Skip to content

Commit 9f7960a

Browse files
krzysz00David Salinas
krzysz00
authored and
David Salinas
committed
Reapply "[AMDGPU][GlobalISel] Fix load/store of pointer vectors, buffer.*.pN (llvm#110714)" v2 (llvm#111708)
This adds `-disable-gisel-legality-check` to some gfx6 and gfx7 test lines to prevent behavior mismatches between debug and release builds The first attempted reapply was llvm#111059 This reverts commit e075dcf.
1 parent 5e7fd37 commit 9f7960a

File tree

1 file changed

+61
-84
lines changed

1 file changed

+61
-84
lines changed

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

Lines changed: 61 additions & 84 deletions
Original file line numberDiff line numberDiff line change
@@ -681,72 +681,61 @@ body: |
681681
; GFX6-LABEL: name: load_global_s128
682682
; GFX6: liveins: $vgpr0_vgpr1
683683
; GFX6-NEXT: {{ $}}
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_]]
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)
692687
;
693688
; GFX7-LABEL: name: load_global_s128
694689
; GFX7: liveins: $vgpr0_vgpr1
695690
; GFX7-NEXT: {{ $}}
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_]]
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)
704694
;
705695
; GFX7-FLAT-LABEL: name: load_global_s128
706696
; GFX7-FLAT: liveins: $vgpr0_vgpr1
707697
; GFX7-FLAT-NEXT: {{ $}}
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_]]
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)
711701
;
712702
; GFX8-LABEL: name: load_global_s128
713703
; GFX8: liveins: $vgpr0_vgpr1
714704
; GFX8-NEXT: {{ $}}
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_]]
705+
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
706+
; GFX8-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
707+
; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
718708
;
719709
; GFX9-LABEL: name: load_global_s128
720710
; GFX9: liveins: $vgpr0_vgpr1
721711
; GFX9-NEXT: {{ $}}
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_]]
712+
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
713+
; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
714+
; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
725715
;
726716
; GFX10-LABEL: name: load_global_s128
727717
; GFX10: liveins: $vgpr0_vgpr1
728718
; GFX10-NEXT: {{ $}}
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_]]
719+
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
720+
; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
721+
; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
732722
;
733723
; GFX11-LABEL: name: load_global_s128
734724
; GFX11: liveins: $vgpr0_vgpr1
735725
; GFX11-NEXT: {{ $}}
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_]]
726+
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
727+
; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
728+
; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
739729
;
740730
; GFX12-LABEL: name: load_global_s128
741731
; GFX12: liveins: $vgpr0_vgpr1
742732
; GFX12-NEXT: {{ $}}
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_]]
733+
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
734+
; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4, addrspace 1)
735+
; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
746736
%0:vgpr(p1) = COPY $vgpr0_vgpr1
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
737+
%1:vgpr(s128) = G_LOAD %0 :: (load (s128), align 4, addrspace 1)
738+
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
750739
751740
...
752741

