return an int. */ ori rRTN, rRTN, 1 blr L(equal): @@ -99,7 +147,7 @@ L(equal): blr L(different): - ldu rWORD1, -8(rSTR1) + ld rWORD1, -8(rSTR1) xor. rBITDIF, rWORD1, rWORD2 sub rRTN, rWORD1, rWORD2 blt- L(highbit) @@ -107,11 +155,10 @@ L(different): ori rRTN, rRTN, 1 blr L(highbit): - srdi rWORD2, rWORD2, 56 - srdi rWORD1, rWORD1, 56 - sub rRTN, rWORD1, rWORD2 + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 diff --git a/sysdeps/powerpc/powerpc64/submul_1.S b/sysdeps/powerpc/powerpc64/submul_1.S new file mode 100644 index 0000000000..145b1d403c --- /dev/null +++ b/sysdeps/powerpc/powerpc64/submul_1.S @@ -0,0 +1,21 @@ +/* PowerPC64 __mpn_addmul_1 -- Multiply a limb vector with a limb and subtract + the result to a second limb vector. + Copyright (C) 2013 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#define USE_AS_SUBMUL +#include "addmul_1.S" diff --git a/sysdeps/powerpc/powerpc64/sysdep.h b/sysdeps/powerpc/powerpc64/sysdep.h index 57fa8ba78f..112e4187c5 100644 --- a/sysdeps/powerpc/powerpc64/sysdep.h +++ b/sysdeps/powerpc/powerpc64/sysdep.h @@ -20,25 +20,67 @@ #ifdef __ASSEMBLER__ +/* Stack frame offsets. */ +#if _CALL_ELF != 2 +#define FRAME_MIN_SIZE 112 +#define FRAME_MIN_SIZE_PARM 112 +#define FRAME_BACKCHAIN 0 +#define FRAME_CR_SAVE 8 +#define FRAME_LR_SAVE 16 +#define FRAME_TOC_SAVE 40 +#define FRAME_PARM_SAVE 48 +#define FRAME_PARM1_SAVE 48 +#define FRAME_PARM2_SAVE 56 +#define FRAME_PARM3_SAVE 64 +#define FRAME_PARM4_SAVE 72 +#define FRAME_PARM5_SAVE 80 +#define FRAME_PARM6_SAVE 88 +#define FRAME_PARM7_SAVE 96 +#define FRAME_PARM8_SAVE 104 +#define FRAME_PARM9_SAVE 112 +#else +#define FRAME_MIN_SIZE 32 +#define FRAME_MIN_SIZE_PARM 96 +#define FRAME_BACKCHAIN 0 +#define FRAME_CR_SAVE 8 +#define FRAME_LR_SAVE 16 +#define FRAME_TOC_SAVE 24 +#define FRAME_PARM_SAVE 32 +#define FRAME_PARM1_SAVE 32 +#define FRAME_PARM2_SAVE 40 +#define FRAME_PARM3_SAVE 48 +#define FRAME_PARM4_SAVE 56 +#define FRAME_PARM5_SAVE 64 +#define FRAME_PARM6_SAVE 72 +#define FRAME_PARM7_SAVE 80 +#define FRAME_PARM8_SAVE 88 +#define FRAME_PARM9_SAVE 96 +#endif + /* Support macros for CALL_MCOUNT. */ +#if _CALL_ELF == 2 +#define call_mcount_parm_offset (-64) +#else +#define call_mcount_parm_offset FRAME_PARM_SAVE +#endif .macro SAVE_ARG NARG .if \NARG SAVE_ARG \NARG-1 - std 2+\NARG,40+8*(\NARG)(1) + std 2+\NARG,call_mcount_parm_offset-8+8*(\NARG)(1) .endif .endm .macro REST_ARG NARG .if \NARG REST_ARG \NARG-1 - ld 2+\NARG,112+40+8*(\NARG)(1) + ld 2+\NARG,FRAME_MIN_SIZE_PARM+call_mcount_parm_offset-8+8*(\NARG)(1) .endif .endm .macro CFI_SAVE_ARG NARG .if \NARG CFI_SAVE_ARG \NARG-1 - cfi_offset(2+\NARG,40+8*(\NARG)) + cfi_offset(2+\NARG,call_mcount_parm_offset-8+8*(\NARG)) .endif .endm @@ -55,25 +97,35 @@ #ifdef PROF mflr r0 SAVE_ARG \NARG - std r0,16(r1) - stdu r1,-112(r1) - cfi_adjust_cfa_offset(112) - cfi_offset(lr,16) + std r0,FRAME_LR_SAVE(r1) + stdu r1,-FRAME_MIN_SIZE_PARM(r1) + cfi_adjust_cfa_offset(FRAME_MIN_SIZE_PARM) + cfi_offset(lr,FRAME_LR_SAVE) CFI_SAVE_ARG \NARG bl JUMPTARGET (_mcount) #ifndef SHARED nop #endif - ld r0,128(r1) + ld r0,FRAME_MIN_SIZE_PARM+FRAME_LR_SAVE(r1) REST_ARG \NARG mtlr r0 - addi r1,r1,112 - cfi_adjust_cfa_offset(-112) + addi r1,r1,FRAME_MIN_SIZE_PARM + cfi_adjust_cfa_offset(-FRAME_MIN_SIZE_PARM) cfi_restore(lr) CFI_REST_ARG \NARG #endif .endm +#if _CALL_ELF != 2 + +/* Macro to prepare for calling via a function pointer. */ + .macro PPC64_LOAD_FUNCPTR PTR + ld r12,0(\PTR) + ld r2,8(\PTR) + mtctr r12 + ld r11,16(\PTR) + .endm + #ifdef USE_PPC64_OVERLAPPING_OPD # define OPD_ENT(name) .quad BODY_LABEL (name), .TOC.@tocbase #else @@ -81,7 +133,6 @@ #endif #define ENTRY_1(name) \ - .section ".text"; \ .type BODY_LABEL(name),@function; \ .globl name; \ .section ".opd","aw"; \ @@ -108,12 +159,37 @@ name##: OPD_ENT (name); \ .size name,.-BODY_LABEL(name); \ .size BODY_LABEL(name),.-BODY_LABEL(name); #endif +#define LOCALENTRY(name) + +#else /* _CALL_ELF */ + +/* Macro to prepare for calling via a function pointer. */ + .macro PPC64_LOAD_FUNCPTR PTR + mr r12,\PTR + mtctr r12 + .endm + +#define DOT_LABEL(X) X +#define BODY_LABEL(X) X +#define ENTRY_2(name) \ + .globl name; \ + .type name,@function; +#define END_2(name) \ + .size name,.-name; +#define LOCALENTRY(name) \ +1: addis r2,r12,.TOC.-1b@ha; \ + addi r2,r2,.TOC.-1b@l; \ + .localentry name,.-name; + +#endif /* _CALL_ELF */ #define ENTRY(name) \ + .section ".text"; \ ENTRY_2(name) \ .align ALIGNARG(2); \ BODY_LABEL(name): \ - cfi_startproc; + cfi_startproc; \ + LOCALENTRY(name) #define EALIGN_W_0 /* No words to insert. */ #define EALIGN_W_1 nop @@ -127,11 +203,13 @@ BODY_LABEL(name): \ /* EALIGN is like ENTRY, but does alignment to 'words'*4 bytes past a 2^alignt boundary. */ #define EALIGN(name, alignt, words) \ + .section ".text"; \ ENTRY_2(name) \ .align ALIGNARG(alignt); \ EALIGN_W_##words; \ BODY_LABEL(name): \ - cfi_startproc; + cfi_startproc; \ + LOCALENTRY(name) /* Local labels stripped out by the linker. */ #undef L @@ -231,15 +309,15 @@ LT_LABELSUFFIX(name,_name_end): ; \ .else; \ .Local_syscall_error: \ mflr 0; \ - std 0,16(1); \ - stdu 1,-112(1); \ - cfi_adjust_cfa_offset(112); \ - cfi_offset(lr,16); \ + std 0,FRAME_LR_SAVE(1); \ + stdu 1,-FRAME_MIN_SIZE(1); \ + cfi_adjust_cfa_offset(FRAME_MIN_SIZE); \ + cfi_offset(lr,FRAME_LR_SAVE); \ bl JUMPTARGET(__syscall_error); \ nop; \ - ld 0,112+16(1); \ - addi 1,1,112; \ - cfi_adjust_cfa_offset(-112); \ + ld 0,FRAME_MIN_SIZE+FRAME_LR_SAVE(1); \ + addi 1,1,FRAME_MIN_SIZE; \ + cfi_adjust_cfa_offset(-FRAME_MIN_SIZE); \ mtlr 0; \ cfi_restore(lr); \ blr; \ @@ -286,27 +364,68 @@ LT_LABELSUFFIX(name,_name_end): ; \ #else /* !