gsdx-ogl: LINUX-ONLY

* Fix the Geomtry shader to output 2 triangles for quad primitive (ie 2R rendering)
- There is an AMD  driver bug on geomtry shader input interface (well could be the spec too). Tell me if it still working on nvidia
* Add a workaroung to a previous AMD bug. It is impossible to unattach a shader so destroy the full shader pipeline...
* Be more strict on FBO management. Would optimize it later
* use a texture insted of a render buffer for depth-stencil management.
* add more dumping capabilities (in particular depth buffer)


git-svn-id: http://pcsx2.googlecode.com/svn/branches/gsdx-ogl@5033 96395faa-99c1-11dd-bbfe-3dabce05a288

plugins/GSdx/GSDeviceOGL.cpp

@@ -53,8 +53,10 @@
 // glUniformBlockBinding(program, block_index, block_binding_point);
 
 //#define LOUD_DEBUGGING
-//#define DUMP_START (380)
+#define SHADER_DEBUG
-#define DUMP_LENGTH (20)
+//#define DUMP_START (500)
+//#define DUMP_LENGTH (40)
+//#define DUMP_ONLY_FRAME  (112)
 
 #ifdef DUMP_START
 static uint32 g_draw_count = 0;
@@ -508,16 +510,27 @@ void GSDeviceOGL::Flip()
 
 void GSDeviceOGL::DrawPrimitive()
 {
-	glDrawArrays(m_state.topology, m_state.vb_state->start, m_state.vb_state->count);
 #ifdef DUMP_START
-	if (g_draw_count > DUMP_START && g_draw_count < (DUMP_START+DUMP_LENGTH)) {
+	bool dump_me = false;
+	if ( (g_draw_count > DUMP_START && g_draw_count < (DUMP_START+DUMP_LENGTH)) )
+		dump_me = true;
+#ifdef DUMP_ONLY_FRAME
+	if (DUMP_ONLY_FRAME != 0 && DUMP_ONLY_FRAME == g_frame_count)
+		dump_me = true;
+	else if (DUMP_ONLY_FRAME != 0)
+		dump_me = false;
+#endif
+#endif
+	
+	// DUMP INPUT
+#ifdef DUMP_START
+	if ( dump_me ) {
 		for (auto i = 0 ; i < 3 ; i++) {
 			if (m_state.ps_srv[i] != NULL) {
-				m_state.ps_srv[i]->Save(format("/tmp/in_%d__%d.bmp", g_draw_count, i),false);
+				m_state.ps_srv[i]->Save(format("/tmp/in_%d__%d.bmp", g_draw_count, i));
 			}
 		}
-		if (m_state.rtv != NULL) m_state.rtv->Save(format("/tmp/out_%d.bmp", g_draw_count),false);
+		if (m_state.dsv != NULL) m_state.dsv->Save(format("/tmp/ds_in_%d.bmp", g_draw_count));
-		if (m_state.dsv != NULL) m_state.dsv->Save(format("/tmp/out_%d_ds.bmp", g_draw_count),false);
 
 		string topo;
 		switch (m_state.topology) {
@@ -530,18 +543,41 @@ void GSDeviceOGL::DrawPrimitive()
 		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, "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);
+	}
+#endif
+
+	glDrawArrays(m_state.topology, m_state.vb_state->start, m_state.vb_state->count);
+
+	// DUMP OUTPUT
+#ifdef DUMP_START
+	if ( dump_me ) {
+		if (m_state.rtv != NULL) m_state.rtv->Save(format("/tmp/out_%d.bmp", g_draw_count));
+		if (m_state.dsv != NULL) m_state.dsv->Save(format("/tmp/ds_out_%d.bmp", g_draw_count));
 
-		//fprintf(stderr, "type: %d, format: 0x%x\n", m_state.rtv->GetType(), m_state.rtv->GetFormat());
 		fprintf(stderr, "\n");
 
