Merge pull request xtensor-stack#1306 from DiamonDinoia/fix/avx2-uint64-mul

DiamonDinoia · web-flow · commit d05d46f52f62 · 2026-04-16T15:20:25.000-04:00
perf: add native AVX2 uint64/int64 mul kernel
diff --git a/include/xsimd/arch/xsimd_avx2.hpp b/include/xsimd/arch/xsimd_avx2.hpp
@@ -912,6 +912,16 @@ namespace xsimd
             {
                 return _mm256_mullo_epi32(self, other);
             }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
+            {
+                return _mm256_add_epi64(
+                    _mm256_mul_epu32(self, other),
+                    _mm256_slli_epi64(
+                        _mm256_add_epi64(
+                            _mm256_mul_epu32(other, _mm256_shuffle_epi32(self, _MM_SHUFFLE(2, 3, 0, 1))),
+                            _mm256_mul_epu32(self, _mm256_shuffle_epi32(other, _MM_SHUFFLE(2, 3, 0, 1)))),
+                        32));
+            }
             else
             {
                 return mul(self, other, avx {});

Original file line number	Diff line number	Diff line change
`@@ -912,6 +912,16 @@ namespace xsimd`
`912`	`912`	`{`
`913`	`913`	`return _mm256_mullo_epi32(self, other);`
`914`	`914`	`}`
	`915`	`+ else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)`
	`916`	`+ {`
	`917`	`+ return _mm256_add_epi64(`
	`918`	`+ _mm256_mul_epu32(self, other),`
	`919`	`+ _mm256_slli_epi64(`
	`920`	`+ _mm256_add_epi64(`
	`921`	`+ _mm256_mul_epu32(other, _mm256_shuffle_epi32(self, _MM_SHUFFLE(2, 3, 0, 1))),`
	`922`	`+ _mm256_mul_epu32(self, _mm256_shuffle_epi32(other, _MM_SHUFFLE(2, 3, 0, 1)))),`
	`923`	`+ 32));`
	`924`	`+ }`
`915`	`925`	`else`
`916`	`926`	`{`
`917`	`927`	`return mul(self, other, avx {});`