mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 00:21:07 +01:00
b21a6ff639
Syncing nvfuser devel branch to upstream master. https://github.com/csarofeen/pytorch/ Code changes includes: - codegen improvements: 1. double support in expression evaluator - bug fixes: 1. dropout fix - rework RNG to support broadcasted dropout (Fixes #82784) 2. expand fix - Patch expand+reduction, expand+view, rework view analysis and guard - scheduler: 1. manual transpose schedule example 2. WIP transpose scheduler Commits that's in this PR from the devel branch: ``` b7435afcd22c917713c2f41a7237bc26e1183f14 Transpose scheduler, step 1 (#1854) 8a45dbf72034684eb8e18b1835b533e90b68f184 Add an example on how to manually schedule transpose (#1889) 83dbf56a9554b2efbd5416461d938fff477b0b27 Patch dropout fix (#1898) 69d3519a532250719b1aa8341b50e067b181b42d Expand+Reduction, Expand+View support, rework View analysis and guards (#1883) 15091c488e96343bdc49e3990acbf238a3b3da51 Rework RNG to correctly support broadcasted dropout (#1888) aafe2d048aaac596e503596a41303423619f3954 Make ExpressionEvaluator support Double (#1885) ``` RUN_TORCHBENCH: nvfuser Differential Revision: [D38657074](https://our.internmc.facebook.com/intern/diff/D38657074) Pull Request resolved: https://github.com/pytorch/pytorch/pull/83239 Approved by: https://github.com/davidberard98
77 lines
1.8 KiB
C++
77 lines
1.8 KiB
C++
#include <torch/csrc/jit/codegen/cuda/instrumentation.h>
|
|
|
|
#include <c10/macros/Export.h>
|
|
|
|
#ifdef _WIN32
|
|
#include <c10/util/win32-headers.h>
|
|
#else
|
|
#include <pthread.h>
|
|
#include <unistd.h>
|
|
#endif
|
|
|
|
namespace torch {
|
|
namespace jit {
|
|
namespace fuser {
|
|
namespace cuda {
|
|
namespace inst {
|
|
|
|
Trace::Trace() {
|
|
const char* trace_filename = getenv("PYTORCH_NVFUSER_TRACE");
|
|
if (trace_filename != nullptr) {
|
|
log_file_ = fopen(trace_filename, "w");
|
|
TORCH_CHECK(log_file_ != nullptr, "Can't open trace file");
|
|
|
|
// Disable the file stream buffering, since it may result
|
|
// in torn writes in multi-threaded tracing
|
|
setbuf(log_file_, nullptr);
|
|
|
|
// Print the trace prologue
|
|
// (including a dummy TRACE_START event)
|
|
fprintf(log_file_, "{\n\"traceEvents\": [\n");
|
|
start_timestamp_ = Clock::now();
|
|
logEvent('I', "TRACE_START");
|
|
}
|
|
|
|
if (isOptionDisabled(DisableOption::Nvtx)) {
|
|
record_nvtx_range_ = false;
|
|
}
|
|
}
|
|
|
|
Trace::~Trace() {
|
|
if (log_file_ != nullptr) {
|
|
// Print trace epilogue
|
|
logEvent('I', "TRACE_END", ' ');
|
|
fprintf(log_file_, "],\n\"displayTimeUnit\": \"ms\"\n}\n");
|
|
fclose(log_file_);
|
|
}
|
|
}
|
|
|
|
void Trace::logEvent(char ph, const char* name, char sep) {
|
|
const std::chrono::duration<double> d = Clock::now() - start_timestamp_;
|
|
const double elapsed = d.count() * 1e6;
|
|
|
|
#ifdef _WIN32
|
|
const unsigned int pid = GetCurrentProcessId();
|
|
const unsigned int tid = GetCurrentThreadId();
|
|
#else
|
|
const unsigned int pid = getpid();
|
|
const unsigned int tid = std::hash<pthread_t>{}(pthread_self());
|
|
#endif // _WIN32
|
|
|
|
fprintf(
|
|
log_file_,
|
|
"{ \"name\": \"%s\", \"ph\": \"%c\", \"pid\": %u, \"tid\": %u, \"ts\": %.0f }%c\n",
|
|
name,
|
|
ph,
|
|
pid,
|
|
tid,
|
|
elapsed,
|
|
sep);
|
|
}
|
|
|
|
} // namespace inst
|
|
} // namespace cuda
|
|
} // namespace fuser
|
|
} // namespace jit
|
|
} // namespace torch
|