| Issue |
155395
|
| Summary |
[Clang] VectorExprEvaluator::VisitCallExpr - add MMX/SSE/AVX PHADD/SUB & HADDPS/D intrinsics to be used in constexpr
|
| Labels |
good first issue,
backend:X86,
clang:frontend,
constexpr
|
| Assignees |
|
| Reporter |
RKSimon
|
The HADD/SUB horizontal intrinsics can all be used in constexpr with suitable handling of the __builtin_ia32_phadd/sub* builtins inside VectorExprEvaluator::VisitCallExpr
```
_mm_hadd_pi16 _mm_hadd_epi16 _mm256_hadd_epi16
_mm_hadd_pi32 _mm_hadd_epi32 _mm256_hadd_epi32
_mm_hadds_pi16 _mm_hadds_epi16 _mm256_hadds_epi16
_mm_hsub_pi16 _mm_hsub_epi16 _mm256_hsub_epi16
_mm_hsub_pi32 _mm_hsub_epi32 _mm256_hsub_epi32
_mm_hsubs_pi16 _mm_hsubs_epi16 _mm256_hsubs_epi16
_mm_hadd_pd _mm256_hadd_pd
_mm_hadd_ps _mm256_hadd_ps
_mm_hsub_pd _mm256_hsub_pd
_mm_hsub_ps _mm256_hsub_ps
```
These are elementwise horizontal pairwise so getting the correct src/dst element indices is vital and use a mixture of regular add/sub, fadd/fsub and signed sat add/sub arithmetic
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
