Huawei_Technology
/
mindspore
Not watched
1
0
Fork 0
Code
Releases 13 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

!10008 [MSLITE] argmin max support mindir 

From: @ling_qiao_min
Reviewed-by: @zhang_xue_tong,@zhanghaibo5
Signed-off-by: @zhang_xue_tong
tags/v1.1.0
mindspore-ci-bot Gitee 5 years ago
parent
c624b04b24 7454b9688c
commit
2ac483c7a3
18 changed files with 218 additions and 1351 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +0
    -102
      mindspore/lite/nnacl/arg_min_max.c
  2. +0
    -29
      mindspore/lite/nnacl/arg_min_max.h
  3. +112
    -358
      mindspore/lite/nnacl/fp32/arg_min_max_fp32.c
  4. +4
    -12
      mindspore/lite/nnacl/fp32/arg_min_max_fp32.h
  5. +0
    -47
      mindspore/lite/nnacl/fp32/instance_norm_fp32.c
  6. +0
    -32
      mindspore/lite/nnacl/fp32/instance_norm_fp32.h
  7. +1
    -0
      mindspore/lite/src/ops/populate/argmax_populate.cc
  8. +1
    -0
      mindspore/lite/src/ops/populate/argmin_populate.cc
  9. +0
    -118
      mindspore/lite/src/runtime/kernel/arm/base/arg_min_max_base.cc
  10. +0
    -41
      mindspore/lite/src/runtime/kernel/arm/base/arg_min_max_base.h
  11. +66
    -12
      mindspore/lite/src/runtime/kernel/arm/fp32/argminmax_fp32.cc
  12. +11
    -3
      mindspore/lite/src/runtime/kernel/arm/fp32/argminmax_fp32.h
  13. +0
    -107
      mindspore/lite/src/runtime/kernel/arm/fp32/instance_norm_fp32.cc
  14. +0
    -53
      mindspore/lite/src/runtime/kernel/arm/fp32/instance_norm_fp32.h
  15. +17
    -9
      mindspore/lite/src/runtime/kernel/arm/int8/argminmax_int8.cc
  16. +6
    -3
      mindspore/lite/src/runtime/kernel/arm/int8/argminmax_int8.h
  17. +0
    -291
      mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/argminmax_fp32_test.cc
  18. +0
    -134
      mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/instance_norm_fp32_tests.cc

+ 0
- 102
mindspore/lite/nnacl/arg_min_max.c View File

@@ -1,102 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "nnacl/arg_min_max.h"
#include "nnacl/fp32/arg_min_max_fp32.h"

#define FLOAT_DATA_TYPE 43

void GetCalcParameter(const int *shape, int dims_number, int axis, int *pre_axis_count, int *axis_count,
int *after_axis_count) {
*pre_axis_count = 1;
for (int i = 0; i < axis; ++i) {
*pre_axis_count = (*pre_axis_count) * shape[i];
}

*axis_count = shape[axis];

*after_axis_count = 1;
for (int i = axis + 1; i < dims_number; ++i) {
*after_axis_count = (*after_axis_count) * shape[i];
}
}

void ArgMinMaxTopk1(const void *input, void *output, const int *shape, const ArgMinMaxParameter *param) {
int pre_axis_count = 1;
int axis_count = 1;
int after_axis_count = 1;
GetCalcParameter(shape, param->dims_size_, param->axis_, &pre_axis_count, &axis_count, &after_axis_count);
if (param->data_type_ != FLOAT_DATA_TYPE) {
return;
}
if (param->get_max_) {
ArgMax(input, output, param, pre_axis_count, axis_count, after_axis_count);
} else {
ArgMin(input, output, param, pre_axis_count, axis_count, after_axis_count);
}
}

void ArgMinMaxTopknFp32(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->get_max_) {
switch (param->axis_) {
case 0:
ArgMaxDim0(input, output, in_shape, param);
break;
case 1:
ArgMaxDim1(input, output, in_shape, param);
break;
case 2:
ArgMaxDim2(input, output, in_shape, param);
break;
case 3:
ArgMaxDim3(input, output, in_shape, param);
break;
}
} else {
switch (param->axis_) {
case 0:
ArgMinDim0(input, output, in_shape, param);
break;
case 1:
ArgMinDim1(input, output, in_shape, param);
break;
case 2:
ArgMinDim2(input, output, in_shape, param);
break;
case 3:
ArgMinDim3(input, output, in_shape, param);
break;
}
}
}

void ArgMinMax(const void *input, void *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->topk_ == 1) {
ArgMinMaxTopk1(input, output, in_shape, param);
return;
}

switch (param->data_type_) {
case FLOAT_DATA_TYPE: {
ArgMinMaxTopknFp32(input, output, in_shape, param);
return;
}
default:
break;
}
}

#undef FLOAT_DATA_TYPE
#undef INT8_DATA_TYPE

+ 0
- 29
mindspore/lite/nnacl/arg_min_max.h View File

@@ -1,29 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef MINDSPORE_LITE_NNACL_ARG_MIN_MAX_H_
#define MINDSPORE_LITE_NNACL_ARG_MIN_MAX_H_

#include "nnacl/arg_min_max_parameter.h"

#ifdef __cplusplus
extern "C" {
#endif
void ArgMinMax(const void *input, void *output, const int *in_shape, const ArgMinMaxParameter *param);
#ifdef __cplusplus
}
#endif

#endif // MINDSPORE_LITE_NNACL_ARG_MIN_MAX_H_

+ 112
- 358
mindspore/lite/nnacl/fp32/arg_min_max_fp32.c View File

@@ -43,87 +43,91 @@ int ArgCompareDescFp32(const void *a, const void *b) {
return 0;
}

void ArgMaxDim0OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
for (int32_t i = 0; i < param->in_strides_[0]; ++i) {
for (int j = 0; j < in_shape[0]; ++j) {
size_t offset = param->in_strides_[0] * j + i;
param->arg_elements_[j].index_ = j;
param->arg_elements_[j].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape[0], sizeof(ArgElement), ArgCompareDescFp32);
for (int j = 0; j < param->topk_; ++j) {
size_t out_offset = j * param->out_strides_[0] + i;
output[out_offset] = param->arg_elements_[j].data_.f_data_;
void ArgMaxTopK1(const float *input, float *output, float *output_value, const ArgMinMaxParameter *param,
int pre_axis_count, int axis_count, int after_axis_count) {
bool out_value = param->out_value_;
for (int i = 0; i < pre_axis_count; ++i) {
size_t output_offset = i * after_axis_count;
size_t input_offset = output_offset * axis_count;
for (int j = 0; j < after_axis_count; ++j) {
float value = -FLT_MAX;
float index = 0.0f;
for (int k = 0; k < axis_count; ++k) {
float value_tmp = input[input_offset + k * after_axis_count + j];
if (value_tmp > value) {
value = value_tmp;
index = k;
}
}
output[output_offset + j] = out_value ? value : index;
if (output_value != NULL) {
output_value[output_offset + j] = value;
}
}
}
}

void ArgMaxDim0OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
for (int32_t i = 0; i < param->in_strides_[0]; ++i) {
for (int j = 0; j < in_shape[0]; ++j) {
size_t offset = param->in_strides_[0] * j + i;
param->arg_elements_[j].index_ = j;
param->arg_elements_[j].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape[0], sizeof(ArgElement), ArgCompareDescFp32);
for (int j = 0; j < param->topk_; ++j) {
size_t out_offset = j * param->out_strides_[0] + i;
output[out_offset] = param->arg_elements_[j].index_;
void ArgMinTopK1(const float *input, float *output, float *output_value, const ArgMinMaxParameter *param,
int pre_axis_count, int axis_count, int after_axis_count) {
bool out_value = param->out_value_;
for (int i = 0; i < pre_axis_count; ++i) {
size_t output_offset = i * after_axis_count;
size_t input_offset = output_offset * axis_count;
for (int j = 0; j < after_axis_count; ++j) {
float value = FLT_MAX;
float index = 0.0f;
for (int k = 0; k < axis_count; ++k) {
float value_tmp = input[input_offset + k * after_axis_count + j];
if (value_tmp < value) {
value = value_tmp;
index = k;
}
}
output[output_offset + j] = out_value ? value : index;
if (output_value != NULL) {
output_value[output_offset + j] = value;
}
}
}
}

