[Xen-devel] [PATCH 3 of 8] Optimize memcpy for x86 arch. If source buffers does not start at a 64

[<steven.smith@xxxxxxxxxx>]

# HG changeset patch
# User root@xxxxxxxxxxxxxxxx
# Date 1221082224 25200
# Node ID 999da363259a122e0c15bf28a0885ec40751e906
# Parent  c630fa8e84bdb0e1a79f7ccc42733ef5ea247cd2
Optimize memcpy for x86 arch. If source buffers does not start at a 64
bit boundary, copy a few bytes at the beginnig up to next 64-bit
boundary and then does an aligned copy for the remaining data. This
can reduce the copy cost by up to 50%.

Signed-off-by: Jose Renato Santos <jsantos@xxxxxxxxxx>

diff -r c630fa8e84bd -r 999da363259a xen/include/asm-x86/string.h
--- a/xen/include/asm-x86/string.h      Mon Sep 22 12:23:27 2008 +0100
+++ b/xen/include/asm-x86/string.h      Wed Sep 10 14:30:24 2008 -0700
@@ -96,13 +96,29 @@
 }
 
 #define __HAVE_ARCH_MEMCPY
+/* align source to a 64-bit boundary */
+static always_inline
+void *__var_memcpy(void *t, const void *f, size_t n)
+{
+    int off = (unsigned long)f & 0x7;
+    /* just do alignment if needed and if size is worth */
+    if ( (n > 32) && off ) {
+        size_t n1 = 8 - off;
+        __variable_memcpy(t, f, n1);
+        __variable_memcpy(t + n1, f + n1, n - n1);
+        return t;
+    } else {
+            return (__variable_memcpy(t, f, n));
+    }
+}
+
 #define memcpy(t,f,n) (__memcpy((t),(f),(n)))
 static always_inline
 void *__memcpy(void *t, const void *f, size_t n)
 {
     return (__builtin_constant_p(n) ?
             __constant_memcpy((t),(f),(n)) :
-            __variable_memcpy((t),(f),(n)));
+            __var_memcpy((t),(f),(n)));
 }
 
 /* Some version of gcc don't have this builtin. It's non-critical anyway. */

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel