llvm/llvm/test/CodeGen/AMDGPU/extend-wwm-virt-reg-liveness.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -start-before=si-lower-sgpr-spills -stop-after=virtregrewriter,1 -verify-machineinstrs %s -o - | FileCheck -check-prefix=GCN %s

# Tests to check the conservative lieness extension for the wwm registers during SGPR spill lowering.

# Even though the VGPR can be shared for the wwm-operand (writelane/readlane get inserted for the SGPR spills)
# and the regular operand (%0), they get different registers as we conservatively extend the liveness of the
# wwm-operands.
---
name:            test_single_block
tracksRegLiveness: true
frameInfo:
  maxAlignment:    4
stack:
  - { id: 0, type: spill-slot, size: 4, alignment: 4, stack-id: sgpr-spill }
machineFunctionInfo:
  isEntryFunction: false
  scratchRSrcReg:  '$sgpr0_sgpr1_sgpr2_sgpr3'
  stackPtrOffsetReg: '$sgpr32'
  frameOffsetReg: '$sgpr33'
  hasSpilledSGPRs: true
body:             |
  bb.0:
    liveins: $sgpr4, $vgpr2_vgpr3
    ; GCN-LABEL: name: test_single_block
    ; GCN: liveins: $sgpr4, $vgpr2_vgpr3
    ; GCN-NEXT: {{  $}}
    ; GCN-NEXT: renamable $vgpr0 = IMPLICIT_DEF
    ; GCN-NEXT: renamable $vgpr0 = SI_SPILL_S32_TO_VGPR $sgpr4, 0, killed $vgpr0
    ; GCN-NEXT: S_NOP 0
    ; GCN-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR $vgpr0, 0
    ; GCN-NEXT: renamable $vgpr1 = V_MOV_B32_e32 20, implicit $exec
    ; GCN-NEXT: GLOBAL_STORE_DWORD $vgpr2_vgpr3, killed renamable $vgpr1, 0, 0, implicit $exec
    ; GCN-NEXT: KILL killed renamable $vgpr0
    ; GCN-NEXT: SI_RETURN
    SI_SPILL_S32_SAVE killed $sgpr4, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    S_NOP 0
    renamable $sgpr4 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    %0:vgpr_32 = V_MOV_B32_e32 20, implicit $exec
    GLOBAL_STORE_DWORD $vgpr2_vgpr3, %0:vgpr_32, 0, 0, implicit $exec
    SI_RETURN
...

# Due to the presence of wwm-operand in the divergent flow, the regular variable (%0) shouldn't get the same register
# allocated for the wwm-operand in writelane/readlane when the SGPR spill is lowered.

---
name:            test_if_else
tracksRegLiveness: true
frameInfo:
  maxAlignment:    4
stack:
  - { id: 0, type: spill-slot, size: 4, alignment: 4, stack-id: sgpr-spill }
machineFunctionInfo:
  isEntryFunction: false
  scratchRSrcReg:  '$sgpr0_sgpr1_sgpr2_sgpr3'
  stackPtrOffsetReg: '$sgpr32'
  frameOffsetReg: '$sgpr33'
  hasSpilledSGPRs: true
body:             |
  ; GCN-LABEL: name: test_if_else
  ; GCN: bb.0:
  ; GCN-NEXT:   successors: %bb.1(0x80000000)
  ; GCN-NEXT:   liveins: $sgpr6, $sgpr10_sgpr11
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr0 = IMPLICIT_DEF
  ; GCN-NEXT:   S_BRANCH %bb.1
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.1:
  ; GCN-NEXT:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
  ; GCN-NEXT:   liveins: $sgpr6, $vgpr0, $sgpr10_sgpr11
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 10, implicit $exec
  ; GCN-NEXT:   S_CBRANCH_EXECZ %bb.3, implicit $exec
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.2:
  ; GCN-NEXT:   successors: %bb.3(0x80000000)
  ; GCN-NEXT:   liveins: $sgpr6, $vgpr0, $sgpr10_sgpr11
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr0 = SI_SPILL_S32_TO_VGPR $sgpr6, 0, killed $vgpr0
  ; GCN-NEXT:   S_NOP 0
  ; GCN-NEXT:   $sgpr6 = SI_RESTORE_S32_FROM_VGPR $vgpr0, 0
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 20, implicit $exec
  ; GCN-NEXT:   S_BRANCH %bb.3
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.3:
  ; GCN-NEXT:   liveins: $vgpr0, $vgpr1, $sgpr10_sgpr11
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   $sgpr5 = V_READFIRSTLANE_B32 killed $vgpr1, implicit $exec
  ; GCN-NEXT:   S_STORE_DWORD_IMM $sgpr5, $sgpr10_sgpr11, 0, 0
  ; GCN-NEXT:   KILL killed renamable $vgpr0
  ; GCN-NEXT:   SI_RETURN
  bb.0:
    liveins: $sgpr6, $sgpr10_sgpr11
    S_BRANCH %bb.1
  bb.1:
    liveins: $sgpr6, $sgpr10_sgpr11
    %0:vgpr_32 = V_MOV_B32_e32 10, implicit $exec
    S_CBRANCH_EXECZ %bb.3, implicit $exec
  bb.2:
    liveins: $sgpr6, $sgpr10_sgpr11
    SI_SPILL_S32_SAVE killed $sgpr6, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    S_NOP 0
    renamable $sgpr6 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    %0:vgpr_32 = V_MOV_B32_e32 20, implicit $exec
    S_BRANCH %bb.3
  bb.3:
    liveins: $sgpr10_sgpr11
    $sgpr5 = V_READFIRSTLANE_B32 %0:vgpr_32, implicit $exec
    S_STORE_DWORD_IMM $sgpr5, $sgpr10_sgpr11, 0, 0
    SI_RETURN
...

