diff options
Diffstat (limited to 'src/thirdparty/VirtualDub/system/source/a_memory.asm')
| -rw-r--r-- | src/thirdparty/VirtualDub/system/source/a_memory.asm | 114 |
1 files changed, 60 insertions, 54 deletions
diff --git a/src/thirdparty/VirtualDub/system/source/a_memory.asm b/src/thirdparty/VirtualDub/system/source/a_memory.asm index 42a8ca998..d2316e55a 100644 --- a/src/thirdparty/VirtualDub/system/source/a_memory.asm +++ b/src/thirdparty/VirtualDub/system/source/a_memory.asm @@ -71,6 +71,65 @@ _VDFastMemcpyPartialMMX: pop edi
ret
+ global _VDFastMemcpyPartialSSE2
+_VDFastMemcpyPartialSSE2:
+ push ebp
+ push edi
+ push esi
+ push ebx
+
+ mov ecx, [esp+12+16]
+ cmp ecx, 128
+ jb _VDFastMemcpyPartialMMX2.MMX2
+ mov edi, [esp+4+16]
+ mov esi, [esp+8+16]
+ mov eax, edi
+ or eax, esi
+ test al, 15
+ jne SHORT _VDFastMemcpyPartialMMX2.MMX2
+
+ shr ecx, 7
+.loop128:
+ prefetchnta [esi+16*8]
+ movaps xmm0, [esi]
+ movaps xmm1, [esi+16*1]
+ movaps xmm2, [esi+16*2]
+ movaps xmm3, [esi+16*3]
+ movaps xmm4, [esi+16*4]
+ movaps xmm5, [esi+16*5]
+ movaps xmm6, [esi+16*6]
+ movaps xmm7, [esi+16*7]
+ movntps [edi], xmm0
+ movntps [edi+16*1], xmm1
+ movntps [edi+16*2], xmm2
+ movntps [edi+16*3], xmm3
+ movntps [edi+16*4], xmm4
+ movntps [edi+16*5], xmm5
+ movntps [edi+16*6], xmm6
+ movntps [edi+16*7], xmm7
+ add esi, 128
+ add edi, 128
+ dec ecx
+ jne .loop128
+.skiploop128:
+ mov ecx, [esp+12+16]
+ and ecx, 127
+ cmp ecx, 0
+ je .nooddballs
+.loop:
+ mov dl, [esi]
+ mov [edi], dl
+ inc esi
+ inc edi
+ dec ecx
+ jne .loop
+.nooddballs:
+ pop ebx
+ pop esi
+ pop edi
+ pop ebp
+ ret
+
global _VDFastMemcpyPartialMMX2
_VDFastMemcpyPartialMMX2:
push ebp
@@ -78,6 +137,7 @@ _VDFastMemcpyPartialMMX2: push esi
push ebx
+.MMX2
mov ebx, [esp+4+16]
mov edx, [esp+8+16]
mov eax, [esp+12+16]
@@ -130,60 +190,6 @@ _VDFastMemcpyPartialMMX2: pop ebp
ret
- global _VDFastMemcpyPartialSSE2
-_VDFastMemcpyPartialSSE2:
- push ebp
- push edi
- push esi
- push ebx
-
- mov edi, [esp+4+16]
- mov esi, [esp+8+16]
- mov ecx, [esp+12+16]
- shr ecx, 7
- cmp ecx, 0
- jbe .skipblastloop
-.blastloop:
- prefetchnta [esi+16*8]
- movaps xmm0, [esi]
- movaps xmm1, [esi+16*1]
- movaps xmm2, [esi+16*2]
- movaps xmm3, [esi+16*3]
- movaps xmm4, [esi+16*4]
- movaps xmm5, [esi+16*5]
- movaps xmm6, [esi+16*6]
- movaps xmm7, [esi+16*7]
- movntps [edi], xmm0
- movntps [edi+16*1], xmm1
- movntps [edi+16*2], xmm2
- movntps [edi+16*3], xmm3
- movntps [edi+16*4], xmm4
- movntps [edi+16*5], xmm5
- movntps [edi+16*6], xmm6
- movntps [edi+16*7], xmm7
- add esi, 128
- add edi, 128
- dec ecx
- jne .blastloop
-.skipblastloop:
- mov ecx, [esp+12+16]
- and ecx, 127
- cmp ecx, 0
- je .nooddballs
-.loop:
- mov dl, [esi]
- mov [edi], dl
- inc esi
- inc edi
- dec ecx
- jne .loop
-.nooddballs:
- pop ebx
- pop esi
- pop edi
- pop ebp
- ret
-
end
|