commitId stringlengths 40 40 | datetime stringlengths 30 31 | subject stringlengths 37 266 | comment stringlengths 109 15.2k | diff stringlengths 238 914k | gitVersion stringclasses 9
values |
|---|---|---|---|---|---|
5a8e9ee7776ba53a768df32a86c1b8a45049fdf | Tue, 30 Apr 2024 16:59:20 -0700 | [PATCH 0877/1000] [inductor] better cache clearing in fx graph cache tests (#125280) | Summary: There's a shortcoming in the FX graph cache tests in that they don't fully clear all inductor in-memory caches when testing the cache-hit path: We were previously accessing the FX graph cache correctly, but when loading the source object using the PyCodeCache.load_by_key_path() method, _that_ path was serving ... | diff --git a/test/inductor/test_codecache.py b/test/inductor/test_codecache.py
index 96ed0d7022..28403bf0df 100644
--- a/test/inductor/test_codecache.py
+++ b/test/inductor/test_codecache.py
@@ -21,7 +21,7 @@ from torch._inductor.codecache import (
)
from torch._inductor.runtime.runtime_utils import cache_dir
from t... | 2.41.0 |
4857e71c261b64cb5265c58ce8eab5b69f87af6 | Wed, 1 May 2024 05:38:01 +0000 | [PATCH 0878/1000] Export `torch.jit.interface` from `torch.jit` package (#125209) | Seems like this symbol was overlooked when other symbols were exported from `torch.jit`. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125209 Approved by: https://github.com/ezyang | diff --git a/torch/jit/__init__.py b/torch/jit/__init__.py
index 8ed5c0727e..a5b9f5627e 100644
--- a/torch/jit/__init__.py
+++ b/torch/jit/__init__.py
@@ -81,6 +81,7 @@ __all__ = [
"export_opnames",
"fork",
"freeze",
+ "interface",
"ignore",
"isinstance",
"load", | 2.41.0 |
81f41a920114d766d303fc4234d05103f1d4d89 | Wed, 1 May 2024 05:43:35 +0000 | [PATCH 0879/1000] Use BFloat16 in distributed quantization when supported by NCCL (#125113) | This PR enables BFloat16 in torch/csrc/distributed/c10d/quantization/quantization_gpu.cu . Pull Request resolved: https://github.com/pytorch/pytorch/pull/125113 Approved by: https://github.com/kwen2501 | diff --git a/torch/csrc/distributed/c10d/quantization/quantization_gpu.cu b/torch/csrc/distributed/c10d/quantization/quantization_gpu.cu
index c9b6185a40..48cc7cfc4f 100644
--- a/torch/csrc/distributed/c10d/quantization/quantization_gpu.cu
+++ b/torch/csrc/distributed/c10d/quantization/quantization_gpu.cu
@@ -69,15 +69... | 2.41.0 |
3c4465f504282f61f962027adf2bb9ea196b976 | Tue, 30 Apr 2024 16:55:24 -0700 | [PATCH 0880/1000] Add has_guarded_code to CompilationMetrics (#125279) | While studying some tlparse, I noticed that CompilationMetrics was reporting that there was no error for frames that have no nodes. I'm pretty sure we don't actually install a frame in this situation. has_guarded_code will tell us if that's the case, because it says if the GuardedCode object is None or not. Actually, ... | diff --git a/torch/_dynamo/convert_frame.py b/torch/_dynamo/convert_frame.py
index f2355ca66d..574c56454d 100644
--- a/torch/_dynamo/convert_frame.py
+++ b/torch/_dynamo/convert_frame.py
@@ -696,6 +696,7 @@ def _compile(
fail_reason: Optional[str] = None
fail_user_frame_filename: Optional[str] = None
... | 2.41.0 |
ead440c62c94d0a5b9055bd34e961be4aabafdb | Wed, 1 May 2024 07:34:04 +0000 | [PATCH 0882/1000] [Inductor] Further tune block size for templated attention on H100 (#125286) | Run a script to enumerate and get the best default block size for templated attention. A100 -> no change, check numbers at #125139 H100 ## torch.bfloat16 Before: ``` | Type | Speedup | batch_size | num_heads | q_seq_len | k_seq_len | head_dim | score_mod | dtype | |---------|-----------|--... | diff --git a/torch/_inductor/kernel/flex_attention.py b/torch/_inductor/kernel/flex_attention.py
index e31dfe0977..635af59f8c 100644
--- a/torch/_inductor/kernel/flex_attention.py
+++ b/torch/_inductor/kernel/flex_attention.py
@@ -3,7 +3,7 @@ import logging
from typing import Any, List
import torch
-from .. import ... | 2.41.0 |
2142192d488f983bdbc6a65de77a0859e21d66d | Tue, 30 Apr 2024 14:32:02 -0700 | [PATCH 0883/1000] [pipelining] Add stage backward function (#124958) | This is a helper function which: 1. computes the gradients for the stage inputs, and 2. accumulates gradients for the stage module's parameters. A unit test for this function is also added. Pull Request resolved: https://github.com/pytorch/pytorch/pull/124958 Approved by: https://github.com/wconstab ghstack dependenc... | diff --git a/docs/source/distributed.pipelining.rst b/docs/source/distributed.pipelining.rst
new file mode 100644
index 0000000000..ec7423f261
--- /dev/null
+++ b/docs/source/distributed.pipelining.rst
@@ -0,0 +1,178 @@
+.. role:: hidden
+ :class: hidden-section
+
+Pipeline Parallelism
+####################
+
+.. no... | 2.41.0 |
59cce38a9b389885c545a6f1b5090aa86539250 | Wed, 1 May 2024 10:39:13 +0000 | [PATCH 0884/1000] [MacOS][CPUInductor] Fix includes to system Python (#125285) | On MacOS 14.4, system Python is configured to point to a non-existing include dir ``` % /usr/bin/python3 -c "import sysconfig;print(sysconfig.get_path('include'))" /Library/Python/3.9/include ``` Workaround the issue by composing path to include folder from `stlib` config, which points to ``` % /usr/bin/python3 -c "im... | diff --git a/torch/_inductor/codecache.py b/torch/_inductor/codecache.py
index 16af022a21..5d884ee62b 100644
--- a/torch/_inductor/codecache.py
+++ b/torch/_inductor/codecache.py
@@ -1456,6 +1456,19 @@ def _set_gpu_runtime_env() -> None:
os.environ["CUDA_HOME"] = os.path.dirname(build_paths.cuda())
+def _g... | 2.41.0 |
7ba7a76e230c9a1355cd16d254a9516d406f95a | Wed, 1 May 2024 10:57:10 +0000 | [PATCH 0885/1000] [ATen][CUDA][AMP] Fix dtype mismatch in linalg_vector_norm (#125175) | Fixes #125174 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125175 Approved by: https://github.com/eqy, https://github.com/lezcano | diff --git a/aten/src/ATen/native/LinearAlgebra.cpp b/aten/src/ATen/native/LinearAlgebra.cpp
index 1849228a63..81f461f6c9 100644
--- a/aten/src/ATen/native/LinearAlgebra.cpp
+++ b/aten/src/ATen/native/LinearAlgebra.cpp
@@ -2839,10 +2839,16 @@ TORCH_IMPL_FUNC(linalg_vector_norm_out)(const Tensor& self, const Scalar& sca... | 2.41.0 |
fbb4dfc125943547df5d484dbb2be4a8b48c0f3 | Wed, 1 May 2024 12:08:02 +0000 | [PATCH 0886/1000] Fix AttributeError when doing mock patch for FileTimerServerTest.test_expired_timers (#125144) | Fix the patch failure, and we should patch the function where it is used, not where it is defined. Failure info: ```bash root@cambricon-PowerEdge-C4140:/workspace# python file_based_timer_test.py -k test_expired_timers /opt/conda/lib/python3.10/site-packages/torch/_custom_ops.py:253: DeprecationWarning: torch.library.i... | diff --git a/test/distributed/elastic/timer/file_based_local_timer_test.py b/test/distributed/elastic/timer/file_based_local_timer_test.py
index 4616ae061b..490e4a9ce3 100644
--- a/test/distributed/elastic/timer/file_based_local_timer_test.py
+++ b/test/distributed/elastic/timer/file_based_local_timer_test.py
@@ -264,7... | 2.41.0 |
2715144c332a3b3e9032b0c9ddee592a1b572da | Wed, 1 May 2024 14:04:46 +0000 | [PATCH 0887/1000] Add NEON-accelerated int8mm for bfloat16 (#125290) | As apparently `vshlq_u32` is faster than `vcvt_f32_f16` Refactor NEON `tinygemm_kernel` to rely on `load_as_float32x4` and `load_as_float32x4x2` and implement them for float16 (using vcvt), bfloat16 (using left shift) and plain float32 (not using anything) As result stories110M run at 60 tokens/sec with f16, but at 6... | diff --git a/aten/src/ATen/native/cpu/int8mm_kernel.cpp b/aten/src/ATen/native/cpu/int8mm_kernel.cpp
index 4ef6cde4a8..bd266030b2 100644
--- a/aten/src/ATen/native/cpu/int8mm_kernel.cpp
+++ b/aten/src/ATen/native/cpu/int8mm_kernel.cpp
@@ -185,17 +185,50 @@ inline void tinygemm_kernel(
#if !defined(C10_MOBILE) && defin... | 2.41.0 |
421f1b4a86ccd0305ac5b3343f8b18097c0adc8 | Wed, 1 May 2024 14:27:37 +0000 | [PATCH 0888/1000] docs: `torch.nn.utils.rnn`: docs improve (#123559) | docs: `torch.nn.utils.rnn`: docs improve Pull Request resolved: https://github.com/pytorch/pytorch/pull/123559 Approved by: https://github.com/mikaylagawarecki | diff --git a/torch/nn/utils/rnn.py b/torch/nn/utils/rnn.py
index 27c902e9aa..1a62254f08 100644
--- a/torch/nn/utils/rnn.py
+++ b/torch/nn/utils/rnn.py
@@ -212,10 +212,10 @@ def pack_padded_sequence(
) -> PackedSequence:
r"""Packs a Tensor containing padded sequences of variable length.
