From bd24fa97138ff1e33a7f8d3c30a4f4482a6482a8 Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Tue, 25 May 2021 19:55:40 -0400 Subject: [PATCH] glsl: Query GL Device for FP16 extension support --- src/shader_recompiler/backend/glsl/emit_context.cpp | 9 +++++++-- src/shader_recompiler/profile.h | 2 ++ src/video_core/renderer_opengl/gl_device.cpp | 2 ++ src/video_core/renderer_opengl/gl_device.h | 10 ++++++++++ src/video_core/renderer_opengl/gl_shader_cache.cpp | 2 ++ 5 files changed, 23 insertions(+), 2 deletions(-) diff --git a/src/shader_recompiler/backend/glsl/emit_context.cpp b/src/shader_recompiler/backend/glsl/emit_context.cpp index 9c3fd44ba..6f769fa10 100644 --- a/src/shader_recompiler/backend/glsl/emit_context.cpp +++ b/src/shader_recompiler/backend/glsl/emit_context.cpp @@ -5,6 +5,7 @@ #include "shader_recompiler/backend/bindings.h" #include "shader_recompiler/backend/glsl/emit_context.h" #include "shader_recompiler/frontend/ir/program.h" +#include "shader_recompiler/profile.h" namespace Shader::Backend::GLSL { @@ -40,8 +41,12 @@ void EmitContext::SetupExtensions(std::string& header) { header += "#extension NV_shader_atomic_fp16_vector : enable\n"; } if (info.uses_fp16) { - // TODO: AMD - header += "#extension GL_NV_gpu_shader5 : enable\n"; + if (profile.support_gl_nv_gpu_shader_5) { + header += "#extension GL_NV_gpu_shader5 : enable\n"; + } + if (profile.support_gl_amd_gpu_shader_half_float) { + header += "#extension GL_AMD_gpu_shader_half_float : enable\n"; + } } } diff --git a/src/shader_recompiler/profile.h b/src/shader_recompiler/profile.h index 3109fb69c..5d269368a 100644 --- a/src/shader_recompiler/profile.h +++ b/src/shader_recompiler/profile.h @@ -83,6 +83,8 @@ struct Profile { bool support_demote_to_helper_invocation{}; bool support_int64_atomics{}; bool support_derivative_control{}; + bool support_gl_nv_gpu_shader_5{}; + bool support_gl_amd_gpu_shader_half_float{}; bool warp_size_potentially_larger_than_guest{}; diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index 3f7929f9e..071133781 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -158,6 +158,8 @@ Device::Device() { has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory; has_debugging_tool_attached = IsDebugToolAttached(extensions); has_depth_buffer_float = HasExtension(extensions, "GL_NV_depth_buffer_float"); + has_nv_gpu_shader_5 = GLAD_GL_NV_gpu_shader5; + has_amd_shader_half_float = GLAD_GL_AMD_gpu_shader_half_float; // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive // uniform buffers as "push constants" diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index 1ffd24883..9b9402c29 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -120,6 +120,14 @@ public: return has_depth_buffer_float; } + bool HasNvGpuShader5() const { + return has_nv_gpu_shader_5; + } + + bool HasAmdShaderHalfFloat() const { + return has_amd_shader_half_float; + } + private: static bool TestVariableAoffi(); static bool TestPreciseBug(); @@ -151,6 +159,8 @@ private: bool use_asynchronous_shaders{}; bool use_driver_cache{}; bool has_depth_buffer_float{}; + bool has_nv_gpu_shader_5{}; + bool has_amd_shader_half_float{}; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 602cf025b..e00d01e34 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -217,6 +217,8 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo .support_demote_to_helper_invocation = false, .support_int64_atomics = false, .support_derivative_control = device.HasDerivativeControl(), + .support_gl_nv_gpu_shader_5 = device.HasNvGpuShader5(), + .support_gl_amd_gpu_shader_half_float = device.HasAmdShaderHalfFloat(), .warp_size_potentially_larger_than_guest = true,