Commit Graph

9 Commits

Author SHA1 Message Date
jjsjann123 b21a6ff639 [NVFuser] Upstream push 0811 (#83239)
Syncing nvfuser devel branch to upstream master. https://github.com/csarofeen/pytorch/

Code changes includes:

- codegen improvements:
  1. double support in expression evaluator
- bug fixes:
  1. dropout fix - rework RNG to support broadcasted dropout (Fixes #82784)
  2. expand fix - Patch expand+reduction, expand+view, rework view analysis and guard
- scheduler:
  1. manual transpose schedule example
  2. WIP transpose scheduler

Commits that's in this PR from the devel branch:

```
b7435afcd2 Transpose scheduler, step 1 (#1854)
8a45dbf720 Add an example on how to manually schedule transpose (#1889)
83dbf56a95 Patch dropout fix (#1898)
69d3519a53 Expand+Reduction, Expand+View support, rework View analysis and guards (#1883)
15091c488e Rework RNG to correctly support broadcasted dropout (#1888)
aafe2d048a Make ExpressionEvaluator support Double (#1885)
```

RUN_TORCHBENCH: nvfuser

Differential Revision: [D38657074](https://our.internmc.facebook.com/intern/diff/D38657074)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/83239
Approved by: https://github.com/davidberard98
2022-08-25 02:23:22 +00:00
Richard Barnes b745e5f115 Check all CUDA API calls for errors in benchmarks/cpp/nvfuser (#74920) (#81817)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/74920

Test Plan: Sandcastle

Differential Revision: D35194656

Pull Request resolved: https://github.com/pytorch/pytorch/pull/81817
Approved by: https://github.com/malfet
2022-08-24 18:59:05 +00:00
jjsjann123 df741c589f [NVFuser] Upstream push 0809 (#83067)
Syncing nvfuser devel branch to upstream master. https://github.com/csarofeen/pytorch/

Code changes includes:

- codegen improvements:
  1. removes un-necessary sync from redundant thread compute analysis
  2. symmetric API for BestEffortReplay
  3. support merge on trivial reductions
  4. Ampere async copy improvements
- bug fixes:
  1. vectorization bug fixes
  2. type inference patch : fixes upstream #81725
  3. segmenter bug fix with deterministic iteration ordering
- parser update
  1. added leaky_relu
- scheduler
  1. normalization scheduler clean up.
  2. simplifies matmul scheduling with new transform propagator
  3. merge all dimensions in PW scheduler
  4. various gemm related improvements
- debuggability
  1. nsight compute support
  2. debug dump for InlinePropagator
  3. Add `UnaryOpType::Print`

Squashed commits to WAR github API
Commits that's actually in this PR from the devel branch:

```
dfe02f3fae Merge remote-tracking branch 'csarofeen/devel' into HEAD
16173732ec Add `TensorViewBuilder::shape(std::vector<Val*> shape)` (#1884)
7cfb7796bd Merge pull request #1887 from csarofeen/upstream_merge_0803
3399f6de62 Merge remote-tracking branch 'origin/viable/strict' into HEAD
01208f5bba Add `UnaryOpType::Print` which can be helpful for debugging (#1878)
0646522454 Remove redundant TORCH_INTERNAL_ASSERT in lower_magic_zero.cpp (#1881)
7bc76aa219 Fix most inlined propagator for mismatched dims (#1875)
501f4aa270 Nonaffine swizzle formulation ep.2: Loop swizzle variant. (#1826)
d863d690f9 Ampere async copy ep.2: circular buffering extension to support pipelined matmul operand load (#1827)
e0ae11a61c Larger sized mma instructions to support full vectorization (#1824)
9bb4cf7a66 fragment iteration to support fully unrolled mma ops (#1823)
a48270a18d Merge all dims in pointwise scheduler (#1872)
172fb3673f Make MostInlined and BestEffort inline propagation no longer assert replayed (#1868)
a64462a5ac Allow trivial reduction to be merged (#1871)
440102bcda Symmetric API for BestEffortReplay (#1870)
d1caf330c0 Some misc cleanups/refactor split out from #1854 (#1867)
1013eda50b Remove some welford specific logic. (#1864)
51589d36be Some cleanups on tests and heuristics params (#1866)
a6b3e70da5 Segmenter bug fix, and deterministic iteration ordering.  (#1865)
1b665b9b5e Add nullptr checks to IrBuilder (#1861)
1cd9451d74 Simplify matmul scheduling with the new transform propagator.  (#1817)
bbc1fb9b8c Add leaky_relu operation (#1852)
e842a9bab5 Minor cleanup in pointwise scheduler (#1858)
9ee850ca2f Fix stringstream usage (#1857)
20a36c1e4f Improve nsight compute support (#1855)
4059103083 Remove debugging `true ||` from getPointwiseHeuristics (#1822)
01117bfe8f Misc cleanup (#1853)
5cc64943dc Apply the magic-zero protection to each indexed domain individually for predicate indexing (#1846)
92e6f0207e Cleanup normalization scheduler (#1845)
db89c6591a Type inference patch (#1848)
102fe93a46 Add debug dump for InlinePropagator (#1847)
b7a4d93d37 Redundant thread compute analysis to avoid un-necessary sync insertion (#1687)
942be5b256 Upstream ci build fixes (#1842)
0b83645915 Fix vectorization bug introduced in #1831 (#1840)
63630f1ae0 Move MaxProducerPosUpdater into InlinePropagator::tearDown (#1825)
9135a963c0 Fix transpose benchmark dtype (#1839)
2c9a6c0231 Add extra configurability to `parallelizeAllLike` (#1831)
```

RUN_TORCHBENCH: nvfuser

Differential Revision: [D38543000](https://our.internmc.facebook.com/intern/diff/D38543000)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/83067
Approved by: https://github.com/davidberard98
2022-08-10 21:02:56 +00:00
jjsjann123 a2802ad0b9 Upstream master bump 0513 (#77471)
Updating nvfuser code base.

