26 Jul
2011
26 Jul
'11
10:15 a.m.
On Monday25/7/11 2:45 PM, Robin Gareus wrote:
Thanks for coming up with and sharing the tests!
Thanks for trying :->
would need to look into asm output.
As Stéphane said, i have to generate the assembler code.
I'll modify the scripts and send the assembler generated assembler code
later.
BTW. I'm pretty much impressed that some tests run
faster on vmware
virtualization compared to native OSX. But yeah it's also no big
surprise either (and may further be related to ubuntu using a never
version of gcc).
Yes, the compiler is different, and the code is a lot faster with more
recent gcc versions. Apple froze the environment to gcc 4.2 waiting for
clang. And, afaik, there is no reason for a vm to be slower in
computational code than the real iron, qs long as no system calls are done.
So are you now considering use some #ifdef to select
float/4 instead of
double/8 vectors in jMax or just change all of them?
Well, at the moment on gcc the perfomance with vector types is the same
as without vector types, so i'll leave the Linux version without vector
types (the code is #ifdef'ed).
It seems that it would be interesting to compile jMax with clang,
vector types and vector size 8 vector; i tried, but i find a lot of
problems with include files (a lot of system level includes are not
found by clang). I'll investigate this later :->.
By the way, i forgot to mentions that all my tests where at 64 bits;
i'll try later on a 32 bit Ubuntu.
About the -funroll-loops and the various buffer strategies, i'll give a
try and report later ..
Maurizio