Skip to content

Commit df1b5ae

Browse files
saiislamsaiislam
authored
[AMDGPU][GlobalISel] Update tests to check for COV5 (#76257)
Update GlobalISel tests to assume ABI to be code object version 5.
1 parent aba40fb commit df1b5ae

18 files changed

+4383
-4458
lines changed

‎llvm/test/CodeGen/AMDGPU/GlobalISel/crash-stack-address-O0.ll

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
define amdgpu_kernel void @stack_write_fi() {
88
; CHECK-LABEL: stack_write_fi:
99
; CHECK: ; %bb.0: ; %entry
10-
; CHECK-NEXT: s_add_u32 s0, s0, s17
10+
; CHECK-NEXT: s_add_u32 s0, s0, s15
1111
; CHECK-NEXT: s_addc_u32 s1, s1, 0
1212
; CHECK-NEXT: s_mov_b32 s5, 0
1313
; CHECK-NEXT: s_mov_b32 s4, 0
@@ -23,3 +23,6 @@ entry:
2323
store volatile i64 0, ptr addrspace(5) %alloca, align 4
2424
ret void
2525
}
26+
27+
!llvm.module.flags = !{!0}
28+
!0 = !{i32 1, !"amdgpu_code_object_version", i32 500}

‎llvm/test/CodeGen/AMDGPU/GlobalISel/dropped_debug_info_assert.ll

Lines changed: 44 additions & 44 deletions
Original file line numberDiff line numberDiff line change
@@ -7,48 +7,47 @@ declare void @callee()
77
define amdgpu_kernel void @call_debug_loc() {
88
; CHECK-LABEL: name: call_debug_loc
99
; CHECK: bb.1.entry:
10-
; CHECK-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr0, $vgpr1, $vgpr2, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11
10+
; CHECK-NEXT: liveins: $sgpr12, $sgpr13, $sgpr14, $vgpr0, $vgpr1, $vgpr2, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9
1111
; CHECK-NEXT: {{ $}}
12-
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2, debug-location !6
13-
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1, debug-location !6
14-
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0, debug-location !6
15-
; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr16, debug-location !6
16-
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr15, debug-location !6
17-
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr14, debug-location !6
18-
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11, debug-location !6
19-
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7, debug-location !6
20-
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5, debug-location !6
21-
; CHECK-NEXT: [[COPY9:%[0-9]+]]:sreg_64 = COPY $sgpr8_sgpr9
22-
; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, debug-location !6
23-
; CHECK-NEXT: [[COPY10:%[0-9]+]]:sreg_64 = COPY [[COPY8]], debug-location !6
24-
; CHECK-NEXT: [[COPY11:%[0-9]+]]:sreg_64 = COPY [[COPY7]], debug-location !6
25-
; CHECK-NEXT: [[COPY12:%[0-9]+]]:sreg_64 = COPY [[COPY6]], debug-location !6
26-
; CHECK-NEXT: [[COPY13:%[0-9]+]]:sreg_32 = COPY [[COPY5]], debug-location !6
27-
; CHECK-NEXT: [[COPY14:%[0-9]+]]:sreg_32 = COPY [[COPY4]], debug-location !6
28-
; CHECK-NEXT: [[COPY15:%[0-9]+]]:sreg_32 = COPY [[COPY3]], debug-location !6
29-
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF debug-location !6
30-
; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 10, debug-location !6
31-
; CHECK-NEXT: [[COPY16:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]], debug-location !6
32-
; CHECK-NEXT: [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY16]], [[COPY1]], implicit $exec, debug-location !6
33-
; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 20, debug-location !6
34-
; CHECK-NEXT: [[COPY17:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]], debug-location !6
35-
; CHECK-NEXT: [[V_LSHLREV_B32_e64_1:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY17]], [[COPY]], implicit $exec, debug-location !6
36-
; CHECK-NEXT: [[V_OR3_B32_e64_:%[0-9]+]]:vgpr_32 = V_OR3_B32_e64 [[COPY2]], [[V_LSHLREV_B32_e64_]], [[V_LSHLREV_B32_e64_1]], implicit $exec, debug-location !6
37-
; CHECK-NEXT: [[COPY18:%[0-9]+]]:sgpr_128 = COPY $sgpr0_sgpr1_sgpr2_sgpr3, debug-location !6
38-
; CHECK-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY18]], debug-location !6
39-
; CHECK-NEXT: $sgpr4_sgpr5 = COPY [[COPY10]], debug-location !6
40-
; CHECK-NEXT: $sgpr6_sgpr7 = COPY [[COPY11]], debug-location !6
41-
; CHECK-NEXT: $sgpr8_sgpr9 = COPY [[COPY9]], debug-location !6
42-
; CHECK-NEXT: $sgpr10_sgpr11 = COPY [[COPY12]], debug-location !6
43-
; CHECK-NEXT: $sgpr12 = COPY [[COPY13]], debug-location !6
44-
; CHECK-NEXT: $sgpr13 = COPY [[COPY14]], debug-location !6
45-
; CHECK-NEXT: $sgpr14 = COPY [[COPY15]], debug-location !6
46-
; CHECK-NEXT: $sgpr15 = COPY [[DEF]], debug-location !6
47-
; CHECK-NEXT: $vgpr31 = COPY [[V_OR3_B32_e64_]], debug-location !6
48-
; CHECK-NEXT: [[SI_PC_ADD_REL_OFFSET:%[0-9]+]]:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-gotprel32-lo) @callee, target-flags(amdgpu-gotprel32-hi) @callee, implicit-def $scc, debug-location !6
49-
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[SI_PC_ADD_REL_OFFSET]], 0, 0, debug-location !6 :: (dereferenceable invariant load (p0) from got, addrspace 4)
