From: Sage Weil <sage@redhat.com>
Date: Tue, 25 Aug 2015 20:40:01 +0000 (-0400)
Subject: include/inline_memcpy: use __builtin_memcpy instead of explicit ptr copies
X-Git-Tag: v9.1.0~296^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=dfd142f2aa2c2b7790fa944edcd8abef3918f614;p=ceph.git

include/inline_memcpy: use __builtin_memcpy instead of explicit ptr copies

The gcc version of this is smart enough to handle any alignment
issues that exist and *should* generate the fastest code that is
general and correct.

On x86_64 this performs identically for the uint64_t
microbenchmark.

Suggested-by: Steve Capper <steve.capper@linaro.org>
Signed-off-by: Sage Weil <sage@redhat.com>
---

diff --git a/src/include/inline_memory.h b/src/include/inline_memory.h
index 3cdcae8859a0..33c6bc0ea0dc 100644
--- a/src/include/inline_memory.h
+++ b/src/include/inline_memory.h
@@ -16,10 +16,6 @@
 
 #if defined(__GNUC__)
 
-typedef struct __attribute__((__packed__)) { uint16_t val; } packed_uint16_t;
-typedef struct __attribute__((__packed__)) { uint32_t val; } packed_uint32_t;
-typedef struct __attribute__((__packed__)) { uint64_t val; } packed_uint64_t;
-
 // optimize for the common case, which is very small copies
 static inline void *maybe_inline_memcpy(void *dest, const void *src, size_t l,
 				       size_t inline_len)
@@ -33,30 +29,26 @@ void *maybe_inline_memcpy(void *dest, const void *src, size_t l,
   }
   switch (l) {
   case 8:
-    ((packed_uint64_t*)dest)->val = ((packed_uint64_t*)src)->val;
-    return dest;
+    return __builtin_memcpy(dest, src, 8);
   case 4:
-    ((packed_uint32_t*)dest)->val = ((packed_uint32_t*)src)->val;
-    return dest;
+    return __builtin_memcpy(dest, src, 4);
   case 3:
-    ((packed_uint16_t*)dest)->val = ((packed_uint16_t*)src)->val;
-    *((uint8_t*)((char*)dest+2)) = *((uint8_t*)((char*)src+2));
-    return dest;
+    return __builtin_memcpy(dest, src, 3);
   case 2:
-    ((packed_uint16_t*)dest)->val = ((packed_uint16_t*)src)->val;
-    return dest;
+    return __builtin_memcpy(dest, src, 2);
   case 1:
-    *((uint8_t*)(dest)) = *((uint8_t*)(src));
-    return dest;
+    return __builtin_memcpy(dest, src, 1);
   default:
     int cursor = 0;
     while (l >= sizeof(uint64_t)) {
-      *((uint64_t*)((char*)dest + cursor)) = *((uint64_t*)((char*)src + cursor));
+      __builtin_memcpy((char*)dest + cursor, (char*)src + cursor,
+		       sizeof(uint64_t));
       cursor += sizeof(uint64_t);
       l -= sizeof(uint64_t);
     }
     while (l >= sizeof(uint32_t)) {
-      *((uint32_t*)((char*)dest + cursor)) = *((uint32_t*)((char*)src + cursor));
+      __builtin_memcpy((char*)dest + cursor, (char*)src + cursor,
+		       sizeof(uint32_t));
       cursor += sizeof(uint32_t);
       l -= sizeof(uint32_t);
     }