Improve exception safety of timer structs by using local scope guards to ensure that cleanup steps, such as signaling blocking kernel to unblock and making sure that the stream is synchronized are performed even launch object throws an exception.
Tests of exception safety were added.
--
* blocking_kernel.unblock_noexcept() noexcept method added
This decouples the logic of signaling to unblock from checking
of the timeout.
* Improve exception safely in kernel_launch_timer
Introduce noexcept cleanup methods. Place body of start()
and stop() methods in the try/catch block and execute
noexcept clean-up on exception before rethrowing.
* Improve exception safety of measure_hot
* Make sure that throwing methods call noexcept ones instead of duplicating functionality
* Use cleanup_guard in measure_cold_base::kernel_launch_timer
Replace try/catch pattern with cleaner use of cleanup_guard
class.
* cpu_timer::start, cpu_timer::stop methods marked noexcept
These methods do not throw, and marking them noexcept explicitly
makes it fine to call them from other noexcept methods, as such
cleanup_noexcept in measure_cold.
* Address remaining exception safety issue in measure_hot
* Renamed guard variables to reflect their purpose, apply arm-then-do to ops queueing kernels
Set m_block_stream_armed = true; before launching the kernel. Doing so signals
cleanup guard that stream must be unblocked, even if launching of the kernel failed.
Same for operation launching time-stamps kernel.
* Add testing/device/exception_safety.cu
This test add benchmark that throws. It verifies that it did not
time-out and control counters the benchmark maintains are at
the expected values.
* Refactor measurement cleanup guards for testability
Extract hot stream cleanup and cold launch timer cleanup into reusable
detail helpers. Keep measure_hot and measure_cold using those helpers through
thin adapters so the tested cleanup logic matches the production path.
Add driver-free cleanup guard tests using a fake measure object to verify
cleanup ordering when exceptions occur after blocking stream setup, after hot
unblock, and around cold GPU frequency start/stop paths.
* Implement cpu_timer_stop_noexcept in terms of cpu_timer_stop
The cpu_timer_stop is already noexcept by nature of implementation,
but we maintain cpu_timer_stop_noexcept method for symmetry with
other pairs sync_stream()/sync_stream_noexcept().
The cpu_timer_stop_noexcept() is implemented via cpu_timer_stop().
These methods are annotated __forceinline__, so the same code should be
generated.
* More readable initialization of bool members
* Moved exception_safety.cu back to testing/ folder
testing/device is reserved for tests that require locking
of GPU frequency per CMake option description.
* Fixed nitpick and bug it discovered
Changed testing/exception_safety.cu:237 so run_benchmark now iterates over every state
from bench.get_states() and checks each one is skipped with a reason
containing "requested".
That exposed a real runner behavior gap, so I also made a minimal fix in
nvbench/runner.cuh:120: after stop_runner_loop, remaining states are now explicitly
marked skipped with a reason instead of only printing a skip notification.
* Move static assertions (pertaining to cleanup guards) to
testing/cleanup_guards.cu
The CI failure with CTK 12.0 and certain version of GCC is caused
by OOM in cudafe++ process tripped by compiling instantiation
of contract verification on cold_launch_timer_probe struct.
As a work-around, this instantiation is excluded for CTK 12.0-12.6
* Add decorators for registering benchmarks and adding axis
cuda.bench.register(fn) continues returning Benchmark, and supports
legacy use.
New signature added:
cuda.bench.register():
Returns a decorator
```
@bench.register()
@bench.axis.float64("Duration (s)", [7e-5, 1e-4, 5e-4])
@bench.option.min_samples(120)
def single_float64_axis(state: bench.State):
...
```
* Remove example/auto_throughput.py
The C++ counterpart's purpose is to demonstrate use of CUPTI
metrics, but these are not supported in Python bindings, so
this example is a duplicate of example/throughput.py
* Add wrong decorator order test for bench.axis.*
* Strengthen type annotation for register function
Acting on code rabbit nit-pick require that function being
registered take cuda.bench.State object as an argument.
