[PyTorch] Specialize Vectorized<Half> for NEON even if FP16 arithmetic isn't available #137911
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…c isn't available We can do most of what this header does (by line count) anyway by converting to and from float. Differential Revision: [D64265757](https://our.internmc.facebook.com/intern/diff/D64265757/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/137911
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 6c24f9c with merge base b9618c9 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
module: cpu
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D64265757 |
This was referenced Oct 14, 2024
…6 arithmetic isn't available" We can do most of what this header does (by line count) anyway by converting to and from float. Differential Revision: [D64265757](https://our.internmc.facebook.com/intern/diff/D64265757/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D64265757 |
…6 arithmetic isn't available" We can do most of what this header does (by line count) anyway by converting to and from float. Differential Revision: [D64265757](https://our.internmc.facebook.com/intern/diff/D64265757/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D64265757 |
Contributor
Author
|
(it's hard to succinctly explain what the benefits of this one are, so I gave it a not-user-facing for release notes) |
swolchok
requested review from
jgong5 and
malfet
and removed request for
malfet
…6 arithmetic isn't available" We can do most of what this header does (by line count) anyway by converting to and from float. Differential Revision: [D64265757](https://our.internmc.facebook.com/intern/diff/D64265757/) cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D64265757 |
This was referenced Oct 28, 2024
Contributor
Author
this failure doesn't seem to repro locally on a linux machine, nor does it pass the sniff test because this diff only affects ARM and it's on Windows... |
…6 arithmetic isn't available" We can do most of what this header does (by line count) anyway by converting to and from float. Differential Revision: [D64265757](https://our.internmc.facebook.com/intern/diff/D64265757/) cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D64265757 |
This was referenced Oct 29, 2024
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 29, 2024
…Vectorized (#137912) Migrated as much as possible and convenient; focusing on fp16 for now. (This is building toward enabling these fast paths on x86 for machines without AVX-512fp16/bf16 to fix pytorch/torchchat#1253 .) Differential Revision: [D64218206](https://our.internmc.facebook.com/intern/diff/D64218206/) Pull Request resolved: #137912 Approved by: https://github.com/malfet ghstack dependencies: #137661, #137911
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 29, 2024
…137913) float16_t is ARM-specific. Half is not. Differential Revision: [D64218427](https://our.internmc.facebook.com/intern/diff/D64218427/) Pull Request resolved: #137913 Approved by: https://github.com/Skylion007, https://github.com/malfet ghstack dependencies: #137661, #137911, #137912
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 29, 2024
…pu/ (#137914) This is in preparation for supporting x86 as well; we need to be in this directory so that we can get rebuilt with different CPU_CAPABILITY settings (AVX2/AVX-512). Also incidentally starts fulfilling request from @malfet to split the ARM64 fast path stuff into its own file. BFloat16 will be in a later diff. Differential Revision: [D64265755](https://our.internmc.facebook.com/intern/diff/D64265755/) Pull Request resolved: #137914 Approved by: https://github.com/Skylion007, https://github.com/malfet ghstack dependencies: #137661, #137911, #137912, #137913
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 29, 2024
In preparation for other vector instruction sets. (NEON and AVX512 have 32 registers, but AVX and AVX2 have only 16.) Differential Revision: [D64265759](https://our.internmc.facebook.com/intern/diff/D64265759/) Pull Request resolved: #137915 Approved by: https://github.com/Skylion007, https://github.com/malfet ghstack dependencies: #137661, #137911, #137912, #137913, #137914
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 29, 2024
…whole vector register instead of half (#137916) The fixup loop doesn't really need to vectorize the last 7 elements, and not doing so will make migrating to x86 simpler. Differential Revision: [D64280689](https://our.internmc.facebook.com/intern/diff/D64280689/) Pull Request resolved: #137916 Approved by: https://github.com/malfet ghstack dependencies: #137661, #137911, #137912, #137913, #137914, #137915
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 29, 2024
…s for non-ARM architectures too (#137917) Remove reasons to gate it on ARM. Differential Revision: [D64280687](https://our.internmc.facebook.com/intern/diff/D64280687/) Pull Request resolved: #137917 Approved by: https://github.com/malfet ghstack dependencies: #137661, #137911, #137912, #137913, #137914, #137915, #137916
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 30, 2024
