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

add int8_t for Resize (#10067)

Browse source 
As we support quantization for format s8s8, we need Resize to support int8_t.
This commit is contained in:
Yufeng Li 2021-12-17 15:36:09 -08:00 committed by GitHub
parent 4fd85cd97a
commit 12ee2e942f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
6 changed files with 132 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
docs/OperatorKernels.md
View file

@ -265,9 +265,9 @@ Do not modify directly.*
|||13|**T** = 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)<br/> **shape** = tensor(int64)|
|||[5, 12]|**T** = 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)<br/> **shape** = tensor(int64)|
|||[1, 4]|**T** = 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)|
|Resize|*in* X:**T**<br> *in* scales:**tensor(float)**<br> *out* Y:**T**<br><br>or<br><br>*in* X:**T1**<br> *in* roi:**T2**<br> *in* scales:**tensor(float)**<br> *in* sizes:**tensor(int64)**<br> *out* Y:**T1**|13+|**T1** = tensor(float), tensor(int32), tensor(uint8)|
|||[11, 12]|**T1** = tensor(float), tensor(int32), tensor(uint8)|
|||10|**T** = tensor(float), tensor(int32), tensor(uint8)|
|Resize|*in* X:**T**<br> *in* scales:**tensor(float)**<br> *out* Y:**T**<br><br>or<br><br>*in* X:**T1**<br> *in* roi:**T2**<br> *in* scales:**tensor(float)**<br> *in* sizes:**tensor(int64)**<br> *out* Y:**T1**|13+|**T1** = tensor(float), tensor(int32), tensor(int8), tensor(uint8)|
|||[11, 12]|**T1** = tensor(float), tensor(int32), tensor(int8), tensor(uint8)|
|||10|**T** = tensor(float), tensor(int32), tensor(int8), tensor(uint8)|
|ReverseSequence|*in* input:**T**<br> *in* sequence_lens:**tensor(int64)**<br> *out* Y:**T**|10+|**T** = 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)|
|RoiAlign|*in* X:**T1**<br> *in* rois:**T1**<br> *in* batch_indices:**T2**<br> *out* Y:**T1**|10+|**T** = tensor(double), tensor(float)<br/> **T2** = tensor(int64)|
|Round|*in* X:**T**<br> *out* Y:**T**|11+|**T** = tensor(double), tensor(float), tensor(float16)|
@ -346,8 +346,8 @@ Do not modify directly.*
|Unsqueeze|*in* data:**T**<br> *in* axes:**tensor(int64)**<br> *out* expanded:**T**<br><br>or<br><br>*in* data:**T**<br> *out* expanded:**T**|13+|**T** = 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]|**T** = 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]|**T** = 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)|
|Upsample|*in* X:**T**<br> *in* scales:**tensor(float)**<br> *out* Y:**T**<br><br>or<br><br>*in* X:**T**<br> *out* Y:**T**|9|**T** = tensor(float), tensor(int32), tensor(uint8)|
|||[7, 8]|**T** = tensor(float), tensor(int32), tensor(uint8)|
|Upsample|*in* X:**T**<br> *in* scales:**tensor(float)**<br> *out* Y:**T**<br><br>or<br><br>*in* X:**T**<br> *out* Y:**T**|9|**T** = tensor(float), tensor(int32), tensor(int8), tensor(uint8)|
|||[7, 8]|**T** = tensor(float), tensor(int32), tensor(int8), tensor(uint8)|
|Where|*in* condition:**B**<br> *in* X:**T**<br> *in* Y:**T**<br> *out* output:**T**|9+|**T** = tensor(double), tensor(float), tensor(int32), tensor(int64), tensor(string), tensor(uint8)|
|Xor|*in* A:**T**<br> *in* B:**T**<br> *out* C:**T1**|7+|**T** = tensor(bool)<br/> **T1** = tensor(bool)|
| |

