You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
graphengine/ge/common/op/ge_op_utils.cc

479 lines
18 KiB

/**
* 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 "framework/common/op/ge_op_utils.h"
#include <list>
#include "common/fp16_t.h"
#include "common/ge/ge_util.h"
#include "external/graph/types.h"
#include "framework/common/debug/ge_log.h"
#include "framework/common/debug/log.h"
#include "framework/common/fmk_error_codes.h"
#include "framework/common/ge_inner_error_codes.h"
#include "framework/common/op/attr_value_util.h"
#include "framework/common/util.h"
#include "framework/common/types.h"
#include "graph/anchor.h"
#include "graph/debug/ge_attr_define.h"
#include "graph/utils/op_desc_utils.h"
#include "graph/utils/tensor_utils.h"
#include "graph/utils/type_utils.h"
#include "mmpa/mmpa_api.h"
using std::vector;
namespace ge {
// General constant
const int32_t kDimSizeZero = 0;
const int32_t kDimSizeOne = 1;
const int32_t kDimSizeTwo = 2;
const int32_t kDimSizeThree = 3;
const uint32_t kSliceDataNum = 2;
// Add Sub Mul
const uint32_t ADD_INPUT_NUM = 2;
const uint32_t MUL_INPUT_NUM = 2;
// Permute
const int32_t PERMUTE_ORDER_NUM = 4;
// Ssd PriroBox
const double SSD_PRIORBOX_ASPECT_RATIO_VALUE = 1.0;
// Switch
const uint32_t SWITCH_INPUT_NUM = 2;
const uint32_t SWITCH_OUTPUT_NUM = 2;
const uint32_t SWITCH_FALSE_OUTPUT = 0;
const uint32_t SWITCH_TRUE_OUTPUT = 1;
const uint32_t SWITCH_DATA_INPUT = 0;
const uint32_t SWITCH_PRED_INPUT = 1;
// FunctionOp
const uint32_t IF_COND_INPUT = 0;
const uint32_t FOR_START_INPUT = 0;
const uint32_t FOR_LIMIT_INPUT = 1;
const uint32_t FOR_DELTA_INPUT = 2;
const uint32_t FOR_DATA_INPUT = 3;
const int NORMAL_TENSOR_SIZE = 4;
// Get the value of key from attr
#define AIPP_GET_ATTR_VALUE(KEY, ATTR_TYPE) \
if (aipp_attr.GetItem(#KEY).GetValue<ATTR_TYPE>(KEY) != SUCCESS) { \
GELOGI("Attr %s will take default value.", #KEY); \
break; \
}
// Converting aippparams and attrdefmap
#define AIPP_CONVERT_FORMAT_EX(KEY, ORG_TYPE, SAVE_TYPE, ATTR_TYPE) \
do { \
SAVE_TYPE KEY = static_cast<SAVE_TYPE>(0); \
AIPP_GET_ATTR_VALUE(KEY, ATTR_TYPE) \
aipp_params->set_##KEY(ORG_TYPE(KEY)); \
} while (0)
// Converting aippparams and attrdefmap
#define AIPP_CONVERT_FORMAT(KEY, KEY_TYPE, ATTR_TYPE) AIPP_CONVERT_FORMAT_EX(KEY, KEY_TYPE, KEY_TYPE, ATTR_TYPE)
#define AIPP_CONVERT_INT(KEY) AIPP_CONVERT_FORMAT(KEY, int64_t, GeAttrValue::INT)
#define AIPP_CONVERT_BOOL(KEY) AIPP_CONVERT_FORMAT(KEY, bool, GeAttrValue::BOOL)
#define AIPP_CONVERT_FLOAT(KEY) AIPP_CONVERT_FORMAT(KEY, float, GeAttrValue::FLOAT)
// Transform aippparams (with repeated decoration) and attrdefmap
#define AIPP_CONVERT_LIST_FORMAT(KEY, KEY_TYPE, REQUIRED, ATTR_TYPE) \
do { \
if (REQUIRED) { \
KEY_TYPE KEY; \
AIPP_GET_ATTR_VALUE(KEY, ATTR_TYPE) \
aipp_params->add_##KEY(KEY); \
} \
} while (0)
#define AIPP_CONVERT_LIST_INT(KEY, REQUIRED) AIPP_CONVERT_LIST_FORMAT(KEY, int64_t, REQUIRED, GeAttrValue::INT)
#define AIPP_CONVERT_LIST_BOOL(KEY, REQUIRED) AIPP_CONVERT_LIST_FORMAT(KEY, bool, REQUIRED, GeAttrValue::BOOL)
