https://bugs.freedesktop.org/show_bug.cgi?id=99919
Matt Turner <[email protected]> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |[email protected]
--- Comment #3 from Matt Turner <[email protected]> ---
Thanks for the report!
I can reproduce on Haswell, in addition to your report on Sandybridge. I expect
this means it will apply to all generations supported by the i965 driver.
The first thing that pops out at me is in the NIR for the variant, we have:
vec1 32 ssa_0 = undefined
vec1 32 ssa_1 = undefined
vec1 32 ssa_2 = undefined
[snip]
loop {
block block_6:
/* preds: block_5 block_15 */
vec1 32 ssa_118 = phi block_5: ssa_0, block_15: ssa_130
vec1 32 ssa_119 = phi block_5: ssa_1, block_15: ssa_131
vec1 32 ssa_120 = phi block_5: ssa_2, block_15: ssa_132
whereas there are no undefined SSA values in the original. Since the undefined
values are flowing into the loop from the entrance, I believe this indeed
indicates a bug in our control flow analysis in NIR.
--
You are receiving this mail because:
You are the assignee for the bug.
You are the QA Contact for the bug._______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
