Re: Does Xorg'x XRender implementation support better-than-bilinear interpolation?
Hi, None of Intel UXA, Radeon EXA or Nouveau EXA accelerate convolutions. Thanks for the clarification. It means I won't have to worry implementing something right now. Though my opinion would be that it would be better to add anisotropic filtering to Render as a standard filter with a software fallback instead of as an NVIDIA specific one. Anisotropic filtering provided by RENDER would be great :) Thanks, Clemens ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
Re: Does Xorg'x XRender implementation support better-than-bilinear interpolation?
Hi, On 11 September 2012 03:17, Pierre-Loup A. Griffais pgriff...@nvidia.com wrote: On 09/10/2012 06:08 PM, Daniel Stone wrote: 'Best' might be bicubic. It might be nearest. It might be anything, it's just whatever the driver thinks will offer the best quality. Bicubic is not required to be offered, but it may be. I'm not very familiar with bicubic interpolation, but couldn't it be achieved using the 'convolution' filter with the adequate kernel? (possibly in several passes at different scales). AFAIK 'convolution' is always provided and often accelerated. None of Intel UXA, Radeon EXA or Nouveau EXA accelerate convolutions. They're not particularly widely used anyway (at least by Cairo et al), probably because no-one accelerates them. Cheers, Daniel ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
Re: Does Xorg'x XRender implementation support better-than-bilinear interpolation?
Pierre-Loup A. Griffais pgriff...@nvidia.com writes: I'm not very familiar with bicubic interpolation, but couldn't it be achieved using the 'convolution' filter with the adequate kernel? (possibly in several passes at different scales). AFAIK 'convolution' is always provided and often accelerated. The convolution filter in Render only allows one phase of the filter to be used, so the quality of bicubic interpolation implemented that way would be terrible. As an aside, a bunch of code could be deleted if we removed the ability for drivers to provide their own filters. As far as I know, no driver has ever done this. I have an old branch here: http://cgit.freedesktop.org/~sandmann/xserver/log/?h=filters that could probably be resurrected. Soren ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
Re: Does Xorg'x XRender implementation support better-than-bilinear interpolation?
On 09/11/2012 10:43 AM, Søren Sandmann wrote: Pierre-Loup A. Griffaispgriff...@nvidia.com writes: I'm not very familiar with bicubic interpolation, but couldn't it be achieved using the 'convolution' filter with the adequate kernel? (possibly in several passes at different scales). AFAIK 'convolution' is always provided and often accelerated. The convolution filter in Render only allows one phase of the filter to be used, so the quality of bicubic interpolation implemented that way would be terrible. As an aside, a bunch of code could be deleted if we removed the ability for drivers to provide their own filters. As far as I know, no driver has ever done this. I have an old branch here: One of the items of my (long) list of things to do is to expose anisotropic filtering to RENDER, so it'd be cool if that didn't get removed. Thanks, - Pierre-Loup http://cgit.freedesktop.org/~sandmann/xserver/log/?h=filters that could probably be resurrected. Soren ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
Re: Does Xorg'x XRender implementation support better-than-bilinear interpolation?
Pierre-Loup A. Griffais pgriff...@nvidia.com writes: On 09/11/2012 10:43 AM, Søren Sandmann wrote: Pierre-Loup A. Griffaispgriff...@nvidia.com writes: I'm not very familiar with bicubic interpolation, but couldn't it be achieved using the 'convolution' filter with the adequate kernel? (possibly in several passes at different scales). AFAIK 'convolution' is always provided and often accelerated. The convolution filter in Render only allows one phase of the filter to be used, so the quality of bicubic interpolation implemented that way would be terrible. As an aside, a bunch of code could be deleted if we removed the ability for drivers to provide their own filters. As far as I know, no driver has ever done this. I have an old branch here: One of the items of my (long) list of things to do is to expose anisotropic filtering to RENDER, so it'd be cool if that didn't get removed. I am unlikely to do anything about this for the forseeable future, and I doubt anyone else cares. Though my opinion would be that it would be better to add anisotropic filtering to Render as a standard filter with a software fallback instead of as an NVIDIA specific one. Søren ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
Does Xorg'x XRender implementation support better-than-bilinear interpolation?
Hi, I just realized that at least on my machine (fedora 17 + intel 20.2), the only interpolation modes offered by Xrender are: {nearest , bilinear , convolution, fast , good, best} I always thought best translates into something better than bilinear, but it seems to be just equal to it. Is there any way to archive bicubic interpolation or any other higher-than-bilinear interpolation using xrender? As Java2D offers bicubic interpolation, it would mean a fallback to software-rendering every time a program requests high-quality interpolation. Thank you in advance, Clemens ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
Re: Does Xorg'x XRender implementation support better-than-bilinear interpolation?
On Mon, Sep 10, 2012 at 11:05 PM, Clemens Eisserer linuxhi...@gmail.com wrote: Hi, I just realized that at least on my machine (fedora 17 + intel 20.2), the only interpolation modes offered by Xrender are: {nearest , bilinear , convolution, fast , good, best} I always thought best translates into something better than bilinear, but it seems to be just equal to it. Is there any way to archive bicubic interpolation or any other higher-than-bilinear interpolation using xrender? As Java2D offers bicubic interpolation, it would mean a fallback to software-rendering every time a program requests high-quality interpolation. Thank you in advance, Clemens ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel I don't see it in the renderproto, so it's probably not possible without a spec extension. http://cgit.freedesktop.org/xorg/proto/renderproto/tree/render.h -- Far away from the primal instinct, the song seems to fade away, the river get wider between your thoughts and the things we do and say. ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
Re: Does Xorg'x XRender implementation support better-than-bilinear interpolation?
On 09/10/2012 06:08 PM, Daniel Stone wrote: Hi, On 10 September 2012 23:33, Maarten Maathuismadman2...@gmail.com wrote: On Mon, Sep 10, 2012 at 11:05 PM, Clemens Eissererlinuxhi...@gmail.com wrote: I just realized that at least on my machine (fedora 17 + intel 20.2), the only interpolation modes offered by Xrender are: {nearest , bilinear , convolution, fast , good, best} I always thought best translates into something better than bilinear, but it seems to be just equal to it. Is there any way to archive bicubic interpolation or any other higher-than-bilinear interpolation using xrender? As Java2D offers bicubic interpolation, it would mean a fallback to software-rendering every time a program requests high-quality interpolation. 'Best' might be bicubic. It might be nearest. It might be anything, it's just whatever the driver thinks will offer the best quality. Bicubic is not required to be offered, but it may be. I'm not very familiar with bicubic interpolation, but couldn't it be achieved using the 'convolution' filter with the adequate kernel? (possibly in several passes at different scales). AFAIK 'convolution' is always provided and often accelerated. Thanks, - Pierre-Loup I don't see it in the renderproto, so it's probably not possible without a spec extension. http://cgit.freedesktop.org/xorg/proto/renderproto/tree/render.h The filters are a list of strings; a protocol extension is not required for a driver to offer (and a client use) more filters than the predefined ones. Cheers, Daniel ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel ___ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel