gsdx ogl: the proof of concept commit

* GL_ARB_shader_subroutine for perf
fix for nvidia => add missing shader declaration. Nvidia got +4fps on colin3 :) 

For the moment only 2 PS parameters are supported. Code need to be extended to support others games that often
switch shader program (like xenosaga).
require GL4 class hardware and the option override_GL_ARB_shader_subroutine = 1
Note: strangely on AMD linux it is slower!

* GL_ARB_shader_image_load_store for accuraccy (Date)
Use a signed integer texture and reenable color buffer writing

Current status: Amagami_transparency.gs & P3_battle_shadows.gs are now working on Nvidia with a small perf impact.
Current implementation detail:
1/ setup the standard stencil as before
2/ on remaining pixel, draw once to compute first primitive that will write a fail alpha value.
3/ final draw based on primitive id of step 2

Note: I think we would get a bad behavior if depth test&mask are enabled on step 2/3
Note2: on my limited testcase the perf impact was on CPU. It would be possible to merge step1&2 to nullifying it (could 
even be faster actually), however it would require more GPU power.

Again require GL4 class hardware. And the option UserHacks_DateGL4 = 1



git-svn-id: http://pcsx2.googlecode.com/svn/trunk@5725 96395faa-99c1-11dd-bbfe-3dabce05a288

plugins/GSdx/GLLoader.cpp

@@ -132,6 +132,7 @@ namespace GLLoader {
 	bool found_only_gl30	= false; // Drop it when mesa support GLSL330
 	bool found_GL_ARB_clear_texture = false; // Don't know if GL3 GPU can support it
 	bool found_GL_ARB_buffer_storage = false;
+	bool found_GL_ARB_explicit_uniform_location = false; // need by subroutine
 	// GL4 hardware
 	bool found_GL_ARB_copy_image = false; // Not sure actually maybe GL3 GPU can do it
 	bool found_GL_ARB_gpu_shader5 = false;
@@ -233,6 +234,8 @@ namespace GLLoader {
 				if (ext.compare("GL_ARB_gpu_shader5") == 0) found_GL_ARB_gpu_shader5 = true;
 				if (ext.compare("GL_ARB_shader_image_load_store") == 0) found_GL_ARB_shader_image_load_store = true;
 #if 0
+				// Erratum: on nvidia implementation, gain is very nice : 42.5 fps => 46.5 fps
+				//
 				// Strangely it doesn't provide the speed boost as expected.
 				// Note: only atst/colclip was replaced with subroutine for the moment. It replace 2000 program switch on
 				// colin mcrae 3 by 2100 uniform, but code is slower!
@@ -243,6 +246,7 @@ namespace GLLoader {
 
 				if (ext.compare("GL_ARB_shader_subroutine") == 0) found_GL_ARB_shader_subroutine = true;
 #endif
+				if (ext.compare("GL_ARB_explicit_uniform_location") == 0) found_GL_ARB_explicit_uniform_location = true;
 #ifdef GL44 // Need to debug the code first
 				if (ext.compare("GL_ARB_clear_texture") == 0) found_GL_ARB_clear_texture = true;
 				if (ext.compare("GL_ARB_multi_bind") == 0) found_GL_ARB_multi_bind = true;
@@ -268,6 +272,7 @@ namespace GLLoader {
 		status &= status_and_override(found_GL_ARB_clear_texture,"GL_ARB_clear_texture");
 		status &= status_and_override(found_GL_ARB_buffer_storage,"GL_ARB_buffer_storage");
 		status &= status_and_override(found_GL_ARB_shader_subroutine,"GL_ARB_shader_subroutine");
+		status &= status_and_override(found_GL_ARB_explicit_uniform_location,"GL_ARB_explicit_uniform_location");
 
 		status &= status_and_override(found_GL_ARB_texture_storage, "GL_ARB_texture_storage", true);
 		status &= status_and_override(found_GL_ARB_shading_language_420pack,"GL_ARB_shading_language_420pack");

plugins/GSdx/GLLoader.h

@@ -279,4 +279,5 @@ namespace GLLoader {
 	extern bool found_GL_ARB_buffer_storage;
 	extern bool found_GL_ARB_shader_subroutine;
 	extern bool found_GL_ARB_bindless_texture;
+	extern bool found_GL_ARB_explicit_uniform_location;
 }

plugins/GSdx/GS.cpp

@@ -826,7 +826,8 @@ EXPORT_C GSgetTitleInfo2(char* dest, size_t length)
 {
 	string s = "GSdx";
 
