Implicitly enable evex512 if avx512 is enabled #121088
Merged
Conversation
This file contains 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
LLVM 18 requires the evex512 feature to allow use of zmm registers. LLVM automatically sets it when using a generic CPU, but not when `-C target-cpu` is specified. This will result either in backend legalization crashes, or code unexpectedly using ymm instead of zmm registers. For now, make sure that `avx512*` features imply `evex512`. Long term we'll probably have to deal with the AVX10 mess somehow.
rustbot
added
S-waiting-on-review
|
@bors r+ |
bors
added
S-waiting-on-bors
This was referenced Feb 14, 2024
bors
added a commit
to rust-lang-ci/rust
that referenced
this pull request
Feb 15, 2024
…llaumeGomez Rollup of 8 pull requests Successful merges: - rust-lang#120449 (Document requirements for unsized {Rc,Arc}::from_raw) - rust-lang#120505 (Fix BTreeMap's Cursor::remove_{next,prev}) - rust-lang#120672 (std::thread update freebsd stack guard handling.) - rust-lang#121088 (Implicitly enable evex512 if avx512 is enabled) - rust-lang#121104 (Ignore unsized types when trying to determine the size of the original type) - rust-lang#121107 (Fix msg for verbose suggestions with confusable capitalization) - rust-lang#121113 (Continue compilation even if inherent impl checks fail) - rust-lang#121120 (Add `ErrorGuaranteed` to `ast::LitKind::Err`, `token::LitKind::Err`.) r? `@ghost` `@rustbot` modify labels: rollup
rust-timer
added a commit
to rust-lang-ci/rust
that referenced
this pull request
Feb 15, 2024
Rollup merge of rust-lang#121088 - nikic:evex512, r=Amanieu Implicitly enable evex512 if avx512 is enabled LLVM 18 requires the evex512 feature to allow use of zmm registers. LLVM automatically sets it when using a generic CPU, but not when `-C target-cpu` is specified. This will result either in backend legalization crashes, or code unexpectedly using ymm instead of zmm registers. For now, make sure that `avx512*` features imply `evex512`. Long term we'll probably have to deal with the AVX10 mess somehow. Fixes rust-lang#121081. r? `@Amanieu`
2 tasks
freebsd-git
pushed a commit
to freebsd/freebsd-src
that referenced
this pull request
May 16, 2024
[X86][Driver] Do not add `-evex512` for `-march=native` when the target doesn't support AVX512 (#91694)
This prevents problems with ports that fail to build with
CPUTYPE=native, if the native CPU supports AVX512F, resulting in errors
like:
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: always_inline function '_mm512_set1_epi8' requires target feature 'evex512', but would be inlined into function 'adler32_x86_avx512_vl512_vnni' that is compiled without support for 'evex512'
197 | const vec_t ones = VSET1_8(1);
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: AVX vector return of type '__m512i' (vector of 8 'long long' values) without 'evex512' enabled changes the ABI
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
Ports known to be affected are: archivers/libdeflate, devel/highway,
www/node20, and lang/rust.
(For rust itself there is also a similar issue reported at
rust-lang/rust#121088).
PR: 276104
Reported by: netchild
MFC after: 3 days
freebsd-git
pushed a commit
to freebsd/freebsd-src
that referenced
this pull request
May 20, 2024
[X86][Driver] Do not add `-evex512` for `-march=native` when the target doesn't support AVX512 (#91694)
This prevents problems with ports that fail to build with
CPUTYPE=native, if the native CPU supports AVX512F, resulting in errors
like:
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: always_inline function '_mm512_set1_epi8' requires target feature 'evex512', but would be inlined into function 'adler32_x86_avx512_vl512_vnni' that is compiled without support for 'evex512'
197 | const vec_t ones = VSET1_8(1);
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: AVX vector return of type '__m512i' (vector of 8 'long long' values) without 'evex512' enabled changes the ABI
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
Ports known to be affected are: archivers/libdeflate, devel/highway,
www/node20, and lang/rust.
(For rust itself there is also a similar issue reported at
rust-lang/rust#121088).
