Jason Ansel comments

Results 199 comments of


                                            Jason Ansel

Use the device interface for detecting Triton availability

Test failure is `RuntimeError: Found Tesla M60 which is too old to be supported by the triton GPU compiler, which is used as the backend. Triton only supports devices of...

LLVM ERROR: Transformation resulted in an invalid module (bfloat16+cuda)

Another example from: ```py def fn(a, b): x = a + b x_view = x.view(dtype=torch.int16) return x_view.mul(2) ``` Generates: ```py import halide as hl from math import inf, nan @hl.generator(name="kernel")...

32 minute compile time for max_pool2d_with_indices

Yeah, it finishes and runs correctly.

Scheduling a fast CUDA reduction

An example schedule for a reduction would be super useful. The best I was able to write was still 3.5x slower than the baseline. For a reduction (sum, prod, max,...

[inductor] Add typing to ir.py 2

@pytorchbot merge

Fix: list index out of range with softmax when using 0 dim

@pytorchbot merge

Fix: list index out of range with softmax when using 0 dim

@krastogi-in can you fix the lint errors?

Fix: list index out of range with softmax when using 0 dim

@pytorchbot merge

Fix: list index out of range with softmax when using 0 dim

@pytorchbot merge

[Inductor] Support tiling reduction dimensions

Heads up this will have some conflicts with https://github.com/pytorch/pytorch/pull/137756