On Wed, 7 Nov 2012 13:27:50 +0000 Sundararaj Prabhu-B36876 <[email protected]> wrote:
> Wondering whether the extension EGL_KHR_image_pixmap can be used instead of > glTex(Sub)Image2D. > This would reduce much of the CPU copying. Do you see any problems? Hi, do you mean creating an EGLImage from a Dispmanx resource, and so eventually a GL texture? That would be ideal for switching between Dispmanx elements and GLESv2, if it works. It would allow removing one or two buffers per surface, saving lots of memory. However considering the current limitations in writing to Dispmanx resources, the data upload itself might be only slightly faster. But I can't imagine any serious negative effects. Maybe you might lose linear texture filtering; mipmaps are not even wanted. The problem is, I don't know what a native pixmap is on RPi, or is it implemented. It looks like the native pixmap type would be a uint32_t[2], but I don't know what to do with that. So yeah, if we had a buffer, that could be used as a texture and put into an overlay (element) as needed without copies, it would be a major win in memory usage, and we would not need to maintain a texture buffer and an overlay buffer separately. Also, the compositor would be able to release the client's buffer sooner, which means the client could do with only one buffer, too, instead of double-buffering towards the compositor. Thanks, pq _______________________________________________ wayland-devel mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/wayland-devel
