OSSForks/tensorflow
1
0
Fork 0
mirror of https://github.com/zebrajr/tensorflow.git synced 2025-12-08 07:38:39 +01:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

Fix discrepancy between measured and analytical cost graph. Use tf_cuda_library for utils.

Browse Source 
PiperOrigin-RevId: 157660745
This commit is contained in:
Yao Zhang 2017-05-31 17:55:46 -07:00 committed by TensorFlower Gardener
parent 787381ca52
commit e5088cb823
4 changed files with 34 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tensorflow/core/grappler/clusters/BUILD
View File

@ -1,5 +1,7 @@
licenses(["notice"]) # Apache 2.0
load("//tensorflow:tensorflow.bzl", "tf_cuda_library")
filegroup(
name = "all_files",
srcs = glob(
@ -20,7 +22,7 @@ config_setting(
},
)
cc_library(
tf_cuda_library(
name = "utils",
srcs = ["utils.cc"],
hdrs = [

14
tensorflow/core/grappler/costs/BUILD
View File

@ -1,5 +1,7 @@
licenses(["notice"]) # Apache 2.0
load("//tensorflow:tensorflow.bzl", "tf_cuda_library")
filegroup(
name = "all_files",
srcs = glob(
@ -108,25 +110,21 @@ cc_test(
],
)
cc_library(
tf_cuda_library(
name = "utils",
srcs = ["utils.cc"],
hdrs = ["utils.h"],
defines = if_cuda(["GOOGLE_CUDA=1"]),
visibility = ["//visibility:public"],
deps = [
":op_performance_data_cc",
"//third_party/eigen3",
"//tensorflow/core/grappler/clusters:utils",
"//tensorflow/core:core_cpu_base",
"//tensorflow/core:framework",
"//tensorflow/core:lib",
"//tensorflow/core:lib_proto_parsing",
"//tensorflow/core:protos_all_cc",
] + if_cuda([
"//tensorflow/core:cuda",
"@local_config_cuda//cuda:cuda_headers",
]),
"//tensorflow/core/grappler/clusters:utils",
"//third_party/eigen3",
],
)
cc_library(

4
tensorflow/core/grappler/costs/utils.cc
View File

@ -167,6 +167,9 @@ std::vector<OpInfo::TensorProperties> FindInputFeatures(
inputs.push_back(UnknownInput());
} else {
const CostGraphDef::Node* input_cost = it->second;
if (input_cost->output_info_size() == 0) {
inputs.push_back(UnknownInput());
} else {
const CostGraphDef::Node::OutputInfo& output =
input_cost->output_info(output_index);
OpInfo::TensorProperties input;
@ -175,6 +178,7 @@ std::vector<OpInfo::TensorProperties> FindInputFeatures(
inputs.push_back(input);
}
}
}
return inputs;
}

23
tensorflow/python/grappler/cost_analyzer.cc
View File

@ -56,8 +56,8 @@ void CostAnalyzer::GatherCosts() {
CostGraphDef cost_graph_measured;
PredictCosts(&measure_estimator_, &cost_graph_measured,
&total_time_measured_);
VLOG(1) << "Graph size: " << item_->graph.node_size();
VLOG(1) << "cost_graph_measured size: " << cost_graph_measured.node_size();
op_perf_ = CostGraphToOpPerformanceData(cost_graph_measured, item_->graph);
CostGraphDef cost_graph_analytical;
PredictCosts(&analytical_estimator_, &cost_graph_analytical,
@ -66,25 +66,32 @@ void CostAnalyzer::GatherCosts() {
<< cost_graph_analytical.node_size();
CostGraphDef cost_graph_analytical_filtered;
std::set<string> cost_nodes;
for (auto& node : cost_graph_measured.node()) {
cost_nodes.insert(node.name());
CostGraphDef cost_graph_measured_filtered;
std::map<string, const CostGraphDef_Node*> measured_nodes;
for (const auto& node : cost_graph_measured.node()) {
measured_nodes[node.name()] = &node;
}
for (const auto& node : cost_graph_analytical.node()) {
auto it = cost_nodes.find(node.name());
auto it = measured_nodes.find(node.name());
// Filter the nodes that are not the cost nodes returned by
// MeasuringCostEstimator.
if (it == cost_nodes.end()) {
if (it == measured_nodes.end()) {
continue;
}
auto added_node = cost_graph_analytical_filtered.add_node();
*added_node = node;
auto added_node_analytical = cost_graph_analytical_filtered.add_node();
auto added_node_measured = cost_graph_measured_filtered.add_node();
*added_node_analytical = node;
*added_node_measured = *(it->second);
}
VLOG(1) << "cost_graph_analytical_filtered size: "
<< cost_graph_analytical_filtered.node_size();
// TODO(yaozhang): add a test to make sure that op_perf_analytical_ and
// op_perf_ cover the same set of nodes.
op_perf_analytical_ = CostGraphToOpPerformanceData(
cost_graph_analytical_filtered, item_->graph);
op_perf_ =
CostGraphToOpPerformanceData(cost_graph_measured_filtered, item_->graph);
}
void CostAnalyzer::PreprocessCosts() {