Merge branch 'develop' into migx-jit-lib

example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt

-add_example_executable(example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp)
+endif()

example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp

@@ -173,6 +173,8 @@ using DeviceGemmInstance =
         8,
         8,
         true,
+        9,           // D0sTransferSrcVectorDim
+        4,           // D0sTransferSrcScalaerPerVector
         S<8, 32, 1>, // B1BlockTransfer
         S<0, 2, 1>,
         S<0, 2, 1>,
@@ -189,7 +191,7 @@ int main(int argc, char* argv[])
 {
     bool do_verification = true;
     int init_method      = 1;
-    bool time_kernel     = false;
+    bool time_kernel     = true;
 
     // GEMM shape
     ck::index_t M              = 1024;

example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -10,4 +11,5 @@ foreach(gpu IN LISTS GPU_TARGETS)
    add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
    set(target 1)
  endif()
-endforeach()
\ No newline at end of file
+endforeach()
+endif()

example/39_permute/CMakeLists.txt

-add_custom_target(example_permute)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_custom_target(example_permute)
 
-add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
-add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
-add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
+    add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
+    add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
+    add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
 
-add_dependencies(example_permute example_permute_1xHxW_fp16)
-add_dependencies(example_permute example_permute_NxHxW_fp16)
-add_dependencies(example_permute example_permute_HxWx4_fp16)
+    add_dependencies(example_permute example_permute_1xHxW_fp16)
+    add_dependencies(example_permute example_permute_NxHxW_fp16)
+    add_dependencies(example_permute example_permute_HxWx4_fp16)
+endif()

example/40_conv2d_fwd_quantization/CMakeLists.txt

+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -9,20 +10,19 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-# Conv perlayer quantization
-add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
 
-# Conv perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
-
-# Conv + bias + relu perlayer quantization
-add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
-
-# Conv + bias + relu perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
-
-# Conv + bias + tanh perlayer quantization
-add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
-
-# Conv + bias + tanh perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
\ No newline at end of file
+  if(DL_KERNELS)
+  # Conv perlayer quantization
+  add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
+  # Conv perchannel quantization
+  add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
+  # Conv + bias + relu perlayer quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
+  # Conv + bias + relu perchannel quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
+  # Conv + bias + tanh perlayer quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
+  # Conv + bias + tanh perchannel quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
+  endif()
+endif()
\ No newline at end of file

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType   = int8_t;
 using WeiDataType  = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType   = int8_t;
 using WeiDataType  = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType  = int8_t;
 using WeiDataType = int8_t;

example/41_grouped_conv_conv_fwd/CMakeLists.txt

@@ -3,9 +3,15 @@ list(APPEND gpu_list2 gfx908 gfx90a)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
+   endif()
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
+   endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
    endif(USE_BITINT_EXTENSION_INT4)
@@ -13,10 +19,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
  endif()
 endforeach()
 
-set(target 0)
-foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
-   set(target 1)
- endif()
-endforeach()
+if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+   endif()
+endif()

example/42_groupnorm/CMakeLists.txt

-add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
-add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
-add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
+    add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
+    add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
+endif()

example/43_splitk_gemm_bias_e_permute/CMakeLists.txt

-add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp16 splitk_gemm_bias_e_permute_xdl_fp16.cpp)
-add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp32 splitk_gemm_bias_e_permute_xdl_fp32.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp16 splitk_gemm_bias_e_permute_xdl_fp16.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp32 splitk_gemm_bias_e_permute_xdl_fp32.cpp)
+endif()

example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp

@@ -8,7 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"

example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp

@@ -8,7 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"

example/44_elementwise_permute/CMakeLists.txt

-add_example_executable(example_elementwise_permute_4D_fp16 elementwise_permute_4D_fp16.cpp)
-add_example_executable(example_elementwise_permute_4D_fp16_2d elementwise_permute_4D_fp16_2d.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_elementwise_permute_4D_fp16 elementwise_permute_4D_fp16.cpp)
+    add_example_executable(example_elementwise_permute_4D_fp16_2d elementwise_permute_4D_fp16_2d.cpp)
+endif()

example/46_gemm_add_multiply/CMakeLists.txt

-add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
-add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    if(DL_KERNELS)
+        add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
+    endif()
+    add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
+endif()

example/48_pool3d_fwd/CMakeLists.txt

-add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
-
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
+endif()

example/48_pool3d_fwd/pool3d_fwd_common.hpp

@@ -18,7 +18,45 @@
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
 
-template <typename InDataType,
+template <typename TensorLayout>
+std::vector<ck::index_t> f_tensor_strides_ncdhw(ck::index_t N_,
+                                                ck::index_t C_,
+                                                ck::index_t D,
+                                                ck::index_t H,
+                                                ck::index_t W,
+                                                TensorLayout layout)
+{
+    using namespace ck::literals;
+    (void)N_;
+    if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+        return {C_ * D * H * W, D * H * W, H * W, W, 1_uz};
+    else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
+        return {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_};
+};
+
+template <typename TensorLayout>
+HostTensorDescriptor f_host_tensor_descriptor(std::size_t N_,
+                                              std::size_t C_,
+                                              std::size_t D,
+                                              std::size_t H,
+                                              std::size_t W,
+                                              TensorLayout layout)
+{
+    using namespace ck::literals;
+
+    if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+    {
+        return HostTensorDescriptor({N_, C_, D, H, W}, {C_ * D * H * W, D * H * W, H * W, W, 1_uz});
+    }
+    else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
+    {
+        return HostTensorDescriptor({N_, C_, D, H, W},
+                                    {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+    }
+};
+
+template <typename DevicePoolFwdInstance,
+          typename InDataType,
           typename OutDataType,
           typename ComputeDataType,
           typename IndexDataType,
@@ -40,6 +78,9 @@ bool pool3d_test(bool do_verification,
                  ck::index_t window_stride_d,
                  ck::index_t window_stride_h,
                  ck::index_t window_stride_w,
+                 ck::index_t window_dilation_d,
+                 ck::index_t window_dilation_h,
+                 ck::index_t window_dilation_w,
                  ck::index_t in_left_pad_d,
                  ck::index_t in_left_pad_h,
                  ck::index_t in_left_pad_w,
@@ -47,53 +88,21 @@ bool pool3d_test(bool do_verification,
                  ck::index_t in_right_pad_h,
                  ck::index_t in_right_pad_w)
 {
-    using DevicePoolFwdInstance =
-        ck::tensor_operation::device::DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<
-            InDataType,      // InDataType
-            OutDataType,     // OutDataType
-            IndexDataType,   // IndexDataType
-            ComputeDataType, // ComputeDataType
-            ReduceOpId,
-            OutputIndex,
-            64, // BlockSize
-            64, // ReduceMThreadClusterSize
-            1,  // ReduceKThreadClusterSize
-            4,  // ReduceMThreadSliceSize
-            1,  // ReduceKThreadSliceSize
-            4>; // InSrcOutDstVectorSize
-
-    const ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Z) / window_stride_d + 1;
-    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
-    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+    const ck::index_t Zs = (Z - 1) * window_dilation_d + 1;
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    const ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Zs) / window_stride_d + 1;
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
 
     const std::vector<ck::index_t> window_spatial_lengths{Z, Y, X};
     const std::vector<ck::index_t> window_strides{
         window_stride_d, window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> window_dilations{
+        window_dilation_d, window_dilation_h, window_dilation_w};
     const std::vector<ck::index_t> input_left_pads{in_left_pad_d, in_left_pad_h, in_left_pad_w};
     const std::vector<ck::index_t> input_right_pads{in_right_pad_d, in_right_pad_h, in_right_pad_w};
 
-    // tensor layout
-    auto f_host_tensor_descriptor = [](std::size_t N_,
-                                       std::size_t C_,
-                                       std::size_t D,
-                                       std::size_t H,
-                                       std::size_t W,
-                                       auto layout) {
-        using namespace ck::literals;
-
-        if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
-        {
-            return HostTensorDescriptor({N_, C_, D, H, W},
-                                        {C_ * D * H * W, D * H * W, H * W, W, 1_uz});
-        }
-        else if constexpr(ck::is_same<decltype(layout),
-                                      ck::tensor_layout::convolution::NDHWC>::value)
-        {
-            return HostTensorDescriptor({N_, C_, D, H, W},
-                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
-        }
-    };
-
     Tensor<InDataType> in_n_c_di_hi_wi(f_host_tensor_descriptor(N, C, Di, Hi, Wi, InLayout{}));
     Tensor<OutDataType> out_n_c_do_ho_wo_host(
         f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
@@ -126,10 +135,11 @@ bool pool3d_test(bool do_verification,
         {N, C, Di, Hi, Wi},
         {Z, Y, X},
         {N, C, Do, Ho, Wo},
-        {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C},
-        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
-        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+        f_tensor_strides_ncdhw(N, C, Di, Hi, Wi, InLayout{}),
+        f_tensor_strides_ncdhw(N, C, Do, Ho, Wo, OutLayout{}),
+        f_tensor_strides_ncdhw(N, C, Do, Ho, Wo, OutLayout{}),
         window_strides,
+        window_dilations,
         input_left_pads,
         input_right_pads,
         {2, 3, 4});
@@ -165,6 +175,7 @@ bool pool3d_test(bool do_verification,
                                                              out_indices_n_c_do_ho_wo_host,
                                                              window_spatial_lengths,
                                                              window_strides,
+                                                             window_dilations,
                                                              input_left_pads,
                                                              input_right_pads);