From dc4acb4ea6a7c45e0ffea2c97388a92807332f68 Mon Sep 17 00:00:00 2001
From: Louis Feng <louis.feng@intel.com>
Date: Thu, 22 Feb 2018 16:01:31 -0800
Subject: [PATCH] added conv+bias forward mkldnn.
---
src/ngraph/runtime/cpu/cpu_emitter.cpp | 117 +++++++++++++++++-----
src/ngraph/runtime/cpu/mkldnn_emitter.cpp | 74 +++++++-------
src/ngraph/runtime/cpu/mkldnn_emitter.hpp | 3 +-
3 files changed, 132 insertions(+), 62 deletions(-)
diff --git a/src/ngraph/runtime/cpu/cpu_emitter.cpp b/src/ngraph/runtime/cpu/cpu_emitter.cpp
index 2dea139b8..5eebfbebe 100644
--- a/src/ngraph/runtime/cpu/cpu_emitter.cpp
+++ b/src/ngraph/runtime/cpu/cpu_emitter.cpp
@@ -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"
@@ -2048,36 +2049,23 @@ namespace ngraph
out[0], mkldnn::memory::format::nchw);
size_t conv_index = 0;
- if (!filter_dilated)
+ // 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())
{
- 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());
+ window_dilation_strides_adjusted.push_back(s - 1);
}
- 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;
-
- for (size_t s : convolution->get_window_dilation_strides())
- {
- window_dilation_strides_adjusted.push_back(s - 1);
- }
- conv_index = mkldnn_emitter->build_convolution_forward(
- input_data_desc,
- weights_desc,
- result_desc,
- convolution->get_window_movement_strides(),
- window_dilation_strides_adjusted,
- convolution->get_padding_below(),
- convolution->get_padding_above());
- }
+ conv_index = mkldnn_emitter->build_convolution_forward(
+ input_data_desc,
+ weights_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])
@@ -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)
{
diff --git a/src/ngraph/runtime/cpu/mkldnn_emitter.cpp b/src/ngraph/runtime/cpu/mkldnn_emitter.cpp
index ffe9e59d5..821b4a6f8 100644
--- a/src/ngraph/runtime/cpu/mkldnn_emitter.cpp
+++ b/src/ngraph/runtime/cpu/mkldnn_emitter.cpp
@@ -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)
@@ -85,19 +86,20 @@ size_t MKLDNNEmitter::build_convolution_forward(const mkldnn::memory::desc& inpu
size_t result_index = build_memory_primitive(result_desc);
size_t conv_index = insert_primitive(new mkldnn::convolution_forward(
- {{mkldnn::prop_kind::forward,
- mkldnn::algorithm::convolution_direct,
- input_data_desc,
- weights_desc,
- result_desc,
- mkldnn::memory::dims(strides.begin(), strides.end()),
- mkldnn::memory::dims(padding_below.begin(), padding_below.end()),
- mkldnn::memory::dims(padding_above.begin(), padding_above.end()),
- mkldnn::padding_kind::zero},
- mkldnn_utils::global_cpu_engine},
- *mkldnn_primitives[input_data_index],
- *mkldnn_primitives[weights_index],
- *mkldnn_primitives[result_index]));
+ {{mkldnn::prop_kind::forward,
+ mkldnn::algorithm::convolution_direct,
+ input_data_desc,
+ 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},
+ mkldnn_utils::global_cpu_engine},
+ *mkldnn_primitives[input_data_index],
+ *mkldnn_primitives[weights_index],
+ *mkldnn_primitives[result_index]));
primitive_deps[conv_index] = {input_data_index, weights_index, result_index};
return conv_index;
@@ -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,
@@ -112,26 +115,29 @@ 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);
-
- size_t conv_index = insert_primitive(new mkldnn::convolution_forward(
- {{mkldnn::prop_kind::forward,
- mkldnn::algorithm::convolution_direct,
- input_data_desc,
- 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},
- mkldnn_utils::global_cpu_engine},
- *mkldnn_primitives[input_data_index],
- *mkldnn_primitives[weights_index],
- *mkldnn_primitives[result_index]));
-
- primitive_deps[conv_index] = {input_data_index, weights_index, result_index};
+ 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);
+
+ 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()),
+ mkldnn::memory::dims(padding_below.begin(), padding_below.end()),
+ mkldnn::memory::dims(padding_above.begin(), padding_above.end()),
+ mkldnn::padding_kind::zero},
+ 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, bias_index, result_index};
return conv_index;
}
diff --git a/src/ngraph/runtime/cpu/mkldnn_emitter.hpp b/src/ngraph/runtime/cpu/mkldnn_emitter.hpp
index 2b4b664a9..c506ef04b 100644
--- a/src/ngraph/runtime/cpu/mkldnn_emitter.hpp
+++ b/src/ngraph/runtime/cpu/mkldnn_emitter.hpp
@@ -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;
--
2.18.0