|
Chris Keating(deleted)
|
glScissor and tiling memory on SGX
|
|
Chris Keating(deleted)
05/14/2012 9:41 AM
post93044
|
glScissor and tiling memory on SGX
Are full-screen and partial-screen glScissor calls handled differently?
I ran into an issue where a lot of partial-screen glScissor calls where being made and the end result was that a lot of
stuff ended up missing from the end rasterization. Just setting the glScissor params to (0,0,800,478) instead of (0,0,
800,480) was enough to show this issue.
It kind of looks like some 64x64 macro-tiles are flat out getting dropped on the final render. Is it possible that the
partial-screen glScissor calls are handled differently than full-screen ones, and are causing the SGX to run out of
internal memory for its tiling lists in this case?
Have you seen cases where tiles/macro-tiles have been dropped altogether?
|
|
|
|
|
|
Derek Leach
|
RE: glScissor and tiling memory on SGX
|
|
Derek Leach
05/14/2012 9:52 AM
post93046
|
RE: glScissor and tiling memory on SGX
You are talking about screen graphics stack I assume, just to be clear?
-----Original Message-----
From: Chris Keating [mailto:community-noreply@qnx.com]
Sent: May 14, 2012 9:41 AM
To: opengles-graphics
Subject: glScissor and tiling memory on SGX
Are full-screen and partial-screen glScissor calls handled differently?
I ran into an issue where a lot of partial-screen glScissor calls where being made and the end result was that a lot of
stuff ended up missing from the end rasterization. Just setting the glScissor params to (0,0,800,478) instead of (0,0,
800,480) was enough to show this issue.
It kind of looks like some 64x64 macro-tiles are flat out getting dropped on the final render. Is it possible that the
partial-screen glScissor calls are handled differently than full-screen ones, and are causing the SGX to run out of
internal memory for its tiling lists in this case?
Have you seen cases where tiles/macro-tiles have been dropped altogether?
_______________________________________________
OpenGL ES
http://community.qnx.com/sf/go/post93044
To cancel your subscription to this discussion, please e-mail opengles-graphics-unsubscribe@community.qnx.com
|
|
|
|
|
|
Joel Pilon(deleted)
|
Re: glScissor and tiling memory on SGX
|
|
Joel Pilon(deleted)
05/14/2012 10:18 AM
post93049
|
Re: glScissor and tiling memory on SGX
Some of our older SGX drivers had issues with non-fullscreen glScissor
rects. However, glScissor performs badly on the SGX, and we recommend
using the stencil buffer to clip content.
-Joel
On 12-05-14 09:41 AM, Chris Keating wrote:
> Are full-screen and partial-screen glScissor calls handled differently?
>
> I ran into an issue where a lot of partial-screen glScissor calls where being made and the end result was that a lot
of stuff ended up missing from the end rasterization. Just setting the glScissor params to (0,0,800,478) instead of (0,
0,800,480) was enough to show this issue.
>
> It kind of looks like some 64x64 macro-tiles are flat out getting dropped on the final render. Is it possible that
the partial-screen glScissor calls are handled differently than full-screen ones, and are causing the SGX to run out of
internal memory for its tiling lists in this case?
>
> Have you seen cases where tiles/macro-tiles have been dropped altogether?
>
>
>
> _______________________________________________
>
> OpenGL ES
> http://community.qnx.com/sf/go/post93044
> To cancel your subscription to this discussion, please e-mail opengles-graphics-unsubscribe@community.qnx.com
|
|
|
|
|
|
Chris Keating(deleted)
|
RE: glScissor and tiling memory on SGX
|
|
Chris Keating(deleted)
05/14/2012 1:34 PM
post93056
|
RE: glScissor and tiling memory on SGX
Deric: Not screen per se. Just GL stack in general.
Thanks Joel. I'll have to play with that some more. I initially was using
the stencil buffer but actually noticed a performance decrease at the time
and so I went back to using the scissor.
Chris Keating
Crank Software Inc.
Office: 613-595-1999 x515, Cell: 613-314-6136
Online: www.cranksoftware.com
Check out: Crank Software's Blog
There is a better way to build user interfaces for embedded devices.
Download a 30 day evaluation of Crank Storyboard Suite today!
-----Original Message-----
From: Joel Pilon [mailto:community-noreply@qnx.com]
Sent: May-14-12 10:10 AM
To: opengles-graphics@community.qnx.com
Cc: Chris Keating; Chris Keating
Subject: Re: glScissor and tiling memory on SGX
Some of our older SGX drivers had issues with non-fullscreen glScissor
rects. However, glScissor performs badly on the SGX, and we recommend using
the stencil buffer to clip content.
