Fix bad alignment crash on SSE

Applied same change to AVX2 and AVX512 impls
BasedInc · Feb 8, 2023 · db0d28e · db0d28e
1 parent 444fc44
commit db0d28e
Show file tree

Hide file tree

Showing 3 changed files with 3 additions and 3 deletions.
diff --git a/src/arch/x86/AVX2.cpp b/src/arch/x86/AVX2.cpp
@@ -31,7 +31,7 @@ namespace hat::detail {
         const auto e = vec + n;
 
         for (; vec != e; vec++) {
-            const auto cmp = _mm256_cmpeq_epi8(firstByte, *vec);
+            const auto cmp = _mm256_cmpeq_epi8(firstByte, _mm256_loadu_si256(vec));
             auto mask = static_cast<uint32_t>(_mm256_movemask_epi8(cmp));
             while (mask) {
                 const auto offset = _tzcnt_u32(mask);

diff --git a/src/arch/x86/AVX512.cpp b/src/arch/x86/AVX512.cpp
@@ -31,7 +31,7 @@ namespace hat::detail {
         const auto e = vec + n;
 
         for (; vec != e; vec++) {
-            auto mask = _mm512_cmpeq_epi8_mask(firstByte, *vec);
+            auto mask = _mm512_cmpeq_epi8_mask(firstByte, _mm512_loadu_si512(vec));
             while (mask) {
                 const auto offset = LIBHAT_TZCNT64(mask);
                 const auto i = reinterpret_cast<const std::byte*>(vec) + offset;

diff --git a/src/arch/x86/SSE.cpp b/src/arch/x86/SSE.cpp
@@ -31,7 +31,7 @@ namespace hat::detail {
         const auto e = vec + n;
 
         for (; vec != e; vec++) {
-            const auto cmp = _mm_cmpeq_epi8(firstByte, *vec);
+            const auto cmp = _mm_cmpeq_epi8(firstByte, _mm_loadu_si128(vec));
             auto mask = static_cast<uint32_t>(_mm_movemask_epi8(cmp));
             while (mask) {
                 const auto offset = LIBHAT_BSF32(mask);