Skip to content

Commit dfef124

Browse files
krzysz00easyonaadit
krzysz00
authored and
easyonaadit
committed
Reapply "[AMDGPU][GlobalISel] Fix load/store of pointer vectors, buffer.*.pN (llvm#110714)" v3 (llvm#114443)
This reverts commit 8a849a2. It seems I missed a spot when trying to ensure the code in the instruction selection tests were actually legalized MIR.
1 parent 16a639a commit dfef124

File tree

1 file changed

+94
-71
lines changed

1 file changed

+94
-71
lines changed

‎llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-global-old-legalization.mir‎

Lines changed: 94 additions & 71 deletions
Original file line numberDiff line numberDiff line change
@@ -681,61 +681,72 @@ body: |
681681
; GFX6-LABEL: name: load_global_s128
682682
; GFX6: liveins: $vgpr0_vgpr1
683683
; GFX6-NEXT: {{ $}}
684-
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
685-
; GFX6-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
686-
; GFX6-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
684+
; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
685+
; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
686+
; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
687+
; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
688+
; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
689+
; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
690+
; GFX6-NEXT: [[BUFFER_LOAD_DWORDX4_ADDR64_:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
691+
; GFX6-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUFFER_LOAD_DWORDX4_ADDR64_]]
687692
;
688693
; GFX7-LABEL: name: load_global_s128
689694
; GFX7: liveins: $vgpr0_vgpr1
690695
; GFX7-NEXT: {{ $}}
691-
; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
692-
; GFX7-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
693-
; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
696+
; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
697+
; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
698+
; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
699+
; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
700+
; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
701+
; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
702+
; GFX7-NEXT: [[BUFFER_LOAD_DWORDX4_ADDR64_:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
703+
; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUFFER_LOAD_DWORDX4_ADDR64_]]
694704
;
695705
; GFX7-FLAT-LABEL: name: load_global_s128
696706
; GFX7-FLAT: liveins: $vgpr0_vgpr1
697707
; GFX7-FLAT-NEXT: {{ $}}
698-
; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
699-
; GFX7-FLAT-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
700-
; GFX7-FLAT-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
708+
; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
709+
; GFX7-FLAT-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<4 x s32>), align 4, addrspace 1)
710+
; GFX7-FLAT-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
701711
;
702712
; GFX8-LABEL: name: load_global_s128
703713
; GFX8: liveins: $vgpr0_vgpr1
704714
; GFX8-NEXT: {{ $}}
705-
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
706-
; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
707-
; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
715+
; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
716+
; GFX8-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<4 x s32>), align 4, addrspace 1)
717+
; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
708718
;
709719
; GFX9-LABEL: name: load_global_s128
710720
; GFX9: liveins: $vgpr0_vgpr1
711721
; GFX9-NEXT: {{ $}}
712-
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
713-
; GFX9-NEXT: [[LOAD:%[0-9]+]]:vgpr(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
714-
; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
722+
; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
723+
; GFX9-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
724+
; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
715725
;
716726
; GFX10-LABEL: name: load_global_s128
717727
; GFX10: liveins: $vgpr0_vgpr1
718728
; GFX10-NEXT: {{ $}}
719-
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
720-
; GFX10-NEXT: [[LOAD:%[0-9]+]]:vgpr(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
721-
; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
729+
; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
730+
; GFX10-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
731+
; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
722732
;
723733
; GFX11-LABEL: name: load_global_s128
724734
; GFX11: liveins: $vgpr0_vgpr1
725735
; GFX11-NEXT: {{ $}}
726-
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
727-
; GFX11-NEXT: [[LOAD:%[0-9]+]]:vgpr(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
728-
; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
736+
; GFX11-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
737+
; GFX11-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
738+
; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
729739
;
730740
; GFX12-LABEL: name: load_global_s128
731741
; GFX12: liveins: $vgpr0_vgpr1
732742
; GFX12-NEXT: {{ $}}
733-
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
734-
; GFX12-NEXT: [[LOAD:%[0-9]+]]:vgpr(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
735-
; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
743+
; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
744+
; GFX12-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
745+
; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
736746
%0:vgpr(p1) = COPY $vgpr0_vgpr1
737-
%1:vgpr(s128) = G_LOAD %0 :: (load (s128), align 4, addrspace 1)
738-
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
747+
%1:vgpr(<4 x s32>) = G_LOAD %0 :: (load (<4 x s32>), align 4, addrspace 1)
748+
%2:vgpr(s128) = G_BITCAST %1(<4 x s32>)
749+
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %2
739750
740751
...
741752

@@ -989,61 +1000,72 @@ body: |
9891000
; GFX6-LABEL: name: load_global_v2p3
9901001
; GFX6: liveins: $vgpr0_vgpr1
9911002
; GFX6-NEXT: {{ $}}
992-
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
993-
; GFX6-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
994-
; GFX6-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1003+
; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1004+
; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
1005+
; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
1006+
; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
1007+
; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
1008+
; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
1009+
; GFX6-NEXT: [[BUFFER_LOAD_DWORDX2_ADDR64_:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<2 x s32>), addrspace 1)
1010+
; GFX6-NEXT: $vgpr0_vgpr1 = COPY [[BUFFER_LOAD_DWORDX2_ADDR64_]]
9951011
;
9961012
; GFX7-LABEL: name: load_global_v2p3
9971013
; GFX7: liveins: $vgpr0_vgpr1
9981014
; GFX7-NEXT: {{ $}}
999-
; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1000-
; GFX7-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1001-
; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1015+
; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1016+
; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
1017+
; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
1018+
; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
1019+
; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
1020+
; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
1021+
; GFX7-NEXT: [[BUFFER_LOAD_DWORDX2_ADDR64_:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<2 x s32>), addrspace 1)
1022+
; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[BUFFER_LOAD_DWORDX2_ADDR64_]]
10021023
;
10031024
; GFX7-FLAT-LABEL: name: load_global_v2p3
10041025
; GFX7-FLAT: liveins: $vgpr0_vgpr1
10051026
; GFX7-FLAT-NEXT: {{ $}}
1006-
; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1007-
; GFX7-FLAT-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1008-
; GFX7-FLAT-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1027+
; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1028+
; GFX7-FLAT-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s32>), addrspace 1)
1029+
; GFX7-FLAT-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
10091030
;
10101031
; GFX8-LABEL: name: load_global_v2p3
10111032
; GFX8: liveins: $vgpr0_vgpr1
10121033
; GFX8-NEXT: {{ $}}
1013-
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1014-
; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1015-
; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1034+
; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1035+
; GFX8-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s32>), addrspace 1)
1036+
; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
10161037
;
10171038
; GFX9-LABEL: name: load_global_v2p3
10181039
; GFX9: liveins: $vgpr0_vgpr1
10191040
; GFX9-NEXT: {{ $}}
1020-
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1021-
; GFX9-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1022-
; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1041+
; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1042+
; GFX9-NEXT: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec :: (load (<2 x s32>), addrspace 1)
1043+
; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
10231044
;
10241045
; GFX10-LABEL: name: load_global_v2p3
10251046
; GFX10: liveins: $vgpr0_vgpr1
10261047
; GFX10-NEXT: {{ $}}
1027-
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1028-
; GFX10-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1029-
; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1048+
; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1049+
; GFX10-NEXT: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec :: (load (<2 x s32>), addrspace 1)
1050+
; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
10301051
;
10311052
; GFX11-LABEL: name: load_global_v2p3
10321053
; GFX11: liveins: $vgpr0_vgpr1
10331054
; GFX11-NEXT: {{ $}}
1034-
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1035-
; GFX11-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1036-
; GFX11-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1055+
; GFX11-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1056+
; GFX11-NEXT: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec :: (load (<2 x s32>), addrspace 1)
1057+
; GFX11-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
10371058
;
10381059
; GFX12-LABEL: name: load_global_v2p3
10391060
; GFX12: liveins: $vgpr0_vgpr1
10401061
; GFX12-NEXT: {{ $}}
1041-
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1042-
; GFX12-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1043-
; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
1062+
; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1063+
; GFX12-NEXT: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec :: (load (<2 x s32>), addrspace 1)
1064+
; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
10441065
%0:vgpr(p1) = COPY $vgpr0_vgpr1
1045-
%1:vgpr(<2 x p3>) = G_LOAD %0 :: (load (<2 x p3>), align 8, addrspace 1)
1046-
$vgpr0_vgpr1 = COPY %1
1066+
%1:vgpr(<2 x s32>) = G_LOAD %0 :: (load (<2 x s32>), align 8, addrspace 1)
1067+
%2:vgpr(<2 x p3>) = G_BITCAST %1(<2 x s32>)
1068+
$vgpr0_vgpr1 = COPY %2
10471069
10481070
...
10491071