- :attr:`input` can be... | 2.41.0 |
09f98c705e4851414cd8ddf21949177af2b13aa | Wed, 1 May 2024 14:31:31 +0000 | [PATCH 0889/1000] =?UTF-8?q?Include=20support=20for=20the=20scatt?= =?UTF-8?q?er=20gather=20cuda=20kernels=20to=20allow=20for=20comp=E2=80=A6?= =?UTF-8?q?=20(#124809)?=MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | Fixes #121965 This PR hopes to add support complex numbers in the scatter/gather related kernels. For brevity, I will only include `complex<float>` for now as `complex<double>`, for example, will be more complicated. C++ unit tests are currently passing alongside tests in `test_scatter_gather_ops.py`. Python test sui... | diff --git a/aten/src/ATen/cuda/Atomic.cuh b/aten/src/ATen/cuda/Atomic.cuh
index 56ee8f87e2..2fa55902f9 100644
--- a/aten/src/ATen/cuda/Atomic.cuh
+++ b/aten/src/ATen/cuda/Atomic.cuh
@@ -35,6 +35,26 @@ struct AtomicFPOp<at::Half> {
}
};
+template <>
+struct AtomicFPOp<c10::complex<float>> {
+ template <typename ... | 2.41.0 |
fde9a988c2b6ba5cd88ee884227563a5f669912 | Wed, 1 May 2024 15:37:48 +0000 | [PATCH 0891/1000] CI: Extending unit test coverage for aarch64 linux (#125255) | Adding core, dynamo and inductor unit tests for aarch64 linux CI runs. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125255 Approved by: https://github.com/malfet, https://github.com/atalman | diff --git a/.ci/pytorch/test.sh b/.ci/pytorch/test.sh
index 5a1e098636..c903a26998 100755
--- a/.ci/pytorch/test.sh
+++ b/.ci/pytorch/test.sh
@@ -1158,8 +1158,23 @@ test_executorch() {
}
test_linux_aarch64(){
- # TODO: extend unit tests list
- python test/run_test.py --include test_modules test_mkldnn test_mkldn... | 2.41.0 |
16f1ee4cc3e58dbd4755021e4b4b87a16b2aac2 | Wed, 1 May 2024 15:48:48 +0000 | [PATCH 0892/1000] [ez][CI] Move test_modules and test_schema_check off CI_SERIAL_LIST (#125193) | * Related https://github.com/pytorch/pytorch/pull/124085 As in title, move test_modules and test_schema_check off CI_SERIAL_LIST If things fail, they can get the serialTest decorator instead Pull Request resolved: https://github.com/pytorch/pytorch/pull/125193 Approved by: https://github.com/huydhn | diff --git a/test/run_test.py b/test/run_test.py
index 9945859987..1b95bcb465 100755
--- a/test/run_test.py
+++ b/test/run_test.py
@@ -227,11 +227,9 @@ CI_SERIAL_LIST = [
"nn/test_pooling",
"nn/test_convolution", # Doesn't respect set_per_process_memory_fraction, results in OOM for other tests in slow gradch... | 2.41.0 |
d410155b241334b32872ff593d3480d10f57d5e | Wed, 1 May 2024 16:02:02 +0000 | [PATCH 0893/1000] =?UTF-8?q?Revert=20"Include=20support=20for=20t?= =?UTF-8?q?he=20scatter=20gather=20cuda=20kernels=20to=20allow=20for=20com?= =?UTF-8?q?p=E2=80=A6=20(#124809)"?=MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | This reverts commit e09f98c705e4851414cd8ddf21949177af2b13aa. Reverted https://github.com/pytorch/pytorch/pull/124809 on behalf of https://github.com/clee2000 due to windows build failure is real, https://github.com/pytorch/pytorch/actions/runs/8910674030/job/24470387612#step:11:11236 is the correct failure line, igno... | diff --git a/aten/src/ATen/cuda/Atomic.cuh b/aten/src/ATen/cuda/Atomic.cuh
index 2fa55902f9..56ee8f87e2 100644
--- a/aten/src/ATen/cuda/Atomic.cuh
+++ b/aten/src/ATen/cuda/Atomic.cuh
@@ -35,26 +35,6 @@ struct AtomicFPOp<at::Half> {
}
};
-template <>
-struct AtomicFPOp<c10::complex<float>> {
- template <typename ... | 2.41.0 |
9eb5d4fa4a09cd462b51f460e71e16768188a44 | Wed, 1 May 2024 16:59:35 +0000 | [PATCH 0894/1000] Add Sanity Testing to Pytorch Profiler (#124773) | Summary: In the recent weeks, we have encountered bugs in both the normal synchronous trace and on-demand tracing. This diff on its own does sanity checking to make sure the profiler does not have spans that extend past the boundaries that we expect. It also checks some basic properties of the tracings we expect to see... | diff --git a/test/profiler/test_profiler.py b/test/profiler/test_profiler.py
index 2ec04c447f..ff6dc640dd 100644
--- a/test/profiler/test_profiler.py
+++ b/test/profiler/test_profiler.py
@@ -22,6 +22,7 @@ import re
import subprocess
import sys
import threading
+import time
import unittest
from dataclasses import d... | 2.41.0 |
3627d05e789c6cd20eeccce89dd733f02a808a9 | Wed, 1 May 2024 17:26:28 +0000 | [PATCH 0895/1000] [CMake] Add NVPL BLAS/LAPACK option (#125268) | This PR add a [NVPL](https://docs.nvidia.com/nvpl/introduction.html) BLAS/LAPACK option to CMake for `aarch64` (ARM) machines. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125268 Approved by: https://github.com/albanD | diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
index 5b30bef4fc..a8313f206f 100644
--- a/cmake/Dependencies.cmake
+++ b/cmake/Dependencies.cmake
@@ -237,6 +237,12 @@ elseif(BLAS STREQUAL "MKL")
set(CAFFE2_USE_EIGEN_FOR_BLAS ON)
set(CAFFE2_USE_MKL OFF)
endif()
+elseif(BLAS STREQUAL "NVPL")... | 2.41.0 |
8d2a55273757c90989fde7c6f05e957aba9a238 | Wed, 1 May 2024 17:45:11 +0000 | [PATCH 0896/1000] Intel GPU: specify the tolerance for torchbench models (#125213) | We encountered some model accuracy failures as the tolerance is critical. In general, we align with CUDA practice. This PR intends to adjust the tolerance for Torchbench models for training mode on Intel GPU devices and aligns with CUDA. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125213 Approved by... | diff --git a/benchmarks/dynamo/torchbench.py b/benchmarks/dynamo/torchbench.py
index a6b4edb3a4..274d04da15 100755
--- a/benchmarks/dynamo/torchbench.py
+++ b/benchmarks/dynamo/torchbench.py
@@ -402,7 +402,7 @@ class TorchBenchmarkRunner(BenchmarkRunner):
if name in self._tolerance["higher_bf16"]:
... | 2.41.0 |
f6acf9add0f52a483c9ea073a7adfe4ab1233dd | Wed, 1 May 2024 18:30:57 +0000 | [PATCH 0897/1000] [ROCm] Add extra cuda_to_hip_mappings.py (#125108) | Adding extra mappings discovered when hipifying the backward CUDA kernel of the Mamba model (https://github.com/state-spaces/mamba/). Co-authored-by: Jeff Daily <jeff.daily@amd.com> Pull Request resolved: https://github.com/pytorch/pytorch/pull/125108 Approved by: https://github.com/Skylion007, https://github.com/jeff... | diff --git a/torch/utils/hipify/cuda_to_hip_mappings.py b/torch/utils/hipify/cuda_to_hip_mappings.py
index e48c595928..d07292a22b 100644
--- a/torch/utils/hipify/cuda_to_hip_mappings.py
+++ b/torch/utils/hipify/cuda_to_hip_mappings.py
@@ -643,7 +643,11 @@ CUDA_INCLUDE_MAP = collections.OrderedDict(
("thrust/sy... | 2.41.0 |
7422fd0b9ad21eb9e44c68172d4a85b3d03769e | Wed, 1 May 2024 18:35:49 +0000 | [PATCH 0898/1000] add missing space to first cmake append (#125294) | the first append not having a space incorrectly merges it to any previous arguments, like `-allow-unsupported-compiler` in my case which results in a silly error: `unrecognized command-line option '-allow-unsupported-compiler-DLIBCUDACXX_ENABLE_SIMPLIFIED_COMPLEX_OPERATIONS'` full log: ``` python setup.py develop Buil... | diff --git a/CMakeLists.txt b/CMakeLists.txt
index 215ec7a81a..9ce1e06bea 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -56,7 +56,7 @@ endif()
# This define is needed to preserve behavior given anticpated changes to cccl/thrust
# https://nvidia.github.io/libcudacxx/standard_api/numerics_library/complex.html
-... | 2.41.0 |
97612c84c37eb99ef6d18f579e7fd8a52a31d28 | Wed, 1 May 2024 19:59:51 +0000 | [PATCH 0902/1000] ProcessGroupWrapper support custom backend (#124447)MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | Fixes #ISSUE_NUMBER In current code, ProcessGroupWrapper works only for `GLOO, NCCL, UCC` when `TORCH_DISTRIBUTED_DEBUG=DETAIL`. I read the ProcessGroupWrapper code，find that communication_op in ProcessGroupWrapper is just communication_op in origin_backend + runCollectiveChecks in gloo, like allreduce: https://github.... | diff --git a/torch/distributed/distributed_c10d.py b/torch/distributed/distributed_c10d.py
index 74f2ed5845..c006fbc08c 100644
--- a/torch/distributed/distributed_c10d.py
+++ b/torch/distributed/distributed_c10d.py
@@ -1602,7 +1602,7 @@ def _new_process_group_helper(
break
# Process group wrappe... | 2.41.0 |
bcf123105a3f11d02f04067ca0cb377ed09e88c | Wed, 1 May 2024 20:59:17 +0000 | [PATCH 0904/1000] Upgrade submodule oneDNN to v3.4 (#122472)MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | ## Improvements This upgrade fixes the following issues: - https://github.com/pytorch/pytorch/issues/120982 This upgrade brings the following new features: - Introduced memory descriptor serialization API. This API is needed to support freezing on CPU in AOTInductor (https://github.com/pytorch/pytorch/issues/114450) ... | diff --git a/third_party/ideep b/third_party/ideep
index 8a6cc4e09d..420c9806c8 160000
--- a/third_party/ideep
+++ b/third_party/ideep
@@ -1 +1 @@
-Subproject commit 8a6cc4e09dc509f04f83c085e38786b1fb44e14d
+Subproject commit 420c9806c870711668a4bbc469f1d40839532fea
diff --git a/third_party/mkl-dnn.BUILD b/third_party/... | 2.41.0 |
46da8755c45c331dc4c7513ff89b531f978c2cd | Wed, 1 May 2024 21:01:26 +0000 | [PATCH 0905/1000] switch tests from constrain_as* to torch._check* (#125253) | To fix data-dependent errors we want to recommend that people use `torch._check*` APIs. The `constrain_as*` APIs should be fully subsumed by them, and in the future we should kill them entirely. Differential Revision: D56774333 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125253 Approved by: https:/... | diff --git a/test/distributed/test_c10d_functional_native.py b/test/distributed/test_c10d_functional_native.py
index 99062d1bab..54030d1f1d 100644
