x86 (32/64): go back to passing SIMD vectors by-ptr #141309
Conversation
rustbot
added
S-waiting-on-review
|
@bors try |
x86 (32/64): go back to passing SIMD vectors by-ptr Fixes rust-lang#139029 by partially reverting rust-lang#135408 and going back to passing SIMD vectors by-ptr on x86. Sadly, by-val confuses the LLVM inliner so much that it's not worth it... r? `@tgross35` Cc `@nikic` try-job: `test-various*` try-job: x86_64-gnu-nopt try-job: dist-i586-gnu-i586-i686-musl try-job: x86_64-msvc-1
|
☀️ Try build successful - checks-actions |
|
Is there an LLVM issue we can reference? |
We got @nikic's +1. Anyone else we should ask? FWIW this will also help with #141848: for people that disable SSE on an i686 target, the lint at #116558 (hard error on nightly and beta) can actually trigger for "Rust" ABI functions. We don't actually support such configurations, but we did not intent for them to become a hard error so quickly either. |
This comment has been minimized.
This comment has been minimized.
|
(That's rust-lang/miri#4367) |
|
We may even consider backporting this, to avoid Rust 1.88 being the only release where a "Rust" ABI function with a by-value SIMD vector on an i686 target modified to disable SSE causes a hard error. (That's not a configuration we intend to support, but it wasn't really meant to hard-error quite so quickly.) See #141848 for context. |
RalfJung
added
the
beta-nominated
I think it's fine to merge with the CI fix 👍 a few other people have looked at this and upvoted the backport, so seems like there's more consensus there anyway. |
|
Which CI fix? This should be good to go as-is.
|
|
Oops I missed that the linked Miri issue has since been fixed. |
|
(mistyped the bors command) @bors r=tgross35,nikic |
bors
added
S-waiting-on-bors
|
lgtm |
…nikic x86 (32/64): go back to passing SIMD vectors by-ptr Fixes rust-lang#139029 by partially reverting rust-lang#135408 and going back to passing SIMD vectors by-ptr on x86. Sadly, by-val confuses the LLVM inliner so much that it's not worth it... Also fixes rust-lang#141848 by no longer actually using vector registers with the "Rust" ABI. r? `@tgross35` Cc `@nikic` try-job: `test-various*` try-job: x86_64-gnu-nopt try-job: dist-i586-gnu-i586-i686-musl try-job: x86_64-msvc-1
Rollup of 6 pull requests Successful merges: - #140715 (Clarify &mut-methods' docs on sync::OnceLock) - #141309 (x86 (32/64): go back to passing SIMD vectors by-ptr) - #141677 (Async drop - type instead of async drop fn, fixes #140484) - #141733 (C-variadic functions must be unsafe) - #141858 (Fix typo in `StructuralPartialEq` docs) - #141874 (add f16_epsilon and f128_epsilon diagnostic items) r? `@ghost` `@rustbot` modify labels: rollup
bors
added
the
S-waiting-on-author
|
@bors rollup=iffy
|
RalfJung
force-pushed
the
x86-simd-abi
branch
2 times, most recently
from
e39a075 to
ca94292
Compare
|
Ah, due to #139261 we can't do a full revert of that test -- the alignment is capped at 4 on 32bit MSVC targets. It's probably fine to skip the test there as the Rust ABI is the same across targets (in terms of the LLVM IR we generate) so having the test covered on Linux and macOS is enough. @bors2 try |
x86 (32/64): go back to passing SIMD vectors by-ptr Fixes #139029 by partially reverting #135408 and going back to passing SIMD vectors by-ptr on x86. Sadly, by-val confuses the LLVM inliner so much that it's not worth it... Also fixes #141848 by no longer actually using vector registers with the "Rust" ABI. r? `@tgross35` Cc `@nikic` try-job: `test-various*` try-job: dist-i586-gnu-i586-i686-musl try-job: `x86_64-*` try-job: `i686-*`
|
💔 Test failed
|
|
@bors try |
x86 (32/64): go back to passing SIMD vectors by-ptr Fixes #139029 by partially reverting #135408 and going back to passing SIMD vectors by-ptr on x86. Sadly, by-val confuses the LLVM inliner so much that it's not worth it... Also fixes #141848 by no longer actually using vector registers with the "Rust" ABI. r? `@tgross35` Cc `@nikic` try-job: `test-various*` try-job: dist-i586-gnu-i586-i686-musl try-job: `x86_64-*` try-job: `i686-msvc*`
This comment has been minimized.
This comment has been minimized.
|
@bors2 try |
x86 (32/64): go back to passing SIMD vectors by-ptr Fixes #139029 by partially reverting #135408 and going back to passing SIMD vectors by-ptr on x86. Sadly, by-val confuses the LLVM inliner so much that it's not worth it... Also fixes #141848 by no longer actually using vector registers with the "Rust" ABI. r? `@tgross35` Cc `@nikic` try-job: `test-various*` try-job: dist-i586-gnu-i586-i686-musl try-job: x86_64-gnu-nopt try-job: `x86_64-msvc*` try-job: `i686-msvc*`
|
@bors r=tgross35,nikic,workingjubilee |
bors
added
S-waiting-on-bors
|
☔ The latest upstream changes (presumably #141984) made this pull request unmergeable. Please resolve the merge conflicts. |
bors
added
S-waiting-on-author
|
@bors r=tgross35,nikic,workingjubilee p=1 |
bors
added
S-waiting-on-bors
Fixes #139029 by partially reverting #135408 and going back to passing SIMD vectors by-ptr on x86. Sadly, by-val confuses the LLVM inliner so much that it's not worth it...
Also fixes #141848 by no longer actually using vector registers with the "Rust" ABI.
r? @tgross35
Cc @nikic
try-job:
test-various*try-job: dist-i586-gnu-i586-i686-musl
try-job: x86_64-gnu-nopt
try-job:
x86_64-msvc*try-job:
i686-msvc*