Merge branch 'develop' into transpose_5d

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp

@@ -9,18 +9,18 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              Empty_Tuple,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances)
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                Empty_Tuple,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                Empty_Tuple,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
                                    device_grouped_conv_fwd_xdl_int8_instances<3,

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/CMakeLists.txt

+set(GROUPED_CONV3D_FWD_SCALEADD_AB
+   xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp)
+
+add_instance_library(device_grouped_conv3d_fwd_scaleadd_ab_instance ${GROUPED_CONV3D_FWD_SCALEADD_AB})

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_scaleadd_ab_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<BF16, BF16>,
+                                                                ck::Tuple<BF16, BF16>,
+                                                                ck::Tuple<>,
+                                                                BF16,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_bf16_instances<3,
+                                                               NDHWGC,
+                                                               GKZYXC,
+                                                               NDHWGK,
+                                                               ConvFwdDefault>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_bf16_instances<3,
+                                                               NDHWGC,
+                                                               GKZYXC,
+                                                               NDHWGK,
+                                                               ConvFwd1x1P0>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_bf16_instances<3,
+                                                               NDHWGC,
+                                                               GKZYXC,
+                                                               NDHWGK,
+                                                               ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_scaleadd_ab_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<F16, F16>,
+                                                                ck::Tuple<F16, F16>,
+                                                                ck::Tuple<>,
+                                                                F16,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_f16_instances<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              NDHWGK,
+                                                              ConvFwdDefault>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_f16_instances<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              NDHWGK,
+                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_f16_instances<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              NDHWGK,
+                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_scaleadd_ab_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<F32, F32>,
+                                                                ck::Tuple<F32, F32>,
+                                                                ck::Tuple<>,
+                                                                F32,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_f32_instances<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              NDHWGK,
+                                                              ConvFwdDefault>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_f32_instances<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              NDHWGK,
+                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_f32_instances<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              NDHWGK,
+                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_scaleadd_ab_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<>,
+                                                                NDHWGK,
+                                                                ck::Tuple<int8_t, int8_t>,
+                                                                ck::Tuple<int8_t, int8_t>,
+                                                                ck::Tuple<>,
+                                                                int8_t,
+                                                                ScaleAdd,
+                                                                ScaleAdd,
+                                                                PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_int8_instances<3,
+                                                               NDHWGC,
+                                                               GKZYXC,
+                                                               NDHWGK,
+                                                               ConvFwdDefault>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_int8_instances<3,
+                                                               NDHWGC,
+                                                               GKZYXC,
+                                                               NDHWGK,
+                                                               ConvFwd1x1P0>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_scaleadd_ab_int8_instances<3,
+                                                               NDHWGC,
+                                                               GKZYXC,
+                                                               NDHWGK,
+                                                               ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_scaleadd_relu/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp

@@ -10,18 +10,18 @@ namespace device {
 namespace instance {
 
 void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              BF16,
-                                                              BF16,
-                                                              ck::Tuple<BF16, BF16>,
-                                                              BF16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& instances)
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                BF16,
+                                                                BF16,
+                                                                ck::Tuple<BF16, BF16>,
+                                                                BF16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances)
 {
     add_device_operation_instances(
         instances,

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_scaleadd_relu/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp

@@ -10,18 +10,18 @@ namespace device {
 namespace instance {
 
 void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              F16,
-                                                              F16,
-                                                              ck::Tuple<half_t, half_t>,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& instances)
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                F16,
+                                                                F16,
+                                                                ck::Tuple<half_t, half_t>,
+                                                                F16,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances)
 {
     add_device_operation_instances(
         instances,

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_scaleadd_relu/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp

@@ -10,18 +10,18 @@ namespace device {
 namespace instance {
 
 void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              F32,
-                                                              F32,
-                                                              ck::Tuple<F32, F32>,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& instances)
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                F32,
+                                                                F32,
+                                                                ck::Tuple<F32, F32>,
+                                                                F32,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances)
 {
     add_device_operation_instances(
         instances,

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_scaleadd_relu/xdl/device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp

@@ -9,18 +9,18 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhwgk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
-                                                              NDHWGC,
-                                                              GKZYXC,
-                                                              ck::Tuple<NDHWGK, NDHWGK>,
-                                                              NDHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              ck::Tuple<F32, F32>,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              ScaleAddScaleAddRelu>>>& instances)
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
+                                                                NDHWGC,
+                                                                GKZYXC,
+                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                NDHWGK,
+                                                                int8_t,
+                                                                int8_t,
+                                                                ck::Tuple<F32, F32>,
+                                                                int8_t,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                ScaleAddScaleAddRelu>>>& instances)
 {
     add_device_operation_instances(
         instances,

library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt

-set(DEVICE_NORMALIZATION_INSTANCES)
-
-list(APPEND DEVICE_NORMALIZATION_INSTANCES 
-    device_layernorm2d_f16_instance.cpp
-    device_layernorm4d_f16_instance.cpp
-    device_groupnorm_f16_instance.cpp
-    device_groupnorm_swish_f16_instance.cpp
-    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
-    device_layernorm2d_f32_instance.cpp
-    device_layernorm4d_f32_instance.cpp
-    device_groupnorm_f32_instance.cpp
-    device_groupnorm_swish_f32_instance.cpp)
-
-add_instance_library(device_normalization_instance ${DEVICE_NORMALIZATION_INSTANCES})

