saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-24 19:43:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix whitespace and style in concat.cc (#9452)

Browse source
This commit is contained in:
Nick Kreeger 2021-10-20 12:43:46 -05:00 committed by GitHub
parent 89a22fb641
commit f1123c2fb3
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 29 additions and 26 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

55
orttraining/orttraining/training_ops/cuda/tensor/concat.cc
View file

@ -3,8 +3,10 @@
#include "orttraining/training_ops/cuda/tensor/concat.h"
#include "core/providers/cuda/tensor/concat_impl.h"
namespace onnxruntime {
namespace cuda {
ONNX_OPERATOR_KERNEL_EX(ConcatTraining,
kMSDomain,
1,
@ -28,8 +30,9 @@ Status ConcatTraining::ComputeInternal(OpKernelContext* ctx) const {
ORT_RETURN_IF_ERROR(PrepareForCompute(ctx, input_tensors, p));
// Return at this point if output tensor is going to be empty
if (p.output_num_elements == 0)
if (p.output_num_elements == 0) {
return Status::OK();
}
std::vector<int64_t> concat_sizes(input_count);
@ -58,28 +61,27 @@ Status ConcatTraining::ComputeInternal(OpKernelContext* ctx) const {
// pass by value to avoid host-to-device copy on same stream
TArray<const void*, 32> input_table(input_count);
for (int i = 0; i < input_count; ++i) {
input_table[i] = input_ptr_cpuspan[i];
input_table[i] = input_ptr_cpuspan[i];
}
ORT_RETURN_IF_ERROR(ConcatSameConcatDimImpl(Stream(),
element_bytes,
block_size_including_axis_dim,
block_size_inside_axis_dim,
concat_sizes[0],
p.output_tensor->MutableDataRaw(),
input_table,
p.output_num_elements));
element_bytes,
block_size_including_axis_dim,
block_size_inside_axis_dim,
concat_sizes[0],
p.output_tensor->MutableDataRaw(),
input_table,
p.output_num_elements));
} else {
// too many inputs, so copy sizes to device memory
ORT_RETURN_IF_ERROR(input_ptr.CopyToGpu());
ORT_RETURN_IF_ERROR(ConcatSameConcatDimImpl(Stream(),
element_bytes,
block_size_including_axis_dim,
block_size_inside_axis_dim,
concat_sizes[0],
p.output_tensor->MutableDataRaw(),
input_ptr.GpuPtr(),
p.output_num_elements));
element_bytes,
block_size_including_axis_dim,
block_size_inside_axis_dim,
concat_sizes[0],
p.output_tensor->MutableDataRaw(),
input_ptr.GpuPtr(),
p.output_num_elements));
}
} else {
// input sizes vary, copy input sizes and range metadata to device
@ -93,15 +95,15 @@ Status ConcatTraining::ComputeInternal(OpKernelContext* ctx) const {
ORT_RETURN_IF_ERROR(concat_sizes_range_gpu.CopyToGpu());
ORT_RETURN_IF_ERROR(ConcatImpl(Stream(),
element_bytes,
block_size_including_axis_dim,
block_size_inside_axis_dim,
concat_sizes_gpu.GpuPtr(),
concat_sizes_range_gpu.GpuPtr(),
axis_dimension_input_output_mapping_gpu.GpuPtr(),
p.output_tensor->MutableDataRaw(),
input_ptr.GpuPtr(),
p.output_num_elements));
element_bytes,
block_size_including_axis_dim,
block_size_inside_axis_dim,
concat_sizes_gpu.GpuPtr(),
concat_sizes_range_gpu.GpuPtr(),
axis_dimension_input_output_mapping_gpu.GpuPtr(),
p.output_tensor->MutableDataRaw(),
input_ptr.GpuPtr(),
p.output_num_elements));
}
// Create optional output tensor for 'per_input_length'
@ -112,5 +114,6 @@ Status ConcatTraining::ComputeInternal(OpKernelContext* ctx) const {
return Status::OK();
}
} // namespace cuda
} // namespace onnxruntime