; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck --check-prefix=GFX11 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck --check-prefix=GFX12 %s
define amdgpu_ps float @vimage_move_to_valu(<8 x i32> %rsrc) {
; GFX11-LABEL: name: vimage_move_to_valu
; GFX11: bb.0.bb:
; GFX11-NEXT: successors: %bb.1(0x80000000)
; GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_256 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3, [[COPY3]], %subreg.sub4, [[COPY2]], %subreg.sub5, [[COPY1]], %subreg.sub6, [[COPY]], %subreg.sub7
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sgpr_32 = S_MOV_B32 0
; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY9]], %subreg.sub1
; GFX11-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.1:
; GFX11-NEXT: successors: %bb.2(0x80000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
; GFX11-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
; GFX11-NEXT: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub4, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub5, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_4]], %subreg.sub0, [[V_READFIRSTLANE_B32_5]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE4]], [[REG_SEQUENCE]].sub4_sub5, implicit $exec
; GFX11-NEXT: [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_]], [[V_CMP_EQ_U64_e64_2]], implicit-def $scc
; GFX11-NEXT: [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub6, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub7, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_6]], %subreg.sub0, [[V_READFIRSTLANE_B32_7]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE5]], [[REG_SEQUENCE]].sub6_sub7, implicit $exec
; GFX11-NEXT: [[S_AND_B32_2:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_1]], [[V_CMP_EQ_U64_e64_3]], implicit-def $scc
; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:sgpr_256 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3, [[V_READFIRSTLANE_B32_4]], %subreg.sub4, [[V_READFIRSTLANE_B32_5]], %subreg.sub5, [[V_READFIRSTLANE_B32_6]], %subreg.sub6, [[V_READFIRSTLANE_B32_7]], %subreg.sub7
; GFX11-NEXT: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_2]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.2:
; GFX11-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[IMAGE_LOAD_V1_V2_gfx11_:%[0-9]+]]:vgpr_32 = IMAGE_LOAD_V1_V2_gfx11 [[REG_SEQUENCE1]], killed [[REG_SEQUENCE6]], 1, 1, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX11-NEXT: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
; GFX11-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.3:
; GFX11-NEXT: $exec_lo = S_MOV_B32 [[S_MOV_B32_1]]
; GFX11-NEXT: $vgpr0 = COPY [[IMAGE_LOAD_V1_V2_gfx11_]]
; GFX11-NEXT: SI_RETURN_TO_EPILOG $vgpr0
;
; GFX12-LABEL: name: vimage_move_to_valu
; GFX12: bb.0.bb:
; GFX12-NEXT: successors: %bb.1(0x80000000)
; GFX12-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
; GFX12-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
; GFX12-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
; GFX12-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX12-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX12-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_256 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3, [[COPY3]], %subreg.sub4, [[COPY2]], %subreg.sub5, [[COPY1]], %subreg.sub6, [[COPY]], %subreg.sub7
; GFX12-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.1:
; GFX12-NEXT: successors: %bb.2(0x80000000)
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
; GFX12-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
; GFX12-NEXT: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub4, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub5, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_4]], %subreg.sub0, [[V_READFIRSTLANE_B32_5]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub4_sub5, implicit $exec
; GFX12-NEXT: [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_]], [[V_CMP_EQ_U64_e64_2]], implicit-def $scc
; GFX12-NEXT: [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub6, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub7, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_6]], %subreg.sub0, [[V_READFIRSTLANE_B32_7]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE4]], [[REG_SEQUENCE]].sub6_sub7, implicit $exec
; GFX12-NEXT: [[S_AND_B32_2:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_1]], [[V_CMP_EQ_U64_e64_3]], implicit-def $scc
; GFX12-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:sgpr_256 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3, [[V_READFIRSTLANE_B32_4]], %subreg.sub4, [[V_READFIRSTLANE_B32_5]], %subreg.sub5, [[V_READFIRSTLANE_B32_6]], %subreg.sub6, [[V_READFIRSTLANE_B32_7]], %subreg.sub7
; GFX12-NEXT: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_2]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.2:
; GFX12-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[IMAGE_LOAD_V1_V2_gfx12_:%[0-9]+]]:vgpr_32 = IMAGE_LOAD_V1_V2_gfx12 [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_]], killed [[REG_SEQUENCE5]], 1, 1, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX12-NEXT: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
; GFX12-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.3:
; GFX12-NEXT: $exec_lo = S_MOV_B32 [[S_MOV_B32_]]
; GFX12-NEXT: $vgpr0 = COPY [[IMAGE_LOAD_V1_V2_gfx12_]]
; GFX12-NEXT: SI_RETURN_TO_EPILOG $vgpr0
bb:
%i = call float @llvm.amdgcn.image.load.2d.f32.i32(i32 1, i32 0, i32 0, <8 x i32> %rsrc, i32 0, i32 0)
ret float %i
}
define amdgpu_ps float @vsample_move_to_valu_rsrc(<8 x i32> %rsrc, <4 x i32> inreg %samp) {
