Messages in this thread | | | From | nathan.zook@amd ... | Subject | RE: [patch]K6-2/K-III compile option | Date | Thu, 30 Sep 1999 15:43:28 -0500 |
| |
I did some checking. In general, the K6 (as opposed to the K6-2/K-III) should optimize identically to a Pentium with MMX.
3DNow! instructions were introduced with the K6-2, so it would be an error to enable them for the K6. I believe this patch has it right, but the comment at the top of what is quoted suggests otherwise.
I also noted what appear to be differences between the options selected for the Athlon and those recommended at http://aboutlinux.com/art_k7opt1_e.html. I wanted to be certain that you were aware of his results, in case he uncovered something extra.
These remarks are my own, based on a quick check and are NOT the voice of my company.
Nathan
-----Original Message----- From: Jeff Garzik [mailto:jgarzik@pobox.com] Sent: Thursday, September 30, 1999 11:49 AM To: Anders Fugmann Cc: linux-kernel@vger.rutgers.edu; alan@redhat.com Subject: Re: [patch]K6-2/K-III compile option
Anders Fugmann wrote: > Jeff Garzik wrote: > > I'm not sure CONFIG_X86_USE_3DNOW, which enables _MMX_ code, is a win on > > all K6's. Wasn't the original K6 rather slow when doing MMX? > > I beleive that the plain K6 is slower using mmx routines, but the > K6-2/K6-III should gain a bit. > This is why the options is only for the K6-2/K6-III, as mentioned in > Configure.help
Missed that, I skipped straight to the code :)
I think there is still room for improvement. '-march=k6' should be a win on all K6's. Also, CONFIG_X86_USE_3DNOW is new and not necessarily mature.
For those reasons I think that CONFIG_X86_USE_3DNOW should be an option, not a feature auto-selected by config.in. That will allow users to choose whether or not they want to use the new MMX routines.
if [ "$CONFIG_EXPERIMENTAL" = "y" ]; then if [ "$CONFIG_MK6" = "y" -o "$CONFIG_MK7" = "y" ]; then bool '3Dnow/MMX support routines' CONFIG_X86_USE_3DNOW fi fi
Take this with a grain of salt, Alan ignored a similar patchlet from me.
> BTW. > in the file arch/i386/lib/usercopy.c there is a "#ifdef > CONFIG_X86_USE_3DNOW_AND_WORKS" > which enables some mmx routines. I cannot find any other references to > "CONFIG_X86_USE_3DNOW_AND_WORKS". > Do you have an idea why this variable is used or is it just a typo? > Anyway... I've changed it to "#ifdef CONFIG_X86_USE_3DNOW", as you can > se below.
This looked to me like the semantic equivalent of '#if 0'. I would not remove "_AND_WORKS" without heavy testing on both K6 and K7.
> Also: > To whom should I send this patch, for it to be merged into the kernel > tree?
Until Linus gets back from vacation, Alan Cox <alan@redhat.com> is temporarily Head Penguin.
Regards,
Jeff
-- Custom driver development | Never worry about theory as long Open source programming | as the machinery does what it's | supposed to do. -- R. A. Heinlein
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |