[virglrenderer-devel] [PATCH v3 0/3] Fix GLES31 copy_image.non-compressed tests

[Gert Wollny]

Dear all, 

I've update the series to also include the 3-component texturs and to use 
the check for format compatibiliy like suggested by Gurchetan. I kept this in 
a seperate function though to have a common place to adress copying to, from, 
and between compressed textures too.

Regarding piglits, I see one regression with 

   fbo-generatemipmap-formats GL_ARB_texture_float

for the RGB16F format, but only when when I run some other piglits before, 
(e.g. fbo-generatemipmap-formats GL_ARB_depth_texture) which points at a state
leak in the GL blit fallback path, because by emulating RGB16* with RGBX16 the
blit that comes into play when the mipmaps are created is now emulated.
I think one can avoid that the code takes the fallback path in this case, but I
prefer to create another patch for this. 

The same state leak makes "fbo-generatemipmap-formats GL_ARB_depth_texture" 
unstable, i.e. all tests pass when running this piglit as the first after 
starting the VM, but later some fail (and not always the same). 

This series needs another series against mesa/virgl to expose the extension, 
and to take care of RGB32* texturs. Since these are needed for ARB_tbo_rgb32 
one can not replace them by RGBX32* texturs and needs to deal with the fact 
that they are not always available as render targets otherwise. 

thanks for reviewing, 
Gert 

Gert Wollny (3):
  vrend: If available use glCopyImageSubData to execute memcopy like
    blits
  vrend_formats: Replace RGB(8|16) formats with RGBX(8|16)
  virgl-caps: Report support for GL_ARB_copy_image to the guest

 src/virgl_hw.h       |  6 +++++
 src/vrend_formats.c  | 27 +++++++++++++++----
 src/vrend_renderer.c | 62 +++++++++++++++++++++++++++++++++++++++++---
 src/vrend_renderer.h |  2 ++
 4 files changed, 89 insertions(+), 8 deletions(-)

-- 
2.17.1