Only call iglp_opt(1) on certain arch (#1195)

bb26118c · Po Yen Chen · GitHub · 62db01c2 · bb26118c
Unverified Commit bb26118c authored Mar 12, 2024 by Po Yen Chen Committed by GitHub Mar 12, 2024
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp ...k/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp +2 -0

No files found.
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
@@ -80,7 +80,9 @@ struct GridwiseGemmPipeline_v2
            do
            {
 #if CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT
+#if defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
                __builtin_amdgcn_iglp_opt(CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT);
+#endif
 #endif

                block_sync_lds();