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

Rename (Add|Get)ProfileResult to something more specific; NFC

Browse Source 
PiperOrigin-RevId: 174084570
This commit is contained in:
Sanjoy Das 2017-10-31 13:41:37 -07:00 committed by TensorFlower Gardener
parent f1916f8f6c
commit 0a7be5a2f5
6 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -234,7 +234,7 @@ Status CpuExecutable::ExecuteComputeFunction(
for (auto hlo_prof_idx : hlo_to_profile_idx_) {
const HloInstruction* hlo = hlo_prof_idx.first;
uint64 cycles_taken = profile_counters[hlo_prof_idx.second];
hlo_execution_profile->AddProfileResult(hlo, cycles_taken);
hlo_execution_profile->SetCyclesTakenBy(hlo, cycles_taken);
}
}
return Status::OK();

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

@ -463,7 +463,7 @@ Status ParallelCpuExecutable::ExecuteComputeFunctions(
for (auto hlo_prof_idx : hlo_to_profile_idx_) {
const HloInstruction* hlo = hlo_prof_idx.first;
uint64 cycles_taken = profile_counters[hlo_prof_idx.second];
hlo_execution_profile->AddProfileResult(hlo, cycles_taken);
hlo_execution_profile->SetCyclesTakenBy(hlo, cycles_taken);
}
}

2
tensorflow/compiler/xla/service/gpu/gpu_executable.cc
View File

@ -88,7 +88,7 @@ class HloExecutionProfiler {
if (do_profile_) {
stream_->ThenStopTimer(per_op_timer_.get());
stream_->BlockHostUntilDone();
profile_->AddProfileResult(
profile_->SetCyclesTakenBy(
hlo_instruction, per_op_timer_->Nanoseconds() * clock_rate_ghz_);
}
}

4
tensorflow/compiler/xla/service/hlo_execution_profile.cc
View File

@ -27,13 +27,13 @@ limitations under the License.
namespace xla {
void HloExecutionProfile::AddProfileResult(const HloInstruction* hlo,
void HloExecutionProfile::SetCyclesTakenBy(const HloInstruction* hlo,
uint64 cycles_taken) {
hlo_to_cycles_taken_[hlo] = cycles_taken;
profiled_computations_.insert(hlo->parent());
}
uint64 HloExecutionProfile::GetProfileResult(const HloInstruction& hlo) const {
uint64 HloExecutionProfile::GetCyclesTakenBy(const HloInstruction& hlo) const {
auto iter = hlo_to_cycles_taken_.find(&hlo);
if (iter == hlo_to_cycles_taken_.end()) {
return 0;

4
tensorflow/compiler/xla/service/hlo_execution_profile.h
View File

@ -36,11 +36,11 @@ class HloExecutionProfile {
using DeviceDescription = perftools::gputools::DeviceDescription;
// Record how many cycles this HLO took to execute.
void AddProfileResult(const HloInstruction* hlo, uint64 cycles_taken);
void SetCyclesTakenBy(const HloInstruction* hlo, uint64 cycles_taken);
// Returns how many cycles this HLO took to execute. Profiling information
// may not be available for some instructions in which case zero is returned.
uint64 GetProfileResult(const HloInstruction& hlo) const;
uint64 GetCyclesTakenBy(const HloInstruction& hlo) const;
// Return the number of cycles this computation took to execute.
uint64 total_cycles_executed(const HloComputation& computation) const {

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

@ -1070,7 +1070,7 @@ string HloDotDumper::GetInstructionNodeExtraInfo(const HloInstruction* instr) {
lines.push_back(Printf("[%p]", instr));
}
if (profile_ != nullptr) {
double hlo_cycles_executed = profile_->GetProfileResult(*instr);
double hlo_cycles_executed = profile_->GetCyclesTakenBy(*instr);
double total_cycles_executed =
profile_->total_cycles_executed(*instr->parent());
if (hlo_cycles_executed > 0 && total_cycles_executed > 0) {