test/CodeGen/AMDGPU/spill384.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s

# Make sure spill/restore of 384 bit registers works.

---
name: spill_restore_sgpr384
tracksRegLiveness: true
machineFunctionInfo:
  scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
  stackPtrOffsetReg: $sgpr32
body: |
  ; SPILLED-LABEL: name: spill_restore_sgpr384
  ; SPILLED: bb.0:
  ; SPILLED-NEXT:   successors: %bb.1(0x80000000)
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT:   S_NOP 0, implicit-def renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
  ; SPILLED-NEXT:   SI_SPILL_S384_SAVE killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, %stack.0, implicit $exec, implicit $sgpr32 :: (store (s384) into %stack.0, align 4, addrspace 5)
  ; SPILLED-NEXT:   S_CBRANCH_SCC1 %bb.1, implicit undef $scc
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT: bb.1:
  ; SPILLED-NEXT:   successors: %bb.2(0x80000000)
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT:   S_NOP 1
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT: bb.2:
  ; SPILLED-NEXT:   $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = SI_SPILL_S384_RESTORE %stack.0, implicit $exec, implicit $sgpr32 :: (load (s384) from %stack.0, align 4, addrspace 5)
  ; SPILLED-NEXT:   S_NOP 0, implicit killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
  ; EXPANDED-LABEL: name: spill_restore_sgpr384
  ; EXPANDED: bb.0:
  ; EXPANDED-NEXT:   successors: %bb.1(0x80000000)
  ; EXPANDED-NEXT:   liveins: $vgpr0
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT:   S_NOP 0, implicit-def renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr4, 0, $vgpr0, implicit-def $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, implicit $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr5, 1, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr6, 2, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr7, 3, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr8, 4, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr9, 5, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr10, 6, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr11, 7, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr12, 8, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr13, 9, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr14, 10, $vgpr0
  ; EXPANDED-NEXT:   $vgpr0 = V_WRITELANE_B32 killed $sgpr15, 11, $vgpr0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
  ; EXPANDED-NEXT:   S_CBRANCH_SCC1 %bb.1, implicit undef $scc
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT: bb.1:
  ; EXPANDED-NEXT:   successors: %bb.2(0x80000000)
  ; EXPANDED-NEXT:   liveins: $vgpr0
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT:   S_NOP 1
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT: bb.2:
  ; EXPANDED-NEXT:   liveins: $vgpr0
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT:   $sgpr4 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
  ; EXPANDED-NEXT:   $sgpr5 = V_READLANE_B32 $vgpr0, 1
  ; EXPANDED-NEXT:   $sgpr6 = V_READLANE_B32 $vgpr0, 2
  ; EXPANDED-NEXT:   $sgpr7 = V_READLANE_B32 $vgpr0, 3
  ; EXPANDED-NEXT:   $sgpr8 = V_READLANE_B32 $vgpr0, 4
  ; EXPANDED-NEXT:   $sgpr9 = V_READLANE_B32 $vgpr0, 5
  ; EXPANDED-NEXT:   $sgpr10 = V_READLANE_B32 $vgpr0, 6
  ; EXPANDED-NEXT:   $sgpr11 = V_READLANE_B32 $vgpr0, 7
  ; EXPANDED-NEXT:   $sgpr12 = V_READLANE_B32 $vgpr0, 8
  ; EXPANDED-NEXT:   $sgpr13 = V_READLANE_B32 $vgpr0, 9
  ; EXPANDED-NEXT:   $sgpr14 = V_READLANE_B32 $vgpr0, 10
  ; EXPANDED-NEXT:   $sgpr15 = V_READLANE_B32 $vgpr0, 11
  ; EXPANDED-NEXT:   S_NOP 0, implicit killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
  bb.0:
    S_NOP 0, implicit-def %0:sgpr_384
    S_CBRANCH_SCC1 implicit undef $scc, %bb.1

  bb.1:
    S_NOP 1

  bb.2:
    S_NOP 0, implicit %0
...

---
name: spill_restore_vgpr384
tracksRegLiveness: true
machineFunctionInfo:
  scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
  stackPtrOffsetReg: $sgpr32
body: |
  ; SPILLED-LABEL: name: spill_restore_vgpr384
  ; SPILLED: bb.0:
  ; SPILLED-NEXT:   successors: %bb.1(0x80000000)
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT:   S_NOP 0, implicit-def renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
  ; SPILLED-NEXT:   SI_SPILL_V384_SAVE killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11, %stack.0, $sgpr32, 0, implicit $exec :: (store (s384) into %stack.0, align 4, addrspace 5)
  ; SPILLED-NEXT:   S_CBRANCH_SCC1 %bb.1, implicit undef $scc
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT: bb.1:
  ; SPILLED-NEXT:   successors: %bb.2(0x80000000)
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT:   S_NOP 1
  ; SPILLED-NEXT: {{  $}}
  ; SPILLED-NEXT: bb.2:
  ; SPILLED-NEXT:   $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11 = SI_SPILL_V384_RESTORE %stack.0, $sgpr32, 0, implicit $exec :: (load (s384) from %stack.0, align 4, addrspace 5)
  ; SPILLED-NEXT:   S_NOP 0, implicit killed renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
  ; EXPANDED-LABEL: name: spill_restore_vgpr384
  ; EXPANDED: bb.0:
  ; EXPANDED-NEXT:   successors: %bb.1(0x80000000)
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT:   S_NOP 0, implicit-def renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
  ; EXPANDED-NEXT:   SI_SPILL_V384_SAVE killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11, %stack.0, $sgpr32, 0, implicit $exec :: (store (s384) into %stack.0, align 4, addrspace 5)
  ; EXPANDED-NEXT:   S_CBRANCH_SCC1 %bb.1, implicit undef $scc
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT: bb.1:
  ; EXPANDED-NEXT:   successors: %bb.2(0x80000000)
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT:   S_NOP 1
  ; EXPANDED-NEXT: {{  $}}
  ; EXPANDED-NEXT: bb.2:
  ; EXPANDED-NEXT:   $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11 = SI_SPILL_V384_RESTORE %stack.0, $sgpr32, 0, implicit $exec :: (load (s384) from %stack.0, align 4, addrspace 5)
  ; EXPANDED-NEXT:   S_NOP 0, implicit killed renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
  bb.0:
    S_NOP 0, implicit-def %0:vreg_384
    S_CBRANCH_SCC1 implicit undef $scc, %bb.1

  bb.1:
    S_NOP 1

  bb.2:
    S_NOP 0, implicit %0
...