Skip to content

Add LpNormalization-22 #27164

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
titaiwangms merged 9 commits into from
Jan 28, 2026
Merged

Add LpNormalization-22 #27164

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
3c6d75b
add lpnorm 22
titaiwangms Jan 26, 2026
8455653
resolve conflict
titaiwangms Jan 26, 2026
d5d7f9e
respect op spec in ONNX
titaiwangms Jan 27, 2026
891ffa8
update the tests
titaiwangms Jan 27, 2026
a4d7b89
revert implemetation changes
titaiwangms Jan 27, 2026
857c046
disable not supported l2norm test for now
titaiwangms Jan 27, 2026
3184cdc
typo
titaiwangms Jan 27, 2026
a470e5a
formatting
titaiwangms Jan 27, 2026
55bcaa9
update the op doc
titaiwangms Jan 27, 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 @@ -240,7 +240,8 @@ Do not modify directly.*
|||[13, 15]|**B** = tensor(bool)<br/> **I** = tensor(int64)<br/> **V** = seq(tensor(bfloat16)), seq(tensor(bool)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[11, 12]|**B** = tensor(bool)<br/> **I** = tensor(int64)<br/> **V** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[1, 10]|**B** = tensor(bool)<br/> **I** = tensor(int64)<br/> **V** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|LpNormalization|*in* input:**T**<br> *out* output:**T**|1+|**T** = tensor(double), tensor(float)|
|LpNormalization|*in* input:**T**<br> *out* output:**T**|22+|**T** = tensor(double), tensor(float)|
|||[1, 21]|**T** = tensor(double), tensor(float)|
|LpPool|*in* X:**T**<br> *out* Y:**T**|22+|**T** = tensor(float)|
|||[18, 21]|**T** = tensor(float)|
|||[11, 17]|**T** = tensor(float)|
Expand Down
18 changes: 12 additions & 6 deletions onnxruntime/core/providers/cpu/cpu_execution_provider.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -194,8 +194,8 @@ class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDoma
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 10, ConvTranspose);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 8, Flatten);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 6, 21, InstanceNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, float, LpNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, double, LpNormalization);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 21, float, LpNormalization);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 21, double, LpNormalization);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 12, LRN);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 9, AveragePool);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 7, MaxPool);
Expand Down Expand Up @@ -1325,6 +1325,8 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain,
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, Softsign);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, ThresholdedRelu);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, AveragePool);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, float, LpNormalization);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, double, LpNormalization);

#ifdef MLAS_F16VEC_INTRINSICS_SUPPORTED
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, MLFloat16, Conv);
Expand Down Expand Up @@ -1728,10 +1730,10 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 8, Flatten)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 6, 21,
InstanceNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, float,
LpNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, double,
LpNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 21, float,
LpNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 21, double,
LpNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 12, LRN)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 9,
AveragePool)>,
Expand Down Expand Up @@ -3384,6 +3386,10 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, Elu)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, HardSigmoid)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, InstanceNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, float,
LpNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, double,
LpNormalization)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, LpPool)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, MaxPool)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 22, MaxUnpool)>,
Expand Down
12 changes: 10 additions & 2 deletions onnxruntime/core/providers/cpu/nn/lp_norm.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,14 +7,22 @@
#include "core/providers/common.h"

namespace onnxruntime {
#define REGISTER_LPNORMALISATION_VERSIONED_KERNEL(type, sinceVersion, endVersion) \
ONNX_CPU_OPERATOR_VERSIONED_TYPED_KERNEL( \
LpNormalization, sinceVersion, endVersion, type, \
KernelDefBuilder().TypeConstraint("T", DataTypeImpl::GetTensorType<type>()), \
LpNorm<type>);

#define REGISTER_LPNORMALISATION_KERNEL(type, sinceVersion) \
ONNX_CPU_OPERATOR_TYPED_KERNEL( \
LpNormalization, sinceVersion, type, \
KernelDefBuilder().TypeConstraint("T", DataTypeImpl::GetTensorType<type>()), \
LpNorm<type>);

REGISTER_LPNORMALISATION_KERNEL(float, 1)
REGISTER_LPNORMALISATION_KERNEL(double, 1)
REGISTER_LPNORMALISATION_VERSIONED_KERNEL(float, 1, 21)
REGISTER_LPNORMALISATION_VERSIONED_KERNEL(double, 1, 21)
REGISTER_LPNORMALISATION_KERNEL(float, 22)
REGISTER_LPNORMALISATION_KERNEL(double, 22)

using InnerStride = Eigen::InnerStride<Eigen::Dynamic>;

Expand Down
5 changes: 2 additions & 3 deletions onnxruntime/test/testdata/onnx_backend_test_series_filters.jsonc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,8 @@

// Tests that are failing temporarily and should be fixed
"current_failing_tests": [
// TODO(titaiwang): onnx 1.21 should fix lpnorm zero norm issue
"^test_l2normalization*", // LpNormalization(22) not implemented
"^test_adagrad",
"^test_adagrad_multiple",
"^test_attention_4d_fp16*", // precision issue: 1 / 192 mismatched elements
Expand All @@ -39,7 +41,6 @@
"^test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal*", // location of infinities
"^test_attention_4d_attn_mask_3d_causal_expanded*", // webgpu
"^test_attention_4d_diff_heads_mask4d_padded_kv*", // Need nonpad_kv_seqlen
"^test_l2normalization*", // LpNormalization(22) not implemented
// TODO: support the following tests in Attention-cuda
"^test_attention_3d_gqa.*_cuda", // GQA not supported in Attention-cuda
"^test_attention_4d_gqa.*_cuda", // GQA not supported in Attention-cuda
Expand Down Expand Up @@ -67,8 +68,6 @@
"^test_attention_4d_attn_mask_4d_causal_cuda",
"^test_attention_4d_causal_cuda",
"^test_attention_4d_diff_heads_sizes_causal_cuda",
"^test_l1normalization*", // LpNormalization(22) not implemented
"^test_lpnormalization*", // LpNormalization(22) not implemented
"^test_tensorscatter*", // TensorScatter(24) not implemented
"^test_castlike_no_saturate_FLOAT_to_FLOAT8*", // ORT does not support ml_dtypes
"^test_castlike_UINT4_to*", // ORT does not support ml_dtypes
Expand Down
Loading