-	if(s_gs != NULL) // TODO: this gets called from a different thread concurrently with GSOpen (on linux)
+	// TODO: this gets called from a different thread concurrently with GSOpen (on linux)
+	if(s_gs == NULL) return;
 
 	if(s_gs->m_GStitleInfoBuffer[0])
 	{

plugins/GSdx/GSDeviceOGL.cpp

@@ -108,6 +108,7 @@ GSDeviceOGL::~GSDeviceOGL()
 	delete m_ps_cb;
 	gl_DeleteSamplers(1, &m_palette_ss);
 	delete m_vb;
+	m_shader->Delete(m_apitrace);
 
 	for (uint32 key = 0; key < VSSelector::size(); key++) m_shader->Delete(m_vs[key]);
 	m_shader->Delete(m_gs);
@@ -296,10 +297,11 @@ bool GSDeviceOGL::Create(GSWnd* wnd)
 	m_date.dss->SetStencil(GL_ALWAYS, GL_REPLACE);
 
 	m_date.bs = new GSBlendStateOGL();
-#ifndef ENABLE_OGL_STENCIL_DEBUG
+	// FIXME impact image load?
-	// Only keep stencil data
+//#ifndef ENABLE_OGL_STENCIL_DEBUG
-	m_date.bs->SetMask(false, false, false, false);
+//	// Only keep stencil data
-#endif
+//	m_date.bs->SetMask(false, false, false, false);
+//#endif
 
 	// ****************************************************************
 	// HW renderer shader
@@ -538,9 +540,9 @@ void GSDeviceOGL::InitPrimDateTexture(int w, int h)
 {
 	// Create a texture to avoid the useless clean@0
 	if (m_date.t == NULL)
-		m_date.t = CreateTexture(w, h, GL_R32UI);
+		m_date.t = CreateTexture(w, h, GL_R32I);
 
-	ClearRenderTarget_ui(m_date.t, 0xFFFFFFFF);
+	ClearRenderTarget_ui(m_date.t, 0x0FFFFFFF);
 
 #ifdef ENABLE_OGL_STENCIL_DEBUG
 	gl_ActiveTexture(GL_TEXTURE0 + 5);
@@ -557,9 +559,9 @@ void GSDeviceOGL::BindDateTexture()
 	// TODO: multibind?
 	// GLuint textures[1] = {static_cast<GSTextureOGL*>(m_date.t)->GetID()};
 	// gl_BindImageTextures(0, 1, textures);
-	//gl_BindImageTexture(0, 0, 0, true, 0, GL_READ_WRITE, GL_R32UI);
+	//gl_BindImageTexture(0, 0, 0, true, 0, GL_READ_WRITE, GL_R32I);
 
-	gl_BindImageTexture(0, static_cast<GSTextureOGL*>(m_date.t)->GetID(), 0, false, 0, GL_READ_WRITE, GL_R32UI);
+	gl_BindImageTexture(0, static_cast<GSTextureOGL*>(m_date.t)->GetID(), 0, false, 0, GL_READ_WRITE, GL_R32I);
 }
 
 void GSDeviceOGL::RecycleDateTexture()

plugins/GSdx/GSDeviceOGL.h

@@ -509,6 +509,7 @@ class GSDeviceOGL : public GSDevice
 	GSDepthStencilOGL* m_om_dss[1<<6];
 	hash_map<uint32, GLuint > m_ps;
 	hash_map<uint32, GSBlendStateOGL* > m_om_bs;
+	GLuint m_apitrace;
 
