From fbb95f85e4ff7e2a221875d5adb59ce888655df7 Mon Sep 17 00:00:00 2001
From: cottonvibes <cottonvibes@96395faa-99c1-11dd-bbfe-3dabce05a288>
Date: Mon, 18 Jan 2010 04:07:57 +0000
Subject: [PATCH] newVif: fixed some bugs with mask/mode modes. i had forgotten
 that mVUmergeRegs() modifies the source reg's vectors, so data was being
 corrupted and breaking some games (sse4.1 users didn't have this problem).

This revision correctly fixes .hack GU Rebirth.
At this point we don't know any games newVif breaks compared to the old vif code. If you know any please leave a comment.

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@2447 96395faa-99c1-11dd-bbfe-3dabce05a288
---
 pcsx2/x86/VifUnpackSSE.cpp         | 12 ++++++++++++
 pcsx2/x86/VifUnpackSSE.h           |  2 ++
 pcsx2/x86/VifUnpackSSE_Dynarec.cpp | 15 ++++++++-------
 pcsx2/x86/newVif_Unpack.cpp        |  3 ++-
 4 files changed, 24 insertions(+), 8 deletions(-)

diff --git a/pcsx2/x86/VifUnpackSSE.cpp b/pcsx2/x86/VifUnpackSSE.cpp
index 1fa850fb1c..3117590278 100644
--- a/pcsx2/x86/VifUnpackSSE.cpp
+++ b/pcsx2/x86/VifUnpackSSE.cpp
@@ -26,6 +26,18 @@
 
 static __pagealigned u8 nVifUpkExec[__pagesize*4];
 
+// Merges xmm vectors without modifying source reg
+void mergeVectors(int dest, int src, int temp, int xyzw) {
+	if (x86caps.hasStreamingSIMD4Extensions  || (xyzw==15) 
+	|| (xyzw==12) || (xyzw==11) || (xyzw==8) || (xyzw==3)) {
+		mVUmergeRegs(dest, src, xyzw);
+	}
+	else {
+		SSE_MOVAPS_XMM_to_XMM(temp, src);
+		mVUmergeRegs(dest, temp, xyzw);
+	}
+}
+
 // =====================================================================================================
 //  VifUnpackSSE_Base Section
 // =====================================================================================================
diff --git a/pcsx2/x86/VifUnpackSSE.h b/pcsx2/x86/VifUnpackSSE.h
index 3a4214d8dd..a3f59a61f6 100644
--- a/pcsx2/x86/VifUnpackSSE.h
+++ b/pcsx2/x86/VifUnpackSSE.h
@@ -26,6 +26,8 @@ using namespace x86Emitter;
 
 #if newVif
 
+extern void mergeVectors(int dest, int src, int temp, int xyzw);
+
 // --------------------------------------------------------------------------------------
 //  VifUnpackSSE_Base
 // --------------------------------------------------------------------------------------
diff --git a/pcsx2/x86/VifUnpackSSE_Dynarec.cpp b/pcsx2/x86/VifUnpackSSE_Dynarec.cpp
index a1373be66e..8c7d29e808 100644
--- a/pcsx2/x86/VifUnpackSSE_Dynarec.cpp
+++ b/pcsx2/x86/VifUnpackSSE_Dynarec.cpp
@@ -111,28 +111,29 @@ _f void VifUnpackSSE_Dynarec::SetMasks(int cS) const {
 
 void VifUnpackSSE_Dynarec::doMaskWrite(const xRegisterSSE& regX) const {
 	pxAssumeDev(regX.Id <= 1, "Reg Overflow! XMM2 thru XMM6 are reserved for masking.");
+	int t  =  regX.Id ? 0 : 1; // Get Temp Reg
 	int cc =  aMin(vCL, 3);
 	u32 m0 = (vB.mask >> (cc * 8)) & 0xff;
-	u32 m1 =  m0 & 0xaaaa;
+	u32 m1 =  m0 & 0xaa;
 	u32 m2 =(~m1>>1) &  m0;
 	u32 m3 = (m1>>1) & ~m0;
 	u32 m4 = (m1>>1) &  m0;
 	makeMergeMask(m2);
 	makeMergeMask(m3);
 	makeMergeMask(m4);
-	if (doMask&&m4) { xMOVAPS(xmmTemp, ptr[dstIndirect]);		 } // Load Write Protect
-	if (doMask&&m2) { mVUmergeRegs(regX.Id, xmmRow.Id,		m2); } // Merge Row
-	if (doMask&&m3) { mVUmergeRegs(regX.Id, xmmCol0.Id+cc,	m3); } // Merge Col
-	if (doMask&&m4) { mVUmergeRegs(regX.Id, xmmTemp.Id,		m4); } // Merge Write Protect
+	if (doMask&&m4) { xMOVAPS(xmmTemp, ptr[dstIndirect]);			} // Load Write Protect
+	if (doMask&&m2) { mergeVectors(regX.Id, xmmRow.Id,		t, m2); } // Merge Row
+	if (doMask&&m3) { mergeVectors(regX.Id, xmmCol0.Id+cc,	t, m3); } // Merge Col
+	if (doMask&&m4) { mergeVectors(regX.Id, xmmTemp.Id,		t, m4); } // Merge Write Protect
 	if (doMode) {
 		u32 m5 = (~m1>>1) & ~m0;
 		if (!doMask)  m5 = 0xf;
 		else		  makeMergeMask(m5);
 		if (m5 < 0xf) {
 			xPXOR(xmmTemp, xmmTemp);
-			mVUmergeRegs(xmmTemp.Id, xmmRow.Id, m5);
+			mergeVectors(xmmTemp.Id, xmmRow.Id, t, m5);
 			xPADD.D(regX, xmmTemp);
-			if (doMode==2) mVUmergeRegs(xmmRow.Id, regX.Id, m5);
+			if (doMode==2) mergeVectors(xmmRow.Id, regX.Id, t, m5);
 		}
 		else if (m5 == 0xf) {
 			xPADD.D(regX, xmmRow);
diff --git a/pcsx2/x86/newVif_Unpack.cpp b/pcsx2/x86/newVif_Unpack.cpp
index 159f7158bb..f81d62a651 100644
--- a/pcsx2/x86/newVif_Unpack.cpp
+++ b/pcsx2/x86/newVif_Unpack.cpp
@@ -141,7 +141,8 @@ int nVifUnpack(int idx, u8* data) {
     }
 
 	if (ret == v.vif->tag.size) { // Full Transfer
-		dVifUnpack(idx, data, size, isFill);
+		if (newVifDynaRec)  dVifUnpack(idx, data, size, isFill);
+		else			   _nVifUnpack(idx, data, size, isFill);
 		vif->tag.size = 0;
 		vif->cmd = 0; 
 	}