#define AIPP_CONVERT_LIST_FLOAT(KEY, REQUIRED) AIPP_CONVERT_LIST_FORMAT(KEY, float, REQUIRED, GeAttrValue::FLOAT)
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status
OpUtils::ConvertAippParams(const GeAttrValue::NAMED_ATTRS &aipp_attr, domi::AippOpParams *aipp_params) {
GE_CHECK_NOTNULL(aipp_params);
AIPP_CONVERT_FORMAT_EX(aipp_mode, domi::AippOpParams::AippMode, int32_t, GeAttrValue::INT);
AIPP_CONVERT_INT(related_input_rank);
if (aipp_params->aipp_mode() == domi::AippOpParams::dynamic) {
AIPP_CONVERT_INT(max_src_image_size);
AIPP_CONVERT_BOOL(support_rotation);
} else {
AIPP_CONVERT_FORMAT_EX(input_format, domi::AippOpParams::InputFormat, int32_t, GeAttrValue::INT);
AIPP_CONVERT_BOOL(csc_switch);
AIPP_CONVERT_BOOL(crop);
AIPP_CONVERT_INT(load_start_pos_w);
AIPP_CONVERT_INT(load_start_pos_h);
AIPP_CONVERT_INT(crop_size_w);
AIPP_CONVERT_INT(crop_size_h);
AIPP_CONVERT_BOOL(resize);
AIPP_CONVERT_INT(resize_output_w);
AIPP_CONVERT_INT(resize_output_h);
AIPP_CONVERT_BOOL(padding);
AIPP_CONVERT_INT(left_padding_size);
AIPP_CONVERT_INT(right_padding_size);
AIPP_CONVERT_INT(top_padding_size);
AIPP_CONVERT_INT(bottom_padding_size);
AIPP_CONVERT_INT(src_image_size_w);
AIPP_CONVERT_INT(src_image_size_h);
AIPP_CONVERT_FLOAT(cpadding_value);
AIPP_CONVERT_BOOL(rbuv_swap_switch);
AIPP_CONVERT_BOOL(ax_swap_switch);
AIPP_CONVERT_BOOL(single_line_mode);
AIPP_CONVERT_INT(mean_chn_0);
AIPP_CONVERT_INT(mean_chn_1);
AIPP_CONVERT_INT(mean_chn_2);
AIPP_CONVERT_FLOAT(min_chn_0);
AIPP_CONVERT_FLOAT(min_chn_1);
AIPP_CONVERT_FLOAT(min_chn_2);
AIPP_CONVERT_LIST_FLOAT(var_reci_chn_0, true);
AIPP_CONVERT_LIST_FLOAT(var_reci_chn_1, true);
AIPP_CONVERT_LIST_FLOAT(var_reci_chn_2, true);
AIPP_CONVERT_LIST_FLOAT(var_reci_chn_3, true);
const bool csc_switch = aipp_params->csc_switch();
AIPP_CONVERT_LIST_INT(matrix_r0c0, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r0c1, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r0c2, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r1c0, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r1c1, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r1c2, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r2c0, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r2c1, csc_switch);
AIPP_CONVERT_LIST_INT(matrix_r2c2, csc_switch);
AIPP_CONVERT_LIST_INT(output_bias_0, csc_switch);
AIPP_CONVERT_LIST_INT(output_bias_1, csc_switch);
AIPP_CONVERT_LIST_INT(output_bias_2, csc_switch);
AIPP_CONVERT_LIST_INT(input_bias_0, csc_switch);
AIPP_CONVERT_LIST_INT(input_bias_1, csc_switch);
AIPP_CONVERT_LIST_INT(input_bias_2, csc_switch);
}
return SUCCESS;
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status OpUtils::TransferDim(const std::vector<int64_t> &dim,
std::vector<int64_t> &dim_vector) {
size_t input_shape_size = dim.size();
std::list<uint32_t> new_dim_list;
for (auto dim_temp : dim) {
new_dim_list.push_back(dim_temp);
}
if (input_shape_size > DIM_DEFAULT_SIZE) {
dim_vector = dim;
GELOGI("Dim_vector size is %zu, do not to transfer dim", input_shape_size);
return SUCCESS;
}
switch (input_shape_size) {
case kDimSizeZero: {
