[HotFix] add config and version files to pass on build info (#856)

* experiment with config file * experiment with version.h config * add more info to version.h * minor updates * minor updates * fix case where DTYPE is not used * large amount of files but minor changes * remove white space * minor changes to add more MACROs * fix cmakedefine01 * fix issue with CK internal conflict * fix define and define value * fix clang-format * fix formatting issue * experiment with cmake * clang format v12 to be consistent with miopen * avoid clang-format for config file
2026-05-03 13:11:25 +00:00 · 2023-08-23 11:36:17 -07:00
parent 350d64f351
commit c8a8385fdd
116 changed files with 709 additions and 456 deletions
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -588,14 +588,18 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
        LoopSched>;

    // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;

    // block-to-e-tile map
    using Block2ETileMap =
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
@@ -378,13 +378,16 @@ struct DeviceBatchedGemmEPermuteXdl : public DeviceBatchedGemmEPermute<ALayout,
        CDEBlockTransferScalarPerVector_NPerBlock,
        LoopSched>;

-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;

-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}));
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}));
    using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;

    // Argument
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -368,14 +368,18 @@ struct DeviceBatchedGemmMultiD_Xdl : public DeviceBatchedGemmMultiD<ALayout,
        LoopSched>;

    // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;

    // block-to-e-tile map
    using Block2ETileMap =
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -510,12 +510,15 @@ struct DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
        CDE1ShuffleBlockTransferScalarPerVector_NPerBlock,
        LoopSched>;

-    using A0GridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultA0GridDescriptor_AK0_M_AK1(A0GridDesc_M_K{}))>;
-    using B0GridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultB0GridDescriptor_BK0_N_BK1(B0GridDesc_N_K{}))>;
-    using B1GridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultB1GridDescriptor_BK0_N_BK1(B1GridDesc_N_K{}))>;
+    using A0GridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultA0GridDescriptor_AK0_M_AK1(
+            A0GridDesc_M_K{}))>;
+    using B0GridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultB0GridDescriptor_BK0_N_BK1(
+            B0GridDesc_N_K{}))>;
+    using B1GridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultB1GridDescriptor_BK0_N_BK1(
+            B1GridDesc_N_K{}))>;

    // Argument
    struct Argument : public BaseArgument
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -355,14 +355,18 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
        LoopSched>;

    // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;

    // block-to-e-tile map
    using Block2ETileMap =
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -364,11 +364,13 @@ struct DeviceGemmMultipleDLayernorm_Xdl_CShuffle
    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
    // We have to separate mean var descriptor for gemm and layernorm bacause of different grid
    // layout(different padding)
-    using GemmMeanVarGridDesc_M_NBlock = decltype(
-        MakeMeanVarDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1, 1));
+    using GemmMeanVarGridDesc_M_NBlock =
+        decltype(MakeMeanVarDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1,
+                                                                                                1));

-    using GemmCountGridDesc_M_NBlock = decltype(
-        MakeCountDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1, 1));
+    using GemmCountGridDesc_M_NBlock =
+        decltype(MakeCountDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1,
+                                                                                              1));

    using LayernormMeanVarGridDesc_M_NBlock =
        decltype(MakeMeanVarDescriptor_M_N<Sequence<true, true>,
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -337,10 +337,12 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
        RThreadTransferDstScalarPerVector_MPerBlock,
        LoopSched>;

-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;

    using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;

--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -288,14 +288,18 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
        PipelineVer>;

    // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;

    // block-to-e-tile map
    using Block2ETileMap =
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -248,10 +248,12 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle : public DeviceGemm<ALayout,
        CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
        CShuffleBlockTransferScalarPerVector_NPerBlock>;

-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;

    using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;

--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -400,14 +400,18 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
        LoopSched>;

    // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;

    struct GroupedContractionBlock2ETileMap
    {
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -422,10 +422,12 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
    using AGridDesc_M_K = decltype(transform_k0_m_k1_to_m_k(AGridDesc_AK0_M_AK1{}));
    using BGridDesc_N_K = decltype(transform_k0_m_k1_to_m_k(BGridDesc_BK0_N_BK1{}));

-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}));
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}));
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}));
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}));

    // block-to-e-tile map
    using Block2ETileMap =
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -381,8 +381,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
    }

    // desc for problem definition
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        MakeAGridDescriptor_AK0_M_AK1<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(MakeAGridDescriptor_AK0_M_AK1<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
    using BGridDesc_BK0_N_BK1 =
        remove_cvref_t<decltype(MakeBGridDescriptor_BK0_N_BK1<BLayout>({}, {}))>;
    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -320,8 +320,8 @@ struct DeviceGroupedConvFwdDl_NHWC_KYXC_NHWK : public DeviceGroupedConvFwd<NDimS
    }

