[CpuInductor] Enable NEON ISA detection on Linux ARM #129075
Conversation
|
Warning: Unknown label
Please add the new label to .github/pytorch-probot.yml |
|
@pytorchbot rebase -b main |
|
@pytorchbot started a rebase job onto refs/remotes/origin/main. Check the current status here |
Also, cleanup code a bit to use `x in [y, z]` instead of `x == y or x == z`
|
Successfully rebased |
c5e565d to
35eddf6
Compare
|
|
||
| __at_align__ float in_out_ptr0[16] = {0.0}; | ||
| #endif | ||
| alignas(64) float in_out_ptr0[16] = {0.0}; |
There was a problem hiding this comment.
@xuhancn Does it work on Windows?
There was a problem hiding this comment.
@jgong5 Is there a CI to test it on Windows? But at least godbolt believes it is supported even by a pretty old MSVC: https://godbolt.org/z/Tr6Wa9WE6
|
@pytorchbot merge -i |
Merge failedReason: This PR needs a If not, please add the To add a label, you can comment to pytorchbot, for example For more information, see Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot merge -i |
I.e. for every platform we support, it should be something but invalid AVX512 on Intel Linux, and NEON on Apple silicon and Linux aarch64 2.4 aarch64 failures are expected, as pytorch/pytorch#129075 has not been picked for the release yet
|
@pytorchbot cherry-pick --onto release/2.4 -c critical |
Also, cleanup code a bit to use `x in [y, z]` instead of `x == y or x == z` And do not redefine `at_align`, but instead use `alignas(64)` as was suggested in https://github.com/pytorch/pytorch/pull/128686/files#r1639365978 Test plan: `python3 -c "import torch._inductor.codecache as cc; isa = cc.valid_vec_isa_list()[0];print(str(isa), bool(isa))"` Pull Request resolved: #129075 Approved by: https://github.com/jansel (cherry picked from commit b2a9b8d)
Cherry picking #129075The cherry pick PR is at #133578 and it is recommended to link a critical cherry pick PR with an issue. The following tracker issues are updated: Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot cherry-pick --onto release/2.4 -c critical |
Cherry picking #129075Command Details for Dev Infra teamRaised by workflow job |
Also, cleanup code a bit to use `x in [y, z]` instead of `x == y or x == z` And do not redefine `at_align`, but instead use `alignas(64)` as was suggested in https://github.com/pytorch/pytorch/pull/128686/files#r1639365978 Test plan: `python3 -c "import torch._inductor.codecache as cc; isa = cc.valid_vec_isa_list()[0];print(str(isa), bool(isa))"` Pull Request resolved: #129075 Approved by: https://github.com/jansel
* [CpuInductor] Enable NEON ISA detection on Linux ARM (#129075) Also, cleanup code a bit to use `x in [y, z]` instead of `x == y or x == z` And do not redefine `at_align`, but instead use `alignas(64)` as was suggested in https://github.com/pytorch/pytorch/pull/128686/files#r1639365978 Test plan: `python3 -c "import torch._inductor.codecache as cc; isa = cc.valid_vec_isa_list()[0];print(str(isa), bool(isa))"` Pull Request resolved: #129075 Approved by: https://github.com/jansel * Fix merge mistakes --------- Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com>
|
Checked that it works with 2.4.1: |
Also, cleanup code a bit to use
x in [y, z]instead ofx == y or x == zAnd do not redefine
at_align, but instead usealignas(64)as was suggested in https://github.com/pytorch/pytorch/pull/128686/files#r1639365978Test plan:
python3 -c "import torch._inductor.codecache as cc; isa = cc.valid_vec_isa_list()[0];print(str(isa), bool(isa))"cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang