nGraph v0.26.0.rc.0 (#249)

CMakeLists.txt

@@ -248,7 +248,7 @@ if (NOT USE_PRE_BUILT_NGRAPH)
     ExternalProject_Add(
         ext_ngraph
         GIT_REPOSITORY https://github.com/NervanaSystems/ngraph
-        GIT_TAG v0.25.1-rc.2
+        GIT_TAG v0.26.0-rc.0
         CMAKE_ARGS 
             -DNGRAPH_DISTRIBUTED_ENABLE=${NGRAPH_DISTRIBUTED_ENABLE}
             -DNGRAPH_INSTALL_PREFIX=${NGRAPH_ARTIFACTS_DIR}

README.md

@@ -88,7 +88,7 @@ Once TensorFlow's dependencies are installed, clone the `ngraph-bridge` repo:
 
         git clone https://github.com/tensorflow/ngraph-bridge.git
         cd ngraph-bridge
-        git checkout v0.19.0-rc1
+        git checkout v0.26.0-rc.0
 
 Run the following Python script to build TensorFlow, nGraph, and the bridge. Use Python 3.5:
 

bazel/WORKSPACE

@@ -55,11 +55,11 @@ tf_workspace(path_prefix = "", tf_repo_name = "org_tensorflow")
 http_archive(
     name = "ngraph",
     build_file = "//:bazel/ngraph.BUILD",
-    sha256 = "d3e0ebf807dd4179c91a41c80639b75f5ea8ca39cdfe3a5697da56a7df894f11",
-    strip_prefix = "ngraph-0.25.1-rc.2",
+    sha256 = "d6832d9f923027aa5cc9f49c319fe537d73102600576e4e69be19a0e37d13a43",
+    strip_prefix = "ngraph-0.26.0-rc.0",
     urls = [
-        "https://mirror.bazel.build/github.com/NervanaSystems/ngraph/archive/v0.25.1-rc.2.tar.gz",
-        "https://github.com/NervanaSystems/ngraph/archive/v0.25.1-rc.2.tar.gz"
+        "https://mirror.bazel.build/github.com/NervanaSystems/ngraph/archive/v0.26.0-rc.0.tar.gz",
+        "https://github.com/NervanaSystems/ngraph/archive/v0.26.0-rc.0.tar.gz"
     ],
 )
 

bazel/ngraph.BUILD

@@ -45,12 +45,10 @@ cc_library(
         "src/ngraph/op/experimental/dyn_reshape.cpp",
         "src/ngraph/op/experimental/dyn_slice.cpp",
         "src/ngraph/op/experimental/generate_mask.cpp",
-        "src/ngraph/op/experimental/quantized_avg_pool.cpp",
         "src/ngraph/op/experimental/quantized_concat.cpp",
         "src/ngraph/op/experimental/quantized_conv.cpp",
         "src/ngraph/op/experimental/quantized_conv_bias.cpp",
         "src/ngraph/op/experimental/quantized_conv_relu.cpp",
-        "src/ngraph/op/experimental/quantized_max_pool.cpp",
         "src/ngraph/op/experimental/shape_of.cpp",
         "src/ngraph/op/experimental/range.cpp",
         "src/ngraph/op/experimental/quantized_dot.cpp",
@@ -84,7 +82,7 @@ cc_library(
         "-fstack-protector-all",
         '-D SHARED_LIB_PREFIX=\\"lib\\"',
         '-D SHARED_LIB_SUFFIX=\\".so\\"',
-        '-D NGRAPH_VERSION=\\"v0.25.1-rc.2\\"',
+        '-D NGRAPH_VERSION=\\"v0.26.0-rc.0\\"',
         "-D NGRAPH_DEX_ONLY",
         '-D PROJECT_ROOT_DIR=\\"\\"',
         '-D NGRAPH_STATIC_LIB_ENABLE'
@@ -117,7 +115,7 @@ cc_library(
         "-fstack-protector-all",
         '-D SHARED_LIB_PREFIX=\\"lib\\"',
         '-D SHARED_LIB_SUFFIX=\\".so\\"',
-        '-D NGRAPH_VERSION=\\"v0.25.1-rc.2\\"',
+        '-D NGRAPH_VERSION=\\"v0.26.0-rc.0\\"',
         "-D NGRAPH_DEX_ONLY",
         '-D PROJECT_ROOT_DIR=\\"\\"',
     ] + CXX_ABI,
@@ -129,6 +127,9 @@ cc_library(
     visibility = ["//visibility:public"],
     alwayslink = 1,
 )
+# TODO: If we update to mkl_dnn v1.0 in future, we should include
+# the source file "src/ngraph/runtime/cpu/pass/cpu_mkldnn_primitive_build.cpp"
+# Currently we use legacy mkl_dnn, NGRAPH_USE_LEGACY_MKLDNN is set to TRUE by default
 
