From 84f20a1cadff41b0500dc0a0bfab2549aaa2b7e7 Mon Sep 17 00:00:00 2001
From: Henrik Rydgard <hrydgard@gmail.com>
Date: Tue, 12 Nov 2013 00:18:49 +0100
Subject: [PATCH] Small optimizations

---
 Core/MIPS/x86/CompVFPU.cpp | 22 ++++++++++++++++++++--
 Core/MemMap.h              | 17 +++++++++++++++++
 Core/MemMapFunctions.cpp   | 17 -----------------
 GPU/GLES/GLES_GPU.cpp      |  7 +++++--
 4 files changed, 42 insertions(+), 21 deletions(-)

diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp
index 4b26b5b47d..bf3196861c 100644
--- a/Core/MIPS/x86/CompVFPU.cpp
+++ b/Core/MIPS/x86/CompVFPU.cpp
@@ -64,6 +64,7 @@ static const float zero = 0.0f;
 const u32 MEMORY_ALIGNED16( noSignMask[4] ) = {0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF};
 const u32 MEMORY_ALIGNED16( signBitLower[4] ) = {0x80000000, 0, 0, 0};
 const float MEMORY_ALIGNED16( oneOneOneOne[4] ) = {1.0f, 1.0f, 1.0f, 1.0f};
+const u32 MEMORY_ALIGNED16( solidOnes[4] ) = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF};
 
 void Jit::Comp_VPFX(MIPSOpcode op)
 {
@@ -881,13 +882,14 @@ void Jit::Comp_Vcmp(MIPSOpcode op) {
 
 	// Some, we just fall back to the interpreter.
 	switch (cond) {
-	case VC_EN: // c = my_isnan(s[i]); break;
 	case VC_EI: // c = my_isinf(s[i]); break;
 	case VC_ES: // c = my_isnan(s[i]) || my_isinf(s[i]); break;   // Tekken Dark Resurrection
-	case VC_NN: // c = !my_isnan(s[i]); break;
 	case VC_NI: // c = !my_isinf(s[i]); break;
 	case VC_NS: // c = !my_isnan(s[i]) && !my_isinf(s[i]); break;
 		DISABLE;
+		break;
+	case VC_EN: // c = my_isnan(s[i]); break;
+	case VC_NN: // c = !my_isnan(s[i]); break;
 	default:
 		break;
 	}
@@ -916,7 +918,20 @@ void Jit::Comp_Vcmp(MIPSOpcode op) {
 		bool compareToZero = false;
 		int comparison = -1;
 		bool flip = false;
+		bool inverse = false;
+
 		switch (cond) {
+		case VC_EN:
+			comparison = CMP_UNORD;
+			compareTwo = true;
+			break;
+
+		case VC_NN:
+			comparison = CMP_UNORD;
+			compareTwo = true;
+			inverse = true;
+			break;
+
 		case VC_EQ: // c = s[i] == t[i]; break;
 			comparison = CMP_EQ;
 			compareTwo = true;
@@ -975,6 +990,9 @@ void Jit::Comp_Vcmp(MIPSOpcode op) {
 			MOVSS(XMM1, fpr.V(sregs[i]));
 			CMPSS(XMM1, R(XMM0), comparison);
 		}
+		if (inverse) {
+			XORPS(XMM1, M((void *)&solidOnes));
+		}
 
 		MOVSS(M((void *) &ssCompareTemp), XMM1);
 		if (i == 0 && n == 1) {
diff --git a/Core/MemMap.h b/Core/MemMap.h
index 3fd45ddf54..7cd098555b 100644
--- a/Core/MemMap.h
+++ b/Core/MemMap.h
@@ -274,6 +274,23 @@ inline void MemcpyUnchecked(void *to_data, const u32 from_address, const u32 len
 	memcpy(to_data, GetPointerUnchecked(from_address), len);
 }
 
+inline bool IsValidAddress(const u32 address) {
+	if ((address & 0x3E000000) == 0x08000000) {
+		return true;
+	}
+	else if ((address & 0x3F800000) == 0x04000000) {
+		return true;
+	}
+	else if ((address & 0xBFFF0000) == 0x00010000) {
+		return true;
+	}
+	else if ((address & 0x3F000000) >= 0x08000000 && (address & 0x3F000000) < 0x08000000 + g_MemorySize) {
+		return true;
+	}
+	else
+		return false;
+}
+
 
 template<class T>
 void ReadStruct(u32 address, T *ptr)
diff --git a/Core/MemMapFunctions.cpp b/Core/MemMapFunctions.cpp
index c8f1ec7089..3938b918cf 100644
--- a/Core/MemMapFunctions.cpp
+++ b/Core/MemMapFunctions.cpp
@@ -145,23 +145,6 @@ inline void WriteToHardware(u32 address, const T data)
 
 // =====================
 
-bool IsValidAddress(const u32 address) {
-	if ((address & 0x3E000000) == 0x08000000) {
-		return true;
-	}
-	else if ((address & 0x3F800000) == 0x04000000) {
-		return true;
-	}
-	else if ((address & 0xBFFF0000) == 0x00010000) {
-		return true;
-	}
-	else if ((address & 0x3F000000) >= 0x08000000 && (address & 0x3F000000) < 0x08000000 + g_MemorySize) {
-		return true;
-	}
-	else
-		return false;
-}
-
 bool IsRAMAddress(const u32 address) {
 	if ((address & 0x3E000000) == 0x08000000) {
 		return true;
diff --git a/GPU/GLES/GLES_GPU.cpp b/GPU/GLES/GLES_GPU.cpp
index 84ca8c4368..47efbcc2c0 100644
--- a/GPU/GLES/GLES_GPU.cpp
+++ b/GPU/GLES/GLES_GPU.cpp
@@ -994,8 +994,6 @@ void GLES_GPU::ExecuteOpInternal(u32 op, u32 diff) {
 		}
 		break;
 
-	case GE_CMD_CLUTADDR:
-	case GE_CMD_CLUTADDRUPPER:
 	case GE_CMD_CLUTFORMAT:
 		if (diff) {
 			gstate_c.textureChanged = true;
@@ -1003,6 +1001,11 @@ void GLES_GPU::ExecuteOpInternal(u32 op, u32 diff) {
 		// This could be used to "dirty" textures with clut.
 		break;
 
+	case GE_CMD_CLUTADDR:
+	case GE_CMD_CLUTADDRUPPER:
+		// Hm, LOADCLUT actually changes the CLUT so no need to dirty here.
+		break;
+
 	case GE_CMD_LOADCLUT:
 		gstate_c.textureChanged = true;
 		textureCache_.LoadClut();