llvm/llvm-project

[X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow SSE/AVX FP2INT conversion intrinsics to be used in constexpr

Open

#169.374 aberto em 24 de nov. de 2025

Ver no GitHub
 (10 comments) (0 reactions) (2 assignees)C++ (10.782 forks)batch import
backend:X86clang:bytecodeclang:frontendclang:headersconstexprgood first issue

Métricas do repositório

Stars
 (26.378 stars)
Métricas de merge de PR
 (Mesclagem média 1d 2h) (1.000 fundiu PRs em 30d)

Description

Similar to #160312 except initially we should only allow EXACT/INBOUNDS and FINITE conversions - any special case (DENORMAL/INF/NAN/etc.) or loss of precision shouldn't be allowed in constexpr for now.

_mm_cvtss_si32
_mm_cvt_ss2si
_mm_cvtss_si64
_mm_cvtps_pi32
_mm_cvt_ps2pi

_mm_cvttss_si32
_mm_cvtt_ss2si
_mm_cvttss_si64
_mm_cvttps_pi32
_mm_cvtt_ps2pi

_mm_cvtps_pi16
_mm_cvtps_pi8

_mm_cvtpd_epi32
_mm_cvtsd_si32
_mm_cvttpd_epi32
_mm_cvttsd_si32

_mm_cvtpd_pi32
_mm_cvttpd_pi32

_mm_cvtsd_si64
_mm_cvttsd_si64

_mm_cvtps_epi32
_mm_cvttps_epi32

_mm256_cvtpd_epi32
_mm256_cvttpd_epi32
_mm256_cvtps_epi32
_mm256_cvttps_epi32

There are also the AVX512 variants, but the SSE/AVX are a good starting point.

Guia do colaborador