From 84f844767c2ec46eacab261cc7c0aa26540d7a34 Mon Sep 17 00:00:00 2001
From: Gregory Hainaut <gregory.hainaut@gmail.com>
Date: Tue, 7 Oct 2014 19:11:43 +0200
Subject: [PATCH] gsdx-ogl: micro optimize PSConstantBuffer cache

Might help to save a cache line on the CPU :)
---
 plugins/GSdx/GSDeviceOGL.h      | 11 ++++++-----
 plugins/GSdx/GSTextureCache.cpp | 24 ++++++++++++------------
 plugins/GSdx/res/glsl_source.h  |  4 ++--
 plugins/GSdx/res/tfx.glsl       |  4 ++--
 4 files changed, 22 insertions(+), 21 deletions(-)

diff --git a/plugins/GSdx/GSDeviceOGL.h b/plugins/GSdx/GSDeviceOGL.h
index 68acabc5e..20fcdbb58 100644
--- a/plugins/GSdx/GSDeviceOGL.h
+++ b/plugins/GSdx/GSDeviceOGL.h
@@ -294,12 +294,12 @@ class GSDeviceOGL : public GSDevice
 	__aligned(struct, 32) PSConstantBuffer
 	{
 		GSVector4 FogColor_AREF;
-		GSVector4 HalfTexel;
 		GSVector4 WH;
-		GSVector4 MinMax;
 		GSVector4 MinF_TA;
 		GSVector4i MskFix;
 
+		GSVector4 HalfTexel;
+		GSVector4 MinMax;
 		GSVector4 TC_OffsetHack;
 
 		PSConstantBuffer()
@@ -319,12 +319,13 @@ class GSDeviceOGL : public GSDevice
 
 			// if WH matches both HalfTexel and TC_OffsetHack do too
 			// MinMax depends on WH and MskFix so no need to check it too
-			if(!((a[0] == b[0]) & (a[2] == b[2]) & (a[4] == b[4]) & (a[5] == b[5])).alltrue())
+			if(!((a[0] == b[0]) & (a[1] == b[1]) & (a[2] == b[2]) & (a[3] == b[3])).alltrue())
 			{
+				// Note previous check uses SSE already, a plain copy will be faster than any memcpy
 				a[0] = b[0];
+				a[1] = b[1];
 				a[2] = b[2];
-				a[4] = b[4];
-				a[5] = b[5];
+				a[3] = b[3];
 
 				return true;
 			}
diff --git a/plugins/GSdx/GSTextureCache.cpp b/plugins/GSdx/GSTextureCache.cpp
index b4607af7b..635c12e70 100644
--- a/plugins/GSdx/GSTextureCache.cpp
+++ b/plugins/GSdx/GSTextureCache.cpp
@@ -26,7 +26,7 @@ GSTextureCache::GSTextureCache(GSRenderer* r)
 	: m_renderer(r)
 {
 	m_spritehack = !!theApp.GetConfig("UserHacks", 0) ? theApp.GetConfig("UserHacks_SpriteHack", 0) : 0;
-	
+
 	UserHacks_HalfPixelOffset = !!theApp.GetConfig("UserHacks", 0) && !!theApp.GetConfig("UserHacks_HalfPixelOffset", 0);
 	UserHacks_NVIDIAHack = !!theApp.GetConfig("UserHacks_NVIDIAHack", 0) && !!theApp.GetConfig("UserHacks", 0);
 	m_paltex = !!theApp.GetConfig("paltex", 0);
@@ -105,10 +105,10 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, con
 
