mindspore-Ecosystem
/
mindspore
21
0
Fork 27
Code Issues Pull Requests 6 Projects Releases Wiki Activity

!20688 [LITE] fix bug 

Merge pull request !20688 from yefeng/134-fix_bug
This commit is contained in:
i-robot 2021-07-23 01:31:24 +00:00 committed by Gitee
parent 5317694497 5e205f6ece
commit d30a0d295c
10 changed files with 30 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
mindspore/lite/tools/anf_exporter/fetch_content.cc
View File

@ -27,6 +27,7 @@
namespace mindspore {
namespace lite {
namespace {
constexpr int kNumWeightIndex = 2;
constexpr size_t kTensorListMinSize = 3 * sizeof(int32_t);
static const std::unordered_map<int, int> TypeToTypeMap = {
{kNumberTypeInt, kNumberTypeInt32}, {kNumberTypeUInt, kNumberTypeUInt32}, {kNumberTypeFloat, kNumberTypeFloat32}};
@ -298,7 +299,7 @@ int FetchDataFromParameterNode(const CNodePtr &cnode, size_t index, converter::F
if ((opt::CheckPrimitiveType(cnode, prim::kPrimConv2DFusion) ||
opt::CheckPrimitiveType(cnode, opt::kPrimConv2DBackpropInputFusion) ||
opt::CheckPrimitiveType(cnode, prim::kPrimConv2dTransposeFusion)) &&
(index == 2 && prim->GetAttr(ops::kFormat) != nullptr)) {
(index == kNumWeightIndex && prim->GetAttr(ops::kFormat) != nullptr)) {
data_info->format_ = mindspore::KHWC;
}
@ -326,7 +327,7 @@ int FetchDataFromValueNode(const CNodePtr &cnode, size_t index, converter::FmkTy
MS_ASSERT(prim != nullptr);
if (value->isa<tensor::Tensor>()) {
ret = FetchFromTensorValue(value_node, prim, fmk_type, train_flag, data_info);
if (index == 2 && prim->GetAttr(ops::kFormat) != nullptr) {
if (index == kNumWeightIndex && prim->GetAttr(ops::kFormat) != nullptr) {
data_info->format_ = GetValue<int64_t>(prim->GetAttr(ops::kFormat));
}
} else if (value->isa<mindspore::Int32Imm>() || value->isa<mindspore::Int64Imm>()) {
@ -366,6 +367,9 @@ int SetFormatForCnode(const CNodePtr &cnode, size_t index, converter::FmkType fm
if (opt::GetTransposePerm(cnode->input(index)->cast<CNodePtr>(), &perm) != RET_OK) {
return RET_ERROR;
}
if (perm.size() < 4) {
return RET_OK;
}
if (perm[0] == 0 && perm[1] == 3 && perm[2] == 1 && perm[3] == 2 &&
(data_info->format_ == NHWC || data_info->format_ == KHWC)) {
data_info->format_ = NCHW;

5
mindspore/lite/tools/converter/legacy_optimizer/graph/infershape_pass.cc
View File

@ -39,6 +39,7 @@ constexpr size_t kInitialSize = 1024;
constexpr int kMainGraphIndex = 0;
constexpr int kCallInputMinSize = 1;
constexpr int kSwitchInputMinSize = 3;
constexpr int kNumDim2 = 2;
void FreeTensors(std::vector<Tensor *> *input_tensors, std::vector<Tensor *> *output_tensors) {
if (input_tensors == nullptr) {
@ -82,9 +83,9 @@ void ConvertTensorList(MetaGraphT *graph, uint32_t index, bool *convert_succ, st
return;
}
for (int j = 0; j < data[1]; ++j) {
element_shape.push_back(data[j + 2]);
element_shape.push_back(data[j + kNumDim2]);
}
tensor_shape = {data[data[1] + 2]};
tensor_shape = {data[data[1] + kNumDim2]};
}
lite_tensor = std::make_unique<TensorList>(tensor_shape, element_shape);
if (lite_tensor == nullptr) {