new_dim_list.push_back(1);
new_dim_list.push_back(1);
new_dim_list.push_back(1);
new_dim_list.push_back(1);
break;
}
case kDimSizeOne: {
new_dim_list.push_front(1);
new_dim_list.push_back(1);
new_dim_list.push_back(1);
break;
}
case kDimSizeTwo: {
new_dim_list.push_front(1);
new_dim_list.push_back(1);
break;
}
case kDimSizeThree: {
new_dim_list.push_front(1);
break;
}
default:
GELOGI("Invalid input_shape_size.");
break;
}
dim_vector.clear();
for (auto dims : new_dim_list) {
dim_vector.push_back(dims);
}
return SUCCESS;
}
template <typename T>
void OpUtils::SliceData(const std::vector<char *> &input, int64_t chunk_size, std::vector<char *> &output,
int64_t begin, int64_t out_dim, int64_t stride) {
char *slice = nullptr;
// chunk_size * (begin + (out_dim-1)*stride) always less than chunk_size * dim_i, no need to check.
for (size_t j = 0; j < input.size(); j++) {
slice = input[j] + sizeof(T) * begin * chunk_size;
for (int64_t i = 0; i < out_dim; i++) {
output.push_back(slice + sizeof(T) * i * chunk_size * stride);
}
}
}
template <typename T>
Status OpUtils::SetDataByDataType(size_t out_size, const std::vector<char *> &chunk_input,
const std::vector<char *> &chunk_output, GeTensor *output) {
unique_ptr<T[]> output_data(new (std::nothrow) T[out_size]());
if (output_data == nullptr) {
GELOGE(MEMALLOC_FAILED, "New buf failed");
return INTERNAL_ERROR;
}
if (!chunk_input.empty()) {
for (size_t j = 0; j < out_size; j++) {
T *value = reinterpret_cast<T *>(chunk_input[j]);
output_data[j] = value[0];
}
} else {
for (size_t j = 0; j < out_size; j++) {
T *value = reinterpret_cast<T *>(chunk_output[j]);
output_data[j] = value[0];
}
}
// output_data != nullptr and out_size > 0, SetData always return success, no need to check value
(void)output->SetData(reinterpret_cast<uint8_t *>(output_data.get()), out_size * sizeof(T));
return SUCCESS;
}
template <typename T>
Status OpUtils::SetOutputSliceDataByDataType(void *data, int64_t data_size, const std::vector<int64_t> &input_dims,
const std::vector<int64_t> &begin, const std::vector<int64_t> &output_dims,
GeTensor *output, const std::vector<int64_t> &stride) {
std::vector<char *> chunk_input;
std::vector<char *> chunk_output;
chunk_input.push_back(reinterpret_cast<char *>(data));
int64_t chunk_size = data_size;
size_t dim_size = input_dims.size();
for (size_t i = 0; i < dim_size; i++) {
int64_t begin_i = begin[i];
int64_t size_i = output_dims[i];
int64_t dim_i = input_dims[i];
int64_t stride_i = stride[i];
if (dim_i == 0) {
GELOGE(PARAM_INVALID, "Dim_i of size tensor can't be 0.");
return PARAM_INVALID;
}
chunk_size = chunk_size / dim_i;
if (i % kSliceDataNum == 0) {
SliceData<T>(chunk_input, chunk_size, chunk_output, begin_i, size_i, stride_i);
chunk_input.clear();
} else {
SliceData<T>(chunk_output, chunk_size, chunk_input, begin_i, size_i, stride_i);
chunk_output.clear();
}
}
size_t out_size = chunk_input.size() + chunk_output.size();
GE_CHK_BOOL_RET_STATUS(out_size > 0, FAILED, "Out_size <= 0");
Status ret = SetDataByDataType<T>(out_size, chunk_input, chunk_output, output);
return ret;
