test/CodeGen/AMDGPU/si-lower-sgpr-spills-cycle-header.ll

; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
; REQUIRES: amdgpu-registered-target
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -O0 -amdgpu-spill-sgpr-to-vgpr=1 \
; RUN:     -verify-machineinstrs -stop-after=si-lower-sgpr-spills -o - %s | FileCheck %s

;; Ensure that si-lower-sgpr-spills prevents IMPLICIT_DEF assignments from clobbering
;; backedge writes by placing the assignment in the cycle preheader and not the header.

define amdgpu_kernel void @loop_sgpr_spill_implicit_def_in_preheader(
  ; CHECK-LABEL: name: loop_sgpr_spill_implicit_def_in_preheader
  ; CHECK: bb.0.entry:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $sgpr8_sgpr9
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   dead renamable $sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM renamable $sgpr8_sgpr9, 0, 0 :: (dereferenceable invariant load (s64) from constant-pool, align 16, addrspace 4)
  ; CHECK-NEXT:   renamable $sgpr4_sgpr5 = S_LOAD_DWORDX2_IMM killed renamable $sgpr8_sgpr9, 0, 0 :: (dereferenceable invariant load (s64) from %ir.out.kernarg.offset, align 16, addrspace 4)
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr4, 0, [[DEF]], implicit-def $sgpr4_sgpr5, implicit $sgpr4_sgpr5
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 1, [[DEF]], implicit killed $sgpr4_sgpr5
  ; CHECK-NEXT:   renamable $sgpr4_sgpr5 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr4 = S_LOAD_DWORD_IMM killed renamable $sgpr4_sgpr5, 0, 0 :: (load (s32) from `ptr addrspace(4) poison`, addrspace 4)
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr4, 2, [[DEF]]
  ; CHECK-NEXT:   renamable $sgpr11 = COPY renamable $sgpr4
  ; CHECK-NEXT:   renamable $sgpr10 = COPY renamable $sgpr4
  ; CHECK-NEXT:   renamable $sgpr9 = COPY renamable $sgpr4
  ; CHECK-NEXT:   renamable $sgpr8 = COPY renamable $sgpr4
  ; CHECK-NEXT:   renamable $sgpr7 = COPY renamable $sgpr4
  ; CHECK-NEXT:   renamable $sgpr6 = COPY renamable $sgpr4
  ; CHECK-NEXT:   renamable $sgpr5 = COPY renamable $sgpr4
  ; CHECK-NEXT:   renamable $sgpr12 = S_MOV_B32 0
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr12, 3, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr11, 4, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr10, 5, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr9, 6, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr8, 7, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr7, 8, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr6, 9, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 10, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr4, 11, [[DEF]]
  ; CHECK-NEXT:   S_BRANCH %bb.1
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1.loop.header:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   $sgpr6 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 3
  ; CHECK-NEXT:   $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 4
  ; CHECK-NEXT:   $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 5
  ; CHECK-NEXT:   $sgpr7 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 6
  ; CHECK-NEXT:   $sgpr8 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 7
  ; CHECK-NEXT:   $sgpr9 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 8
  ; CHECK-NEXT:   $sgpr10 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 9
  ; CHECK-NEXT:   $sgpr11 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 10
  ; CHECK-NEXT:   $sgpr12 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 11
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr12, 12, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr11, 13, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr10, 14, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr9, 15, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr8, 16, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr7, 17, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 18, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr4, 19, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr6, 20, [[DEF]]
  ; CHECK-NEXT:   renamable $sgpr4_sgpr5 = S_MOV_B64 -1
  ; CHECK-NEXT:   renamable $sgpr7 = S_MOV_B32 0
  ; CHECK-NEXT:   S_CMP_EQ_U32 renamable $sgpr6, killed renamable $sgpr7, implicit-def $scc
  ; CHECK-NEXT:   renamable $sgpr6_sgpr7 = S_CSELECT_B64 -1, 0, implicit killed $scc
  ; CHECK-NEXT:   renamable $sgpr6_sgpr7 = S_XOR_B64 killed renamable $sgpr6_sgpr7, renamable $sgpr4_sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $vcc = S_AND_B64 $exec, killed renamable $sgpr6_sgpr7, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   renamable $sgpr6 = IMPLICIT_DEF
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr4, 21, [[DEF]], implicit-def $sgpr4_sgpr5, implicit $sgpr4_sgpr5
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 22, [[DEF]], implicit killed $sgpr4_sgpr5
  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.3, implicit killed $vcc
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2.loop.latch:
  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 20
  ; CHECK-NEXT:   $sgpr6 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 12
  ; CHECK-NEXT:   $sgpr13 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 13
  ; CHECK-NEXT:   $sgpr12 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 14
  ; CHECK-NEXT:   $sgpr11 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 15
  ; CHECK-NEXT:   $sgpr10 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 16
  ; CHECK-NEXT:   $sgpr9 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 17
  ; CHECK-NEXT:   $sgpr8 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 18
  ; CHECK-NEXT:   $sgpr7 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 19