 		// Arc the Lad finds the wrong surface here when looking for a depth stencil.
 		// Since we're currently not caching depth stencils (check ToDo in CreateSource) we should not look for it here.
-		
+
 		// (Simply not doing this code at all makes a lot of previsouly missing stuff show (but breaks pretty much everything
 		// else.)
-		
+
 		//for(int type = 0; type < 2 && dst == NULL; type++)
 		for(int type = 0; type < 1 && dst == NULL; type++) // Only look for render target, no depth stencil
 		{
@@ -320,7 +320,7 @@ void GSTextureCache::InvalidateVideoMem(GSOffset* o, const GSVector4i& rect, boo
 	GSVector4i r;
 
 	uint32* pages = (uint32*)m_temp;
-	
+
 	o->GetPages(rect, pages, &r);
 
 	bool found = false;
@@ -349,7 +349,7 @@ void GSTextureCache::InvalidateVideoMem(GSOffset* o, const GSVector4i& rect, boo
 					if(s->m_repeating)
 					{
 						vector<GSVector2i>& l = s->m_p2t[page];
-						
+
 						for(vector<GSVector2i>::iterator k = l.begin(); k != l.end(); k++)
 						{
 							valid[k->x] &= k->y;
@@ -468,7 +468,7 @@ void GSTextureCache::InvalidateLocalMem(GSOffset* o, const GSVector4i& r)
 			}
 		}
 	}
-	
+
 	//GSTextureCache::Target* rt2 = NULL;
 	//int ymin = INT_MAX;
 	//for(list<Target*>::iterator i = m_dst[RenderTarget].begin(); i != m_dst[RenderTarget].end(); )
@@ -535,7 +535,7 @@ void GSTextureCache::InvalidateLocalMem(GSOffset* o, const GSVector4i& r)
 	//	Read(rt2, GSVector4i(r.left, r.top + ymin, r.right, r.bottom + ymin));
 	//}
 
-	
+
 	// TODO: ds
 }
 
@@ -594,13 +594,13 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 
 	bool hack = false;
 
-	if(m_spritehack && (TEX0.PSM == PSM_PSMT8 || TEX0.PSM == PSM_PSMT8H)) 
+	if(m_spritehack && (TEX0.PSM == PSM_PSMT8 || TEX0.PSM == PSM_PSMT8H))
 	{
 		src->m_spritehack_t = true;
-		
-		if(m_spritehack == 2 && TEX0.CPSM != PSM_PSMCT16) 
-			src->m_spritehack_t = false;		
-	}			
+
+		if(m_spritehack == 2 && TEX0.CPSM != PSM_PSMCT16)
+			src->m_spritehack_t = false;
+	}
 	else
 		src->m_spritehack_t = false;
 
diff --git a/plugins/GSdx/res/glsl_source.h b/plugins/GSdx/res/glsl_source.h
index 907a88468..f5b682de1 100644
--- a/plugins/GSdx/res/glsl_source.h
+++ b/plugins/GSdx/res/glsl_source.h
@@ -787,12 +787,12 @@ static const char* tfx_glsl =
 	"{\n"
 	"    vec3 FogColor;\n"
 	"    float AREF;\n"
-	"    vec4 HalfTexel;\n"
 	"    vec4 WH;\n"
-	"    vec4 MinMax;\n"
 	"    vec2 MinF;\n"
 	"    vec2 TA;\n"
 	"    uvec4 MskFix;\n"
+	"    vec4 HalfTexel;\n"
+	"    vec4 MinMax;\n"
 	"    vec4 TC_OffsetHack;\n"
 	"};\n"
 	"\n"
diff --git a/plugins/GSdx/res/tfx.glsl b/plugins/GSdx/res/tfx.glsl
index acf436b15..c4f3536f4 100644
--- a/plugins/GSdx/res/tfx.glsl
+++ b/plugins/GSdx/res/tfx.glsl
@@ -336,12 +336,12 @@ layout(std140, binding = 21) uniform cb21
 {
     vec3 FogColor;
     float AREF;
-    vec4 HalfTexel;
     vec4 WH;
-    vec4 MinMax;
     vec2 MinF;
     vec2 TA;
     uvec4 MskFix;
+    vec4 HalfTexel;
+    vec4 MinMax;
     vec4 TC_OffsetHack;
 };