test/CodeGen/AMDGPU/indirect-addressing-undef.mir

# RUN: llc -march=amdgcn -verify-machineinstrs -run-pass si-lower-control-flow -o - %s | FileCheck %s
# Getting an undef that is specifically a VGPR is tricky from IR

# CHECK-LABEL: name: extract_undef_offset_vgpr{{$}}
# CHECK: bb.1:
# CHECK: successors: %bb.2(0x40000000 / 0x80000000 = 50.00%), %bb.1(0x40000000 / 0x80000000 = 50.00%)
# CHECK: liveins: %vgpr0_vgpr1_vgpr2_vgpr3{{$}}

# CHECK: V_READFIRSTLANE_B32 undef %vgpr10, implicit %exec
# CHECK: %vgpr0 = V_MOVRELS_B32_e32 %vgpr0, implicit %m0, implicit %exec, implicit %vgpr0_vgpr1_vgpr2_vgpr3
# CHECK: S_CBRANCH_EXECNZ %bb.1, implicit %exec

# CHECK: bb.2:
# CHECK: liveins: %sgpr6_sgpr7, %sgpr4_sgpr5_sgpr6_sgpr7, %sgpr4, %sgpr5, %sgpr6, %sgpr7, %sgpr4_sgpr5, %vgpr0_vgpr1_vgpr2_vgpr3, %vgpr0, %vgpr1, %vgpr2, %vgpr3, %vgpr0_vgpr1, %vgpr2_vgpr3, %vgpr0_vgpr1_vgpr2, %vgpr1_vgpr2, %vgpr1_vgpr2_vgpr3, %sgpr0_sgpr1, %sgpr0, %sgpr1{{$}}


--- |
  target datalayout = "e-p:32:32-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64"

  define void @extract_undef_offset_vgpr(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
  entry:
    %ld = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
    %value = extractelement <4 x i32> %ld, i32 undef
    store i32 %value, i32 addrspace(1)* %out
    ret void
  }

  define void @extract_undef_neg_offset_vgpr(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
  entry:
    %ld = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
    %value = extractelement <4 x i32> %ld, i32 undef
    store i32 %value, i32 addrspace(1)* %out
    ret void
  }

  define void @insert_undef_offset_vgpr(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
  entry:
    %ld = load <4 x i32>, <4 x i32> addrspace(1)* %in
    %value = insertelement <4 x i32> %ld, i32 5, i32 undef
    store <4 x i32> %value, <4 x i32> addrspace(1)* %out
    ret void
  }

  define void @insert_undef_neg_offset_vgpr(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
  entry:
    %ld = load <4 x i32>, <4 x i32> addrspace(1)* %in
    %value = insertelement <4 x i32> %ld, i32 5, i32 undef
    store <4 x i32> %value, <4 x i32> addrspace(1)* %out
    ret void
  }

  define void @insert_undef_value_offset_vgpr(<4 x i32> addrspace(1)*%out, <4 x i32> addrspace(1)* %in, i32 %idx) {
  entry:
    %ld = load <4 x i32>, <4 x i32> addrspace(1)* %in
    %value = insertelement <4 x i32> %ld, i32 undef, i32 %idx
    store <4 x i32> %value, <4 x i32> addrspace(1)* %out
    ret void
  }

...
---
name:            extract_undef_offset_vgpr
alignment:       0
exposesReturnsTwice: false
hasInlineAsm:    false
allVRegsAllocated: true
isSSA:           false
tracksRegLiveness: true
tracksSubRegLiveness: true
liveins:
  - { reg: '%sgpr0_sgpr1' }
frameInfo:
  isFrameAddressTaken: false
  isReturnAddressTaken: false
  hasStackMap:     false
  hasPatchPoint:   false
  stackSize:       0
  offsetAdjustment: 0
  maxAlignment:    0
  adjustsStack:    false
  hasCalls:        false
  maxCallFrameSize: 0
  hasOpaqueSPAdjustment: false
  hasVAStart:      false
  hasMustTailInVarArgFunc: false
body:             |
  bb.0.entry:
    liveins: %sgpr0_sgpr1

    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM %sgpr0_sgpr1, 11
    %sgpr7 = S_MOV_B32 61440
    %sgpr6 = S_MOV_B32 -1
    S_WAITCNT 127
    %vgpr0_vgpr1_vgpr2_vgpr3 = BUFFER_LOAD_DWORDX4_OFFSET %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec
    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM killed %sgpr0_sgpr1, 9
    S_WAITCNT 3952
    %vgpr0, dead %sgpr0_sgpr1 = SI_INDIRECT_SRC_V4 killed %vgpr0_vgpr1_vgpr2_vgpr3, undef %vgpr10, 0, implicit-def dead %exec, implicit-def dead %vcc, implicit-def dead %m0, implicit %exec
    S_WAITCNT 127
    BUFFER_STORE_DWORD_OFFSET killed %vgpr0, killed %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec
    S_ENDPGM

...

