Align to 32B instead of 64B

No benefit to full 64B alignment, just more scalar ops.
zbjornson · Oct 23, 2018 · 1e9a3c8 · 1e9a3c8
1 parent 8c81322
commit 1e9a3c8
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/src/bswap.cc b/src/bswap.cc
@@ -69,9 +69,9 @@ static void shuffle(Local<TypedArray> data_ta) {
 	size_t byteLength = data_ta->ByteLength();
 	size_t elemLength = byteLength / sizeof(STYPE);
 
-	// Scalar until aligned to cache line
+	// Scalar until aligned
 	size_t sIdx = 0;
-	size_t preLength = ((uintptr_t)(void *)(bytes) % 64) / sizeof(STYPE);
+	size_t preLength = ((uintptr_t)(void *)(bytes) % 32) / sizeof(STYPE);
 	if (elemLength < preLength) preLength = elemLength;
 	while (sIdx < preLength) swap(&(*data)[sIdx++]);