Skip to content

N
ngraph
Commit dc4acb4e authored by Louis Feng's avatar Louis Feng
Browse files
Options

added conv+bias forward mkldnn.

parent 38b28c13
Show whitespace changes
Inline Side-by-side
Showing with 89 additions and 19 deletions
src/ngraph/runtime/cpu/cpu_emitter.cpp
View file @ dc4acb4e
......@@ -86,6 +86,7 @@
#include "ngraph/ops/tanh.hpp"
#include "ngraph/runtime/cpu/cpu_emitter.hpp"
#include "ngraph/runtime/cpu/cpu_kernel_emitters.hpp"
#include "ngraph/runtime/cpu/ops/conv_bias.hpp"
#include "ngraph/runtime/cpu/ops/convert_layout.hpp"
#include "ngraph/runtime/cpu/ops/matmul_bias.hpp"
#include "ngraph/types/element_type.hpp"
......@@ -2048,18 +2049,6 @@ namespace ngraph
out[0], mkldnn::memory::format::nchw);
size_t conv_index = 0;
if (!filter_dilated)
{
conv_index = mkldnn_emitter->build_convolution_forward(
input_data_desc,
weights_desc,
result_desc,
convolution->get_window_movement_strides(),
convolution->get_padding_below(),
convolution->get_padding_above());
}
else
{
// For dilation, MKLDNN wants to know how many elements to insert between, not how far
// apart to space the elements like nGraph. So we have to subtract 1 from each pos.
Strides window_dilation_strides_adjusted;
......@@ -2077,7 +2066,6 @@ namespace ngraph
window_dilation_strides_adjusted,
convolution->get_padding_below(),
convolution->get_padding_above());
}
auto& deps = mkldnn_emitter->get_primitive_deps(conv_index);
writer << "cpu::mkldnn_utils::set_memory_ptr(ctx, " << to_string(deps[0])
......@@ -2340,6 +2328,81 @@ namespace ngraph
}
}
template <>
void CPU_Emitter::EMITTER_DECL(ngraph::op::ConvolutionBias)
{
auto convolution = static_cast<const ngraph::op::Convolution*>(node);
const TensorViewWrapper& data = args[0];
const TensorViewWrapper& weights = args[1];
const TensorViewWrapper& bias = args[2];
const TensorViewWrapper& result = out[0];
const vector<size_t>& data_shape = data.get_shape();
const vector<size_t>& weights_shape = weights.get_shape();
const vector<size_t>& bias_shape = bias.get_shape();
const vector<size_t>& result_shape = result.get_shape();
const size_t data_rank = data_shape.size();
const size_t weights_rank = weights_shape.size();
const element::Type& elem_type = data.get_element_type();
bool data_dilated = false;
for (size_t s : convolution->get_data_dilation_strides())
{
data_dilated = data_dilated || (s != 1);
}
if (!data_dilated && data_rank == 4 && weights_rank == 4 &&
elem_type == element::f32)
{
auto& mkldnn_emitter = external_function->get_mkldnn_emitter();
auto data_desc = mkldnn_emitter->build_memory_descriptor(
data, mkldnn::memory::format::nchw);
auto weights_desc = mkldnn_emitter->build_memory_descriptor(
weights, mkldnn::memory::format::oihw);
auto bias_desc = mkldnn_emitter->build_memory_descriptor(
bias, mkldnn::memory::format::x);
auto result_desc = mkldnn_emitter->build_memory_descriptor(
result, mkldnn::memory::format::nchw);
size_t conv_index = 0;
// For dilation, MKLDNN wants to know how many elements to insert between, not how far
// apart to space the elements like nGraph. So we have to subtract 1 from each pos.
Strides window_dilation_strides_adjusted;
for (size_t s : convolution->get_window_dilation_strides())
{
window_dilation_strides_adjusted.push_back(s - 1);
}
conv_index = mkldnn_emitter->build_convolution_forward(
data_desc,
weights_desc,
bias_desc,
result_desc,
convolution->get_window_movement_strides(),
window_dilation_strides_adjusted,
convolution->get_padding_below(),
convolution->get_padding_above());
auto& deps = mkldnn_emitter->get_primitive_deps(conv_index);
writer << "cpu::mkldnn_utils::set_memory_ptr(ctx, " << to_string(deps[0])
<< ", " << data.get_name() << ");\n";
writer << "cpu::mkldnn_utils::set_memory_ptr(ctx, " << to_string(deps[1])
<< ", " << weights.get_name() << ");\n";
writer << "cpu::mkldnn_utils::set_memory_ptr(ctx, " << to_string(deps[2])
<< ", " << bias.get_name() << ");\n";
writer << "cpu::mkldnn_utils::set_memory_ptr(ctx, " << to_string(deps[3])