# CHECK-LABEL: name: extract_undef_neg_offset_vgpr{{$}}
# CHECK: bb.1:
# CHECK: successors: %bb.2(0x40000000 / 0x80000000 = 50.00%), %bb.1(0x40000000 / 0x80000000 = 50.00%)
# CHECK: liveins: %vgpr0_vgpr1_vgpr2_vgpr3{{$}}

# CHECK: %vcc_lo = V_READFIRSTLANE_B32 undef %vgpr10, implicit %exec
# CHECK: %m0 = S_MOV_B32 %vcc_lo
# CHECK: %m0 = S_ADD_I32 %m0, -7, implicit-def %scc
# CHECK: %vgpr0 = V_MOVRELS_B32_e32 %vgpr0, implicit %m0, implicit %exec, implicit %vgpr0_vgpr1_vgpr2_vgpr3
# CHECK: S_CBRANCH_EXECNZ %bb.1, implicit %exec

# CHECK: bb.2:
# CHECK: liveins: %sgpr6_sgpr7, %sgpr4_sgpr5_sgpr6_sgpr7, %sgpr4, %sgpr5, %sgpr6, %sgpr7, %sgpr4_sgpr5, %vgpr0_vgpr1_vgpr2_vgpr3, %vgpr0, %vgpr1, %vgpr2, %vgpr3, %vgpr0_vgpr1, %vgpr2_vgpr3, %vgpr0_vgpr1_vgpr2, %vgpr1_vgpr2, %vgpr1_vgpr2_vgpr3, %sgpr0_sgpr1, %sgpr0, %sgpr1

name:            extract_undef_neg_offset_vgpr
alignment:       0
exposesReturnsTwice: false
hasInlineAsm:    false
allVRegsAllocated: true
isSSA:           false
tracksRegLiveness: true
tracksSubRegLiveness: true
liveins:
  - { reg: '%sgpr0_sgpr1' }
frameInfo:
  isFrameAddressTaken: false
  isReturnAddressTaken: false
  hasStackMap:     false
  hasPatchPoint:   false
  stackSize:       0
  offsetAdjustment: 0
  maxAlignment:    0
  adjustsStack:    false
  hasCalls:        false
  maxCallFrameSize: 0
  hasOpaqueSPAdjustment: false
  hasVAStart:      false
  hasMustTailInVarArgFunc: false
body:             |
  bb.0.entry:
    liveins: %sgpr0_sgpr1

    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM %sgpr0_sgpr1, 11
    %sgpr7 = S_MOV_B32 61440
    %sgpr6 = S_MOV_B32 -1
    S_WAITCNT 127
    %vgpr0_vgpr1_vgpr2_vgpr3 = BUFFER_LOAD_DWORDX4_OFFSET %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec
    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM killed %sgpr0_sgpr1, 9
    S_WAITCNT 3952
    %vgpr0, dead %sgpr0_sgpr1 = SI_INDIRECT_SRC_V4 killed %vgpr0_vgpr1_vgpr2_vgpr3, undef %vgpr10, -7, implicit-def dead %exec, implicit-def dead %vcc, implicit-def dead %m0, implicit %exec
    S_WAITCNT 127
    BUFFER_STORE_DWORD_OFFSET killed %vgpr0, killed %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec
    S_ENDPGM

...

# CHECK-LABEL: name: insert_undef_offset_vgpr{{$}}
# CHECK: bb.1:
# CHECK: successors: %bb.2(0x40000000 / 0x80000000 = 50.00%), %bb.1(0x40000000 / 0x80000000 = 50.00%)
# CHECK: liveins: %vgpr4, %vgpr0_vgpr1_vgpr2_vgpr3{{$}}

# CHECK: %vcc_lo = V_READFIRSTLANE_B32 undef %vgpr10, implicit %exec
# CHECK: %m0 = S_MOV_B32 %vcc_lo
# CHECK: %vgpr0 = V_MOVRELD_B32_e32 %vgpr4, implicit %m0, implicit %exec, implicit %vgpr0_vgpr1_vgpr2_vgpr3
# CHECK: S_CBRANCH_EXECNZ %bb.1, implicit %exec