void ArgMinDim0OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
for (int32_t i = 0; i < param->in_strides_[0]; ++i) {
for (int j = 0; j < in_shape[0]; ++j) {
size_t offset = param->in_strides_[0] * j + i;
param->arg_elements_[j].index_ = j;
param->arg_elements_[j].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape[0], sizeof(ArgElement), ArgCompareAscFp32);
for (int j = 0; j < param->topk_; ++j) {
size_t out_offset = j * param->out_strides_[0] + i;
output[out_offset] = param->arg_elements_[j].data_.f_data_;
}
void GetCalcParameter(const int *shape, int dims_number, int axis, int *pre_axis_count, int *axis_count,
int *after_axis_count) {
*pre_axis_count = 1;
for (int i = 0; i < axis; ++i) {
*pre_axis_count = (*pre_axis_count) * shape[i];
}
*axis_count = shape[axis];
*after_axis_count = 1;
for (int i = axis + 1; i < dims_number; ++i) {
*after_axis_count = (*after_axis_count) * shape[i];
}
}

void ArgMinDim0OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
void ArgMinMaxDim0(const float *input, float *output, float *output_value, const int *in_shape,
const ArgMinMaxParameter *param, COMPARE_FUNCTION compare_func) {
for (int32_t i = 0; i < param->in_strides_[0]; ++i) {
for (int j = 0; j < in_shape[0]; ++j) {
size_t offset = param->in_strides_[0] * j + i;
param->arg_elements_[j].index_ = j;
param->arg_elements_[j].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape[0], sizeof(ArgElement), ArgCompareAscFp32);
qsort(param->arg_elements_, in_shape[0], sizeof(ArgElement), *compare_func);
for (int j = 0; j < param->topk_; ++j) {
size_t out_offset = j * param->out_strides_[0] + i;
output[out_offset] = param->arg_elements_[j].index_;
}
}
}

void ArgMaxDim1OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < param->in_strides_[1]; ++j) {
for (int k = 0; k < in_shape1; ++k) {
size_t offset = param->in_strides_[1] * k + in_dim0_offset + j;
param->arg_elements_[k].index_ = k;
param->arg_elements_[k].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape1, sizeof(ArgElement), ArgCompareDescFp32);
for (int k = 0; k < param->topk_; ++k) {
size_t out_offset = out_dim0_offset + j + k * param->out_strides_[1];
output[out_offset] = param->arg_elements_[k].data_.f_data_;
output[out_offset] = param->out_value_ ? param->arg_elements_[j].data_.f_data_ : param->arg_elements_[j].index_;
if (output_value != NULL) {
output_value[out_offset] = param->arg_elements_[j].data_.f_data_;
}
}
}
return;
}

void ArgMaxDim1OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
void ArgMinMaxDim1(const float *input, float *output, float *output_value, const int *in_shape,
const ArgMinMaxParameter *param, COMPARE_FUNCTION compare_func) {
int in_shape1 = in_shape[1];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
@@ -134,81 +138,21 @@ void ArgMaxDim1OutIndex(const float *input, float *output, const int *in_shape,
param->arg_elements_[k].index_ = k;
param->arg_elements_[k].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape1, sizeof(ArgElement), ArgCompareDescFp32);
qsort(param->arg_elements_, in_shape1, sizeof(ArgElement), *compare_func);
for (int k = 0; k < param->topk_; ++k) {
size_t out_offset = out_dim0_offset + j + k * param->out_strides_[1];
output[out_offset] = param->arg_elements_[k].index_;
}
}
}
}

void ArgMinDim1OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < param->in_strides_[1]; ++j) {
for (int k = 0; k < in_shape1; ++k) {
size_t offset = param->in_strides_[1] * k + in_dim0_offset + j;
param->arg_elements_[k].index_ = k;
param->arg_elements_[k].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape1, sizeof(ArgElement), ArgCompareAscFp32);
for (int k = 0; k < param->topk_; ++k) {
size_t out_offset = out_dim0_offset + j + k * param->out_strides_[1];
output[out_offset] = param->arg_elements_[k].data_.f_data_;
}
}
}
}

void ArgMinDim1OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < param->in_strides_[1]; ++j) {
for (int k = 0; k < in_shape1; ++k) {
size_t offset = param->in_strides_[1] * k + in_dim0_offset + j;
param->arg_elements_[k].index_ = k;
param->arg_elements_[k].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape1, sizeof(ArgElement), ArgCompareAscFp32);
for (int k = 0; k < param->topk_; ++k) {
size_t out_offset = out_dim0_offset + j + k * param->out_strides_[1];
output[out_offset] = param->arg_elements_[k].index_;
}
}
}
}

void ArgMaxDim2OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < in_shape1; ++j) {
size_t in_dim1_offset = j * param->in_strides_[1] + in_dim0_offset;
size_t out_dim1_offset = j * param->out_strides_[1] + out_dim0_offset;
for (int k = 0; k < param->in_strides_[2]; ++k) {
for (int l = 0; l < in_shape2; ++l) {
size_t offset = param->in_strides_[2] * l + k + in_dim1_offset;
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape2, sizeof(ArgElement), ArgCompareDescFp32);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim1_offset + k + l * param->out_strides_[2];
output[out_offset] = param->arg_elements_[l].data_.f_data_;
output[out_offset] = param->out_value_ ? param->arg_elements_[k].data_.f_data_ : param->arg_elements_[k].index_;
if (output_value != NULL) {
output_value[out_offset] = param->arg_elements_[k].data_.f_data_;
}
}
}
}
return;
}

void ArgMaxDim2OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
void ArgMinMaxDim2(const float *input, float *output, float *output_value, const int *in_shape,
const ArgMinMaxParameter *param, COMPARE_FUNCTION compare_func) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
for (int i = 0; i < in_shape[0]; ++i) {
@@ -223,67 +167,23 @@ void ArgMaxDim2OutIndex(const float *input, float *output, const int *in_shape,
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape2, sizeof(ArgElement), ArgCompareDescFp32);
qsort(param->arg_elements_, in_shape2, sizeof(ArgElement), *compare_func);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim1_offset + k + l * param->out_strides_[2];
output[out_offset] = param->arg_elements_[l].index_;
}
}
}
}
}

void ArgMinDim2OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < in_shape1; ++j) {
size_t in_dim1_offset = j * param->in_strides_[1] + in_dim0_offset;
size_t out_dim1_offset = j * param->out_strides_[1] + out_dim0_offset;
for (int k = 0; k < param->in_strides_[2]; ++k) {
for (int l = 0; l < in_shape2; ++l) {
size_t offset = param->in_strides_[2] * l + k + in_dim1_offset;
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape2, sizeof(ArgElement), ArgCompareAscFp32);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim1_offset + k + l * param->out_strides_[2];
output[out_offset] = param->arg_elements_[l].data_.f_data_;
output[out_offset] =
param->out_value_ ? param->arg_elements_[l].data_.f_data_ : param->arg_elements_[l].index_;
if (output_value != NULL) {
output_value[out_offset] = param->arg_elements_[l].data_.f_data_;
}
}
}
}
}
}

void ArgMinDim2OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < in_shape1; ++j) {
size_t in_dim1_offset = j * param->in_strides_[1] + in_dim0_offset;
size_t out_dim1_offset = j * param->out_strides_[1] + out_dim0_offset;
for (int k = 0; k < param->in_strides_[2]; ++k) {
for (int l = 0; l < in_shape2; ++l) {
size_t offset = param->in_strides_[2] * l + k + in_dim1_offset;
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape2, sizeof(ArgElement), ArgCompareAscFp32);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim1_offset + k + l * param->out_strides_[2];
output[out_offset] = param->arg_elements_[l].index_;
}
}
}
}
}