 	}
 
 	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)
 {
+	GLuint fbo_old = m_state.fbo;
 	if (static_cast<GSTextureOGL*>(t)->IsBackbuffer()) {
 		// FIXME I really not sure
 		OMSetFBO(0);
@@ -557,6 +593,7 @@ void GSDeviceOGL::ClearRenderTarget(GSTexture* t, const GSVector4& c)
 		static_cast<GSTextureOGL*>(t)->Attach(GL_COLOR_ATTACHMENT0);
 		glClearBufferfv(GL_COLOR, 0, c.v);
 	}
+	OMSetFBO(fbo_old);
 }
 
 void GSDeviceOGL::ClearRenderTarget(GSTexture* t, uint32 c)
@@ -567,16 +604,19 @@ void GSDeviceOGL::ClearRenderTarget(GSTexture* t, uint32 c)
 
 void GSDeviceOGL::ClearDepth(GSTexture* t, float c)
 {
+	GLuint fbo_old = m_state.fbo;
 	// FIXME I need to clarify this FBO attachment stuff
 	// I would like to avoid FBO for a basic clean operation
 	OMSetFBO(m_fbo);
 	static_cast<GSTextureOGL*>(t)->Attach(GL_DEPTH_STENCIL_ATTACHMENT);
 	// FIXME can you clean depth and stencil separately
 	glClearBufferfv(GL_DEPTH, 0, &c);
+	OMSetFBO(fbo_old);
 }
 
 void GSDeviceOGL::ClearStencil(GSTexture* t, uint8 c)
 {
+	GLuint fbo_old = m_state.fbo;
 	// FIXME I need to clarify this FBO attachment stuff
 	// I would like to avoid FBO for a basic clean operation
 	OMSetFBO(m_fbo);
@@ -584,6 +624,7 @@ void GSDeviceOGL::ClearStencil(GSTexture* t, uint8 c)
 	GLint color = c;
 	// FIXME can you clean depth and stencil separately
 	glClearBufferiv(GL_STENCIL, 0, &color);
+	OMSetFBO(fbo_old);
 }
 
 GSTexture* GSDeviceOGL::CreateRenderTarget(int w, int h, bool msaa, int format)
@@ -666,6 +707,11 @@ void GSDeviceOGL::CopyRect(GSTexture* st, GSTexture* dt, const GSVector4i& r)
 	// Maybe opengl 4.3 !
 	// FIXME check those function work as expected
 	
+	// FIXME FBO
+	GLuint fbo_old = m_state.fbo;
+	OMSetFBO(m_fbo);
+
+
 	// Set the input of glCopyTexSubImage2D
 	static_cast<GSTextureOGL*>(st)->Attach(GL_COLOR_ATTACHMENT1);
 	glReadBuffer(GL_COLOR_ATTACHMENT1);
@@ -674,6 +720,8 @@ void GSDeviceOGL::CopyRect(GSTexture* st, GSTexture* dt, const GSVector4i& r)
 	static_cast<GSTextureOGL*>(dt)->EnableUnit(0);
 	glCopyTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 0, 0, dt->GetWidth(), dt->GetHeight());
 
+	OMSetFBO(fbo_old);
+
 #if 0
 	// FIXME attach the texture to the FBO
 	GSTextureOGL* st_ogl = (GSTextureOGL*) st;
@@ -879,6 +927,9 @@ void GSDeviceOGL::SetupDATE(GSTexture* rt, GSTexture* ds, const GSVertexPT1* ver
 
 		//
 
+#ifdef DUMP_START
+		fprintf(stderr, "draw date!!!\n");
+#endif
 		DrawPrimitive();
 
 		//
@@ -982,7 +1033,6 @@ void GSDeviceOGL::GSSetShader(GLuint gs)
 	if(m_state.gs != gs)
 	{
 		m_state.gs = gs;
-		// FIXME AMD driver bug !!!!!!!!
 		glUseProgramStages(m_pipeline, GL_GEOMETRY_SHADER_BIT, gs);
 	}
 }
@@ -1244,6 +1294,7 @@ void GSDeviceOGL::CompileShaderFromSource(const std::string& glsl_file, const st
 	free(header_str);
 	free(sources_array);
 
+#ifdef SHADER_DEBUG
 	// Print a nice debug log
 	GLint log_length = 0;
 	glGetProgramiv(*program, GL_INFO_LOG_LENGTH, &log_length);
@@ -1254,6 +1305,7 @@ 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);
 }
 