# CHECK: bb.2:
# CHECK: liveins: %sgpr6_sgpr7, %sgpr7, %sgpr4_sgpr5, %sgpr5, %sgpr4_sgpr5_sgpr6_sgpr7, %sgpr6, %sgpr4, %vgpr0_vgpr1_vgpr2_vgpr3, %vgpr0, %vgpr1, %vgpr2, %vgpr3, %vgpr0_vgpr1, %vgpr2_vgpr3, %vgpr0_vgpr1_vgpr2, %vgpr1_vgpr2, %vgpr1_vgpr2_vgpr3, %vgpr4, %sgpr0_sgpr1, %sgpr0, %sgpr1

name:            insert_undef_offset_vgpr
alignment:       0
exposesReturnsTwice: false
hasInlineAsm:    false
allVRegsAllocated: true
isSSA:           false
tracksRegLiveness: true
tracksSubRegLiveness: true
liveins:
  - { reg: '%sgpr0_sgpr1' }
frameInfo:
  isFrameAddressTaken: false
  isReturnAddressTaken: false
  hasStackMap:     false
  hasPatchPoint:   false
  stackSize:       0
  offsetAdjustment: 0
  maxAlignment:    0
  adjustsStack:    false
  hasCalls:        false
  maxCallFrameSize: 0
  hasOpaqueSPAdjustment: false
  hasVAStart:      false
  hasMustTailInVarArgFunc: false
body:             |
  bb.0.entry:
    liveins: %sgpr0_sgpr1

    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM %sgpr0_sgpr1, 11 :: (non-temporal invariant load 8 from `i64 addrspace(2)* undef`)
    %sgpr7 = S_MOV_B32 61440
    %sgpr6 = S_MOV_B32 -1
    %vgpr4 = V_MOV_B32_e32 5, implicit %exec
    S_WAITCNT 127
    %vgpr0_vgpr1_vgpr2_vgpr3 = BUFFER_LOAD_DWORDX4_OFFSET %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec :: (load 16 from %ir.in)
    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM killed %sgpr0_sgpr1, 9 :: (non-temporal invariant load 8 from `i64 addrspace(2)* undef`)
    S_WAITCNT 3952
    %vgpr0_vgpr1_vgpr2_vgpr3, dead %sgpr0_sgpr1 = SI_INDIRECT_DST_V4 %vgpr0_vgpr1_vgpr2_vgpr3, undef %vgpr10, 0, killed %vgpr4, implicit-def dead %exec, implicit-def dead %vcc, implicit-def dead %m0, implicit %exec
    S_WAITCNT 127
    BUFFER_STORE_DWORDX4_OFFSET killed %vgpr0_vgpr1_vgpr2_vgpr3, killed %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec :: (store 16 into %ir.out)
    S_ENDPGM

...

# CHECK-LABEL: name: insert_undef_neg_offset_vgpr{{$}}
# CHECK: bb.1:
# CHECK: successors: %bb.2(0x40000000 / 0x80000000 = 50.00%), %bb.1(0x40000000 / 0x80000000 = 50.00%)
# CHECK: liveins: %vgpr4, %vgpr0_vgpr1_vgpr2_vgpr3{{$}}

# CHECK: %vcc_lo = V_READFIRSTLANE_B32 undef %vgpr10, implicit %exec
# CHECK: %m0 = S_MOV_B32 %vcc_lo
# CHECK: %m0 = S_ADD_I32 %m0, -7, implicit-def %scc
# CHECK: %vgpr0 = V_MOVRELD_B32_e32 %vgpr4, implicit %m0, implicit %exec, implicit %vgpr0_vgpr1_vgpr2_vgpr3
# CHECK: S_CBRANCH_EXECNZ %bb.1, implicit %exec

# CHECK: bb.2:
# CHECK: liveins: %sgpr6_sgpr7, %sgpr7, %sgpr4_sgpr5, %sgpr5, %sgpr4_sgpr5_sgpr6_sgpr7, %sgpr6, %sgpr4, %vgpr0_vgpr1_vgpr2_vgpr3, %vgpr0, %vgpr1, %vgpr2, %vgpr3, %vgpr0_vgpr1, %vgpr2_vgpr3, %vgpr0_vgpr1_vgpr2, %vgpr1_vgpr2, %vgpr1_vgpr2_vgpr3, %vgpr4, %sgpr0_sgpr1, %sgpr0, %sgpr1{{$}}

name:            insert_undef_neg_offset_vgpr
alignment:       0
exposesReturnsTwice: false
hasInlineAsm:    false
allVRegsAllocated: true
isSSA:           false
tracksRegLiveness: true
tracksSubRegLiveness: true
liveins:
  - { reg: '%sgpr0_sgpr1' }
frameInfo:
  isFrameAddressTaken: false
  isReturnAddressTaken: false
  hasStackMap:     false
  hasPatchPoint:   false
  stackSize:       0
  offsetAdjustment: 0
  maxAlignment:    0
  adjustsStack:    false
  hasCalls:        false
  maxCallFrameSize: 0
  hasOpaqueSPAdjustment: false
  hasVAStart:      false
  hasMustTailInVarArgFunc: false
body:             |
  bb.0.entry:
    liveins: %sgpr0_sgpr1

    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM %sgpr0_sgpr1, 11 :: (non-temporal invariant load 8 from `i64 addrspace(2)* undef`)
    %sgpr7 = S_MOV_B32 61440
    %sgpr6 = S_MOV_B32 -1
    %vgpr4 = V_MOV_B32_e32 5, implicit %exec
    S_WAITCNT 127
    %vgpr0_vgpr1_vgpr2_vgpr3 = BUFFER_LOAD_DWORDX4_OFFSET %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec :: (load 16 from %ir.in)
    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM killed %sgpr0_sgpr1, 9 :: (non-temporal invariant load 8 from `i64 addrspace(2)* undef`)
    S_WAITCNT 3952
    %vgpr0_vgpr1_vgpr2_vgpr3, dead %sgpr0_sgpr1 = SI_INDIRECT_DST_V4 %vgpr0_vgpr1_vgpr2_vgpr3, undef %vgpr10, -7, killed %vgpr4, implicit-def dead %exec, implicit-def dead %vcc, implicit-def dead %m0, implicit %exec
    S_WAITCNT 127
    BUFFER_STORE_DWORDX4_OFFSET killed %vgpr0_vgpr1_vgpr2_vgpr3, killed %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec :: (store 16 into %ir.out)
    S_ENDPGM

...

# CHECK-LABEL: insert_undef_value_offset_vgpr{{$}}
# CHECK: bb.1:
# CHECK: successors: %bb.2(0x40000000 / 0x80000000 = 50.00%), %bb.1(0x40000000 / 0x80000000 = 50.00%)
# CHECK: liveins: %vgpr4, %vgpr0_vgpr1_vgpr2_vgpr3{{$}}

# CHECK: %vcc_lo = V_READFIRSTLANE_B32 %vgpr4, implicit %exec
# CHECK: %m0 = S_MOV_B32 %vcc_lo
# CHECK: %vgpr0 = V_MOVRELD_B32_e32 undef %vgpr10, implicit %m0, implicit %exec, implicit %vgpr0_vgpr1_vgpr2_vgpr3
# CHECK: S_CBRANCH_EXECNZ %bb.1, implicit %exec

# CHECK: bb.2:
# CHECK: liveins: %sgpr6_sgpr7, %sgpr7, %sgpr4_sgpr5, %sgpr5, %sgpr4_sgpr5_sgpr6_sgpr7, %sgpr6, %sgpr4, %vgpr0_vgpr1_vgpr2_vgpr3, %vgpr0, %vgpr1, %vgpr2, %vgpr3, %vgpr0_vgpr1, %vgpr2_vgpr3, %vgpr0_vgpr1_vgpr2, %vgpr1_vgpr2, %vgpr1_vgpr2_vgpr3, %vgpr4, %sgpr0_sgpr1, %sgpr0, %sgpr1{{$}}

name:            insert_undef_value_offset_vgpr
alignment:       0
exposesReturnsTwice: false
hasInlineAsm:    false
allVRegsAllocated: true
isSSA:           false
tracksRegLiveness: true
tracksSubRegLiveness: true
liveins:
  - { reg: '%sgpr0_sgpr1' }
frameInfo:
  isFrameAddressTaken: false
  isReturnAddressTaken: false
  hasStackMap:     false
  hasPatchPoint:   false
  stackSize:       0
  offsetAdjustment: 0
  maxAlignment:    0
  adjustsStack:    false
  hasCalls:        false
  maxCallFrameSize: 0
  hasOpaqueSPAdjustment: false
  hasVAStart:      false
  hasMustTailInVarArgFunc: false
body:             |
  bb.0.entry:
    liveins: %sgpr0_sgpr1

    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM %sgpr0_sgpr1, 11 :: (non-temporal invariant load 8 from `i64 addrspace(2)* undef`)
    %sgpr7 = S_MOV_B32 61440
    %sgpr6 = S_MOV_B32 -1
    %vgpr4 = V_MOV_B32_e32 2, implicit %exec
    S_WAITCNT 127
    %vgpr0_vgpr1_vgpr2_vgpr3 = BUFFER_LOAD_DWORDX4_OFFSET %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec :: (load 16 from %ir.in)
    %sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM killed %sgpr0_sgpr1, 9 :: (non-temporal invariant load 8 from `i64 addrspace(2)* undef`)
    S_WAITCNT 3952
    %vgpr0_vgpr1_vgpr2_vgpr3, dead %sgpr0_sgpr1 = SI_INDIRECT_DST_V4 %vgpr0_vgpr1_vgpr2_vgpr3, killed %vgpr4, 0, undef %vgpr10, implicit-def dead %exec, implicit-def dead %vcc, implicit-def dead %m0, implicit %exec
    S_WAITCNT 127
    BUFFER_STORE_DWORDX4_OFFSET killed %vgpr0_vgpr1_vgpr2_vgpr3, killed %sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, implicit %exec :: (store 16 into %ir.out)
    S_ENDPGM

...