--- a/test/distributed/test_c10d_functional_native.py
+++ b/test/distributed/test_c10d_functional_native.py
@@ -703,7 +703,7 @@ class CompileTest(TestCase):
def _to... | 2.41.0 |
d794bcb8a9e7dec578b1e8587dc6513cd24c2e7 | Wed, 1 May 2024 13:00:06 -0400 | [PATCH 0906/1000] Delete NegateSource handling, I think it's dead (#125311) | Signed-off-by: Edward Z. Yang <ezyang@meta.com> Pull Request resolved: https://github.com/pytorch/pytorch/pull/125311 Approved by: https://github.com/Skylion007 | diff --git a/torch/fx/experimental/symbolic_shapes.py b/torch/fx/experimental/symbolic_shapes.py
index 748737a9e1..fdaeb3a060 100644
--- a/torch/fx/experimental/symbolic_shapes.py
+++ b/torch/fx/experimental/symbolic_shapes.py
@@ -3398,7 +3398,7 @@ class ShapeEnv:
# TODO: Make this more efficient by binding al... | 2.41.0 |
058563078e707b848b061a222d0195e7472be7d | Wed, 1 May 2024 09:51:21 -0700 | [PATCH 0907/1000] support as_python_constant on PlacementClassVariable (#124398) | Fixes an error for torchtitan + internal Pull Request resolved: https://github.com/pytorch/pytorch/pull/124398 Approved by: https://github.com/ezyang, https://github.com/wanchaol, https://github.com/yoyoyocmu | diff --git a/torch/_dynamo/variables/distributed.py b/torch/_dynamo/variables/distributed.py
index 2f4a2eb91e..6816ea9b1e 100644
--- a/torch/_dynamo/variables/distributed.py
+++ b/torch/_dynamo/variables/distributed.py
@@ -107,6 +107,9 @@ class PlacementClassVariable(DistributedVariable):
return type(value) ... | 2.41.0 |
173cbe260e8cfc9173da005a7b272c31511b05a | Wed, 1 May 2024 09:51:21 -0700 | [PATCH 0908/1000] fix FakeTensor creation on noncontiguous subclasses (#124399) | Fixes https://github.com/pytorch/pytorch/issues/125287 Fixes https://github.com/pytorch/pytorch/issues/124090, context on the issue Pull Request resolved: https://github.com/pytorch/pytorch/pull/124399 Approved by: https://github.com/soulitzer ghstack dependencies: #124398 | diff --git a/test/distributed/_tensor/test_dtensor_compile.py b/test/distributed/_tensor/test_dtensor_compile.py
index f9ad0278d7..21ca8ae8f0 100644
--- a/test/distributed/_tensor/test_dtensor_compile.py
+++ b/test/distributed/_tensor/test_dtensor_compile.py
@@ -252,6 +252,43 @@ class TestDTensorCompile(torch._dynamo.t... | 2.41.0 |
e9ba61fde8f97fc05ceaee6f4ccbf23fe6bf9e8 | Wed, 1 May 2024 09:51:22 -0700 | [PATCH 0909/1000] AOTAutograd: force tangents to be contiguous when subclass inner tensor is noncontiguous (#124400) | Fixes https://github.com/pytorch/pytorch/issues/124397 Pull Request resolved: https://github.com/pytorch/pytorch/pull/124400 Approved by: https://github.com/ezyang, https://github.com/yoyoyocmu ghstack dependencies: #124398, #124399 | diff --git a/test/distributed/_tensor/test_dtensor_compile.py b/test/distributed/_tensor/test_dtensor_compile.py
index 21ca8ae8f0..f726e35153 100644
--- a/test/distributed/_tensor/test_dtensor_compile.py
+++ b/test/distributed/_tensor/test_dtensor_compile.py
@@ -289,6 +289,41 @@ class TestDTensorCompile(torch._dynamo.t... | 2.41.0 |
99a2e25f132b6ad7cc060d1dad4a4f374eb100d | Wed, 1 May 2024 09:56:32 -0700 | [PATCH 0910/1000] Reland "make sure dynamo doesn't inline DTensor __new__ or __torch_dispatch__ (#123347)" (#125288) | Re-land of https://github.com/pytorch/pytorch/pull/123347. The original PR broke internal because of a circular import due to importing dynamo in the DTensor code. The new version uses `torch._dynamo_disable` to work around This reverts commit 9d88339b535f57cd0e2926c9ac4c2542e4490aac. Pull Request resolved: https://... | diff --git a/test/distributed/_tensor/test_dtensor_compile.py b/test/distributed/_tensor/test_dtensor_compile.py
index f726e35153..b26c8900dd 100644
--- a/test/distributed/_tensor/test_dtensor_compile.py
+++ b/test/distributed/_tensor/test_dtensor_compile.py
@@ -191,6 +191,47 @@ class TestDTensorCompile(torch._dynamo.t... | 2.41.0 |
06eda538b85c9fec7a35a4c8684cdef05fc5e9d | Wed, 1 May 2024 22:06:47 +0000 | [PATCH 0911/1000] Fix windows build error not propagating (#125306) | * Fixes https://github.com/pytorch/pytorch/issues/124886 * Kind of similar to https://github.com/pytorch/pytorch/pull/109393 I think what happens is `exit` and `exit /b` propagate the errorlevel correctly, but `exit /b` only exists the currently running batch script and not the entire cmd.exe (or whatever program is r... | diff --git a/.ci/pytorch/win-test-helpers/build_pytorch.bat b/.ci/pytorch/win-test-helpers/build_pytorch.bat
index 4b7bbad744..28bd083f98 100644
--- a/.ci/pytorch/win-test-helpers/build_pytorch.bat
+++ b/.ci/pytorch/win-test-helpers/build_pytorch.bat
@@ -17,22 +17,22 @@ set PATH=C:\Program Files\CMake\bin;C:\Program Fi... | 2.41.0 |
451d108da6b903e5a4d21d55be7988080cac684 | Wed, 1 May 2024 23:14:05 +0000 | [PATCH 0912/1000] Implemented isin_Tensor_Tensor_out for MPS backend (#124896) | Addresses issue #124518, adds isin_Tensor_Tensor_out. Tests added to test_mps.py. Pull Request resolved: https://github.com/pytorch/pytorch/pull/124896 Approved by: https://github.com/malfet, https://github.com/kulinseth | diff --git a/aten/src/ATen/native/mps/operations/TensorCompare.mm b/aten/src/ATen/native/mps/operations/TensorCompare.mm
index 78fa67e57d..f378af1326 100644
--- a/aten/src/ATen/native/mps/operations/TensorCompare.mm
+++ b/aten/src/ATen/native/mps/operations/TensorCompare.mm
@@ -12,7 +12,9 @@
#include <ATen/ops/clamp_m... | 2.41.0 |
cfb55dd5dfcffc06e039c9af2342f455dfd1007 | Wed, 1 May 2024 23:19:07 +0000 | [PATCH 0913/1000] Add a variable for some testcases. (#124708) | Some testcases can use 'TEST_PRIVATEUSE1_DEVICE_TYPE' to make adapting these testcases on others device more convenient. Fixes #ISSUE_NUMBER Pull Request resolved: https://github.com/pytorch/pytorch/pull/124708 Approved by: https://github.com/albanD | diff --git a/test/test_shape_ops.py b/test/test_shape_ops.py
index 189187b582..47acfff9c6 100644
--- a/test/test_shape_ops.py
+++ b/test/test_shape_ops.py
@@ -12,7 +12,7 @@ import unittest
from torch import nan
from torch.testing import make_tensor
from torch.testing._internal.common_utils import (
- TestCase, ru... | 2.41.0 |
f5f405b057c7de0f5fce0b1432cb74468f96f95 | Wed, 1 May 2024 23:29:55 +0000 | [PATCH 0914/1000] [ncclx] Rename NCCL-EXP to NCCLX (#125238) | Reviewed By: kryanchun Differential Revision: D56534548 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125238 Approved by: https://github.com/kwen2501 | diff --git a/test/cpp/c10d/ProcessGroupNCCLTest.cpp b/test/cpp/c10d/ProcessGroupNCCLTest.cpp
index d2dc02d323..d1c2380274 100644
--- a/test/cpp/c10d/ProcessGroupNCCLTest.cpp
+++ b/test/cpp/c10d/ProcessGroupNCCLTest.cpp
@@ -840,7 +840,7 @@ TEST_F(ProcessGroupNCCLTest, testSplittingCommunicator) {
multiThreadRun(testS... | 2.41.0 |
6f326eff56250d97e0df1f5e75943a72ebaa5c2 | Wed, 1 May 2024 11:38:49 -0700 | [PATCH 0915/1000] explicitly reset stderr/stdout in precompilation (#125289) | I was seeing a weird bug where after running max-autotune my stdout would be misdirected. other people have not been able to repro this. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125289 Approved by: https://github.com/shunting314, https://github.com/mlazos | diff --git a/torch/_inductor/select_algorithm.py b/torch/_inductor/select_algorithm.py
index 577a1c318b..8fcb441ed8 100644
--- a/torch/_inductor/select_algorithm.py
+++ b/torch/_inductor/select_algorithm.py
@@ -40,6 +40,7 @@ from .runtime.runtime_utils import do_bench
from .utils import (
get_dtype_size,
Pla... | 2.41.0 |
043ccafdf133002ce6c6993429c419054ee9ad7 | Tue, 30 Apr 2024 22:55:30 +0300 | [PATCH 0916/1000] Require nnz==0 in sparse meta tensors (#125221) | As in the title and per discussion starting at https://github.com/pytorch/pytorch/pull/117907#issuecomment-2082426468 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125221 Approved by: https://github.com/amjames, https://github.com/ezyang | diff --git a/aten/src/ATen/native/TensorConversions.cpp b/aten/src/ATen/native/TensorConversions.cpp
index a6c1118c4e..c70da8334a 100644
--- a/aten/src/ATen/native/TensorConversions.cpp
+++ b/aten/src/ATen/native/TensorConversions.cpp
@@ -259,6 +259,9 @@ Tensor _to_copy(
memory_format == MemoryFormat::Preser... | 2.41.0 |
281d3a0cb435792c88f5015fafd33af40e2004e | Wed, 1 May 2024 23:44:53 +0000 | [PATCH 0917/1000] Enable UFMT on test_indexing&test_view_ops (#125112) | Part of https://github.com/pytorch/pytorch/issues/123062 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125112 Approved by: https://github.com/ezyang | diff --git a/.lintrunner.toml b/.lintrunner.toml
index a945814b23..21dfdf5b3f 100644
--- a/.lintrunner.toml
+++ b/.lintrunner.toml
@@ -1071,7 +1071,6 @@ exclude_patterns = [
'test/test_fx_reinplace_pass.py',
'test/test_hub.py',
'test/test_import_stats.py',
- 'test/test_indexing.py',
'test/test_it... | 2.41.0 |
022f131b54065ffbfd8e17b4be73a5228f1139a | Sun, 28 Apr 2024 12:27:29 -0700 | [PATCH 0918/1000] [inductor] switch assume_aligned_inputs to False (#124336) | In #123319, we guard some behavior behind the `assume_aligned_inputs` config option. If we set this to `False`, then the behavior added in #123319 becomes the default behavior. See the referenced PR for more details about the behavior affected. Side effects: * It's possible that this will hurt performance in some scen... | diff --git a/test/inductor/test_cudagraph_trees.py b/test/inductor/test_cudagraph_trees.py
index 33ee5247bd..c8877d4a8e 100644
--- a/test/inductor/test_cudagraph_trees.py
+++ b/test/inductor/test_cudagraph_trees.py
@@ -634,7 +634,7 @@ if HAS_CUDA and not TEST_WITH_ASAN:
new_id = self.get_manager().new_grap... | 2.41.0 |
