diff options
| author | degasus <wickmarkus@web.de> | 2023-01-31 18:10:48 +0100 |
|---|---|---|
| committer | degasus <wickmarkus@web.de> | 2023-02-09 13:27:02 +0100 |
| commit | 4b2aa948e662aa037ad39f5ee81a64256a4d61aa (patch) | |
| tree | c6bc567e6ee91b30e4317d726f0643779f916745 /Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp | |
| parent | aaad0cd39fca1ad020864ad7a141052179e7feb4 (diff) | |
VideoBackend/OGL: Prefer KHR_subgroup over NV_shader_thread.
While the NV extension is totally fine, the KHR extension should be able to support more hardware.
For NVIDIA, the hardware either supports both or neither, it just needs a driver from the last two years.
For AMD, the drivers from late 2022-12 seems to bring support for the KHR extension.
For Intel, the KHR is also supported for some years.
Diffstat (limited to 'Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp')
| -rw-r--r-- | Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp | 27 |
1 files changed, 11 insertions, 16 deletions
diff --git a/Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp b/Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp index 289e99cae0..eff1497e7f 100644 --- a/Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp +++ b/Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp @@ -78,6 +78,8 @@ static std::string GetGLSLVersionString() return "#version 400"; case Glsl430: return "#version 430"; + case Glsl450: + return "#version 450"; default: // Shouldn't ever hit this return "#version ERROR"; @@ -720,25 +722,18 @@ void ProgramShaderCache::CreateHeader() } std::string shader_shuffle_string; - if (g_ogl_config.bSupportsShaderThreadShuffleNV) + if (g_ogl_config.bSupportsKHRShaderSubgroup) { shader_shuffle_string = R"( -#extension GL_NV_shader_thread_group : enable -#extension GL_NV_shader_thread_shuffle : enable -#define SUPPORTS_SUBGROUP_REDUCTION 1 +#extension GL_KHR_shader_subgroup_basic : enable +#extension GL_KHR_shader_subgroup_arithmetic : enable +#extension GL_KHR_shader_subgroup_ballot : enable -// The xor shuffle below produces incorrect results if all threads in a warp are not active. -#define CAN_USE_SUBGROUP_REDUCTION (ballotThreadNV(true) == 0xFFFFFFFFu) - -#define IS_HELPER_INVOCATION gl_HelperThreadNV -#define IS_FIRST_ACTIVE_INVOCATION (gl_ThreadInWarpNV == findLSB(ballotThreadNV(!gl_HelperThreadNV))) -#define SUBGROUP_REDUCTION(func, value) value = func(value, shuffleXorNV(value, 16, 32)); \ - value = func(value, shuffleXorNV(value, 8, 32)); \ - value = func(value, shuffleXorNV(value, 4, 32)); \ - value = func(value, shuffleXorNV(value, 2, 32)); \ - value = func(value, shuffleXorNV(value, 1, 32)); -#define SUBGROUP_MIN(value) SUBGROUP_REDUCTION(min, value) -#define SUBGROUP_MAX(value) SUBGROUP_REDUCTION(max, value) +#define SUPPORTS_SUBGROUP_REDUCTION 1 +#define IS_HELPER_INVOCATION gl_HelperInvocation +#define IS_FIRST_ACTIVE_INVOCATION (subgroupElect()) +#define SUBGROUP_MIN(value) value = subgroupMin(value) +#define SUBGROUP_MAX(value) value = subgroupMax(value) )"; } |
