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

Merge pull request #304 from Tilka/nop 

Optimize NOPs
This commit is contained in:
Pierre Bourdon 2014-04-27 11:52:05 +02:00
parent a093b41231 957649b7af
commit aef24d509b
2 changed files with 55 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

68
Source/Core/Common/x64Emitter.cpp
View File

@ -515,23 +515,63 @@ void XEmitter::INT3() {Write8(0xCC);}
void XEmitter::RET() {Write8(0xC3);} void XEmitter::RET() {Write8(0xC3);}
void XEmitter::RET_FAST() {Write8(0xF3); Write8(0xC3);} //two-byte return (rep ret) - recommended by AMD optimization manual for the case of jumping to a ret void XEmitter::RET_FAST() {Write8(0xF3); Write8(0xC3);} //two-byte return (rep ret) - recommended by AMD optimization manual for the case of jumping to a ret
void XEmitter::NOP(int count) // The first sign of decadence: optimized NOPs.
void XEmitter::NOP(int size)
{ {
// TODO: look up the fastest nop sleds for various sizes while (true)
int i; {
switch (count) { switch (size)
case 1: {
Write8(0x90); case 0:
break; return;
case 2: case 1:
Write8(0x66);
Write8(0x90);
break;
default:
for (i = 0; i < count; i++) {
Write8(0x90); Write8(0x90);
return;
case 2:
Write8(0x66); Write8(0x90);
return;
case 3:
Write8(0x0F); Write8(0x1F); Write8(0x00);
return;
case 4:
Write8(0x0F); Write8(0x1F); Write8(0x40); Write8(0x00);
return;
case 5:
Write8(0x0F); Write8(0x1F); Write8(0x44); Write8(0x00);
Write8(0x00);
return;
case 6:
Write8(0x66); Write8(0x0F); Write8(0x1F); Write8(0x44);
Write8(0x00); Write8(0x00);
return;
case 7:
Write8(0x0F); Write8(0x1F); Write8(0x80); Write8(0x00);
Write8(0x00); Write8(0x00); Write8(0x00);
return;
case 8:
Write8(0x0F); Write8(0x1F); Write8(0x84); Write8(0x00);
Write8(0x00); Write8(0x00); Write8(0x00); Write8(0x00);
return;
case 9:
Write8(0x66); Write8(0x0F); Write8(0x1F); Write8(0x84);
Write8(0x00); Write8(0x00); Write8(0x00); Write8(0x00);
Write8(0x00);
return;
case 10:
Write8(0x66); Write8(0x66); Write8(0x0F); Write8(0x1F);
Write8(0x84); Write8(0x00); Write8(0x00); Write8(0x00);
Write8(0x00); Write8(0x00);
return;
default:
// Even though x86 instructions are allowed to be up to 15 bytes long,
// AMD advises against using NOPs longer than 11 bytes because they
// carry a performance penalty on CPUs older than AMD family 16h.
Write8(0x66); Write8(0x66); Write8(0x66); Write8(0x0F);
Write8(0x1F); Write8(0x84); Write8(0x00); Write8(0x00);
Write8(0x00); Write8(0x00); Write8(0x00);
size -= 11;
continue;
} }
break;
} }
} }

2
Source/Core/Common/x64Emitter.h
View File

@ -290,7 +290,7 @@ public:
void INT3(); void INT3();
// Do nothing // Do nothing
void NOP(int count = 1); //nop padding - TODO: fast nop slides, for amd and intel (check their manuals) void NOP(int count = 1);
// Save energy in wait-loops on P4 only. Probably not too useful. // Save energy in wait-loops on P4 only. Probably not too useful.
void PAUSE(); void PAUSE();