3
mindspore/lite/tools/converter/parser/caffe/caffe_model_parser.cc
View File

@ -138,8 +138,7 @@ STATUS CaffeModelParser::WeightFormatTransform(const FuncGraphPtr &graph) {
MS_LOG(ERROR) << "weight node must param value";
return RET_OK;
}
lite::STATUS status;
status = HardCodeCaffe(conv_cnode, tensor_info, graph);
auto status = HardCodeCaffe(conv_cnode, tensor_info, graph);
if (status != lite::RET_OK) {
MS_LOG(ERROR) << "Format hard code failed: " << status << ", node: " << node->fullname_with_scope();
return RET_ERROR;

3
mindspore/lite/tools/converter/parser/conv1d_inout_adjust.cc
View File

@ -29,6 +29,7 @@ namespace mindspore::lite {
namespace {
constexpr size_t kTripleNum = 3;
constexpr size_t kConvWeightIndex = 2;
constexpr int64_t kNumDim2 = 2;
} // namespace
CNodePtr Conv1DInOutAdjust::NewUnsqueezeOpNode(const FuncGraphPtr &func_graph, const AnfNodePtr input_node,
const std::vector<int64_t> &axis) {
@ -75,7 +76,7 @@ lite::STATUS Conv1DInOutAdjust::ExpandFilterShape(const AnfNodePtr &weight_node,
switch (format) {
case schema::Format_NCHW:
case schema::Format_KCHW:
new_shape.insert(new_shape.begin() + 2, 1);
new_shape.insert(new_shape.begin() + kNumDim2, 1);
break;
case schema::Format_NHWC:
case schema::Format_KHWC:

3
mindspore/lite/tools/converter/parser/onnx/onnx_model_parser.cc
View File

@ -122,8 +122,7 @@ STATUS OnnxModelParser::WeightFormatTransform(const std::set<FuncGraphPtr> &all_
auto weight_node = conv_cnode->input(kConvWeightIndex);
MS_ASSERT(weight_node != nullptr);
auto tensor_info = opt::GetTensorInfo(weight_node);
lite::STATUS status;
status = HardCodeONNX(conv_cnode, tensor_info, graph);
auto status = HardCodeONNX(conv_cnode, tensor_info, graph);
if (status != lite::RET_OK) {
MS_LOG(ERROR) << "Format hard code failed: " << status << ", node: " << node->fullname_with_scope();
return RET_ERROR;

7
mindspore/lite/tools/converter/parser/parser_utils.cc
View File

@ -28,13 +28,15 @@
#include "tools/optimizer/common/gllo_utils.h"
namespace mindspore::lite {
namespace {
constexpr size_t kNumWeightIndex = 2;
}
void GetAllFuncGraph(const FuncGraphPtr &func_graph, std::set<FuncGraphPtr> *all_func_graphs) {
if (all_func_graphs->find(func_graph) == all_func_graphs->end()) {
all_func_graphs->insert(func_graph);
} else {
return;
}
auto nodes = func_graph->nodes();
for (auto &node : nodes) {
if (IsValueNode<FuncGraph>(node)) {
@ -221,7 +223,7 @@ int TransposeInsertForWeightConst(const FuncGraphPtr &graph, const CNodePtr &con
prim->AddAttr("quant_params", std::make_shared<QuantParamHolder>(1, 1));
auto transpose_node = graph->NewCNode(prim, {weight_node, perm_node});
transpose_node->set_fullname_with_scope(weight_node->fullname_with_scope() + "_const_post");
conv_node->set_input(2, transpose_node);
conv_node->set_input(kNumWeightIndex, transpose_node);
return lite::RET_OK;
}
@ -244,5 +246,4 @@ int HandleWeightConst(const FuncGraphPtr &graph, const CNodePtr &conv_node, cons
}
return status;
}
} // namespace mindspore::lite