 	GLuint m_palette_ss;
 	GLuint m_rt_ss;

plugins/GSdx/GSRendererOGL.cpp

@@ -268,10 +268,11 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 	om_bsel.wrgba = ~GSVector4i::load((int)context->FRAME.FBMSK).eq8(GSVector4i::xffffffff()).mask();
 
 	// TODO
-	//if (UserHacks_DateGL4 && DATE && om_bsel.wa && (!context->TEST.ATE || context->TEST.ATST == ATST_ALWAYS)) {
+	if (UserHacks_DateGL4 && DATE && om_bsel.wa && (!context->TEST.ATE || context->TEST.ATST == ATST_ALWAYS)) {
-	if (UserHacks_DateGL4 && DATE) {
 		//if (!(context->FBA.FBA && context->TEST.DATM == 1))
-		advance_DATE = true;
+
+		//advance_DATE = true;
+		advance_DATE = GLLoader::found_GL_ARB_shader_image_load_store;
 	}
 
 	// vs
@@ -500,8 +501,8 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 	dev->SetupCB(&vs_cb, &ps_cb);
 
 	if (advance_DATE) {
-		// Create an r32ui image that will contain primitive ID
+		// Create an r32i image that will contain primitive ID
-		// Note: do it at the beginning because the clean will dirty the state
+		// Note: do it at the beginning because the clean will dirty the FBO state
 		//dev->InitPrimDateTexture(rtsize.x, rtsize.y);
 
 		// Don't write anything on the color buffer

plugins/GSdx/GSShaderOGL.cpp

@@ -89,6 +89,7 @@ void GSShaderOGL::PS(GLuint s, GLuint sub_count)
 
 		GLState::ps = s;
 		GLState::dirty_prog = true;
+		GLState::dirty_subroutine_ps = true;
 #ifndef ENABLE_GLES
 		if (GLLoader::found_GL_ARB_separate_shader_objects) {
 			gl_UseProgramStages(m_pipeline, GL_FRAGMENT_SHADER_BIT, s);
@@ -278,10 +279,10 @@ GLuint GSShaderOGL::LinkNewProgram()
 void GSShaderOGL::UseProgram()
 {
 	if (GLState::dirty_prog) {
+		if (!GLLoader::found_GL_ARB_separate_shader_objects) {
 			GLState::dirty_subroutine_ps = true;
 			GLState::dirty_ressources = true;
 
-		if (!GLLoader::found_GL_ARB_separate_shader_objects) {
 			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.
@@ -340,19 +341,22 @@ std::string GSShaderOGL::GenGlslHeader(const std::string& entry, GLenum type, co
 	}
 	if (GLLoader::found_GL_ARB_separate_shader_objects) {
 		// Need GL version 410
-		header += "#extension GL_ARB_separate_shader_objects : require\n";
+		header += "#extension GL_ARB_separate_shader_objects: require\n";
 	} else {
 		header += "#define DISABLE_SSO\n";
 	}
 	if (GLLoader::found_only_gl30) {
 		// Need version 330
-		header += "#extension GL_ARB_explicit_attrib_location : require\n";
+		header += "#extension GL_ARB_explicit_attrib_location: require\n";
 		// Need version 140
-		header += "#extension GL_ARB_uniform_buffer_object : require\n";
+		header += "#extension GL_ARB_uniform_buffer_object: require\n";
 	}
-	if (GLLoader::found_GL_ARB_shader_subroutine) {
+	if (GLLoader::found_GL_ARB_shader_subroutine && GLLoader::found_GL_ARB_explicit_uniform_location) {
 		// Need GL version 400
 		header += "#define SUBROUTINE_GL40 1\n";
+		header += "#extension GL_ARB_shader_subroutine: require\n";
+		// Need GL version 430
+		header += "#extension GL_ARB_explicit_uniform_location: require\n";
 	}
 #ifdef ENABLE_OGL_STENCIL_DEBUG
 	header += "#define ENABLE_OGL_STENCIL_DEBUG 1\n";
@@ -413,7 +417,7 @@ GLuint GSShaderOGL::Compile(const std::string& glsl_file, const std::string& ent
 