PR: 276104
Reported by: netchild
MFC after: 3 days
(cherry picked from commit fcfb2d6)
freebsd-git
pushed a commit
to freebsd/freebsd-src
that referenced
this pull request
May 20, 2024
[X86][Driver] Do not add `-evex512` for `-march=native` when the target doesn't support AVX512 (#91694)
This prevents problems with ports that fail to build with
CPUTYPE=native, if the native CPU supports AVX512F, resulting in errors
like:
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: always_inline function '_mm512_set1_epi8' requires target feature 'evex512', but would be inlined into function 'adler32_x86_avx512_vl512_vnni' that is compiled without support for 'evex512'
197 | const vec_t ones = VSET1_8(1);
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: AVX vector return of type '__m512i' (vector of 8 'long long' values) without 'evex512' enabled changes the ABI
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
Ports known to be affected are: archivers/libdeflate, devel/highway,
www/node20, and lang/rust.
(For rust itself there is also a similar issue reported at
rust-lang/rust#121088).
PR: 276104
Reported by: netchild
MFC after: 3 days
(cherry picked from commit fcfb2d6)
freebsd-git
pushed a commit
to freebsd/freebsd-src
that referenced
this pull request
May 20, 2024
[X86][Driver] Do not add `-evex512` for `-march=native` when the target doesn't support AVX512 (#91694)
This prevents problems with ports that fail to build with
CPUTYPE=native, if the native CPU supports AVX512F, resulting in errors
like:
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: always_inline function '_mm512_set1_epi8' requires target feature 'evex512', but would be inlined into function 'adler32_x86_avx512_vl512_vnni' that is compiled without support for 'evex512'
197 | const vec_t ones = VSET1_8(1);
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: AVX vector return of type '__m512i' (vector of 8 'long long' values) without 'evex512' enabled changes the ABI
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
Ports known to be affected are: archivers/libdeflate, devel/highway,
www/node20, and lang/rust.
(For rust itself there is also a similar issue reported at
rust-lang/rust#121088).
PR: 276104
Reported by: netchild
Approved by: re (cperciva)
MFC after: 3 days
(cherry picked from commit fcfb2d6)
(cherry picked from commit 14efb9b)
bsdjhb
pushed a commit
to bsdjhb/cheribsd
that referenced
this pull request
Aug 28, 2024
[X86][Driver] Do not add `-evex512` for `-march=native` when the target doesn't support AVX512 (#91694)
This prevents problems with ports that fail to build with
CPUTYPE=native, if the native CPU supports AVX512F, resulting in errors
like:
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: always_inline function '_mm512_set1_epi8' requires target feature 'evex512', but would be inlined into function 'adler32_x86_avx512_vl512_vnni' that is compiled without support for 'evex512'
197 | const vec_t ones = VSET1_8(1);
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: AVX vector return of type '__m512i' (vector of 8 'long long' values) without 'evex512' enabled changes the ABI
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
Ports known to be affected are: archivers/libdeflate, devel/highway,
www/node20, and lang/rust.
(For rust itself there is also a similar issue reported at
rust-lang/rust#121088).
PR: 276104
Reported by: netchild
MFC after: 3 days
bsdjhb
pushed a commit
to bsdjhb/cheribsd
that referenced
this pull request
Aug 29, 2024
[X86][Driver] Do not add `-evex512` for `-march=native` when the target doesn't support AVX512 (#91694)
This prevents problems with ports that fail to build with
CPUTYPE=native, if the native CPU supports AVX512F, resulting in errors
like:
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: always_inline function '_mm512_set1_epi8' requires target feature 'evex512', but would be inlined into function 'adler32_x86_avx512_vl512_vnni' that is compiled without support for 'evex512'
197 | const vec_t ones = VSET1_8(1);
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:197:21: error: AVX vector return of type '__m512i' (vector of 8 'long long' values) without 'evex512' enabled changes the ABI
/wrkdirs/usr/ports/archivers/libdeflate/work/libdeflate-1.20/lib/x86/adler32_template.h:116:23: note: expanded from macro 'VSET1_8'
116 | # define VSET1_8(a) _mm512_set1_epi8(a)
| ^
Ports known to be affected are: archivers/libdeflate, devel/highway,
www/node20, and lang/rust.
(For rust itself there is also a similar issue reported at
rust-lang/rust#121088).
PR: 276104
Reported by: netchild
MFC after: 3 days
3 tasks
# for free
to join this conversation on GitHub.
Already have an account?
# 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.
LLVM 18 requires the evex512 feature to allow use of zmm registers. LLVM automatically sets it when using a generic CPU, but not when
-C target-cpuis specified. This will result either in backend legalization crashes, or code unexpectedly using ymm instead of zmm registers.For now, make sure that
avx512*features implyevex512. Long term we'll probably have to deal with the AVX10 mess somehow.Fixes #121081.
r? @Amanieu