diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc64/power4/memset.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/power4/memset.S | 79 |
1 files changed, 27 insertions, 52 deletions
diff --git a/sysdeps/powerpc/powerpc64/power4/memset.S b/sysdeps/powerpc/powerpc64/power4/memset.S index c86a68a042..3a1e9dc76a 100644 --- a/sysdeps/powerpc/powerpc64/power4/memset.S +++ b/sysdeps/powerpc/powerpc64/power4/memset.S @@ -1,6 +1,5 @@ /* Optimized memset implementation for PowerPC64. - Copyright (C) 1997, 1999, 2000, 2002, 2003, 2007 - Free Software Foundation, Inc. + Copyright (C) 1997-2014 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 @@ -18,8 +17,6 @@ <http://www.gnu.org/licenses/>. */ #include <sysdep.h> -#include <bp-sym.h> -#include <bp-asm.h> /* __ptr_t [r3] memset (__ptr_t s [r3], int c [r4], size_t n [r5])); Returns 's'. @@ -29,22 +26,15 @@ to 0, to take advantage of the dcbz instruction. */ .machine power4 -EALIGN (BP_SYM (memset), 5, 0) +EALIGN (memset, 5, 0) CALL_MCOUNT 3 #define rTMP r0 #define rRTN r3 /* Initial value of 1st argument. */ -#if __BOUNDED_POINTERS__ -# define rMEMP0 r4 /* Original value of 1st arg. */ -# define rCHR r5 /* Char to set in each byte. */ -# define rLEN r6 /* Length of region to set. */ -# define rMEMP r10 /* Address at which we are storing. */ -#else -# define rMEMP0 r3 /* Original value of 1st arg. */ -# define rCHR r4 /* Char to set in each byte. */ -# define rLEN r5 /* Length of region to set. */ -# define rMEMP r6 /* Address at which we are storing. */ -#endif +#define rMEMP0 r3 /* Original value of 1st arg. */ +#define rCHR r4 /* Char to set in each byte. */ +#define rLEN r5 /* Length of region to set. */ +#define rMEMP r6 /* Address at which we are storing. */ #define rALIGN r7 /* Number of bytes we are setting now (when aligning). */ #define rMEMP2 r8 @@ -52,14 +42,6 @@ EALIGN (BP_SYM (memset), 5, 0) #define rCLS r8 /* Cache line size obtained from static. */ #define rCLM r9 /* Cache line size mask to check for cache alignment. */ L(_memset): -#if __BOUNDED_POINTERS__ - cmpldi cr1, rRTN, 0 - CHECK_BOUNDS_BOTH_WIDE (rMEMP0, rTMP, rTMP2, rLEN) - beq cr1, L(b0) - STORE_RETURN_VALUE (rMEMP0) - STORE_RETURN_BOUNDS (rTMP, rTMP2) -L(b0): -#endif /* Take care of case for size <= 4. */ cmpldi cr1, rLEN, 8 andi. rALIGN, rMEMP0, 7 @@ -68,14 +50,14 @@ L(b0): /* 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) @@ -97,14 +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 @@ -164,24 +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. @@ -218,7 +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 @@ -250,25 +232,18 @@ L(medium_27f): L(medium_28t): std rCHR, -8(rMEMP) blr -END_GEN_TB (BP_SYM (memset),TB_TOCLESS) +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 (BP_SYM (__bzero)) +ENTRY (__bzero) CALL_MCOUNT 3 -#if __BOUNDED_POINTERS__ - mr r6,r4 - li r5,0 - mr r4,r3 - /* Tell memset that we don't want a return value. */ - li r3,0 - b L(_memset) -#else mr r5,r4 li r4,0 b L(_memset) -#endif -END_GEN_TB (BP_SYM (__bzero),TB_TOCLESS) +END_GEN_TB (__bzero,TB_TOCLESS) -weak_alias (BP_SYM (__bzero), BP_SYM (bzero)) +weak_alias (__bzero, bzero) +#endif |