experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Replace NumSchedulableCPUs() with MaxParallelism()

Browse Source 
PiperOrigin-RevId: 255652572
This commit is contained in:
Martijn Vels 2019-06-28 12:57:27 -07:00 committed by TensorFlower Gardener
parent c1ad328572
commit 870fb5c489
5 changed files with 6 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tensorflow/compiler/aot/test.cc
View File

@ -59,7 +59,7 @@ void zero_buffers(XlaCompiledCpuFunction* computation) {
// Trivial test that runs the generated function to ensure it doesn't crash.
TEST(TEST_NAME, NoCrash) {
Eigen::ThreadPool pool(port::NumSchedulableCPUs());
Eigen::ThreadPool pool(port::MaxParallelism());
Eigen::ThreadPoolDevice device(&pool, pool.NumThreads());
CPP_CLASS computation;
@ -73,7 +73,7 @@ TEST(TEST_NAME, NoCrash) {
void BM_NAME(int iters) {
testing::StopTiming();
Eigen::ThreadPool pool(port::NumSchedulableCPUs());
Eigen::ThreadPool pool(port::MaxParallelism());
Eigen::ThreadPoolDevice device(&pool, pool.NumThreads());
CPP_CLASS computation;

2
tensorflow/compiler/xla/service/backend.cc
View File

@ -137,7 +137,7 @@ Backend::Backend(se::Platform* platform, Compiler* compiler,
if (platform->id() == se::host::kHostPlatformId) {
const int num_threads = intra_op_parallelism_threads > 0
? intra_op_parallelism_threads
: tensorflow::port::NumSchedulableCPUs();
: tensorflow::port::MaxParallelism();
intra_op_thread_pool_.reset(new IntraOpThreadPool(num_threads));
}
}

2
tensorflow/compiler/xla/service/cpu/parallel_task_assignment.cc
View File

@ -77,7 +77,7 @@ class DefaultCostModel : public ParallelCostModel {
// TODO(b/29630486) Develop system bandwidth model.
max_parallelism = std::min<int64>(
max_parallelism_,
std::ceil(std::sqrt(tensorflow::port::NumSchedulableCPUs())));
std::ceil(std::sqrt(tensorflow::port::MaxParallelism())));
// Use shape size instruction cost and L2 cache size min per-thread cost.
instruction_cost = shape_size_(instruction->shape());
min_cost_per_thread = 256LL << 10; // 256KB L2 Cache size.

2
tensorflow/compiler/xla/shape_util.h
View File

@ -751,7 +751,7 @@ class ShapeUtil {
// once with the proper empty indexes.
int64 n = -1;
std::vector<int64> indexes(base.begin(), base.end());
const int kNumThreads = tensorflow::port::NumSchedulableCPUs();
const int kNumThreads = tensorflow::port::MaxParallelism();
absl::optional<tensorflow::thread::ThreadPool> pool;
if (parallel) {
pool.emplace(tensorflow::Env::Default(), "foreach", kNumThreads);

3
tensorflow/compiler/xla/tools/replay_computation.cc
View File

@ -86,8 +86,7 @@ namespace {
// Command-line opts to this tool. See main() for descriptions of these
// fields.
struct Options {
Options()
: intra_op_thread_pool_size(tensorflow::port::NumSchedulableCPUs()) {}
Options() : intra_op_thread_pool_size(tensorflow::port::MaxParallelism()) {}
bool NeedsRealData() const { return !use_fake_data && !compile_only; }