add a marco to turn off denorm fix by default

include/ck/ck.hpp

@@ -168,6 +168,9 @@
 // flag to enable (1) or disable (0) the debugging output in some kernels
 #define DEBUG_LOG 0
 
+// denorm test fix, required to work around dissue
+#define CK_WORKAROUND_DENORM_FIX 0
+
 namespace ck {
 
 enum struct InMemoryDataOperationEnum

include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp

@@ -96,7 +96,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
     // when mfma if fixed, remove this section and update
     // ABDataTypeAdjusted -> ABDataType throughout this file
-#if defined(__gfx90a__)
+#if CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
     using ABDataTypeAdjusted =
         conditional_t<is_same_v<ABDataType, ck::half_t>, ck::bhalf_t, ABDataType>;
 #else