dmypy silently ignores follow_imports = skip, so to get parity between
dmypy and mypy we have to suck it up and type: ignore all of the sympy
typing problems.
The suppressions were added automatically with the following script generated by GPT-4:
```
import re
# Read the error file
with open("error_file.txt", "r") as f:
errors = f.readlines()
# Parse the lines with errors and error types
error_lines = {}
for error in errors:
match = re.match(r"(.*):(\d+):\d+: error:.*\[(.*)\]", error)
if match:
file_path, line_number, error_type = match.groups()
if file_path not in error_lines:
error_lines[file_path] = {}
error_lines[file_path][int(line_number)] = error_type
# Insert ignore comments in the source files
for file_path, lines in error_lines.items():
with open(file_path, "r") as f:
code = f.readlines()
for line_number, error_type in sorted(lines.items(), key=lambda x: x[0], reverse=True):
code[line_number - 1] = code[line_number - 1].rstrip() + f" # type: ignore[{error_type}]\n"
with open(file_path, "w") as f:
f.writelines(code)
```
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/118469
Approved by: https://github.com/Skylion007
ghstack dependencies: #118414, #118418, #118432, #118467, #118468
This PR brings in a few inductor changes required for ROCm
~**1 - Introduction of a toggle for enforced channel last convolution fallbacks**~
This addition is split off into its own PR after some cleanup by @pragupta https://github.com/pytorch/pytorch/pull/107812
**2 - Addition of ROCm specific block sizes**
We are now able to support the MAX_AUTOTUNE mode on ROCm, we are proposing conditions to allow us to finetune our own block tuning. Currently triton on ROCm does not benefit from pipelining so we are setting all configs to `num_stages=1` and we have removed some upstream tunings on ROCm to avoid running out of shared memory resources.
In the future we will provide more optimised tunings for ROCm but for now this should mitigate any issues
~**3 - Addition of device_type to triton's compile_meta**~
~Proposing this addition to `triton_heuristics.py`, Triton on ROCm requires device_type to be set to hip https://github.com/ROCmSoftwarePlatform/triton/pull/284 suggesting to bring this change in here so we can pass down the correct device type to triton.~
This change is split off and will arrive in the wheel update PR https://github.com/pytorch/pytorch/pull/107600 leaving this PR to focus on the ROCm specific block sizes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107584
Approved by: https://github.com/jithunnair-amd, https://github.com/jansel, https://github.com/eellison
The guard functions require you to ALREADY KNOW that a particular
condition holds. If you don't know (you want to guard on an expression
being a particular value, and then get access to that value), use
the evaluate functions.
I renamed the functions that don't abide by this:
```
guard_min -> evaluate_min
guard_max (deleted, no uses)
guard_static_shape -> evaluate_static_shape
guard_static_shapes -> evaluate_static_shapes
```
Some added comments.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105585
Approved by: https://github.com/voznesenskym
The changes in this PR include:
- Support ConvTranspose in cpp wrapper
- Fix cpp wrapper support for aten convolution when bias is `not None`: bias is in `args` instead of `kwargs` when it is `not None`. The change is covered by ConvTranspose dynamic shapes UT since we'll fall back to aten convolution in dynamic shape cases.
- Fix cpp wrapper support for `inf`. This is a UT added in https://github.com/pytorch/pytorch/issues/101865. The cpp wrapper UT is covered in `test_conv2d_unary` of `test_cpp_wrapper.py`. It's in `slowTest` category and seems not captured in the CI of that PR.
I will submit another PR to remove the hard-coded schema in these `ExternKernel`s.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103308
Approved by: https://github.com/jgong5, https://github.com/desertfire
This changes `TritonKernel` to have an `index_dtype` property which is
used as the dtype in indexing calculations. By default it is
`tl.int32` but if any input or output buffer is larger than `INT_MAX`
then we use `tl.int64` instead.
should fix#96978, #93606 (need to double check)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97447
Approved by: https://github.com/ngimel
