Skip to content

Commit

Permalink
Remove tets code
Browse files Browse the repository at this point in the history
  • Loading branch information
@riverlijunjie
riverlijunjie committed Jul 15, 2023
1 parent b2da7f4 commit 60cdef6
Show file tree
Hide file tree
Showing 2 changed files with 46 additions and 134 deletions.
52 changes: 1 addition & 51 deletions src/plugins/intel_cpu/src/compiled_model.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -68,7 +68,6 @@ CompiledModel::CompiledModel(const std::shared_ptr<ov::Model>& model,
m_cfg.isLegacyApi = !core->is_new_api();

bool support_all_precision = true;
#if 0
auto get_graph_supported_precisions = [&]() -> std::set<ov::element::Type_t> {
std::set<ov::element::Type_t> supported_precisions = {ov::element::Type_t::u8,
ov::element::Type_t::i8,
Expand All @@ -93,7 +92,6 @@ CompiledModel::CompiledModel(const std::shared_ptr<ov::Model>& model,
support_all_precision = false;
}


if (!support_all_precision) {
m_graph_model = m_model->clone();
Transformations graph_transformations(m_graph_model,
Expand All @@ -106,55 +104,7 @@ CompiledModel::CompiledModel(const std::shared_ptr<ov::Model>& model,
} else {
m_graph_model = m_model;
}
#else
auto get_convert_precisions = []() {
precisions_map map = {{ov::element::i64, ov::element::i32},
{ov::element::u64, ov::element::i32},
{ov::element::i16, ov::element::i32},
{ov::element::u16, ov::element::i32},
{ov::element::u32, ov::element::i32},
{ov::element::f64, ov::element::f32},
{ov::element::f16, ov::element::f32},
{ov::element::boolean, ov::element::u8},
{ov::element::i4, ov::element::i8},
{ov::element::u4, ov::element::u8}};

if (!dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core))
map.insert({ov::element::bf16, ov::element::f32});

return map;
};
static const auto convert_precisions = get_convert_precisions();

m_graph_model = m_model->clone();
ov::preprocess::PrePostProcessor preproc(m_graph_model);
for (size_t i = 0; i < m_graph_model->inputs().size(); i++) {
auto port = m_graph_model->input(i);
auto prec = port.get_element_type();
std::cout << "input " << i << ", prec = " << prec << std::endl;
auto it = convert_precisions.find(prec);
if (it != convert_precisions.end()) {
preproc.input(i).tensor().set_element_type(it->second);
support_all_precision = false;
std::cout << "preproc: input precision convert " << it->first << " to " << it->second << std::endl;
}
}
for (size_t i = 0; i < m_graph_model->outputs().size(); i++) {
auto port = m_graph_model->output(i);
auto prec = port.get_element_type();
std::cout << "input " << i << ", prec = " << prec << std::endl;
auto it = convert_precisions.find(prec);
if (it != convert_precisions.end()) {
preproc.output(i).tensor().set_element_type(it->second);
support_all_precision = false;
std::cout << "preproc: output precision covert " << it->first << " to " << it->second << std::endl;
}
}
if (!support_all_precision)
preproc.build();
else
m_graph_model = m_model;
#endif

std::cout << std::endl << "support_all_precision = " << support_all_precision << std::endl;
for (const auto& in : m_graph_model->inputs()) {
auto port_name = get_port_name(in, false);
Expand Down
128 changes: 45 additions & 83 deletions src/plugins/intel_cpu/src/graph.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -867,99 +867,61 @@ void Graph::PushInputData(const std::string& name, const ov::SoPtr<ITensor> &in)

auto input = inputNodesMap.find(name);
if (input != inputNodesMap.end()) {
#if 0
auto create_mem_desc = [&](const ov::SoPtr<ITensor>& tensor,
InferenceEngine::Precision& precision) -> CpuBlockedMemoryDesc {
auto create_mem_desc = [&](const ov::SoPtr<ITensor>& tensor) -> CpuBlockedMemoryDesc {
InferenceEngine::TensorDesc tensorDesc(
precision,
ie::details::convertPrecision(tensor->get_element_type()),
tensor->get_shape(),
InferenceEngine::TensorDesc::getLayoutByRank(tensor->get_shape().size()));
return MemoryDescUtils::convertToCpuBlockedMemoryDesc(tensorDesc);
};
#else
auto create_mem_desc = [&](const ov::SoPtr<ITensor>& tensor) -> CpuBlockedMemoryDesc {
auto element_type = tensor->get_element_type();
auto shape = tensor->get_shape();
std::vector<size_t> blk_order(shape.size());
std::iota(blk_order.begin(), blk_order.end(), 0);
std::vector<size_t> dim_offset(shape.size(), 0);
std::vector<size_t> blk_strides;
auto byte_strides = element_type.bitwidth() >= 8 ? tensor->get_strides() : Strides{};
Shape mem_shape(shape);
if (mem_shape.hasZeroDims()) {
blk_strides = std::vector<size_t>(shape.size(), 0);
} else if (byte_strides.empty()) {
blk_strides = ov::row_major_strides(shape);
} else {
blk_strides.resize(byte_strides.size());
std::transform(byte_strides.begin(),
byte_strides.end(),
blk_strides.begin(),
[&element_type](size_t byte_stride) {
OPENVINO_ASSERT(byte_stride % element_type.size() == 0,
"Limitation: Stride in bytes ",
byte_stride,
" should be divisible by size of element ",
element_type.size());
return byte_stride / element_type.size();
});

auto node = input->second;
auto childEdge = node->getChildEdgeAt(0);
const auto& outDims = node->getOutputShapeAtPort(0);

const void* ext_data_ptr = in->data();
void* inter_data_ptr = childEdge->getMemory().getData();

