diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc64/strncmp.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/strncmp.S | 86 |
1 files changed, 64 insertions, 22 deletions
diff --git a/sysdeps/powerpc/powerpc64/strncmp.S b/sysdeps/powerpc/powerpc64/strncmp.S index 7d8ab4d471..b3f12a8292 100644 --- a/sysdeps/powerpc/powerpc64/strncmp.S +++ b/sysdeps/powerpc/powerpc64/strncmp.S @@ -1,5 +1,5 @@ /* Optimized strcmp implementation for PowerPC64. - Copyright (C) 2003, 2011 Free Software Foundation, Inc. + Copyright (C) 2003-2014 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 @@ -17,30 +17,26 @@ <http://www.gnu.org/licenses/>. */ #include <sysdep.h> -#include <bp-sym.h> -#include <bp-asm.h> /* See strlen.s for comments on how the end-of-string testing works. */ /* int [r3] strncmp (const char *s1 [r3], const char *s2 [r4], size_t size [r5]) */ -EALIGN (BP_SYM(strncmp), 4, 0) +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 */ #define rN r5 /* max string length */ -/* Note: The Bounded pointer support in this code is broken. This code - was inherited from PPC32 and that support was never completed. - Current PPC gcc does not support -fbounds-check or -fbounded-pointers. */ #define rWORD1 r6 /* current word in s1 */ #define rWORD2 r7 /* current word in s2 */ #define rFEFE r8 /* constant 0xfefefefefefefeff (-0x0101010101010101) */ #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 @@ -50,13 +46,13 @@ EALIGN (BP_SYM(strncmp), 4, 0) cmpldi cr1, rN, 0 lis rFEFE, -0x101 bne L(unaligned) -/* We are doubleword alligned so set up for two loops. first a double word +/* We are doubleword aligned so set up for two loops. first a double word loop, then fall into the byte loop if any residual. */ srdi. rTMP, rN, 3 clrldi rN, rN, 61 addi rFEFE, rFEFE, -0x101 addi r7F7F, r7F7F, 0x7f7f - cmpldi cr1, rN, 0 + cmpldi cr1, rN, 0 beq L(unaligned) mtctr rTMP /* Power4 wants mtctr 1st in dispatch group. */ @@ -67,7 +63,7 @@ EALIGN (BP_SYM(strncmp), 4, 0) add rFEFE, rFEFE, rTMP b L(g1) -L(g0): +L(g0): ldu rWORD1, 8(rSTR1) bne- cr1, L(different) ldu rWORD2, 8(rSTR2) @@ -77,17 +73,64 @@ L(g1): add rTMP, rFEFE, rWORD1 and. rTMP, rTMP, rNEG cmpd cr1, rWORD1, rWORD2 beq+ L(g0) - + /* 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 @@ -96,7 +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): @@ -104,7 +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) @@ -112,11 +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 @@ -157,8 +199,8 @@ L(u1): lbzu rWORD1, 1(rSTR1) bne+ cr1, L(u0) -L(u2): lbzu rWORD1, -1(rSTR1) +L(u2): lbzu rWORD1, -1(rSTR1) L(u3): sub rRTN, rWORD1, rWORD2 blr -END (BP_SYM (strncmp)) +END (strncmp) libc_hidden_builtin_def (strncmp) |