gsdx-ogl: only compile useful VS

logz is a constant
wildhack is only compatbile with TME/FST

Compilation goes down from 64 to 20 vertex shaders.

plugins/GSdx/GSDeviceOGL.cpp

@@ -653,10 +653,10 @@ void GSDeviceOGL::Barrier(GLbitfield b)
 }
 
 /* Note: must be here because tfx_glsl is static */
-GLuint GSDeviceOGL::CompileVS(VSSelector sel)
+GLuint GSDeviceOGL::CompileVS(VSSelector sel, int logz)
 {
 	std::string macro = format("#define VS_BPPZ %d\n", sel.bppz)
-		+ format("#define VS_LOGZ %d\n", sel.logz)
+		+ format("#define VS_LOGZ %d\n", logz)
 		+ format("#define VS_TME %d\n", sel.tme)
 		+ format("#define VS_FST %d\n", sel.fst)
 		+ format("#define VS_WILDHACK %d\n", sel.wildhack)

plugins/GSdx/GSDeviceOGL.h

@@ -275,7 +275,6 @@ class GSDeviceOGL : public GSDevice
 			{
 				uint32 wildhack:1;
 				uint32 bppz:2;
-				uint32 logz:1;
 				// Next param will be handle by subroutine
 				uint32 tme:1;
 				uint32 fst:1;
@@ -284,12 +283,12 @@ class GSDeviceOGL : public GSDevice
 			uint32 key;
 		};
 
-		operator uint32() {return key & 0x7f;}
+		operator uint32() {return key & 0x3f;}
 
 		VSSelector() : key(0) {}
 		VSSelector(uint32 k) : key(k) {}
 
-		static uint32 size() { return 1 << 6; }
+		static uint32 size() { return 1 << 5; }
 	};
 
 	__aligned(struct, 32) PSConstantBuffer
@@ -612,7 +611,7 @@ class GSDeviceOGL : public GSDevice
 	void OMSetWriteBuffer(GLenum buffer = GL_COLOR_ATTACHMENT0);
 
 	void CreateTextureFX();
-	GLuint CompileVS(VSSelector sel);
+	GLuint CompileVS(VSSelector sel, int logz);
 	GLuint CompileGS();
 	GLuint CompilePS(PSSelector sel);
 	GLuint CreateSampler(bool bilinear, bool tau, bool tav);

plugins/GSdx/GSRendererOGL.cpp

@@ -27,7 +27,6 @@
 GSRendererOGL::GSRendererOGL()
 	: GSRendererHW(new GSTextureCacheOGL(this))
 {
-	m_logz = !!theApp.GetConfig("logz", 1);
 	m_fba = !!theApp.GetConfig("fba", 1);
 	UserHacks_AlphaHack = !!theApp.GetConfig("UserHacks_AlphaHack", 0) && !!theApp.GetConfig("UserHacks", 0);
 	UserHacks_AlphaStencil = !!theApp.GetConfig("UserHacks_AlphaStencil", 0) && !!theApp.GetConfig("UserHacks", 0);
@@ -262,7 +261,6 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 
 	vs_sel.tme = PRIM->TME;
 	vs_sel.fst = PRIM->FST;
-	vs_sel.logz = m_logz ? 1 : 0;
 	vs_sel.wildhack = (UserHacks_WildHack && !isPackedUV_HackFlag) ? 1 : 0;
 
 	// The real GS appears to do no masking based on the Z buffer format and writing larger Z values

plugins/GSdx/GSRendererOGL.h

@@ -35,7 +35,6 @@ class GSRendererOGL : public GSRendererHW
 {
 	private:
 		GSVector2 m_pixelcenter;
-		bool m_logz;
 		bool m_fba;
 		bool UserHacks_AlphaHack;
 		bool UserHacks_AlphaStencil;

plugins/GSdx/GSShaderOGL.cpp

@@ -263,7 +263,7 @@ void GSShaderOGL::UseProgram()
 			hash_map<uint64, GLuint >::iterator it;
 			// Note: shader are integer lookup pointer. They start from 1 and incr
 			// every time you create a new shader OR a new program.
-			// Note2: vs & gs are precompiled at startup. FGLRX and radeon got value < 128.
+			// Note2: vs & gs are precompiled at startup. FGLRX and radeon got value < 128. GS has only 2 programs
 			// We migth be able to pack the value in a 32bits int
 			// I would need to check the behavior on Nvidia (pause/resume).
 			uint64 sel = (uint64)GLState::vs << 40 | (uint64)GLState::gs << 20 | GLState::ps;

plugins/GSdx/GSTextureFXOGL.cpp

@@ -39,8 +39,15 @@ void GSDeviceOGL::CreateTextureFX()
 	// It might cost a seconds at startup but it would reduce benchmark pollution
 	m_gs = CompileGS();
 
-	for (uint32 key = 0; key < VSSelector::size(); key++)
+	int logz = theApp.GetConfig("logz", 1);
-		m_vs[key] = CompileVS(VSSelector(key));
+	for (uint32 key = 0; key < VSSelector::size(); key++) {
+		// wildhack is only useful if both TME and FST are enabled.
+		VSSelector sel(key);
+		if (sel.wildhack && (!sel.tme || !sel.fst))
+			m_vs[key] = 0;
+		else
+			m_vs[key] = CompileVS(sel, logz);
+	}
 
 	for (uint32 key = 0; key < OMDepthStencilSelector::size(); key++)
 		m_om_dss[key] = CreateDepthStencil(OMDepthStencilSelector(key));