p83571920
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

fix bug of mirror pad

This commit is contained in:
fuzhiye 2020-12-31 09:18:42 +08:00
parent 882a582eb4
commit cf82c3e15e
2 changed files with 54 additions and 60 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
mindspore/lite/nnacl/pad_parameter.h
View File

@ -39,8 +39,8 @@ typedef struct PadParameter {
typedef struct MirrorPadBlock {
int out_offset_;
int out_stride_[3];
int size_[3];
int out_stride_[DEFAULT_PAD_NDIMS];
int size_[DEFAULT_PAD_NDIMS];
} MirrorPadBlock;
#endif // MINDSPORE_LITE_NNACL_PAD_PARAMETER_H_

110
mindspore/lite/src/runtime/kernel/arm/fp32/pad_fp32.cc
View File

@ -71,66 +71,57 @@ int PadCPUKernel::ReSize() {
void PadCPUKernel::InitMirrorPadBlock() {
mirror_pad_block_.clear();
auto input = in_tensors_.at(0);
std::vector<int> left_pads(input->shape().size());
for (size_t i = 0; i < input->shape().size(); ++i) {
std::vector<int> left_pads(DEFAULT_PAD_NDIMS);
for (size_t i = 0; i < DEFAULT_PAD_NDIMS; ++i) {
left_pads[i] = pad_param_->paddings_[2 * i];
}
std::vector<int> input_seperate_dims;
std::vector<int> output_seperate_dims;
std::vector<int> seperate_offset;
std::vector<int> input_separate_dims;
std::vector<int> output_separate_dims;
std::vector<int> separate_offset;
/* init seperate dims */
/* init separate dims */
int cur_input = 1;
int cur_output = 1;
for (size_t i = 0; i < input->shape().size(); ++i) {
if (in_[i] != out_[i]) {
if (1 < cur_input) {
input_seperate_dims.emplace_back(cur_input);
output_seperate_dims.emplace_back(cur_output);
seperate_offset.emplace_back(0);
}
input_seperate_dims.emplace_back(in_[i]);
output_seperate_dims.emplace_back(out_[i]);
seperate_offset.emplace_back(left_pads[i]);
cur_input = 1;
cur_output = 1;
} else {
cur_input *= in_[i];
cur_output *= out_[i];
for (size_t i = 0; i < DEFAULT_PAD_NDIMS; ++i) {
if (1 < cur_input) {
input_separate_dims.emplace_back(cur_input);
output_separate_dims.emplace_back(cur_output);
separate_offset.emplace_back(0);
}
input_separate_dims.emplace_back(in_[i]);
output_separate_dims.emplace_back(out_[i]);
separate_offset.emplace_back(left_pads[i]);
cur_input = 1;
cur_output = 1;
}
if (cur_input != 1 || cur_output != 1) {
input_seperate_dims.emplace_back(cur_input);
output_seperate_dims.emplace_back(cur_output);
seperate_offset.emplace_back(0);
input_separate_dims.emplace_back(cur_input);
output_separate_dims.emplace_back(cur_output);
separate_offset.emplace_back(0);
}
/* init seperate stride */
std::vector<int> output_seperate_stride;
output_seperate_stride.resize(output_seperate_dims.size());
GetStride(output_seperate_stride.data(), output_seperate_dims.data(), output_seperate_dims.size());
/* init separate stride */
std::vector<int> output_separate_stride;
output_separate_stride.resize(output_separate_dims.size());
GetStride(output_separate_stride.data(), output_separate_dims.data(), output_separate_dims.size());
/* init seperate stride */
/* init separate stride */
std::vector<int> remain_stride;
int remain_stride_size = seperate_offset.size() > 3 ? static_cast<int>(seperate_offset.size()) - 3 : 0;
remain_stride.resize(remain_stride_size);
int remain_size = GetStride(remain_stride.data(), output_seperate_dims.data(), remain_stride.size());
remain_stride.resize(0);
int remain_size = GetStride(remain_stride.data(), output_separate_dims.data(), remain_stride.size());
std::vector<int> right_pads(seperate_offset.size());
std::vector<int> right_pads(separate_offset.size());
for (size_t i = 0; i < right_pads.size(); ++i) {
right_pads[i] = output_seperate_dims[i] - input_seperate_dims[i] - seperate_offset[i];
right_pads[i] = output_separate_dims[i] - input_separate_dims[i] - separate_offset[i];
}
/* init pad region */
std::vector<int> pad_region;
for (size_t i = remain_stride.size(); i < output_seperate_stride.size(); ++i) {
for (size_t i = remain_stride.size(); i < output_separate_stride.size(); ++i) {
// 0: center, 1: left, 2: right
int r = 1;
if (seperate_offset[i] > 0) {
if (separate_offset[i] > 0) {
r++;
}
if (right_pads[i] > 0) {
