# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1031 -run-pass=register-coalescer -verify-coalescing -o - %s | FileCheck %s
# Make sure coalescing doesn't produce "no live segment at def" when
# there is a live out implicit_def with subranges.
# %1 will be coalesced into %0. %0 is a cross block implicit_def that
# cannot be deleted. The def of %0 in %bb.2 is a live out subregister
# def of the same register. We need to ensure that the resulting
# subrange for %0.sub0 includes the def in %bb.1
---
name: liveout_implicit_def_super_reg_redefine_sub0_implicit_def
tracksRegLiveness: true
body: |
; CHECK-LABEL: name: liveout_implicit_def_super_reg_redefine_sub0_implicit_def
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.2(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: S_CBRANCH_SCC0 %bb.2, implicit undef $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub0:sgpr_128 = S_MOV_B32 0
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub0:sgpr_128 = IMPLICIT_DEF
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: S_NOP 0, implicit [[S_MOV_B32_]]
; CHECK-NEXT: S_NOP 0, implicit [[S_MOV_B32_]].sub0
; CHECK-NEXT: S_ENDPGM 0
bb.0:
S_CBRANCH_SCC0 %bb.2, implicit undef $scc
bb.1:
%0:sgpr_128 = IMPLICIT_DEF
%1:sgpr_32 = S_MOV_B32 0
S_BRANCH %bb.3
bb.2:
undef %0.sub0:sgpr_128 = IMPLICIT_DEF
%1:sgpr_32 = COPY %0.sub0
bb.3:
S_NOP 0, implicit %0
S_NOP 0, implicit %1
S_ENDPGM 0
...
# Redef of sub0 is a meaningful value.
---
name: liveout_implicit_def_redefine_sub0_undef_other
tracksRegLiveness: true
body: |
; CHECK-LABEL: name: liveout_implicit_def_redefine_sub0_undef_other
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.2(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: S_CBRANCH_SCC0 %bb.2, implicit undef $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub0:sgpr_128 = S_MOV_B32 0
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub0:sgpr_128 = S_MOV_B32 9
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: S_NOP 0, implicit [[S_MOV_B32_]]
; CHECK-NEXT: S_NOP 0, implicit [[S_MOV_B32_]].sub0
; CHECK-NEXT: S_ENDPGM 0
bb.0:
S_CBRANCH_SCC0 %bb.2, implicit undef $scc
bb.1:
%0:sgpr_128 = IMPLICIT_DEF
%1:sgpr_32 = S_MOV_B32 0
S_BRANCH %bb.3
bb.2:
undef %0.sub0:sgpr_128 = S_MOV_B32 9
%1:sgpr_32 = COPY %0.sub0
bb.3:
S_NOP 0, implicit %0
S_NOP 0, implicit %1
S_ENDPGM 0
...
# The initial def of the register doesn't doesn't cover the redefined
# lanes. This had no error but was useful to compare against the
# failing cases.
---
name: only_redefine_undefined_lanes
tracksRegLiveness: true
body: |
; CHECK-LABEL: name: only_redefine_undefined_lanes
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.2(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: S_CBRANCH_SCC0 %bb.2, implicit undef $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: S_NOP 0, implicit-def undef %0.sub1_sub2_sub3
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]].sub0:vreg_128 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: undef [[V_MOV_B32_e32_:%[0-9]+]].sub0:vreg_128 = V_MOV_B32_e32 9, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: S_NOP 0, implicit [[V_MOV_B32_e32_]]
; CHECK-NEXT: S_NOP 0, implicit [[V_MOV_B32_e32_]].sub0
; CHECK-NEXT: S_ENDPGM 0
bb.0:
S_CBRANCH_SCC0 %bb.2, implicit undef $scc
bb.1:
S_NOP 0, implicit-def undef %0.sub1_sub2_sub3:vreg_128
%1:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
S_BRANCH %bb.3
bb.2:
undef %0.sub0:vreg_128 = V_MOV_B32_e32 9, implicit $exec
%1:vgpr_32 = COPY %0.sub0
bb.3:
S_NOP 0, implicit %0
S_NOP 0, implicit %1
S_ENDPGM 0
...
Codebase Browser
llvm