gsdx-ogl:

* Use a geometry shader pass-through to replace previous AMD workaround
* various cosmetic change


git-svn-id: http://pcsx2.googlecode.com/svn/branches/gsdx-ogl@5038 96395faa-99c1-11dd-bbfe-3dabce05a288

plugins/GSdx/CMakeLists.txt

@@ -19,6 +19,7 @@ set(CommonFlags
     -std=c++0x
     -fno-strict-aliasing
     -DOGL_DEBUG
+	-DAMD_DRIVER_WORKAROUND
     )
 
 set(OptimizationFlags

plugins/GSdx/GSDeviceOGL.cpp

@@ -54,8 +54,8 @@
 
 //#define LOUD_DEBUGGING
 #define SHADER_DEBUG
-//#define DUMP_START (13000)
+//#define DUMP_START (70)
-//#define DUMP_LENGTH (200)
+//#define DUMP_LENGTH (130)
 //#define DUMP_ONLY_FRAME  (112)
 
 #ifdef DUMP_START
@@ -237,6 +237,7 @@ bool GSDeviceOGL::Create(GSWnd* wnd)
 	// convert
 	// ****************************************************************
 	CompileShaderFromSource("convert.glsl", "vs_main", GL_VERTEX_SHADER, &m_convert.vs);
+	CompileShaderFromSource("convert.glsl", "gs_main", GL_GEOMETRY_SHADER, &m_convert.gs);
 	for(int i = 0; i < countof(m_convert.ps); i++)
 		CompileShaderFromSource("convert.glsl", format("ps_main%d", i), GL_FRAGMENT_SHADER, &m_convert.ps[i]);
 
@@ -540,9 +541,11 @@ void GSDeviceOGL::DrawPrimitive()
 			case GL_POINTS: topo = "point"; break;
 			default: topo = "!!!!";
 		}
-		fprintf(stderr, "Draw %d (Frame %d), %d elem of %s\n", g_draw_count, g_frame_count, m_state.vb_state->count, topo.c_str() );
+		fprintf(stderr, "Draw %d (Frame %d), %d elem of %s\n", g_draw_count, g_frame_count, /*m_state.vb_state->count*/ 0, topo.c_str() );
 		fprintf(stderr, "vs: %d ; gs: %d ; ps: %d\n", m_state.vs, m_state.gs, m_state.ps);
 		fprintf(stderr, "Blend: %d, Depth: %d, Stencil: %d \n",m_state.bs->m_enable, m_state.dss->m_depth_enable, m_state.dss->m_stencil_enable);
+		m_state.bs->debug();
+		m_state.dss->debug_depth();
 	}
 #endif
 
@@ -560,19 +563,6 @@ void GSDeviceOGL::DrawPrimitive()
 
 	g_draw_count++;
 #endif
-
-
-	// FIXME AMD driver bug workaround
-	// You cannot unattach shader. So destroy everythings and recreate the shader pipeline...
-	// Slow and painful...
-	glBindProgramPipeline(0);
-	glDeleteProgramPipelines(1, &m_pipeline);
-	m_state.gs = 0;
-	m_state.ps = 0;
-	m_state.vs = 0;
-
-	glGenProgramPipelines(1, &m_pipeline);
-	glBindProgramPipeline(m_pipeline);
 }
 
 void GSDeviceOGL::ClearRenderTarget(GSTexture* t, const GSVector4& c)
@@ -822,7 +812,11 @@ void GSDeviceOGL::StretchRect(GSTexture* st, const GSVector4& sr, GSTexture* dt,
 	// gs
 	// ************************************
 
+#ifdef AMD_DRIVER_WORKAROUND
+	GSSetShader(m_convert.gs);
+#else
 	GSSetShader(0);
+#endif
 
 	// ************************************
 	// ps
@@ -914,7 +908,12 @@ void GSDeviceOGL::SetupDATE(GSTexture* rt, GSTexture* ds, const GSVertexPT1* ver
 