50-
; CHECK-NEXT: $sgpr30_sgpr31 = noconvergent SI_CALL [[S_LOAD_DWORDX2_IMM]], @callee, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $vgpr31, debug-location !6
51-
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, debug-location !6
12+
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2, debug-location !7
13+
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1, debug-location !7
14+
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0, debug-location !7
15+
; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr14, debug-location !7
16+
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr13, debug-location !7
17+
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr12, debug-location !7
18+
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9, debug-location !7
19+
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5, debug-location !7
20+
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sreg_64 = COPY $sgpr6_sgpr7
21+
; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, debug-location !7
22+
; CHECK-NEXT: [[COPY9:%[0-9]+]]:sreg_64 = COPY [[COPY7]], debug-location !7
23+
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF debug-location !7
24+
; CHECK-NEXT: [[COPY10:%[0-9]+]]:sreg_64 = COPY [[COPY6]], debug-location !7
25+
; CHECK-NEXT: [[COPY11:%[0-9]+]]:sreg_32 = COPY [[COPY5]], debug-location !7
26+
; CHECK-NEXT: [[COPY12:%[0-9]+]]:sreg_32 = COPY [[COPY4]], debug-location !7
27+
; CHECK-NEXT: [[COPY13:%[0-9]+]]:sreg_32 = COPY [[COPY3]], debug-location !7
28+
; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF debug-location !7
29+
; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 10, debug-location !7
30+
; CHECK-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]], debug-location !7
31+
; CHECK-NEXT: [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY14]], [[COPY1]], implicit $exec, debug-location !7
32+
; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 20, debug-location !7
33+
; CHECK-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]], debug-location !7
34+
; CHECK-NEXT: [[V_LSHLREV_B32_e64_1:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY15]], [[COPY]], implicit $exec, debug-location !7
35+
; CHECK-NEXT: [[V_OR3_B32_e64_:%[0-9]+]]:vgpr_32 = V_OR3_B32_e64 [[COPY2]], [[V_LSHLREV_B32_e64_]], [[V_LSHLREV_B32_e64_1]], implicit $exec, debug-location !7
36+
; CHECK-NEXT: [[COPY16:%[0-9]+]]:sgpr_128 = COPY $sgpr0_sgpr1_sgpr2_sgpr3, debug-location !7
37+
; CHECK-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY16]], debug-location !7
38+
; CHECK-NEXT: $sgpr4_sgpr5 = COPY [[COPY9]], debug-location !7
39+
; CHECK-NEXT: $sgpr6_sgpr7 = COPY [[DEF]], debug-location !7
40+
; CHECK-NEXT: $sgpr8_sgpr9 = COPY [[COPY8]], debug-location !7
41+
; CHECK-NEXT: $sgpr10_sgpr11 = COPY [[COPY10]], debug-location !7
42+
; CHECK-NEXT: $sgpr12 = COPY [[COPY11]], debug-location !7
43+
; CHECK-NEXT: $sgpr13 = COPY [[COPY12]], debug-location !7
44+
; CHECK-NEXT: $sgpr14 = COPY [[COPY13]], debug-location !7
45+
; CHECK-NEXT: $sgpr15 = COPY [[DEF1]], debug-location !7
46+
; CHECK-NEXT: $vgpr31 = COPY [[V_OR3_B32_e64_]], debug-location !7
47+
; CHECK-NEXT: [[SI_PC_ADD_REL_OFFSET:%[0-9]+]]:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-gotprel32-lo) @callee, target-flags(amdgpu-gotprel32-hi) @callee, implicit-def $scc, debug-location !7
48+
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[SI_PC_ADD_REL_OFFSET]], 0, 0, debug-location !7 :: (dereferenceable invariant load (p0) from got, addrspace 4)
49+
; CHECK-NEXT: $sgpr30_sgpr31 = noconvergent SI_CALL [[S_LOAD_DWORDX2_IMM]], @callee, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $vgpr31, debug-location !7
50+
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, debug-location !7
5251
; CHECK-NEXT: S_ENDPGM 0
5352
entry:
5453
call void @callee(), !dbg !6
@@ -60,11 +59,11 @@ define void @returnaddress_debug_loc(ptr addrspace(1) %ptr) {
6059
; CHECK: bb.1.entry:
6160
; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $sgpr30_sgpr31
6261
; CHECK-NEXT: {{ $}}
63-
; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr30_sgpr31, debug-location !6
62+
; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr30_sgpr31, debug-location !7
6463
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
6564
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
6665
; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
67-
; CHECK-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]], debug-location !6
66+
; CHECK-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]], debug-location !7
6867
; CHECK-NEXT: GLOBAL_STORE_DWORDX2 [[REG_SEQUENCE]], [[COPY3]], 0, 0, implicit $exec :: (store (p0) into %ir.ptr, addrspace 1)
6968
; CHECK-NEXT: SI_RETURN
7069
entry:
@@ -78,7 +77,7 @@ declare ptr @llvm.returnaddress(i32 immarg) #0
7877
attributes #0 = { nofree nosync nounwind readnone willreturn }
7978