void ArgMaxDim3OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
void ArgMinMaxDim3(const float *input, float *output, float *output_value, const int *in_shape,
const ArgMinMaxParameter *param, COMPARE_FUNCTION compare_func) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
int in_shape3 = in_shape[3];
@@ -301,202 +201,56 @@ void ArgMaxDim3OutValue(const float *input, float *output, const int *in_shape,
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape3, sizeof(ArgElement), ArgCompareDescFp32);
qsort(param->arg_elements_, in_shape3, sizeof(ArgElement), *compare_func);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim2_offset + l;
output[out_offset] = param->arg_elements_[l].data_.f_data_;
output[out_offset] =
param->out_value_ ? param->arg_elements_[l].data_.f_data_ : param->arg_elements_[l].index_;
if (output_value != NULL) {
output_value[out_offset] = param->arg_elements_[l].data_.f_data_;
}
}
}
}
}
}

void ArgMaxDim3OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
int in_shape3 = in_shape[3];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < in_shape1; ++j) {
size_t in_dim1_offset = j * param->in_strides_[1] + in_dim0_offset;
size_t out_dim1_offset = j * param->out_strides_[1] + out_dim0_offset;
for (int k = 0; k < in_shape2; ++k) {
size_t in_dim2_offset = k * param->in_strides_[2] + in_dim1_offset;
size_t out_dim2_offset = k * param->out_strides_[2] + out_dim1_offset;
for (int l = 0; l < in_shape3; ++l) {
size_t offset = l + in_dim2_offset;
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape3, sizeof(ArgElement), ArgCompareDescFp32);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim2_offset + l;
output[out_offset] = param->arg_elements_[l].index_;
}
}
}
}
}

void ArgMinDim3OutValue(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
int in_shape3 = in_shape[3];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < in_shape1; ++j) {
size_t in_dim1_offset = j * param->in_strides_[1] + in_dim0_offset;
size_t out_dim1_offset = j * param->out_strides_[1] + out_dim0_offset;
for (int k = 0; k < in_shape2; ++k) {
size_t in_dim2_offset = k * param->in_strides_[2] + in_dim1_offset;
size_t out_dim2_offset = k * param->out_strides_[2] + out_dim1_offset;
for (int l = 0; l < in_shape3; ++l) {
size_t offset = l + in_dim2_offset;
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape3, sizeof(ArgElement), ArgCompareAscFp32);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim2_offset + l;
output[out_offset] = param->arg_elements_[l].data_.f_data_;
}
}
}
}
}
void ArgMinMaxFp32(const float *input, float *output, float *output_value, const int *in_shape,
const ArgMinMaxParameter *param) {
if (param->topk_ == 1) {
int pre_axis_count = 1;
int axis_count = 1;
int after_axis_count = 1;
GetCalcParameter(in_shape, param->dims_size_, param->axis_, &pre_axis_count, &axis_count, &after_axis_count);

void ArgMinDim3OutIndex(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
int in_shape1 = in_shape[1];
int in_shape2 = in_shape[2];
int in_shape3 = in_shape[3];
for (int i = 0; i < in_shape[0]; ++i) {
size_t in_dim0_offset = i * param->in_strides_[0];
size_t out_dim0_offset = i * param->out_strides_[0];
for (int j = 0; j < in_shape1; ++j) {
size_t in_dim1_offset = j * param->in_strides_[1] + in_dim0_offset;
size_t out_dim1_offset = j * param->out_strides_[1] + out_dim0_offset;
for (int k = 0; k < in_shape2; ++k) {
size_t in_dim2_offset = k * param->in_strides_[2] + in_dim1_offset;
size_t out_dim2_offset = k * param->out_strides_[2] + out_dim1_offset;
for (int l = 0; l < in_shape3; ++l) {
size_t offset = l + in_dim2_offset;
param->arg_elements_[l].index_ = l;
param->arg_elements_[l].data_.f_data_ = input[offset];
}
qsort(param->arg_elements_, in_shape3, sizeof(ArgElement), ArgCompareAscFp32);
for (int l = 0; l < param->topk_; ++l) {
size_t out_offset = out_dim2_offset + l;
output[out_offset] = param->arg_elements_[l].index_;
}
}
if (param->get_max_) {
ArgMaxTopK1(input, output, output_value, param, pre_axis_count, axis_count, after_axis_count);
} else {
ArgMinTopK1(input, output, output_value, param, pre_axis_count, axis_count, after_axis_count);
}
return;
}
}

void ArgMaxDim0(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMaxDim0OutValue(input, output, in_shape, param);
} else {
ArgMaxDim0OutIndex(input, output, in_shape, param);
}
}

void ArgMinDim0(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMinDim0OutValue(input, output, in_shape, param);
COMPARE_FUNCTION compare_function = NULL;
if (param->get_max_) {
compare_function = ArgCompareDescFp32;
} else {
ArgMinDim0OutIndex(input, output, in_shape, param);
compare_function = ArgCompareAscFp32;
}
}

void ArgMaxDim1(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMaxDim1OutValue(input, output, in_shape, param);
} else {
ArgMaxDim1OutIndex(input, output, in_shape, param);
}
}

void ArgMinDim1(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMinDim1OutValue(input, output, in_shape, param);
} else {
ArgMinDim1OutIndex(input, output, in_shape, param);
}
}

void ArgMaxDim2(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMaxDim2OutValue(input, output, in_shape, param);
} else {
ArgMaxDim2OutIndex(input, output, in_shape, param);
}
}

void ArgMinDim2(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMinDim2OutValue(input, output, in_shape, param);
} else {
ArgMinDim2OutIndex(input, output, in_shape, param);
}
}

void ArgMaxDim3(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMaxDim3OutValue(input, output, in_shape, param);
} else {
ArgMaxDim3OutIndex(input, output, in_shape, param);
}
}

void ArgMinDim3(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param) {
if (param->out_value_) {
ArgMinDim3OutValue(input, output, in_shape, param);
} else {
ArgMinDim3OutIndex(input, output, in_shape, param);
}
}

void ArgMax(const float *input, float *output, const ArgMinMaxParameter *param, int pre_axis_count, int axis_count,
int after_axis_count) {
bool out_value = param->out_value_;
for (int i = 0; i < pre_axis_count; ++i) {
size_t output_offset = i * after_axis_count;
size_t input_offset = output_offset * axis_count;
for (int j = 0; j < after_axis_count; ++j) {
float value = -FLT_MAX;
float index = 0.0f;
for (int k = 0; k < axis_count; ++k) {
float value_tmp = input[input_offset + k * after_axis_count + j];
if (value_tmp > value) {
value = value_tmp;
index = k;
}
}
output[output_offset + j] = out_value ? value : index;
}
}
}

void ArgMin(const float *input, float *output, const ArgMinMaxParameter *param, int pre_axis_count, int axis_count,
int after_axis_count) {
bool out_value = param->out_value_;
for (int i = 0; i < pre_axis_count; ++i) {
size_t output_offset = i * after_axis_count;
size_t input_offset = output_offset * axis_count;
for (int j = 0; j < after_axis_count; ++j) {
float value = FLT_MAX;
float index = 0.0f;
for (int k = 0; k < axis_count; ++k) {
float value_tmp = input[input_offset + k * after_axis_count + j];
if (value_tmp < value) {
value = value_tmp;
index = k;
}
}
output[output_offset + j] = out_value ? value : index;
}
switch (param->axis_) {
case 0:
ArgMinMaxDim0(input, output, output_value, in_shape, param, compare_function);
break;
case 1:
ArgMinMaxDim1(input, output, output_value, in_shape, param, compare_function);
break;
case 2:
ArgMinMaxDim2(input, output, output_value, in_shape, param, compare_function);
break;
case 3:
ArgMinMaxDim3(input, output, output_value, in_shape, param, compare_function);
break;
}
return;
}

