gl: Fall back to renderpass decoder on ATI drivers

rpcs3/Emu/RSX/GL/GLOverlays.cpp

@@ -631,13 +631,17 @@ namespace gl
 		#include "../Program/GLSLSnippets/CopyBufferToGenericImage.glsl"
 		;
 
-		const bool stencil_export_supported = gl::get_driver_caps().ARB_shader_stencil_export_supported;
+		const auto& caps = gl::get_driver_caps();
+		const bool stencil_export_supported = caps.ARB_shader_stencil_export_supported;
+		const bool legacy_format_support = caps.subvendor_ATI;
+
 		std::pair<std::string_view, std::string> repl_list[] =
 		{
 			{ "%set, ", "" },
 			{ "%loc", std::to_string(GL_COMPUTE_BUFFER_SLOT(0)) },
 			{ "%push_block", fmt::format("binding=%d, std140", GL_COMPUTE_BUFFER_SLOT(1)) },
-			{ "%stencil_export_supported", stencil_export_supported ? "1" : "0" }
+			{ "%stencil_export_supported", stencil_export_supported ? "1" : "0" },
+			{ "%legacy_format_support", legacy_format_support ? "1" : "0" }
 		};
 
 		fs_src = fmt::replace_all(fs_src, repl_list);

rpcs3/Emu/RSX/GL/GLTexture.cpp

@@ -384,6 +384,7 @@ namespace gl
 		bool skip_barrier = false;
 		u32 in_offset = static_cast<u32>(reinterpret_cast<u64>(src_offset));
 		u32 out_offset = in_offset;
+		const auto& caps = gl::get_driver_caps();
 
 		auto initialize_scratch_mem = [&]()
 		{
@@ -409,7 +410,6 @@ namespace gl
 			transfer_buf = &scratch_mem;
 		};
 
-		const auto caps = gl::get_driver_caps();
 		if ((dst->aspect() & image_aspect::stencil) == 0 || caps.ARB_shader_stencil_export_supported)
 		{
 			// We do not need to use the driver's builtin transport mechanism
@@ -469,18 +469,24 @@ namespace gl
 			}
 
 			// If possible, decode using a compute transform to potentially have asynchronous scheduling
-			bool use_compute_transform = (dst->aspect() == gl::image_aspect::color);
+			bool use_compute_transform = (
-			switch (dst->get_internal_format())
+				dst->aspect() == gl::image_aspect::color &&  // Cannot use image_load_store with depth images
+				caps.subvendor_ATI == false);                // The old AMD/ATI driver does not support image writeonly without format specifier
+
+			if (use_compute_transform)
 			{
-			case texture::internal_format::bgr5a1:
+				switch (dst->get_internal_format())
-			case texture::internal_format::rgb5a1:
+				{
-			case texture::internal_format::rgb565:
+				case texture::internal_format::bgr5a1:
-			case texture::internal_format::rgba4:
+				case texture::internal_format::rgb5a1:
-				// Packed formats are a problem with image_load_store
+				case texture::internal_format::rgb565:
-				use_compute_transform = false;
+				case texture::internal_format::rgba4:
-				break;
+					// Packed formats are a problem with image_load_store
-			default:
+					use_compute_transform = false;
-				break;
+					break;
+				default:
+					break;
+				}
 			}
 
 			if (use_compute_transform)

rpcs3/Emu/RSX/GL/glutils/capabilities.hpp

@@ -32,6 +32,7 @@ namespace gl
 		bool vendor_MESA = false;   // requires CLIENT_STORAGE bit set for streaming buffers
 		bool subvendor_RADEONSI = false;
 		bool subvendor_NOUVEAU = false;
+		bool subvendor_ATI = false;
 
 		bool check(const std::string& ext_name, const char* test)
 		{
@@ -230,6 +231,7 @@ namespace gl
 			else if (vendor_string.find("amd") != umax || vendor_string.find("ati") != umax)
 			{
 				vendor_AMD = true;
+				subvendor_ATI = vendor_string.find("ati") != umax;
 			}
 #endif
 

rpcs3/Emu/RSX/Program/GLSLSnippets/CopyBufferToGenericImage.glsl

@@ -3,12 +3,27 @@ R"(
 #extension GL_ARB_shader_stencil_export : enable
 
 #define ENABLE_DEPTH_STENCIL_LOAD %stencil_export_supported
+#define LEGACY_FORMAT_SUPPORT %legacy_format_support
 
 #define FMT_GL_DEPTH_COMPONENT16      0x81A5
 #define FMT_GL_DEPTH_COMPONENT32F     0x8CAC
 #define FMT_GL_DEPTH24_STENCIL8       0x88F0
 #define FMT_GL_DEPTH32F_STENCIL8      0x8CAD
 
+#if LEGACY_FORMAT_SUPPORT
+  #define FMT_GL_RGBA8                  0x8058
+  #define FMT_GL_BGRA8                  0x80E1
+  #define FMT_GL_R8                     0x8229
+  #define FMT_GL_R16                    0x822A
+  #define FMT_GL_R32F                   0x822E
+  #define FMT_GL_RG8                    0x822B
+  #define FMT_GL_RG8_SNORM              0x8F95
+  #define FMT_GL_RG16                   0x822C
+  #define FMT_GL_RG16F                  0x822F
+  #define FMT_GL_RGBA16F                0x881A
+  #define FMT_GL_RGBA32F                0x8814
+#endif
+
 #define FMT_GL_RGB565                 0x8D62
 #define FMT_GL_RGB5_A1                0x8057
 #define FMT_GL_BGR5_A1                0x99F0
@@ -144,6 +159,45 @@ void main()
 		gl_FragStencilRefARB = int(utmp2.y);
 		break;
 
+#endif
+
+#if LEGACY_FORMAT_SUPPORT
+
+	// Simple color. Provided for compatibility with old drivers.
+	case FMT_GL_RGBA8:
+		outColor = readFixed8x4(invocation_id);
+		break;
+	case FMT_GL_BGRA8:
+		outColor = readFixed8x4(invocation_id).bgra;
+		break;
+	case FMT_GL_R8:
+		outColor.r = readFixed8(invocation_id);
+		break;
+	case FMT_GL_R16:
+		outColor.r = readFixed16(invocation_id);
+		break;
+	case FMT_GL_R32F:
+		outColor.r = readFloat32(invocation_id);
+		break;
+	case FMT_GL_RG8:
+		outColor.rg = readFixed8x2(invocation_id);
+		break;
+	case FMT_GL_RG8_SNORM:
+		outColor.rg = readFixed8x2Snorm(invocation_id);
+		break;
+	case FMT_GL_RG16:
+		outColor.rg = readFixed16x2(invocation_id);
+		break;
+	case FMT_GL_RG16F:
+		outColor.rg = readFloat16x2(invocation_id);
+		break;
+	case FMT_GL_RGBA16F:
+		outColor = readFloat16x4(invocation_id);
+		break;
+	case FMT_GL_RGBA32F:
+		outColor = readFloat32x4(invocation_id);
+		break;
+
 #endif
 
 	// Packed color