avutil/mem: Optimize fill32() by unrolling and using 64bit
authorMichael Niedermayer <michael@niedermayer.cc>
Thu, 17 Jan 2019 21:35:10 +0000 (22:35 +0100)
committerMichael Niedermayer <michael@niedermayer.cc>
Sun, 24 Mar 2019 09:39:03 +0000 (10:39 +0100)
Reviewed-by: Marton Balint <cus@passwd.hu>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
(cherry picked from commit 12b1338be376a3e5fb606d9fe41b58dc4a9e62c7)
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
libavutil/mem.c

index 1a8fc21..3b9b14f 100644 (file)
@@ -415,6 +415,18 @@ static void fill32(uint8_t *dst, int len)
 {
     uint32_t v = AV_RN32(dst - 4);
 
+#if HAVE_FAST_64BIT
+    uint64_t v2= v + ((uint64_t)v<<32);
+    while (len >= 32) {
+        AV_WN64(dst   , v2);
+        AV_WN64(dst+ 8, v2);
+        AV_WN64(dst+16, v2);
+        AV_WN64(dst+24, v2);
+        dst += 32;
+        len -= 32;
+    }
+#endif
+
     while (len >= 4) {
         AV_WN32(dst, v);
         dst += 4;