diff --git a/onnxruntime/core/providers/cpu/object_detection/non_max_suppression_helper.h b/onnxruntime/core/providers/cpu/object_detection/non_max_suppression_helper.h index 6417c125fb..5cfd1ecee6 100644 --- a/onnxruntime/core/providers/cpu/object_detection/non_max_suppression_helper.h +++ b/onnxruntime/core/providers/cpu/object_detection/non_max_suppression_helper.h @@ -10,6 +10,11 @@ #define ORT_DEVICE __device__ #define HelperMin(a, b) _Min(a, b) #define HelperMax(a, b) _Max(a, b) +#elif defined(__HIPCC__) +#include "core/providers/rocm/cu_inc/common.cuh" +#define ORT_DEVICE __host__ __device__ +#define HelperMin(a, b) _Min(a, b) +#define HelperMax(a, b) _Max(a, b) #else #include #define ORT_DEVICE @@ -45,6 +50,8 @@ struct SelectedIndex { #ifdef __NVCC__ namespace cuda { +#elif defined(__HIPCC__) +namespace rocm { #endif namespace nms_helpers { @@ -141,8 +148,10 @@ inline bool SuppressByIOU(const float* boxes_data, int64_t box_index1, int64_t b return intersection_over_union > iou_threshold; } +} // namespace nms_helpers #ifdef __NVCC__ } // namespace cuda +#elif defined(__HIPCC__) +} // namespace rocm #endif -} // namespace nms_helpers } // namespace onnxruntime diff --git a/onnxruntime/core/providers/rocm/rocm_execution_provider.cc b/onnxruntime/core/providers/rocm/rocm_execution_provider.cc index fef51b55cd..0d60177c3b 100644 --- a/onnxruntime/core/providers/rocm/rocm_execution_provider.cc +++ b/onnxruntime/core/providers/rocm/rocm_execution_provider.cc @@ -1566,15 +1566,15 @@ static Status RegisterRocmKernels(KernelRegistry& kernel_registry) { // BuildKernelCreateInfo, // BuildKernelCreateInfo, // BuildKernelCreateInfo, - // BuildKernelCreateInfo, + BuildKernelCreateInfo, // BuildKernelCreateInfo, // BuildKernelCreateInfo, // BuildKernelCreateInfo, // BuildKernelCreateInfo, // BuildKernelCreateInfo, BuildKernelCreateInfo, - // BuildKernelCreateInfo, - // BuildKernelCreateInfo, + BuildKernelCreateInfo, + BuildKernelCreateInfo, BuildKernelCreateInfo, BuildKernelCreateInfo, BuildKernelCreateInfo, @@ -1605,7 +1605,7 @@ static Status RegisterRocmKernels(KernelRegistry& kernel_registry) { BuildKernelCreateInfo, // BuildKernelCreateInfo, // BuildKernelCreateInfo, - // BuildKernelCreateInfo, + BuildKernelCreateInfo, BuildKernelCreateInfo, BuildKernelCreateInfo, // BuildKernelCreateInfo, diff --git a/tools/ci_build/amd_hipify.py b/tools/ci_build/amd_hipify.py index 3c6afb0b78..99de6b213c 100644 --- a/tools/ci_build/amd_hipify.py +++ b/tools/ci_build/amd_hipify.py @@ -110,14 +110,6 @@ provider_excluded_files = [ 'nn/max_pool_with_index.h', 'nn/pool.cc', 'nn/pool.h', - 'object_detection/non_max_suppression.cc', - 'object_detection/non_max_suppression.h', - 'object_detection/non_max_suppression_impl.cu', - 'object_detection/non_max_suppression_impl.h', - 'object_detection/roialign.cc', - 'object_detection/roialign.h', - 'object_detection/roialign_impl.cu', - 'object_detection/roialign_impl.h', 'reduction/reduction_ops.cc', 'reduction/reduction_ops.h', 'rnn/cudnn_rnn_base.cc',