* Reworked cupti_profiler to use Host + Range Profiler APIs end-to-end
NVPW_* API has been deprecated since CTK 13.0. Followed advice in compliation
message to replace NVPW_* API with CUPTI Profiler Host API.
`libnvbench.so` no longer links to `nvperf_host` directly, only to `libcupti`.
NVBench uses the **CUPTI Host API** to build a config image from metric names,
and the **Range Profiler API** to collect and decode counters. The host API never
collects data directly; it prepares and evaluates data produced by range profiling.
Introduce `host_impl`/`profiler_init_guard` to manage CUPTI Host object and
initialization/deinitialization, including safe move-assignment cleanup.
`profiler_init_guard` initializes profiler, and throws if CUPTI returns
an error code. `profiler_init_guard::finalize_profiler()` de-inits profiler
and returns the error code. Destructor calls finalize_profiler, but ignores
the status code. If user wants to explicitly de-initialize profiler and handle
the error, he/she is advised to call `finalize_profiler()` directly. The guard
has a boolean member variable to allow destructor to work even if user explicitly
called finalize_profiler() method.
The old counter-data prefix/scratch flow was replaced with the Range Profiler counter
data image sizing/initialization path and decode flow.
Host API metric filtering (base metrics + context scope) and Host-side evaluation to
GPU values via cuptiProfilerHostEvaluateToGpuValues is implemented.
- **Host object**: `host_impl::object` in `nvbench/cupti_profiler.cxx`.
- **Range profiler object**: `host_impl::range_profiler_object`.
- **Config image**: `m_config_image`.
- **Counter data image**: `m_data_image`.
1) **Host init + config image**
- `initialize_profiler_host()` creates the host object.
- `initialize_config_image_host()` adds metrics and builds the config image.
2) **Range profiler enable + counter data image**
- `enable_range_profiler()` creates the range profiler object.
- `initialize_counter_data_image()` sizes and initializes the data image using
the range profiler object, matching the CUPTI samples.
3) **Config + collect + decode**
- `set_range_profiler_config()` binds the config image + data image.
- `start_user_loop()` / `stop_user_loop()` push/pop the user range and
start/stop the range profiler.
- `process_user_loop()` decodes counter data via
`cuptiRangeProfilerDecodeData()`.
4) **Evaluate metrics**
- `get_counter_values()` calls `cuptiProfilerHostEvaluateToGpuValues()` to
convert counter data into metric values.
The
* Use class instead of struct in profiler_init_guard; forward declaration
* Add SFINAE guards before accessing members not present in earlier CTK versions
* Check if cupti_profiler_host.h exists, use old/new implementation based on that check
1. Reintroduced legacy `cupti_profiler_nvpw.cuh` and `cupti_profiler_nvpw.cuh`.
2. Moved profiler-host-API implementation to `cupti_profiler_host.cuh`, `cupti_profiler_host.cxx`.
3. Add `nvbench/cupti_profiler.cuh` which checks if `cupti_profiler_host.h` header is known and
includes `cupti_profiler_host.cuh` or `cupti_profiler_nvpw.cuh` respectively.
4. In cmake, we check if ${nvbench_cupti_root}/include/cupti_profiler_host.h file exists.
If it does not, `libnvbench.so` would have dependency on libnvperf_host and libnvperf_target
in addition to dependency on libcupti. If the header exists, it would only depend on libcupti
* Build native arch by default, update rapids-cmake.
* Add check that CXX and CUDA_HOST compiler match.
Similar to CCCL, we need these to match to ensure that our warning flag detection functions properly.
* GCC only recognizes `unused-local-typedefs`.
Clang recognizes both. Ensure that we set this for both compilers.
Switched away from the rapids-cmake provided version and manually CPM'd it.
rapids-cmake will stop providing fmtlib later this year, and the version currently supported is rather old.
Included the same logic that rapids-cmake currently uses to hopefully provide a smooth transition for edge cases (external fmt, etc).
Added `FMT_SYSTEM_HEADERS=ON` to mark fmt headers as system includes, suppressing any internal warnings.
To do this we need to ensure that the nvml init handler is
both contained in the library/executable that uses nvbench.
The original implementation fails since the singleton can be dropped
since it has no usages. So instead we move to a function static
which we ensure will always be used.
* write_git_revision_file must be used in same CMakeLists as consumer
So we can't have this in the rapids-cmake init function.
* Fix whitespace damage
---------
Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>
The formatting of `{}` can be incorrect under 7.X when given
doubles and compiled with the latest conda toolchain. While both
fmt 8 and 9 don't show this issue move to the latest version
to leverage all the improvements in fmt 9.
Fixes#103
This will provide functionality such as clock locking (--lgm),
persistance mode (--pm), device querying (--list), version checking
(--version), and documentation (--help).
This is possible already with any nvbench executable, but having
one with a reliable name will be helpful for scripting and writing
documentation.
- /W4 on MSVC
- -Wall -Wextra + others on gcc/clang
- New NVBench_ENABLE_WERROR option to toggle "warnings as errors"
- Mark the nlohmann_json library as IMPORTED to switch to system includes
- Rename nvbench_main -> nvbench.main to follow target name conventions
- Explicitly suppress some cudafe warnings when compiling templates in
nlohmann_json headers.
- Explicitly suppress some warnings from Thrust headers.
- Various fixes for warnings exposed by new flags.
- Disable CUPTI on CTK < 11.3 (See #52).
Locking clocks is currently only implemented for Volta+ devices.
Example usage:
my_bench -d [0,1,3] --persistence-mode 1 --lock-gpu-clocks base
See the cli_help.md docs for more info.
- Add export sets
- Add install rules
- Remove manual CPM import, port to rapids_cpm_*, etc
- Organize CMake code into cmake/*.cmake files.
- NVBench is now a shared library.
