purgatory/yuzu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

glsl: Clamp shared mem size to GL_MAX_COMPUTE_SHARED_MEMORY_SIZE

Browse Source
This commit is contained in:
ameerj 2021-07-12 02:03:25 -04:00
parent 8c166c68d4
commit 94af0a00f6
3 changed files with 12 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
src/shader_recompiler/backend/glsl/emit_glsl.cpp
View File

@ -218,8 +218,15 @@ std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info, IR
const std::string version{fmt::format("#version 450{}\n", GlslVersionSpecifier(ctx))}; const std::string version{fmt::format("#version 450{}\n", GlslVersionSpecifier(ctx))};
ctx.header.insert(0, version); ctx.header.insert(0, version);
if (program.shared_memory_size > 0) { if (program.shared_memory_size > 0) {
ctx.header += const auto requested_size{program.shared_memory_size};
fmt::format("shared uint smem[{}];", Common::DivCeil(program.shared_memory_size, 4U)); const auto max_size{profile.gl_max_compute_smem_size};
const bool needs_clamp{requested_size > max_size};
if (needs_clamp) {
LOG_WARNING(Shader_GLSL, "Requested shared memory size ({}) exceeds device limit ({})",
requested_size, max_size);
}
const auto smem_size{needs_clamp ? max_size : requested_size};
ctx.header += fmt::format("shared uint smem[{}];", Common::DivCeil(smem_size, 4U));
} }
ctx.header += "void main(){\n"; ctx.header += "void main(){\n";
if (program.local_memory_size > 0) { if (program.local_memory_size > 0) {

2
src/shader_recompiler/profile.h
View File

@ -67,6 +67,8 @@ struct Profile {
bool has_gl_precise_bug{}; bool has_gl_precise_bug{};
/// Ignores SPIR-V ordered vs unordered using GLSL semantics /// Ignores SPIR-V ordered vs unordered using GLSL semantics
bool ignore_nan_fp_comparisons{}; bool ignore_nan_fp_comparisons{};
u32 gl_max_compute_smem_size{};
}; };
} // namespace Shader } // namespace Shader

1
src/video_core/renderer_opengl/gl_shader_cache.cpp
View File

@ -211,6 +211,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
.has_gl_component_indexing_bug = device.HasComponentIndexingBug(), .has_gl_component_indexing_bug = device.HasComponentIndexingBug(),
.has_gl_precise_bug = device.HasPreciseBug(), .has_gl_precise_bug = device.HasPreciseBug(),
.ignore_nan_fp_comparisons = true, .ignore_nan_fp_comparisons = true,
.gl_max_compute_smem_size = device.GetMaxComputeSharedMemorySize(),
}, },
host_info{ host_info{
.support_float16 = false, .support_float16 = false,