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

Final pass of today's vtlb optimizations: Improved the codegen for const-propagated direct reads and writes (very minor optimization). 

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@885 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
Jake.Stine 2009-04-02 15:17:47 +00:00
parent d60718e79d
commit a8d9cbc25d
2 changed files with 111 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

116
pcsx2/x86/ix86-32/recVTLB.cpp
View File

@ -248,8 +248,46 @@ void vtlb_DynGenRead64_Const( u32 bits, u32 addr_const )
s32 ppf = addr_const + vmv_ptr;
if( ppf >= 0 )
{
MOV32ItoR( ECX, ppf );
_vtlb_DynGen_DirectRead( bits, false );
switch( bits )
{
case 64:
if( _hasFreeMMXreg() )
{
const int freereg = _allocMMXreg(-1, MMX_TEMP, 0);
MOVQMtoR(freereg,ppf);
MOVQRtoRmOffset(EDX,freereg,0);
_freeMMXreg(freereg);
}
else
{
MOV32MtoR(EAX,ppf);
MOV32RtoRm(EDX,EAX);
MOV32MtoR(EAX,ppf+4);
MOV32RtoRmOffset(EDX,EAX,4);
}
break;
case 128:
if( _hasFreeXMMreg() )
{
const int freereg = _allocTempXMMreg( XMMT_INT, -1 );
SSE2_MOVDQA_M128_to_XMM( freereg, ppf );
SSE2_MOVDQARtoRmOffset(EDX,freereg,0);
_freeXMMreg(freereg);
}
else
{
// Could put in an MMX optimization here as well, but no point really.
// It's almost never used since there's almost always a free XMM reg.
MOV32ItoR( ECX, ppf );
MOV128_MtoM( EDX, ECX ); // dest <- src!
}
break;
jNO_DEFAULT
}
}
else
{
@ -281,8 +319,26 @@ void vtlb_DynGenRead32_Const( u32 bits, bool sign, u32 addr_const )
s32 ppf = addr_const + vmv_ptr;
if( ppf >= 0 )
{
MOV32ItoR( ECX, ppf );
_vtlb_DynGen_DirectRead( bits, sign );
switch( bits )
{
case 8:
if( sign )
MOVSX32M8toR(EAX,ppf);
else
MOVZX32M8toR(EAX,ppf);
break;
case 16:
if( sign )
MOVSX32M16toR(EAX,ppf);
else
MOVZX32M16toR(EAX,ppf);
break;
case 32:
MOV32MtoR(EAX,ppf);
break;
}
}
else
{
@ -429,8 +485,56 @@ void vtlb_DynGenWrite_Const( u32 bits, u32 addr_const )
s32 ppf = addr_const + vmv_ptr;
if( ppf >= 0 )
{
MOV32ItoR( ECX, ppf );
_vtlb_DynGen_DirectWrite( bits );
switch(bits)
{
//8 , 16, 32 : data on EDX
case 8:
MOV8RtoM(ppf,EDX);
break;
case 16:
MOV16RtoM(ppf,EDX);
break;
case 32:
MOV32RtoM(ppf,EDX);
break;
case 64:
if( _hasFreeMMXreg() )
{
const int freereg = _allocMMXreg(-1, MMX_TEMP, 0);
MOVQRmtoROffset(freereg,EDX,0);
MOVQRtoM(ppf,freereg);
_freeMMXreg( freereg );
}
else
{
MOV32RmtoR(EAX,EDX);
MOV32RtoM(ppf,EAX);
MOV32RmtoROffset(EAX,EDX,4);
MOV32RtoM(ppf+4,EAX);
}
break;
case 128:
if( _hasFreeXMMreg() )
{
const int freereg = _allocTempXMMreg( XMMT_INT, -1 );
SSE2_MOVDQARmtoROffset(freereg,EDX,0);
SSE2_MOVDQA_XMM_to_M128(ppf,freereg);
_freeXMMreg( freereg );
}
else
{
// Could put in an MMX optimization here as well, but no point really.
// It's almost never used since there's almost always a free XMM reg.
MOV32ItoR( ECX, ppf );
MOV128_MtoM( ECX, EDX ); // dest <- src!
}
break;
}
}
else
{

2
pcsx2/x86/ix86/ix86_macros.h
View File

@ -708,7 +708,7 @@
#define SSE2_MOVDQA_XMM_to_XMM eSSE2_MOVDQA_XMM_to_XMM<_EmitterId_>
#define SSE2_MOVDQU_M128_to_XMM eSSE2_MOVDQU_M128_to_XMM<_EmitterId_>
#define SSE2_MOVDQU_XMM_to_M128 eSSE2_MOVDQU_XMM_to_M128<_EmitterId_>
#define SSE2_MOVDQU_XMM_to_XMM eSSE2_MOVDQU_XMM_to_XMM<_EmitterId_>
#define SSE2_MOVDQU_XMM_to_XMM eSSE2_MOVDQA_XMM_to_XMM<_EmitterId_>
#define SSE2_PSRLW_XMM_to_XMM eSSE2_PSRLW_XMM_to_XMM<_EmitterId_>
#define SSE2_PSRLW_M128_to_XMM eSSE2_PSRLW_M128_to_XMM<_EmitterId_>
#define SSE2_PSRLW_I8_to_XMM eSSE2_PSRLW_I8_to_XMM<_EmitterId_>