 	std::string header = GenGlslHeader(entry, type, macro_sel);
 	int shader_nb = 1;
-#if 0
+#if 1
 	sources[0] = header.c_str();
 	sources[1] = glsl_h_code;
 	shader_nb++;

plugins/GSdx/GSTextureFXOGL.cpp

@@ -56,6 +56,9 @@ void GSDeviceOGL::CreateTextureFX()
 
 	for (uint32 key = 0; key < OMDepthStencilSelector::size(); key++)
 		m_om_dss[key] = CreateDepthStencil(OMDepthStencilSelector(key));
+
+	// Help to debug FS in apitrace
+	m_apitrace = CompilePS(PSSelector());
 }
 
 GSDepthStencilOGL* GSDeviceOGL::CreateDepthStencil(OMDepthStencilSelector dssel)

plugins/GSdx/GSTextureOGL.cpp

@@ -204,9 +204,9 @@ GSTextureOGL::GSTextureOGL(int type, int w, int h, int format, GLuint fbo_read)
 
 	// Bunch of constant parameter
 	switch (m_format) {
-		case GL_R32UI:
+		case GL_R32I:
 			m_int_format    = GL_RED_INTEGER;
-			m_int_type      = GL_UNSIGNED_INT;
+			m_int_type      = GL_INT;
 			m_int_alignment = 4;
 			m_int_shift     = 2;
 			break;
@@ -559,7 +559,7 @@ void GSTextureOGL::SaveRaw(const string& fn, const void* image, uint32 pitch)
 
 	for(int h = m_size.y; h > 0; h--) {
 		for (int w = m_size.x; w > 0; w--, data += 1) {
-			if (*data == 0xffffffff)
+			if (*data > 0xffffff)
 				fprintf(fp, "");
 			else {
 				fprintf(fp, "%x", *data);
@@ -594,11 +594,11 @@ bool GSTextureOGL::Save(const string& fn, bool dds)
 		glReadPixels(0, 0, m_size.x, m_size.y, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, image);
 
 		gl_BindFramebuffer(GL_READ_FRAMEBUFFER, 0);
-	} else if(m_format == GL_R32UI) {
+	} else if(m_format == GL_R32I) {
 		gl_ActiveTexture(GL_TEXTURE0 + 6);
 		glBindTexture(GL_TEXTURE_2D, m_texture_id);
 
-		glGetTexImage(GL_TEXTURE_2D, 0, GL_RED_INTEGER, GL_UNSIGNED_INT, image);
+		glGetTexImage(GL_TEXTURE_2D, 0, GL_RED_INTEGER, GL_INT, image);
 		SaveRaw(fn, image, pitch);
 
