saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-30 20:18:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Register opset13 flatten, LRN for cuda. (#5694)

Browse source 
* Register opset13 flatten, LRN, ArgMax and ArgMin for cuda.

* Fix build
This commit is contained in:
Pranav Sharma 2020-11-05 14:13:15 -08:00 committed by GitHub
parent 11fe683471
commit 28197b1460
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 48 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

24
onnxruntime/core/providers/cuda/cuda_execution_provider.cc
View file

@ -481,9 +481,9 @@ class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kO
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, float, BatchNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, double, BatchNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, MLFloat16, BatchNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, float, LRN);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, double, LRN);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, MLFloat16, LRN);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, float, LRN);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, double, LRN);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, MLFloat16, LRN);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, float, Conv);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, double, Conv);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, MLFloat16, Conv);
@ -679,7 +679,7 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain,
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, MLFloat16, ArgMin);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, Compress);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, Concat);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, Flatten);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, Flatten);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, Gather);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, GatherElements);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, float, Gemm);
@ -1000,6 +1000,10 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain,
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, MLFloat16, Resize);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, int32_t, Resize);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, uint8_t, Resize);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, Flatten);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, float, LRN);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, double, LRN);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, MLFloat16, LRN);
template <>
KernelCreateInfo BuildKernelCreateInfo<void>() {
@ -1169,9 +1173,9 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, float, BatchNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, double, BatchNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, MLFloat16, BatchNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, float, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, double, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, MLFloat16, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, float, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, double, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, MLFloat16, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, float, Conv)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, double, Conv)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, MLFloat16, Conv)>,
@ -1371,7 +1375,7 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, MLFloat16, ArgMin)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, Compress)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, Concat)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, Flatten)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, Flatten)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, Gather)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, GatherElements)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, 12, MLFloat16, Gemm)>,
@ -1688,6 +1692,10 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, MLFloat16, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, int32_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, uint8_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, Flatten)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, float, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, double, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, MLFloat16, LRN)>,
};
for (auto& function_table_entry : function_table) {

27
onnxruntime/core/providers/cuda/nn/lrn.cc
View file

@ -6,19 +6,34 @@
namespace onnxruntime {
namespace cuda {
#define REGISTER_KERNEL_TYPED(T) \
ONNX_OPERATOR_TYPED_KERNEL_EX( \
#define REGISTER_KERNEL_VERSIONED_TYPED(START_VER, END_VER, T) \
ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_EX( \
LRN, \
kOnnxDomain, \
1, \
START_VER, \
END_VER, \
T, \
kCudaExecutionProvider, \
KernelDefBuilder().TypeConstraint("T", DataTypeImpl::GetTensorType<T>()), \
LRN<T>);
REGISTER_KERNEL_TYPED(float)
REGISTER_KERNEL_TYPED(double)
REGISTER_KERNEL_TYPED(MLFloat16)
#define REGISTER_KERNEL_TYPED(VER, T) \
ONNX_OPERATOR_TYPED_KERNEL_EX( \
LRN, \
kOnnxDomain, \
VER, \
T, \
kCudaExecutionProvider, \
KernelDefBuilder().TypeConstraint("T", DataTypeImpl::GetTensorType<T>()), \
LRN<T>);
REGISTER_KERNEL_VERSIONED_TYPED(1, 12, float)
REGISTER_KERNEL_VERSIONED_TYPED(1, 12, double)
REGISTER_KERNEL_VERSIONED_TYPED(1, 12, MLFloat16)
REGISTER_KERNEL_TYPED(13, float)
REGISTER_KERNEL_TYPED(13, double)
REGISTER_KERNEL_TYPED(13, MLFloat16)
template <typename T>
LRN<T>::LRN(const OpKernelInfo& info) : CudaKernel(info) {

12
onnxruntime/core/providers/cuda/tensor/flatten.cc
View file

@ -28,10 +28,20 @@ ONNX_OPERATOR_VERSIONED_KERNEL_EX(
Flatten);
// explicitly support negative axis
ONNX_OPERATOR_VERSIONED_KERNEL_EX(
Flatten,
kOnnxDomain,
11, 12,
kCudaExecutionProvider,
KernelDefBuilder()
.Alias(0, 0)
.TypeConstraint("T", DataTypeImpl::AllFixedSizeTensorTypes()),
Flatten);
ONNX_OPERATOR_KERNEL_EX(
Flatten,
kOnnxDomain,
11,
13,
kCudaExecutionProvider,
KernelDefBuilder()
.Alias(0, 0)