From b1389b5bf79340e1f348eb387c3ff6d2dd390644 Mon Sep 17 00:00:00 2001
From: kinddragon <kinddragon@users.sourceforge.net>
Date: Fri, 28 May 2010 01:23:17 +0000
Subject: Minor code cleanup Added SSE2 VD memcpy implementation

git-svn-id: https://mpc-hc.svn.sourceforge.net/svnroot/mpc-hc/trunk@1965 10f7b99b-c216-0410-bff0-8a66a9350fd8
---
 .../VirtualDub/system/source/a_memory.asm          | 54 ++++++++++++++++++++++
 src/thirdparty/VirtualDub/system/source/memory.cpp | 20 +++++++-
 2 files changed, 73 insertions(+), 1 deletion(-)

(limited to 'src/thirdparty')

diff --git a/src/thirdparty/VirtualDub/system/source/a_memory.asm b/src/thirdparty/VirtualDub/system/source/a_memory.asm
index e4b6cac8b..42a8ca998 100644
--- a/src/thirdparty/VirtualDub/system/source/a_memory.asm
+++ b/src/thirdparty/VirtualDub/system/source/a_memory.asm
@@ -130,6 +130,60 @@ _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
 
diff --git a/src/thirdparty/VirtualDub/system/source/memory.cpp b/src/thirdparty/VirtualDub/system/source/memory.cpp
index 3e03b5d34..6edce7873 100644
--- a/src/thirdparty/VirtualDub/system/source/memory.cpp
+++ b/src/thirdparty/VirtualDub/system/source/memory.cpp
@@ -365,6 +365,7 @@ void VDMemset32Rect(void *dst, ptrdiff_t pitch, uint32 value, size_t w, size_t h
 	extern "C" void __cdecl VDFastMemcpyPartialScalarAligned8(void *dst, const void *src, size_t bytes);
 	extern "C" void __cdecl VDFastMemcpyPartialMMX(void *dst, const void *src, size_t bytes);
 	extern "C" void __cdecl VDFastMemcpyPartialMMX2(void *dst, const void *src, size_t bytes);
+	extern "C" void __cdecl VDFastMemcpyPartialSSE2(void *dst, const void *src, size_t bytes);
 
 	void VDFastMemcpyPartialScalar(void *dst, const void *src, size_t bytes) {
 		if (!(((int)dst | (int)src | bytes) & 7))
@@ -391,7 +392,13 @@ void VDMemset32Rect(void *dst, ptrdiff_t pitch, uint32 value, size_t w, size_t h
 	void VDFastMemcpyAutodetect() {
 		long exts = CPUGetEnabledExtensions();
 
-		if (exts & CPUF_SUPPORTS_SSE) {
+		// MPC custom code (begin)
+		if (exts & CPUF_SUPPORTS_SSE2) {
+			VDFastMemcpyPartial = VDFastMemcpyPartialSSE2;
+			VDFastMemcpyFinish	= VDFastMemcpyFinishMMX2;
+			VDSwapMemory		= VDSwapMemorySSE;
+		// MPC custom code (end)
+		} else if (exts & CPUF_SUPPORTS_SSE) {
 			VDFastMemcpyPartial = VDFastMemcpyPartialMMX2;
 			VDFastMemcpyFinish	= VDFastMemcpyFinishMMX2;
 			VDSwapMemory		= VDSwapMemorySSE;
@@ -426,6 +433,17 @@ void VDMemcpyRect(void *dst, ptrdiff_t dststride, const void *src, ptrdiff_t src
 	if (w <= 0 || h <= 0)
 		return;
 
+	// MPC custom code (begin)
+#ifdef _DEBUG
+	if (CPUGetEnabledExtensions() & CPUF_SUPPORTS_SSE2) {
+		_ASSERT(!(((UINT_PTR)dst | (UINT_PTR)src) & 0xF));
+		if (h > 1)	{
+			_ASSERT(!((dststride | srcstride) & 0xF));
+		}
+	}
+#endif
+	// MPC custom code (end)
+
 	if (w == srcstride && w == dststride)
 		VDFastMemcpyPartial(dst, src, w*h);
 	// MPC custom code (begin)
-- 
cgit v1.2.3