1f142c44f81384afbdba5e451fc15744868bf26 | Wed, 1 May 2024 23:56:00 +0000 | [PATCH 0919/1000] Revert "Fakify script object inputs and attributes for non-strict export (#124239)" | This reverts commit ecc2e034f7e55bf9ff7f4e5df4e9086a5c92caaa. Reverted https://github.com/pytorch/pytorch/pull/124239 on behalf of https://github.com/kit1980 due to breaking internal builds ([comment](https://github.com/pytorch/pytorch/pull/124239#issuecomment-2089305447)) | diff --git a/test/export/test_passes.py b/test/export/test_passes.py
index e2724ead88..41597a6030 100644
--- a/test/export/test_passes.py
+++ b/test/export/test_passes.py
@@ -13,10 +13,6 @@ from typing import List, Set
import torch
from functorch.experimental.control_flow import cond
from torch._dynamo.eval_frame im... | 2.41.0 |
e24c263f998819f849bb8293323213101e9aefc | Wed, 1 May 2024 23:58:33 +0000 | [PATCH 0920/1000] =?UTF-8?q?Include=20support=20for=20the=20scatt?= =?UTF-8?q?er=20gather=20cuda=20kernels=20to=20allow=20for=20comp=E2=80=A6?= =?UTF-8?q?=20(#124809)?=MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | Fixes #121965 This PR hopes to add support complex numbers in the scatter/gather related kernels. For brevity, I will only include `complex<float>` for now as `complex<double>`, for example, will be more complicated. C++ unit tests are currently passing alongside tests in `test_scatter_gather_ops.py`. Python test sui... | diff --git a/aten/src/ATen/NumericUtils.h b/aten/src/ATen/NumericUtils.h
index 788da64b4e..421fe0efab 100644
--- a/aten/src/ATen/NumericUtils.h
+++ b/aten/src/ATen/NumericUtils.h
@@ -38,7 +38,11 @@ inline C10_HOST_DEVICE bool _isnan(T val) {
template <typename T, std::enable_if_t<c10::is_complex<T>::value, int> = 0>... | 2.41.0 |
731130ea8a83a33b14e5b566130b2f6888c2e6b | Tue, 30 Apr 2024 20:09:36 -0700 | [PATCH 0923/1000] Add a code comment about torch._check_is_size in tensor_split (#125292) | Signed-off-by: Edward Z. Yang <ezyang@meta.com> Pull Request resolved: https://github.com/pytorch/pytorch/pull/125292 Approved by: https://github.com/albanD | diff --git a/torch/_decomp/decompositions.py b/torch/_decomp/decompositions.py
index 124ed8fb72..6cfccbab0d 100644
--- a/torch/_decomp/decompositions.py
+++ b/torch/_decomp/decompositions.py
@@ -1413,6 +1413,15 @@ def tensor_split_tensor_indices_or_sections_py_impl(
return self.tensor_split(sections, dim)
... | 2.41.0 |
119e1bcc2aa7e7d25bbad46cdcdfa06e8d3cc20 | Thu, 2 May 2024 02:34:30 +0000 | [PATCH 0924/1000] Fix refcount handling for dtype, layout and memory format (#125271) | Finish fixing https://github.com/pytorch/pytorch/issues/124868 re-use our wrap() utils as much as possible and NewRef in other places. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125271 Approved by: https://github.com/colesbury | diff --git a/tools/autograd/templates/python_nn_functions.cpp b/tools/autograd/templates/python_nn_functions.cpp
index f311cfebe4..4877df6584 100644
--- a/tools/autograd/templates/python_nn_functions.cpp
+++ b/tools/autograd/templates/python_nn_functions.cpp
@@ -60,14 +60,14 @@ static PyObject * THPVariable__parse_to(P... | 2.41.0 |
ea54839c90896d9ba1fcf1f26472f8b477954f0 | Wed, 1 May 2024 07:02:48 -0700 | [PATCH 0925/1000] Make min(stride, strides[idx]) in collapse_view_helper size oblivious (#125301) | Signed-off-by: Edward Z. Yang <ezyang@meta.com> Pull Request resolved: https://github.com/pytorch/pytorch/pull/125301 Approved by: https://github.com/albanD | diff --git a/torch/_prims/__init__.py b/torch/_prims/__init__.py
index 43ba508175..116671703e 100644
--- a/torch/_prims/__init__.py
+++ b/torch/_prims/__init__.py
@@ -1366,7 +1366,10 @@ def _collapse_view_helper(
continue
length = length * shape[idx]
- stride = min(stride, strides[idx])
+... | 2.41.0 |
ff7a31800296669d0dc880282934efa45d4c339 | Thu, 2 May 2024 03:38:32 +0000 | [PATCH 0927/1000] fix torchdeploy issue on sharddim_alltoall op (#125344) | Summary: fix torchdeploy issues when registering the distributed op, similar to what functional collective did Differential Revision: D56850434 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125344 Approved by: https://github.com/XilunWu, https://github.com/fegin | diff --git a/torch/distributed/_tensor/_collective_utils.py b/torch/distributed/_tensor/_collective_utils.py
index ce4809d996..fe712f3cb2 100644
--- a/torch/distributed/_tensor/_collective_utils.py
+++ b/torch/distributed/_tensor/_collective_utils.py
@@ -23,11 +23,20 @@ from torch.distributed.distributed_c10d import (
... | 2.41.0 |
a5d2d9b3ed8c61fb463226854ba92cd0fb682bd | Wed, 1 May 2024 16:41:09 -0700 | [PATCH 0928/1000] Hotfix: restore CPP guard string in structured trace (#125303) | Signed-off-by: Edward Z. Yang <ezyang@meta.com> Pull Request resolved: https://github.com/pytorch/pytorch/pull/125303 Approved by: https://github.com/albanD | diff --git a/test/dynamo/test_structured_trace.py b/test/dynamo/test_structured_trace.py
index 07f541edbe..1213a48ffc 100644
--- a/test/dynamo/test_structured_trace.py
+++ b/test/dynamo/test_structured_trace.py
@@ -141,6 +141,7 @@ class StructuredTraceTest(TestCase):
{"inductor_post_grad_graph": {}, "frame_id": 0, "fr... | 2.41.0 |
b70026d3b549782ee194c85a0a0f6805cceaed7 | Wed, 1 May 2024 16:55:30 -0700 | [PATCH 0930/1000] Do not pass none to has_pending_mutation (#125359) | #fix https://github.com/pytorch/pytorch/issues/125315 Several failures when inlining nn module is enabled are due to passing None to has_pending_mutation from previous code, it sounds like its expected for variable to be none when not found, In that case we should skip it and not call has_pending_mutation this is test... | diff --git a/torch/_dynamo/variables/optimizer.py b/torch/_dynamo/variables/optimizer.py
index e183f7a5e5..f7d2056aa2 100644
--- a/torch/_dynamo/variables/optimizer.py
+++ b/torch/_dynamo/variables/optimizer.py
@@ -112,9 +112,8 @@ class OptimizerVariable(UserDefinedObjectVariable):
for g in self.value.param_gr... | 2.41.0 |
13a0a247906941438717e74b7990c26b2246e0c | Thu, 2 May 2024 00:22:14 -0700 | [PATCH 0931/1000] [dynamo][easy] Simple fixes to prepare for nn module guards (#125316) | Pull Request resolved: https://github.com/pytorch/pytorch/pull/125316 Approved by: https://github.com/williamwen42 ghstack dependencies: #125275 | diff --git a/test/dynamo/test_modules.py b/test/dynamo/test_modules.py
index de0e66e59f..b46844f219 100644
--- a/test/dynamo/test_modules.py
+++ b/test/dynamo/test_modules.py
@@ -1961,6 +1961,7 @@ class OptimizedModuleTest(torch._dynamo.test_case.TestCase):
self.assertEqual(compiled_func(inp).item(), 16)
... | 2.41.0 |
b1bfe115693d4dd3cc2be68a2ea62717bc7f18e | Thu, 2 May 2024 15:17:10 +0000 | [PATCH 0933/1000] Get cutlass_library import working under fbcode (#125257) | Differential Revision: D56764089 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125257 Approved by: https://github.com/chenyang78 | diff --git a/torch/_inductor/codegen/cuda/cutlass_utils.py b/torch/_inductor/codegen/cuda/cutlass_utils.py
index ff60525548..62465b0883 100644
--- a/torch/_inductor/codegen/cuda/cutlass_utils.py
+++ b/torch/_inductor/codegen/cuda/cutlass_utils.py
@@ -10,6 +10,7 @@ from typing import Any, List, Optional
import sympy
... | 2.41.0 |
93b57a57031b893139a6b334cf64b6201dde8c4 | Tue, 30 Apr 2024 17:08:35 -0700 | [PATCH 0934/1000] Add propagate_real_tensors mode for unbacked (#125115) | A common complaint when working with data-dependent code in PyTorch is that it's hard to tell how far you are from the finish line: every time a GuardOnDataDependentSymNode error is hit, you have to somehow fix or workaround it to see the next one. This PR adds a new mode `torch._functorch.config.fake_tensor_propagate... | diff --git a/benchmarks/dynamo/common.py b/benchmarks/dynamo/common.py
index 2ecadef60e..48ea0496da 100644
--- a/benchmarks/dynamo/common.py
+++ b/benchmarks/dynamo/common.py
@@ -75,6 +75,7 @@ except ImportError:
graph_break_reasons,
maybe_enable_compiled_autograd,
)
+import torch._functorch.conf... | 2.41.0 |
eb7b8eb6085322c96ba0bbacb039ef6f3b2875b | Wed, 1 May 2024 23:19:07 -0700 | [PATCH 0935/1000] [PT2D] Ensure the trace rules are correct with distributed (#125333) | Summary: 1. Avoid using `torch._dynamo.disable`. 2. Clear the LRU cache of the trace rules. This won't do anything if rules are not evluated before PG initilization. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125333 Approved by: https://github.com/yanboliang | diff --git a/torch/_dynamo/trace_rules.py b/torch/_dynamo/trace_rules.py
index ab6f52c7fb..954f3997e2 100644
--- a/torch/_dynamo/trace_rules.py
+++ b/torch/_dynamo/trace_rules.py
@@ -3541,3 +3541,11 @@ def lookup_inner(
return SkipFunctionVariable
else:
return UserFunctionVariable
+
+
+def clear_... | 2.41.0 |
a991fac221e8c03ed8924498e58c0ea48416a5c | Thu, 2 May 2024 17:16:02 +0000 | [PATCH 0936/1000] [ROCm][CI] upgrade CI to ROCm 6.1 (#124300) | Pull Request resolved: https://github.com/pytorch/pytorch/pull/124300 Approved by: https://github.com/malfet | diff --git a/.ci/docker/build.sh b/.ci/docker/build.sh
index 1b8ed8df93..a45516f46a 100755
--- a/.ci/docker/build.sh
+++ b/.ci/docker/build.sh
@@ -204,7 +204,7 @@ case "$image" in
PROTOBUF=yes
DB=yes
VISION=yes
- ROCM_VERSION=5.7
+ ROCM_VERSION=6.0
NINJA_VERSION=1.9.0
CONDA_CMAKE=yes
... | 2.41.0 |
741fb36803febc7e5db28e722746a872b3a4884 | Thu, 2 May 2024 08:53:52 -0700 | [PATCH 0937/1000] [DCP] Introduce async staging extension points (#122965) | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #124944 * #124939 * __->__ #122965 Differential Revision: [D55493240](https://our.internmc.facebook.com/intern/diff/D55493240/) *This PR is now ready for merge and is not an RFC* Major choices are: -- the introduction of the AsyncStager p... | diff --git a/docs/source/distributed.checkpoint.rst b/docs/source/distributed.checkpoint.rst