// Convert data if precision mismatch
auto& inter_mem_desc = childEdge->getMemory().getDesc();
auto inter_precision = inter_mem_desc.getPrecision();
auto ext_precision = ie::details::convertPrecision(in->get_element_type());
auto ext_tensor_desc = create_mem_desc(in);
if (ext_precision != inter_precision) {
if ((inter_data_ptr == nullptr) || (ext_data_ptr == nullptr)) {
OPENVINO_THROW("Get tensor has no allocated memory");
}
return CpuBlockedMemoryDesc(ie::details::convertPrecision(element_type),
mem_shape,
shape,
blk_order,
0,
dim_offset,
blk_strides);
};
#endif
auto node = input->second;
auto childEdge = node->getChildEdgeAt(0);
const auto& outDims = node->getOutputShapeAtPort(0);

const void* ext_data_ptr = in->data();
void* inter_data_ptr = childEdge->getMemory().getData();

// Convert data if precision mismatch
auto& inter_mem_desc = childEdge->getMemory().getDesc();
auto inter_precision = inter_mem_desc.getPrecision();
auto ext_precision = ie::details::convertPrecision(in->get_element_type());
auto ext_tensor_desc = create_mem_desc(in);
if (ext_precision != inter_precision) {
if ((inter_data_ptr == nullptr) || (ext_data_ptr == nullptr)) {
OPENVINO_THROW("Get tensor has no allocated memory");
}
cpu_convert(ext_data_ptr, inter_data_ptr, ext_precision, inter_precision, in->get_size());
std::cout << "push_input: convert data " << ext_precision << " to " << inter_precision << std::endl;
cpu_convert(ext_data_ptr, inter_data_ptr, ext_precision, inter_precision, in->get_size());
std::cout << "push_input: convert data " << ext_precision << " to " << inter_precision << std::endl;

std::cout << "input data: " << std::endl;
float* p = static_cast<float*>(inter_data_ptr);
for (size_t i = 0; i < in->get_size(); i++) {
std::cout << " " << p[i];
}
std::cout << std::endl;

Memory mem(getEngine(), inter_mem_desc, inter_data_ptr, false);
childEdge->getMemory().load(mem, false);
} else if (ext_data_ptr != inter_data_ptr) {
Memory ext_mem(getEngine(), inter_mem_desc, ext_data_ptr, false);
childEdge->getMemory().load(ext_mem, false);
std::cout << "push_input: update data " << ext_precision << " to " << inter_precision << std::endl;
std::cout << "input data: " << std::endl;
float* p = static_cast<float*>(inter_data_ptr);
for (size_t i = 0; i < in->get_size() && i < 10; i++) {
std::cout << " " << p[i];
}
std::cout << std::endl;

Memory mem(getEngine(), inter_mem_desc, inter_data_ptr, false);
childEdge->getMemory().load(mem, false);
} else if (ext_data_ptr != inter_data_ptr) {
Memory ext_mem(getEngine(), inter_mem_desc, ext_data_ptr, false);
childEdge->getMemory().load(ext_mem, false);
std::cout << "push_input: update data " << ext_precision << " to " << inter_precision << std::endl;
}

// todo: make sure 'name' exists in this map...
if (_normalizePreprocMap.find(name) != _normalizePreprocMap.end()) {
if (ext_tensor_desc.getPrecision() == InferenceEngine::Precision::FP32) {
_normalizePreprocMap[name].NormalizeImage(outDims,
reinterpret_cast<float*>(inter_data_ptr),
TensorDesc::getLayoutByDims(in->get_shape()));
} else {
OPENVINO_THROW("Mean image of type ", ext_tensor_desc.getPrecision().name(), " is unsupported");
}
// todo: make sure 'name' exists in this map...
if (_normalizePreprocMap.find(name) != _normalizePreprocMap.end()) {
if (ext_tensor_desc.getPrecision() == InferenceEngine::Precision::FP32) {
_normalizePreprocMap[name].NormalizeImage(outDims,
reinterpret_cast<float*>(inter_data_ptr),
TensorDesc::getLayoutByDims(in->get_shape()));
} else {
OPENVINO_THROW("Mean image of type ", ext_tensor_desc.getPrecision().name(), " is unsupported");
}
} else {
OPENVINO_THROW("Input blob for infer '", name, "' doesn't correspond to input in network");
}
} else {
OPENVINO_THROW("Input blob for infer '", name, "' doesn't correspond to input in network");
}
}

void Graph::PullOutputData(std::unordered_map<std::string, ov::SoPtr<ITensor>>& out) {
Expand Down Expand Up @@ -1029,7 +991,7 @@ void Graph::PullOutputData(std::unordered_map<std::string, ov::SoPtr<ITensor>>&
std::cout << "pull_output: " << srcPrec << " to " << dstPrec << ", same_ptr "<< (ext_blob_ptr == intr_blob_ptr) << std::endl;
std::cout << "output data: " << std::endl;
float* p = static_cast<float*>(intr_blob_ptr);
for (size_t i = 0; i < ext_blob->get_size(); i++) {
for (size_t i = 0; i < ext_blob->get_size() && i < 10; i++) {
std::cout << " " << p[i];
}
std::cout << std::endl;
Expand Down

0 comments on commit 60cdef6

Please sign in to comment.