 		// gs
 
+#ifdef AMD_DRIVER_WORKAROUND
+		GSSetShader(m_convert.gs);
+#else
 		GSSetShader(0);
+#endif
+
 
 		// ps
 
@@ -1044,7 +1043,6 @@ void GSDeviceOGL::PSSetSamplerState(GLuint ss0, GLuint ss1, GLuint ss2)
 
 void GSDeviceOGL::PSSetShader(GLuint ps)
 {
-
 	if(m_state.ps != ps)
 	{
 		m_state.ps = ps;
@@ -1279,8 +1277,8 @@ void GSDeviceOGL::CompileShaderFromSource(const std::string& glsl_file, const st
 	fprintf(stderr, "%s (entry %s, prog %d) :", glsl_file.c_str(), entry.c_str(), *program);
 	fprintf(stderr, "\n%s", macro_sel.c_str());
 	fprintf(stderr, "%s\n", log);
-#endif
 	free(log);
+#endif
 }
 
 void GSDeviceOGL::CheckDebugLog()

plugins/GSdx/GSDeviceOGL.h

@@ -54,6 +54,13 @@ struct GSBlendStateOGL {
 		, m_a_msk(GL_TRUE)
 	{}
 
+	void debug()
+	{
+		if (!m_enable) return;
+		fprintf(stderr,"Blend RGB: %x src:%x dst:%x\n", m_equation_RGB, m_func_sRGB, m_func_dRGB);
+		fprintf(stderr,"Blend ALPHA: %x src:%x dst:%x\n", m_equation_ALPHA, m_func_sALPHA, m_func_dALPHA);
+		fprintf(stderr,"Mask. R:%d B:%d G:%d A:%d\n", m_r_msk, m_b_msk, m_g_msk, m_a_msk);
+	}
 };
 
 struct GSDepthStencilOGL {
@@ -81,6 +88,11 @@ struct GSDepthStencilOGL {
 		, m_stencil_spass_dpass_op(GL_KEEP)
 	{}
 
+	void debug_depth()
+	{
+		if (!m_depth_enable) return;
+		fprintf(stderr, "Depth %x, %x\n", m_depth_func, m_depth_mask);
+	}
 };
 
 class GSUniformBufferOGL {
@@ -503,6 +515,7 @@ class GSDeviceOGL : public GSDevice
 		GLuint ps[8];	// program object
 		GLuint ln;		// sampler object
 		GLuint pt;		// sampler object
+		GLuint gs;
 		GSDepthStencilOGL* dss;
 		GSBlendStateOGL* bs;
 	} m_convert;

plugins/GSdx/GSTextureFXOGL.cpp

@@ -120,7 +120,12 @@ void GSDeviceOGL::SetupGS(GSSelector sel)
 	// Static
 	// *************************************************************
 	GLuint gs = 0;
-	if(sel.prim > 0 && (sel.iip == 0 || sel.prim == 3)) {
+#ifdef AMD_DRIVER_WORKAROUND
+	if (true)
+#else
+	if(sel.prim > 0 && (sel.iip == 0 || sel.prim == 3))
+#endif
+	{
 		auto i = m_gs.find(sel);
 
 		if(i == m_gs.end()) {

plugins/GSdx/GSTextureOGL.cpp

@@ -22,8 +22,8 @@
 #pragma once
 
 #include "GSTextureOGL.h"
-static uint g_state_texture_unit = 0;
+static int g_state_texture_unit = -1;
-static uint g_state_texture_id = 0;
+static int g_state_texture_id = -1;
 
 GSTextureOGL::GSTextureOGL(int type, int w, int h, bool msaa, int format)
 	: m_extra_buffer_id(0),
@@ -317,6 +317,68 @@ struct BITMAPINFOHEADER
 #pragma pack(pop)
 