This should fix the indexing issue observed in https://github.com/pytorch/vision/issues/6015.

Running tests locally as well. Will update the description here at a later point

@bypass-github-export-checks
Pull Request resolved: https://github.com/pytorch/pytorch/pull/77471
Approved by: https://github.com/seemethere, https://github.com/eellison
2022-05-18 11:48:50 -07:00
jiej 2d110d514f Nvfuser code bump 2_1_2022 (#72127)
Summary:
Things changed in this PR that requires review:
1. aten/src/ATen/core/interned_strings.h
2. torch/csrc/jit/ir/alias_analysis.h : exposing createValue to allow efficient mutation
3. torch/csrc/jit/runtime/symbolic_shape_registry.cpp : added gelu/tanh/erf in registry
4. torch/jit/_script.py : throws scripting model sees autocast as decorator since it's not supported

nvfuser code update:
1. codegen improvements and performance tuning
2. integration bug fixes for shape expression logic
3. kernel segmentation update to address perf regression from horizontal fusion
4. scalar cpu tensor promotion to support inter-device operation between cpu scalar tensor and cuda tensor

Things reverted from local changes:
aten::gelu with approximation (tracked in PR: https://github.com/pytorch/pytorch/pull/61439)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/72127

Reviewed By: HamidShojanazeri

Differential Revision: D34113233

Pulled By: jbschlosser

fbshipit-source-id: b82cde32b71e324eca0ea57cb8c9f9647278ca74
(cherry picked from commit e009bc5c4e)
2022-02-15 00:43:16 +00:00
jjsjann123 0dc3f829d9 Nvfuser code bump 11 5 (#67943)
Summary:
nvfuser code update:
1. Tuning heuristics on schedulers for reduction/normalization kernels;
2. bfloat16 on IO tensor support;
3. Refactored memory format support, now we can support dimension collapsing with non-coherent input tensors with different memory format. e.g. channels last tensor input to batch normalization. Note that we are currently limiting memory format to only Contiguous and Channels last;
4. Refactored nvfuser graph partitioning in `graph_fuser.cpp`, separated node merge and profile node API. Updated `profiling_record.cpp`.

Things that are reverted from our local branch:
1. changes on some entries in autodiff
2. aten::gelu with approximation
3. native_dropout(_backward)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/67943

Reviewed By: ngimel

Differential Revision: D32288709

Pulled By: dzhulgakov

fbshipit-source-id: fc9491182ea7e0158bc112c66f096823c588eaf1
2021-11-17 01:22:17 -08:00
jiej 127c9402d0 Revert "Revert D30752939: [pytorch][PR] nvfuser update" (#65137)
Summary:
This reverts commit 03389dc851.

Attempt again for PR: https://github.com/pytorch/pytorch/issues/63745
Fixes the windows build failure.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/65137

Reviewed By: seemethere, dzhulgakov, heitorschueroff

Differential Revision: D30994556

Pulled By: malfet

fbshipit-source-id: f1925b6c5cc1a1a441a96499667c91e8dfc1b53d
2021-09-22 04:54:51 -07:00
Eli Uriegas 03389dc851 Revert D30752939: [pytorch][PR] nvfuser update
Test Plan: revert-hammer

Differential Revision:
D30752939 (https://github.com/pytorch/pytorch/commit/cfaecaf40bd6cabd3f4e0ef0d8c7252655349b61)

Original commit changeset: ce122e80f01b

fbshipit-source-id: 57685df8f9946032a06eff1de8a3d1498500d2d2
2021-09-15 17:38:47 -07:00
jiej cfaecaf40b nvfuser update (#63745)
Summary:
Syncing nvfuser code base from devel branch, Listing a few of our development since last sync:

- Extends support to normalization and reduction kernels.
- Multiple kernel launch for single `CudaFusionGroup`. Hierarchical caching system has been updated to cache graph segmentation.
- profile_ivalue is enabled to convert dynamic scalar into compile time constants, which are required by the codegen. (e.g. reduction axes).

To keep this PR simple and relatively review-free. We stripped most external changes and submitted them as separate PRs, so this gigantic PR is easier to handle.

internal updates are files located in:
1. updates in nvfuser codegen `torch/csrc/jit/coddgen/cuda`
2. added nvfuser specific benchmarks `benchmarks/cpp/nvfuser`
3. nvfuser jit cpp tests `test/cpp/jit/test_gpu.cpp` `test/cpp/jit/test_gpu_shift.cpp` `test/cpp/jit/test_gpu_validator.h`

updates affecting integration:

1. profile_ivalue enabled for nvfuser. related changes are in `torch/csrc/jit/runtime/*`,
2. exposed a few more symbols `aten/src/ATen/core/*` used by codegen

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63745

Reviewed By: saketh-are

Differential Revision: D30752939

Pulled By: malfet

fbshipit-source-id: ce122e80f01bcd3865f5bd3c4dfde660665fd84c
2021-09-15 14:42:55 -07:00