; GFX11-LABEL: name: vsample_move_to_valu_rsrc
; GFX11: bb.0.main_body:
; GFX11-NEXT: successors: %bb.1(0x80000000)
; GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $sgpr0, $sgpr1, $sgpr2, $sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr7
; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr6
; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr5
; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr4
; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_256 = REG_SEQUENCE [[COPY11]], %subreg.sub0, [[COPY10]], %subreg.sub1, [[COPY9]], %subreg.sub2, [[COPY8]], %subreg.sub3, [[COPY7]], %subreg.sub4, [[COPY6]], %subreg.sub5, [[COPY5]], %subreg.sub6, [[COPY4]], %subreg.sub7
; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2, [[COPY]], %subreg.sub3
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.1:
; GFX11-NEXT: successors: %bb.2(0x80000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
; GFX11-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
; GFX11-NEXT: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub4, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub5, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_4]], %subreg.sub0, [[V_READFIRSTLANE_B32_5]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE4]], [[REG_SEQUENCE]].sub4_sub5, implicit $exec
; GFX11-NEXT: [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_]], [[V_CMP_EQ_U64_e64_2]], implicit-def $scc
; GFX11-NEXT: [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub6, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub7, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_6]], %subreg.sub0, [[V_READFIRSTLANE_B32_7]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE5]], [[REG_SEQUENCE]].sub6_sub7, implicit $exec
; GFX11-NEXT: [[S_AND_B32_2:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_1]], [[V_CMP_EQ_U64_e64_3]], implicit-def $scc
; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:sgpr_256 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3, [[V_READFIRSTLANE_B32_4]], %subreg.sub4, [[V_READFIRSTLANE_B32_5]], %subreg.sub5, [[V_READFIRSTLANE_B32_6]], %subreg.sub6, [[V_READFIRSTLANE_B32_7]], %subreg.sub7
; GFX11-NEXT: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_2]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.2:
; GFX11-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[IMAGE_SAMPLE_V1_V1_gfx11_:%[0-9]+]]:vgpr_32 = IMAGE_SAMPLE_V1_V1_gfx11 [[V_MOV_B32_e32_]], killed [[REG_SEQUENCE6]], [[REG_SEQUENCE1]], 1, 0, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX11-NEXT: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
; GFX11-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.3:
; GFX11-NEXT: $exec_lo = S_MOV_B32 [[S_MOV_B32_]]
; GFX11-NEXT: $vgpr0 = COPY [[IMAGE_SAMPLE_V1_V1_gfx11_]]
; GFX11-NEXT: SI_RETURN_TO_EPILOG $vgpr0
;
; GFX12-LABEL: name: vsample_move_to_valu_rsrc
; GFX12: bb.0.main_body:
; GFX12-NEXT: successors: %bb.1(0x80000000)
; GFX12-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $sgpr0, $sgpr1, $sgpr2, $sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX12-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX12-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX12-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX12-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr7
; GFX12-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr6
; GFX12-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr5
; GFX12-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr4
; GFX12-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX12-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX12-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_256 = REG_SEQUENCE [[COPY11]], %subreg.sub0, [[COPY10]], %subreg.sub1, [[COPY9]], %subreg.sub2, [[COPY8]], %subreg.sub3, [[COPY7]], %subreg.sub4, [[COPY6]], %subreg.sub5, [[COPY5]], %subreg.sub6, [[COPY4]], %subreg.sub7
; GFX12-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2, [[COPY]], %subreg.sub3
; GFX12-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.1:
; GFX12-NEXT: successors: %bb.2(0x80000000)
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
; GFX12-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
; GFX12-NEXT: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub4, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub5, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_4]], %subreg.sub0, [[V_READFIRSTLANE_B32_5]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE4]], [[REG_SEQUENCE]].sub4_sub5, implicit $exec
; GFX12-NEXT: [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_]], [[V_CMP_EQ_U64_e64_2]], implicit-def $scc
; GFX12-NEXT: [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub6, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub7, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_6]], %subreg.sub0, [[V_READFIRSTLANE_B32_7]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE5]], [[REG_SEQUENCE]].sub6_sub7, implicit $exec
; GFX12-NEXT: [[S_AND_B32_2:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_1]], [[V_CMP_EQ_U64_e64_3]], implicit-def $scc
; GFX12-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:sgpr_256 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3, [[V_READFIRSTLANE_B32_4]], %subreg.sub4, [[V_READFIRSTLANE_B32_5]], %subreg.sub5, [[V_READFIRSTLANE_B32_6]], %subreg.sub6, [[V_READFIRSTLANE_B32_7]], %subreg.sub7
; GFX12-NEXT: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_2]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.2:
