sse/avx

serge-sans-paille · serge-sans-paille · commit 3a9e9027bd02 · 2025-10-11T23:48:16.000+02:00
diff --git a/include/xsimd/arch/xsimd_avx2.hpp b/include/xsimd/arch/xsimd_avx2.hpp
@@ -1023,13 +1023,13 @@ namespace xsimd
         template <class A>
         XSIMD_INLINE void store(batch_bool<float, A> b, bool* mem, requires_arch<avx2>) noexcept
         {
-            return store(batch_bool<uint32_t, A>(b.data), mem, A {});
+            return store(batch_bool<uint32_t, A>(_mm256_castps_si256(b.data)), mem, A {});
         }
 
         template <class A>
         XSIMD_INLINE void store(batch_bool<double, A> b, bool* mem, requires_arch<avx2>) noexcept
         {
-            return store(batch_bool<uint64_t, A>(b.data), mem, A {});
+            return store(batch_bool<uint64_t, A>(_mm256_castpd_si256(b.data)), mem, A {});
         }
 
         // ssub
diff --git a/include/xsimd/arch/xsimd_sse2.hpp b/include/xsimd/arch/xsimd_sse2.hpp
@@ -1761,13 +1761,13 @@ namespace xsimd
         template <class A>
         XSIMD_INLINE void store(batch_bool<float, A> b, bool* mem, requires_arch<sse2>) noexcept
         {
-            store(batch_bool<uint32_t, A>(b.data), mem, A {});
+            store(batch_bool<uint32_t, A>(_mm_castsi128_ps(b.data)), mem, A {});
         }
 
         template <class A>
         XSIMD_INLINE void store(batch_bool<double, A> b, bool* mem, requires_arch<sse2>) noexcept
         {
-            store(batch_bool<uint64_t, A>(b.data), mem, A {});
+            store(batch_bool<uint64_t, A>(_mm_castsi128_pd(b.data)), mem, A {});
         }
 
         // store_aligned

Original file line number	Diff line number	Diff line change
`@@ -1023,13 +1023,13 @@ namespace xsimd`
`1023`	`1023`	`template <class A>`
`1024`	`1024`	`XSIMD_INLINE void store(batch_bool<float, A> b, bool* mem, requires_arch<avx2>) noexcept`
`1025`	`1025`	`{`
`1026`		`- return store(batch_bool<uint32_t, A>(b.data), mem, A {});`
	`1026`	`+ return store(batch_bool<uint32_t, A>(_mm256_castps_si256(b.data)), mem, A {});`
`1027`	`1027`	`}`
`1028`	`1028`
`1029`	`1029`	`template <class A>`
`1030`	`1030`	`XSIMD_INLINE void store(batch_bool<double, A> b, bool* mem, requires_arch<avx2>) noexcept`
`1031`	`1031`	`{`
`1032`		`- return store(batch_bool<uint64_t, A>(b.data), mem, A {});`
	`1032`	`+ return store(batch_bool<uint64_t, A>(_mm256_castpd_si256(b.data)), mem, A {});`
`1033`	`1033`	`}`
`1034`	`1034`
`1035`	`1035`	`// ssub`
Original file line number	Diff line number	Diff line change
`@@ -1761,13 +1761,13 @@ namespace xsimd`
`1761`	`1761`	`template <class A>`
`1762`	`1762`	`XSIMD_INLINE void store(batch_bool<float, A> b, bool* mem, requires_arch<sse2>) noexcept`
`1763`	`1763`	`{`
`1764`		`- store(batch_bool<uint32_t, A>(b.data), mem, A {});`
	`1764`	`+ store(batch_bool<uint32_t, A>(_mm_castsi128_ps(b.data)), mem, A {});`
`1765`	`1765`	`}`
`1766`	`1766`
`1767`	`1767`	`template <class A>`
`1768`	`1768`	`XSIMD_INLINE void store(batch_bool<double, A> b, bool* mem, requires_arch<sse2>) noexcept`
`1769`	`1769`	`{`
`1770`		`- store(batch_bool<uint64_t, A>(b.data), mem, A {});`
	`1770`	`+ store(batch_bool<uint64_t, A>(_mm_castsi128_pd(b.data)), mem, A {});`
`1771`	`1771`	`}`
`1772`	`1772`
`1773`	`1773`	`// store_aligned`