gl_arb_decompiler: Implement an assembly shader decompiler

Emit code compatible with NV_gpu_program5.
This should emit code compatible with Fermi, but it wasn't tested on
that architecture. Pascal has some issues not present on Turing GPUs.

src/video_core/CMakeLists.txt

@@ -54,6 +54,8 @@ add_library(video_core STATIC
     rasterizer_interface.h
     renderer_base.cpp
     renderer_base.h
+    renderer_opengl/gl_arb_decompiler.cpp
+    renderer_opengl/gl_arb_decompiler.h
     renderer_opengl/gl_buffer_cache.cpp
     renderer_opengl/gl_buffer_cache.h
     renderer_opengl/gl_device.cpp

src/video_core/renderer_opengl/gl_arb_decompiler.h

@@ -0,0 +1,29 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <string>
+#include <string_view>
+
+#include "common/common_types.h"
+
+namespace Tegra::Engines {
+enum class ShaderType : u32;
+}
+
+namespace VideoCommon::Shader {
+class ShaderIR;
+class Registry;
+} // namespace VideoCommon::Shader
+
+namespace OpenGL {
+
+class Device;
+
+std::string DecompileAssemblyShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
+                                    const VideoCommon::Shader::Registry& registry,
+                                    Tegra::Engines::ShaderType stage, std::string_view identifier);
+
+} // namespace OpenGL

src/video_core/renderer_opengl/gl_device.cpp

@@ -213,6 +213,7 @@ Device::Device()
     has_component_indexing_bug = is_amd;
     has_precise_bug = TestPreciseBug();
     has_fast_buffer_sub_data = is_nvidia && !disable_fast_buffer_sub_data;
+    has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2;
     use_assembly_shaders = Settings::values.use_assembly_shaders && GLAD_GL_NV_gpu_program5 &&
                            GLAD_GL_NV_compute_program5 && GLAD_GL_NV_transform_feedback &&
                            GLAD_GL_NV_transform_feedback2;

src/video_core/renderer_opengl/gl_device.h

@@ -88,6 +88,10 @@ public:
         return has_fast_buffer_sub_data;
     }
 
+    bool HasNvViewportArray2() const {
+        return has_nv_viewport_array2;
+    }
+
     bool UseAssemblyShaders() const {
         return use_assembly_shaders;
     }
@@ -111,6 +115,7 @@ private:
     bool has_component_indexing_bug{};
     bool has_precise_bug{};
     bool has_fast_buffer_sub_data{};
+    bool has_nv_viewport_array2{};
     bool use_assembly_shaders{};
 };
 

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -20,6 +20,7 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/memory_manager.h"
+#include "video_core/renderer_opengl/gl_arb_decompiler.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
@@ -147,7 +148,8 @@ ProgramSharedPtr BuildShader(const Device& device, ShaderType shader_type, u64 u
     auto program = std::make_shared<ProgramHandle>();
 
     if (device.UseAssemblyShaders()) {
-        const std::string arb = "Not implemented";
+        const std::string arb =
+            DecompileAssemblyShader(device, ir, registry, shader_type, shader_id);
 
         GLuint& arb_prog = program->assembly_program.handle;