index 86f69f9d22..e948b66e1e 100644
--- a/docs/source/distributed.checkpoint.rst
+++ b/docs/source/distributed.checkpoint.rst
@@ -29,6 +29,12 @@ The entrypoints to load and save a checkpoint are the following:
.. autofunction:... | 2.41.0 |
99f1460af159a1417bb3ef1f088347cb8d89434 | Thu, 2 May 2024 08:53:54 -0700 | [PATCH 0938/1000] [DCP] Provides default AsyncStager (#124939) | Differential Revision: [D56575987](https://our.internmc.facebook.com/intern/diff/D56575987/) Pull Request resolved: https://github.com/pytorch/pytorch/pull/124939 Approved by: https://github.com/fegin ghstack dependencies: #122965 | diff --git a/docs/source/distributed.checkpoint.rst b/docs/source/distributed.checkpoint.rst
index e948b66e1e..573faa429b 100644
--- a/docs/source/distributed.checkpoint.rst
+++ b/docs/source/distributed.checkpoint.rst
@@ -36,6 +36,9 @@ The following module is also useful for additional customization of the staging
..... | 2.41.0 |
bbfb708315137ad709656f7df76dc7871346f59 | Thu, 2 May 2024 20:30:49 +0000 | [PATCH 0939/1000] ignore unsupported module from flop counter (#125346) | Summary: Torchscript modules do not support forward hooks and thus can't work with flop_counter context manager for hierarchical output by passing a module to FlopCounterMode on construction. Currently any module that includes a script module causes an exception to be thrown so adding a try/catch to ignore any script ... | diff --git a/torch/utils/flop_counter.py b/torch/utils/flop_counter.py
index 42868dc359..09bb039ea6 100644
--- a/torch/utils/flop_counter.py
+++ b/torch/utils/flop_counter.py
@@ -388,11 +388,19 @@ class FlopCounterMode(TorchDispatchMode):
else:
name = ".".join([prefix, name])
- ... | 2.41.0 |
f62494bf9c6915189ffd2ec1dbef0c5dd8b91ae | Thu, 2 May 2024 08:53:56 -0700 | [PATCH 0940/1000] [DCP] Move async logic into filesystem for better encapsulation (#124944) | This logic is specific to FilesystemWriter, and now has a better place to live due to the new AsyncStager class Differential Revision: [D56578436](https://our.internmc.facebook.com/intern/diff/D56578436/) Pull Request resolved: https://github.com/pytorch/pytorch/pull/124944 Approved by: https://github.com/fegin ghsta... | diff --git a/torch/distributed/checkpoint/filesystem.py b/torch/distributed/checkpoint/filesystem.py
index 3672e2401b..07bc85560e 100644
--- a/torch/distributed/checkpoint/filesystem.py
+++ b/torch/distributed/checkpoint/filesystem.py
@@ -29,8 +29,8 @@ import torch
from torch import Tensor
from torch._utils import _g... | 2.41.0 |
ae574c7133de4ee36b1802f1ac113be6d644c1e | Thu, 2 May 2024 11:22:54 -0400 | [PATCH 0941/1000] Don't make replacements for i variables (#125398) | This was introduced in https://github.com/pytorch/pytorch/pull/110262 but actually it looks like they were trying to hit unbacked SymInt. Now that unbacked SymInt is renamed to u, this code is no longer necessary Signed-off-by: Edward Z. Yang <ezyang@meta.com> Pull Request resolved: https://github.com/pytorch/pytorch... | diff --git a/torch/_inductor/codegen/common.py b/torch/_inductor/codegen/common.py
index 7126d565cf..a044fd3210 100644
--- a/torch/_inductor/codegen/common.py
+++ b/torch/_inductor/codegen/common.py
@@ -1683,7 +1683,6 @@ class Kernel(CodeGen):
x: self.args.size(x)
for x in sorted_symbols
... | 2.41.0 |
c76764a567c3e2425432671de04c4cfc705b745 | Thu, 2 May 2024 20:43:17 +0000 | [PATCH 0942/1000] Always pass down kernel_file and grid as string (#125384) | From my test with Ads production workload, I found sometime kernel_file is None and grid is a tuple. It will crash since ExecutionTraceObserver expects string for both kernel_file and grid. This PR is to make sure kernel_file and grid are always passed down as string. Need to find the root cause why kernel_file is non... | diff --git a/torch/_inductor/runtime/triton_heuristics.py b/torch/_inductor/runtime/triton_heuristics.py
index 6d49c60b0d..b66bdbf393 100644
--- a/torch/_inductor/runtime/triton_heuristics.py
+++ b/torch/_inductor/runtime/triton_heuristics.py
@@ -794,14 +794,15 @@ class CachingAutotuner(KernelInterface):
# man... | 2.41.0 |
199ce8d6c7149fb4ba9e1d92cb133abe4ac0667 | Tue, 30 Apr 2024 17:04:50 -0700 | [PATCH 0944/1000] [pipelining] Add microbatch split and merge utils (#125273) | Pull Request resolved: https://github.com/pytorch/pytorch/pull/125273 Approved by: https://github.com/H-Huang ghstack dependencies: #124776, #124875, #124958 | diff --git a/docs/source/distributed.pipelining.rst b/docs/source/distributed.pipelining.rst
index ec7423f261..0083fe4226 100644
--- a/docs/source/distributed.pipelining.rst
+++ b/docs/source/distributed.pipelining.rst
@@ -176,3 +176,16 @@ Note that since we split our model into three stages, we must run this script wi... | 2.41.0 |
dad82fc906f7e738bca1f918a97ccbc534c43ab | Thu, 2 May 2024 21:28:23 +0000 | [PATCH 0945/1000] Add private helper for determining which version of FA2 closest matches kernel version (#123653) | Fixes #ISSUE_NUMBER Pull Request resolved: https://github.com/pytorch/pytorch/pull/123653 Approved by: https://github.com/mikaylagawarecki | diff --git a/torch/nn/attention/__init__.py b/torch/nn/attention/__init__.py
index fca8055ad2..fc4835f046 100644
--- a/torch/nn/attention/__init__.py
+++ b/torch/nn/attention/__init__.py
@@ -115,3 +115,8 @@ def sdpa_kernel(backends: Union[List[SDPBackend], SDPBackend]):
enable_flash_sdp(previous_flash)
... | 2.41.0 |
1b03992d02911ac59f8320ba08eed8bf002ad84 | Thu, 2 May 2024 21:29:28 +0000 | [PATCH 0946/1000] Merge the pyi files into py files of optimizer (#125153) | Merge the interfaces in pyi files into py files in `torch/optim`. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125153 Approved by: https://github.com/janeyx99 | diff --git a/torch/optim/adadelta.py b/torch/optim/adadelta.py
index 4061a6b68f..d0cf2b32dd 100644
--- a/torch/optim/adadelta.py
+++ b/torch/optim/adadelta.py
@@ -1,4 +1,4 @@
-from typing import List, Optional
+from typing import Any, Dict, List, Optional
import torch
from torch import Tensor
@@ -13,6 +13,7 @@ from... | 2.41.0 |
0e2f62edd79ebdc21c3b8c16bf480db1af168a0 | Thu, 2 May 2024 21:36:18 +0000 | [PATCH 0947/1000] =?UTF-8?q?Revert=20"Include=20support=20for=20t?= =?UTF-8?q?he=20scatter=20gather=20cuda=20kernels=20to=20allow=20for=20com?= =?UTF-8?q?p=E2=80=A6=20(#124809)"?=MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | This reverts commit 9e24c263f998819f849bb8293323213101e9aefc. Reverted https://github.com/pytorch/pytorch/pull/124809 on behalf of https://github.com/kit1980 due to breaking internal builds ([comment](https://github.com/pytorch/pytorch/pull/124809#issuecomment-2091751002)) | diff --git a/aten/src/ATen/NumericUtils.h b/aten/src/ATen/NumericUtils.h
index 421fe0efab..788da64b4e 100644
--- a/aten/src/ATen/NumericUtils.h
+++ b/aten/src/ATen/NumericUtils.h
@@ -38,11 +38,7 @@ inline C10_HOST_DEVICE bool _isnan(T val) {
template <typename T, std::enable_if_t<c10::is_complex<T>::value, int> = 0>... | 2.41.0 |
c8237c6aa32ab7470e76bde03f2d3dcb9dd42a1 | Thu, 2 May 2024 22:26:39 +0000 | [PATCH 0949/1000] [ATen-VK] Resolve compiler_flags to allow Mac build (#125361) | Summary: ## `-Wmissing-prototypes` In ATen-Vulkan, we often define functions in `.cpp` files without declaring them in `.h` files without hiding them in an anonymous namespace. Example: [`Packing.cpp`'s channel_image_repacking()](https://github.com/pytorch/pytorch/blob/f1f142c44f81384afbdba5e451fc15744868bf26/aten/sr... | diff --git a/aten/src/ATen/native/vulkan/api/Adapter.cpp b/aten/src/ATen/native/vulkan/api/Adapter.cpp
index ff1ea41cf6..173479a0c2 100644
--- a/aten/src/ATen/native/vulkan/api/Adapter.cpp
+++ b/aten/src/ATen/native/vulkan/api/Adapter.cpp
@@ -44,10 +44,10 @@ PhysicalDevice::PhysicalDevice(VkPhysicalDevice physical_devi... | 2.41.0 |
b5f6b10add7515064f20dace8f80b6226d26ada | Thu, 2 May 2024 22:51:03 +0000 | [PATCH 0951/1000] [Inductor] default block size for head_dim = 256 for flex attention (#125380) | ## H100 ### torch.bfloat16 No major change, as expected. ``` | Type | Speedup | batch_size | num_heads | q_seq_len | k_seq_len | head_dim | score_mod | dtype | |---------|-----------|--------------|-------------|-------------|-------------|------------|-------------|----------------| | Average... | diff --git a/benchmarks/transformer/score_mod.py b/benchmarks/transformer/score_mod.py
index e337049707..0e9e8d11a3 100644
--- a/benchmarks/transformer/score_mod.py
+++ b/benchmarks/transformer/score_mod.py
@@ -211,7 +211,7 @@ def generate_experiment_configs() -> List[ExperimentConfig]:
batch_sizes = [1, 8, 16]
... | 2.41.0 |
551755cec99adef3b9210fd85996fc304bed349 | Thu, 2 May 2024 23:34:18 +0000 | [PATCH 0952/1000] Update tolerance for flex fp32 (#125444) | # Summary Updates the tolerances to account for internal failure Pull Request resolved: https://github.com/pytorch/pytorch/pull/125444 Approved by: https://github.com/kit1980 | diff --git a/test/inductor/test_flex_attention.py b/test/inductor/test_flex_attention.py
index e5f31e7bcb..e5aa2aeb86 100644
--- a/test/inductor/test_flex_attention.py
+++ b/test/inductor/test_flex_attention.py
@@ -143,7 +143,7 @@ class TestTemplatedSDPA(InductorTestCase):
