[ARM] Fix DAG combine loop in reduction distribution
authorDavid Green <david.green@arm.com>
Thu, 12 Aug 2021 15:37:39 +0000 (16:37 +0100)
committerDavid Green <david.green@arm.com>
Thu, 12 Aug 2021 15:37:39 +0000 (16:37 +0100)
commitae9a346ef8620851b54682b4b6738788af709578
treee8589d5878ddfe0b157bdf10e09e45cc76dbf071
parent876664317a4d116abc70ca686311b6a95aded834
[ARM] Fix DAG combine loop in reduction distribution

Given a constant operand, the MVE and DAGCombine combines could fight,
each redistributing in the opposite order. Add a guard to the MVE
vecreduce distribution to prevent that.
llvm/lib/Target/ARM/ARMISelLowering.cpp
llvm/test/CodeGen/Thumb2/mve-vecreduce-slp.ll