Thomas Fletcher
|
RE: Performance versus compiler version
|
Thomas Fletcher
09/21/2007 1:34 PM
post1503
|
RE: Performance versus compiler version
> -----Original Message-----
> From: Mario Charest [mailto:mcharest@zinformatic.com]
> Sent: September 21, 2007 11:55 AM
> To: ostech-core_os
> Subject: Performance versus compiler version
>
>
> As anyone done a comparison of kernel's performance when
> compile with Watcom/2.95.3/3.3.5/4.2.1 ?
We have some numbers internally. I can see what we can do
about getting them up on the Wiki. It is more a matter of
format conversion than anything else.
Short story seems to be:
* Watcom rules size/speed/codegen for x86 over 2.*, 3.* series
* Starting with GCC 4.* we'll likely switch production builds
Thomas
|
|
|
Sada Murugan
|
Re: Performance versus compiler version
|
Sada Murugan
09/21/2007 1:57 PM
post1505
|
Re: Performance versus compiler version
I actually did a rather large and comprehensive comparison a few months ago, including the difference between
optimization levels.
If it's ok to post the full results here, then I will. I'm the only one with the original spreadsheet, because I posted
it as HTML before. I can tweak as necessary.
For x86, the watcom compiler performs much better for size and speed than 2.x or 3.x with any optimization level. 4.x
consistently outperforms 2.x and 3.x for speed by 2 or 3 percent. With the -Os option, it actually produces smaller
code than the watcom compiler.
There was negligible difference between 2.x and 3.x for all architectures, but 4.x consistency outperformed - usually 1
to 3 percent speed wise, and 8 percent or so for size. (-O2 and -Os only)
Interestingly, -O2 flag often had better speed performance than -O3 using gcc 4.2. The theory is that using -O3
generates huge binaries, and the pipeline suffers.
We'll see what we can do about posting the big chart of the results if people are interested.
|
|
|