Skip to contentSkip to conversation

fs/inode: iunique() Optimize Performance

PATCH In review Liuweni 3 participants 3 messages Nov 25, 2009 VFS

lore
Share & tools
lore

Liuweni

author
#0 lore raw

PATCH 2/3 fs/inode: iunique() Optimize Performance

---
Change log:
Change the hash operation from division to shift. It will cost less time. Also, I change the divisor from L1_CACHE_BYTES to L1_CACHE_SHIFT. In the cache.h, the most L1_CACHE_BYTES defined as "(1 << L1_CACHE_SHIFT)".
---

---

fs/inode.c +2 -2
index 4d8e3be..397d65f 100644
@@ -605,8 +605,8 @@ static unsigned long hash(struct super_block *sb, unsigned long hashval)
 {
 	unsigned long tmp;
 
-	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
-			L1_CACHE_BYTES;
+	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) >>
+			L1_CACHE_SHIFT;
 	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
 	return tmp & I_HASHMASK;
 }

--------------
Best Regards,
Liuweni
2009-11-25

#1 lore raw

fs/inode.c static unsigned long hash(struct super_block *sb, unsigned long hashval) · line 612

Quoted 11 lines · Liuweni
fs/inode.c +2 -2
@@ -605,8 +605,8 @@ static unsigned long hash(struct super_block *sb, unsigned long hashval)
 {
 	unsigned long tmp;
 
-	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
-			L1_CACHE_BYTES;
+	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) >>
+			L1_CACHE_SHIFT;
 	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
 	return tmp & I_HASHMASK;
 }

Have you compared the compiler output before/after your change? I'd be amazed if GCC isn't able to optimise division-by-a-constant-power-of-two into shift-by-constant.

Signature · 4 lines
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."
#2 lore raw
Quoted 5 lines · Matthew Wilcox
Quoted 4 lines · Liuweni
fs/inode.c +0 -0
@@ -605,8 +605,8 @@ static unsigned long hash(struct super_block *sb, unsigned long hashval)
 {
 	unsigned long tmp;
 
Quoted 11 lines

fs/inode.c static unsigned long hash(struct super_block *sb, unsigned long hashval) · line 612

Quoted 7 lines
fs/inode.c +2 -2
-	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
-			L1_CACHE_BYTES;
+	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) >>
+			L1_CACHE_SHIFT;
 	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
 	return tmp & I_HASHMASK;
 }

Have you compared the compiler output before/after your change? I'd be amazed if GCC isn't able to optimise division-by-a-constant-power-of-two into shift-by-constant.

If a compiler can't do this nowadays, I'd consider it's a bug.

Signature · 1 line
Live like a child, think like the god.