8079
!llvm.dbg.cu = !{!0}
81-
!llvm.module.flags = !{!2, !3, !4, !5}
80+
!llvm.module.flags = !{!2, !3, !4, !5, !10}
8281

8382
!0 = distinct !DICompileUnit(language: DW_LANG_OpenCL, file: !1, producer: "clang version 14.0.0 ([email protected]:llvm/llvm-project.git 4132dc917eddb446405cc5afef41167b8bce360b)", isOptimized: false, runtimeVersion: 0, emissionKind: NoDebug, splitDebugInlining: false, nameTableKind: None)
8483
!1 = !DIFile(filename: "gisel_1_gfx1031.cl", directory: "/home/matt/builds/conformance/2.0")
@@ -90,3 +89,4 @@ attributes #0 = { nofree nosync nounwind readnone willreturn }
9089
!7 = distinct !DISubprogram(name: "call_debug_loc", scope: !1, file: !1, line: 8, type: !8, scopeLine: 9, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !9)
9190
!8 = !DISubroutineType(types: !9)
9291
!9 = !{}
92+
!10 = !{i32 1, !"amdgpu_code_object_version", i32 500}

‎llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-assert-align.ll

Lines changed: 11 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -32,13 +32,13 @@ define void @call_result_align_1() {
3232
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr12
3333
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
3434
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
35-
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
35+
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr6_sgpr7
3636
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
3737
; CHECK-NEXT: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
3838
; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc
3939
; CHECK-NEXT: [[GV:%[0-9]+]]:_(p0) = G_GLOBAL_VALUE @returns_ptr
4040
; CHECK-NEXT: [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY8]]
41-
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]]
41+
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]](p4)
4242
; CHECK-NEXT: [[COPY11:%[0-9]+]]:_(p4) = COPY [[COPY6]]
4343
; CHECK-NEXT: [[COPY12:%[0-9]+]]:_(s64) = COPY [[COPY5]]
4444
; CHECK-NEXT: [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY4]]
@@ -81,13 +81,13 @@ define void @call_result_align_8() {
8181
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr12
8282
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
8383
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
84-
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
84+
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr6_sgpr7
8585
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
8686
; CHECK-NEXT: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
8787
; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc
8888
; CHECK-NEXT: [[GV:%[0-9]+]]:_(p0) = G_GLOBAL_VALUE @returns_ptr
8989
; CHECK-NEXT: [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY8]]
90-
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]]
90+
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]](p4)
9191
; CHECK-NEXT: [[COPY11:%[0-9]+]]:_(p4) = COPY [[COPY6]]
9292
; CHECK-NEXT: [[COPY12:%[0-9]+]]:_(s64) = COPY [[COPY5]]
9393
; CHECK-NEXT: [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY4]]
@@ -131,13 +131,13 @@ define void @declaration_result_align_8() {
131131
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr12
132132
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
133133
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
134-
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
134+
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr6_sgpr7
