Skip to content

Fill CUDA Transpose operator opset gap (23 → 25) #27740

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tianleiwu merged 4 commits into from
Apr 6, 2026
Merged

Fill CUDA Transpose operator opset gap (23 → 25) #27740

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
50bd44a
Update Transpose CUDA operator to fill opset gap from 23 to 25
Copilot Mar 18, 2026
087c116
Merge remote-tracking branch 'origin/main' into copilot/update-transp…
Copilot Mar 18, 2026
e47282a
Update docs/OperatorKernels.md for Transpose opset 25 and merge main
Copilot Mar 18, 2026
58955c7
Merge remote-tracking branch 'origin/main' into copilot/update-transp…
tianleiwu Mar 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion docs/OperatorKernels.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -998,7 +998,8 @@ Do not modify directly.*
|TopK|*in* X:**T**<br> *in* K:**tensor(int64)**<br> *out* Values:**T**<br> *out* Indices:**I**<br><br>or<br><br>*in* X:**T**<br> *out* Values:**T**<br> *out* Indices:**I**|11+|**I** = tensor(int64)<br/> **T** = tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64)|
|||10|**I** = tensor(int64)<br/> **T** = tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64)|
|||[1, 9]|**T** = tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64)|
|Transpose|*in* data:**T**<br> *out* transposed:**T**|23+|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|Transpose|*in* data:**T**<br> *out* transposed:**T**|25+|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[23, 24]|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[21, 22]|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[13, 20]|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[1, 12]|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
Expand Down
10 changes: 6 additions & 4 deletions onnxruntime/core/providers/cuda/cuda_execution_provider.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -852,9 +852,9 @@ class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDom
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 5, 12, Reshape);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, Shape);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, Size);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, Transpose);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 6, 12, Tile);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, Tile);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 12, Transpose);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 6, float, InstanceNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 6, double, InstanceNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 6, MLFloat16, InstanceNormalization);
Expand Down Expand Up @@ -1580,7 +1580,6 @@ class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kO
#endif
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Shape);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Reshape);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Transpose);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Squeeze);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Unsqueeze);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, float, Pad);
Expand All @@ -1593,6 +1592,7 @@ class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDom
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Loop);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Scan);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Size);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Transpose);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 21, 22, Flatten);

// Opset 22.
Expand Down Expand Up @@ -1667,7 +1667,6 @@ class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kO
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 23, bool, Pad);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, Shape);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, Reshape);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, Transpose);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 23, Squeeze);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 23, Unsqueeze);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, ConstantOfShape);
Expand All @@ -1676,6 +1675,7 @@ class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDom
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, Loop);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, Scan);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, Size);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, Transpose);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, Flatten);

#if !defined(DISABLE_FLOAT4_TYPES)
Expand Down Expand Up @@ -1707,6 +1707,7 @@ class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, I
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Loop);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Scan);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Size);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Transpose);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Flatten);

#endif
Expand Down Expand Up @@ -2824,7 +2825,7 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
#endif
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, Shape)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, Reshape)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, Transpose)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, Transpose)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 23, Squeeze)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 23, Unsqueeze)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 23, 24, ConstantOfShape)>,
Expand Down Expand Up @@ -2860,6 +2861,7 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Loop)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Scan)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Size)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Transpose)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 25, Flatten)>,
#endif
};
Expand Down
11 changes: 10 additions & 1 deletion onnxruntime/core/providers/cuda/tensor/transpose.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,10 +37,19 @@ ONNX_OPERATOR_VERSIONED_KERNEL_EX(
.TypeConstraint("T", DataTypeImpl::AllFixedSizeTensorTypes()),
Transpose);

ONNX_OPERATOR_VERSIONED_KERNEL_EX(
Transpose,
kOnnxDomain,
23, 24,
kCudaExecutionProvider,
(*KernelDefBuilder::Create())
.TypeConstraint("T", DataTypeImpl::AllFixedSizeTensorTypes()),
Transpose);

ONNX_OPERATOR_KERNEL_EX(
Transpose,
kOnnxDomain,
23,
25,
kCudaExecutionProvider,
(*KernelDefBuilder::Create())
.TypeConstraint("T", DataTypeImpl::AllFixedSizeTensorTypes()),
Expand Down
Loading