The current main would fail on `math.scalbn` because we implicitly cast
the first argument from `int32` to `float32`, while the function only
accepts `int32` as the first argument and `float32` as the second
argument.
So we update the type matching logic as follows:
1. Check if there's a type tuple that matches the types of the input
arguments
2. If yes, we don't allow arithmetic check.
3. If not, we will do arithmetic check to implicitly cast types among
arguments.
4. If we still don't find a corresponding function that accepts the
casted types, throwing an error.
---------
Co-authored-by: Philippe Tillet <phil@openai.com>
Related to #1271 . I am currently working on adding support for
Pre-volta GPUs in Triton.
---------
Co-authored-by: Himanshu Pathak <himanshu@mtatva.com>
Co-authored-by: Philippe Tillet <phil@openai.com>
We have had complaints/issues randomly where a zombie python process is
holding this lock. We don't need it since renames are atomic on posix.
So refactor this to make temp files unique and then use replace
(https://docs.python.org/3/library/os.html#os.replace )
Fixes#1545
`build_temp` is a temporary directory which `distutils` used to keep in
the `./build` directory, but when `pyproject.toml` is present `pip` now
puts it in `/tmp` and removes it at the end of the build.
Instead, this creates a new permanent directory like
`python/build/cmake.linux_x86_64-cpython-3.8` (the old name but with
cmake instead of temp).
While I was looking at the verbose pip output, I also noticed a bunch of
warnings like
```
Python recognizes 'triton/runtime.backends' as an importable package,
but it is not listed in the `packages` configuration of setuptools.
'triton/runtime.backends' has been automatically added to the distribution only
because it may contain data files, but this behavior is likely to change
in future versions of setuptools (and therefore is considered deprecated).
```
So I've also added these to the packages list.
---------
Co-authored-by: Keren Zhou <kerenzhou@openai.com>
Change the usage of LRU cache decorator from @functools.lru_cache to
@functools.lru_cache().
The former raises an error TypeError('Expected maxsize to be an integer
or None' for Python 3.7 or older.
This way reduces build time with assertions enabled LLVM and
dramatically speeds up triton's build with a "debug" LLVM.
Co-authored-by: Philippe Tillet <phil@openai.com>
`tl.reduction` is currently tested indirectly through the existing
reduction operators, but it's good to have a direct test for the
function itself.
---------
Co-authored-by: Philippe Tillet <phil@openai.com>
When running python 3.8, I've found that process creation gets slower
over time (e.g. after creating a CUDA context, it can take 50-300ms per
subprocess.run), and we do one of these calls to `ptxas --version` for
every kernel, so a model with thousands of kernels can end up spending
substantial time just calling ptxas redundantly.
Co-authored-by: Philippe Tillet <phil@openai.com>
This PR contains:
- Several fixes for the matrix multiplication (M and N dimensions may
have out-of-bound access)
- A type check for block-based store
- The tutorial for block pointers
- Fix some formats
A small oversight in #1305, since `view` can rearrange elements it
should be avoided here. Instead I use indexing with `None` to create new
dimensions.
Co-authored-by: Philippe Tillet <phil@openai.com>
Additional context: https://github.com/ROCmSoftwarePlatform/frameworks-internal/issues/3367#issuecomment-1505072217
When triton is installed via `python setup.py install` the required cuda2gcn.bc file is not copied over to the package location. This results in UT failures in pytorch `Failed to load /opt/conda/envs/py_3.8/lib/python3.8/site-packages/triton/language/cuda2gcn.bcTranslate to LLVM IR failedLLVM ERROR: Failed to translate TritonGPU to LLVM IR.`
To alleviate this issue I have proposed to add the .bc file to package_data of setup.py to ensure the file is copied over.
Reproducing torch UT:
`pytest test/inductor/test_torchinductor_dynamic_shapes.py -k "test_any_dynamic_shapes_cuda" --verbose`
