ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Memcpy (linux): Mangle asm label name avoid symbol already defined when inline the function 

Note: the function can be moved into a .h ;)

Note2: %= is replaced by a number so it is a bad idea to put it after a digit (reason why I put underscore before)



git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3563 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
gregory.hainaut 2010-07-25 09:38:26 +00:00
parent 4c4cf75432
commit f965b0af46
1 changed files with 10 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
common/src/Utilities/x86/MemcpyVibes.cpp
View File

@ -185,12 +185,12 @@ __forceinline void memcpy_vibes(void * dest, const void * src, int size) {
".intel_syntax noprefix\n"
"mov eax, %[qwc]\n" // keep a copy of count for looping
"shr eax, 1\n"
"jz memcpy_qwc_1\n" // only one 16 byte block to copy?
"jz memcpy_qwc_1_%=\n" // only one 16 byte block to copy?
"cmp eax, 64\n" // "IN_CACHE_COPY/32"
"jb memcpy_qwc_loop1\n" // small copies should be cached (definite speedup --air)
"jb memcpy_qwc_loop1_%=\n" // small copies should be cached (definite speedup --air)
"memcpy_qwc_loop2:\n" // 32-byte blocks, uncached copy
"memcpy_qwc_loop2_%=:\n" // 32-byte blocks, uncached copy
"prefetchnta [%[src] + 568]\n" // start reading ahead (tested: it helps! --air)
"movq mm0,[%[src]+0]\n" // read 64 bits
@ -205,15 +205,15 @@ __forceinline void memcpy_vibes(void * dest, const void * src, int size) {
"add %[src],32\n" // update source pointer
"add %[dest],32\n" // update destination pointer
"sub eax,1\n"
"jnz memcpy_qwc_loop2\n" // last 64-byte block?
"jnz memcpy_qwc_loop2_%=\n" // last 64-byte block?
"sfence\n" // flush the write buffer
"jmp memcpy_qwc_1\n"
"jmp memcpy_qwc_1_%=\n"
// 32-byte blocks, cached!
// This *is* important. Removing this and using exclusively non-temporal stores
// results in noticeable speed loss!
"memcpy_qwc_loop1:\n"
"memcpy_qwc_loop1_%=:\n"
"prefetchnta [%[src] + 568]\n" // start reading ahead (tested: it helps! --air)
"movq mm0,[%[src]+0]\n" // read 64 bits
@ -228,17 +228,17 @@ __forceinline void memcpy_vibes(void * dest, const void * src, int size) {
"add %[src],32\n" // update source pointer
"add %[dest],32\n" // update destination pointer
"sub eax,1\n"
"jnz memcpy_qwc_loop1\n" // last 64-byte block?
"jnz memcpy_qwc_loop1_%=\n" // last 64-byte block?
"memcpy_qwc_1:\n"
"memcpy_qwc_1_%=:\n"
"test %[qwc],1\n"
"jz memcpy_qwc_final\n"
"jz memcpy_qwc_final_%=\n"
"movq mm0,[%[src]]\n"
"movq mm1,[%[src]+8]\n"
"movq [%[dest]], mm0\n"
"movq [%[dest]+8], mm1\n"
"memcpy_qwc_final:\n"
"memcpy_qwc_final_%=:\n"
"emms\n" // clean up the MMX state
".att_syntax\n"
: "=&r"(dest), "=&r"(src), "=&r"(qwc)