From 56d3982dc51a9eb1c7ecafe4c8eb2c3d7afff2d0 Mon Sep 17 00:00:00 2001
From: "Jake.Stine" <Jake.Stine@96395faa-99c1-11dd-bbfe-3dabce05a288>
Date: Tue, 13 Jul 2010 04:36:39 +0000
Subject: [PATCH] * Minor optimization to GIFpath, by utilizing the precached
value for numregs in place of a convoluted test against NREG. * Disabled the
SSE store version of memzero (no performance benefits and it was messy
anyway)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3473 96395faa-99c1-11dd-bbfe-3dabce05a288
---
common/include/Utilities/win_memzero.h | 66 ++++++++++----------------
pcsx2/ps2/GIFpath.cpp | 2 +-
2 files changed, 25 insertions(+), 43 deletions(-)
diff --git a/common/include/Utilities/win_memzero.h b/common/include/Utilities/win_memzero.h
index 5207002cdc..639fba6b71 100644
--- a/common/include/Utilities/win_memzero.h
+++ b/common/include/Utilities/win_memzero.h
@@ -73,6 +73,10 @@ static __forceinline void memzero_ptr( void *dest )
return;
}
+#if 0
+ // SSE-based memory clear. Currently disabled so to avoid unnecessary dependence on
+ // SSE cpu instruction sets. (memzero typically isn't used in any performance critical
+ // situations anyway)
enum
{
remainder = MZFbytes & 127,
@@ -86,8 +90,6 @@ static __forceinline void memzero_ptr( void *dest )
if( (MZFbytes & 0xf) == 0 )
{
- u64 _xmm_backup[2];
-
if( ((uptr)dest & 0xf) != 0 )
{
// UNALIGNED COPY MODE.
@@ -97,24 +99,21 @@ static __forceinline void memzero_ptr( void *dest )
{
__asm
{
- movups _xmm_backup,xmm0;
mov ecx,dest
pxor xmm0,xmm0
mov eax,bytes128
- align 16
-
_loop_6:
- movups [ecx],xmm0;
- movups [ecx+0x10],xmm0;
- movups [ecx+0x20],xmm0;
- movups [ecx+0x30],xmm0;
- movups [ecx+0x40],xmm0;
- movups [ecx+0x50],xmm0;
- movups [ecx+0x60],xmm0;
- movups [ecx+0x70],xmm0;
+ movups [ecx],xmm0
+ movups [ecx+0x10],xmm0
+ movups [ecx+0x20],xmm0
+ movups [ecx+0x30],xmm0
+ movups [ecx+0x40],xmm0
+ movups [ecx+0x50],xmm0
+ movups [ecx+0x60],xmm0
+ movups [ecx+0x70],xmm0
sub ecx,-128
- dec eax;
+ sub eax,1
jnz _loop_6;
}
if( remainder != 0 )
@@ -130,10 +129,6 @@ static __forceinline void memzero_ptr( void *dest )
jnz _loop_5;
}
}
- __asm
- {
- movups xmm0,[_xmm_backup];
- }
return;
}
}
@@ -145,24 +140,21 @@ static __forceinline void memzero_ptr( void *dest )
__asm
{
- movups _xmm_backup,xmm0;
mov ecx,dest
pxor xmm0,xmm0
mov eax,bytes128
- align 16
-
_loop_8:
- movaps [ecx],xmm0;
- movaps [ecx+0x10],xmm0;
- movaps [ecx+0x20],xmm0;
- movaps [ecx+0x30],xmm0;
- movaps [ecx+0x40],xmm0;
- movaps [ecx+0x50],xmm0;
- movaps [ecx+0x60],xmm0;
- movaps [ecx+0x70],xmm0;
+ movaps [ecx],xmm0
+ movaps [ecx+0x10],xmm0
+ movaps [ecx+0x20],xmm0
+ movaps [ecx+0x30],xmm0
+ movaps [ecx+0x40],xmm0
+ movaps [ecx+0x50],xmm0
+ movaps [ecx+0x60],xmm0
+ movaps [ecx+0x70],xmm0
sub ecx,-128
- dec eax;
+ sub eax,1
jnz _loop_8;
}
if( remainder != 0 )
@@ -173,18 +165,15 @@ static __forceinline void memzero_ptr( void *dest )
mov eax, remainder
_loop_10:
- movaps [ecx+eax],xmm0;
+ movaps [ecx+eax],xmm0
sub eax,16;
jnz _loop_10;
}
}
- __asm
- {
- movups xmm0,[_xmm_backup];
- }
return;
}
}
+ #endif
// This function only works on 32-bit alignments.
pxAssume( (MZFbytes & 0x3) == 0 );
@@ -271,8 +260,6 @@ static __forceinline void memset_8( void *dest )
return;
}
- //u64 _xmm_backup[2];
-
/*static const size_t remainder = MZFbytes & 127;
static const size_t bytes128 = MZFbytes / 128;
if( bytes128 > 32 )
@@ -283,7 +270,6 @@ static __forceinline void memset_8( void *dest )
__asm
{
- movups _xmm_backup,xmm0;
mov eax,bytes128
mov ecx,dest
movss xmm0,data
@@ -316,10 +302,6 @@ static __forceinline void memset_8( void *dest )
jnz _loop_10;
}
}
- __asm
- {
- movups xmm0,[_xmm_backup];
- }
}*/
// This function only works on 32-bit alignments of data copied.
diff --git a/pcsx2/ps2/GIFpath.cpp b/pcsx2/ps2/GIFpath.cpp
index 02b6551e4f..b311361c84 100644
--- a/pcsx2/ps2/GIFpath.cpp
+++ b/pcsx2/ps2/GIFpath.cpp
@@ -252,7 +252,7 @@ __forceinline void GIFPath::Reset()
__forceinline bool GIFPath::StepReg()
{
- if ((++curreg & 0xf) == tag.NREG) {
+ if (++curreg >= numregs) {
curreg = 0;
if (--nloop == 0) {
return false;