 #endif
+void GSTextureOGL::Save(const string& fn, const void* image, uint32 pitch)
+{
+	// Build a BMP file
+	FILE* fp = fopen(fn.c_str(), "wb");
+
+	BITMAPINFOHEADER bih;
+
+	memset(&bih, 0, sizeof(bih));
+
+	bih.biSize = sizeof(bih);
+	bih.biWidth = m_size.x;
+	bih.biHeight = m_size.y;
+	bih.biPlanes = 1;
+	bih.biBitCount = 32;
+	bih.biCompression = BI_RGB;
+	bih.biSizeImage = m_size.x * m_size.y << 2;
+
+	BITMAPFILEHEADER bfh;
+
+	memset(&bfh, 0, sizeof(bfh));
+
+	uint8* bfType = (uint8*)&bfh.bfType;
+
+	// bfh.bfType = 'MB';
+	bfType[0] = 0x42;
+	bfType[1] = 0x4d;
+	bfh.bfOffBits = sizeof(bfh) + sizeof(bih);
+	bfh.bfSize = bfh.bfOffBits + bih.biSizeImage;
+	bfh.bfReserved1 = bfh.bfReserved2 = 0;
+
+	fwrite(&bfh, 1, sizeof(bfh), fp);
+	fwrite(&bih, 1, sizeof(bih), fp);
+
+	uint8* data = (uint8*)image + (m_size.y - 1) * pitch;
+
+	for(int h = m_size.y; h > 0; h--, data -= pitch)
+	{
+		if (IsDss()) {
+			// Only get the depth and convert it to an integer
+			uint8* better_data = data;
+			for (int w = m_size.x; w > 0; w--, better_data += 8) {
+				float* input = (float*)better_data;
+				// FIXME how to dump 32 bits value into 8bits component color
+				uint32 depth = (uint32)ldexpf(*input, 32);
+				uint8 small_depth = depth >> 24;
+				uint8 better_data[4] = {small_depth, small_depth, small_depth, 0 };
+				fwrite(&better_data, 1, 4, fp);
+			}
+		} else {
+			// swap red and blue
+			uint8* better_data = data;
+			for (int w = m_size.x; w > 0; w--, better_data += 4) {
+				uint8 red = better_data[2];
+				better_data[2] = better_data[0];
+				better_data[0] = red;
+				fwrite(better_data, 1, 4, fp);
+			}
+		}
+	}
+
+	fclose(fp);
+}
 
 bool GSTextureOGL::Save(const string& fn, bool dds)
 {
@@ -325,6 +387,8 @@ bool GSTextureOGL::Save(const string& fn, bool dds)
 	if (IsDss()) pitch *= 2;
 	char* image = (char*)malloc(pitch * m_size.y);
 
+	// FIXME instead of swapping manually B and R maybe you can request the driver to do it
+	// for us
 	if (IsBackbuffer()) {
 		glReadBuffer(GL_BACK);
 		glReadPixels(0, 0, m_size.x, m_size.y, GL_RGBA, GL_UNSIGNED_BYTE, image);
@@ -336,70 +400,9 @@ bool GSTextureOGL::Save(const string& fn, bool dds)
 		glGetTexImage(m_texture_target, 0, GL_RGBA, GL_UNSIGNED_BYTE, image);
 	}
 
-	// Build a BMP file
+	Save(fn, image, pitch);
-	if(FILE* fp = fopen(fn.c_str(), "wb"))
+	free(image);
-	{
-		BITMAPINFOHEADER bih;
 
