[9fans] faster vesa driver

[9fans] faster vesa driver

[geoff]

I've just pushed out kernel sources and binaries to incorporate
Aki's mtrr and vesa changes.  The combination makes monitor=vesa
run quite a bit faster; we saw a factor of three speed improvement
in one case.

It's still limited to a single processor, but we've got someone
investigating ways to fix that.

Re: [9fans] faster vesa driver

[erik quanstrom]

On Tue Jun  9 17:20:12 EDT 2009, geoff@plan9.bell-labs.com wrote:
> I've just pushed out kernel sources and binaries to incorporate
> Aki's mtrr and vesa changes.  The combination makes monitor=vesa
> run quite a bit faster; we saw a factor of three speed improvement
> in one case.
>
> It's still limited to a single processor, but we've got someone
> investigating ways to fix that.

very cool.  thanks, geoff.

- erik

Re: [9fans] faster vesa driver

[Venkatesh Srinivas]

> On Tue Jun  9 17:20:12 EDT 2009, geoff@plan9.bell-labs.com wrote:
>> I've just pushed out kernel sources and binaries to incorporate
>> Aki's mtrr and vesa changes.  The combination makes monitor=vesa
>> run quite a bit faster; we saw a factor of three speed improvement
>> in one case.

What got three times faster? Do you have some kind of tool to measure
draw performance overall?

(I ask because I'm currently working on a 'drawperf', some that runs
tests similar to x11perf and xrender on drawing speed, but I'm stuck
and was hoping something like this exists....)

>> It's still limited to a single processor, but we've got someone
>> investigating ways to fix that.

We have rsc's 8i in contrib; would executing VESA BIOS operations in
it be the best way to support MP?

Thanks very much!
-- vs

Re: [9fans] faster vesa driver

[erik quanstrom]

On Tue Jun  9 19:30:32 EDT 2009, me@acm.jhu.edu wrote:
> > On Tue Jun  9 17:20:12 EDT 2009, geoff@plan9.bell-labs.com wrote:
> >> I've just pushed out kernel sources and binaries to incorporate
> >> Aki's mtrr and vesa changes.  The combination makes monitor=vesa
> >> run quite a bit faster; we saw a factor of three speed improvement
> >> in one case.
>
> What got three times faster? Do you have some kind of tool to measure
> draw performance overall?

using andrey's screen saver hacks, i measured about an even greater
speedup using PAT, a technique similar to mtrrs but one that
uses page tables and not fixed physical ranges.  i did my tests with
andrey's xscreensaver hacks and timing catting large files to a fixed-sized
window.

i would attribute the perf difference between PAT and mtrr to be
a function of measurement and hardware — as in my crappy hardware.
if someone sees that one or the other is faster on the same hardware,
that would be very interesting to me.

- erik

Re: [9fans] faster vesa driver

[Kenji Arisawa]

Thanks Geoff,

That made great effect to my VIA LN10000EG!

Kenji Arisawa

On 2009/06/10, at 6:18, geoff@plan9.bell-labs.com wrote:

> I've just pushed out kernel sources and binaries to incorporate
> Aki's mtrr and vesa changes.  The combination makes monitor=vesa
> run quite a bit faster; we saw a factor of three speed improvement
> in one case.
>
> It's still limited to a single processor, but we've got someone
> investigating ways to fix that.
>

Re: [9fans] faster vesa driver

[Skip Tavakkolian]

> I've just pushed out kernel sources and binaries to incorporate
> Aki's mtrr and vesa changes.  The combination makes monitor=vesa
> run quite a bit faster; we saw a factor of three speed improvement
> in one case.

thank you!  i'm using it on a lenovo T61p; it definitely feels faster.

Re: [9fans] faster vesa driver

[Balwinder S Dheeman]

On 06/10/2009 02:51 AM, geoff@plan9.bell-labs.com wrote:
> I've just pushed out kernel sources and binaries to incorporate
> Aki's mtrr and vesa changes.  The combination makes monitor=vesa
> run quite a bit faster; we saw a factor of three speed improvement
> in one case.
>
> It's still limited to a single processor, but we've got someone
> investigating ways to fix that.