plugins/GSdx/GSDeviceOGL.h

@@ -180,10 +180,16 @@ struct GSVertexBufferState {
 		for (int i = 0; i < layout_nbr; i++) {
 			// Note this function need both a vertex array object and a GL_ARRAY_BUFFER buffer
 			glEnableVertexAttribArray(layout[i].index);
-			if (layout[i].type == GL_UNSIGNED_INT || layout[i].type == GL_UNSIGNED_SHORT)
+			switch (layout[i].type) {
+				case GL_UNSIGNED_SHORT:
+				case GL_UNSIGNED_INT:
+					// Rule: when shader use integral (not normalized) you must use glVertexAttribIPointer (note the extra I)
 					glVertexAttribIPointer(layout[i].index, layout[i].size, layout[i].type, layout[i].stride, layout[i].offset);
-			else
+					break;
+				default:
 					glVertexAttribPointer(layout[i].index, layout[i].size, layout[i].type, layout[i].normalize,  layout[i].stride, layout[i].offset);
+					break;
+			}
 		}
 	}
 

plugins/GSdx/GSTextureOGL.cpp

@@ -95,14 +95,10 @@ GSTextureOGL::GSTextureOGL(int type, int w, int h, bool msaa, int format)
 			// corollary we can maybe use it for multisample stuff
 		case GSTexture::Texture:
 		case GSTexture::RenderTarget:
+		case GSTexture::DepthStencil:
 			glGenTextures(1, &m_texture_id);
 			m_texture_target = GL_TEXTURE_2D;
 			break;
-		case GSTexture::DepthStencil:
-			glGenRenderbuffers(1, &m_texture_id);
-			m_texture_target = GL_RENDERBUFFER;
-			break;
-			break;
 		case GSTexture::Backbuffer:
 			m_texture_target = 0;
 			m_texture_id = 0;
@@ -123,8 +119,8 @@ GSTextureOGL::GSTextureOGL(int type, int w, int h, bool msaa, int format)
 	// Allocate the buffer
 	switch (m_type) {
 		case GSTexture::DepthStencil:
-			glBindRenderbuffer(m_texture_target, m_texture_id);
+			EnableUnit(1);
-			glRenderbufferStorageMultisample(m_texture_target, msaa_level, m_format, m_size.y, m_size.x);
+			glTexImage2D(m_texture_target, 0, m_format, w, h, 0, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV, NULL);
 			break;
 		case GSTexture::RenderTarget:
 		case GSTexture::Texture:
@@ -155,27 +151,11 @@ GSTextureOGL::GSTextureOGL(int type, int w, int h, bool msaa, int format)
 GSTextureOGL::~GSTextureOGL()
 {
 	glDeleteBuffers(1, &m_extra_buffer_id);
-	switch (m_type) {
-		case GSTexture::Texture:
-		case GSTexture::RenderTarget:
 	glDeleteTextures(1, &m_texture_id);
-			break;
-		case GSTexture::DepthStencil:
-			glDeleteRenderbuffers(1, &m_texture_id);
-			break;
-		case GSTexture::Offscreen:
-			assert(0);
-			break;
-		default:
-			break;
-	}
 }
 
 void GSTextureOGL::Attach(GLenum attachment)
 {
-	if (m_type == GSTexture::DepthStencil)
-		glFramebufferRenderbuffer(GL_FRAMEBUFFER, attachment, m_texture_target, m_texture_id);
-	else
 	glFramebufferTexture2D(GL_FRAMEBUFFER, attachment, m_texture_target, m_texture_id, 0);
 }
 
@@ -245,13 +225,7 @@ bool GSTextureOGL::Update(const GSVector4i& r, const void* data, int pitch)
 
 void GSTextureOGL::EnableUnit(uint unit)
 {
-	switch (m_type) {
+	if (!IsBackbuffer()) {
-		case GSTexture::DepthStencil:
-		case GSTexture::Offscreen:
-			assert(0);
-			break;
-		case GSTexture::RenderTarget:
-		case GSTexture::Texture:
 		// FIXME
 		// Howto allocate the texture unit !!!
 		// In worst case the HW renderer seems to use 3 texture unit
@@ -266,7 +240,6 @@ void GSTextureOGL::EnableUnit(uint unit)
 			g_state_texture_id = m_texture_id;
 			glBindTexture(m_texture_target, m_texture_id);
 		}
-			break;
 	}
 }
 