@@ -1000,72 +989,61 @@ body: |
1000989
; GFX6-LABEL: name: load_global_v2p3
1001990
; GFX6: liveins: $vgpr0_vgpr1
1002991
; GFX6-NEXT: {{ $}}
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_]]
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>)
1011995
;
1012996
; GFX7-LABEL: name: load_global_v2p3
1013997
; GFX7: liveins: $vgpr0_vgpr1
1014998
; GFX7-NEXT: {{ $}}
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_]]
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>)
10231002
;
10241003
; GFX7-FLAT-LABEL: name: load_global_v2p3
10251004
; GFX7-FLAT: liveins: $vgpr0_vgpr1
10261005
; GFX7-FLAT-NEXT: {{ $}}
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_]]
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>)
10301009
;
10311010
; GFX8-LABEL: name: load_global_v2p3
10321011
; GFX8: liveins: $vgpr0_vgpr1
10331012
; GFX8-NEXT: {{ $}}
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_]]
1013+
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1014+
; GFX8-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1015+
; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
10371016
;
10381017
; GFX9-LABEL: name: load_global_v2p3
10391018
; GFX9: liveins: $vgpr0_vgpr1
10401019
; GFX9-NEXT: {{ $}}
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_]]
1020+
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1021+
; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1022+
; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
10441023
;
10451024
; GFX10-LABEL: name: load_global_v2p3
10461025
; GFX10: liveins: $vgpr0_vgpr1
10471026
; GFX10-NEXT: {{ $}}
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_]]
1027+
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1028+
; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1029+
; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
10511030
;
10521031
; GFX11-LABEL: name: load_global_v2p3
10531032
; GFX11: liveins: $vgpr0_vgpr1
10541033
; GFX11-NEXT: {{ $}}
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_]]
1034+
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1035+
; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1036+
; GFX11-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
10581037
;
10591038
; GFX12-LABEL: name: load_global_v2p3
10601039
; GFX12: liveins: $vgpr0_vgpr1
10611040
; GFX12-NEXT: {{ $}}
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_]]
1041+
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
1042+
; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_64(<2 x p3>) = G_LOAD [[COPY]](p1) :: (load (<2 x p3>), addrspace 1)
1043+
; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
10651044
%0:vgpr(p1) = COPY $vgpr0_vgpr1
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
1045+
%1:vgpr(<2 x p3>) = G_LOAD %0 :: (load (<2 x p3>), align 8, addrspace 1)
1046+
$vgpr0_vgpr1 = COPY %1
10691047
10701048
...
10711049

@@ -1253,7 +1231,7 @@ body: |
12531231
; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
12541232
; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
12551233
; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
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)
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)
12571235
; GFX6-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUFFER_LOAD_DWORDX4_ADDR64_]]
12581236
;
12591237
; GFX7-LABEL: name: load_global_v8s16
@@ -1265,53 +1243,52 @@ body: |
12651243
; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
12661244
; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
12671245
; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
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)
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)
12691247
; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUFFER_LOAD_DWORDX4_ADDR64_]]
12701248
;
12711249
; GFX7-FLAT-LABEL: name: load_global_v8s16
12721250
; GFX7-FLAT: liveins: $vgpr0_vgpr1
12731251
; GFX7-FLAT-NEXT: {{ $}}
12741252
; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
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)
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)
12761254
; GFX7-FLAT-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
12771255
;
12781256
; GFX8-LABEL: name: load_global_v8s16
12791257
; GFX8: liveins: $vgpr0_vgpr1
12801258
; GFX8-NEXT: {{ $}}
12811259
; 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)
1260+
; GFX8-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)
12831261
; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
12841262
;
12851263
; GFX9-LABEL: name: load_global_v8s16
12861264
; GFX9: liveins: $vgpr0_vgpr1
12871265
; GFX9-NEXT: {{ $}}
12881266
; 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)
1267+
; GFX9-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<8 x s16>), align 4, addrspace 1)
12901268
; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
12911269
;
12921270
; GFX10-LABEL: name: load_global_v8s16
12931271
; GFX10: liveins: $vgpr0_vgpr1
12941272
; GFX10-NEXT: {{ $}}
12951273
; 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)
1274+
; GFX10-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<8 x s16>), align 4, addrspace 1)
12971275
; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
12981276
;
12991277
; GFX11-LABEL: name: load_global_v8s16
13001278
; GFX11: liveins: $vgpr0_vgpr1
13011279
; GFX11-NEXT: {{ $}}
13021280
; 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)
1281+
; GFX11-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<8 x s16>), align 4, addrspace 1)
13041282
; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
13051283
;
13061284
; GFX12-LABEL: name: load_global_v8s16
13071285
; GFX12: liveins: $vgpr0_vgpr1
13081286
; GFX12-NEXT: {{ $}}
13091287
; 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)
1288+
; GFX12-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = GLOBAL_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec :: (load (<8 x s16>), align 4, addrspace 1)
13111289
; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[GLOBAL_LOAD_DWORDX4_]]
13121290
%0:vgpr(p1) = COPY $vgpr0_vgpr1
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>)
1291+
%1:vgpr(<8 x s16>) = G_LOAD %0 :: (load (<8 x s16>), align 4, addrspace 1)
13151292
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
13161293
13171294
...

0 commit comments

Comments
 (0)