Optimize avx2 store of batch_bool for 64 bit registers

serge-sans-paille · serge-sans-paille · commit 62a008c797a4 · 2025-10-11T20:26:51.000Z
diff --git a/include/xsimd/arch/xsimd_avx2.hpp b/include/xsimd/arch/xsimd_avx2.hpp
@@ -1013,12 +1013,8 @@ namespace xsimd
                 }
                 else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
                 {
-                    const auto bmask = _mm_set_epi8(
-                        -1, -1, -1, -1, -1, -1, -1, -1,
-                        -1, -1, -1, -1, -1, -1, 8, 0);
-                    auto pack = _mm_unpacklo_epi16(_mm_shuffle_epi8(b_lo, bmask), _mm_shuffle_epi8(b_hi, bmask));
-                    uint32_t val = _mm_cvtsi128_si32(_mm_sub_epi8(_mm_set1_epi8(0), pack));
-                    memcpy(mem, &val, sizeof(val));
+                    uint32_t mask = _mm256_movemask_epi8(_mm256_srli_epi64(b, 56));
+                    memcpy(mem, &mask, sizeof(mask));
                 }
                 else
                 {

Original file line number	Diff line number	Diff line change
`@@ -1013,12 +1013,8 @@ namespace xsimd`
`1013`	`1013`	`}`
`1014`	`1014`	`else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)`
`1015`	`1015`	`{`
`1016`		`- const auto bmask = _mm_set_epi8(`
`1017`		`- -1, -1, -1, -1, -1, -1, -1, -1,`
`1018`		`- -1, -1, -1, -1, -1, -1, 8, 0);`
`1019`		`- auto pack = _mm_unpacklo_epi16(_mm_shuffle_epi8(b_lo, bmask), _mm_shuffle_epi8(b_hi, bmask));`
`1020`		`- uint32_t val = _mm_cvtsi128_si32(_mm_sub_epi8(_mm_set1_epi8(0), pack));`
`1021`		`- memcpy(mem, &val, sizeof(val));`
	`1016`	`+ uint32_t mask = _mm256_movemask_epi8(_mm256_srli_epi64(b, 56));`
	`1017`	`+ memcpy(mem, &mask, sizeof(mask));`
`1022`	`1018`	`}`
`1023`	`1019`	`else`
`1024`	`1020`	`{`