@@ -303,13 +276,6 @@ bool GSTextureOGL::Map(GSMap& m, const GSVector4i* r)
 
 	return false;
 #if 0
-	if(r != NULL)
-	{
-		// ASSERT(0); // not implemented
-
-		return false;
-	}
-
 	if(m_texture && m_desc.Usage == D3D11_USAGE_STAGING)
 	{
 		D3D11_MAPPED_SUBRESOURCE map;
@@ -370,24 +336,20 @@ struct BITMAPINFOHEADER
 
 bool GSTextureOGL::Save(const string& fn, bool dds)
 {
-	// Code not yet working
-	if (IsDss()) return false;
-
 	// Collect the texture data
 	uint32 pitch = 4 * m_size.x;
 	if (IsDss()) pitch *= 2;
 	char* image = (char*)malloc(pitch * m_size.y);
 
 	if (IsBackbuffer()) {
-		// TODO backbuffer
 		glReadBuffer(GL_BACK);
 		glReadPixels(0, 0, m_size.x, m_size.y, GL_RGBA, GL_UNSIGNED_BYTE, image);
 	} else if(IsDss()) {
-		Attach(GL_DEPTH_STENCIL_ATTACHMENT);
+		EnableUnit(1);
-		glGetTexImage(GL_TEXTURE_2D, 0, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV, image);
+		glGetTexImage(m_texture_target, 0, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV, image);
 	} else {
 		EnableUnit(0);
-		glGetTexImage(GL_TEXTURE_2D, 0, GL_RGBA, GL_UNSIGNED_BYTE, image);
+		glGetTexImage(m_texture_target, 0, GL_RGBA, GL_UNSIGNED_BYTE, image);
 	}
 
 	// Build a BMP file
@@ -430,8 +392,11 @@ bool GSTextureOGL::Save(const string& fn, bool dds)
 				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);
-					fwrite(&depth, 1, 4, fp);
+					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
@@ -443,7 +408,6 @@ bool GSTextureOGL::Save(const string& fn, bool dds)
 					fwrite(better_data, 1, 4, fp);
 				}
 			}
-			// fwrite(data, 1, m_size.x << 2, fp); // TODO: swap red-blue?
 		}
 
 		fclose(fp);
@@ -453,61 +417,5 @@ bool GSTextureOGL::Save(const string& fn, bool dds)
 	}
 
 	return false;
-#if 0
-	CComPtr<ID3D11Resource> res;
-
-	if(m_desc.BindFlags & D3D11_BIND_DEPTH_STENCIL)
-	{
-		HRESULT hr;
-
-		D3D11_TEXTURE2D_DESC desc;
-
-		memset(&desc, 0, sizeof(desc));
-
-		m_texture->GetDesc(&desc);
-
-		desc.Usage = D3D11_USAGE_STAGING;
-		desc.BindFlags = 0;
-		desc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
-
-		CComPtr<ID3D11Texture2D> src, dst;
-
-		hr = m_dev->CreateTexture2D(&desc, NULL, &src);
-
-		m_ctx->CopyResource(src, m_texture);
-
-		desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
-		desc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
-
-		hr = m_dev->CreateTexture2D(&desc, NULL, &dst);
-
-		D3D11_MAPPED_SUBRESOURCE sm, dm;
-
-		hr = m_ctx->Map(src, 0, D3D11_MAP_READ, 0, &sm);
-		hr = m_ctx->Map(dst, 0, D3D11_MAP_WRITE, 0, &dm);
-
-		uint8* s = (uint8*)sm.pData;
-		uint8* d = (uint8*)dm.pData;
-
-		for(uint32 y = 0; y < desc.Height; y++, s += sm.RowPitch, d += dm.RowPitch)
-		{
-			for(uint32 x = 0; x < desc.Width; x++)
-			{
-				((uint32*)d)[x] = (uint32)(ldexpf(((float*)s)[x*2], 32));
-			}
-		}
-
-		m_ctx->Unmap(src, 0);
-		m_ctx->Unmap(dst, 0);
-
-		res = dst;
-	}
-	else
-	{
-		res = m_texture;
-	}
-
-	return SUCCEEDED(D3DX11SaveTextureToFile(m_ctx, res, dds ? D3DX11_IFF_DDS : D3DX11_IFF_BMP, fn.c_str()));
-#endif
 }
 

plugins/GSdx/res/tfx.glsl

@@ -87,6 +87,9 @@ void vs_main()
 	
 	vec4 p = vec4(i_p, z, 0) - vec4(0.05f, 0.05f, 0, 0); 
 	vec4 final_p = p * VertexScale - VertexOffset;
+    // FIXME
+    // FLIP vertically
+    final_p.y *= -1.0f;
 
 	OUT.p = final_p;
     gl_Position = final_p; // NOTE I don't know if it is possible to merge POSITION_OUT and gl_Position
@@ -132,7 +135,10 @@ out gl_PerVertex {
     float gl_ClipDistance[];
 };
 
