Comments on: Google’s “go” simple & stupid benchmark (2nd round: memspeed)

By: Justin L

Justin L — Tue, 12 Jan 2010 03:39:24 +0000

no error there with python, anything you do in python will either be A) logic written in C that’s just invoked by python, or B) shockingly slow if your domain-specific logic is actually written in python

By: stenyak

Fri, 13 Nov 2009 08:01:04 +0000

I’ve tried zeroing the memory in the C version, and it doesn’t provoke any noticeable change on the results. Maybe if i run it a million times i’ll find out it’s 0.01% slower, but after 10 tests it does not make much difference:

$ cat hello.cpp && g++ hello.cpp && time for i in $(seq 10); do ./a.out >/dev/null ; done
#include
int main (int argc, char** argv)
{
const int size = 250;
int a[size] = {0};
int b[size] = {0};
int c[size] = {0};
for(int i=0;i for(int j=0;j for(int k=0;k c[k]+=a[i]*b[j];
}

real 0m0.782s
user 0m0.740s
sys 0m0.016s

It’s correct that there’s overhead loading the runtime, and it’s intentionally also part of these tests (that’s why i run the program several times). FYI, the most difference is in runtime, not in loadtime, even though the binnary sizes are spectacularly different.

It’s also true that the compiler was born just a year ago, but what can i do… it makes no sense to use gcc version v0.1 and python v0.1 in the benchmark, does it?

Anyway, remember the post title: simple and stupid

By: Matthias

Matthias — Fri, 13 Nov 2009 07:17:43 +0000

These aren’t equivalent tests. The C code doesn’t zero the memory, there’s some overhead loading the runtime, etc. Testing longer runs would get you closer, too. The 8g compiler isn’t exactly their speedy implementation either. It’s their research implementation feeding into a 10-year-old backend. GCC is a wee bit more optimized. Their gcc frontend will be faster.

By: binder

binder — Fri, 13 Nov 2009 02:25:55 +0000

managed to get some of the benchmark game tests run.
This is the go runtime as a percentage of the c runtime.
nbody 1087
threadring 6.32
binarytree 175
fannkuch 388
spectralnorm 397
mandelbrot 472
meteor 232
chameneos 49

By: kazbayadum

kazbayadum — Fri, 13 Nov 2009 00:33:15 +0000

C++ wins again! ^^

By: binder

binder — Thu, 12 Nov 2009 22:01:53 +0000

Realistically 3x slower is the same speed. Over a range of benchmarks they will probably swap places.

They have merged python and erlang… ehh… not that exciting.

By: stenyak.com » Blog Archive » Google’s “go” simple & stupid benchmark (1st round: I/O)

Thu, 12 Nov 2009 15:32:41 +0000

[...] in Round 2: memspeed Tags: computing, rant [en, es] | November 11th, 2009 [...]

By: stenyak

Thu, 12 Nov 2009 12:43:46 +0000

Yeah, python really sucked in this one.

Some pythoners will probably blame me for not knowing how to code (which is probably true). But hey, I just learned Go and it’s already 20 times faster. Let’s say development time is also a part of this benchmark

Anyway, out of curiosity, I researched how to optimize the python version. Tried a different syntax, tried some lambda functions, looked for alternative math libs, and i finally stumbled upon NumPy, which improves the situation. Ironically, NumPy core is coded in C. Pwnage.

By: juanval

juanval — Thu, 12 Nov 2009 11:34:24 +0000

Holy crap, python really chokes on that one! I suspect that might be because of the “for i in a” syntax.

Anyway, Go here takes “just” 3 times that of C++. Not half bad… but definitely not for high performance computing…