 cc_library(
     name = 'cpu_backend',
@@ -188,12 +189,10 @@ cc_library(
         "src/ngraph/runtime/cpu/builder/reduce_function.cpp",
         "src/ngraph/runtime/cpu/builder/replace_slice.cpp",
         "src/ngraph/runtime/cpu/builder/quantization.cpp",
-        "src/ngraph/runtime/cpu/builder/quantized_avg_pool.cpp",
         "src/ngraph/runtime/cpu/builder/quantized_conv.cpp",
         "src/ngraph/runtime/cpu/builder/quantized_concat.cpp",
         "src/ngraph/runtime/cpu/builder/quantized_dot.cpp",
         "src/ngraph/runtime/cpu/builder/quantized_matmul.cpp",
-        "src/ngraph/runtime/cpu/builder/quantized_max_pool.cpp",
         "src/ngraph/runtime/cpu/builder/reshape.cpp",
         "src/ngraph/runtime/cpu/builder/reverse.cpp",
         "src/ngraph/runtime/cpu/builder/reverse_sequence.cpp",
@@ -245,7 +244,6 @@ cc_library(
         "src/ngraph/runtime/cpu/pass/cpu_mat_fusion.cpp",
         "src/ngraph/runtime/cpu/pass/cpu_memory_assignment.cpp",
         "src/ngraph/runtime/cpu/pass/cpu_memory_optimization.cpp",
-        "src/ngraph/runtime/cpu/pass/cpu_mkldnn_primitive_build.cpp",
         "src/ngraph/runtime/cpu/pass/cpu_post_layout_optimizations.cpp",
         "src/ngraph/runtime/cpu/pass/cpu_rnn_fusion.cpp",
         "src/ngraph/runtime/cpu/pass/cpu_workspace_insertion.cpp",
@@ -269,7 +267,7 @@ cc_library(
         "-fstack-protector-all",
         '-D SHARED_LIB_PREFIX=\\"lib\\"',
         '-D SHARED_LIB_SUFFIX=\\".so\\"',
-        '-D NGRAPH_VERSION=\\"0.25.1-rc.2\\"',
+        '-D NGRAPH_VERSION=\\"v0.26.0-rc.0\\"',
         "-D NGRAPH_DEX_ONLY",
         "-D NGRAPH_TBB_ENABLE",
         '-D PROJECT_ROOT_DIR=\\"\\"',

build_ngtf.py

@@ -53,7 +53,7 @@ def main():
     '''
 
     # Component versions
-    ngraph_version = "v0.25.1-rc.2"
+    ngraph_version = "v0.26.0-rc.0"
     tf_version = "v1.14.0"
 
     # Command line parser options

ngraph_bridge/ngraph_builder.cc

@@ -22,8 +22,9 @@
 #include "tensorflow/core/lib/core/errors.h"
 
 #include "ngraph/builder/autobroadcast.hpp"
+#include "ngraph/builder/dequantize_builder.hpp"
 #include "ngraph/builder/numpy_transpose.hpp"
-#include "ngraph/builder/quantization.hpp"
+#include "ngraph/builder/quantize_builder.hpp"
 #include "ngraph/op/argmax.hpp"
 #include "ngraph/op/argmin.hpp"
 #include "ngraph/op/util/logical_reduction.hpp"
@@ -513,14 +514,14 @@ static Status TranslateQuantizedPoolOp(const Node* op,
   if (is_quantizedAvgPool) {
     // QuantizeAvgPool
     // TF doesn't include padding in avg calculation
-    ng_quant_pool = ng::builder::ScaledQuantizedAvgPool(
-        ng_input, ng_kernel_shape, ng_strides, ng_padding_below,
-        ng_padding_above, false, dummy_min, dummy_max);
+    ng_quant_pool = ConstructNgNode<ng::op::AvgPool>(
+        op->name(), ng_input, ng_kernel_shape, ng_strides, ng_padding_below,
+        ng_padding_above, false);
   } else {
     // QuantizeMaxPool
-    ng_quant_pool = ng::builder::ScaledQuantizedMaxPool(
-        ng_input, ng_kernel_shape, ng_strides, ng_padding_below,
-        ng_padding_above, dummy_min, dummy_max);
+    ng_quant_pool = ConstructNgNode<ng::op::MaxPool>(
+        op->name(), ng_input, ng_kernel_shape, ng_strides, ng_padding_below,
+        ng_padding_above);
   }
 