    // desc for problem definition
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        MakeAGridDescriptor_AK0_M_AK1<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(MakeAGridDescriptor_AK0_M_AK1<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
    using BGridDesc_BK0_N_BK1 =
        remove_cvref_t<decltype(MakeBGridDescriptor_BK0_N_BK1<BLayout>({}, {}))>;
    using CGridDesc_M_N = remove_cvref_t<decltype(MakeCGridDescriptor_M_N<CLayout>({}, {}))>;
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -446,8 +446,8 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
        return GetPaddedRGridDescriptor(r_grid_desc_mraw, NHoWo);
    }

-    using AGridDesc_M_K = remove_cvref_t<decltype(
-        MakeAGridDescriptor_M_K<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_M_K = remove_cvref_t<decltype(MakeAGridDescriptor_M_K<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
    using BGridDesc_N_K = remove_cvref_t<decltype(MakeBGridDescriptor_N_K<BLayout>({}, {}))>;
    using EGridDesc_M_N = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<DELayout>({}, {}))>;
    using RGridDesc_M   = remove_cvref_t<decltype(MakeRGridDescriptor_M<RLayout>({}, {}))>;
@@ -507,10 +507,12 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
        RThreadTransferDstScalarPerVector_MPerBlock,
        LoopSched>;

-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;

    using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;

--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
@@ -245,8 +245,8 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
    }

    // desc for problem definition
-    using AGridDesc_M_K  = remove_cvref_t<decltype(
-        MakeAGridDescriptor_M_K<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_M_K  = remove_cvref_t<decltype(MakeAGridDescriptor_M_K<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
    using BGridDesc_N_K  = remove_cvref_t<decltype(MakeBGridDescriptor_N_K<BLayout>({}, {}))>;
    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
    using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -361,8 +361,8 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
    }

    // desc for problem definition
-    using AGridDesc_M_K  = remove_cvref_t<decltype(
-        MakeAGridDescriptor_M_K<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_M_K  = remove_cvref_t<decltype(MakeAGridDescriptor_M_K<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
    using BGridDesc_N_K  = remove_cvref_t<decltype(MakeBGridDescriptor_N_K<BLayout>({}, {}))>;
    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
    using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
@@ -412,14 +412,18 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
        LoopSched>;

    // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;

    // block-to-e-tile map
    using Block2ETileMap =
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -735,12 +735,12 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
            }

            // Check vector load/store requirement
-            const auto a_stride_lowest = ABlockTransferSrcVectorDim == 2
-                                             ? device_arg.a_mz_kz_strides_[1]
-                                             : device_arg.a_mz_kz_strides_[0];
-            const auto b_stride_lowest = BBlockTransferSrcVectorDim == 2
-                                             ? device_arg.b_nz_kz_strides_[1]
-                                             : device_arg.b_nz_kz_strides_[0];
+            const auto a_stride_lowest  = ABlockTransferSrcVectorDim == 2
+                                              ? device_arg.a_mz_kz_strides_[1]
+                                              : device_arg.a_mz_kz_strides_[0];
+            const auto b_stride_lowest  = BBlockTransferSrcVectorDim == 2
+                                              ? device_arg.b_nz_kz_strides_[1]
+                                              : device_arg.b_nz_kz_strides_[0];
            const auto b1_stride_lowest = B1BlockTransferSrcVectorDim == 2
                                              ? device_arg.b1_nz_kz_strides_[1]
                                              : device_arg.b1_nz_kz_strides_[0];
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -272,14 +272,18 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
        CDEBlockTransferScalarPerVector_NPerBlock,
        LoopSched>;

-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;

    struct GroupedGemmBlock2ETileMap
    {
--- a/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -617,10 +617,12 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
        CDEBlockTransferScalarPerVector_NPerBlock,
        LoopSched>;

-    using AGridDesc_AKB_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AKB_AK0_M_AK1(AGridDesc_M_K{}, 1))>;
-    using BGridDesc_BKB_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BKB_BK0_N_BK1(BGridDesc_N_K{}, 1))>;
+    using AGridDesc_AKB_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AKB_AK0_M_AK1(
+            AGridDesc_M_K{}, 1))>;
+    using BGridDesc_BKB_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BKB_BK0_N_BK1(
+            BGridDesc_N_K{}, 1))>;

    using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;