Merge remote-tracking branch 'rosenrodt/gemm-standalone-bench' into wavelet_model

library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp

@@ -5,7 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp

@@ -5,7 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp

@@ -5,7 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp

@@ -5,7 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp

@@ -5,7 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp

@@ -6,7 +6,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 

library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp

@@ -6,7 +6,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 

library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp

@@ -6,7 +6,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 

library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp

@@ -6,7 +6,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 

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

 add_instance_library(device_normalization_instance
-    device_layernorm_f16_instance.cpp
-    device_layernorm_f32_instance.cpp
-    device_softmax_f32_f32_instance.cpp
-    device_softmax_f16_f16_instance.cpp
+    device_normalization_f16_instance.cpp
+    device_normalization_f32_instance.cpp
 )

library/src/tensor_operation_instance/gpu/normalization/device_layernorm_f16_instance.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_layernorm_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;
-
-using Pass = ck::tensor_operation::element_wise::PassThrough;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_layernorm_f16_instances = std::tuple<
-    // clang-format off
-        // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 4, 1, 4, 1, 4, 4>, // fallback kernel
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 4, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceLayernormImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 2, 1, 2, 1, 2, 2>
-    // clang-format on
-    >;
-
-void add_device_layernorm_rank_2_1_f16_instances(
-    std::vector<std::unique_ptr<DeviceLayernorm<F16, F16, F16, F32, F16, Pass, 2, 1>>>& instances)
-{
-    add_device_operation_instances(instances, device_layernorm_f16_instances<Pass, 2, 1>{});
-}
-
-void add_device_layernorm_rank_4_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceLayernorm<F16, F16, F16, F32, F16, Pass, 4, 3>>>& instances)
-{
-    add_device_operation_instances(instances, device_layernorm_f16_instances<Pass, 4, 3>{});
-}
-
-void add_device_layernorm_rank_5_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceLayernorm<F16, F16, F16, F32, F16, Pass, 5, 3>>>& instances)
-{
-    add_device_operation_instances(instances, device_layernorm_f16_instances<Pass, 5, 3>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck

library/src/tensor_operation_instance/gpu/normalization/device_normalization_f16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_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;
+
+using Pass = ck::tensor_operation::element_wise::PassThrough;
+
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+// clang-format off
+using device_normalization_f16_instances =
+    std::tuple <
+        // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 4, 1, 4, 1, 4, 4>, // fallback kernel
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 4, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 2, 1, 2, 1, 2, 2>
+    >;
+// clang-format on
+
+void add_device_normalization_rank_2_1_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 2, 1>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 2, 1>{});
+}
+
+void add_device_normalization_rank_4_3_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 4, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 4, 3>{});
+}
+
+void add_device_normalization_rank_5_3_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 5, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 5, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/normalization/device_layernorm_f32_instance.cpp → library/src/tensor_operation_instance/gpu/normalization/device_normalization_f32_instance.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_layernorm_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
 #include "ck/utility/data_type.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
@@ -20,33 +20,36 @@ template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_layernorm_f32_instances = std::tuple<
     // clang-format off
         // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 4, 64, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceLayernormImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 4, 64, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>
     // clang-format on
     >;
 
-void add_device_layernorm_rank_2_1_f32_instances(
-    std::vector<std::unique_ptr<DeviceLayernorm<F32, F32, F32, F32, F32, Pass, 2, 1>>>& instances)
+void add_device_normalization_rank_2_1_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 2, 1>>>&
+        instances)
 {
     add_device_operation_instances(instances, device_layernorm_f32_instances<Pass, 2, 1>{});
 }
 
-void add_device_layernorm_rank_4_3_f32_instances(
-    std::vector<std::unique_ptr<DeviceLayernorm<F32, F32, F32, F32, F32, Pass, 4, 3>>>& instances)
+void add_device_normalization_rank_4_3_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 4, 3>>>&
+        instances)
 {
     add_device_operation_instances(instances, device_layernorm_f32_instances<Pass, 4, 3>{});
 }
 
-void add_device_layernorm_rank_5_3_f32_instances(
-    std::vector<std::unique_ptr<DeviceLayernorm<F32, F32, F32, F32, F32, Pass, 5, 3>>>& instances)
+void add_device_normalization_rank_5_3_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 5, 3>>>&
+        instances)
 {
     add_device_operation_instances(instances, device_layernorm_f32_instances<Pass, 5, 3>{});
 }

