Make it explicit in README that we build and locally install NVBench first,
and then build Python package use the library as a dependency.
The nvbench library is installed into Python layout alongside the native
extension.
Add comments stating the need to keep implementation and Python stub
file in sync to both files. In the stub file to comment documents
use of mypy's stubgen to generate stubs and calls to compare that against
current stubs. It also calls out the need to keep docstrings and
doctring examples in sync with implementation.
- Add license header to each example file.
- Fixed broken runs caused by type declarations.
- Fixed hang in throughput.py when --run-once by doing a
manual warm-up step, like in auto_throughput.py
Removed use of __all__ per PR feedback. Emit warnings.warn if
version information could not be retrieved from the package metadata,
e.g., package has been renamed by source code was not updated.
Introduce get_int64_or_default method, and counterparts for
float64 and string.
Provided names for Python arguments.
Tried generating Python stubs automatically with
```
stubgen -m cuda.nvbench._nvbench
```
Gave up on this, since it does not include doc-strings.
It would be nice to compare auto-generated _nvbench.pyi with
__init__.pyi for discrepancies though.
Edit wheel.packages metadata to include namespace package "cuda".
Updated README to remove the work-around of setting PYTHONPATH,
as it is no longer necessary.
Change example to illustrate timing CPU work.
First example does only CPU work (sleeps), use CPU-only timer.
Second examples does both CPU and GPU work (sleeps in either case).
Use cold-run timer with/without sync tag to measure both CPU and GPU times.
Benchmark function that sleeps for 1 seconda on the host using CPU-only
timer, as well as CPU/GPU timer that does/doesn't use blocking kernel.
All three methods must report consistent values close to 1 second.
python examples/cpu_only.py --run-once -d 0 --output foo.md
used to trip SystemError, returned a result with an exception set.
It now returns a clean NVBenchmarkError exception.
Change explicit constructor of benchmark_wrapper_t to use move-constructor
of py::object instead of copy constructor by replacing `py::object(o)` with
`py::object(std::move(o))`.
Add throughput.py example, which is based on the same kernel as
auto_throughput.py but records global memory reads/writes amounts
to output BWUtil metric measuring %SOL in bandwidth utilization.
state.add_summary(column_name: str, value: Union[int, float, str])
This is used in examples/axes.py to map integral value from Int64Axis
to string description.
make batch/sync arguments of State.exec keyword-only
Provide default column_name value for State.addElementCount method,
so that it can be called state.addElementCount(count), or as
state.addElementCount(count, column_name="Descriptive Name")
Fix run-time exception:
```
Fail: Unexpected error: RuntimeError: return_value_policy = copy, but type is non-copyable! (#define PYBIND11_DETAILED_ERROR_MESSAGES or compile in debug mode for details)
```
caused by attempt to returning move-only `nvbench::cuda_stream` class
instance using default `pybind11::return_value_policy::copy`.