@@ -1231,7 +1253,7 @@ body: |
12311253
; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
12321254
; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
12331255
; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
1234-
; GFX6-NEXT: [[BUFFER_LOAD_DWORDX4_ADDR64_:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<8 x s16>), align 4, addrspace 1)
1256+
; GFX6-NEXT: [[BUFFER_LOAD_DWORDX4_ADDR64_:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
12351257
; GFX6-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUFFER_LOAD_DWORDX4_ADDR64_]]
12361258
;
12371259
; GFX7-LABEL: name: load_global_v8s16
@@ -1243,52 +1265,53 @@ body: |
12431265
; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
12441266
; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
12451267
; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
1246-
; GFX7-NEXT: [[BUFFER_LOAD_DWORDX4_ADDR64_:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<8 x s16>), align 4, addrspace 1)
1268+
; GFX7-NEXT: [[BUFFER_LOAD_DWORDX4_ADDR64_:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
12471269
; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUFFER_LOAD_DWORDX4_ADDR64_]]
12481270
;
12491271
; GFX7-FLAT-LABEL: name: load_global_v8s16
12501272
; GFX7-FLAT: liveins: $vgpr0_vgpr1
12511273
; GFX7-FLAT-NEXT: {{ $}}
12521274
; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1253-
; GFX7-FLAT-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<8 x s16>), align 4, addrspace 1)
1275+
; GFX7-FLAT-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<4 x s32>), align 4, addrspace 1)
12541276
; GFX7-FLAT-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
12551277
;
12561278
; GFX8-LABEL: name: load_global_v8s16
12571279
; GFX8: liveins: $vgpr0_vgpr1
12581280
; GFX8-NEXT: {{ $}}
1259-
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1260-
; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(<8 x s16>) = G_LOAD [[COPY]](p1) :: (load (<8 x s16>), align 4, addrspace 1)
1261-
; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<8 x s16>)
1281+
; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1282+
; GFX8-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<4 x s32>), align 4, addrspace 1)
1283+
; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
12621284
;
12631285
; GFX9-LABEL: name: load_global_v8s16
12641286
; GFX9: liveins: $vgpr0_vgpr1
12651287
; GFX9-NEXT: {{ $}}
1266-
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1267-
; GFX9-NEXT: [[LOAD:%[0-9]+]]:vgpr(<8 x s16>) = G_LOAD [[COPY]](p1) :: (load (<8 x s16>), align 4, addrspace 1)
1268-
; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<8 x s16>)
1288+
; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1289+
; GFX9-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
1290+
; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
12691291
;
12701292
; GFX10-LABEL: name: load_global_v8s16
12711293
; GFX10: liveins: $vgpr0_vgpr1
12721294
; GFX10-NEXT: {{ $}}
1273-
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1274-
; GFX10-NEXT: [[LOAD:%[0-9]+]]:vgpr(<8 x s16>) = G_LOAD [[COPY]](p1) :: (load (<8 x s16>), align 4, addrspace 1)
1275-
; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<8 x s16>)
1295+
; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1296+
; GFX10-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
1297+
; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
12761298
;
12771299
; GFX11-LABEL: name: load_global_v8s16
12781300
; GFX11: liveins: $vgpr0_vgpr1
12791301
; GFX11-NEXT: {{ $}}
1280-
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1281-
; GFX11-NEXT: [[LOAD:%[0-9]+]]:vgpr(<8 x s16>) = G_LOAD [[COPY]](p1) :: (load (<8 x s16>), align 4, addrspace 1)
1282-
; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<8 x s16>)
1302+
; GFX11-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1303+
; GFX11-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
1304+
; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
12831305
;
12841306
; GFX12-LABEL: name: load_global_v8s16
12851307
; GFX12: liveins: $vgpr0_vgpr1
12861308
; GFX12-NEXT: {{ $}}
1287-
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1288-
; GFX12-NEXT: [[LOAD:%[0-9]+]]:vgpr(<8 x s16>) = G_LOAD [[COPY]](p1) :: (load (<8 x s16>), align 4, addrspace 1)
1289-
; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<8 x s16>)
1309+
; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
1310+
; GFX12-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<4 x s32>), align 4, addrspace 1)
1311+
; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
12901312
%0:vgpr(p1) = COPY $vgpr0_vgpr1
1291-
%1:vgpr(<8 x s16>) = G_LOAD %0 :: (load (<8 x s16>), align 4, addrspace 1)
1313+
%1:vgpr(<4 x s32>) = G_LOAD %0 :: (load (<4 x s32>), align 4, addrspace 1)
1314+
%2:vgpr(<8 x s16>) = G_BITCAST %1(<4 x s32>)
12921315
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
12931316
12941317
...

0 commit comments

Comments
 (0)