Skip to content

Fix HAS_CUPTI_RANGE_PROFILER check #1131

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 5 commits into
base: main
Choose a base branch
from
Open

Fix HAS_CUPTI_RANGE_PROFILER check #1131

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
0d1d426
Fix HAS_CUPTI_RANGE_PROFILER check
cyyever Aug 25, 2025
fff0715
Always print NVPERF
cyyever Aug 25, 2025
be6457b
Simplify CUDA paths
cyyever Aug 25, 2025
23710ac
More fixes
cyyever Aug 30, 2025
7c767d4
Merge branch 'main' into windows_build
cyyever Sep 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions libkineto/src/CuptiRangeProfilerApi.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,6 @@
* LICENSE file in the root directory of this source tree.
*/

#include <cstdio>
#include <cstdlib>
#include "ILoggerObserver.h"
#ifdef HAS_CUPTI
Expand All @@ -24,6 +23,7 @@

// TODO(T90238193)
// @lint-ignore-every CLANGTIDY facebook-hte-RelativeInclude
#include "CuptiCallbackApi.h"
#include "CuptiRangeProfilerApi.h"

#define STRINGIFY(x) #x
Expand Down Expand Up @@ -218,7 +218,7 @@ void __trackCudaKernelLaunch(
profiler->numCallbacks_++;
}

bool enableKernelCallbacks() {
static bool enableKernelCallbacks() {
auto cbapi = CuptiCallbackApi::singleton();

bool status = cbapi->enableCallback(
Expand Down
11 changes: 4 additions & 7 deletions libkineto/src/CuptiRangeProfilerApi.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,14 +13,13 @@
#include <cuda_runtime_api.h>
// Using CUDA 11 and above due to usage of API:
// cuptiProfilerGetCounterAvailability.
// Starting from CUDA 12.06 the Profiler API is superseded by Range Profiler API
// Starting from CUDA 12.6 the Profiler API is superseded by Range Profiler API
// This needs significant rework. See
// https://docs.nvidia.com/cupti/main/main.html#evolution-of-the-profiling-apis
#if defined(USE_CUPTI_RANGE_PROFILER) && defined(CUDART_VERSION) && \
CUDART_VERSION >= 10000 && CUDA_VERSION >= 11000 && CUDA_VERSION <= 12060
CUDART_VERSION >= 10000 && CUDA_VERSION >= 12060
Copy link
Contributor

@sraikund16 sraikund16 Aug 28, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not sure this is correct. I recall the range profiler breaking on later versions

Copy link
Contributor Author

@cyyever cyyever Aug 28, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@sraikund16 Can reproduce? Nevertheless, the guard is wrong according to the above comment. If the profile still breaks, we should fix it rather than incorrectly disabling it.

Copy link
Contributor

@sraikund16 sraikund16 Aug 29, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@cyyever based on the NV link provided, it looks like we can use the current APIs until 13.0 although it is not recommended. However, by 13.0 we will need to use the new APIs as the old ones are deprecated. Either way it would require a rework.

Copy link
Contributor Author

@cyyever cyyever Aug 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If the current APIs exist during the whole CUDA 13 cycle, then we have lots of time for refactoring.

#define HAS_CUPTI_RANGE_PROFILER 1
#endif // CUDART_VERSION > 10.00 and CUDA_VERSION >= 11.00 and CUDA_VERSION
// <= 12.06
#endif // CUDART_VERSION > 10.00 and CUDA_VERSION >= 12.6
#endif // HAS_CUPTI

#if HAS_CUPTI_RANGE_PROFILER
Expand All @@ -40,14 +39,13 @@ enum CUpti_ProfilerReplayMode {
#endif // HAS_CUPTI_RANGE_PROFILER

#include <chrono>
#include <mutex>
#include <set>
#include <string>
#include <vector>

// TODO(T90238193)
// @lint-ignore-every CLANGTIDY facebook-hte-RelativeInclude
#include "CuptiCallbackApi.h"
// #include "CuptiCallbackApi.h"
#include "CuptiNvPerfMetric.h"
#include "TraceSpan.h"

Expand Down Expand Up @@ -214,7 +212,6 @@ class CuptiRBProfilerSession {
struct ICuptiRBProfilerSessionFactory {
virtual std::unique_ptr<CuptiRBProfilerSession> make(
const CuptiRangeProfilerOptions& opts) = 0;
virtual ~ICuptiRBProfilerSessionFactory() {}
};

struct CuptiRBProfilerSessionFactory : ICuptiRBProfilerSessionFactory {
Expand Down