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status OpUtils::SetOutputSliceData(
void *data, int64_t data_size, int32_t data_type, std::vector<int64_t> &input_dims, std::vector<int64_t> &begin,
std::vector<int64_t> &output_dims, GeTensor *output, std::vector<int64_t> &stride) {
if (data == nullptr || output == nullptr) {
GELOGE(PARAM_INVALID, "Input param is nullptr.");
return PARAM_INVALID;
}
Status ret;
switch (data_type) {
case DT_INT32:
ret = SetOutputSliceDataByDataType<int32_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_FLOAT:
ret = SetOutputSliceDataByDataType<float>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_DOUBLE:
ret = SetOutputSliceDataByDataType<double>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_FLOAT16:
ret = SetOutputSliceDataByDataType<fp16_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_UINT8:
ret = SetOutputSliceDataByDataType<uint8_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_INT8:
ret = SetOutputSliceDataByDataType<int8_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_UINT16:
ret = SetOutputSliceDataByDataType<uint16_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_INT16:
ret = SetOutputSliceDataByDataType<int16_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_UINT32:
ret = SetOutputSliceDataByDataType<uint32_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_UINT64:
ret = SetOutputSliceDataByDataType<uint64_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
case DT_INT64:
ret = SetOutputSliceDataByDataType<int64_t>(data, data_size, input_dims, begin, output_dims, output, stride);
break;
default:
GELOGW("Unsupported data type: %s", TypeUtils::DataTypeToSerialString(static_cast<DataType>(data_type)).c_str());
return PARAM_INVALID;
}
return ret;
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY void OpUtils::TransDataHWCK2KCHW(const void *input, int64_t h,
int64_t w, int64_t c, int64_t k,
void **output) {
if (input == nullptr) {
return;
}
if (output == nullptr) {
return;
}
const char *w_data = (const char *)input;
int64_t count = h * w * c * k;
GE_IF_BOOL_EXEC(count <= 0, GELOGW("Count value must be greater than 0, but count = %ld", count); return);
float *buf = new (std::nothrow) float[count]();
GE_RT_VOID_CHECK_NOTNULL(buf);
float *src_buff = nullptr;
float *dst_buff = nullptr;
for (int h_i = 0; h_i < h; ++h_i) {
for (int w_i = 0; w_i < w; ++w_i) {
for (int c_i = 0; c_i < c; ++c_i) {
for (int k_i = 0; k_i < k; ++k_i) {
src_buff = reinterpret_cast<float *>(const_cast<char *>(w_data)) +
((h_i * w * c * k) + (w_i * c * k) + (c_i * k) + (k_i));
dst_buff = buf + ((k_i * c * h * w) + (c_i * h * w) + (h_i * w) + (w_i));
*dst_buff = *src_buff;
}
}
}
}
*output = buf;
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY void OpUtils::TransDataKCHW2HWCK(const void *input, int64_t k,
int64_t c, int64_t h, int64_t w,
void *output) {
if ((input == nullptr) || (output == nullptr)) {
GELOGD("%s[%d]: input param is nullptr.", __FILE__, __LINE__);
return;
}
const char *w_data = (const char *)input;
float *buf = reinterpret_cast<float *>(output);