<< ", " << result.get_name() << ");\n";
writer << "cpu::mkldnn_utils::mkldnn_invoke_primitive(ctx, "
<< to_string(conv_index) << ");\n";
}
else
{
throw ngraph_error("ConvolutionBias does not yet support this layout rank: "+std::to_string(data_rank));
}
}
template <>
void CPU_Emitter::EMITTER_DECL(ngraph::op::Not)
{
......
src/ngraph/runtime/cpu/mkldnn_emitter.cpp
View file @ dc4acb4e
......@@ -76,6 +76,7 @@ size_t MKLDNNEmitter::build_convolution_forward(const mkldnn::memory::desc& inpu
const mkldnn::memory::desc& weights_desc,
const mkldnn::memory::desc& result_desc,
const ngraph::Strides& strides,
const ngraph::Strides& dilation_strides,
const ngraph::CoordinateDiff& padding_below,
const ngraph::CoordinateDiff& padding_above)
......@@ -91,6 +92,7 @@ size_t MKLDNNEmitter::build_convolution_forward(const mkldnn::memory::desc& inpu
weights_desc,
result_desc,
mkldnn::memory::dims(strides.begin(), strides.end()),
mkldnn::memory::dims(dilation_strides.begin(), dilation_strides.end()),
mkldnn::memory::dims(padding_below.begin(), padding_below.end()),
mkldnn::memory::dims(padding_above.begin(), padding_above.end()),
mkldnn::padding_kind::zero},
......@@ -105,6 +107,7 @@ size_t MKLDNNEmitter::build_convolution_forward(const mkldnn::memory::desc& inpu
size_t MKLDNNEmitter::build_convolution_forward(const mkldnn::memory::desc& input_data_desc,
const mkldnn::memory::desc& weights_desc,
const mkldnn::memory::desc& bias_desc,
const mkldnn::memory::desc& result_desc,
const ngraph::Strides& strides,
const ngraph::Strides& dilation_strides,
......@@ -112,15 +115,17 @@ size_t MKLDNNEmitter::build_convolution_forward(const mkldnn::memory::desc& inpu
const ngraph::CoordinateDiff& padding_above)
{
size_t input_data_index = build_memory_primitive(input_data_desc);
size_t weights_index = build_memory_primitive(weights_desc);
size_t result_index = build_memory_primitive(result_desc);
const size_t input_data_index = build_memory_primitive(input_data_desc);
const size_t weights_index = build_memory_primitive(weights_desc);
const size_t bias_index = build_memory_primitive(bias_desc);
const size_t result_index = build_memory_primitive(result_desc);
size_t conv_index = insert_primitive(new mkldnn::convolution_forward(
const size_t conv_index = insert_primitive(new mkldnn::convolution_forward(
{{mkldnn::prop_kind::forward,
mkldnn::algorithm::convolution_direct,
input_data_desc,
weights_desc,
bias_desc,
result_desc,
mkldnn::memory::dims(strides.begin(), strides.end()),
mkldnn::memory::dims(dilation_strides.begin(), dilation_strides.end()),
......@@ -130,8 +135,9 @@ size_t MKLDNNEmitter::build_convolution_forward(const mkldnn::memory::desc& inpu
mkldnn_utils::global_cpu_engine},
*mkldnn_primitives[input_data_index],
*mkldnn_primitives[weights_index],
*mkldnn_primitives[bias_index],
*mkldnn_primitives[result_index]));
primitive_deps[conv_index] = {input_data_index, weights_index, result_index};
primitive_deps[conv_index] = {input_data_index, weights_index, bias_index, result_index};
return conv_index;
}
src/ngraph/runtime/cpu/mkldnn_emitter.hpp
View file @ dc4acb4e
......@@ -57,17 +57,18 @@ namespace ngraph
const mkldnn::memory::desc& weights_desc,
const mkldnn::memory::desc& result_desc,
const ngraph::Strides& strides,
const ngraph::Strides& dilation_strides,
const ngraph::CoordinateDiff& padding_below,
const ngraph::CoordinateDiff& padding_above);
size_t build_convolution_forward(const mkldnn::memory::desc& input_data_desc,
const mkldnn::memory::desc& weights_desc,
const mkldnn::memory::desc& bias_desc,
const mkldnn::memory::desc& result_desc,
const ngraph::Strides& strides,
const ngraph::Strides& dilation_strides,
const ngraph::CoordinateDiff& padding_below,
const ngraph::CoordinateDiff& padding_above);
private:
std::shared_ptr<CPU_ExternalFunction> external_function;
std::vector<mkldnn::primitive*> mkldnn_primitives;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment