Fixes piglit test glsl-vs-arrays-3 on Sandybridge, as well as garbage
rendering in 3DMarkMobileES 2.0's Taiji demo and GLBenchmark 2.0's
Egypt and PRO demos.
NOTE: This a candidate for stable release branches. It depends on
commit 9a21bc6401.
This was open-coded in three different places, and more are necessary.
Extract this into a function so it can be reused.
Unfortunately, not all variations were the same: in particular, one set
compression control and checked that the source register was not
ARF_NULL. This seemed like a good idea, so all cases now do so.
Eventually the miptree refcounting interface should be cleaned up.
The assymmetry dramatically increases the probability of bugs like
this. It should be made to like like libdrm refcounting or the
refcounting style used in other parts of Mesa.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=33046
Reviewed-by: Eric Anholt <eric@anholt.net>
Specifically, this ensures things like the front buffer actually exist. This
fixes piglt fbo/fbo-sys-blit and fd.o bug 35483.
Signed-off-by: Henri Verbeet <hverbeet@gmail.com>
GLSL 1.30 states clearly that only float and int are allowed, while the
GLSL ES specification's issues section states that sampler types may
take precision qualifiers.
Fixes compilation failures in 3DMarkMobileES 2.0 and GLBenchmark 2.0.
NOTE: This is a candidate for stable release branches.
Civilization 4's shaders make heavy use of gl_Color and don't use
perspective interpolation. This resulted in rivers, units, trees, and
so on being rendered almost entirely white. This is a regression
compared to the old fragment shader backend.
Found by inspection (comparing the old and new FS backend code).
References: https://bugs.freedesktop.org/show_bug.cgi?id=32949
NOTE: This is a candidate for the 7.10 branch.
Since GLSL IR allows multiple ir_variables to share the same name, we
need to generate unique names when printing the IR. Previously, we
always used %s@%p, appending the ir_variable's memory address.
While this worked, it had two drawbacks:
- When there aren't duplicates, the extra "@0x669a3e88" is useless
and makes the code harder to read.
- Real duplicates were hard to tell apart:
channel_expressions@0x6699e3c8 vs. channel_expressions@0x6699ddd8
We now append @2, @3, @4, and so on, but only where necessary to
distinguish duplicates. Since we only do this at print time, any
performance impact is irrelevant.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <idr@freedesktop.org>
While making some other changes in this area I was finding it annoying
each of these functions took mostly the same set of parameters in
differing orders.
'i' is already used for the outer loop. This caused some problems
while doing other work in this area. No bug exists here... until you
want to use the outer loop counter in the inner loop.
We were using typecasts because the functions pointers are polymorphic
in the second argument type, which.
Surprisingly the wrong calling convention didn't cause crashes on Windows,
but it was causing certain registers to be trashed in MSVC optimized
builds, when processing callists in the ClearView RC Flight Simulator.
I think the code ends up a lot more legible this way, though we've
still got the overloads in the fs_inst as well (even though there's
only one caller left currently).
If set to year X, only report extensions up to that year. This is a
work-around for games that try to copy the extensions string to a fixed
size buffer and overflow. If a game was released in year X, setting
MESA_EXTENSION_MAX_YEAR to that year will likely fix the problem.