-		memset(&bih, 0, sizeof(bih));
+	return true;
-
-		bih.biSize = sizeof(bih);
-		bih.biWidth = m_size.x;
-		bih.biHeight = m_size.y;
-		bih.biPlanes = 1;
-		bih.biBitCount = 32;
-		bih.biCompression = BI_RGB;
-		bih.biSizeImage = m_size.x * m_size.y << 2;
-
-		BITMAPFILEHEADER bfh;
-
-		memset(&bfh, 0, sizeof(bfh));
-
-		uint8* bfType = (uint8*)&bfh.bfType;
-
-		// bfh.bfType = 'MB';
-		bfType[0] = 0x42;
-		bfType[1] = 0x4d;
-		bfh.bfOffBits = sizeof(bfh) + sizeof(bih);
-		bfh.bfSize = bfh.bfOffBits + bih.biSizeImage;
-		bfh.bfReserved1 = bfh.bfReserved2 = 0;
-
-		fwrite(&bfh, 1, sizeof(bfh), fp);
-		fwrite(&bih, 1, sizeof(bih), fp);
-
-		uint8* data = (uint8*)image + (m_size.y - 1) * pitch;
-
-		for(int h = m_size.y; h > 0; h--, data -= pitch)
-		{
-			if (IsDss()) {
-				// Only get the depth and convert it to an integer
-				uint8* better_data = data;
-				for (int w = m_size.x; w > 0; w--, better_data += 8) {
-					float* input = (float*)better_data;
-					// FIXME how to dump 32 bits value into 8bits component color
-					uint32 depth = (uint32)ldexpf(*input, 32);
-					uint8 small_depth = depth >> 24;
-					uint8 better_data[4] = {small_depth, small_depth, small_depth, 0 };
-					fwrite(&better_data, 1, 4, fp);
-				}
-			} else {
-				// swap red and blue
-				uint8* better_data = data;
-				for (int w = m_size.x; w > 0; w--, better_data += 4) {
-					uint8 red = better_data[2];
-					better_data[2] = better_data[0];
-					better_data[0] = red;
-					fwrite(better_data, 1, 4, fp);
-				}
-			}
-		}
-
-		fclose(fp);
-
-		free(image);
-		return true;
-	}
-
-	return false;
 }
 

plugins/GSdx/GSTextureOGL.h

@@ -39,6 +39,7 @@ class GSTextureOGL : public GSTexture
 		bool Map(GSMap& m, const GSVector4i* r = NULL);
 		void Unmap();
 		bool Save(const string& fn, bool dds = false);
+		void Save(const string& fn, const void* image, uint32 pitch);
 
 		void EnableUnit(uint unit);
 		void Attach(GLenum attachment);

plugins/GSdx/res/convert.glsl

@@ -1,5 +1,12 @@
 //#version 420 // Keep it for editor detection
 
+struct vertex_basic
+{
+    vec4 p;
+    vec2 t;
+};
+
+
 #ifdef VERTEX_SHADER
 
 out gl_PerVertex {
@@ -19,24 +26,54 @@ layout(location = 1) in vec2 TEXCOORD0;
 // smooth, the default, means to do perspective-correct interpolation.
 //
 // The centroid qualifier only matters when multisampling. If this qualifier is not present, then the value is interpolated to the pixel's center, anywhere in the pixel, or to one of the pixel's samples. This sample may lie outside of the actual primitive being rendered, since a primitive can cover only part of a pixel's area. The centroid qualifier is used to prevent this; the interpolation point must fall within both the pixel's area and the primitive's area.
-// FIXME gl_Position
+layout(location = 0) out vertex_basic VSout;
-layout(location = 0) out vec4 POSITION_OUT;
-layout(location = 1) out vec2 TEXCOORD0_OUT;
 
 void vs_main()
 {
-    POSITION_OUT = POSITION;
+    VSout.p = POSITION;
-    TEXCOORD0_OUT = TEXCOORD0;
+    VSout.t = TEXCOORD0;
     gl_Position = POSITION; // NOTE I don't know if it is possible to merge POSITION_OUT and gl_Position
 }
 
