argument supported

a6b30591 · root · 3c3a3753 · a6b30591
Commit a6b30591 authored Dec 30, 2024 by root
Show whitespace changes
Inline Side-by-side

Showing with 4 additions and 1 deletion

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp ...n/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp +4 -1

No files found.
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp
@@ -469,7 +469,10 @@ struct DeviceGemm_Xdl_CShuffle_Streamk_V3 : public DeviceGemm_Streamk_V2<ALayout
        {
            return false;
        }
-
+        if(!is_bf16_atomic_supported() && std::is_same_v<CDataType, ck::bhalf_t> && arg.Streamk_sel > 0)
+        {
+            return false;
+        }
        if((arg.K % AK1 != 0 || arg.K % BK1 != 0) && !(GemmSpec == GemmSpecialization::MKPadding ||
                                                       GemmSpec == GemmSpecialization::NKPadding ||
                                                       GemmSpec == GemmSpecialization::MNKPadding ||