   BatchToTensorflow(is_nhwc, ng_quant_pool);
@@ -3305,7 +3306,7 @@ static Status TranslateQuantizedConcatOpHelper(
   shared_ptr<ng::Node> ng_max_of_maxs =
       make_shared<ng::op::Constant>(ng::element::f32, ng::Shape{}, max_of_maxs);
 
-  auto ng_qconcat = ng::builder::ScaledQuantizedConcat(
+  auto ng_qconcat = ng::builder::QuantizedConcatBuilder(
       ng_args, size_t(concat_axis), ng_all_mins, ng_all_maxs);
 
   SaveNgOp(ng_op_map, op->name(), ng_qconcat);
@@ -3373,7 +3374,7 @@ static Status TranslateQuantizedConv(
                        ng_strides, ng_dilations, ng_padding_below,
                        ng_padding_above);
 
-  // It is expected by ScaledQuantizedConvolutionBias (and other builder
+  // It is expected by QuantizedConvolutionBiasBuilder (and other builder
   // functions) that the min max inputs be constant nodes
   // Hence declaring them static, reading their values and converting to
   // constant nodes
@@ -3401,7 +3402,7 @@ static Status TranslateQuantizedConv2DWithBiasMaybeReluAndRequantizeOp(
       std::vector<std::shared_ptr<ng::Node>> node_inps, ng::Strides ng_strides,
       ng::Strides ng_dilations, ng::CoordinateDiff ng_padding_below,
       ng::CoordinateDiff ng_padding_above, ng::Strides ng_data_dilations) {
-    return ng::builder::ScaledQuantizedConvolutionBias(
+    return ng::builder::QuantizedConvolutionBiasBuilder(
         node_inps[0], node_inps[1], node_inps[2], ng_strides, ng_dilations,
         ng_padding_below, ng_padding_above, ng_data_dilations, node_inps[3],
         node_inps[4], node_inps[5], node_inps[6], node_inps[7], node_inps[8],
@@ -3417,7 +3418,7 @@ static Status TranslateQuantizedConv2DWithBiasSumAndReluAndRequantizeOp(
       std::vector<std::shared_ptr<ng::Node>> node_inps, ng::Strides ng_strides,
       ng::Strides ng_dilations, ng::CoordinateDiff ng_padding_below,
       ng::CoordinateDiff ng_padding_above, ng::Strides ng_data_dilations) {
-    return ng::builder::ScaledQuantizedConvolutionBiasAdd(
+    return ng::builder::QuantizedConvolutionBiasAddBuilder(
         node_inps[0], node_inps[1], node_inps[2], node_inps[9], ng_strides,
         ng_dilations, ng_padding_below, ng_padding_above, ng_data_dilations,
         node_inps[3], node_inps[4], node_inps[5], node_inps[6], node_inps[7],
@@ -3433,7 +3434,7 @@ static Status TranslateQuantizedConv2DWithBiasSignedSumAndReluAndRequantizeOp(
       std::vector<std::shared_ptr<ng::Node>> node_inps, ng::Strides ng_strides,
       ng::Strides ng_dilations, ng::CoordinateDiff ng_padding_below,
       ng::CoordinateDiff ng_padding_above, ng::Strides ng_data_dilations) {
-    return ng::builder::ScaledQuantizedConvolutionBiasSignedAdd(
+    return ng::builder::QuantizedConvolutionBiasSignedAddBuilder(
         node_inps[0], node_inps[1], node_inps[2], node_inps[9], ng_strides,
         ng_dilations, ng_padding_below, ng_padding_above, ng_data_dilations,
         node_inps[3], node_inps[4], node_inps[5], node_inps[6], node_inps[7],
@@ -3467,8 +3468,8 @@ static Status TranslateQuantizeV2Op(const Node* op,
       ng::op::Quantize::RoundMode::ROUND_NEAREST_TOWARD_EVEN;
 