# Note, it seems like we really are l... | 2.41.0 |
440d0755ab4659c450de1bc901839e675f253ad | Thu, 2 May 2024 23:44:09 +0000 | [PATCH 0953/1000] Support custom layout call under torch dispatch mode (#125379) | Fixes #ISSUE_NUMBER Pull Request resolved: https://github.com/pytorch/pytorch/pull/125379 Approved by: https://github.com/jbschlosser | diff --git a/test/test_nestedtensor.py b/test/test_nestedtensor.py
index 3df4b63801..31ea10a930 100644
--- a/test/test_nestedtensor.py
+++ b/test/test_nestedtensor.py
@@ -3878,6 +3878,14 @@ class TestNestedTensorSubclass(TestCase):
self.assertTrue(not nt_noncontiguous.is_contiguous(memory_format=torch.contiguo... | 2.41.0 |
18a6f46d01db665cf092d501fb9597cdac91a33 | Fri, 3 May 2024 00:50:49 +0000 | [PATCH 0954/1000] Adding Compare in torch.utils.benchmark documentation (#125009) | `torch.utils.benchmark.Compare` is not directly exposed in torch.utils.benchmark documentation. I think this is a valuable resource to add since it can help people embracing the torch benchmark way of doing things, and help people building documentation towards it. Pull Request resolved: https://github.com/pytorch/py... | diff --git a/docs/source/benchmark_utils.rst b/docs/source/benchmark_utils.rst
index c93fbfd66c..7546179c50 100644
--- a/docs/source/benchmark_utils.rst
+++ b/docs/source/benchmark_utils.rst
@@ -19,6 +19,9 @@ Benchmark Utils - torch.utils.benchmark
.. autoclass:: FunctionCounts
:members:
+.. autoclass:: Compare... | 2.41.0 |
cac7aa70ffc62090ab12211c36a3ac37e73ed0b | Fri, 3 May 2024 01:18:52 +0000 | [PATCH 0955/1000] [CI] Unskip Linalg tests on ARM (#125377) | Removes obscure "Issue with numpy version on arm" added by https://github.com/pytorch/pytorch/pull/82213 And replaces it with 4 targeted skips: - test_addmv for `float16` - test_vector_norm for `float16`, `bfloat16` and `float32` Followups to fix them are tracked in https://github.com/pytorch/pytorch/issues/125438 Pul... | diff --git a/test/test_linalg.py b/test/test_linalg.py
index 8976d81c5a..0a5577d54c 100644
--- a/test/test_linalg.py
+++ b/test/test_linalg.py
@@ -53,7 +53,6 @@ def blaslt_supported_device():
return True
return False
-@unittest.skipIf(IS_ARM64, "Issue with numpy version on arm")
class TestLinalg(Te... | 2.41.0 |
15da7856caf2deacbf0d9adabd53e5e3f0d9a1d | Fri, 3 May 2024 01:19:21 +0000 | [PATCH 0956/1000] [MPS] Fix overflow in cumsum when dtype is bool (#125318) | `cumsum` and `cumprod` was (is?) buggy for MPS: https://github.com/pytorch/pytorch/blob/c8d2a55273757c90989fde7c6f05e957aba9a238/aten/src/ATen/native/mps/operations/UnaryOps.mm#L435-L436 A workaround casts the input to int32 prior to performing the op to prevent overflow for certain numeric types. It turns out this i... | diff --git a/aten/src/ATen/native/mps/operations/UnaryOps.mm b/aten/src/ATen/native/mps/operations/UnaryOps.mm
index c76a60e4a4..7709c79db6 100644
--- a/aten/src/ATen/native/mps/operations/UnaryOps.mm
+++ b/aten/src/ATen/native/mps/operations/UnaryOps.mm
@@ -434,7 +434,7 @@ static void cumulative_op_impl(const Tensor& ... | 2.41.0 |
40d6df448de1acb263ed8f6ff9e7d26f5a1a161 | Fri, 3 May 2024 03:50:55 +0000 | [PATCH 0957/1000] [MPS] Native nonzero implementation (#125355) | Fixes https://github.com/pytorch/pytorch/issues/124850 Replace previous MPSGraph nonzero construction with native nonzero op. For older OSes, fallback to CPU (previous implementation was not reliable and was comparable to CPU in speed). Pull Request resolved: https://github.com/pytorch/pytorch/pull/125355 Approved by... | diff --git a/aten/src/ATen/native/mps/operations/Indexing.mm b/aten/src/ATen/native/mps/operations/Indexing.mm
index d86f57c49f..38c3212500 100644
--- a/aten/src/ATen/native/mps/operations/Indexing.mm
+++ b/aten/src/ATen/native/mps/operations/Indexing.mm
@@ -241,14 +241,20 @@ static void index_put_kernel_mps(TensorIter... | 2.41.0 |
156cb2e1279bd44d140fdb469d41dda5d0c40c2 | Fri, 3 May 2024 04:42:38 +0000 | [PATCH 0958/1000] Fix mem size mismatch from split/chunk in const folding (#125199) | Summary: The chunk/split ops on the weights/constants is folded in a fx pass and each output tensor has the same storage size of the original tensor (which is 3x of its actual size if chunk(3)). However Backend calculates the mem size on device from tensor shape/stride/dtype. This causes the mismatch when copying weig... | diff --git a/torch/fx/experimental/const_fold.py b/torch/fx/experimental/const_fold.py
index 548d1d3852..8176ccb562 100644
--- a/torch/fx/experimental/const_fold.py
+++ b/torch/fx/experimental/const_fold.py
@@ -60,7 +60,7 @@ class FoldedGraphModule(torch.fx.GraphModule):
def _create_param(i):
re... | 2.41.0 |
706da2bad835f521a1062b80aa6c631730a2064 | Thu, 2 May 2024 16:29:05 -0700 | [PATCH 0959/1000] [dynamo][cpp-guards] Improve recompilation reason logic for NO_TENSOR_ALIASING guard (#125439) | Pull Request resolved: https://github.com/pytorch/pytorch/pull/125439 Approved by: https://github.com/williamwen42 | diff --git a/torch/_dynamo/guards.py b/torch/_dynamo/guards.py
index fb4cb7a039..c04d114336 100644
--- a/torch/_dynamo/guards.py
+++ b/torch/_dynamo/guards.py
@@ -132,6 +132,7 @@ class GuardManager:
self.cache_entry = None
self.extra_state = None
self.id_matched_objs = None
+ self.no_t... | 2.41.0 |
f757a5c004641c7ef4781aad4a20196e7dd8b9e | Fri, 3 May 2024 04:59:17 +0000 | [PATCH 0961/1000] [export] use tree_map for _flatten_dynamic_shapes (#125415) | Summary: Fixing the implementation of `_flatten_dynamic_shapes()`, to follow how `_process_dynamic_shapes()` does it. The previous implementation would misinterpret some nested dynamic shapes specs, causing it to miss out on some shapes specs, for example with nested inputs/constant input tuples: ``` inputs = ( (2, 1)... | diff --git a/test/export/test_export.py b/test/export/test_export.py
index c7b6f53aaf..8aede91977 100644
--- a/test/export/test_export.py
+++ b/test/export/test_export.py
@@ -4561,6 +4561,35 @@ def forward(self, x):
self.assertEqual(div_spec.arg.name, "div")
self.assertEqual(div_spec.arg.value, "floor... | 2.41.0 |
71ee40793c21858cc7ccb9855f930549c35777d | Thu, 2 May 2024 16:29:05 -0700 | [PATCH 0962/1000] [dynamo][nn module] Check for duplicate tensors in register_attr_or_module (#125421) | Pull Request resolved: https://github.com/pytorch/pytorch/pull/125421 Approved by: https://github.com/mlazos ghstack dependencies: #125439 | diff --git a/torch/_dynamo/output_graph.py b/torch/_dynamo/output_graph.py
index 0be89d59e2..7e0d2e6197 100644
--- a/torch/_dynamo/output_graph.py
+++ b/torch/_dynamo/output_graph.py
@@ -767,22 +767,32 @@ class OutputGraph:
# are registered as get_attr nodes in the root graph.
tracer =... | 2.41.0 |
c8478974367753eb1ff3c82dbd8922193ab8416 | Fri, 3 May 2024 05:55:25 +0000 | [PATCH 0963/1000] [vision hash update] update the pinned vision hash (#123227) | This PR is auto-generated nightly by [this action](https://github.com/pytorch/pytorch/blob/main/.github/workflows/nightly.yml). Update the pinned vision hash. Pull Request resolved: https://github.com/pytorch/pytorch/pull/123227 Approved by: https://github.com/pytorchbot | diff --git a/.github/ci_commit_pins/vision.txt b/.github/ci_commit_pins/vision.txt
index 15cb85caca..50412178ab 100644
--- a/.github/ci_commit_pins/vision.txt
+++ b/.github/ci_commit_pins/vision.txt
@@ -1 +1 @@
-2c4665ffbb64f03f5d18016d3398af4ac4da5f03
+06ad737628abc3a1e617571dc03cbdd5b36ea96a | 2.41.0 |
d92637f445d2787f83829079276f71b1ad1fc7c | Thu, 2 May 2024 21:19:48 -0700 | [PATCH 0964/1000] Add `write_record_metadata` to PyTorchFileWriter (#125184) | Add `PyTorchFileWriter.write_record_metadata(record_name, num_bytes)` that - writes the zipfile header/end of central directory metadata for an entry* - reserves `num_bytes` in the zipfile for the payload. *Since the payload is not provided, the CRC32 computation is skipped and 0s are written in the corresponding entr... | diff --git a/caffe2/serialize/inline_container.cc b/caffe2/serialize/inline_container.cc
index 533fd42a04..173153e805 100644
--- a/caffe2/serialize/inline_container.cc
+++ b/caffe2/serialize/inline_container.cc
@@ -612,15 +612,35 @@ size_t ostream_write_func(
return ret;
}
+// This func will not update combined_u... | 2.41.0 |
5cc7ada6706f737959d8488d96028b3eb29aeea | Thu, 2 May 2024 20:50:18 -0700 | [PATCH 0965/1000] skip triton template precompilation in 311.0-3.11.7 to workaround 311 cpython bug (#125446) | Fix for https://github.com/pytorch/pytorch/issues/125374. We dont have CI for this specific versions, but I verified locally. THere is a cpython bug from 3.11.0->3.11.7 where the ast parsing state is global, and errors with multiple threads. when dust settles a little around the new process based compilation we can loo... | diff --git a/torch/_inductor/select_algorithm.py b/torch/_inductor/select_algorithm.py
index 8fcb441ed8..23f73c6207 100644
--- a/torch/_inductor/select_algorithm.py
+++ b/torch/_inductor/select_algorithm.py
@@ -976,6 +976,14 @@ class AlgorithmSelectorCache(PersistentCache):
if num_workers <= 0:
... | 2.41.0 |
ebefcf84520e46e4b156a9426ef76c6e110668d | Fri, 3 May 2024 09:10:11 +0000 | [PATCH 0966/1000] Driver folder check (#117548) | Added extra check for driver folders for Libtorch, as stat struct does not recognize driver folders, so torch.save should work for them as well. (e.g. save model.pt directly under C: ) Fixes [#111121](https://github.com/pytorch/pytorch/issues/111121) and #105488 Co-authored-by: Ozan Aydin <148207261+ozanMSFT@users.no... | diff --git a/caffe2/serialize/inline_container.cc b/caffe2/serialize/inline_container.cc
index 173153e805..d3bba2c797 100644
--- a/caffe2/serialize/inline_container.cc
+++ b/caffe2/serialize/inline_container.cc
@@ -93,7 +93,15 @@ static std::string parentdir(const std::string& name) {
end = name.find_last_of('\\')... | 2.41.0 |
89b4586e95752dc65a1821a4383b9679ccd5b6b | Fri, 3 May 2024 13:57:32 +0800 | [PATCH 0967/1000] [optim]fix ut and sgd kernel (#124904) | - Original `test_grad_scaling_autocast_fused_optimizers` does not work since there is no "fused" in `optim_inputs` - We should use different `grad_scaler`, they should not share 1 `scale`, there is no issue exposed here because the default `_growth_interval` is 2000 so it will not growth and there is also no inf is fou... | diff --git a/aten/src/ATen/native/cpu/FusedSGDKernel.cpp b/aten/src/ATen/native/cpu/FusedSGDKernel.cpp
index 3383585675..c19aa249a1 100644
--- a/aten/src/ATen/native/cpu/FusedSGDKernel.cpp
+++ b/aten/src/ATen/native/cpu/FusedSGDKernel.cpp
@@ -52,8 +52,8 @@ typename std::enable_if<
grad_vec2 = grad_vec2 * fVec(op... | 2.41.0 |
a98c2a932132e49559bf777c02798633d585e66 | Fri, 3 May 2024 10:24:14 +0000 | [PATCH 0968/1000] inductor: Add Conv3d support (#124361)MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | This PR is to add Conv3d support in inductor. Basicly reuse and expand Conv2d logic and unit tests to Conv3d. Conv3d inductor support will improve the performance of C2D_R50, I3D_R50, I3D_R101, Slow and SlowFast-R50 from OOB models. | C2D_R50 | I3D_R50 | I3D_R101 | Slow | SlowFast-R50 -- | -- | -- | -- | -- | -- eage... | diff --git a/aten/src/ATen/native/mkldnn/MKLDNNConversions.cpp b/aten/src/ATen/native/mkldnn/MKLDNNConversions.cpp
index ea5464f285..b2901bc522 100644
--- a/aten/src/ATen/native/mkldnn/MKLDNNConversions.cpp
+++ b/aten/src/ATen/native/mkldnn/MKLDNNConversions.cpp
@@ -198,24 +198,40 @@ Tensor mkldnn_reorder_conv3d_weight... | +- func: mkldnn_reorder_conv3d_weight(Tensor self, SymInt[3] padding=0, SymInt[3] stride=1, SymInt[3] dilation=1, SymInt groups=1, SymInt[]? input_size=None) -> Tensor |
9af8143692571c7dec07465fe93605f82815ce8 | Thu, 2 May 2024 18:04:01 -0700 | [PATCH 0969/1000] [FSDP] Added private `_unshard` API (#124304)MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | Some toy example: <img width="998" alt="Screenshot 2024-04-17 at 2 00 05 PM" src="https://github.com/pytorch/pytorch/assets/31054793/b5665a63-beb0-4ca1-92c6-c57a052812fd"> We define `FullyShardedDataParallel._unshard(async_op: bool = False)` that can be used to prefetch all-gathers. The user should make sure: 1. Run l... | diff --git a/test/distributed/fsdp/test_fsdp_comm.py b/test/distributed/fsdp/test_fsdp_comm.py
index 2e5993b23e..c20637061d 100644
--- a/test/distributed/fsdp/test_fsdp_comm.py
+++ b/test/distributed/fsdp/test_fsdp_comm.py
@@ -3,18 +3,23 @@
import sys
from contextlib import nullcontext
from enum import auto, Enum
-f... | 2.41.0 |
99ada5b27a1032e4bda4849b34411031d893b5e | Fri, 3 May 2024 14:54:15 +0000 | [PATCH 0970/1000] call `super().__post_init__` in `ForeachFuncinfo.__post_init__` (#125457) | obviously the current main branch's `ForeachFuncInfo`'s dunder post init doesn't `super().__post_init__()` which does some setup including setting `dtypesIfCUDA` and `dtypesIfROCM`. Fixes #125295 related: #125001 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125457 Approved by: https://github.com/jan... | diff --git a/torch/testing/_internal/common_methods_invocations.py b/torch/testing/_internal/common_methods_invocations.py
index 182bf42cce..001d93de18 100644
--- a/torch/testing/_internal/common_methods_invocations.py
+++ b/torch/testing/_internal/common_methods_invocations.py
@@ -9479,7 +9479,6 @@ foreach_unary_op_db... | 2.41.0 |
6052a35d462ed783a32d37f5278cab2f5cfa17d | Thu, 2 May 2024 07:38:49 -0700 | [PATCH 0975/1000] [RFC][FSDP2] Added `register_fsdp_forward_method` for user fwd methods (#125394) | FSDP only runs its pre/post-forward hooks on `nn.Module.forward`. This means that if the user runs a custom method meant as a forward pass, then FSDP will not all-gather the parameters. Examples include HuggingFace models' `generate()` (https://github.com/pytorch/pytorch/issues/123962, https://github.com/pytorch/pytorc... | diff --git a/test/distributed/_composable/fsdp/test_fully_shard_training.py b/test/distributed/_composable/fsdp/test_fully_shard_training.py
index bf90bd165c..8d1e1dbfb6 100644
--- a/test/distributed/_composable/fsdp/test_fully_shard_training.py
+++ b/test/distributed/_composable/fsdp/test_fully_shard_training.py
@@ -1... | 2.41.0 |
84a5b6cc0ecf0d75dbdbe438f2dee4909dc8db4 | Wed, 1 May 2024 13:02:42 -0700 | [PATCH 0976/1000] [AOTI] Add missing std::move for constant args (#125329) | Summary: fix https://github.com/pytorch/pytorch/issues/123187 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125329 Approved by: https://github.com/angelayi, https://github.com/chenyang78 | diff --git a/test/inductor/test_cpu_cpp_wrapper.py b/test/inductor/test_cpu_cpp_wrapper.py
index 828ed8eb0e..c28451613c 100644
--- a/test/inductor/test_cpu_cpp_wrapper.py
+++ b/test/inductor/test_cpu_cpp_wrapper.py
@@ -88,7 +88,6 @@ if config.abi_compatible:
"test_qlinear_cpu",
"test_qlinear_dequant_p... | 2.41.0 |
1a7455b996a2bed2f716f54315509b6b9a6dcce | Fri, 3 May 2024 17:19:43 +0200 | [PATCH 0978/1000] [Inductor cutlass backend] Fix cutlass_utils.get_max_alignment() for strided layouts. (#124930) | Fixes cutlass_utils.get_max_alignment() which was so far not checking the alignment properly. Basically the method so far assumed that the passed layout is contiguous and row-major, which does not have to be true. Test Plan: CI - test_cutlass_backend.py to prevent regressions Added unit test Pull Request resolved: ht... | diff --git a/test/inductor/test_cutlass_backend.py b/test/inductor/test_cutlass_backend.py
index 7da02e7077..aed07919a8 100644
--- a/test/inductor/test_cutlass_backend.py
+++ b/test/inductor/test_cutlass_backend.py
@@ -9,7 +9,8 @@ import torch
from torch._dynamo.utils import counters
from torch._inductor import confi... | 2.41.0 |
45baef05da3b2624382b9c3242233bba2a61c83 | Fri, 3 May 2024 20:52:01 +0000 | [PATCH 0979/1000] s390x: remove workaround for sleef issue (#124730) | This workaround is no longer needed since sleef was updated. Pull Request resolved: https://github.com/pytorch/pytorch/pull/124730 Approved by: https://github.com/soulitzer | diff --git a/aten/src/ATen/cpu/vec/vec256/zarch/vec256_zarch.h b/aten/src/ATen/cpu/vec/vec256/zarch/vec256_zarch.h
index 9b53745b03..b70b494649 100644
--- a/aten/src/ATen/cpu/vec/vec256/zarch/vec256_zarch.h
+++ b/aten/src/ATen/cpu/vec/vec256/zarch/vec256_zarch.h
@@ -13,8 +13,6 @@
#include <ATen/cpu/vec/vec_base.h>
#i... | 2.41.0 |
941fee7ea6d069cd79c420dbe5528decb86ac3e | Fri, 3 May 2024 21:10:36 +0000 | [PATCH 0980/1000] [CPP extention] Baton lock is called regardless the code version (#125404) | Greetings! Fixes #125403 Please assist me with the testing as it is possible for my reproducer to miss the error in the code. Several (at least two) threads should enter the same part of the code at the same time to check file lock is actually working Pull Request resolved: https://github.com/pytorch/pytorch/pull/12... | diff --git a/torch/utils/cpp_extension.py b/torch/utils/cpp_extension.py
index 8961dfcae3..ee5584315e 100644
--- a/torch/utils/cpp_extension.py
+++ b/torch/utils/cpp_extension.py
@@ -1691,10 +1691,10 @@ def _jit_compile(name,
file=sys.stderr)
name = f'{name}_v{version}'
- if version != ... | 2.41.0 |
a578df57cc0f417f671634e564c62ef5d9a97e2 | Fri, 3 May 2024 11:19:09 -0700 | [PATCH 0981/1000] [FSDP2] Added test to show rank 0 broadcast for HSDP replicas (#125431) | This PR shows a simple utility to broadcast the parameters across replicas for HSDP: ``` replicate_group = mesh.get_group("replicate") for param in model.parameters(): # E.g. for mesh [[0, 1, 2, 3], [4, 5, 6, 7]] sharding on dim-1 and # replicating on dim-0, broadcast with sources 0, 1, 2, 3 src_rank = dist.get_process... | diff --git a/test/distributed/_composable/fsdp/test_fully_shard_init.py b/test/distributed/_composable/fsdp/test_fully_shard_init.py
index 2e2e97e3a3..958f375fe2 100644
--- a/test/distributed/_composable/fsdp/test_fully_shard_init.py
+++ b/test/distributed/_composable/fsdp/test_fully_shard_init.py
@@ -725,5 +725,72 @@ ... | 2.41.0 |
503c29357f8ec7a5f97779a83f3a752eeb1654e | Fri, 3 May 2024 11:13:52 -0700 | [PATCH 0982/1000] Introduce torch.utils._sympy.symbol (#125395) | This provides utilities for creating and querying properties on sympy.Symbol. I want to use this refactor to get a better handle on how the 's' prefix is being used in Inductor. To start, I only do symbolic_shapes code because that's what I'm familiar with. Signed-off-by: Edward Z. Yang <ezyang@meta.com> Pull Reque... | diff --git a/test/inductor/test_compiled_autograd.py b/test/inductor/test_compiled_autograd.py
index d0e6949450..4c52630f8f 100644
--- a/test/inductor/test_compiled_autograd.py
+++ b/test/inductor/test_compiled_autograd.py
@@ -1516,11 +1516,13 @@ def wrap_test_class(orig_cls):
elif name.startswith("test_"):
... | 2.41.0 |
b5ae2611e22d992565f202df9267fe66469efaa | Fri, 3 May 2024 21:34:34 +0000 | [PATCH 0983/1000] s390x: use runtime detection for vectorization support (#123936) | s390x: use runtime detection for vectorization support Pull Request resolved: https://github.com/pytorch/pytorch/pull/123936 Approved by: https://github.com/malfet, https://github.com/jansel, https://github.com/xuhancn | diff --git a/aten/src/ATen/native/DispatchStub.cpp b/aten/src/ATen/native/DispatchStub.cpp
index c7db94889c..93c004acdc 100644
--- a/aten/src/ATen/native/DispatchStub.cpp
+++ b/aten/src/ATen/native/DispatchStub.cpp
@@ -10,8 +10,19 @@
#include <cstdlib>
#include <cstring>
+#ifdef HAVE_ZVECTOR_CPU_DEFINITION