-Joel
On 12-05-14 09:41 AM, Chris Keating wrote:
> Are full-screen and partial-screen glScissor calls handled differently?
>
> I ran into an issue where a lot of partial-screen glScissor calls where
being made and the end result was that a lot of stuff ended up missing from
the end rasterization. Just setting the glScissor params to (0,0,800,478)
instead of (0,0,800,480) was enough to show this issue.
>
> It kind of looks like some 64x64 macro-tiles are flat out getting dropped
on the final render. Is it possible that the partial-screen glScissor calls
are handled differently than full-screen ones, and are causing the SGX to
run out of internal memory for its tiling lists in this case?
>
> Have you seen cases where tiles/macro-tiles have been dropped altogether?
>
>
>
> _______________________________________________
>
> OpenGL ES
> http://community.qnx.com/sf/go/post93044
> To cancel your subscription to this discussion, please e-mail
> opengles-graphics-unsubscribe@community.qnx.com
_______________________________________________
OpenGL ES
http://community.qnx.com/sf/go/post93049
To cancel your subscription to this discussion, please e-mail
opengles-graphics-unsubscribe@community.qnx.com
-----
No virus found in this message.
Checked by AVG - www.avg.com
Version: 2012.0.2171 / Virus Database: 2425/4998 - Release Date: 05/14/12
|
|
|
|
|
|
Joel Pilon(deleted)
|
Re: glScissor and tiling memory on SGX
|
|
Joel Pilon(deleted)
05/14/2012 2:00 PM
post93060
|
Re: glScissor and tiling memory on SGX
Keep in mind, you should try to cache as many rects/masks in the 8 bit
stencil buffer. Multiple rects can be stored on each plane of the
stencil. Rendering the rects/mask is more expensive then the actual
stencil test. The stencil test is extremely cheap on PVR compared to
other architectures.
-Joel
On 12-05-14 01:17 PM, Chris Keating wrote:
> Deric: Not screen per se. Just GL stack in general.
>
> Thanks Joel. I'll have to play with that some more. I initially was using
> the stencil buffer but actually noticed a performance decrease at the time
> and so I went back to using the scissor.
>
> Chris Keating
> Crank Software Inc.
> Office: 613-595-1999 x515, Cell: 613-314-6136
> Online: www.cranksoftware.com
> Check out: Crank Software's Blog
> There is a better way to build user interfaces for embedded devices.
> Download a 30 day evaluation of Crank Storyboard Suite today!
>
> -----Original Message-----
> From: Joel Pilon [mailto:community-noreply@qnx.com]
> Sent: May-14-12 10:10 AM
> To: opengles-graphics@community.qnx.com
> Cc: Chris Keating; Chris Keating
> Subject: Re: glScissor and tiling memory on SGX
>
> Some of our older SGX drivers had issues with non-fullscreen glScissor
> rects. However, glScissor performs badly on the SGX, and we recommend using
> the stencil buffer to clip content.
>
> -Joel
>
> On 12-05-14 09:41 AM, Chris Keating wrote:
>> Are full-screen and partial-screen glScissor calls handled differently?
>>
>> I ran into an issue where a lot of partial-screen glScissor calls where
> being made and the end result was that a lot of stuff ended up missing from
> the end rasterization. Just setting the glScissor params to (0,0,800,478)
> instead of (0,0,800,480) was enough to show this issue.
>>
>> It kind of looks like some 64x64 macro-tiles are flat out getting dropped
> on the final render. Is it possible that the partial-screen glScissor calls
> are handled differently than full-screen ones, and are causing the SGX to
> run out of internal memory for its tiling lists in this case?
>>
>> Have you seen cases where tiles/macro-tiles have been dropped altogether?
>>
>>
>>
>> _______________________________________________
>>
>> OpenGL ES
>> http://community.qnx.com/sf/go/post93044
>> To cancel your subscription to this discussion, please e-mail
>> opengles-graphics-unsubscribe@community.qnx.com
>
>
>
>
> _______________________________________________
>
> OpenGL ES
> http://community.qnx.com/sf/go/post93049
> To cancel your subscription to this discussion, please e-mail
> opengles-graphics-unsubscribe@community.qnx.com
> -----
> No virus found in this message.
> Checked by AVG - www.avg.com
> Version: 2012.0.2171 / Virus Database: 2425/4998 - Release Date: 05/14/12
>
>
>
>
>
> _______________________________________________
>
> OpenGL ES
> http://community.qnx.com/sf/go/post93056
> To cancel your subscription to this discussion, please e-mail opengles-graphics-unsubscribe@community.qnx.com
|
|
|
|
|
|