Verified the fix as
```
(py313) :~/repos/nvbench/python$ python -m mypy --ignore-missing-import /tmp/t.py
/tmp/t.py:8: error: Argument 1 has incompatible type "Callable[[], None]"; expected "Callable[[State], None]" [arg-type]
Found 1 error in 1 file (checked 1 source file)
(py313) :~/repos/nvbench/python$ nl -ba /tmp/t.py
1 # /tmp/check_nvbench_register.py
2 import cuda.bench as bench
3
4 @bench.register()
5 def good(state: bench.State) -> None:
6 pass
7
8 @bench.register()
9 def bad() -> None:
10 pass
```
* Replace use of global variable with thread-safe lru_cache
This improves thread-safety of module initialization.
* Abide by RUF005 linting rule
* Expand docstrings regarding cuda.bench.register() decorator
It explains to the user what the decorator does and provides
a concise usage example.
* Sharpen wording on exception maybe-thrown by decorator
Implements `cuda.bench.results.BenchmarkResult` class to represent data from JSON output of benchmark execution.
The contains implements two class methods `BenchmarkResult.from_json(filename : str | os.PathLike, *, metadata : Any = None)` which expects well-formed JSON filename and `BenchmarkResult.empty(*, metadata : Any = None)` intended to represent failed result with reasons that can be recorded in metadata at user's discretion.
The `BenchmarkResult` implements mapping interface, supporting `.keys()`, `.values()`, `.items()` methods, `__len__`, `__contains__`, `__getitem__` and `__iter__` special methods.
Values in `BenchmarkResult` has type `cuda.bench.results.SubBenchmarkResult` which implements a list-like interface, i.e. implements `__len__`, `__getitem__`, and `__iter__` special methods. Values in this list-like structure correspond to measurements of individual states of a particular benchmark (the key in `BenchmarkResult`).
Elements of `SubBenchmarkResult` structure have type `SubBenchmarkState` that supports mapping protocol with axis_values as a key and represent data corresponding to measurements for a particular state (combination of settings for each axis).
The state provides `.samples` and `.frequencies` attributes storing raw execution duration values and estimates for average GPU frequencies.
Example usage:
```
import array, numpy as np, cuda.bench.results
r = cuda.bench.results.BenchmarkResult("perf_data/axes_run1.json")
r["copy_sweep_grid_shape"].centers_with_frequencies(
lambda t, f: np.median(np.asarray(t)*np.asarray(f)))
```
```
In [1]: import array, numpy as np, cuda.bench.results
In [2]: r = cuda.bench.results.BenchmarkResult("temp_data/axes_run1.json")
In [3]: list(r)
Out[3]:
['simple',
'single_float64_axis',
'copy_sweep_grid_shape',
'copy_type_sweep',
'copy_type_conversion_sweep',
'copy_type_and_block_size_sweep']
In [4]: r["simple"].centers(lambda t: np.percentile(t, [25,75]))
Out[4]: {'Device=0': array([0.00100966, 0.00101299])}
In [5]: r.centers(lambda t: np.percentile(t, [25,75]))["simple"]
Out[5]: {'Device=0': array([0.00100966, 0.00101299])}
In [6]: len(r)
Out[6]: 6
In [7]: "fake" in r
Out[7]: False
```
Each `SubBenchmarkState` implements
`.summaries` attribute - rich object that retains tag/name/hint/hide/description metadata.
* Add nvbench-json-summary to render NVBench JSON output as an NVBench-style
markdown summary table, including axis formatting, device sections, hidden
summary filtering, and summary hint formatting.
Update packaging, type stubs, and tests for the new namespace, renamed
classes, Python 3.10-compatible annotations, and summary-table generation.
* Split tests in test_benchmark_result into smaller tests
* Fix break due to file name change
* Add python/examples/benchmark_result_autotune.py
This example demonstrates using cuda.bench and cuda.bench.results
to implement simple auto-tuning, demonstrated on selecting of
tile shape hyperparameter for naive stencil kernel implemented
in numba-cuda.
* Resolve ruff PLE0604
* Fix for format_axis_value in json format script to handle None value
Add tests to cover such input.
* Address code rabbit review feedback
* Fix license header, add validation
* Addressed both issues raised in review
Malformed values are now represented in result as None.
Skipped benchmarks are no longer dropped, i.e., they are present
in BenchmarkResult data, but they are not reflected in summary
table in line with what NVBench-instrumented benchmarks do.
