On 27/11/17 13:01, Eric Anholt wrote:
Gert Wollny <gw.foss...@gmail.com> writes:
Array who's elements are only accessed directly are replaced by the
according number of temporary registers. By doing so the otherwise
reserved register range becomes subject to further optimizations like
copy propagation and register merging.
Thanks to the resulting reduced register pressure this patch makes
the piglits
spec/glsl-1.50/execution -
variable-indexing/vs-output-array-vec3-index-wr-before-gs
geometry/max-input-components
pass on r600 (barts) where they would fail before with a "GPR limit exceeded"
error.
Signed-off-by: Gert Wollny <gw.foss...@gmail.com>
---
No further changes are observed with the piglit shader subset.
PS: Submitter has no write access to mesa-git
This looks like something that should be done with (at most) a small
change to opt_array_splitting.cpp, rather than reimplementing it here.
Not sure how much it matters since I also have a NIR version of this
waiting review on the list, so a GLSL IR version is probably not going
to help with much. Either way there will be duplication.
Ultimately it would be nice to see the following, although I feel its
probably unlikely:
1. The ARB_gl_spriv work result partial support for a NIR based GLSL linker.
2. Remaining (non spriv) NIR GLSL linker support added.
3. Gallium drivers switch to nir->tgsi (probably solves the soft fp64
r600 support issues).
4. i915 gets a NIR backend.
5. Drop any unrequired GLSL IR linker support and now obsolete GLSL IR opts.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