; GFX12-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[IMAGE_SAMPLE_V1_V1_gfx12_:%[0-9]+]]:vgpr_32 = IMAGE_SAMPLE_V1_V1_gfx12 [[V_MOV_B32_e32_]], killed [[REG_SEQUENCE6]], [[REG_SEQUENCE1]], 1, 0, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX12-NEXT: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
; GFX12-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.3:
; GFX12-NEXT: $exec_lo = S_MOV_B32 [[S_MOV_B32_]]
; GFX12-NEXT: $vgpr0 = COPY [[IMAGE_SAMPLE_V1_V1_gfx12_]]
; GFX12-NEXT: SI_RETURN_TO_EPILOG $vgpr0
main_body:
%v = call float @llvm.amdgcn.image.sample.1d.f32.f32(i32 1, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
ret float %v
}
define amdgpu_ps float @vsample_move_to_valu_samp(<8 x i32> inreg %rsrc, <4 x i32> %samp) {
; GFX11-LABEL: name: vsample_move_to_valu_samp
; GFX11: bb.0.main_body:
; GFX11-NEXT: successors: %bb.1(0x80000000)
; GFX11-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr7
; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr6
; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr5
; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX11-NEXT: [[COPY8:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_256 = REG_SEQUENCE [[COPY11]], %subreg.sub0, [[COPY10]], %subreg.sub1, [[COPY9]], %subreg.sub2, [[COPY8]], %subreg.sub3, [[COPY7]], %subreg.sub4, [[COPY6]], %subreg.sub5, [[COPY5]], %subreg.sub6, [[COPY4]], %subreg.sub7
; GFX11-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2, [[COPY]], %subreg.sub3
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.1:
; GFX11-NEXT: successors: %bb.2(0x80000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub0, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub1, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE1]].sub0_sub1, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub2, implicit $exec
; GFX11-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub3, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; GFX11-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE1]].sub2_sub3, implicit $exec
; GFX11-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
; GFX11-NEXT: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.2:
; GFX11-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[IMAGE_SAMPLE_V1_V1_gfx11_:%[0-9]+]]:vgpr_32 = IMAGE_SAMPLE_V1_V1_gfx11 [[V_MOV_B32_e32_]], [[REG_SEQUENCE]], killed [[REG_SEQUENCE4]], 1, 0, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX11-NEXT: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
; GFX11-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.3:
; GFX11-NEXT: $exec_lo = S_MOV_B32 [[S_MOV_B32_]]
; GFX11-NEXT: $vgpr0 = COPY [[IMAGE_SAMPLE_V1_V1_gfx11_]]
; GFX11-NEXT: SI_RETURN_TO_EPILOG $vgpr0
;
; GFX12-LABEL: name: vsample_move_to_valu_samp
; GFX12: bb.0.main_body:
; GFX12-NEXT: successors: %bb.1(0x80000000)
; GFX12-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX12-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr7
; GFX12-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr6
; GFX12-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr5
; GFX12-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX12-NEXT: [[COPY8:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX12-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX12-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX12-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX12-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_256 = REG_SEQUENCE [[COPY11]], %subreg.sub0, [[COPY10]], %subreg.sub1, [[COPY9]], %subreg.sub2, [[COPY8]], %subreg.sub3, [[COPY7]], %subreg.sub4, [[COPY6]], %subreg.sub5, [[COPY5]], %subreg.sub6, [[COPY4]], %subreg.sub7
; GFX12-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX12-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2, [[COPY]], %subreg.sub3
; GFX12-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.1:
; GFX12-NEXT: successors: %bb.2(0x80000000)
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub0, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub1, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE1]].sub0_sub1, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub2, implicit $exec
; GFX12-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE1]].sub3, implicit $exec
; GFX12-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
; GFX12-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE1]].sub2_sub3, implicit $exec
; GFX12-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
; GFX12-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
; GFX12-NEXT: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_]], implicit-def $exec, implicit-def $scc, implicit $exec
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.2:
; GFX12-NEXT: successors: %bb.1(0x40000000), %bb.3(0x40000000)
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[IMAGE_SAMPLE_V1_V1_gfx12_:%[0-9]+]]:vgpr_32 = IMAGE_SAMPLE_V1_V1_gfx12 [[V_MOV_B32_e32_]], [[REG_SEQUENCE]], killed [[REG_SEQUENCE4]], 1, 0, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX12-NEXT: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
; GFX12-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: bb.3:
; GFX12-NEXT: $exec_lo = S_MOV_B32 [[S_MOV_B32_]]
; GFX12-NEXT: $vgpr0 = COPY [[IMAGE_SAMPLE_V1_V1_gfx12_]]
; GFX12-NEXT: SI_RETURN_TO_EPILOG $vgpr0
main_body:
%v = call float @llvm.amdgcn.image.sample.1d.f32.f32(i32 1, float 0.0, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
ret float %v
}
declare float @llvm.amdgcn.image.load.2d.f32.i32(i32 immarg, i32, i32, <8 x i32>, i32 immarg, i32 immarg)
declare float @llvm.amdgcn.image.sample.1d.f32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32)
Codebase Browser
llvm