diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc64/memset.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/memset.S | 54 |
1 files changed, 24 insertions, 30 deletions
diff --git a/sysdeps/powerpc/powerpc64/memset.S b/sysdeps/powerpc/powerpc64/memset.S index 4bfe20d7b1..53a4a2753d 100644 --- a/sysdeps/powerpc/powerpc64/memset.S +++ b/sysdeps/powerpc/powerpc64/memset.S @@ -1,5 +1,5 @@ /* Optimized memset implementation for PowerPC64. - Copyright (C) 1997, 1999, 2000, 2002 Free Software Foundation, Inc. + Copyright (C) 1997, 1999, 2000, 2002, 2003 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 @@ -21,17 +21,12 @@ #include <bp-sym.h> #include <bp-asm.h> -/* Define a global static that can hold the cache line size. The +/* Define a global static that can hold the cache line size. The assumption is that startup code will access the "aux vector" to - to obtain the value set by the kernel and store it into this + to obtain the value set by the kernel and store it into this variable. */ .globl __cache_line_size - .section ".data" - .align 2 - .type __cache_line_size,@object - .size __cache_line_size,4 -__cache_line_size: - .long 0 + .lcomm __cache_line_size,4,4 .section ".toc","aw" .LC0: .tc __cache_line_size[TC],__cache_line_size @@ -81,7 +76,7 @@ L(b0): andi. rALIGN, rMEMP0, 7 mr rMEMP, rMEMP0 ble- cr1, L(small) - + /* Align to doubleword boundary. */ cmpldi cr5, rLEN, 31 rlwimi rCHR, rCHR, 8, 16, 23 /* Replicate byte to halfword. */ @@ -108,9 +103,9 @@ L(g4): bf+ 31, L(g0) stb rCHR, 0(rMEMP0) bt 30, L(aligned) -L(g0): - sth rCHR, -2(rMEMP) - +L(g0): + sth rCHR, -2(rMEMP) + /* Handle the case of size < 31. */ L(aligned2): rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ @@ -138,9 +133,9 @@ L(a2): L(caligned): cmpldi cr1, rCHR, 0 clrrdi. rALIGN, rLEN, 5 - mtcrf 0x01, rLEN + mtcrf 0x01, rLEN beq cr1, L(zloopstart) /* Special case for clearing memory using dcbz. */ -L(nondcbz): +L(nondcbz): srdi rTMP, rALIGN, 5 mtctr rTMP beq L(medium) /* We may not actually get to do a full line. */ @@ -168,21 +163,21 @@ L(cloopdone): .align 5 /* Clear lines of memory in 128-byte chunks. */ L(zloopstart): -/* If the remaining length is less the 32 bytes, don't bother getting +/* If the remaining length is less the 32 bytes, don't bother getting the cache line size. */ beq L(medium) ld rCLS,.LC0@toc(r2) - lwz rCLS,0(rCLS) -/* If the cache line size was not set just goto to L(nondcbz) which is - safe for any cache line size. */ + lwz rCLS,0(rCLS) +/* If the cache line size was not set just goto to L(nondcbz) which is + safe for any cache line size. */ cmpldi cr1,rCLS,0 beq cr1,L(nondcbz) - - + + /* Now we know the cache line size, and it is not 32-bytes, but - we may not yet be aligned to the cache line. May have a partial - line to fill, so touch it 1st. */ - dcbt 0,rMEMP + we may not yet be aligned to the cache line. May have a partial + line to fill, so touch it 1st. */ + dcbt 0,rMEMP addi rCLM,rCLS,-1 L(getCacheAligned): cmpldi cr1,rLEN,32 @@ -196,8 +191,8 @@ L(getCacheAligned): std rCHR,-16(rMEMP) std rCHR,-8(rMEMP) b L(getCacheAligned) - -/* Now we are aligned to the cache line and can use dcbz. */ + +/* Now we are aligned to the cache line and can use dcbz. */ L(cacheAligned): cmpld cr1,rLEN,rCLS blt cr1,L(handletail32) @@ -208,7 +203,7 @@ 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. - So set up the preconditions for L(nondcbz) and go there. */ + So set up the preconditions for L(nondcbz) and go there. */ L(handletail32): clrrwi. rALIGN, rLEN, 5 b L(nondcbz) @@ -264,7 +259,7 @@ L(medium_30t): bf- 29, L(medium_29f) L(medium_29t): stwu rCHR, -4(rMEMP) - blt- cr1, L(medium_27f) + blt- cr1, L(medium_27f) L(medium_27t): std rCHR, -8(rMEMP) stdu rCHR, -16(rMEMP) @@ -275,7 +270,7 @@ L(medium_28t): blr END_GEN_TB (BP_SYM (memset),TB_TOCLESS) -/* Copied from bzero.S to prevent the linker from inserting a stub +/* Copied from bzero.S to prevent the linker from inserting a stub between bzero and memset. */ ENTRY (BP_SYM (__bzero)) #if __BOUNDED_POINTERS__ @@ -293,4 +288,3 @@ ENTRY (BP_SYM (__bzero)) END_GEN_TB (BP_SYM (__bzero),TB_TOCLESS) weak_alias (BP_SYM (__bzero), BP_SYM (bzero)) - |