This is the mail archive of the
mailing list for the GCC project.
Re: [OT] GCC vs Intel C++ compiler benchmark
- From: Tim Prince <tprinceusa at mindspring dot com>
- To: Claus Fischer <claus dot fischer at clausfischer dot com>,gcc at gcc dot gnu dot org
- Date: Sun, 27 Jan 2002 08:41:56 -0800
- Subject: Re: [OT] GCC vs Intel C++ compiler benchmark
- References: <20020127124821.A25764@clausfischer.com>
- Reply-to: tprince at computer dot org
On Sunday 27 January 2002 03:48, Claus Fischer wrote:
> The rather biased article on http://www.open-mag.com/754088105111.htm
> compares Intel's C++ favourably to GCC, without disclosing many
> enlightening details.
> I'm not concerned with that but with the notable OS difference
> between Linux (SuSE 7.3) and Windows (XP Pro). IMHO a CPU bound
> benchmark should see less than 1 % influence from OS and C library.
> The graphics shows roughly 7% better performance on Windows for the
> same (Intel) compiler.
Both gcc and icc (beta versions available to customers) have improved
significantly since then.
> Are there any ABI differences which would justify such a difference?
> Register use conventions? Other stuff that you could identify?
glibc vs Windows libraries; greater maturity of the Windows version under
test; different default compiler options
> Unfortunately the benchmark gives no details about compiler options,
> so presumably the answer is `the benchmark is flawed' and `go ask
> them'. In this case don't bother to answer.
Probably many of the comparisons are deficient in matching compiler options
gcc icc icl
-fstrict-aliasing (default on) -ansi (default off) -Qansi (default off)
according to data type -pc64 (default 80) pc64, no choice
-march=pentiumpro -tpp6 -G6
-march=pentium4 (gcc-3.1) -tpp7 -G7
-msse2 (gcc-3.1) -xW -QxW
and the default for the Intel compilers resembles
gcc -ffast-math -Os -funroll-loops -mpreferred-stack-boundary=4
to mention only a few of the simpler comparisons which will affect performance