# The wwm-register usage outside the loop should have the interference marked with
# all the regular virtual registers used in the test. The divergent loop index value (%1)
# can actually share the same VGPR as the wwm-operand. But since we extend the liveness of
# the wwm operand, an interference will always exist between them.

---
name:            test_loop
tracksRegLiveness: true
frameInfo:
  maxAlignment:    4
stack:
  - { id: 0, type: spill-slot, size: 4, alignment: 4, stack-id: sgpr-spill }
machineFunctionInfo:
  isEntryFunction: false
  scratchRSrcReg:  '$sgpr0_sgpr1_sgpr2_sgpr3'
  stackPtrOffsetReg: '$sgpr32'
  frameOffsetReg: '$sgpr33'
  hasSpilledSGPRs: true
body:             |
  ; GCN-LABEL: name: test_loop
  ; GCN: bb.0:
  ; GCN-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; GCN-NEXT:   liveins: $sgpr4, $sgpr10_sgpr11
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr0 = IMPLICIT_DEF
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 10, implicit $exec
  ; GCN-NEXT:   S_CBRANCH_EXECZ %bb.2, implicit $exec
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.1:
  ; GCN-NEXT:   successors: %bb.2(0x80000000)
  ; GCN-NEXT:   liveins: $sgpr4, $vgpr0, $sgpr10_sgpr11
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr0 = SI_SPILL_S32_TO_VGPR $sgpr4, 0, killed $vgpr0
  ; GCN-NEXT:   S_NOP 0
  ; GCN-NEXT:   $sgpr4 = SI_RESTORE_S32_FROM_VGPR $vgpr0, 0
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 20, implicit $exec
  ; GCN-NEXT:   S_BRANCH %bb.2
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.2:
  ; GCN-NEXT:   successors: %bb.3(0x80000000)
  ; GCN-NEXT:   liveins: $sgpr4, $vgpr0, $vgpr1, $sgpr10_sgpr11
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   S_STORE_DWORD_IMM $sgpr4, $sgpr10_sgpr11, 0, 0
  ; GCN-NEXT:   $sgpr5 = V_READFIRSTLANE_B32 killed $vgpr1, implicit $exec
  ; GCN-NEXT:   S_STORE_DWORD_IMM $sgpr5, $sgpr10_sgpr11, 0, 4
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 5, implicit $exec
  ; GCN-NEXT:   S_CBRANCH_EXECZ %bb.3, implicit $exec
  ; GCN-NEXT:   S_BRANCH %bb.3
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.3:
  ; GCN-NEXT:   successors: %bb.5(0x40000000), %bb.4(0x40000000)
  ; GCN-NEXT:   liveins: $vgpr0, $vgpr1
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   $vcc = V_CMP_EQ_U32_e64 0, $vgpr1, implicit $exec
  ; GCN-NEXT:   $sgpr6_sgpr7 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
  ; GCN-NEXT:   S_CBRANCH_SCC1 %bb.5, implicit $scc
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.4:
  ; GCN-NEXT:   successors: %bb.3(0x80000000)
  ; GCN-NEXT:   liveins: $vgpr0, $vgpr1, $sgpr6_sgpr7
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr1 = V_SUB_U32_e32 1, killed $vgpr1, implicit $exec
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 killed $vgpr1, implicit $exec
  ; GCN-NEXT:   S_BRANCH %bb.3
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.5:
  ; GCN-NEXT:   liveins: $vgpr0, $sgpr6_sgpr7
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   $exec = S_OR_B64 $exec, $sgpr6_sgpr7, implicit-def $scc
  ; GCN-NEXT:   KILL killed renamable $vgpr0
  ; GCN-NEXT:   SI_RETURN
  bb.0:
    liveins: $sgpr4, $sgpr10_sgpr11
    %0:vgpr_32 = V_MOV_B32_e32 10, implicit $exec
    S_CBRANCH_EXECZ %bb.2, implicit $exec
  bb.1:
    liveins: $sgpr4, $sgpr10_sgpr11
    SI_SPILL_S32_SAVE killed $sgpr4, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    S_NOP 0
    renamable $sgpr4 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    %0:vgpr_32 = V_MOV_B32_e32 20, implicit $exec
    S_BRANCH %bb.2
  bb.2:
    liveins: $sgpr4, $sgpr10_sgpr11
    S_STORE_DWORD_IMM $sgpr4, $sgpr10_sgpr11, 0, 0
    $sgpr5 = V_READFIRSTLANE_B32 %0:vgpr_32, implicit $exec
    S_STORE_DWORD_IMM $sgpr5, $sgpr10_sgpr11, 0, 4
    %1:vgpr_32 = V_MOV_B32_e32 5, implicit $exec
    S_CBRANCH_EXECZ %bb.3, implicit $exec
    S_BRANCH %bb.3
  bb.3:
    $vcc = V_CMP_EQ_U32_e64 0, %1:vgpr_32, implicit $exec
    $sgpr6_sgpr7 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
    S_CBRANCH_SCC1 %bb.5, implicit $scc
  bb.4:
    liveins: $sgpr6_sgpr7
    %2:vgpr_32 = V_SUB_U32_e32 1, %1:vgpr_32, implicit $exec
    %1:vgpr_32 = V_MOV_B32_e32 %2:vgpr_32, implicit $exec
    S_BRANCH %bb.3
  bb.5:
    liveins: $sgpr6_sgpr7
    $exec = S_OR_B64 $exec, $sgpr6_sgpr7, implicit-def $scc
    SI_RETURN
...