15
onnxruntime/core/providers/cpu/cpu_execution_provider.cc
View file

@ -219,6 +219,7 @@ class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDoma
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 10, Unsqueeze);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8, float, Upsample);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8, int32_t, Upsample);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8, int8_t, Upsample);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8, uint8_t, Upsample);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 8, 12, float, Expand);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 8, 12, double, Expand);
@ -298,6 +299,7 @@ class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOn
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, PRelu);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 9, float, Upsample);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 9, int32_t, Upsample);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 9, int8_t, Upsample);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 9, uint8_t, Upsample);
// Opset 10
@ -308,6 +310,7 @@ class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDoma
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 12, Mod);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10, float, Resize);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10, int32_t, Resize);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10, int8_t, Resize);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10, uint8_t, Resize);
class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, ThresholdedRelu);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 12, uint8_t, DequantizeLinear);
@ -442,6 +445,7 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain,
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, int64_t_float_int32_t, OneHot);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, float, Resize);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, int32_t, Resize);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, int8_t, Resize);
class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, uint8_t, Resize);
// opset 12
@ -659,6 +663,7 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain,
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, int64_t, ReduceSum);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, float, Resize);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, int32_t, Resize);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, int8_t, Resize);
class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, uint8_t, Resize);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, 15, Loop);
class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, 15, If);
@ -1042,6 +1047,8 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
float, Upsample)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8,
int32_t, Upsample)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8,
int8_t, Upsample)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8,
uint8_t, Upsample)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 8, 12, float,
@ -1185,6 +1192,8 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
float, Upsample)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 9,
int32_t, Upsample)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 9,
int8_t, Upsample)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 9,
uint8_t, Upsample)>,
@ -1201,6 +1210,8 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
float, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10,
int32_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10,
int8_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10,
uint8_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, ThresholdedRelu)>,
@ -1366,6 +1377,8 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
float, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12,
int32_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12,
int8_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12,
uint8_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 11,
@ -1782,6 +1795,8 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
float, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13,
int32_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13,
int8_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13,
uint8_t, Resize)>,
BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, 15, Loop)>,

22
onnxruntime/core/providers/cpu/tensor/resize.cc
View file

@ -21,6 +21,14 @@ ONNX_CPU_OPERATOR_VERSIONED_TYPED_KERNEL(
KernelDefBuilder().TypeConstraint("T", DataTypeImpl::GetTensorType<int32_t>()),
Resize<int32_t>);
ONNX_CPU_OPERATOR_VERSIONED_TYPED_KERNEL(
Resize,
10,
10,
int8_t,
KernelDefBuilder().TypeConstraint("T", DataTypeImpl::GetTensorType<int8_t>()),
Resize<int8_t>);
ONNX_CPU_OPERATOR_VERSIONED_TYPED_KERNEL(
Resize,
10,
@ -43,6 +51,13 @@ ONNX_CPU_OPERATOR_VERSIONED_TYPED_KERNEL(
KernelDefBuilder().TypeConstraint("T1", DataTypeImpl::GetTensorType<int32_t>()),
Resize<int32_t>);
ONNX_CPU_OPERATOR_VERSIONED_TYPED_KERNEL(
Resize,
11, 12,
int8_t,
KernelDefBuilder().TypeConstraint("T1", DataTypeImpl::GetTensorType<int8_t>()),
Resize<int8_t>);
ONNX_CPU_OPERATOR_VERSIONED_TYPED_KERNEL(
Resize,
11, 12,
@ -64,6 +79,13 @@ ONNX_CPU_OPERATOR_TYPED_KERNEL(
KernelDefBuilder().TypeConstraint("T1", DataTypeImpl::GetTensorType<int32_t>()),
Resize<int32_t>);
ONNX_CPU_OPERATOR_TYPED_KERNEL(
Resize,
13,
int8_t,
KernelDefBuilder().TypeConstraint("T1", DataTypeImpl::GetTensorType<int8_t>()),
Resize<int8_t>);
ONNX_CPU_OPERATOR_TYPED_KERNEL(
Resize,
13,

2
onnxruntime/core/providers/cpu/tensor/upsample.cc
View file

@ -20,11 +20,13 @@ namespace onnxruntime {
REGISTER_VERSIONED_TYPED_KERNEL(float, 7, 8);
REGISTER_VERSIONED_TYPED_KERNEL(int32_t, 7, 8);
REGISTER_VERSIONED_TYPED_KERNEL(int8_t, 7, 8);
REGISTER_VERSIONED_TYPED_KERNEL(uint8_t, 7, 8);
// Upsample was deprecated in opset 10
REGISTER_VERSIONED_TYPED_KERNEL(float, 9, 9);
REGISTER_VERSIONED_TYPED_KERNEL(int32_t, 9, 9);
REGISTER_VERSIONED_TYPED_KERNEL(int8_t, 9, 9);
REGISTER_VERSIONED_TYPED_KERNEL(uint8_t, 9, 9);
template <typename T>