+ 4
- 12
mindspore/lite/nnacl/fp32/arg_min_max_fp32.h View File

@@ -18,21 +18,13 @@

#include "nnacl/arg_min_max_parameter.h"

typedef int (*COMPARE_FUNCTION)(const void *a, const void *b);

#ifdef __cplusplus
extern "C" {
#endif
void ArgMax(const float *input, float *output, const ArgMinMaxParameter *param, int pre_axis_count, int axis_count,
int after_axis_count);
void ArgMin(const float *input, float *output, const ArgMinMaxParameter *param, int pre_axis_count, int axis_count,
int after_axis_count);
void ArgMaxDim0(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMinDim0(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMaxDim1(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMinDim1(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMaxDim2(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMinDim2(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMaxDim3(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMinDim3(const float *input, float *output, const int *in_shape, const ArgMinMaxParameter *param);
void ArgMinMaxFp32(const float *input, float *output, float *output_value, const int *in_shape,
const ArgMinMaxParameter *param);
#ifdef __cplusplus
}
#endif


+ 0
- 47
mindspore/lite/nnacl/fp32/instance_norm_fp32.c View File

@@ -1,47 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "nnacl/fp32/instance_norm_fp32.h"
#include <math.h>
#include "nnacl/errorcode.h"
#include "nnacl/op_base.h"

int InstanceNorm(int outer_size, int inner_size, const float *src_data, const float *scale_data, const float *bias_data,
const InstanceNormParameter *param, float *dst_data, int task_id, int thread_num) {
if (src_data == NULL || dst_data == NULL || scale_data == NULL || bias_data == NULL) {
return NNACL_NULL_PTR;
}
for (int j = task_id; j < outer_size; j += thread_num) {
int offset = (j / param->channel_) * inner_size * param->channel_;
const float *src = src_data + offset;
float *dst = dst_data + offset;
float mean = 0.0f;
float square_mean = 0.0f;
for (int i = 0; i < inner_size; i++) {
int idx = j % param->channel_ + i * param->channel_;
mean += src[idx];
square_mean += src[idx] * src[idx];
}
mean /= (float)inner_size;
square_mean /= (float)inner_size;
const float deno = 1 / sqrtf(square_mean - mean * mean + param->epsilon_);
for (int i = 0; i < inner_size; ++i) {
int idx = j % param->channel_ + i * param->channel_;
int scale_idx = (j / param->channel_) * param->channel_ + j % param->channel_;
dst[idx] = ((src[idx] - mean) * deno) * scale_data[scale_idx] + bias_data[scale_idx];
}
}
return NNACL_OK;
}

+ 0
- 32
mindspore/lite/nnacl/fp32/instance_norm_fp32.h View File

@@ -1,32 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef MINDSPORE_LITE_NNACL_FP32_INSTANCE_NORM_H_
#define MINDSPORE_LITE_NNACL_FP32_INSTANCE_NORM_H_

#include "nnacl/op_base.h"
#include "nnacl/instance_norm_parameter.h"

#ifdef __cplusplus
extern "C" {
#endif

int InstanceNorm(int outer_size, int inner_size, const float *src_data, const float *scale_data, const float *bias_data,
const InstanceNormParameter *param, float *dst_data, int task_id, int thread_num);
#ifdef __cplusplus
}
#endif

#endif // MINDSPORE_LITE_NNACL_FP32_INSTANCE_NORM_H_

+ 1
- 0
mindspore/lite/src/ops/populate/argmax_populate.cc View File

@@ -35,6 +35,7 @@ OpParameter *PopulateArgMaxParameter(const mindspore::lite::PrimitiveC *primitiv
arg_param->axis_type_ = param->GetAxisType();
arg_param->out_value_ = param->GetOutMaxValue();
arg_param->keep_dims_ = param->GetKeepDims();
arg_param->get_max_ = true;
return reinterpret_cast<OpParameter *>(arg_param);
}



+ 1
- 0
mindspore/lite/src/ops/populate/argmin_populate.cc View File

@@ -35,6 +35,7 @@ OpParameter *PopulateArgMinParameter(const mindspore::lite::PrimitiveC *primitiv
arg_param->axis_type_ = param->GetAxisType();
arg_param->out_value_ = param->GetOutMaxValue();
arg_param->keep_dims_ = param->GetKeepDims();
arg_param->get_max_ = false;
return reinterpret_cast<OpParameter *>(arg_param);
}



+ 0
- 118
mindspore/lite/src/runtime/kernel/arm/base/arg_min_max_base.cc View File

@@ -1,118 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "src/runtime/kernel/arm/base/arg_min_max_base.h"
#include "nnacl/arg_min_max.h"
#include "src/runtime/kernel/arm/fp32/argminmax_fp32.h"
#include "nnacl/arithmetic_common.h"
#include "schema/model_generated.h"
#include "src/kernel_registry.h"
#include "include/errorcode.h"
#include "include/context.h"

using mindspore::lite::KernelRegistrar;
using mindspore::lite::RET_ERROR;
using mindspore::lite::RET_FORMAT_ERR;
using mindspore::lite::RET_OK;
using mindspore::lite::RET_PARAM_INVALID;
using mindspore::schema::PrimitiveType_ArgMax;
using mindspore::schema::PrimitiveType_ArgMin;

namespace mindspore::kernel {
int ArgMinMaxBaseCPUKernel::Init() {
auto param = reinterpret_cast<ArgMinMaxParameter *>(op_parameter_);
switch (op_parameter_->type_) {
case PrimitiveType_ArgMax:
param->get_max_ = true;
break;
case PrimitiveType_ArgMin:
param->get_max_ = false;
break;
default:
MS_LOG(ERROR) << "Unexpected type " << op_parameter_->type_;
return RET_ERROR;
}

return RET_OK;
}

int ArgMinMaxBaseCPUKernel::ReSize() {
auto in_shape = in_tensors_.at(0)->shape();
auto dims_size = in_shape.size();
auto param = reinterpret_cast<ArgMinMaxParameter *>(op_parameter_);
int axis = param->axis_ < 0 ? param->axis_ + dims_size : param->axis_;
param->axis_ = axis;
param->dims_size_ = dims_size;
if (param->topk_ <= 0) {
MS_LOG(ERROR) << "Invalid topk " << param->topk_;
return RET_PARAM_INVALID;
}
param->topk_ = MSMIN(param->topk_, in_shape.at(axis));
ComputeStrides(in_shape.data(), param->in_strides_, in_shape.size());
auto out_shape = out_tensors_.at(0)->shape();
ComputeStrides(out_shape.data(), param->out_strides_, out_shape.size());
return RET_OK;
}

int ArgMinMaxBaseCPUKernel::Run() {
auto input_data = in_tensors_.at(0)->MutableData();
auto output_data = out_tensors_.at(0)->MutableData();

auto shape = in_tensors_.at(0)->shape();

auto param = reinterpret_cast<ArgMinMaxParameter *>(op_parameter_);
MS_ASSERT(context_->allocator != nullptr);
if (param->topk_ > 1 || param->keep_dims_) {
param->arg_elements_ =
reinterpret_cast<ArgElement *>(context_->allocator->Malloc(sizeof(ArgElement) * shape[param->axis_]));
if (param->arg_elements_ == nullptr) {
MS_LOG(ERROR) << "malloc memroy fail!";
return RET_ERROR;
}
}
ArgMinMax(input_data, output_data, reinterpret_cast<const int *>(shape.data()), param);
context_->allocator->Free(param->arg_elements_);
param->arg_elements_ = nullptr;
return RET_OK;
}

kernel::LiteKernel *CpuArgMinMaxFp32KernelCreator(const std::vector<lite::Tensor *> &inputs,
const std::vector<lite::Tensor *> &outputs, OpParameter *op_parameter,
const lite::InnerContext *ctx, const kernel::KernelKey &desc,
const mindspore::lite::PrimitiveC *primitive) {
if (op_parameter == nullptr) {
MS_LOG(ERROR) << "Input op_parameter is nullptr!";
return nullptr;
}
auto kernel = new (std::nothrow) ArgMinMaxCPUKernel(op_parameter, inputs, outputs, ctx, primitive);
if (kernel == nullptr) {
MS_LOG(ERROR) << "new ArgMinMaxCPUKernel fail!";
free(op_parameter);
return nullptr;
}
auto ret = kernel->Init();
if (ret != RET_OK) {
MS_LOG(ERROR) << "Init kernel failed, name: " << op_parameter->name_ << ", type: "
<< schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(op_parameter->type_));
delete kernel;
return nullptr;
}
return kernel;
}

REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_ArgMax, CpuArgMinMaxFp32KernelCreator)
REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_ArgMin, CpuArgMinMaxFp32KernelCreator)

} // namespace mindspore::kernel

+ 0
- 41
mindspore/lite/src/runtime/kernel/arm/base/arg_min_max_base.h View File

@@ -1,41 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_BASE_ARG_MIN_MAX_BASE_H_
#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_BASE_ARG_MIN_MAX_BASE_H_

#include <vector>
#include "src/lite_kernel.h"

namespace mindspore::kernel {
class ArgMinMaxBaseCPUKernel : public LiteKernel {
public:
ArgMinMaxBaseCPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
const mindspore::lite::PrimitiveC *primitive)
: LiteKernel(parameter, inputs, outputs, ctx, primitive) {}

virtual ~ArgMinMaxBaseCPUKernel() = default;

int Init() override;
int ReSize() override;
int Run() override;

private:
};
} // namespace mindspore::kernel

#endif // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_BASE_ARG_MIN_MAX_BASE_H_

+ 66
- 12
mindspore/lite/src/runtime/kernel/arm/fp32/argminmax_fp32.cc View File

@@ -15,11 +15,8 @@
*/

#include "src/runtime/kernel/arm/fp32/argminmax_fp32.h"
#include <vector>
#include "schema/model_generated.h"
#include "src/kernel_registry.h"
#include "nnacl/arg_min_max.h"
#include "include/errorcode.h"

using mindspore::kernel::KERNEL_ARCH::kCPU;
using mindspore::lite::KernelRegistrar;
@@ -30,22 +27,79 @@ using mindspore::schema::PrimitiveType_ArgMin;

namespace mindspore::kernel {
int ArgMinMaxCPUKernel::Init() {
auto ret = ArgMinMaxBaseCPUKernel::Init();
if (ret != RET_OK) {
return ret;
}
auto param = reinterpret_cast<ArgMinMaxParameter *>(op_parameter_);
param->data_type_ = kNumberTypeFloat32;
arg_param_->data_type_ = kNumberTypeFloat32;
if (!InferShapeDone()) {
return RET_OK;
}
return ReSize();
}

int ArgMinMaxCPUKernel::ReSize() { return ArgMinMaxBaseCPUKernel::ReSize(); }
int ArgMinMaxCPUKernel::ReSize() {
auto in_shape = in_tensors_.at(0)->shape();
auto dims_size = in_shape.size();
int axis = arg_param_->axis_ < 0 ? arg_param_->axis_ + dims_size : arg_param_->axis_;
arg_param_->axis_ = axis;
arg_param_->dims_size_ = dims_size;
if (arg_param_->topk_ <= 0) {
MS_LOG(ERROR) << "Invalid topk " << arg_param_->topk_;
return RET_ERROR;
}
arg_param_->topk_ = MSMIN(arg_param_->topk_, in_shape.at(axis));
ComputeStrides(in_shape.data(), arg_param_->in_strides_, in_shape.size());
auto out_shape = out_tensors_.at(0)->shape();
ComputeStrides(out_shape.data(), arg_param_->out_strides_, out_shape.size());
return RET_OK;
}

int ArgMinMaxCPUKernel::Run() {
auto ret = ArgMinMaxBaseCPUKernel::Run();
return ret;
float *input_data = reinterpret_cast<float *>(in_tensors_.at(0)->data_c());
float *output_data = reinterpret_cast<float *>(out_tensors_.at(0)->data_c());
float *output_value = nullptr;
if (out_tensors_.size() == 2) {
output_value = reinterpret_cast<float *>(out_tensors_.at(1)->data_c());
}

auto shape = in_tensors_.at(0)->shape();

MS_ASSERT(context_->allocator != nullptr);
if (arg_param_->topk_ > 1 || arg_param_->keep_dims_) {
arg_param_->arg_elements_ =
reinterpret_cast<ArgElement *>(context_->allocator->Malloc(sizeof(ArgElement) * shape[arg_param_->axis_]));
if (arg_param_->arg_elements_ == nullptr) {
MS_LOG(ERROR) << "malloc memroy fail!";
return RET_ERROR;
}
}
ArgMinMaxFp32(input_data, output_data, output_value, reinterpret_cast<const int *>(shape.data()), arg_param_);
context_->allocator->Free(arg_param_->arg_elements_);
arg_param_->arg_elements_ = nullptr;
return RET_OK;
}

kernel::LiteKernel *CpuArgMinMaxFp32KernelCreator(const std::vector<lite::Tensor *> &inputs,
const std::vector<lite::Tensor *> &outputs, OpParameter *op_parameter,
const lite::InnerContext *ctx, const kernel::KernelKey &desc,
const mindspore::lite::PrimitiveC *primitive) {
if (op_parameter == nullptr) {
MS_LOG(ERROR) << "Input op_parameter is nullptr!";
return nullptr;
}
auto kernel = new (std::nothrow) ArgMinMaxCPUKernel(op_parameter, inputs, outputs, ctx, primitive);
if (kernel == nullptr) {
MS_LOG(ERROR) << "new ArgMinMaxCPUKernel fail!";
free(op_parameter);
return nullptr;
}
auto ret = kernel->Init();
if (ret != RET_OK) {
MS_LOG(ERROR) << "Init kernel failed, name: " << op_parameter->name_ << ", type: "
<< schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(op_parameter->type_));
delete kernel;
return nullptr;
}
return kernel;
}

REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_ArgMax, CpuArgMinMaxFp32KernelCreator)
REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_ArgMin, CpuArgMinMaxFp32KernelCreator)
} // namespace mindspore::kernel

+ 11
- 3
mindspore/lite/src/runtime/kernel/arm/fp32/argminmax_fp32.h View File

@@ -17,21 +17,29 @@
#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_ARGMINMAX_H_

#include <vector>
#include "src/runtime/kernel/arm/base/arg_min_max_base.h"
#include "include/errorcode.h"
#include "nnacl/fp32/arg_min_max_fp32.h"
#include "nnacl/arithmetic_common.h"
#include "src/lite_kernel.h"

namespace mindspore::kernel {
class ArgMinMaxCPUKernel : public ArgMinMaxBaseCPUKernel {
class ArgMinMaxCPUKernel : public LiteKernel {
public:
ArgMinMaxCPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
const mindspore::lite::PrimitiveC *primitive)
: ArgMinMaxBaseCPUKernel(parameter, inputs, outputs, ctx, primitive) {}
: LiteKernel(parameter, inputs, outputs, ctx, primitive) {
arg_param_ = reinterpret_cast<ArgMinMaxParameter *>(op_parameter_);
}

~ArgMinMaxCPUKernel() = default;

int Init() override;
int ReSize() override;
int Run() override;

private:
ArgMinMaxParameter *arg_param_;
};
} // namespace mindspore::kernel



+ 0
- 107
mindspore/lite/src/runtime/kernel/arm/fp32/instance_norm_fp32.cc View File