Previously, it corresponded to main.cu.o object file. Now it corresponds to
static library libnvbench_main.a which is archive file with main.cu.o object
in it.
This closes#349
* Implement warmup-runs count, supported as CLI
CLI option --warmup-runs implemented and documented.
The warm-up counts is enforced to always be positive.
This is necessary to ensure that JIT-ting has occurred,
and use of blocking kernel would not result in time-outs.
Test is option parser is added.
* Ensure that measure_cold::run_warmup instantiates blocking kernel
Because warm-up runs are executed without use of blocking kernel,
the blocking kernel was not jitted until actual measurements were
collected. The module loading cost incurred during the first run
shows as elevated CPU time noise value for the first measurement
as noted in https://github.com/NVIDIA/nvbench/pull/339
This PR adds `this->block_stream(); this->unblock_stream();` prior
to executing warm-up loop with use of blocking kernel disabled.
This ensures that blocking kernel is instantiated during the warm-up,
but it no other kernel is launched between its launch and stream sync
thus avoiding deadlocking.
* Rename --warmup-runs to --cold-warmup-runs, add --cold-max-warmup-walltime
Since configurable number of warmups only applies to measure_cold.cuh
rename the CLI option to reflect that.
Also add --cold-max-warmup-walltime (defaults to -1, i.e. disabled).
If enabled, exits warmup loop before request count is reached if
the wall-time expanded executign warmups exceeds this max-warmup-walltime
value.
* Correct Python API signature of State.get_axis_values_as_strings
The C++ API has default boolean argument color, but Python API
declared no arguments.
Closes#345
* Also exercise invocation of get_axis_values_as_string with keyword argument value
* Remove use of cuda.core.experimental
Fixed relative text alignment in docstrings to fix autodoc warnigns
Renamed cuda.bench.test_cpp_exception and cuda.bench.test_py_exception functions
to start with underscore, signaling that these functions are internal and should
not be documented
Account for test_cpp_exceptions -> _test_cpp_exception, same for *_py_*
Make sure to reset __module__ of reexported symbols to be cuda.bench
* Introduce function colorize to modularize colorization/no-color handling
* Use sns.set_theme instead of deprecated sns.set()
* Use str.format instead of legacy % syntax
* Simplified iteration over list
Use f-string (supported since Python 3.6) instead of str.format for
better readability and performance
Fix GCC16 sfinae incomplete warnings.
GCC16 started requiring that the type `T` used in `std::reference_wrapper<T>` is complete where using `-std=c++17`. Since NVBench has to forward declare some types in header files to break circular dependency, use of incomplete type breaks build due to use of `-Werror` flag due to `-Wsfinae-incomplete` warning emitted by GCC16.
This commit replaced affected uses of `std::reference_wrapper<const nvbench::benchmark_base>` in state.cxx, and `std::reference_wrapper<nvbench::printer_base>` in benchmark_base.cxx with raw pointers.
* Mark NVBench headers as SYSTEM for consuming targets.
Fixes#30.
* As nvbench.main links to nvbench as INTERFACE only, it no longer consumes usage reqs of nvbench
Because of this nvbench.main was no longer consuming dependence on CUDA::toolkit include dirs.
This PR links nvbench.main to ${ctk_libraries} privately to reestablish that dependency
* Implement use of pragma system_header in NVBench
1. Add code to nvbench/config.cuh.in to define NVBENCH_IMPLICIT_SYSTEM_HEADER_*
preprocessor variable dependending on compiler, unless NVBENCH_NO_IMPLICIT_SYSTEM_HEADER
was defined.
2. Build NVBench targets with -DNVBENCH_NO_IMPLICIT_SYSTEM_HEADER
3. Modify each header file in nvbench/ folder to
- include <nvbench/config.cuh>
- Execute pragma <OPTIONAL_CMPLR> system_header guarded
by checks for defined preprocessor variables
- Do the above two steps before any other headers are included
---------
Co-authored-by: Allison Piper <apiper@nvidia.com>
* 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
measure_cold class now directly inherits m_check_throttling from state.
This ensures that when `--jsonbin` is specified frequency data corresponding
to timing data are available to write out.
This places all std::vector members together. Added default initialization
to all std::vector members, and all other members with default constructors.
Exceptions are references and nvbench::launch m_launch; member
