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

gsdx sw x64: shuffle memory allocation 

tex address is a3
vm address is a1

Could help to avoid REX prefix
Reduce prologue/epilogue register copy

Byte code size 41893 => 38912 (on my testcase)
This commit is contained in:
Gregory Hainaut 2016-11-20 19:21:24 +01:00
parent b76305a0d2
commit 37379d5d1d
1 changed files with 24 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
plugins/GSdx/GSDrawScanlineCodeGenerator.x64.avx.cpp
View File

@ -26,11 +26,11 @@
#if _M_SSE < 0x501 && (defined(_M_AMD64) || defined(_WIN64)) #if _M_SSE < 0x501 && (defined(_M_AMD64) || defined(_WIN64))
// Ease the reading of the code // Ease the reading of the code
#define _m_local r11 #define _m_local r12
#define _m_local__gd r12 #define _m_local__gd r13
#define _m_local__gd__vm r13 #define _m_local__gd__vm a1
#define _m_local__gd__clut r14 #define _m_local__gd__clut r11
#define _m_local__gd__tex r15 #define _m_local__gd__tex a3
// More pretty name // More pretty name
#define _z xmm8 #define _z xmm8
#define _f xmm9 #define _f xmm9
@ -89,23 +89,22 @@ void GSDrawScanlineCodeGenerator::Generate_AVX()
// No reservation on the stack as a red zone is available // No reservation on the stack as a red zone is available
push(rbp); push(rbp);
mov(ptr[rsp + _rz_rbx], rbx); mov(ptr[rsp + _rz_rbx], rbx);
mov(ptr[rsp + _rz_r12], r12); if (!m_rip)
mov(ptr[rsp + _rz_r13], r13); {
if(need_clut) mov(ptr[rsp + _rz_r12], r12);
mov(ptr[rsp + _rz_r14], r14); mov(ptr[rsp + _rz_r13], r13);
if(need_tex) }
mov(ptr[rsp + _rz_r15], r15);
#endif #endif
mov(r10, (size_t)&m_test[0]); mov(r10, (size_t)&m_test[0]);
mov(_m_local, (size_t)&m_local); if (!m_rip)
mov(_m_local__gd, _rip_local(gd)); {
mov(_m_local, (size_t)&m_local);
mov(_m_local__gd, _rip_local(gd));
}
mov(_m_local__gd__vm, _rip_global(vm));
if(need_clut) if(need_clut)
mov(_m_local__gd__clut, _rip_global(clut)); mov(_m_local__gd__clut, _rip_global(clut));
if(need_tex)
mov(_m_local__gd__tex, _rip_global(tex));
Init_AVX(); Init_AVX();
@ -272,12 +271,11 @@ L("exit");
pop(rbx); pop(rbx);
#else #else
mov(rbx, ptr[rsp + _rz_rbx]); mov(rbx, ptr[rsp + _rz_rbx]);
mov(r12, ptr[rsp + _rz_r12]); if (!m_rip)
mov(r13, ptr[rsp + _rz_r13]); {
if(need_clut) mov(r12, ptr[rsp + _rz_r12]);
mov(r14, ptr[rsp + _rz_r14]); mov(r13, ptr[rsp + _rz_r13]);
if(need_tex) }
mov(r15, ptr[rsp + _rz_r15]);
pop(rbp); pop(rbp);
#endif #endif
@ -482,13 +480,16 @@ void GSDrawScanlineCodeGenerator::Init_AVX()
} }
} }
if(m_sel.fwrite && m_sel.fpsm == 2 && m_sel.dthe) if(m_sel.fwrite && m_sel.fpsm == 2 && m_sel.dthe)
{ {
// On linux, a2 is edx which will be used for fzm // On linux, a2 is edx which will be used for fzm
// In all case, it will require a mov in dthe code, so let's keep the value on the stack // In all case, it will require a mov in dthe code, so let's keep the value on the stack
mov(ptr[rsp + _rz_top], a2); mov(ptr[rsp + _rz_top], a2);
} }
mov(_m_local__gd__vm, _rip_global(vm));
if(m_sel.fb && m_sel.tfx != TFX_NONE)
mov(_m_local__gd__tex, _rip_global(tex));
} }
void GSDrawScanlineCodeGenerator::Step_AVX() void GSDrawScanlineCodeGenerator::Step_AVX()