# There must be one KILL instruction for the wwm-operand in every return block.
# Due to that, the wwm-register allocated should be different from the ones
# allocated for the regular virtual registers.

---
name:            test_multiple_return_blocks
tracksRegLiveness: true
frameInfo:
  maxAlignment:    4
stack:
  - { id: 0, type: spill-slot, size: 4, alignment: 4, stack-id: sgpr-spill }
machineFunctionInfo:
  isEntryFunction: false
  scratchRSrcReg:  '$sgpr0_sgpr1_sgpr2_sgpr3'
  stackPtrOffsetReg: '$sgpr32'
  frameOffsetReg: '$sgpr33'
  hasSpilledSGPRs: true
body:             |
  ; GCN-LABEL: name: test_multiple_return_blocks
  ; GCN: bb.0:
  ; GCN-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; GCN-NEXT:   liveins: $sgpr4, $vgpr2_vgpr3
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr0 = IMPLICIT_DEF
  ; GCN-NEXT:   S_CBRANCH_EXECZ %bb.2, implicit $exec
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.1:
  ; GCN-NEXT:   liveins: $sgpr4, $vgpr0, $vgpr2_vgpr3
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr0 = SI_SPILL_S32_TO_VGPR $sgpr4, 0, killed $vgpr0
  ; GCN-NEXT:   S_NOP 0
  ; GCN-NEXT:   $sgpr4 = SI_RESTORE_S32_FROM_VGPR $vgpr0, 0
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 10, implicit $exec
  ; GCN-NEXT:   GLOBAL_STORE_DWORD $vgpr2_vgpr3, killed renamable $vgpr1, 0, 0, implicit $exec
  ; GCN-NEXT:   KILL killed renamable $vgpr0
  ; GCN-NEXT:   SI_RETURN
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT: bb.2:
  ; GCN-NEXT:   liveins: $vgpr0, $vgpr2_vgpr3
  ; GCN-NEXT: {{  $}}
  ; GCN-NEXT:   renamable $vgpr1 = V_MOV_B32_e32 20, implicit $exec
  ; GCN-NEXT:   GLOBAL_STORE_DWORD $vgpr2_vgpr3, killed renamable $vgpr1, 0, 0, implicit $exec
  ; GCN-NEXT:   KILL killed renamable $vgpr0
  ; GCN-NEXT:   SI_RETURN
  bb.0:
    liveins: $sgpr4, $vgpr2_vgpr3
    S_CBRANCH_EXECZ %bb.2, implicit $exec
  bb.1:
    liveins: $sgpr4, $vgpr2_vgpr3
    SI_SPILL_S32_SAVE killed $sgpr4, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    S_NOP 0
    renamable $sgpr4 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
    %0:vgpr_32 = V_MOV_B32_e32 10, implicit $exec
    GLOBAL_STORE_DWORD $vgpr2_vgpr3, %0:vgpr_32, 0, 0, implicit $exec
    SI_RETURN
  bb.2:
    liveins: $vgpr2_vgpr3
    %1:vgpr_32 = V_MOV_B32_e32 20, implicit $exec
    GLOBAL_STORE_DWORD $vgpr2_vgpr3, %1:vgpr_32, 0, 0, implicit $exec
    SI_RETURN
...