Fix UBSAN's cry afoul

Technically, we weren't actually doing this the way C wants us to,
legally.  The zmemcpy's turn into NOPs for pretty much all > 0
optimization levels and this gets us defined behavior with the
sanitizer, putting the optimized load by arbitrary alignment into the
compiler's hands instead of ours.

Backport note: Replaced zmemcpy with direct memcpy, as that is what we
end up with in a later commit anyway.

diff --git a/arch/x86/chunkset_avx.c b/arch/x86/chunkset_avx.c
index 7a9a56a096d395c56e818e89d5e94a1b271c854b..82bf0ebbbfaa56726189828e70ce14b238730d87 100644 (file)
--- a/arch/x86/chunkset_avx.c
+++ b/arch/x86/chunkset_avx.c
@@ -21,15 +21,21 @@ static inline void chunkmemset_1(uint8_t *from, chunk_t *chunk) {
 }
 
 static inline void chunkmemset_2(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm256_set1_epi16(*(int16_t *)from);
+    int16_t tmp;
+    memcpy(&tmp, from, sizeof(tmp));
+    *chunk = _mm256_set1_epi16(tmp);
 }
 
 static inline void chunkmemset_4(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm256_set1_epi32(*(int32_t *)from);
+    int32_t tmp;
+    memcpy(&tmp, from, sizeof(tmp));
+    *chunk = _mm256_set1_epi32(tmp);
 }
 
 static inline void chunkmemset_8(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm256_set1_epi64x(*(int64_t *)from);
+    int64_t tmp;
+    memcpy(&tmp, from, sizeof(tmp));
+    *chunk = _mm256_set1_epi64x(tmp);
 }
 
 static inline void loadchunk(uint8_t const *s, chunk_t *chunk) {

diff --git a/arch/x86/chunkset_sse.c b/arch/x86/chunkset_sse.c
index d38e99dad6dcddbca7bef81ff655a59efaf04cdd..aa3478c09dafed63b995845b56481989609c3aa9 100644 (file)
--- a/arch/x86/chunkset_sse.c
+++ b/arch/x86/chunkset_sse.c
@@ -22,15 +22,21 @@ static inline void chunkmemset_1(uint8_t *from, chunk_t *chunk) {
 }
 
 static inline void chunkmemset_2(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm_set1_epi16(*(int16_t *)from);
+    int16_t tmp;
+    memcpy(&tmp, from, sizeof(tmp));
+    *chunk = _mm_set1_epi16(tmp);
 }
 
 static inline void chunkmemset_4(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm_set1_epi32(*(int32_t *)from);
+    int32_t tmp;
+    memcpy(&tmp, from, sizeof(tmp));
+    *chunk = _mm_set1_epi32(tmp);
 }
 
 static inline void chunkmemset_8(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm_set1_epi64x(*(int64_t *)from);
+    int64_t tmp;
+    memcpy(&tmp, from, sizeof(tmp));
+    *chunk = _mm_set1_epi64x(tmp);
 }
 
 static inline void loadchunk(uint8_t const *s, chunk_t *chunk) {