@@ -1,107 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "src/runtime/kernel/arm/fp32/instance_norm_fp32.h"
#include <vector>
#include "schema/model_generated.h"
#include "src/kernel_registry.h"
#include "include/errorcode.h"

using mindspore::kernel::KERNEL_ARCH::kCPU;
using mindspore::lite::KernelRegistrar;
using mindspore::lite::RET_ERROR;
using mindspore::lite::RET_OK;
using mindspore::schema::PrimitiveType_InstanceNorm;

namespace mindspore::kernel {
int InstanceNormCPUKernel::Init() {
if (!InferShapeDone()) {
return RET_OK;
}
return ReSize();
}

int InstanceNormCPUKernel::ReSize() {
auto input_shapes = in_tensors_.front()->shape();
auto n_dim = input_shapes.size();
outer_size_ = input_shapes.at(0) * input_shapes.at(n_dim - 1);
inner_size_ = 1;
for (size_t i = 0; i < n_dim - 1; ++i) {
inner_size_ *= input_shapes.at(i);
}
param_->channel_ = input_shapes.at(n_dim - 1);
return RET_OK;
}

int InstanceNormCPUKernel::DoInstanceNorm(int task_id) {
int ret = InstanceNorm(outer_size_, inner_size_, src_data_, scale_data_, bias_data_, param_, dst_data_, task_id,
op_parameter_->thread_num_);
if (ret != RET_OK) {
MS_LOG(ERROR) << "DoInstanceNorm error error_code[" << ret << "]";
return ret;
}
return RET_OK;
}

int InstanceNormRun(void *cdata, int task_id) {
auto kernel = reinterpret_cast<InstanceNormCPUKernel *>(cdata);
auto ret = kernel->DoInstanceNorm(task_id);
if (ret != RET_OK) {
MS_LOG(ERROR) << "InstanceNormRun error task_id[" << task_id << "] error_code[" << ret << "]";
}
return ret;
}

int InstanceNormCPUKernel::Run() {
src_data_ = reinterpret_cast<float *>(in_tensors_.at(0)->MutableData());
scale_data_ = reinterpret_cast<float *>(in_tensors_.at(1)->MutableData());
bias_data_ = reinterpret_cast<float *>(in_tensors_.at(2)->MutableData());
dst_data_ = reinterpret_cast<float *>(out_tensors_.at(0)->MutableData());
auto ret = ParallelLaunch(this->context_->thread_pool_, InstanceNormRun, this, op_parameter_->thread_num_);
if (ret != RET_OK) {
MS_LOG(ERROR) << "FillRun error error_code[" << ret << "]";
return ret;
}
return RET_OK;
}

kernel::LiteKernel *CpuInstanceNormFp32KernelCreator(const std::vector<lite::Tensor *> &inputs,
const std::vector<lite::Tensor *> &outputs,
OpParameter *opParameter, const lite::InnerContext *ctx,
const kernel::KernelKey &desc,
const mindspore::lite::PrimitiveC *primitive) {
if (opParameter == nullptr) {
MS_LOG(ERROR) << "Create kernel failed, opParameter is nullptr, type: PrimitiveType_InstanceNorm. ";
return nullptr;
}
MS_ASSERT(desc.type == schema::PrimitiveType_InstanceNorm);
auto *kernel = new (std::nothrow) InstanceNormCPUKernel(opParameter, inputs, outputs, ctx, primitive);
if (kernel == nullptr) {
MS_LOG(ERROR) << "new InstanceNormCPUKernel fail!";
free(opParameter);
return nullptr;
}
auto ret = kernel->Init();
if (ret != RET_OK) {
MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
<< schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
delete kernel;
return nullptr;
}
return kernel;
}

REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_InstanceNorm, CpuInstanceNormFp32KernelCreator)
} // namespace mindspore::kernel

+ 0
- 53
mindspore/lite/src/runtime/kernel/arm/fp32/instance_norm_fp32.h View File

@@ -1,53 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_INSTANCE_NORM_H_
#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_INSTANCE_NORM_H_

#include <vector>
#include "src/lite_kernel.h"
#include "include/context.h"
#include "nnacl/fp32/instance_norm_fp32.h"

using mindspore::lite::InnerContext;

namespace mindspore::kernel {
class InstanceNormCPUKernel : public LiteKernel {
public:
InstanceNormCPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
const mindspore::lite::PrimitiveC *primitive)
: LiteKernel(parameter, inputs, outputs, ctx, primitive) {
param_ = reinterpret_cast<InstanceNormParameter *>(parameter);
}
~InstanceNormCPUKernel() override{};

int Init() override;
int ReSize() override;
int Run() override;
int DoInstanceNorm(int thread_id);

private:
InstanceNormParameter *param_ = nullptr;
int outer_size_;
int inner_size_;
float *src_data_ = nullptr;
float *dst_data_ = nullptr;
float *scale_data_ = nullptr;
float *bias_data_ = nullptr;
};
} // namespace mindspore::kernel

#endif // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_INSTANCE_NORM_H_

+ 17
- 9
mindspore/lite/src/runtime/kernel/arm/int8/argminmax_int8.cc View File

@@ -14,11 +14,8 @@
* limitations under the License.
*/
#include "src/runtime/kernel/arm/int8/argminmax_int8.h"
#include <vector>
#include "schema/model_generated.h"
#include "src/kernel_registry.h"
#include "nnacl/int8/arg_min_max_int8.h"
#include "include/errorcode.h"

using mindspore::lite::RET_ERROR;
using mindspore::lite::RET_OK;
@@ -31,10 +28,6 @@ using mindspore::schema::PrimitiveType_ArgMin;

namespace mindspore::kernel {
int ArgMinMaxInt8CPUKernel::Init() {
auto ret = ArgMinMaxBaseCPUKernel::Init();
if (ret != RET_OK) {
return ret;
}
auto param = reinterpret_cast<ArgMinMaxParameter *>(op_parameter_);
param->data_type_ = kNumberTypeInt8;
auto *input_tensor = in_tensors_.at(kInputIndex);
@@ -52,7 +45,23 @@ int ArgMinMaxInt8CPUKernel::Init() {
return ReSize();
}

int ArgMinMaxInt8CPUKernel::ReSize() { return ArgMinMaxBaseCPUKernel::ReSize(); }
int ArgMinMaxInt8CPUKernel::ReSize() {
auto in_shape = in_tensors_.at(0)->shape();
auto dims_size = in_shape.size();
auto param = reinterpret_cast<ArgMinMaxParameter *>(op_parameter_);
int axis = param->axis_ < 0 ? param->axis_ + dims_size : param->axis_;
param->axis_ = axis;
param->dims_size_ = dims_size;
if (param->topk_ <= 0) {
MS_LOG(ERROR) << "Invalid topk " << param->topk_;
return RET_ERROR;
}
param->topk_ = MSMIN(param->topk_, in_shape.at(axis));
ComputeStrides(in_shape.data(), param->in_strides_, in_shape.size());
auto out_shape = out_tensors_.at(0)->shape();
ComputeStrides(out_shape.data(), param->out_strides_, out_shape.size());
return RET_OK;
}

int ArgMinMaxInt8CPUKernel::Run() {
auto input = in_tensors_.at(0);
@@ -110,5 +119,4 @@ kernel::LiteKernel *CpuArgMinMaxInt8KernelCreator(const std::vector<lite::Tensor

REG_KERNEL(kCPU, kNumberTypeInt8, PrimitiveType_ArgMax, CpuArgMinMaxInt8KernelCreator)
REG_KERNEL(kCPU, kNumberTypeInt8, PrimitiveType_ArgMin, CpuArgMinMaxInt8KernelCreator)

} // namespace mindspore::kernel

+ 6
- 3
mindspore/lite/src/runtime/kernel/arm/int8/argminmax_int8.h View File

@@ -17,16 +17,19 @@
#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_ARGMINMAX_INT8_H_

#include <vector>
#include "src/runtime/kernel/arm/base/arg_min_max_base.h"
#include "nnacl/quantization/quantize.h"
#include "nnacl/int8/arg_min_max_int8.h"
#include "nnacl/arithmetic_common.h"
#include "include/errorcode.h"
#include "src/lite_kernel.h"

namespace mindspore::kernel {
class ArgMinMaxInt8CPUKernel : public ArgMinMaxBaseCPUKernel {
class ArgMinMaxInt8CPUKernel : public LiteKernel {
public:
ArgMinMaxInt8CPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
const mindspore::lite::PrimitiveC *primitive)
: ArgMinMaxBaseCPUKernel(parameter, inputs, outputs, ctx, primitive) {}
: LiteKernel(parameter, inputs, outputs, ctx, primitive) {}