   SaveNgOp(ng_op_map, op->name(),
-           ng::builder::ScaledQuantize(ng_input, ng_min, ng_max, ng_et,
-                                       ng::AxisSet(), ng_round_mode));
+           ng::builder::QuantizeBuilder(ng_input, ng_min, ng_max, ng_et,
+                                        ng::AxisSet(), ng_round_mode));
   SaveNgOp(ng_op_map, op->name(), ng_min);
   SaveNgOp(ng_op_map, op->name(), ng_max);
 
@@ -3483,8 +3484,8 @@ static Status TranslateDequantizeOp(const Node* op,
 
   // TF only dequantizes to fp32
   SaveNgOp(ng_op_map, op->name(),
-           ng::builder::ScaledDequantize(ng_input, ng_min, ng_max,
-                                         ng::element::f32, ng::AxisSet()));
+           ng::builder::DequantizeBuilder(ng_input, ng_min, ng_max,
+                                          ng::element::f32, ng::AxisSet()));
   return Status::OK();
 }
 

ngraph_bridge/ngraph_encapsulate_impl.cc

@@ -336,7 +336,7 @@ Status NGraphEncapsulateImpl::AllocateNGInputTensors(
         std::unique_ptr<ngraph::Event> event_copy_input_next(
             new ngraph::Event(event_name, m_name, ""));
         current_ng_tensor->write(
-            current_src_ptr, 0,
+            current_src_ptr,
             current_ng_tensor->get_element_count() * ng_element_type.size());
 
         event_copy_input_next->Stop();

ngraph_bridge/ngraph_encapsulate_op.cc

@@ -630,8 +630,8 @@ void NGraphEncapsulateOp::Compute(OpKernelContext* ctx) {
             "Output_" + to_string(i) + "_" + to_string(copy_size);
         std::unique_ptr<ngraph::Event> event_copy_output_next(
             new ngraph::Event(event_name, name(), ""));
-        dst_ng_tensor->read(dst_ptr, 0, dst_ng_tensor->get_element_count() *
-                                            ng_element_type.size());
+        dst_ng_tensor->read(dst_ptr, dst_ng_tensor->get_element_count() *
+                                         ng_element_type.size());
         event_copy_output_next->Stop();
         output_copy_events.push_back(std::move(event_copy_output_next));
       }
@@ -648,8 +648,8 @@ void NGraphEncapsulateOp::Compute(OpKernelContext* ctx) {
              std::to_string(dst_ng_tensor->get_element_count() *
                             ng_element_type.size())),
             name(), ""));
-        dst_ng_tensor->read(dst_ptr, 0, dst_ng_tensor->get_element_count() *
-                                            ng_element_type.size());
+        dst_ng_tensor->read(dst_ptr, dst_ng_tensor->get_element_count() *
+                                         ng_element_type.size());
         event_copy_output_next->Stop();
         output_copy_events.push_back(std::move(event_copy_output_next));
       }

ngraph_bridge/ngraph_utils.cc

@@ -89,8 +89,8 @@ void ReadNGTensor(shared_ptr<ng::runtime::Tensor> ng_tensor,
                   Tensor* tf_tensor) {
   ngraph::Event event_sync_ng_tf_tensors("Tensor Read D2H", "", "");
   void* tf_src_ptr = (void*)DMAHelper::base(tf_tensor);
-  ng_tensor->read(tf_src_ptr, 0, ng_tensor->get_element_count() *
-                                     ng_tensor->get_element_type().size());
+  ng_tensor->read(tf_src_ptr, ng_tensor->get_element_count() *
+                                  ng_tensor->get_element_type().size());
   event_sync_ng_tf_tensors.Stop();
   ngraph::Event::write_trace(event_sync_ng_tf_tensors);
 }
@@ -100,8 +100,8 @@ void WriteNGTensor(shared_ptr<ng::runtime::Tensor> ng_tensor,
                    Tensor* tf_tensor) {
   ngraph::Event event_sync_ng_tf_tensors("Tensor Write H2D", "", "");
   void* tf_src_ptr = (void*)DMAHelper::base(tf_tensor);
-  ng_tensor->write(tf_src_ptr, 0, ng_tensor->get_element_count() *
-                                      ng_tensor->get_element_type().size());
+  ng_tensor->write(tf_src_ptr, ng_tensor->get_element_count() *
+                                   ng_tensor->get_element_type().size());
   event_sync_ng_tf_tensors.Stop();
   ngraph::Event::write_trace(event_sync_ng_tf_tensors);
 }