`mask` is already defined as `uint16x8_t` no need to reinterpret it https://github.com/pytorch/pytorch/blob/bd369bb18258fc3be5ee91f8fcaf06a4b6fc41a7/aten/src/ATen/cpu/vec/vec128/vec128_half_neon.h#L220 Fixes ``` var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec128/vec128_half_neon.h: In static member function 'static at::vec::DEFAULT::Vectorized<c10::Half> at::vec::DEFAULT::Vectorized<c10::Half>::set(const at::vec::DEFAULT::Vectorized<c10::Half>&, const at::vec::DEFAULT::Vectorized<c10::Half>&, int64_t)': /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec128/vec128_half_neon.h:227:39: error: cannot convert 'uint16x8_t' to 'float16x8_t' 227 | vreinterpretq_u16_f16(mask), | ^~~~ | | | uint16x8_t In file included from /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/intrinsics.h:23, from /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec128/vec128.h:4, from /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec.h:6, from /var/lib/jenkins/workspace/aten/src/ATen/test/vec_test_all_types.h:2, from /var/lib/jenkins/workspace/aten/src/ATen/test/vec_test_all_types.cpp:1: /usr/lib/gcc/aarch64-linux-gnu/11/include/arm_neon.h:5841:36: note: initializing argument 1 of 'uint16x8_t vreinterpretq_u16_f16(float16x8_t)' 5841 | vreinterpretq_u16_f16 (float16x8_t __a) | ~~~~~~~~~~~~^~~ ``` introduced by #137911 Also, guard any use of NEON intrinsics in `ReducedPrecisionFloatGemvFastPathKernel.cpp` with `!defined(CPU_CAPABILITY_SVE)` otherwise compilation fails with ``` /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp: In function 'float at::native::SVE256::reduce(at::vec::SVE256::VectorizedN<c10::Half, 16>&)': /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp:77:24: error: cannot convert 'at::vec::SVE256::Vectorized<float>' to 'float32x4_t' 77 | return vaddvq_f32(t0 + t1); | ~~~^~~~ | | | at::vec::SVE256::Vectorized<float> In file included from /var/lib/jenkins/workspace/c10/util/Half.h:51, from /var/lib/jenkins/workspace/c10/util/Float8_e5m2.h:17, from /var/lib/jenkins/workspace/c10/core/ScalarType.h:8, from /var/lib/jenkins/workspace/c10/core/TensorImpl.h:11, from /var/lib/jenkins/workspace/c10/core/GeneratorImpl.h:8, from /var/lib/jenkins/workspace/aten/src/ATen/core/Generator.h:18, from /var/lib/jenkins/workspace/aten/src/ATen/CPUGeneratorImpl.h:3, from /var/lib/jenkins/workspace/aten/src/ATen/Context.h:4, from /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp:2, from /var/lib/jenkins/workspace/build/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp.SVE256.cpp:1: /usr/lib/gcc/aarch64-linux-gnu/11/include/arm_neon.h:10423:25: note: initializing argument 1 of 'float32_t vaddvq_f32(float32x4_t)' 10423 | vaddvq_f32 (float32x4_t __a) | ~~~~~~~~~~~~^~~ In file included from /var/lib/jenkins/workspace/build/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp.SVE256.cpp:1: /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp: In function 'float at::native::SVE256::reduce(at::vec::SVE256::Vectorized<float>)': /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp:119:21: error: cannot convert 'at::vec::SVE256::Vectorized<float>' to 'float32x4_t' 119 | return vaddvq_f32(x); | ^ | | | at::vec::SVE256::Vectorized<float> ``` Pull Request resolved: #139235 Approved by: https://github.com/huydhn
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…c isn't available (pytorch#137911) We can do most of what this header does (by line count) anyway by converting to and from float. Differential Revision: [D64265757](https://our.internmc.facebook.com/intern/diff/D64265757/) Pull Request resolved: pytorch#137911 Approved by: https://github.com/jgong5, https://github.com/malfet ghstack dependencies: pytorch#137661
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…Vectorized (pytorch#137912) Migrated as much as possible and convenient; focusing on fp16 for now. (This is building toward enabling these fast paths on x86 for machines without AVX-512fp16/bf16 to fix pytorch/torchchat#1253 .) Differential Revision: [D64218206](https://our.internmc.facebook.com/intern/diff/D64218206/) Pull Request resolved: pytorch#137912 Approved by: https://github.com/malfet ghstack dependencies: pytorch#137661, pytorch#137911
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…ytorch#137913) float16_t is ARM-specific. Half is not. Differential Revision: [D64218427](https://our.internmc.facebook.com/intern/diff/D64218427/) Pull Request resolved: pytorch#137913 Approved by: https://github.com/Skylion007, https://github.com/malfet ghstack dependencies: pytorch#137661, pytorch#137911, pytorch#137912
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…pu/ (pytorch#137914) This is in preparation for supporting x86 as well; we need to be in this directory so that we can get rebuilt with different CPU_CAPABILITY settings (AVX2/AVX-512). Also incidentally starts fulfilling request from @malfet to split the ARM64 fast path stuff into its own file. BFloat16 will be in a later diff. Differential Revision: [D64265755](https://our.internmc.facebook.com/intern/diff/D64265755/) Pull Request resolved: pytorch#137914 Approved by: https://github.com/Skylion007, https://github.com/malfet ghstack dependencies: pytorch#137661, pytorch#137911, pytorch#137912, pytorch#137913