+#includ... | 2.41.0 |
783fef9904ea785ed1489cdcc0d4c7f55af4a83 | Fri, 3 May 2024 14:58:27 -0700 | [PATCH 0984/1000] [AOTI] Add a missing mypy ignore (#125508) | Summary: Caused by https://github.com/pytorch/pytorch/pull/125397, but somehow was not caught by CI. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125508 Approved by: https://github.com/izaitsevfb | diff --git a/torch/_inductor/ir.py b/torch/_inductor/ir.py
index f240efa7ae..49261cfd58 100644
--- a/torch/_inductor/ir.py
+++ b/torch/_inductor/ir.py
@@ -5547,7 +5547,7 @@ class FallbackKernel(ExternKernelAlloc):
self.set_cpp_kernel(kernel)
else:
self.cpp_kern... | 2.41.0 |
9abd1dccbbf6a20c76f82a2e9d7670126299a99 | Fri, 3 May 2024 22:58:20 +0000 | [PATCH 0986/1000] Fix lint after PR 122611 (#125512) | Fix lint after https://github.com/pytorch/pytorch/pull/122611 Pull Request resolved: https://github.com/pytorch/pytorch/pull/125512 Approved by: https://github.com/clee2000 | diff --git a/c10/util/Exception.h b/c10/util/Exception.h
index 55de996683..bc01ad8d4e 100644
--- a/c10/util/Exception.h
+++ b/c10/util/Exception.h
@@ -68,7 +68,10 @@ class C10_API Error : public std::exception {
const void* caller = nullptr);
// Base constructor
- Error(std::string msg, std::string backtra... | 2.41.0 |
2ab96a57e5b19c72cf11bc098fd690d238c1d86 | Wed, 1 May 2024 14:45:19 -0700 | [PATCH 0987/1000] [dynamo] fix crash when context manager is passed to a function (#125321) | Fix https://github.com/pytorch/pytorch/issues/125274. Main change was to reconstruct `ContextWrappingVariables` as objects in general, but we can replace them with the class on the caller side when generating the resume function. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125321 Approved by: https:... | diff --git a/test/dynamo/test_ctx_manager.py b/test/dynamo/test_ctx_manager.py
index cc6e39de4d..eab8fdb41a 100644
--- a/test/dynamo/test_ctx_manager.py
+++ b/test/dynamo/test_ctx_manager.py
@@ -1304,7 +1304,7 @@ class GraphModule(torch.nn.Module):
self.assertEqual(fn(x), opt_fn(x))
self.asser... | 2.41.0 |
96bb74077ddc4349ba928e2f6b011d7ed35058c | Fri, 3 May 2024 11:19:09 -0700 | [PATCH 0989/1000] [FSDP2] Added HSDP grad acc tests and some minor changes (#125479) | This adds HSDP to the existing gradient accumulation tests and includes some minor changes to simplify things a tiny bit. Pull Request resolved: https://github.com/pytorch/pytorch/pull/125479 Approved by: https://github.com/wanchaol ghstack dependencies: #125431 | diff --git a/.ci/pytorch/test.sh b/.ci/pytorch/test.sh
index c903a26998..72a3943039 100755
--- a/.ci/pytorch/test.sh
+++ b/.ci/pytorch/test.sh
@@ -322,6 +322,7 @@ test_inductor_distributed() {
pytest test/distributed/_composable/fsdp/test_fully_shard_training.py -k test_train_parity_2d_mlp
pytest test/distributed... | 2.41.0 |
aa7699185e4ec39077e3046dfd63244dffa9ddb | Fri, 3 May 2024 12:09:18 -0700 | [PATCH 0990/1000] [FSDP2] Computed grad divide factors at runtime (#125484) | **Context** We are interested in supporting the case where HSDP reduce-scatters but does not all-reduce in a microbatch backward. This saves communication while still saving memory. Only on the last microbatch do we need to both reduce-scatter and all-reduce. This is not implemented yet and will hopefully come in a fut... | diff --git a/test/distributed/_composable/fsdp/test_fully_shard_comm.py b/test/distributed/_composable/fsdp/test_fully_shard_comm.py
index 33c6e61ac4..115c1f9322 100644
--- a/test/distributed/_composable/fsdp/test_fully_shard_comm.py
+++ b/test/distributed/_composable/fsdp/test_fully_shard_comm.py
@@ -18,6 +18,8 @@ fro... | 2.41.0 |
a1af95b0979d85c4fe32a75e797323ad81f298d | Sat, 4 May 2024 00:10:53 +0000 | [PATCH 0991/1000] [Inductor] Properly package target info for triton.compile (#125241) | Triton updated the interface for `triton.compile` https://github.com/openai/triton/commit/5162346487b3e3ebc062d9697429bafad25f22f6 The `target` argument to compile needs to be wrapped in a `GPUTarget` object. Without proper wrapping, we hit an assert in `compile`. If that assert is removed, Triton attempts to read dev... | diff --git a/torch/_inductor/runtime/triton_heuristics.py b/torch/_inductor/runtime/triton_heuristics.py
index b66bdbf393..0dc43f0649 100644
--- a/torch/_inductor/runtime/triton_heuristics.py
+++ b/torch/_inductor/runtime/triton_heuristics.py
@@ -55,11 +55,17 @@ if triton is not None:
from triton.compiler.comp... | 2.41.0 |
325c55896648ccecc5cfec11d3220dd93f75f23 | Sat, 4 May 2024 00:29:35 +0000 | [PATCH 0992/1000] Add CUDA paths to `CODEOWNERS` (#125409) | CC @ptrblck @albanD Pull Request resolved: https://github.com/pytorch/pytorch/pull/125409 Approved by: https://github.com/albanD | diff --git a/CODEOWNERS b/CODEOWNERS
index 6999f8553b..e481e66112 100644
--- a/CODEOWNERS
+++ b/CODEOWNERS
@@ -144,3 +144,14 @@ caffe2/utils/hip @jeffdaily @jithunnair-amd
/torch/csrc/Storage* @mikaylagawarecki
# subscribing for PyTorchFileWriter/PyTorchFileReader changes
/torch/csrc/jit/python/init.cpp @mikaylagawa... | 2.41.0 |
fd5bb0c44dfaef36961eaa67008df4d6134199d | Fri, 3 May 2024 13:38:04 -0700 | [PATCH 0993/1000] [c10d] only PG0 should dump when monitoring thread timed out (#125356) | Summary: We found that some dumps are missing when monitoring thread timeout. This is likely due to multiple PGs could still dump the same records at the same time. So we should allow only PG0 to actualy dump Test Plan: unit test python test/run_test.py --cpp --verbose -i cpp/ProcessGroupNCCLErrorsTest Tags: Pull Requ... | diff --git a/test/cpp/c10d/ProcessGroupNCCLErrorsTest.cpp b/test/cpp/c10d/ProcessGroupNCCLErrorsTest.cpp
index 991a8c5dc7..aef97daae2 100644
--- a/test/cpp/c10d/ProcessGroupNCCLErrorsTest.cpp
+++ b/test/cpp/c10d/ProcessGroupNCCLErrorsTest.cpp
@@ -389,6 +389,7 @@ TEST_F(ProcessGroupNCCLErrorsTest, testNCCLErrorsNoHeartb... | 2.41.0 |
302dc68bf76a0af6dd4bb0488aaf22998374a0e | Sat, 4 May 2024 02:39:39 +0000 | [PATCH 0994/1000] =?UTF-8?q?[Reland]=20Fakify=20script=20object?= =?UTF-8?q?=20inputs=20and=20attributes=20for=20non-strict=20ex=E2=80=A6?= =?UTF-8?q?=20(#125490)?=MIME-Version: 1.0Content-Type: text/plain; charset=UTF-8Content-Transfer-Encoding: 8bit | A re-land of #124239. This PR fakify ScriptObject inputs and attributes in export non-strict mode by default. The basic idea is to only fakify the script object during tracing (i.e. aot_export). After we get the traced graph module, eagerly executing, serializing, or running more passes will use the real script objec... | diff --git a/test/export/test_passes.py b/test/export/test_passes.py
index 41597a6030..e2724ead88 100644
--- a/test/export/test_passes.py
+++ b/test/export/test_passes.py
@@ -13,6 +13,10 @@ from typing import List, Set
import torch
from functorch.experimental.control_flow import cond
from torch._dynamo.eval_frame im... | 2.41.0 |
4727fd4ebd42936a1bae7b7f44ee9a038fd643e | Sat, 4 May 2024 03:08:44 +0000 | [PATCH 0995/1000] [TD][ez] Better check for is pr or not (#125485) | You can trigger ciflow tags on main branch commits, so we should be more conservative when checking to see if a workflow is a PR/on the main branch. get_pr_number checks for the pr number based on the PR_NUMBER env var or a tag of the for `ciflow/workflow/pr number` If we fail to find something like this, then assume... | diff --git a/test/run_test.py b/test/run_test.py
index 1b95bcb465..5120ac6513 100755
--- a/test/run_test.py
+++ b/test/run_test.py
@@ -59,6 +59,7 @@ from tools.testing.discover_tests import (
)
from tools.testing.do_target_determination_for_s3 import import_results
from tools.testing.target_determination.gen_artifac... | 2.41.0 |
f061baa94f437e54c5f4a12f48bcb1dfa662918 | Fri, 3 May 2024 16:03:08 -0700 | [PATCH 0996/1000] [comm_mode] adding some initial c10d ops to CommDebugMode (#125475) | looks like we can make it work :) Pull Request resolved: https://github.com/pytorch/pytorch/pull/125475 Approved by: https://github.com/awgu | diff --git a/test/distributed/_tensor/debug/test_comm_mode.py b/test/distributed/_tensor/debug/test_comm_mode.py
index 893131fe4e..d674905f2b 100644
--- a/test/distributed/_tensor/debug/test_comm_mode.py
+++ b/test/distributed/_tensor/debug/test_comm_mode.py
@@ -9,11 +9,13 @@ from torch.distributed._tensor import Devic... | 2.41.0 |
62e89c1b8b27e848f30991773d689db3af24532 | Fri, 3 May 2024 11:47:00 -0700 | [PATCH 0997/1000] [dynamo] Do not turn on record relay with TORCH_COMPILE_DEBUG (#125488) | Pull Request resolved: https://github.com/pytorch/pytorch/pull/125488 Approved by: https://github.com/yanboliang, https://github.com/mlazos | diff --git a/torch/_dynamo/config.py b/torch/_dynamo/config.py
index 1360782db1..6a870a6d12 100644
--- a/torch/_dynamo/config.py
+++ b/torch/_dynamo/config.py
@@ -130,7 +130,7 @@ suppress_errors = bool(os.environ.get("TORCHDYNAMO_SUPPRESS_ERRORS", False))
# Record and write an execution record of the current frame to ... | 2.41.0 |
b41e1d6fc05428008875e3cfe8be17184e57491 | Fri, 3 May 2024 16:22:16 -0700 | [PATCH 0998/1000] try to fix the warning in distribute_tensor (#125476) | Pull Request resolved: https://github.com/pytorch/pytorch/pull/125476 Approved by: https://github.com/albanD, https://github.com/awgu ghstack dependencies: #125475 | diff --git a/test/distributed/_tensor/test_api.py b/test/distributed/_tensor/test_api.py
index 196bd6407b..4318054117 100644
--- a/test/distributed/_tensor/test_api.py
+++ b/test/distributed/_tensor/test_api.py
@@ -70,6 +70,7 @@ class DTensorAPITest(DTensorTestBase):
)
tensor_shape = [3 * self.world_s... | 2.41.0 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.