 #endif
 
+#ifdef GEOMETRY_SHADER
+in gl_PerVertex {
+    vec4 gl_Position;
+    float gl_PointSize;
+    float gl_ClipDistance[];
+} gl_in[];
+
+out gl_PerVertex {
+    vec4 gl_Position;
+    float gl_PointSize;
+    float gl_ClipDistance[];
+};
+
+// FIXME
+// AMD Driver bug again !!!!
+//layout(location = 0) in vertex GSin[];
+in vertex_basic GSin[];
+
+layout(location = 0) out vertex_basic GSout;
+layout(triangles) in;
+layout(triangle_strip, max_vertices = 3) out;
+
+void gs_main()
+{
+    for(int i = 0; i < gl_in.length(); i++) {
+        gl_Position = gl_in[i].gl_Position;
+        GSout = GSin[i];
+        EmitVertex();
+    }
+    EndPrimitive();
+}
+#endif
+
 #ifdef FRAGMENT_SHADER
 // NOTE: pixel can be clip with "discard"
 
-layout(location = 0) in vec4 SV_Position;
+layout(location = 0) in vertex_basic PSin;
-layout(location = 1) in vec2 TEXCOORD0;
 
 layout(location = 0) out vec4 SV_Target0;
 layout(location = 1) out uint SV_Target1;
@@ -45,7 +82,7 @@ layout(binding = 0) uniform sampler2D TextureSampler;
 
 vec4 sample_c()
 {
-    return texture(TextureSampler, vec2(TEXCOORD0.x,TEXCOORD0.y) );
+    return texture(TextureSampler, PSin.t );
 }
 
 vec4 ps_crt(uint i)
@@ -88,7 +125,7 @@ void ps_main7()
 
 void ps_main5() // triangular
 {
-	highp uvec4 p = uvec4(SV_Position);
+	highp uvec4 p = uvec4(PSin.p);
 
 	vec4 c = ps_crt(((p.x + ((p.y >> 1u) & 1u) * 3u) >> 1u) % 3u);
 
@@ -97,7 +134,7 @@ void ps_main5() // triangular
 
 void ps_main6() // diagonal
 {
-	uvec4 p = uvec4(SV_Position);
+	uvec4 p = uvec4(PSin.p);
 
 	vec4 c = ps_crt((p.x + (p.y % 3)) % 3);
 

plugins/GSdx/res/interlace.glsl

@@ -1,8 +1,13 @@
 //#version 420 // Keep it for editor detection
 
+struct vertex_basic
+{
+    vec4 p;
+    vec2 t;
+};
+
 #ifdef FRAGMENT_SHADER
-layout(location = 0) in vec4 SV_Position;
+layout(location = 0) in vertex_basic PSin;
-layout(location = 1) in vec2 TEXCOORD0;
 
 layout(location = 0) out vec4 SV_Target0;
 
@@ -19,8 +24,8 @@ void ps_main0()
 {
     // I'm not sure it impact us but be safe to lookup texture before conditional if
     // see: http://www.opengl.org/wiki/GLSL_Sampler#Non-uniform_flow_control
-    vec4 c = texture(TextureSampler, TEXCOORD0);
+    vec4 c = texture(TextureSampler, PSin.t);
-    if (fract(TEXCOORD0.y * hH) - 0.5 < 0.0)
+    if (fract(PSin.t.y * hH) - 0.5 < 0.0)
         discard;
 
     SV_Target0 = c;
@@ -30,8 +35,8 @@ void ps_main1()
 {
     // I'm not sure it impact us but be safe to lookup texture before conditional if
     // see: http://www.opengl.org/wiki/GLSL_Sampler#Non-uniform_flow_control
-    vec4 c = texture(TextureSampler, TEXCOORD0);
+    vec4 c = texture(TextureSampler, PSin.t);
-    if (0.5 - fract(TEXCOORD0.y * hH) < 0.0)
+    if (0.5 - fract(PSin.t.y * hH) < 0.0)
         discard;
 