library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
-#include "ck/utility/data_type.hpp"
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_f16_instances =
-    // clang-format off
-    std::tuple <
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8, 2>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>
-        // clang-format on
-        >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_splitk_f16_instances =
-    // clang-format off
-    std::tuple <
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8, 2>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>
-        // clang-format on
-        >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_f16_generic_instance = std::tuple<
-    // clang-format off
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
-    // clang-format on
-    >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_f32_instances = std::tuple<
-    // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
-    // clang-format on
-    >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_splitk_f32_instances = std::tuple<
-    // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
-    // clang-format on
-    >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_f32_generic_instance = std::tuple<
-    // clang-format off
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
-    // clang-format on
-    >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_f16_f32_f32_f16_instances = std::tuple<
-    // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
-    // clang-format on
-    >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_splitk_f16_f32_f32_f16_instances = std::tuple<
-    // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
-    // clang-format on
-    >;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_normalization_f16_f32_f32_f16_generic_instance = std::tuple<
-    // clang-format off
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
-    // clang-format on
-    >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck

library/src/tensor_operation_instance/gpu/normalization_fwd/CMakeLists.txt

+set(DEVICE_NORMALIZATION_FWD_INSTANCES)
+
+list(APPEND DEVICE_NORMALIZATION_FWD_INSTANCES
+    device_layernorm2d_fwd_f16_instance.cpp
+    device_layernorm4d_fwd_f16_instance.cpp
+    device_groupnorm_fwd_f16_instance.cpp
+    device_groupnorm_fwd_swish_f16_instance.cpp
+    device_groupnorm_fwd_swish_f16_f32_f32_f16_instance.cpp
+    device_layernorm2d_fwd_f32_instance.cpp
+    device_layernorm4d_fwd_f32_instance.cpp
+    device_groupnorm_fwd_f32_instance.cpp
+    device_groupnorm_fwd_swish_f32_instance.cpp)
+
+add_instance_library(device_normalization_fwd_instance ${DEVICE_NORMALIZATION_FWD_INSTANCES})

library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp → library/src/tensor_operation_instance/gpu/normalization_fwd/device_groupnorm_fwd_f16_instance.cpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "normalization_instance_common.hpp"
+#include "normalization_fwd_instance_common.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -10,8 +10,8 @@ namespace instance {
 
 using Pass = ck::tensor_operation::element_wise::PassThrough;
 
-void add_device_normalization_rank_5_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Pass, 5, 3>>>&
+void add_device_normalization_fwd_rank_5_3_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F16, F16, F16, F16, F32, Pass, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(instances,

library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp → library/src/tensor_operation_instance/gpu/normalization_fwd/device_groupnorm_fwd_f32_instance.cpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "normalization_instance_common.hpp"
+#include "normalization_fwd_instance_common.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -10,8 +10,8 @@ namespace instance {
 
 using Pass = ck::tensor_operation::element_wise::PassThrough;
 
-void add_device_normalization_rank_5_3_f32_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 5, 3>>>&
+void add_device_normalization_fwd_rank_5_3_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F32, F32, F32, F32, F32, Pass, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(instances,

library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp → library/src/tensor_operation_instance/gpu/normalization_fwd/device_groupnorm_fwd_swish_f16_f32_f32_f16_instance.cpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "normalization_instance_common.hpp"
+#include "normalization_fwd_instance_common.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -10,8 +10,8 @@ namespace instance {
 
 using Swish = ck::tensor_operation::element_wise::Swish;
 
-void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F16, F32, Swish, 5, 3>>>&
+void add_device_normalization_fwd_rank_5_3_swish_f16_f32_f32_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F16, F32, F32, F16, F32, Swish, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(

library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp → library/src/tensor_operation_instance/gpu/normalization_fwd/device_groupnorm_fwd_swish_f16_instance.cpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "normalization_instance_common.hpp"
+#include "normalization_fwd_instance_common.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -10,8 +10,8 @@ namespace instance {
 
 using Swish = ck::tensor_operation::element_wise::Swish;
 
-void add_device_normalization_rank_5_3_swish_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Swish, 5, 3>>>&
+void add_device_normalization_fwd_rank_5_3_swish_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F16, F16, F16, F16, F32, Swish, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(instances,

library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp → library/src/tensor_operation_instance/gpu/normalization_fwd/device_groupnorm_fwd_swish_f32_instance.cpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "normalization_instance_common.hpp"
+#include "normalization_fwd_instance_common.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -10,8 +10,8 @@ namespace instance {
 
 using Swish = ck::tensor_operation::element_wise::Swish;
 
-void add_device_normalization_rank_5_3_swish_f32_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Swish, 5, 3>>>&
+void add_device_normalization_fwd_rank_5_3_swish_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F32, F32, F32, F32, F32, Swish, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(instances,

library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp → library/src/tensor_operation_instance/gpu/normalization_fwd/device_layernorm2d_fwd_f16_instance.cpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "normalization_instance_common.hpp"
+#include "normalization_fwd_instance_common.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -10,8 +10,8 @@ namespace instance {
 
 using Pass = ck::tensor_operation::element_wise::PassThrough;
 
-void add_device_normalization_rank_2_1_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Pass, 2, 1>>>&
+void add_device_normalization_fwd_rank_2_1_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F16, F16, F16, F16, F32, Pass, 2, 1>>>&
         instances)
 {
     add_device_operation_instances(instances,

library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp → library/src/tensor_operation_instance/gpu/normalization_fwd/device_layernorm2d_fwd_f32_instance.cpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "normalization_instance_common.hpp"
+#include "normalization_fwd_instance_common.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -10,8 +10,8 @@ namespace instance {
 
 using Pass = ck::tensor_operation::element_wise::PassThrough;
 
-void add_device_normalization_rank_2_1_f32_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 2, 1>>>&
+void add_device_normalization_fwd_rank_2_1_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F32, F32, F32, F32, F32, Pass, 2, 1>>>&
         instances)
 {
     add_device_operation_instances(instances,