Comment 1:

It looks like your C code is very quick.
From:
# hg st
# go test -run=NONE -bench=xxh
PASS
Benchmark_xxhash32      10000000               137 ns/op
Benchmark_goxxhash32    20000000                99.5 ns/op
ok      bitbucket.org/StephaneBunel/xxhash-go   3.618s
if I just put "return" at the start of you C function, like:
# hg st
M C/xxhash.c
# hg diff
diff -r c77f1e3b34a6 C/xxhash.c
--- a/C/xxhash.c        Sun Mar 10 18:58:14 2013 +0100
+++ b/C/xxhash.c        Mon Mar 25 13:57:49 2013 +1100
@@ -135,6 +135,7 @@
 unsigned int XXH32(const void* input, int len, unsigned int seed)
 {
+       return 0;
 #if 0
        // Simple version, good for code maintenance, but unfortunately slow for small inputs
        void* state = XXH32_init(seed);
# go test -run=NONE -bench=xxh
PASS
Benchmark_xxhash32      20000000               107 ns/op
Benchmark_goxxhash32    20000000                99.2 ns/op
ok      bitbucket.org/StephaneBunel/xxhash-go   4.349s
#
benchmark does not change much. So really you are not benchmarking your code, but how
fast Go to C interface is. You have discovered that linux is running switching code
faster then windows, and, perhaps, it is. Feel free to improve that, if you like.
Alex

Comment 2 by Vova616:

Yeah you are right, I'm searching right now why this is happening exactly
1.0.3 switch is faster than tip switch btw
go version go1.0.3 (386)
Benchmark_xxhash32      50000000                45.4 ns/op
Benchmark_goxxhash32     5000000               364 ns/op
go version devel +f78c5036b7f6 Mon Mar 25 08:58:13 2013 -0700 windows/386
Benchmark_xxhash32      50000000                69.6 ns/op
Benchmark_goxxhash32    50000000                61.0 ns/op

Comment 3:

Labels changed: added priority-later, removed priority-triage.

Status changed to Accepted.

Comment 4:

Labels changed: added performance.

Comment 5:

It's pretty hard to get excited about 50 nanoseconds.

Labels changed: added priority-someday, removed priority-later.

Comment 6:

Status changed to Unfortunate.