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

+add_instance_library(device_softmax_instance
+    device_softmax_f16_f16_instance.cpp
+    device_softmax_f32_f32_instance.cpp
+)

profiler/CMakeLists.txt

@@ -25,7 +25,7 @@ set(PROFILER_SOURCE
     src/profile_reduce.cpp
     src/profile_groupnorm.cpp
     src/profile_layernorm.cpp
-    src/profile_normalization.cpp
+    src/profile_softmax.cpp
 )
 
 add_executable(ckProfiler ${PROFILER_SOURCE})
@@ -55,4 +55,5 @@ target_link_libraries(ckProfiler PRIVATE device_conv3d_bwd_weight_instance)
 target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_instance)
 target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instance)
 target_link_libraries(ckProfiler PRIVATE device_normalization_instance)
+target_link_libraries(ckProfiler PRIVATE device_softmax_instance)
 target_link_libraries(ckProfiler PRIVATE device_reduce_instance)

profiler/include/profile_batched_gemm_masking_scale_softmax_gemm_permute_impl.hpp

@@ -7,7 +7,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/gpu/batched_gemm_masking_scale_softmax_gemm_permute.hpp"

profiler/include/profile_groupnorm_impl.hpp

@@ -7,7 +7,7 @@
 
 #include "ck/ck.hpp"
 
-#include "ck/library/tensor_operation_instance/gpu/layernorm.hpp"
+#include "ck/library/tensor_operation_instance/gpu/normalization.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
@@ -75,14 +75,14 @@ bool profile_groupnorm_impl(int do_verification,
     beta_dev.ToDevice(beta.mData.data());
 
     // add device normalization instances
-    using DeviceOp = ck::tensor_operation::device::DeviceLayernorm<XDataType,
-                                                                   GammaDataType,
-                                                                   BetaDataType,
-                                                                   AccDataType,
-                                                                   YDataType,
-                                                                   PassThrough,
-                                                                   5,
-                                                                   3>;
+    using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
+                                                                       GammaDataType,
+                                                                       BetaDataType,
+                                                                       AccDataType,
+                                                                       YDataType,
+                                                                       PassThrough,
+                                                                       5,
+                                                                       3>;
 
     // get device op instances
     const auto instance_ptrs =

profiler/include/profile_layernorm_impl.hpp

@@ -6,9 +6,7 @@
 #include <iomanip>
 
 #include "ck/ck.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/layernorm.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/normalization.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
@@ -28,27 +26,29 @@ void profile_layernorm_impl(int do_verification,
                             int init_method,
                             bool do_log,
                             bool time_kernel,
-                            std::vector<index_t> length,
-                            std::vector<index_t> strideXY,
-                            std::vector<index_t> strideGamma,
-                            std::vector<index_t> strideBeta)
+                            std::vector<index_t> length)
 {
     using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
     if(length.size() < 2)
         return;
 
-    // Assume normalize dimension except for first dimension
+    // Assume normalize dimension except for batch (first) dimension
     std::vector<index_t> reduce_length{length.begin() + 1, length.end()};
     std::vector<index_t> reduce_dim;
     for(int i = 1; i < Rank; ++i)
         reduce_dim.push_back(i);
 
     Tensor<XDataType> x(length);
-    Tensor<GammaDataType> gamma(reduce_length, strideGamma);
-    Tensor<BetaDataType> beta(reduce_length, strideBeta);
-    Tensor<YDataType> y(length, strideXY);
-    Tensor<YDataType> host_y(length, strideXY);
+    Tensor<GammaDataType> gamma(reduce_length);
+    Tensor<BetaDataType> beta(reduce_length);
+    Tensor<YDataType> y(length);
+    Tensor<YDataType> host_y(length);
+
+    std::vector<index_t> strideXY =
+        std::vector<ck::index_t>{x.mDesc.GetStrides().begin(), x.mDesc.GetStrides().end()};
+    std::vector<index_t> strideGammaBeta = strideXY;
+    strideGammaBeta[0]                   = 0;
 
     switch(init_method)
     {
@@ -84,14 +84,14 @@ void profile_layernorm_impl(int do_verification,
     constexpr int NumReduceDim = Rank - 1;
 
     // add device normalization instances
-    using DeviceOp = ck::tensor_operation::device::DeviceLayernorm<XDataType,
-                                                                   GammaDataType,
-                                                                   BetaDataType,
-                                                                   AccDataType,
-                                                                   YDataType,
-                                                                   PassThrough,
-                                                                   Rank,
-                                                                   NumReduceDim>;
+    using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
+                                                                       GammaDataType,
+                                                                       BetaDataType,
+                                                                       AccDataType,
+                                                                       YDataType,
+                                                                       PassThrough,
+                                                                       Rank,
+                                                                       NumReduceDim>;
 
     // get device op instances
     const auto instance_ptrs =
@@ -126,8 +126,8 @@ void profile_layernorm_impl(int do_verification,
     {
         auto argument_ptr = inst_ptr->MakeArgumentPointer(length,
                                                           strideXY,
-                                                          strideGamma,
-                                                          strideBeta,
+                                                          strideGammaBeta,
+                                                          strideGammaBeta,
                                                           strideXY,
                                                           reduce_dim,
                                                           1e-4,