Great, but is in-compatible with QEMU.

Moreover, Plan9 does not boot off of a SiS SATA/USB2.0 DVD-RW or HDD's,
that's why I was forced to use QEMU.
/
--
Balwinder S "bdheeman" Dheeman        Registered Linux User: #229709
Anu'z Linux@HOME (Unix Shoppe)        Machines: #168573, 170593, 259192
Chandigarh, UT, 160062, India         Plan9, T2, Arch/Debian/FreeBSD/XP
Home: http://werc.homelinux.net/      Visit: http://counter.li.org/

Re: [9fans] faster vesa driver

[erik quanstrom]

> > I've just pushed out kernel sources and binaries to incorporate
> > Aki's mtrr and vesa changes.  The combination makes monitor=vesa
> > run quite a bit faster; we saw a factor of three speed improvement
> > in one case.
> >
> > It's still limited to a single processor, but we've got someone
> > investigating ways to fix that.
>
> Great, but is in-compatible with QEMU.

qemu already marks this memory WB with PAT
bits in the shadow PT.

http://www.slideshare.net/xen_com_mgr/xs-boston-2008-cache

> Moreover, Plan9 does not boot off of a SiS SATA/USB2.0 DVD-RW or HDD's,
> that's why I was forced to use QEMU.

does your chipset support ahci, and if so, is it in ahci mode?
that should work.  if it doesn't could you send me your
information off-line?

- erik

Re: [9fans] faster vesa driver

[James Tomaschke]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Balwinder S Dheeman wrote:
> Great, but is in-compatible with QEMU.

Using KVM(QEMU) it is working for me. At first I got some aux/vga error
on init, after recompiling /sys/src/cmd it seemed to work just fine.
Did not investigate any further.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkoyy7oACgkQl3JZTBu3/9lywwCdEU1uZ33njnS7PI8xMuLKPDfG
o+8AoJb+/9bdaZq/2eSKsVbY0WYV0NiF
=2cUR
-----END PGP SIGNATURE-----

Re: [9fans] faster vesa driver

[Kenji Arisawa]

Hello Geoff,

I am glad if mtrr makes effect only to single processor
until the mtrr can support multi-core processor.

Kenji Arisawa

On 2009/06/10, at 6:18, geoff@plan9.bell-labs.com wrote:

> I've just pushed out kernel sources and binaries to incorporate
> Aki's mtrr and vesa changes.  The combination makes monitor=vesa
> run quite a bit faster; we saw a factor of three speed improvement
> in one case.
>
> It's still limited to a single processor, but we've got someone
> investigating ways to fix that.
>

Re: [9fans] faster vesa driver

[erik quanstrom]

On Mon Jun 15 20:21:36 EDT 2009, arisawa@ar.aichi-u.ac.jp wrote:
> Hello Geoff,
>
> I am glad if mtrr makes effect only to single processor
> until the mtrr can support multi-core processor.
>
> Kenji Arisawa

mtrrs work on any number of processors.  it is plan 9's
vesa which does not.

- erik

Re: [9fans] faster vesa driver

[Venkatesh Srinivas]

> mtrrs work on any number of processors.  it is plan 9's
> vesa which does not.

Is this just because the driver uses realmode()? Or is there something else?

To solve the realmode() problem, we could link rsc's 8i into the
kernel and use that to execute the vesa bios code, right?

-- vs

Re: [9fans] faster vesa driver

[Ori Bernstein]

On Mon, 15 Jun 2009 22:46:42 -0400
Venkatesh Srinivas <me@acm.jhu.edu> wrote:

> > mtrrs work on any number of processors.  it is plan 9's
> > vesa which does not.
>
> Is this just because the driver uses realmode()? Or is there something else?
>
> To solve the realmode() problem, we could link rsc's 8i into the
> kernel and use that to execute the vesa bios code, right?
>
8i needs some work to add coverage for more instructions (for example,
movz). Other than that, it should be callable from userspace (say,
aux/vga) instead of being linked into the kernel.

--
    Ori Bernstein