llvm/llvm-project

[X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - add AVX512 integer comparison intrinsics to be used in constexpr

Open

#162.054 aberto em 6 de out. de 2025

Ver no GitHub
 (15 comments) (0 reactions) (2 assignees)C++ (10.782 forks)batch import
backend:X86clang:bytecodeclang: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

The core integer comparison intrinsics could be used in constexpr:

_mm_cmp_epi8_mask _mm_cmp_epu8_mask
_mm_cmp_epi16_mask _mm_cmp_epu16_mask
_mm_cmp_epi32_mask _mm_cmp_epu32_mask
_mm_cmp_epi64_mask _mm_cmp_epu64_mask

_mm256_cmp_epi8_mask _mm256_cmp_epu8_mask
_mm256_cmp_epi16_mask _mm256_cmp_epu16_mask
_mm256_cmp_epi32_mask _mm256_cmp_epu32_mask
_mm256_cmp_epi64_mask _mm256_cmp_epu64_mask

_mm512_cmp_epi8_mask _mm512_cmp_epu8_mask
_mm512_cmp_epi16_mask _mm512_cmp_epu16_mask
_mm512_cmp_epi32_mask _mm512_cmp_epu32_mask
_mm512_cmp_epi64_mask _mm512_cmp_epu64_mask

These need to handle the comparison modes detailed in the Intel Intrinsics Guide:

0: OP := _MM_CMPINT_EQ
1: OP := _MM_CMPINT_LT
2: OP := _MM_CMPINT_LE
3: OP := _MM_CMPINT_FALSE
4: OP := _MM_CMPINT_NE
5: OP := _MM_CMPINT_NLT
6: OP := _MM_CMPINT_NLE
7: OP := _MM_CMPINT_TRUE

There are a LOT of wrappers to these intrinsics (e.g. eq/ge/gt/le/lt/neq variants) - ideally we need test coverage for all of them. To avoid incredible boredom you should probably write a script to autogen the TEST_CONSTEXPR lines somehow that you then copy+paste into the relevant locations.

Guia do colaborador