     SV_Target0 = c;
@@ -39,16 +44,16 @@ void ps_main1()
 
 void ps_main2()
 {
-    vec4 c0 = texture(TextureSampler, TEXCOORD0 - ZrH);
+    vec4 c0 = texture(TextureSampler, PSin.t - ZrH);
-    vec4 c1 = texture(TextureSampler, TEXCOORD0);
+    vec4 c1 = texture(TextureSampler, PSin.t);
-    vec4 c2 = texture(TextureSampler, TEXCOORD0 + ZrH);
+    vec4 c2 = texture(TextureSampler, PSin.t + ZrH);
 
     SV_Target0 = (c0 + c1 * 2 + c2) / 4;
 }
 
 void ps_main3()
 {
-    SV_Target0 = texture(TextureSampler, TEXCOORD0);
+    SV_Target0 = texture(TextureSampler, PSin.t);
 }
 
 #endif

plugins/GSdx/res/merge.glsl

@@ -1,8 +1,13 @@
 //#version 420 // Keep it for editor detection
 
+struct vertex_basic
+{
+    vec4 p;
+    vec2 t;
+};
+
 #ifdef FRAGMENT_SHADER
-layout(location = 0) in vec4 SV_Position;
+layout(location = 0) in vertex_basic PSin;
-layout(location = 1) in vec2 TEXCOORD0;
 
 layout(location = 0) out vec4 SV_Target0;
 
@@ -15,14 +20,14 @@ layout(binding = 0) uniform sampler2D TextureSampler;
 
 void ps_main0()
 {
-    vec4 c = texture(TextureSampler, TEXCOORD0);
+    vec4 c = texture(TextureSampler, PSin.t);
 	c.a = min(c.a * 2, 1.0);
     SV_Target0 = c;
 }
 
 void ps_main1()
 {
-    vec4 c = texture(TextureSampler, TEXCOORD0);
+    vec4 c = texture(TextureSampler, PSin.t);
 	c.a = BGColor.a;
     SV_Target0 = c;
 }

plugins/GSdx/res/tfx.glsl

@@ -55,7 +55,7 @@ layout(location = 3) in uvec2 i_p;
 layout(location = 4) in uint  i_z;
 layout(location = 5) in vec4  i_f;
 
-layout(location = 0) out vertex OUT;
+layout(location = 0) out vertex VSout;
 
 out gl_PerVertex {
     vec4 gl_Position;
@@ -91,33 +91,33 @@ void vs_main()
     // FLIP vertically
     final_p.y *= -1.0f;
 
-	OUT.p = final_p;
+	VSout.p = final_p;
     gl_Position = final_p; // NOTE I don't know if it is possible to merge POSITION_OUT and gl_Position
 #if VS_RTCOPY
-	OUT.tp = final_p * vec4(0.5, -0.5, 0, 0) + 0.5;
+	VSout.tp = final_p * vec4(0.5, -0.5, 0, 0) + 0.5;
 #endif
 
 	if(VS_TME != 0)
 	{
 		if(VS_FST != 0)
 		{
-			OUT.t.xy = i_t * TextureScale;
+			VSout.t.xy = i_t * TextureScale;
-			OUT.t.w = 1.0f;
+			VSout.t.w = 1.0f;
 		}
 		else
 		{
-			OUT.t.xy = i_t;
+			VSout.t.xy = i_t;
-			OUT.t.w = i_q;
+			VSout.t.w = i_q;
 		}
 	}
 	else
 	{
-		OUT.t.xy = vec2(0.0f, 0.0f);
+		VSout.t.xy = vec2(0.0f, 0.0f);
-		OUT.t.w = 1.0f;
+		VSout.t.w = 1.0f;
 	}
 
-	OUT.c = i_c;
+	VSout.c = i_c;
-	OUT.t.z = i_f.a;
+	VSout.t.z = i_f.a;
 }
 
 #endif
@@ -633,8 +633,5 @@ void ps_main()
 	}
 
     SV_Target1 = c;
-
-    //SV_Target0 = vec4(1.0f,0.0f,0.0f, 1.0f);
-    //SV_Target1 = vec4(0.0f,1.0f,0.0f, 1.0f);
 }
 #endif