Restrict 4gemm to PassThrough + bug fix

include/ck/tensor_operation/gpu/device/device_cgemm_4gemm_xdl_cshuffle.hpp

@@ -42,7 +42,8 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-template <typename ALayout,
+template <
+    typename ALayout,
     typename BLayout,
     typename CLayout,
     typename ADataType,
@@ -83,7 +84,12 @@ template <typename ALayout,
     index_t CShuffleNXdlPerWavePerShuffle,
     typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
     index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
-          LoopScheduler LoopSched = make_default_loop_scheduler()>
+    LoopScheduler LoopSched = make_default_loop_scheduler(),
+    enable_if_t<
+        is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+            is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+            is_same_v<CElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
+        bool> = false>
 struct DeviceCGemm_4Gemm_Xdl_CShuffle
     : public DeviceCGemm<AElementwiseOperation, BElementwiseOperation, CElementwiseOperation>
 {

include/ck/tensor_operation/gpu/device/device_gemm_dl.hpp

@@ -60,8 +60,8 @@ template <
     index_t CThreadTransferDstScalarPerVector,
     enable_if_t<
         is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
-            is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
-            is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
+            is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+            is_same_v<CElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
         bool> = false>
 struct DeviceGemmDl
     : public DeviceGemm<AElementwiseOperation, BElementwiseOperation, CElementwiseOperation>