msvcrt: Use unaligned data types in memset.

This fixes a regression in memset on ARM described by Martin: ARM can do 64 bit writes with the STRD instruction, but that instruction requires a 32 bit aligned address - while these stores are unaligned. Signed-off-by: Piotr Caban <piotr@codeweavers.com> Signed-off-by: Martin Storsjö <martin@martin.st> Signed-off-by: Alexandre Julliard <julliard@winehq.org>

msvcrt: Use unaligned data types in memset.
2c01333d · Piotr Caban · Alexandre Julliard · b14ba22a · 2c01333d
Commit 2c01333d authored Sep 16, 2021 by Piotr Caban Committed by Alexandre Julliard Sep 16, 2021
Hide whitespace changes
Inline Side-by-side

Showing with 18 additions and 14 deletions

string.c dlls/msvcrt/string.c +18 -14

No files found.
--- a/dlls/msvcrt/string.c
+++ b/dlls/msvcrt/string.c
@@ -2872,21 +2872,25 @@ static inline void memset_aligned_32(unsigned char *d, uint64_t v, size_t n)
 */
 void *__cdecl memset(void *dst, int c, size_t n)
 {
+    typedef uint64_t DECLSPEC_ALIGN(1) unaligned_ui64;
+    typedef uint32_t DECLSPEC_ALIGN(1) unaligned_ui32;
+    typedef uint16_t DECLSPEC_ALIGN(1) unaligned_ui16;
+
    uint64_t v = 0x101010101010101ull * (unsigned char)c;
    unsigned char *d = (unsigned char *)dst;
    size_t a = 0x20 - ((uintptr_t)d & 0x1f);

    if (n >= 16)
    {
-        *(uint64_t *)(d + 0) = v;
-        *(uint64_t *)(d + 8) = v;
-        *(uint64_t *)(d + n - 16) = v;
-        *(uint64_t *)(d + n - 8) = v;
+        *(unaligned_ui64 *)(d + 0) = v;
+        *(unaligned_ui64 *)(d + 8) = v;
+        *(unaligned_ui64 *)(d + n - 16) = v;
+        *(unaligned_ui64 *)(d + n - 8) = v;
        if (n <= 32) return dst;
-        *(uint64_t *)(d + 16) = v;
-        *(uint64_t *)(d + 24) = v;
-        *(uint64_t *)(d + n - 32) = v;
-        *(uint64_t *)(d + n - 24) = v;
+        *(unaligned_ui64 *)(d + 16) = v;
+        *(unaligned_ui64 *)(d + 24) = v;
+        *(unaligned_ui64 *)(d + n - 32) = v;
+        *(unaligned_ui64 *)(d + n - 24) = v;
        if (n <= 64) return dst;

        n = (n - a) & ~0x1f;
@@ -2895,20 +2899,20 @@ void *__cdecl memset(void *dst, int c, size_t n)
    }
    if (n >= 8)
    {
-        *(uint64_t *)d = v;
-        *(uint64_t *)(d + n - 8) = v;
+        *(unaligned_ui64 *)d = v;
+        *(unaligned_ui64 *)(d + n - 8) = v;
        return dst;
    }
    if (n >= 4)
    {
-        *(uint32_t *)d = v;
-        *(uint32_t *)(d + n - 4) = v;
+        *(unaligned_ui32 *)d = v;
+        *(unaligned_ui32 *)(d + n - 4) = v;
        return dst;
    }
    if (n >= 2)
    {
-        *(uint16_t *)d = v;
-        *(uint16_t *)(d + n - 2) = v;
+        *(unaligned_ui16 *)d = v;
+        *(unaligned_ui16 *)(d + n - 2) = v;
        return dst;
    }
    if (n >= 1)