@ -158,29 +149,29 @@ void PadCPUKernel::InitMirrorPadBlock() {
}
MirrorPadBlock block;
int size_offset = 3 - static_cast<int>(pad_region.size());
int size_offset = DEFAULT_PAD_NDIMS - static_cast<int>(pad_region.size());
for (size_t i = 0; i < pad_region.size(); ++i) {
int di = size_offset + i;
int si = remain_dim_offset + i;
switch (pad_cord[i]) {
case 0:
dst_offset += seperate_offset[si] * output_seperate_stride[si];
block.size_[di] = input_seperate_dims[si];
block.out_stride_[di] = output_seperate_stride[si];
dst_offset += separate_offset[si] * output_separate_stride[si];
block.size_[di] = input_separate_dims[si];
block.out_stride_[di] = output_separate_stride[si];
break;
case 2:
dst_offset += (seperate_offset[si] + input_seperate_dims[si]) * output_seperate_stride[si];
dst_offset += (separate_offset[si] + input_separate_dims[si]) * output_separate_stride[si];
block.size_[di] = right_pads[si];
block.out_stride_[di] = output_seperate_stride[si];
block.out_stride_[di] = output_separate_stride[si];
break;
case 1:
if (seperate_offset[si] > 0) {
block.size_[di] = seperate_offset[si];
block.out_stride_[di] = output_seperate_stride[si];
if (separate_offset[si] > 0) {
block.size_[di] = separate_offset[si];
block.out_stride_[di] = output_separate_stride[si];
} else {
dst_offset += (seperate_offset[si] + input_seperate_dims[si]) * output_seperate_stride[si];
dst_offset += (separate_offset[si] + input_separate_dims[si]) * output_separate_stride[si];
block.size_[di] = right_pads[si];
block.out_stride_[di] = output_seperate_stride[si];
block.out_stride_[di] = output_separate_stride[si];
}
break;
default:
@ -235,8 +226,8 @@ int PadCPUKernel::RunImpl(int task_id) {
auto output = out_tensors_.at(0);
MS_ASSERT(input);
MS_ASSERT(output);
auto input_data = reinterpret_cast<float *>(input->MutableData());
auto output_data = reinterpret_cast<float *>(output->MutableData());
auto input_data = reinterpret_cast<float *>(input->data_c());
auto output_data = reinterpret_cast<float *>(output->data_c());
MS_ASSERT(input_data);
MS_ASSERT(output_data);
Pad(input_data, output_data, in_, out_, pad_param_->paddings_, task_id, context_->thread_num_);
@ -257,8 +248,8 @@ int MirrorPadImpl(void *cdata, int task_id) {
int PadCPUKernel::RunMirrorPadImpl(int task_id) {
auto input = in_tensors_.at(0);
auto output = out_tensors_.at(0);
auto input_data = reinterpret_cast<float *>(input->MutableData());
auto output_data = reinterpret_cast<float *>(output->MutableData());
auto input_data = reinterpret_cast<float *>(input->data_c());
auto output_data = reinterpret_cast<float *>(output->data_c());
/* Fast Mirror pad */
if (mirror_pad_block_.size() != 0) {
@ -272,8 +263,11 @@ int PadCPUKernel::RunMirrorPadImpl(int task_id) {
for (int a = 0; a < block.size_[0]; a++) {
int out_a_index = block.out_offset_ + a * block.out_stride_[0];
for (int b = 0; b < block.size_[1]; b++) {
int output_index = out_a_index + b * block.out_stride_[1];
MirrorPad(input_data, output_data, in_, pad_param_, output_index, output_index + block.size_[2]);
int out_b_index = out_a_index + b * block.out_stride_[1];
for (int c = 0; c < block.size_[2]; ++c) {
int output_index = out_b_index + c * block.out_stride_[2];
MirrorPad(input_data, output_data, in_, pad_param_, output_index, output_index + block.size_[3]);
}
}
}
}
@ -321,7 +315,7 @@ int PadCPUKernel::CopyPaddingFromInput() {
return RET_ERROR;
}
auto padding_tensor = in_tensors_.at(1);
auto paddings = reinterpret_cast<int *>(padding_tensor->MutableData());
auto paddings = reinterpret_cast<int *>(padding_tensor->data_c());
if (paddings == nullptr) {
MS_LOG(ERROR) << "Pad second input data nullptr";
return RET_ERROR;
@ -389,7 +383,7 @@ int PadCPUKernel::Run() {
if (pad_param_->pad_mode_ == static_cast<int>(schema::PaddingMode_CONSTANT)) {
auto output = out_tensors_.at(0);
int output_size = output->ElementsNum();
auto output_data = reinterpret_cast<float *>(output->MutableData());
auto output_data = reinterpret_cast<float *>(output->data_c());
if (abs(pad_param_->constant_value_ - 0.0f) < 1e-5) {
memset(output_data, 0, output_size * sizeof(float));
} else {