The previous behaviour was incorrect, and resulted in EXT_tfp being broken
for DDX drivers that implement the correct behaviour (intel/radeon/nouveau).
In the cases where a fake front buffer is required, the X server will
return one when requesting __DRI_BUFFER_FRONT_LEFT.
The Xorg state tracker (aka modesetting_drv) is likely broken now until
it's modified to match the other drivers.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This ensures that an assertion like
assert(expensive_test());
won't have any penalty on release builds. It also implies that no vital
code should be in assert expressions.
This is just a temporary change until we code generate the tile read/write
functions in runtime. The new code avoids an extra memcpy that exists in
u_tile.c functions, from which lp_tile_soa.c was originally based.
This achieves up to 5% improvement, particularly in frames with
little geometry overlap.
intel_pixel_read.c: In function ‘do_blit_readpixels’:
intel_pixel_read.c:221: warning: cast from pointer to integer of
different size
Cast via an intermediate (GLintptr) instead and hope the result fits
within GLuint... [It should as we simply do not support textures *that*
large!]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This should catch the common programming error where we attempt to
emit a relocation to beyond the end of the target buffer.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Same as in src/gallium/winsys/drm/intel/xorg/Makefile
Thanks MrCooper for explanations on IRC
[ Summary amended by Michel Dänzer to clarify that this is related to the xorg
state tracker ]
Besides from being necessary to use that extension, it also fixes a crash
when deleting the currently bound vertex array object.
Signed-off-by: Nicolai Hähnle <nhaehnle@gmail.com>
The non-KMS interface is to blame here. In theory, a proper fix
could be produced that works for the KMS interface only, but it
require cleaning a lot of mess. Easier to just do it right in r300g.
Signed-off-by: Nicolai Hähnle <nhaehnle@gmail.com>
This reverts commit df058298e1. It didn't
explain why it was required, doesnt appear to be required, and is a
significant performance penalty for cairo-gl firefox.
Conflicts:
src/mesa/main/fbobject.c
Currently using max_slots > 1 will cause synchronous rendering
if the driver flushes its command buffers between one bitmap and the
next.
Need to improve buffer_write to allow NO_WAIT (as well as no_flush)
updates to buffers where we know there is no conflict with previous
data.
Always test for PIPE_TRANSFER_READ/WRITE using the bit-wise and operator, and
add a pipe_transfer_buffer_flags() helper for getting the buffer usage flags
corresponding to them.
If we deleted a currently bound texture, we were always reverting the texture
binding to the default 1D texture rather than the proper default texture.
Need to check if we'll take the software path so which requires mapping the
src texture image.
Fixes crash in piglit gen-compressed-teximage, bug 24219. However, the
test still does not pass (it may never have).
