From 8a089ee583f4d1d5b435bd2b59dd29a49a47d0bd Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sat, 3 Jan 2015 10:39:26 -0800
Subject: [PATCH 1/3] x86jit: Handle -0.0 in vavg/vfad.

Should result in a positive result.
---
 Core/MIPS/x86/CompVFPU.cpp | 23 +++++++++++++++--------
 1 file changed, 15 insertions(+), 8 deletions(-)

diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp
index 9d64ab12f6..5b252cd710 100644
--- a/Core/MIPS/x86/CompVFPU.cpp
+++ b/Core/MIPS/x86/CompVFPU.cpp
@@ -3123,8 +3123,12 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
 				MOVAPS(fpr.VSX(dregs), R(XMM0));
 				break;
 			case V_Quad:
+				XORPS(XMM1, R(XMM1));
 				MOVAPS(XMM0, fpr.VS(sregs));
 				DPPS(XMM0, M(&oneOneOneOne), 0xF1);
+				// In every other case, +0.0 is selected by the mask and added.
+				// But, here we need to manually add it to the result.
+				ADDPS(XMM0, R(XMM1));
 				MOVAPS(fpr.VSX(dregs), R(XMM0));
 				break;
 			default:
@@ -3133,15 +3137,17 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
 		} else {
 			switch (sz) {
 			case V_Pair:
+				XORPS(XMM1, R(XMM1));
 				MOVAPS(XMM0, fpr.VS(sregs));
-				MOVAPS(XMM1, R(XMM0));
+				ADDPS(XMM1, R(XMM0));
 				SHUFPS(XMM1, R(XMM1), _MM_SHUFFLE(3, 2, 1, 1));
 				ADDPS(XMM0, R(XMM1));
 				MOVAPS(fpr.VSX(dregs), R(XMM0));
 				break;
 			case V_Triple:
+				XORPS(XMM1, R(XMM1));
 				MOVAPS(XMM0, fpr.VS(sregs));
-				MOVAPS(XMM1, R(XMM0));
+				ADDPS(XMM1, R(XMM0));
 				SHUFPS(XMM1, R(XMM1), _MM_SHUFFLE(3, 2, 1, 1));
 				ADDPS(XMM0, R(XMM1));
 				SHUFPS(XMM1, R(XMM1), _MM_SHUFFLE(3, 2, 1, 2));
@@ -3149,7 +3155,10 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
 				MOVAPS(fpr.VSX(dregs), R(XMM0));
 				break;
 			case V_Quad:
+				XORPS(XMM1, R(XMM1));
 				MOVAPS(XMM0, fpr.VS(sregs));
+				// This flips the sign of any -0.000.
+				ADDPS(XMM0, R(XMM1));
 				MOVHLPS(XMM1, XMM0);
 				ADDPS(XMM0, R(XMM1));
 				MOVAPS(XMM1, R(XMM0));
@@ -3174,17 +3183,15 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
 	fpr.SimpleRegsV(dregs, V_Single, MAP_NOINIT | MAP_DIRTY);
 
 	X64Reg reg = XMM0;
-	if (IsOverlapSafeAllowS(dregs[0], 0, n, sregs)) {
+	if (IsOverlapSafe(dregs[0], 0, n, sregs)) {
 		fpr.MapRegV(dregs[0], dregs[0] == sregs[0] ? MAP_DIRTY : MAP_NOINIT);
 		fpr.SpillLockV(dregs[0]);
 		reg = fpr.VX(dregs[0]);
 	}
 
-	// We use a temp reg in case of overlap.
-	if (!fpr.V(sregs[0]).IsSimpleReg(reg)) {
-		MOVSS(reg, fpr.V(sregs[0]));
-	}
-	for (int i = 1; i < n; ++i) {
+	// We have to start zt +0.000 in case any values are -0.000.
+	XORPS(reg, R(reg));
+	for (int i = 0; i < n; ++i) {
 		ADDSS(reg, fpr.V(sregs[i]));
 	}
 

From a1864b2c789c5fa16884bcf645b69299cb4b62a0 Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sat, 3 Jan 2015 10:39:58 -0800
Subject: [PATCH 2/3] x86jit: Fix vavg scale.

Was completely wrong, darn it.
---
 Core/MIPS/x86/CompVFPU.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp
index 5b252cd710..eaa4bf572b 100644
--- a/Core/MIPS/x86/CompVFPU.cpp
+++ b/Core/MIPS/x86/CompVFPU.cpp
@@ -3171,7 +3171,7 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
 			}
 		}
 		if (((op >> 16) & 31) == 7) { // vavg
-			MULSS(fpr.VSX(dregs), M(&vavg_table[n]));
+			MULSS(fpr.VSX(dregs), M(&vavg_table[n - 1]));
 		}
 		ApplyPrefixD(dregs, V_Single);
 		fpr.ReleaseSpillLocks();
@@ -3199,7 +3199,7 @@ void Jit::Comp_Vhoriz(MIPSOpcode op) {
 	case 6:  // vfad
 		break;
 	case 7:  // vavg
-		MULSS(reg, M(&vavg_table[n]));
+		MULSS(reg, M(&vavg_table[n - 1]));
 		break;
 	}
 

From e6a50a13717394b07c2391c33bd3c878519093d8 Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sat, 3 Jan 2015 10:40:21 -0800
Subject: [PATCH 3/3] Initialize vfpu regs are NAN, not 0.0.

This matches tests, the fpu, and seems logical.
---
 Core/HLE/sceKernelThread.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Core/HLE/sceKernelThread.cpp b/Core/HLE/sceKernelThread.cpp
index 6a0fd56547..197a4559b6 100644
--- a/Core/HLE/sceKernelThread.cpp
+++ b/Core/HLE/sceKernelThread.cpp
@@ -1982,7 +1982,7 @@ void ThreadContext::reset()
 	r[0] = 0;
 	for (int i = 0; i<128; i++)
 	{
-		v[i] = 0.0f;
+		vi[i] = 0x7f800001;
 	}
 	for (int i = 0; i<15; i++)
 	{