__ASSEMBLER__ */ +#if _CALL_ELF != 2 + +#define PPC64_LOAD_FUNCPTR(ptr) \ + "ld 12,0(" #ptr ");\n" \ + "ld 2,8(" #ptr ");\n" \ + "mtctr 12;\n" \ + "ld 11,16(" #ptr ");" + #ifdef USE_PPC64_OVERLAPPING_OPD # define OPD_ENT(name) ".quad " BODY_PREFIX #name ", .TOC.@tocbase;" #else # define OPD_ENT(name) ".quad " BODY_PREFIX #name ", .TOC.@tocbase, 0;" #endif +#define ENTRY_1(name) \ + ".type " BODY_PREFIX #name ",@function;\n" \ + ".globl " #name ";\n" \ + ".pushsection \".opd\",\"aw\";\n" \ + ".align 3;\n" \ +#name ":\n" \ + OPD_ENT (name) "\n" \ + ".popsection;" + #ifdef HAVE_ASM_GLOBAL_DOT_NAME # define DOT_PREFIX "." # define BODY_PREFIX "." # define ENTRY_2(name) \ ".globl " BODY_PREFIX #name ";\n" \ + ENTRY_1(name) "\n" \ ".size " #name ", 24;" # define END_2(name) \ ".size " BODY_PREFIX #name ",.-" BODY_PREFIX #name ";" #else # define DOT_PREFIX "" # define BODY_PREFIX ".LY" -# define ENTRY_2(name) ".type " #name ",@function;" +# define ENTRY_2(name) \ + ".type " #name ",@function;\n" \ + ENTRY_1(name) # define END_2(name) \ ".size " #name ",.-" BODY_PREFIX #name ";\n" \ ".size " BODY_PREFIX #name ",.-" BODY_PREFIX #name ";" #endif +#define LOCALENTRY(name) + +#else /* _CALL_ELF */ + +#define PPC64_LOAD_FUNCPTR(ptr) \ + "mr 12," #ptr ";\n" \ + "mtctr 12;" + +#define DOT_PREFIX "" +#define BODY_PREFIX "" +#define ENTRY_2(name) \ + ".type " #name ",@function;\n" \ + ".globl " #name ";" +#define END_2(name) \ + ".size " #name ",.-" #name ";" +#define LOCALENTRY(name) \ + "1: addis 2,12,.TOC.-1b@ha;\n" \ + "addi 2,2,.TOC.-1b@l;\n" \ + ".localentry " #name ",.-" #name ";" + +#endif /* _CALL_ELF */ #endif /* __ASSEMBLER__ */ diff --git a/sysdeps/powerpc/powerpc64/tst-audit.h b/sysdeps/powerpc/powerpc64/tst-audit.h index ad6545ed3a..0fbe1fec52 100644 --- a/sysdeps/powerpc/powerpc64/tst-audit.h +++ b/sysdeps/powerpc/powerpc64/tst-audit.h @@ -18,8 +18,16 @@ License along with the GNU C Library. If not, see <http://www.gnu.org/licenses/>. */ +#if _CALL_ELF != 2 #define pltenter la_ppc64_gnu_pltenter #define pltexit la_ppc64_gnu_pltexit #define La_regs La_ppc64_regs #define La_retval La_ppc64_retval #define int_retval lrv_r3 +#else +#define pltenter la_ppc64v2_gnu_pltenter +#define pltexit la_ppc64v2_gnu_pltexit +#define La_regs La_ppc64v2_regs +#define La_retval La_ppc64v2_retval +#define int_retval lrv_r3 +#endif |