6 Commits

Author	SHA1	Message	Date
Tom Ritchford	d8c8ba2440	Fix unused Python variables in test/[e-z]* (#136964) ... Pull Request resolved: https://github.com/pytorch/pytorch/pull/136964 Approved by: https://github.com/justinchuby, https://github.com/albanD	2024-12-18 23:02:30 +00:00
Jeff Daily	0711bff9aa	[ROCm] add skipCUDAIfVersionLessThan to unskip test_jiterator for ROCm (#99197) ... This unskips 121 tests that the decorator `@skipCUDAIf(_get_torch_cuda_version() < (11, 6))` was unintentionally skipping for ROCm. Other decorators such as `skipCUDAVersionIn` will only activate for the CUDA device, not the CPU or ROCm-as-CUDA. Pull Request resolved: https://github.com/pytorch/pytorch/pull/99197 Approved by: https://github.com/ngimel	2023-04-17 16:05:16 +00:00
Catherine Lee	eea0733045	Reduce pytest blocklist (#96016) ... `TestCase = object` or variations of it get switched to `TestCase = NoTest`. unittest collects test based on subclassing unittest.TestCase, so setting TestCase = object removes it from unittest test collection. pytest collects based on name (https://docs.pytest.org/en/7.1.x/reference/reference.html#confval-python_classes) but can be told to ignore a class (bottom of https://docs.pytest.org/en/7.1.x/example/pythoncollection.html#changing-naming-conventions) Pull Request resolved: https://github.com/pytorch/pytorch/pull/96016 Approved by: https://github.com/ZainRizvi, https://github.com/huydhn	2023-03-07 18:30:27 +00:00
Jeff Daily	ff5fe9e622	[ROCm] enable jiterator (#77982) ... ### Description Enables jiterator for ROCm builds. This includes necessary porting when hiprtc and nvrtc behavior differed. This also ported ROCm versus CUDA differences w.r.t. MAX_DIMS and NUM_THREADS from the non-jiterator code paths into jiterator. ### Testing CI with ciflow/trunk label to force running ROCm workflows that are currently trunk-only. Pull Request resolved: https://github.com/pytorch/pytorch/pull/77982 Approved by: https://github.com/ngimel	2022-08-15 16:04:09 +00:00
Sherlock Huang	6db8440f35	Python Jiterator supports multiple outputs (#78139) ... This PR is part3. Part1: https://github.com/pytorch/pytorch/pull/77902 Part2: https://github.com/pytorch/pytorch/pull/77921 Python Jiterator now supports returning multiple outputs ``` fn = torch.cuda.jiterator._create_multi_output_jit_fn( """ template <typename T> T binary_2outputs(T i0, T i1, T& out0, T& out1) { out0 = i0 + i1; out1 = i0 - i1; } """, num_outputs=2) x = torch.rand(3, device='cuda') y = torch.rand(3, device='cuda') out0, out1 = fn(x, y) torch.allclose(out0, x+y) torch.allclose(out1, x-y) ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/78139 Approved by: https://github.com/ngimel	2022-05-24 21:52:56 +00:00
Sherlockk Huang	8b6a78f39f	Python Interface for Jiterator ... This PR allows user to author a CUDA kernel in python. ``` from torch.cuda.jiterator import create_jit_fn code_string = "template <typename T> T my_kernel(T x, T y, T alpha) { return -x * y + x - y + alpha; }" jitted_fn = create_jit_fn(code_string, alpha=0) a = torch.rand(3, device='cuda') b = torch.rand(3, device='cuda') result = jitted_fn(a, b, alpha=1.0) ``` Limitations: - Only supports elementwise kernel - 1~8 tensor inputs (empty input, e.g. factory methods, is not supported) - inputs tensors must live in cuda device - cpu Scalar is not supported - kwargs must be pre-declared when calling create_jit_fn - kwargs must be convertible to at::Scalar, one of float64, int64_t, bool. (complex not support for now) TODOs: - [x] consolidate union and c10::variant implementation - [x] plug into existing op testing framework - [ ] rename files, place files in the right folder - [ ] place util functions in the right file - [x] enforce assumptions in python interface e.g <8 inputs, kwargs types - [x] Add user-facing documentation Pull Request resolved: https://github.com/pytorch/pytorch/pull/76394 Approved by: https://github.com/mruberry	2022-05-06 18:44:28 +00:00