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…137915) In preparation for other vector instruction sets. (NEON and AVX512 have 32 registers, but AVX and AVX2 have only 16.) Differential Revision: [D64265759](https://our.internmc.facebook.com/intern/diff/D64265759/) Pull Request resolved: pytorch#137915 Approved by: https://github.com/Skylion007, https://github.com/malfet ghstack dependencies: pytorch#137661, pytorch#137911, pytorch#137912, pytorch#137913, pytorch#137914
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…whole vector register instead of half (pytorch#137916) The fixup loop doesn't really need to vectorize the last 7 elements, and not doing so will make migrating to x86 simpler. Differential Revision: [D64280689](https://our.internmc.facebook.com/intern/diff/D64280689/) Pull Request resolved: pytorch#137916 Approved by: https://github.com/malfet ghstack dependencies: pytorch#137661, pytorch#137911, pytorch#137912, pytorch#137913, pytorch#137914, pytorch#137915
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…s for non-ARM architectures too (pytorch#137917) Remove reasons to gate it on ARM. Differential Revision: [D64280687](https://our.internmc.facebook.com/intern/diff/D64280687/) Pull Request resolved: pytorch#137917 Approved by: https://github.com/malfet ghstack dependencies: pytorch#137661, pytorch#137911, pytorch#137912, pytorch#137913, pytorch#137914, pytorch#137915, pytorch#137916
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
`mask` is already defined as `uint16x8_t` no need to reinterpret it https://github.com/pytorch/pytorch/blob/bd369bb18258fc3be5ee91f8fcaf06a4b6fc41a7/aten/src/ATen/cpu/vec/vec128/vec128_half_neon.h#L220 Fixes ``` var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec128/vec128_half_neon.h: In static member function 'static at::vec::DEFAULT::Vectorized<c10::Half> at::vec::DEFAULT::Vectorized<c10::Half>::set(const at::vec::DEFAULT::Vectorized<c10::Half>&, const at::vec::DEFAULT::Vectorized<c10::Half>&, int64_t)': /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec128/vec128_half_neon.h:227:39: error: cannot convert 'uint16x8_t' to 'float16x8_t' 227 | vreinterpretq_u16_f16(mask), | ^~~~ | | | uint16x8_t In file included from /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/intrinsics.h:23, from /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec128/vec128.h:4, from /var/lib/jenkins/workspace/aten/src/ATen/cpu/vec/vec.h:6, from /var/lib/jenkins/workspace/aten/src/ATen/test/vec_test_all_types.h:2, from /var/lib/jenkins/workspace/aten/src/ATen/test/vec_test_all_types.cpp:1: /usr/lib/gcc/aarch64-linux-gnu/11/include/arm_neon.h:5841:36: note: initializing argument 1 of 'uint16x8_t vreinterpretq_u16_f16(float16x8_t)' 5841 | vreinterpretq_u16_f16 (float16x8_t __a) | ~~~~~~~~~~~~^~~ ``` introduced by pytorch#137911 Also, guard any use of NEON intrinsics in `ReducedPrecisionFloatGemvFastPathKernel.cpp` with `!defined(CPU_CAPABILITY_SVE)` otherwise compilation fails with ``` /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp: In function 'float at::native::SVE256::reduce(at::vec::SVE256::VectorizedN<c10::Half, 16>&)': /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp:77:24: error: cannot convert 'at::vec::SVE256::Vectorized<float>' to 'float32x4_t' 77 | return vaddvq_f32(t0 + t1); | ~~~^~~~ | | | at::vec::SVE256::Vectorized<float> In file included from /var/lib/jenkins/workspace/c10/util/Half.h:51, from /var/lib/jenkins/workspace/c10/util/Float8_e5m2.h:17, from /var/lib/jenkins/workspace/c10/core/ScalarType.h:8, from /var/lib/jenkins/workspace/c10/core/TensorImpl.h:11, from /var/lib/jenkins/workspace/c10/core/GeneratorImpl.h:8, from /var/lib/jenkins/workspace/aten/src/ATen/core/Generator.h:18, from /var/lib/jenkins/workspace/aten/src/ATen/CPUGeneratorImpl.h:3, from /var/lib/jenkins/workspace/aten/src/ATen/Context.h:4, from /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp:2, from /var/lib/jenkins/workspace/build/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp.SVE256.cpp:1: /usr/lib/gcc/aarch64-linux-gnu/11/include/arm_neon.h:10423:25: note: initializing argument 1 of 'float32_t vaddvq_f32(float32x4_t)' 10423 | vaddvq_f32 (float32x4_t __a) | ~~~~~~~~~~~~^~~ In file included from /var/lib/jenkins/workspace/build/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp.SVE256.cpp:1: /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp: In function 'float at::native::SVE256::reduce(at::vec::SVE256::Vectorized<float>)': /var/lib/jenkins/workspace/aten/src/ATen/native/cpu/ReducedPrecisionFloatGemvFastPathKernel.cpp:119:21: error: cannot convert 'at::vec::SVE256::Vectorized<float>' to 'float32x4_t' 119 | return vaddvq_f32(x); | ^ | | | at::vec::SVE256::Vectorized<float> ``` Pull Request resolved: pytorch#139235 Approved by: https://github.com/huydhn
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
We can do most of what this header does (by line count) anyway by converting to and from float.
Differential Revision: D64265757
cc @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10