[Mesa-dev] [PATCH] r600g: pad IBs to a multiple of 4 DWs on r6xx
Christian König
deathsimple at vodafone.de
Fri Sep 6 01:50:18 PDT 2013
Actually Type 2 packets are handled much faster on the R6xx compared to
most type 3 packets, cause they are handled by the PFP/fetch hw and
doesn't need to be forwarded to the ME.
Christian.
Am 06.09.2013 02:31, schrieb Dominik Behr:
> 0x80000000 is Type 2 NOP.
> You could make it a little better/faster by inserting single
> multi-DWORD Type 3 NOP
> And pad to 8 DWORDs. CP fetches are 32 bytes each and R600 has
> requires padding. Same with padding CP ring buffer updates to 32 bytes
> (pad to 32bytes before you update CP_RB_WPTR).
>
>
> On Thu, Sep 5, 2013 at 3:56 PM, Marek Olšák <maraeo at gmail.com
> <mailto:maraeo at gmail.com>> wrote:
>
> Reviewed-by: Marek Olšák <marek.olsak at amd.com
> <mailto:marek.olsak at amd.com>>
>
> Though I'm not sure if 0x80000000 is correct.
>
> Marek
>
> On Wed, Sep 4, 2013 at 11:55 PM, Alex Deucher
> <alexdeucher at gmail.com <mailto:alexdeucher at gmail.com>> wrote:
> > IBs need to be a multiple of 4 dwords on r6xx asics
> > to avoid a hw bug.
> >
> > Signed-off-by: Alex Deucher <alexander.deucher at amd.com
> <mailto:alexander.deucher at amd.com>>
> > CC: "9.2" <mesa-stable at lists.freedesktop.org
> <mailto:mesa-stable at lists.freedesktop.org>>
> > CC: "9.1" <mesa-stable at lists.freedesktop.org
> <mailto:mesa-stable at lists.freedesktop.org>>
> > ---
> > src/gallium/drivers/r600/r600_hw_context.c | 13 +++++++++++++
> > 1 file changed, 13 insertions(+)
> >
> > diff --git a/src/gallium/drivers/r600/r600_hw_context.c
> b/src/gallium/drivers/r600/r600_hw_context.c
> > index 97b0f9c..0a219af 100644
> > --- a/src/gallium/drivers/r600/r600_hw_context.c
> > +++ b/src/gallium/drivers/r600/r600_hw_context.c
> > @@ -347,6 +347,19 @@ void r600_context_flush(struct r600_context
> *ctx, unsigned flags)
> > flags |= RADEON_FLUSH_KEEP_TILING_FLAGS;
> > }
> >
> > + /* Pad the GFX CS to a multiple of 4 dwords on rv6xx
> > + * to avoid a hw bug.
> > + */
> > + if (ctx->chip_class < R700) {
> > + unsigned i;
> > + unsigned padding_dw = 4 - cs->cdw % 4;
> > + if (padding_dw < 4) {
> > + for (i = 0; i < padding_dw; i++) {
> > + cs->buf[cs->cdw++] = 0x80000000;
> > + }
> > + }
> > + }
> > +
> > /* Flush the CS. */
> > ctx->ws->cs_flush(ctx->rings.gfx.cs, flags,
> ctx->screen->cs_count++);
> > }
> > --
> > 1.8.3.1
> >
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> <mailto:mesa-dev at lists.freedesktop.org>
> > http://lists.freedesktop.org/mailman/listinfo/mesa-dev
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org <mailto:mesa-dev at lists.freedesktop.org>
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
>
>
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20130906/03d9b857/attachment-0001.html>
More information about the mesa-dev
mailing list