frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-27 17:54:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

[c10] Move profiler clock to libc10 for timestamps (#111972)

Browse Source 
Summary:
Move the profiler's Approximate Clock from libtorch to libc10. The main reason is to allow c10 features to get time.

The clock is using TSC when available for performance. CUDA Caching Allocator's implementation of memory snapshot will add the timestamps to memory events with this same clock in subsequent diff.

Test Plan: CI

Differential Revision: D50601935

Pulled By: aaronenyeshi

Pull Request resolved: https://github.com/pytorch/pytorch/pull/111972
Approved by: https://github.com/davidberard98
This commit is contained in:
Aaron Enye Shi
2023-10-27 16:18:40 +00:00
committed by PyTorch MergeBot
parent fdbb73fa4e
commit 63c089b09d
16 changed files with 283 additions and 242 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
test/cpp/jit/test_backend_compiler_lib.cpp
View File

@ -1,5 +1,6 @@
#include <ATen/Utils.h>
#include <c10/core/TensorImpl.h>
#include <c10/util/ApproximateClock.h>
#include <torch/csrc/jit/backends/backend.h>
#include <torch/csrc/jit/backends/backend_exception.h>
@ -112,14 +113,14 @@ class BackendWithCompiler : public PyTorchBackendInterface {
c10::List<at::Tensor> output_list;
#ifndef NO_PROFILING
auto start_us = torch::profiler::impl::getTime() / 1000;
auto start_us = c10::getTime() / 1000;
#endif
for (const auto& token : handle.toList()) {
IValue val = token;
auto instruction = val.toTupleRef().elements()[0].toStringRef();
auto debug_handle = val.toTupleRef().elements()[1].toInt();
#ifndef NO_PROFILING
auto start_time_us = torch::profiler::impl::getTime() / 1000;
auto start_time_us = c10::getTime() / 1000;
#endif
try {
if (instruction.rfind("prim::Constant", 0) == 0) {
@ -171,7 +172,7 @@ class BackendWithCompiler : public PyTorchBackendInterface {
TORCH_DELEGATED_BACKEND_THROW(false, e.what(), debug_handle);
}
#ifndef NO_PROFILING
auto end_time_us = torch::profiler::impl::getTime() / 1000;
auto end_time_us = c10::getTime() / 1000;
auto duration = end_time_us - start_time_us;
op_runtimes_us.emplace_back(duration, debug_handle, instruction);
#endif