diff options
Diffstat (limited to 'src/utilities/compile.cpp')
| -rw-r--r-- | src/utilities/compile.cpp | 5 |
1 files changed, 5 insertions, 0 deletions
diff --git a/src/utilities/compile.cpp b/src/utilities/compile.cpp index c1d921a4..65131cca 100644 --- a/src/utilities/compile.cpp +++ b/src/utilities/compile.cpp @@ -57,6 +57,11 @@ Program CompileFromSource(const std::string &source_string, const Precision prec header_string += "#define GLOBAL_MEM_FENCE 1\n"; } + // For Intel GPUs with subgroup support, use subgroup shuffling. + if (device.IsGPU() && device.HasExtension(kKhronosIntelSubgroups)) { + header_string += "#define USE_SUBGROUP_SHUFFLING 1\n"; + } + // Optionally adds a translation header from OpenCL kernels to CUDA kernels #ifdef CUDA_API header_string += |