Merge branch 'develop' of...

Merge branch 'develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into e2e_kernellib

include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp

@@ -822,7 +822,14 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp

@@ -956,7 +956,19 @@ struct
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp

@@ -913,7 +913,19 @@ struct DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

@@ -880,7 +880,17 @@ struct DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_W
             << MPerBlock << ", "
             << NPerBlock << ", "
             << K0PerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << K1 << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp

@@ -720,7 +720,16 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
             << MPerBlock << ", "
             << NPerBlock << ", "
             << K0PerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp

@@ -630,7 +630,19 @@ struct DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp

@@ -1567,7 +1567,8 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Dl
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1
             << ">";
         if constexpr(ConvBackwardDataSpecialization ==
                      ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0){

include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp

@@ -1552,7 +1552,14 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">";
         if constexpr(ConvBackwardDataSpecialization ==
                      ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0){

include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp

@@ -862,7 +862,15 @@ struct DeviceGemmBiasAddReduce_Xdl_CShuffle : public DeviceGemmReduce<1, ReduceO
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp

@@ -559,7 +559,19 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp

@@ -51,7 +51,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
-    defined(__gfx1030__))
+    defined(__gfx90a__) || defined(__gfx1030__))
 
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);
@@ -552,7 +552,7 @@ struct DeviceGemmMultipleD_Dl : public DeviceGemmMultipleD<ALayout,
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
-           ck::get_device_name() == "gfx1030")
+           ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030")
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_);

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp

@@ -669,7 +669,15 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
             << KPerBlock << ", "
             << AK1 << ", "
             << BK1 << ", "
-            << getGemmSpecializationString(GemmSpec)
+            << getGemmSpecializationString(GemmSpec) << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp

@@ -166,7 +166,7 @@ struct DeviceGemmMultipleD_Wmma_CShuffle : public DeviceGemmMultipleD<ALayout,
 
                 return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
             }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
             {
                 const auto b_grid_desc_nraw_kraw =
                     make_naive_tensor_descriptor(make_tuple(NRaw, KRaw), make_tuple(StrideB, I1));
@@ -531,7 +531,8 @@ struct DeviceGemmMultipleD_Wmma_CShuffle : public DeviceGemmMultipleD<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(ck::get_device_name() == "gfx1100")
+        if(ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
             {

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp

@@ -680,6 +680,14 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
             << KPerBlock << ", "
             << AK1 << ", "
             << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
             << getGemmSpecializationString(GemmSpec)
             << ">"
             << " LoopScheduler: "

include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp

@@ -822,7 +822,15 @@ struct DeviceGemmReduce_Xdl_CShuffle : public DeviceGemmReduce<0, ReduceOperatio
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle 
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp

@@ -158,7 +158,7 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
 
                 return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
             }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
             {
                 const auto b_grid_desc_nraw_kraw =
                     make_naive_tensor_descriptor(make_tuple(NRaw, KRaw), make_tuple(StrideB, I1));
@@ -454,7 +454,8 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(ck::get_device_name() == "gfx1100")
+        if(ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
             {

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp

@@ -77,8 +77,6 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
 
     static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
     {
-        assert(K % K1 == 0);
-
         const index_t K0 = K / K1;
 
         const auto a_grid_desc_m_k = [&]() {
@@ -116,8 +114,6 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
 
     static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
     {
-        assert(K % K1 == 0);
-
         const index_t K0 = K / K1;
 
         const auto b_grid_desc_k_n = [&]() {
@@ -551,7 +547,11 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
             << MPerXDL << ", "
             << NPerXDL << ", "
             << MXdlPerWave << ", "
-            << NXdlPerWave
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">"
             << " NumPrefetch: "
             << NumPrefetch << ", "

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp

@@ -682,7 +682,15 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">"
             << " LoopScheduler: "
             << LoopSchedToString[LoopSched] << ", "

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp

@@ -760,7 +760,15 @@ struct DeviceGemmLayerNorm_Xdl_CShuffle : public BaseOperator
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp

@@ -640,7 +640,16 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">";
         // clang-format on