[Bug 109081] [bisected] [HSW] Regression in clipping.user_defined.clip_* vulkancts tests

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=109081

--- Comment #2 from Ian Romanick <idr at freedesktop.org> ---
Ken and I looked at 
dEQP-VK.clipping.user_defined.clip_cull_distance_dynamic_index.vert_tess.1_7. 
There are two problems that appear to just be uncovered by this commit.

1. A loop that was previously unrolled is no longer unrolled.  There is code in
nir_loop_analysis to handle cases like

    ssa_50 = phi ssa_0, ssa_54

    if ssa_52 {
        ssa_53 = iadd ssa_50, ssa_1
    }
    ssa_54 = phi ssa_50, ssa_53

but it does not appear handle

    ssa_50 = phi ssa_0, ssa_54
    ssa_53 = iadd ssa_50, ssa_1
    ssa_54 = bcsel ssa_52, ssa_53, ssa_50

2. There is a local, compact array that does not get lowered by
nir_lower_indirect_derefs.  When the loop is unrolled, the temporary array is
eliminated, so this problem is not observable.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20181218/e2dc4979/attachment.html>