[PATCH] Simplify XOR template selection

[PATCH] Simplify XOR template selection

[Josh Litherland]

[-- Attachment #1: Type: text/plain, Size: 664 bytes --]


I found this output confusing:

raid5: measuring checksumming speed
   8regs     :  1896.000 MB/sec
   8regs_prefetch:  2396.000 MB/sec
   32regs    :  1772.000 MB/sec
   32regs_prefetch:  1664.000 MB/sec
   pIII_sse  :  2352.000 MB/sec
   pII_mmx   :  4696.000 MB/sec
   p5_mmx    :  6252.000 MB/sec
raid5: using function: pIII_sse (2352.000 MB/sec)

since it seemed to be choosing a sub-optimal function.  I read the code
and figured out why this is the way it is.  However, to save others the
same confusion I had, and maybe even save a few picoseconds on startup,
I humbly submit the attached patch.

-- 
Josh Litherland <josh@emperorlinux.com>
Emperor Linux

[-- Attachment #2: xor_selection.diff --]
[-- Type: text/x-patch, Size: 1719 bytes --]

diff -drauN linux-2.6.6/drivers/md/xor.c linux-2.6.6.new/drivers/md/xor.c
--- linux-2.6.6/drivers/md/xor.c	2004-05-09 22:32:37.000000000 -0400
+++ linux-2.6.6.new/drivers/md/xor.c	2004-06-20 14:26:21.000000000 -0400
@@ -108,29 +108,40 @@
 	}
 	b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
 
-	printk(KERN_INFO "raid5: measuring checksumming speed\n");
+	/*
+	 * If this arch/cpu has a short-circuited selection, don't loop through all
+	 * the possible functions, just test the best one
+	 */
 
+	fastest = NULL;
+
+#ifdef XOR_SELECT_TEMPLATE
+		fastest = XOR_SELECT_TEMPLATE(fastest);
+#endif
+	
 #define xor_speed(templ)	do_xor_speed((templ), b1, b2)
 
-	XOR_TRY_TEMPLATES;
+	if (fastest) {
+		xor_speed(fastest);
+		printk("KERN_INFO raid5: automatically using best checksumming function: %s (%d.%03d MB/sec)\n",
+		       fastest->name, fastest->speed / 1000, fastest->speed % 1000);
+	} else {
+		printk(KERN_INFO "raid5: measuring checksumming speed\n");
+		XOR_TRY_TEMPLATES;
+		fastest = template_list;
+		for (f = fastest; f; f = f->next) 
+			if (f->speed > fastest->speed)
+				fastest = f;
+		printk("raid5: using function: %s (%d.%03d MB/sec)\n",
+		       fastest->name, fastest->speed / 1000, fastest->speed % 1000);
+
+	}
 
 #undef xor_speed
 
 	free_pages((unsigned long)b1, 2);
 
-	fastest = template_list;
-	for (f = fastest; f; f = f->next)
-		if (f->speed > fastest->speed)
-			fastest = f;
-
-#ifdef XOR_SELECT_TEMPLATE
-	fastest = XOR_SELECT_TEMPLATE(fastest);
-#endif
-
 	active_template = fastest;
-	printk("raid5: using function: %s (%d.%03d MB/sec)\n",
-	       fastest->name, fastest->speed / 1000, fastest->speed % 1000);
-
 	return 0;
 }
 

Re: [PATCH] Simplify XOR template selection

[Josh Litherland]

On Sun, 2004-06-20 at 14:47, Josh Litherland wrote:
> I humbly submit the attached patch.

Grrr... must fix the obvious typo in my printk...

-- 
Josh Litherland <josh@emperorlinux.com>
Emperor Linux