68
onnxruntime/test/providers/cpu/tensor/resize_op_test.cc
View file

@ -1224,5 +1224,73 @@ TEST(ResizeOpTest, ResizeOp_MissingRoiAndMissingScalesOptionalInputs) {
test.AddOutput<float>("Y", {H, W}, X);
test.Run();
}
template <typename T>
void ResizeOpTypeCheck_Ver_10() {
OpTester test("Resize", 10);
std::vector<float> scales{1.0f, 1.0f, 2.0f, 3.0f};
test.AddAttribute("mode", "nearest");
constexpr int64_t N = 1, C = 1, H = 2, W = 2;
std::vector<T> X = {1, 2, 3, 4};
test.AddInput<T>("X", {N, C, H, W}, X);
test.AddInput<float>("scales", {4}, scales);
std::vector<T> Y = {1, 1, 1, 2, 2, 2,
1, 1, 1, 2, 2, 2,
3, 3, 3, 4, 4, 4,
3, 3, 3, 4, 4, 4};
test.AddOutput<T>("Y", {N, C, static_cast<int64_t>(H * scales[2]), static_cast<int64_t>(W * scales[3])}, Y);
test.Run();
}
TEST(ResizeOpTest, ResizeOpTypeCheck_Ver_10) {
ResizeOpTypeCheck_Ver_10<float>();
ResizeOpTypeCheck_Ver_10<int32_t>();
ResizeOpTypeCheck_Ver_10<int8_t>();
ResizeOpTypeCheck_Ver_10<uint8_t>();
}
template <typename T>
void ResizeOpTypeCheck_Ver_11_13(int opset_version) {
OpTester test("Resize", opset_version);
std::vector<float> roi{};
std::vector<float> scales{1.0f, 1.0f, 2.0f, 3.0f};
test.AddAttribute("mode", "nearest");
constexpr int64_t N = 1, C = 1, H = 2, W = 2;
std::vector<T> X = {1, 2, 3, 4};
test.AddInput<T>("X", {N, C, H, W}, X);
test.AddInput<float>("roi", {0}, roi);
test.AddInput<float>("scales", {4}, scales);
std::vector<T> Y = {1, 1, 1, 2, 2, 2,
1, 1, 1, 2, 2, 2,
3, 3, 3, 4, 4, 4,
3, 3, 3, 4, 4, 4};
test.AddOutput<T>("Y", {N, C, static_cast<int64_t>(H * scales[2]), static_cast<int64_t>(W * scales[3])}, Y);
test.Run();
}
TEST(ResizeOpTest, ResizeOpTypeCheck_Ver11) {
ResizeOpTypeCheck_Ver_11_13<float>(11);
ResizeOpTypeCheck_Ver_11_13<int32_t>(11);
ResizeOpTypeCheck_Ver_11_13<int8_t>(11);
ResizeOpTypeCheck_Ver_11_13<uint8_t>(11);
}
TEST(ResizeOpTest, ResizeOpTypeCheck_Ver13) {
ResizeOpTypeCheck_Ver_11_13<float>(13);
ResizeOpTypeCheck_Ver_11_13<int32_t>(13);
ResizeOpTypeCheck_Ver_11_13<int8_t>(13);
ResizeOpTypeCheck_Ver_11_13<uint8_t>(13);
}
} // namespace test
} // namespace onnxruntime

20
onnxruntime/test/testdata/kernel_def_hashes/onnx.cpu.json vendored
View file

@ -2095,6 +2095,18 @@
"Resize ai.onnx CPUExecutionProvider",
16654410418270805888
],
[
"Resize ai.onnx CPUExecutionProvider",
11612050915202267528
],
[
"Resize ai.onnx CPUExecutionProvider",
17391179092212899176
],
[
"Resize ai.onnx CPUExecutionProvider",
18197536715869401872
],
[
"ReverseSequence ai.onnx CPUExecutionProvider",
777686786711298232
@ -2543,6 +2555,14 @@
"Upsample ai.onnx CPUExecutionProvider",
17854969343220970496
],
[
"Upsample ai.onnx CPUExecutionProvider",
4881414272874428968
],
[
"Upsample ai.onnx CPUExecutionProvider",
7886741677344006832
],
[
"Where ai.onnx CPUExecutionProvider",
4809895774227191528