On 11/7/19 1:06 PM, Peter Bergner wrote: > Before, LRA, we have an insn that sets a TImode pseudo with an integer > constant and a following insn that copies that TImode pseudo to a PTImode > pseudo. During LRA spilling, we generate a new insn that sets a PTImode > pseudo to that constant directly and we ICE because we do not recognize > that as a valid insn. The fix below fixes the ICE reported in PR92090 by > modifying our input_operand predicate to allow MODE_PARTIAL_INT modes for > integer constant input operands. > > This patch (preapproved by Segher) passed bootstrap and regtesting > with no errors. Committed.
I have now committed the backports to the GCC 9 and GCC 8 branches, including the test case updates from David. Peter