~ArgMinMaxInt8CPUKernel() = default;



+ 0
- 291
mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/argminmax_fp32_test.cc View File

@@ -1,291 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "src/common/log_adapter.h"
#include "common/common_test.h"
#include "mindspore/lite/nnacl/fp32/arg_min_max_fp32.h"
#include "mindspore/lite/nnacl/arg_min_max.h"
#include "mindspore/lite/nnacl/arithmetic_common.h"

namespace mindspore {

class TestArgMinMaxTestFp32 : public mindspore::CommonTest {
public:
TestArgMinMaxTestFp32() = default;
};

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest1) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {2, 2, 0, 2, 0};
std::vector<int> shape = {3, 5};
float out[5];
ArgMinMaxParameter param;
param.topk_ = 1;
param.out_value_ = false;
param.axis_ = 0;
param.data_type_ = 43;
param.dims_size_ = 2;
param.get_max_ = true;
param.keep_dims_ = false;
ArgMinMax(in.data(), out, shape.data(), &param);
for (size_t i = 0; i < except_out.size(); ++i) {
std::cout << out[i] << " ";
}
std::cout << "\n";
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.000001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest1_keep_dim) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {2, 2, 0, 2, 0};
std::vector<int> shape = {3, 5};
float out[5];
ArgMinMaxParameter param;
param.topk_ = 1;
param.out_value_ = false;
param.axis_ = 0;
param.data_type_ = 43;
param.dims_size_ = 2;
param.get_max_ = true;
param.keep_dims_ = true;
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(shape[param.axis_] * sizeof(ArgElement)));
std::vector<int> out_shape = {1, 5};
ComputeStrides(shape.data(), param.in_strides_, shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
ArgMinMax(in.data(), out, shape.data(), &param);
for (size_t i = 0; i < except_out.size(); ++i) {
std::cout << out[i] << " ";
}
std::cout << "\n";
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.000001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest_axis2_keep_dim) {
std::vector<float> in = {10, 20, 30, 11, 15, 10, 5, 10, 12, 10, 20, 30, 11, 15,
10, 5, 10, 12, 10, 20, 30, 11, 15, 10, 5, 10, 12};
std::vector<float> except_out = {1, 0, 0, 1, 0, 0, 1, 0, 0};
std::vector<int> shape = {1, 3, 3, 3};
float out[9];
ArgMinMaxParameter param;
param.topk_ = 1;
param.out_value_ = false;
param.axis_ = 2;
param.data_type_ = 43;
param.dims_size_ = 4;
param.get_max_ = true;
param.keep_dims_ = true;
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(shape[param.axis_] * sizeof(ArgElement)));
std::vector<int> out_shape = {1, 3, 1, 3};
ComputeStrides(shape.data(), param.in_strides_, shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
ArgMinMax(in.data(), out, shape.data(), &param);
for (size_t i = 0; i < except_out.size(); ++i) {
std::cout << out[i] << " ";
}
std::cout << "\n";
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.000001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest2) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {30, 45, 30, 50, 90};
std::vector<int> shape = {3, 5};
float out[5];
ArgMinMaxParameter param;
param.topk_ = 1;
param.out_value_ = true;
param.axis_ = 0;
param.data_type_ = 43;
param.dims_size_ = 2;
param.get_max_ = true;
param.keep_dims_ = false;
ArgMinMax(in.data(), out, shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.000001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMinTest2) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {10, 11, 15, 1, 30};
std::vector<int> shape = {3, 5};
float out[5];
ArgMinMaxParameter param;
param.topk_ = 1;
param.out_value_ = true;
param.axis_ = 0;
param.data_type_ = 43;
param.dims_size_ = 2;
param.get_max_ = false;
param.keep_dims_ = false;
ArgMinMax(in.data(), out, shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.000001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest3_axis2_out_data) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {30, 45, 30, 50, 90, 20, 20, 25, 40, 50};
ArgMinMaxParameter param;
param.axis_ = 2;
std::vector<int> in_shape = {1, 1, 3, 5};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = true;
param.topk_ = 2;
std::vector<int> out_shape = {1, 1, 2, 5};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[10];
ArgMaxDim2(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest3_axis2_out_index) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {2, 2, 0, 2, 0, 1, 0, 2, 0, 1};
ArgMinMaxParameter param;
param.axis_ = 2;
std::vector<int> in_shape = {1, 1, 3, 5};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = false;
param.topk_ = 2;
std::vector<int> out_shape = {1, 1, 2, 5};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[10];
ArgMaxDim2(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest4_axis3_out_data) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {90, 40, 50, 20, 50, 45};
ArgMinMaxParameter param;
param.axis_ = 3;
std::vector<int> in_shape = {1, 1, 3, 5};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = true;
param.topk_ = 2;
std::vector<int> out_shape = {1, 1, 3, 2};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[6];
ArgMaxDim3(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest4_axis3_out_index) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {4, 3, 4, 0, 3, 1};
ArgMinMaxParameter param;
param.axis_ = 3;
std::vector<int> in_shape = {1, 1, 3, 5};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = false;
param.topk_ = 2;
std::vector<int> out_shape = {1, 1, 3, 2};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[6];
ArgMaxDim3(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest5_axis1_out_index) {
std::vector<float> in = {100, 2, 300, 4, 50, 6, 11, 12, 13, 34, 35, 36, 9, 6, 17, 10, 20, 30,
10, 20, 30, 40, 5, 60, 7, 80, 90, 10, 11, 120, 18, 5, 16, 9, 22, 23};
std::vector<float> except_out = {0, 1, 0, 1, 0, 1, 1, 2, 2, 2, 1, 2, 2, 1, 1, 0, 2, 1, 0, 0, 0, 1, 1, 0};
ArgMinMaxParameter param;
param.axis_ = 1;
std::vector<int> in_shape = {2, 3, 2, 3};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = false;
param.topk_ = 2;
std::vector<int> out_shape = {2, 2, 2, 3};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[24];
ArgMaxDim1(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest5_axis1_out_data) {
std::vector<float> in = {100, 2, 300, 4, 50, 6, 11, 12, 13, 34, 35, 36, 9, 6, 17, 10, 20, 30,
10, 20, 30, 40, 5, 60, 7, 80, 90, 10, 11, 120, 18, 5, 16, 9, 22, 23};
std::vector<float> except_out = {100, 12, 300, 34, 50, 36, 11, 6, 17, 10, 35, 30,
18, 80, 90, 40, 22, 120, 10, 20, 30, 10, 11, 60};
ArgMinMaxParameter param;
param.axis_ = 1;
std::vector<int> in_shape = {2, 3, 2, 3};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = true;
param.topk_ = 2;
std::vector<int> out_shape = {2, 2, 2, 3};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[24];
ArgMaxDim1(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest6_axis0_out_index) {
std::vector<float> in = {100, 2, 4, 50, 11, 12, 34, 35, 10, 20, 40, 5, 7, 80, 10, 11, 55, 25, 5, 15, 18, 8, 15, 16};
std::vector<float> except_out = {0, 2, 1, 0, 2, 1, 0, 0, 2, 1, 2, 2, 0, 0, 2, 2};
ArgMinMaxParameter param;
param.axis_ = 1;
std::vector<int> in_shape = {3, 2, 2, 2};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = false;
param.topk_ = 2;
std::vector<int> out_shape = {2, 2, 2, 2};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[16];
ArgMaxDim0(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMaxTest6_axis0_out_data) {
std::vector<float> in = {100, 2, 4, 50, 11, 12, 34, 35, 10, 20, 40, 5, 7, 80, 10, 11, 55, 25, 5, 15, 18, 8, 15, 16};
std::vector<float> except_out = {100, 25, 40, 50, 18, 80, 34, 35, 55, 20, 5, 15, 11, 12, 15, 16};
ArgMinMaxParameter param;
param.axis_ = 1;
std::vector<int> in_shape = {3, 2, 2, 2};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = true;
param.topk_ = 2;
std::vector<int> out_shape = {2, 2, 2, 2};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[16];
ArgMaxDim0(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

