piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-23 06:09:50 -06:00
Code Issues Packages Projects Releases Wiki Activity

x64Emitter: add LZCNT/TZCNT support and detection

Browse Source 
Also add a unit test.
This commit is contained in:
Fiora
2014-09-14 05:31:22 -07:00
parent 84a564a304
commit 997c5c2d0e
4 changed files with 51 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
Source/Core/Common/x64CPUDetect.cpp
View File

@ -197,6 +197,7 @@ void CPUInfo::Detect()
// Check for more features.
__cpuid(cpu_id, 0x80000001);
if (cpu_id[2] & 1) bLAHFSAHF64 = true;
if ((cpu_id[2] >> 5) & 1) bLZCNT = true;
if ((cpu_id[3] >> 29) & 1) bLongMode = true;
}

17
Source/Core/Common/x64Emitter.cpp
View File

@ -750,12 +750,14 @@ void XEmitter::IDIV(int bits, OpArg src) {WriteMulDivType(bits, src, 7);}
void XEmitter::NEG(int bits, OpArg src) {WriteMulDivType(bits, src, 3);}
void XEmitter::NOT(int bits, OpArg src) {WriteMulDivType(bits, src, 2);}
void XEmitter::WriteBitSearchType(int bits, X64Reg dest, OpArg src, u8 byte2)
void XEmitter::WriteBitSearchType(int bits, X64Reg dest, OpArg src, u8 byte2, bool rep)
{
_assert_msg_(DYNA_REC, !src.IsImm(), "WriteBitSearchType - Imm argument");
src.operandReg = (u8)dest;
if (bits == 16)
Write8(0x66);
if (rep)
Write8(0xF3);
src.WriteRex(this, bits, bits);
Write8(0x0F);
Write8(byte2);
@ -772,6 +774,19 @@ void XEmitter::MOVNTI(int bits, OpArg dest, X64Reg src)
void XEmitter::BSF(int bits, X64Reg dest, OpArg src) {WriteBitSearchType(bits,dest,src,0xBC);} //bottom bit to top bit
void XEmitter::BSR(int bits, X64Reg dest, OpArg src) {WriteBitSearchType(bits,dest,src,0xBD);} //top bit to bottom bit
void XEmitter::TZCNT(int bits, X64Reg dest, OpArg src)
{
if (!cpu_info.bBMI1)
PanicAlert("Trying to use BMI1 on a system that doesn't support it. Bad programmer.");
WriteBitSearchType(bits, dest, src, 0xBC, true);
}
void XEmitter::LZCNT(int bits, X64Reg dest, OpArg src)
{
if (!cpu_info.bLZCNT)
PanicAlert("Trying to use LZCNT on a system that doesn't support it. Bad programmer.");
WriteBitSearchType(bits, dest, src, 0xBD, true);
}
void XEmitter::MOVSX(int dbits, int sbits, X64Reg dest, OpArg src)
{
_assert_msg_(DYNA_REC, !src.IsImm(), "MOVSX - Imm argument");

7
Source/Core/Common/x64Emitter.h
View File

@ -266,7 +266,7 @@ private:
void WriteSimple1Byte(int bits, u8 byte, X64Reg reg);
void WriteSimple2Byte(int bits, u8 byte1, u8 byte2, X64Reg reg);
void WriteMulDivType(int bits, OpArg src, int ext);
void WriteBitSearchType(int bits, X64Reg dest, OpArg src, u8 byte2);
void WriteBitSearchType(int bits, X64Reg dest, OpArg src, u8 byte2, bool rep = false);
void WriteShift(int bits, OpArg dest, OpArg &shift, int ext);
void WriteBitTest(int bits, OpArg &dest, OpArg &index, int ext);
void WriteMXCSR(OpArg arg, int ext);
@ -454,6 +454,11 @@ public:
// Available only on Atom or >= Haswell so far. Test with cpu_info.bMOVBE.
void MOVBE(int dbits, const OpArg& dest, const OpArg& src);
// Available only on AMD >= Phenom or Intel >= Haswell
void LZCNT(int bits, X64Reg dest, OpArg src);
// Note: this one is actually part of BMI1
void TZCNT(int bits, X64Reg dest, OpArg src);
// WARNING - These two take 11-13 cycles and are VectorPath! (AMD64)
void STMXCSR(OpArg memloc);
void LDMXCSR(OpArg memloc);