 		// Not supported in Save function

plugins/GSdx/res/glsl_source.h

@@ -817,7 +817,7 @@ static const char* tfx_glsl =
 	"#ifndef DISABLE_GL42_image\n"
 	"#if PS_DATE > 0\n"
 	"// FIXME how to declare memory access\n"
-	"layout(r32ui, binding = 0) coherent uniform uimage2D img_prim_min;\n"
+	"layout(r32i, binding = 0) coherent uniform iimage2D img_prim_min;\n"
 	"#endif\n"
 	"#else\n"
 	"// use basic stencil\n"
@@ -1316,6 +1316,16 @@ static const char* tfx_glsl =
 	"#if !GL_ES\n"
 	"void ps_main()\n"
 	"{\n"
+	"#if PS_DATE == 3 && !defined(DISABLE_GL42_image)\n"
+	"    int stencil_ceil = imageLoad(img_prim_min, ivec2(gl_FragCoord.xy));\n"
+	"    // Note gl_PrimitiveID == stencil_ceil will be the primitive that will update\n"
+	"    // the bad alpha value so we must keep it.\n"
+	"\n"
+	"	if (gl_PrimitiveID > stencil_ceil) {\n"
+	"		discard;\n"
+	"	}\n"
+	"#endif\n"
+	"\n"
 	"    vec4 c = ps_color();\n"
 	"\n"
 	"    float alpha = c.a * 2.0;\n"
@@ -1347,33 +1357,11 @@ static const char* tfx_glsl =
 	"    }\n"
 	"#endif\n"
 	"\n"
-	"    // TODO\n"
-	"    // warning non uniform flow ???\n"
-	"#if PS_DATE == 3 && !defined(DISABLE_GL42_image)\n"
-	"    uint stencil_ceil = imageLoad(img_prim_min, ivec2(gl_FragCoord.xy));\n"
-	"    // Note gl_PrimitiveID == stencil_ceil will be the primitive that will update\n"
-	"    // the bad alpha value so we must keep it.\n"
-	"#if 0\n"
-	"	if (stencil_ceil > 0)\n"
-	"		c = vec4(1.0, 0.0, 0.0, 1.0);\n"
-	"	else\n"
-	"		c = vec4(0.0, 1.0, 0.0, 1.0);\n"
-	"#endif\n"
-	"\n"
-	"#if 1\n"
-	"	if (gl_PrimitiveID > stencil_ceil) {\n"
-	"		discard;\n"
-	"	}\n"
-	"#endif\n"
-	"\n"
-	"#endif\n"
-	"\n"
 	"\n"
 	"#if (PS_DATE == 2 || PS_DATE == 1) && !defined(DISABLE_GL42_image)\n"
 	"    // Don't write anything on the framebuffer\n"
 	"    // Note: you can't use discard because it will also drop\n"
 	"    // image operation\n"
-	"    // Note2: output will be disabled too in opengl\n"
 	"#else\n"
 	"    SV_Target0 = c;\n"
 	"    SV_Target1 = vec4(alpha, alpha, alpha, alpha);\n"

plugins/GSdx/res/tfx.glsl

@@ -307,7 +307,7 @@ layout(binding = 1) uniform sampler2D PaletteSampler;
 #ifndef DISABLE_GL42_image
 #if PS_DATE > 0
 // FIXME how to declare memory access
-layout(r32ui, binding = 0) coherent uniform uimage2D img_prim_min;
+layout(r32i, binding = 0) coherent uniform iimage2D img_prim_min;
 #endif
 #else
 // use basic stencil
@@ -806,6 +806,16 @@ void ps_main()
 #if !GL_ES
 void ps_main()
 {
+#if PS_DATE == 3 && !defined(DISABLE_GL42_image)
+    int stencil_ceil = imageLoad(img_prim_min, ivec2(gl_FragCoord.xy));
+    // Note gl_PrimitiveID == stencil_ceil will be the primitive that will update
+    // the bad alpha value so we must keep it.
+
+	if (gl_PrimitiveID > stencil_ceil) {
+		discard;
+	}
+#endif
+
     vec4 c = ps_color();
 
     float alpha = c.a * 2.0;
@@ -837,33 +847,11 @@ void ps_main()
     }
 #endif
 
-    // TODO
-    // warning non uniform flow ???
-#if PS_DATE == 3 && !defined(DISABLE_GL42_image)
-    uint stencil_ceil = imageLoad(img_prim_min, ivec2(gl_FragCoord.xy));
-    // Note gl_PrimitiveID == stencil_ceil will be the primitive that will update
-    // the bad alpha value so we must keep it.
-#if 0
-	if (stencil_ceil > 0)
-		c = vec4(1.0, 0.0, 0.0, 1.0);
-	else
-		c = vec4(0.0, 1.0, 0.0, 1.0);
-#endif
-
-#if 1
-	if (gl_PrimitiveID > stencil_ceil) {
-		discard;
-	}
-#endif
-
-#endif
-
 
 #if (PS_DATE == 2 || PS_DATE == 1) && !defined(DISABLE_GL42_image)
     // Don't write anything on the framebuffer
     // Note: you can't use discard because it will also drop
     // image operation
-    // Note2: output will be disabled too in opengl
 #else
     SV_Target0 = c;
     SV_Target1 = vec4(alpha, alpha, alpha, alpha);