TEST_F(TestArgMinMaxTestFp32, ArgMinTest1_axis3_out_data) {
std::vector<float> in = {10, 20, 30, 40, 90, 20, 11, 15, 1, 50, 30, 45, 25, 50, 30};
std::vector<float> except_out = {10, 20, 1, 11, 25, 30};
ArgMinMaxParameter param;
param.axis_ = 3;
std::vector<int> in_shape = {1, 1, 3, 5};
param.arg_elements_ = reinterpret_cast<ArgElement *>(malloc(in_shape[param.axis_] * sizeof(ArgElement)));
param.out_value_ = true;
param.topk_ = 2;
std::vector<int> out_shape = {1, 1, 3, 2};
ComputeStrides(in_shape.data(), param.in_strides_, in_shape.size());
ComputeStrides(out_shape.data(), param.out_strides_, out_shape.size());
float out[6];
ArgMinDim3(in.data(), out, in_shape.data(), &param);
ASSERT_EQ(0, CompareOutputData(out, except_out.data(), except_out.size(), 0.00001));
}

} // namespace mindspore

+ 0
- 134
mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/instance_norm_fp32_tests.cc View File

@@ -1,134 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <iostream>
#include "src/common/log_adapter.h"
#include "common/common_test.h"
#include "mindspore/lite/nnacl/fp32/instance_norm_fp32.h"
#include "mindspore/lite/src/kernel_registry.h"
#include "mindspore/lite/src/lite_kernel.h"

namespace mindspore {
class TestInstanceNormFp32 : public mindspore::CommonTest {
public:
TestInstanceNormFp32() {}
};

TEST_F(TestInstanceNormFp32, INTest1) {
std::vector<float> in_data = {-11.18675, 11.433986, 11.386012, 11.245945, -2.7614849, 14.692399,
-1.1983503, -6.6790967, 6.383416, -13.3213005, -8.693595, 9.476344};
std::vector<float> in_data1 = {12.352293, 5.122387, 14.249514};
std::vector<float> in_data2 = {14.632595, 0.70900035, 11.179003};

InstanceNormParameter op_param;
op_param.op_parameter_.type_ = schema::PrimitiveType_InstanceNorm;
op_param.epsilon_ = 0.001f;

lite::Tensor input0_tensor(kNumberTypeFloat32, {1, 2, 2, 3});
lite::Tensor input1_tensor(kNumberTypeFloat32, {3});
lite::Tensor input2_tensor(kNumberTypeFloat32, {3});
input0_tensor.set_data(in_data.data());
input1_tensor.set_data(in_data1.data());
input2_tensor.set_data(in_data2.data());
std::vector<lite::Tensor *> inputs_tensor = {&input0_tensor, &input1_tensor, &input2_tensor};

std::vector<float> output(12);
std::vector<float> corr_out = {5.0145645, 9.248516, 15.439679, 33.51017, 0.0012711287, 31.0666883,
17.70254, -2.5507483, -8.204435, 2.3031063, -3.8630369, 6.4138837};

lite::Tensor output0_tensor(kNumberTypeFloat32, {1, 2, 2, 3});
output0_tensor.set_data(output.data());
std::vector<lite::Tensor *> outputs_tensor = {&output0_tensor};

kernel::KernelKey desc = {kernel::KERNEL_ARCH::kCPU, kNumberTypeFloat32, schema::PrimitiveType_InstanceNorm};
auto creator = lite::KernelRegistry::GetInstance()->GetCreator(desc);
ASSERT_NE(creator, nullptr);
lite::InnerContext ctx;
ctx.thread_num_ = 4;
ASSERT_EQ(lite::RET_OK, ctx.Init());
kernel::LiteKernel *kernel =
creator(inputs_tensor, outputs_tensor, reinterpret_cast<OpParameter *>(&op_param), &ctx, desc, nullptr);
ASSERT_NE(kernel, nullptr);
auto output_tensor_shape = output0_tensor.shape();
kernel->Run();

printf("==================output data=================\n");
for (int i = 0; i < output0_tensor.ElementsNum(); i++) {
std::cout << output[i] << " ,";
}
std::cout << std::endl;
ASSERT_EQ(0, CompareOutputData(output.data(), corr_out.data(), output0_tensor.ElementsNum(), 0.001));

input0_tensor.set_data(nullptr);
input1_tensor.set_data(nullptr);
input2_tensor.set_data(nullptr);
output0_tensor.set_data(nullptr);
}

TEST_F(TestInstanceNormFp32, INTest2) {
std::vector<float> in_data = {-11.18675, 11.433986, 11.386012, 11.245945, -2.7614849, 14.692399,
-1.1983503, -6.6790967, 6.383416, -13.3213005, -8.693595, 9.476344,
-12.18675, 12.433986, 12.386012, 12.245945, -3.7614849, 15.692399,
-2.1983503, -7.6790967, 7.383416, -14.3213005, -9.693595, 10.476344};
std::vector<float> in_data1 = {12.352293, 5.122387, 14.249514, 12.352293, 5.122387, 14.249514};
std::vector<float> in_data2 = {14.632595, 0.70900035, 11.179003, 14.632595, 0.70900035, 11.179003};

InstanceNormParameter op_param;
op_param.op_parameter_.type_ = schema::PrimitiveType_InstanceNorm;
op_param.epsilon_ = 0.001f;

lite::Tensor input0_tensor(kNumberTypeFloat32, {2, 2, 2, 3});
lite::Tensor input1_tensor(kNumberTypeFloat32, {2, 3});
lite::Tensor input2_tensor(kNumberTypeFloat32, {2, 3});
input0_tensor.set_data(in_data.data());
input1_tensor.set_data(in_data1.data());
input2_tensor.set_data(in_data2.data());
std::vector<lite::Tensor *> inputs_tensor = {&input0_tensor, &input1_tensor, &input2_tensor};

std::vector<float> output(24);
std::vector<float> corr_out = {5.0145645, 9.248516, 15.439679, 33.51017, 0.0012711287, 31.0666883,
17.70254, -2.5507483, -8.204435, 2.3031063, -3.8630369, 6.4138837,
5.133601, 9.310399, 15.439679, 33.886883, -0.22505027, 31.066883,
16.888313, -2.5316327, -8.204435, 2.6215858, -3.717714, 6.4138837};

lite::Tensor output0_tensor(kNumberTypeFloat32, {2, 2, 2, 3});
output0_tensor.set_data(output.data());
std::vector<lite::Tensor *> outputs_tensor = {&output0_tensor};

kernel::KernelKey desc = {kernel::KERNEL_ARCH::kCPU, kNumberTypeFloat32, schema::PrimitiveType_InstanceNorm};
auto creator = lite::KernelRegistry::GetInstance()->GetCreator(desc);
ASSERT_NE(creator, nullptr);
lite::InnerContext ctx;
ctx.thread_num_ = 4;
ASSERT_EQ(lite::RET_OK, ctx.Init());
kernel::LiteKernel *kernel =
creator(inputs_tensor, outputs_tensor, reinterpret_cast<OpParameter *>(&op_param), &ctx, desc, nullptr);
ASSERT_NE(kernel, nullptr);
auto output_tensor_shape = output0_tensor.shape();
kernel->Run();

printf("==================output data=================\n");
for (int i = 0; i < output0_tensor.ElementsNum(); i++) {
std::cout << output[i] << " ,";
}
std::cout << std::endl;
ASSERT_EQ(0, CompareOutputData(output.data(), corr_out.data(), output0_tensor.ElementsNum(), 0.001));

input0_tensor.set_data(nullptr);
input1_tensor.set_data(nullptr);
input2_tensor.set_data(nullptr);
output0_tensor.set_data(nullptr);
}
} // namespace mindspore

Write Preview
Loading…
Cancel
Save