-layout(location = 0) in vertex GSin[];
+// FIXME
+// AMD Driver bug again !!!!
+//layout(location = 0) in vertex GSin[];
+in vertex GSin[];
 
 layout(location = 0) out vertex GSout;
 
@@ -188,46 +194,59 @@ void gs_main()
 
 #elif GS_PRIM == 3
 layout(lines) in;
-layout(triangle_strip, max_vertices = 4) out;
+layout(triangle_strip, max_vertices = 6) out;
 
 void gs_main()
 {
     // left top     => GSin[0];
     // right bottom => GSin[1];
+    vertex rb = GSin[1];
+    vertex lt = GSin[0];
 
-
+    lt.p.z = rb.p.z;
-    // left top
+    lt.t.zw = rb.t.zw;
-	GSout = GSin[0];
-
-    GSout.p.z = GSin[1].p.z;
-    GSout.t.zw = GSin[1].t.zw;
-    gl_Position = GSout.p; // FIXME is it useful
 	#if GS_IIP == 0
-	GSout.c = GSin[1].c;
+	lt.c = rb.c;
 	#endif
+
+    vertex lb = rb;
+	lb.p.x = lt.p.x;
+	lb.t.x = lt.t.x;
+
+    vertex rt = rb;
+	rt.p.y = lt.p.y;
+	rt.t.y = lt.t.y;
+
+    // Triangle 1
+    gl_Position = lt.p;
+    GSout = lt;
     EmitVertex();
 
-    // left bottom
+    gl_Position = lb.p;
-	GSout = GSin[1];
+    GSout = lb;
-    gl_Position = gl_in[1].gl_Position; // FIXME is it useful
-    gl_Position.x = GSin[0].p.x;
-    GSout.p.x = GSin[0].p.x;
-    GSout.t.x = GSin[0].t.x;
     EmitVertex();
 
-    // rigth top
+    gl_Position = rt.p;
-	GSout = GSin[1];
+    GSout = rt;
-    gl_Position = gl_in[1].gl_Position; // FIXME is it useful
-    gl_Position.y = GSin[0].p.y;
-    GSout.p.y = GSin[0].p.y;
-    GSout.t.y = GSin[0].t.y;
     EmitVertex();
 
-    // rigth bottom
+	EndPrimitive();
-	GSout = GSin[1];
+
-    gl_Position = GSin[1].p; // FIXME is it useful
+	// Triangle 2
+    gl_Position = lb.p;
+    GSout = lb;
     EmitVertex();
 
+    gl_Position = rt.p;
+    GSout = rt;
+    EmitVertex();
+
+    gl_Position = rb.p;
+    GSout = rb;
+    EmitVertex();
+
+    EndPrimitive();
+
 }
 
 #endif
@@ -238,13 +257,8 @@ void gs_main()
 layout(location = 0) in vertex PSin;
 
 // Same buffer but 2 colors for dual source blending
-//FIXME
-#if 1
 layout(location = 0, index = 0) out vec4 SV_Target0;
 layout(location = 0, index = 1) out vec4 SV_Target1;
-#else
-	layout(location = 0)		    out vec4 SV_Target;
-#endif
 
 layout(binding = 0) uniform sampler2D TextureSampler;
 layout(binding = 1) uniform sampler2D PaletteSampler;
@@ -264,7 +278,10 @@ layout(std140, binding = 5) uniform cb1
 
 vec4 sample_c(vec2 uv)
 {
+    // FIXME I'm not sure it is a good solution to flip texture
 	return texture(TextureSampler, uv);
+    //FIXME another way to FLIP vertically
+	//return texture(TextureSampler, vec2(uv.x, 1.0f-uv.y) );
 }
 
 vec4 sample_p(float u)
@@ -598,7 +615,6 @@ vec4 ps_color()
 void ps_main()
 {
 	//FIXME
-#if 1
 	vec4 c = ps_color();
 
     // FIXME: I'm not sure about the value of others field
@@ -620,8 +636,5 @@ void ps_main()
 
     //SV_Target0 = vec4(1.0f,0.0f,0.0f, 1.0f);
     //SV_Target1 = vec4(0.0f,1.0f,0.0f, 1.0f);
-#else
-    SV_Target = vec4(1.0f,0.0f,0.0f, 1.0f);
-#endif
 }
 #endif