On 3 M?r, Michal Seliga wrote:
> Greetings
Hi, please activate a linewrap after ~72 characters in your mailer.
> I tried to compile lame 3.91 with gcc-3.0.1 and I tried to allow as
> much optimizations as possible while having as small result in 'make
> test' as possible
You didn't have to do that. "make test" is for development use only. It
varies depending on the used compiler, the used operating system, the
used libraries and on the version numbers of all of them. The number of
"make test" doesn't matter, a change of the number while still using the
same compiler/OS/libs matters. But this will only happen, if you modify
the source.
> I understand that -ffast-math may change something, but this
> -fno-caller-saves, may it be the compiler bug?
---snip---
-fcaller-saves
Enable values to be allocated in registers that
will be clobbered by function calls, by emitting
extra instructions to save and restore the regis-
ters around such calls. Such allocation is done
only when it seems to result in better code than
would otherwise be produced.
---snip---
The precission of a register may exceed the precission of the used C
type, e.g. floating point arithmetic on a Intel CPU may be done with
80bit precission, whereas a "double" in C is specified for 64bit
precission. If a value of some arithmetic gets converted to 64bit after
performing the entire arithmetic with 80bit may result in a slighly
different result than converting to 64bit at some stages while doing the
arithmetic.
Don't worry about it.
[-mcpu=athlon]
> Result was very bad, with 'pentiumpro' I have playtime/cputime 2.5 (or
> near by), with 'athlon' I have playtime/cputime 0.6 (or near by). Can
> be compiler bug or it simply doesn't make sense to use 3dnow for mp3
> encoding?
There are contraints the compiler has to obeye if he wants to use 3dnow.
Depending on them the 3dnow result may be slower than a non 3dnow
compile. It is some sort of a "bug" in the compiler (think of it like a
missing optimisation or "not as good as it should be" optimisation), but
the gcc guys know about it.
Bye,
Alexander.
--
The computer revolution is over. The computers won.
http://www.Leidinger.net Alexander @ Leidinger.net
GPG fingerprint = C518 BC70 E67F 143F BE91 3365 79E2 9C60 B006 3FE7
_______________________________________________
mp3encoder mailing list
[EMAIL PROTECTED]
http://minnie.tuhs.org/mailman/listinfo/mp3encoder
