fixes

include/ck/tensor_description/multi_index_transform.hpp

@@ -1553,6 +1553,14 @@ struct UnMerge
     }
 };
 
+/**
+ * @brief Transformation struct for convolution backward data output indices to GEMM indices.
+ *
+ * This struct is responsible for mapping the output tensor indices (N, Ho, Wo, K) from the
+ * convolution backward data operation to the corresponding indices (K0, M, K1) used in the
+ * implicit GEMM computation. It encapsulates the necessary parameters and transformation logic
+ * required to efficiently perform the index conversion.
+ */
 struct ConvBwdDataImplicitGemmOutTransform
 {
     static constexpr auto I0 = Number<0>{};

include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp

@@ -13,6 +13,12 @@
 namespace ck {
 namespace tensor_operation {
 
+/**
+ * @brief Enable custom tensor transform for convolution backward data output.
+ *
+ * When set to 1, this macro enables a custom transformation of the output tensor
+ * in convolution backward data operations.
+ */
 #define CK_USE_CUSTOM_TENSOR_TRANSFORM_FOR_BWD_DATA_OUT 1
 
 template <

profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp

@@ -137,8 +137,8 @@ bool profile_grouped_conv_bwd_data_impl(int do_verification,
 
             auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
-            float avg_time = invoker_ptr->Run(argument_ptr.get(),
-                                              StreamConfig{nullptr, time_kernel, 0, 500, 100});
+            float avg_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
 
             std::size_t flop      = conv_param.GetFlops();
             std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();