ngraph_bridge/ngraph_utils.h

@@ -178,7 +178,7 @@ template <typename T>
 T GetScalarFromTensor(const std::shared_ptr<ngraph::runtime::Tensor>& t,
                       size_t element_offset = 0) {
   T result;
-  t->read(&result, element_offset * sizeof(T), sizeof(T));
+  t->read(&result, sizeof(T));
   return result;
 }
 

ngraph_bridge/version.cc

@@ -24,15 +24,15 @@
 // nGraph-TensorFlow bridge uses semantic versioning: see http://semver.org/
 
 #define NG_TF_MAJOR_VERSION 0
-#define NG_TF_MINOR_VERSION 19
+#define NG_TF_MINOR_VERSION 20
 #define NG_TF_PATCH_VERSION 0
 
 // The version suffix is used for pre-release version numbers
 // For example before v0.7.0 we may do a pre-release i.e., a release
 // candidate such as v0.7.0-rc0
 // The code in master will always have the last released version number
 // with a suffix of '-master'
-#define NG_TF_VERSION_SUFFIX "-rc1"
+#define NG_TF_VERSION_SUFFIX "-rc0"
 
 #define VERSION_STR_HELPER(x) #x
 #define VERSION_STR(x) VERSION_STR_HELPER(x)

python/setup.in.py

@@ -59,7 +59,7 @@ def get_tag(self):
 
 setup( 
     name='ngraph_tensorflow_bridge',
-    version='0.19.0rc1',
+    version='0.20.0rc0',
     description='Intel nGraph compiler and runtime for TensorFlow',
     long_description=long_description,
     long_description_content_type="text/markdown",

test/graph_exec.cpp

@@ -85,10 +85,10 @@ TEST(graph_exec, axpy) {
 
   auto t_x = backend->create_tensor(ng::element::f32, ng_shape_x);
   float v_x[2][3] = {{1, 1, 1}, {1, 1, 1}};
-  t_x->write(&v_x, 0, sizeof(v_x));
+  t_x->write(&v_x, sizeof(v_x));
 
   auto t_y = backend->create_tensor(ng::element::f32, ng_shape_y);
-  t_y->write(&v_x, 0, sizeof(v_x));
+  t_y->write(&v_x, sizeof(v_x));
 
   // Allocate tensor for the result(s)
   vector<shared_ptr<ng::runtime::Tensor>> outputs;

test/opexecuter.cpp

@@ -376,7 +376,7 @@ void OpExecuter::ExecuteOnNGraph(vector<Tensor>& ngraph_outputs,
     std::shared_ptr<ngraph::runtime::Tensor> result;
     if (ng_backend_type != "CPU") {
       result = backend->create_tensor(ng_et, ng_shape);
-      result->write(src_ptr, 0, result->get_element_count() * ng_et.size());
+      result->write(src_ptr, result->get_element_count() * ng_et.size());
     } else {
       result = backend->create_tensor(ng_et, ng_shape, src_ptr);
     }
@@ -433,7 +433,7 @@ void OpExecuter::ExecuteOnNGraph(vector<Tensor>& ngraph_outputs,
     // Convert to tf tensor
     Tensor output_tensor(expected_output_datatypes_[i], tf_op_shapes[i]);
     void* dst_ptr = DMAHelper::base(&output_tensor);
-    ng_op_tensors[i]->read(dst_ptr, 0, output_tensor.TotalBytes());
+    ng_op_tensors[i]->read(dst_ptr, output_tensor.TotalBytes());
     ngraph_outputs.push_back(output_tensor);
     NGRAPH_VLOG(5) << " NGRAPH op " << i << ngraph_outputs[i].DebugString();
   }