4
mindspore/lite/tools/converter/parser/tf/tf_model_parser.cc
View File

@ -608,8 +608,7 @@ STATUS TFModelParser::WeightFormatTransform(const FuncGraphPtr &graph) {
auto weight_node = conv_cnode->input(kConvWeightIndex);
MS_ASSERT(weight_node != nullptr);
auto tensor_info = opt::GetTensorInfo(weight_node);
lite::STATUS status;
status = HardCodeTF(conv_cnode, tensor_info, graph);
auto status = HardCodeTF(conv_cnode, tensor_info, graph);
if (status != lite::RET_OK) {
MS_LOG(ERROR) << "Format hard code failed: " << status << ", node: " << node->fullname_with_scope();
return RET_ERROR;
@ -647,7 +646,6 @@ STATUS TFModelParser::HardCodeTF(const CNodePtr &conv_node, const tensor::Tensor
prim->AddAttr(ops::kFormat, MakeValue<int64_t>(weight_dst_format));
weight_src_format = schema::Format::Format_HWKC;
}
} else if (opt::CheckPrimitiveType(conv_node, prim::kPrimConv2dTransposeFusion) && !is_depth_wise) {
prim->AddAttr(ops::kFormat, MakeValue<int64_t>(weight_dst_format));
weight_src_format = schema::Format::Format_HWCK;

3
mindspore/lite/tools/converter/parser/tflite/tflite_model_parser.cc
View File

@ -134,8 +134,7 @@ STATUS TfliteModelParser::WeightFormatTransform(const FuncGraphPtr &graph) {
auto weight_node = conv_cnode->input(kConvWeightIndex);
MS_ASSERT(weight_node != nullptr);
auto tensor_info = opt::GetTensorInfo(weight_node);
lite::STATUS status;
status = HardCodeTflite(conv_cnode, tensor_info, graph);
auto status = HardCodeTflite(conv_cnode, tensor_info, graph);
if (status != lite::RET_OK) {
MS_LOG(ERROR) << "Format hard code failed: " << status << ", node: " << node->fullname_with_scope();
return RET_ERROR;

4
mindspore/lite/tools/optimizer/fusion/conv_transform_fusion.cc
View File

@ -28,6 +28,8 @@ constexpr size_t kConvWeightIndex = 2;
constexpr size_t kConvBiasIndex = 3;
constexpr size_t kConvNoBiasLen = 3;
constexpr size_t kConvWithBiasLen = 4;
constexpr size_t kNumDim1 = 1;
constexpr size_t kNumDim2 = 2;
int GetOutChannels(const CNodePtr &conv_node) {
MS_ASSERT(conv_node != nullptr);
auto value_node = conv_node->input(0);
@ -72,7 +74,7 @@ void GenerateNewWeightConv2DTranspose(float *dst_weight, const float *scale_weig
MS_ASSERT(group > 0);
auto weight_data = reinterpret_cast<float *>(weight_tensor->data_c());
auto cin_group = weight_tensor->shape()[0] / group;
int area_size = weight_tensor->shape()[1] * weight_tensor->shape()[2];
int area_size = weight_tensor->shape()[kNumDim2] * weight_tensor->shape()[kNumDim2];
for (int k = 0; k < cin_group; ++k) {
for (int j = 0; j < area_size; j++) {
for (int i = 0; i < kernel_num; ++i) {

12
mindspore/lite/tools/optimizer/graph/update_conv2d_param_pass.cc
View File

@ -21,8 +21,12 @@
namespace mindspore::opt {
namespace {
constexpr size_t kNumDim0 = 0;
constexpr size_t kNumDim1 = 1;
constexpr size_t kNumDim2 = 2;
constexpr size_t kNumDim3 = 3;
constexpr int kAnfPopulaterInputNumTwo = 2;
}
} // namespace
lite::STATUS UpdateConv2DParamPass::UpdateCommonConv2D(const CNodePtr &cnode) {
MS_ASSERT(cnode != nullptr);
@ -55,10 +59,10 @@ lite::STATUS UpdateConv2DParamPass::UpdateCommonConv2D(const CNodePtr &cnode) {
auto default_param = weight_param->default_param();
auto weight_tensor = std::dynamic_pointer_cast<tensor::Tensor>(default_param);
auto weight_shape = weight_tensor->shape();
std::vector<int64_t> kernel_size = {weight_shape[1], weight_shape[2]};
std::vector<int64_t> kernel_size = {weight_shape[kNumDim1], weight_shape[kNumDim2]};
conv->set_kernel_size(kernel_size);
conv->set_in_channel(weight_shape[3]);
conv->set_out_channel(weight_shape[0]);
conv->set_in_channel(weight_shape[kNumDim3]);
conv->set_out_channel(weight_shape[kNumDim0]);
return lite::RET_OK;
}