135135
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
136136
; CHECK-NEXT: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
137137
; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc
138138
; CHECK-NEXT: [[GV:%[0-9]+]]:_(p0) = G_GLOBAL_VALUE @returns_ptr_align8
139139
; CHECK-NEXT: [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY8]]
140-
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]]
140+
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]](p4)
141141
; CHECK-NEXT: [[COPY11:%[0-9]+]]:_(p4) = COPY [[COPY6]]
142142
; CHECK-NEXT: [[COPY12:%[0-9]+]]:_(s64) = COPY [[COPY5]]
143143
; CHECK-NEXT: [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY4]]
@@ -181,11 +181,11 @@ define ptr addrspace(1) @tail_call_assert_align() {
181181
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr12
182182
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
183183
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
184-
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
184+
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr6_sgpr7
185185
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
186186
; CHECK-NEXT: [[GV:%[0-9]+]]:ccr_sgpr_64(p0) = G_GLOBAL_VALUE @returns_ptr_align8
187187
; CHECK-NEXT: [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY8]]
188-
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]]
188+
; CHECK-NEXT: [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY7]](p4)
189189
; CHECK-NEXT: [[COPY11:%[0-9]+]]:_(p4) = COPY [[COPY6]]
190190
; CHECK-NEXT: [[COPY12:%[0-9]+]]:_(s64) = COPY [[COPY5]]
191191
; CHECK-NEXT: [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY4]]
@@ -209,3 +209,6 @@ entry:
209209
%call = tail call ptr addrspace(1) @returns_ptr_align8()
210210
ret ptr addrspace(1) %call
211211
}
212+
213+
!llvm.module.flags = !{!0}
214+
!0 = !{i32 1, !"amdgpu_code_object_version", i32 500}

‎llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-atomicrmw.ll

Lines changed: 5 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -30,8 +30,8 @@ define float @test_atomicrmw_fsub(ptr addrspace(3) %addr) {
3030
; CHECK-NEXT: bb.2.atomicrmw.start:
3131
; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.2(0x40000000)
3232
; CHECK-NEXT: {{ $}}
33-
; CHECK-NEXT: [[PHI:%[0-9]+]]:_(s64) = G_PHI %16(s64), %bb.2, [[C1]](s64), %bb.1
34-
; CHECK-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[LOAD]](s32), %bb.1, %14(s32), %bb.2
33+
; CHECK-NEXT: [[PHI:%[0-9]+]]:_(s64) = G_PHI %15(s64), %bb.2, [[C1]](s64), %bb.1
34+
; CHECK-NEXT: [[PHI1:%[0-9]+]]:_(s32) = G_PHI [[LOAD]](s32), %bb.1, %13(s32), %bb.2
3535
; CHECK-NEXT: [[FSUB:%[0-9]+]]:_(s32) = G_FSUB [[PHI1]], [[C]]
3636
; CHECK-NEXT: [[ATOMIC_CMPXCHG_WITH_SUCCESS:%[0-9]+]]:_(s32), [[ATOMIC_CMPXCHG_WITH_SUCCESS1:%[0-9]+]]:_(s1) = G_ATOMIC_CMPXCHG_WITH_SUCCESS [[COPY]](p3), [[PHI1]], [[FSUB]] :: (load store seq_cst seq_cst (s32) on %ir.addr, addrspace 3)
3737
; CHECK-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:_(s64) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.if.break), [[ATOMIC_CMPXCHG_WITH_SUCCESS1]](s1), [[PHI]](s64)
@@ -48,3 +48,6 @@ define float @test_atomicrmw_fsub(ptr addrspace(3) %addr) {
4848
%oldval = atomicrmw fsub ptr addrspace(3) %addr, float 1.0 seq_cst
4949
ret float %oldval
5050
}
51+
52+
!llvm.module.flags = !{!0}
53+
!0 = !{i32 1, !"amdgpu_code_object_version", i32 500}

0 commit comments

Comments
 (0)