[PATCH] D54231: AMDGPU/InsertWaitcnts: Remove the dependence on MachineLoopInfo

[Nicolai Hähnle via Phabricator via llvm-commits]

nhaehnle created this revision.
nhaehnle added reviewers: msearles, rampitec, scott.linder, kanarayan.
Herald added subscribers: t-tye, tpr, dstuttard, yaxunl, wdng, jvesely, kzhuravl, arsenm.

MachineLoopInfo cannot be relied on for correctness, because it cannot
properly recognize loops in irreducible control flow which can be
introduced by late machine basic block optimization passes. See the new
test case for the reduced form of an example that occurred in practice.

Use a simple fixpoint iteration instead.

In order to facilitate this change, refactor WaitcntBrackets so that it
only tracks pending events and registers, rather than also maintaining
state that is relevant for the high-level algorithm. Various accessor
methods can be removed or made private as a consequence.

Affects (in radv):

- dEQP-VK.glsl.loops.special.{for,while}_uniform_iterations.select_iteration_count_{fragment,vertex}

Fixes: r345719 ("AMDGPU: Rewrite SILowerI1Copies to always stay on SALU")


Repository:
  rL LLVM

https://reviews.llvm.org/D54231

Files:
  lib/Target/AMDGPU/SIInsertWaitcnts.cpp
  test/CodeGen/AMDGPU/waitcnt-loop-irreducible.mir

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D54231.173032.patch
Type: text/x-patch
Size: 44229 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20181107/eb936298/attachment-0001.bin>