c298f71ea6
It seems that there can be other cases with this that also can lead to wrong code (discovered with csmith). This time it involved not the kill flag but the undef flag. Use the intersection of the flags from both MachineOperand:s instead of the RegState from just one of them.
44 lines
1.3 KiB
YAML
44 lines
1.3 KiB
YAML
# RUN: llc -o - %s -mtriple=s390x-linux-gnu -mcpu=z15 -start-before=systemz-post-rewrite \
|
|
# RUN: -stop-after=machine-cp -verify-machineinstrs 2>&1 | FileCheck %s
|
|
|
|
# The chained SELRMux:es both has two operands with the same register but
|
|
# where one of the operands have been marked as undef (resulting from
|
|
# early-ifcvt). Check that the resulting COPY after machine-cp is from $r0l
|
|
# to $r2l.
|
|
|
|
# CHECK: name: fun0
|
|
# CHECK: $r2l = COPY $r0l
|
|
--- |
|
|
|
|
@Res = global i32 0, align 4
|
|
@Z = global i32 0, align 4
|
|
define signext i32 @fun0() { ret i32 0 }
|
|
...
|
|
---
|
|
name: fun0
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
successors: %bb.1(0x80000000)
|
|
|
|
renamable $r0l = LRL @Z :: (dereferenceable load (s32) from @Z)
|
|
renamable $r1l = LHIMux 1
|
|
|
|
bb.1:
|
|
successors: %bb.1(0x7c000000), %bb.2(0x04000000)
|
|
liveins: $r0l, $r1l
|
|
|
|
CHIMux renamable $r1l, 0, implicit-def $cc
|
|
renamable $r2l = SELRMux undef renamable $r0l, renamable $r0l, 14, 6, implicit $cc
|
|
renamable $r2l = SELRMux undef renamable $r2l, killed renamable $r2l, 14, 6, implicit $cc
|
|
BRC 14, 8, %bb.1, implicit killed $cc
|
|
J %bb.2
|
|
|
|
bb.2:
|
|
liveins: $r2l
|
|
|
|
STRL renamable $r2l, @Res :: (store (s32) into @Res)
|
|
renamable $r2d = LGFR killed renamable $r2l
|
|
Return implicit $r2d
|
|
...
|