float *src_buff = nullptr;
float *dst_buff = nullptr;
for (int k_i = 0; k_i < k; ++k_i) {
for (int c_i = 0; c_i < c; ++c_i) {
for (int h_i = 0; h_i < h; ++h_i) {
for (int w_i = 0; w_i < w; ++w_i) {
src_buff = reinterpret_cast<float *>(const_cast<char *>(w_data)) +
((k_i * c * h * w) + (c_i * h * w) + (h_i * w) + (w_i));
dst_buff = buf + ((h_i * w * c * k) + (w_i * c * k) + (c_i * k) + (k_i));
*dst_buff = *src_buff;
}
}
}
}
}
vector<ConstGeTensorPtr> OpUtils::GetWeights(const ge::Node &node) { return OpDescUtils::GetWeights(node); }
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY vector<ConstGeTensorPtr> OpUtils::GetWeights(ge::ConstNodePtr node) {
return OpDescUtils::GetWeights(node);
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY vector<GeTensorPtr> OpUtils::MutableWeights(const ge::Node &node) {
return OpDescUtils::MutableWeights(node);
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY vector<GeTensorPtr> OpUtils::MutableWeights(const ge::NodePtr node) {
return OpDescUtils::MutableWeights(node);
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status OpUtils::SetWeights(ge::Node &node,
const vector<ge::GeTensorPtr> &weights) {
return OpDescUtils::SetWeights(node, weights);
}
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status OpUtils::SetWeights(ge::NodePtr node,
const vector<ge::GeTensorPtr> &weights) {
return OpDescUtils::SetWeights(node, weights);
}
// The caller guarantees that the input sensor is constant
FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status
OpUtils::GetShapeDataFromConstTensor(const ConstGeTensorPtr &tensor, DataType type, std::vector<int64_t> &dims) {
if (tensor == nullptr) {
GELOGE(PARAM_INVALID, "Input tensor is nullptr");
return PARAM_INVALID;
}
// If the tensor data is a vector, the shape dimension must be 1
if (tensor->GetTensorDesc().GetShape().GetDims().size() > 1) {
GELOGE(PARAM_INVALID, "The dimension of the input tensor shape cannot be more than 1, it is %zu",
tensor->GetTensorDesc().GetShape().GetDims().size());
return PARAM_INVALID;
}
if (type == DT_INT32) {
int32_t *shape_data = const_cast<int32_t *>(reinterpret_cast<const int32_t *>(tensor->GetData().GetData()));
GE_CHECK_NOTNULL(shape_data);
size_t dims_num = tensor->GetData().size() / sizeof(int32_t);
for (size_t i = 0; i < dims_num; i++) {
dims.push_back(static_cast<int64_t>(shape_data[i]));
}
} else if (type == DT_INT64) {
int64_t *shape_data = const_cast<int64_t *>(reinterpret_cast<const int64_t *>(tensor->GetData().GetData()));
GE_CHECK_NOTNULL(shape_data);
size_t dims_num = tensor->GetData().size() / sizeof(int64_t);
for (size_t i = 0; i < dims_num; i++) {
dims.push_back(shape_data[i]);
}
} else {
GELOGE(PARAM_INVALID, "Data type only can be DT_INT32 or DT_INT64. type is %s",
TypeUtils::DataTypeToSerialString(type).c_str());
return PARAM_INVALID;
}
return SUCCESS;
}
uint32_t OpUtils::GetRealDimCnt(const GeTensorDesc &tensor_desc) {
uint32_t real_dim_cnt = 0;
domi::Status ret = TensorUtils::GetRealDimCnt(tensor_desc, real_dim_cnt);
return (ret == domi::SUCCESS) ? real_dim_cnt : 0;
}
} // namespace ge