Icemist commented on code in PR #11465:
URL: https://github.com/apache/tvm/pull/11465#discussion_r900011107
##########
src/runtime/profiling.cc:
##########
@@ -822,25 +823,37 @@ PackedFunc WrapTimeEvaluator(PackedFunc pf, Device dev,
int number, int repeat,
f_preproc.CallPacked(args, &temp);
}
double duration_ms = 0.0;
-
+ std::vector<double> results;
+ int64_t final_number = number;
do {
+ results.clear();
if (duration_ms > 0.0) {
- number = static_cast<int>(std::max((min_repeat_ms / (duration_ms /
number) + 1),
- number * 1.618)); // 1.618 is
chosen by random
+ final_number =
+ static_cast<int>(std::max((min_repeat_ms / (duration_ms /
final_number) + 1),
+ final_number * 1.618)); // 1.618 is
chosen by random
}
+ results.resize(final_number);
- Timer t = Timer::Start(dev);
- // start timing
- for (int i = 0; i < number; ++i) {
+ for (int i = 0; i < final_number; ++i) {
+ // start timing
+ Timer t = Timer::Start(dev);
Review Comment:
Reverted
##########
src/runtime/profiling.cc:
##########
@@ -822,25 +823,37 @@ PackedFunc WrapTimeEvaluator(PackedFunc pf, Device dev,
int number, int repeat,
f_preproc.CallPacked(args, &temp);
}
double duration_ms = 0.0;
-
+ std::vector<double> results;
+ int64_t final_number = number;
do {
+ results.clear();
if (duration_ms > 0.0) {
- number = static_cast<int>(std::max((min_repeat_ms / (duration_ms /
number) + 1),
- number * 1.618)); // 1.618 is
chosen by random
+ final_number =
+ static_cast<int>(std::max((min_repeat_ms / (duration_ms /
final_number) + 1),
+ final_number * 1.618)); // 1.618 is
chosen by random
}
+ results.resize(final_number);
- Timer t = Timer::Start(dev);
- // start timing
- for (int i = 0; i < number; ++i) {
+ for (int i = 0; i < final_number; ++i) {
+ // start timing
+ Timer t = Timer::Start(dev);
pf.CallPacked(args, &temp);
+ t->Stop();
+ int64_t t_nanos = t->SyncAndGetElapsedNanos();
+ results[i] = t_nanos / 1e6;
}
- t->Stop();
- int64_t t_nanos = t->SyncAndGetElapsedNanos();
- duration_ms = t_nanos / 1e6;
+ duration_ms = std::accumulate(results.begin(), results.end(), 0.0);
} while (duration_ms < min_repeat_ms);
- double speed = duration_ms / 1e3 / number;
- os.write(reinterpret_cast<char*>(&speed), sizeof(speed));
+ // write the size value.
+ os.write(reinterpret_cast<char*>(&final_number), sizeof(int64_t));
+ // write the array
+ for (int i = 0; i < final_number; ++i) {
+ results[i] /= 1e3;
+ os.write(reinterpret_cast<char*>(&results[i]), sizeof(double));
+ }
+
+
std::this_thread::sleep_for(std::chrono::milliseconds(cooldown_interval_ms));
Review Comment:
Done
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