  ; CHECK-NEXT:   renamable $sgpr5 = S_MOV_B32 1
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr5, 23, [[DEF]]
  ; CHECK-NEXT:   renamable $sgpr7 = S_ADD_I32 renamable $sgpr7, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr8 = S_ADD_I32 renamable $sgpr8, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr9 = S_ADD_I32 renamable $sgpr9, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr10 = S_ADD_I32 renamable $sgpr10, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr11 = S_ADD_I32 renamable $sgpr11, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr12 = S_ADD_I32 renamable $sgpr12, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr13 = S_ADD_I32 renamable $sgpr13, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr14 = S_ADD_I32 renamable $sgpr6, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr6 = nuw nsw S_ADD_I32 renamable $sgpr4, renamable $sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   renamable $sgpr4_sgpr5 = S_MOV_B64 0
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr14, 24, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr13, 25, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr12, 26, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr11, 27, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr10, 28, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr9, 29, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr8, 30, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr7, 31, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr6, 32, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr4, 21, [[DEF]], implicit-def $sgpr4_sgpr5, implicit $sgpr4_sgpr5
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 22, [[DEF]], implicit killed $sgpr4_sgpr5
  ; CHECK-NEXT:   S_BRANCH %bb.3
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.3.Flow:
  ; CHECK-NEXT:   successors: %bb.4(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   $vcc_lo = SI_RESTORE_S32_FROM_VGPR [[DEF]], 21, implicit-def $vcc
  ; CHECK-NEXT:   $vcc_hi = SI_RESTORE_S32_FROM_VGPR [[DEF]], 22
  ; CHECK-NEXT:   $sgpr6 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 24
  ; CHECK-NEXT:   $sgpr7 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 25
  ; CHECK-NEXT:   $sgpr8 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 26
  ; CHECK-NEXT:   $sgpr9 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 27
  ; CHECK-NEXT:   $sgpr10 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 28
  ; CHECK-NEXT:   $sgpr11 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 29
  ; CHECK-NEXT:   $sgpr12 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 30
  ; CHECK-NEXT:   $sgpr13 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 31
  ; CHECK-NEXT:   $sgpr14 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 32
  ; CHECK-NEXT:   renamable $sgpr4_sgpr5 = S_MOV_B64 -1
  ; CHECK-NEXT:   renamable $vcc = S_XOR_B64 killed renamable $vcc, killed renamable $sgpr4_sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 19
  ; CHECK-NEXT:   $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 18
  ; CHECK-NEXT:   renamable $vcc = S_AND_B64 $exec, killed renamable $vcc, implicit-def dead $scc
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr14, 3, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr13, 4, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr12, 5, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr11, 6, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr10, 7, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr9, 8, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr8, 9, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr7, 10, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr6, 11, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 33, [[DEF]]
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr4, 34, [[DEF]]
  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc
  ; CHECK-NEXT:   S_BRANCH %bb.4
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.4.exit:
  ; CHECK-NEXT:   $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 0, implicit-def $sgpr4_sgpr5
  ; CHECK-NEXT:   $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 1
  ; CHECK-NEXT:   $sgpr7 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 34
  ; CHECK-NEXT:   $sgpr6 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 33
  ; CHECK-NEXT:   renamable $sgpr6 = S_ADD_I32 killed renamable $sgpr6, killed renamable $sgpr7, implicit-def dead $scc
  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY killed renamable $sgpr6
  ; CHECK-NEXT:   GLOBAL_STORE_DWORD_SADDR [[V_MOV_B32_e32_]], [[COPY]], killed renamable $sgpr4_sgpr5, 0, 0, implicit $exec :: (volatile store (s32) into %ir.out.load, addrspace 1)
  ; CHECK-NEXT:   S_ENDPGM 0
    ptr addrspace(1) %out) local_unnamed_addr #0 {
entry:
  %a0 = load i32, ptr addrspace(4) poison
  %a1 = load i32, ptr addrspace(4) poison
  %a2 = load i32, ptr addrspace(4) poison
  %a3 = load i32, ptr addrspace(4) poison
  %a4 = load i32, ptr addrspace(4) poison
  %a5 = load i32, ptr addrspace(4) poison
  %a6 = load i32, ptr addrspace(4) poison
  %a7 = load i32, ptr addrspace(4) poison
  br label %loop.header

loop.header:
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %v0 = phi i32 [ %a0, %entry ], [ %t0, %loop.latch ]
  %v1 = phi i32 [ %a1, %entry ], [ %t1, %loop.latch ]
  %v2 = phi i32 [ %a2, %entry ], [ %t2, %loop.latch ]
  %v3 = phi i32 [ %a3, %entry ], [ %t3, %loop.latch ]
  %v4 = phi i32 [ %a4, %entry ], [ %t4, %loop.latch ]
  %v5 = phi i32 [ %a5, %entry ], [ %t5, %loop.latch ]
  %v6 = phi i32 [ %a6, %entry ], [ %t6, %loop.latch ]
  %v7 = phi i32 [ %a7, %entry ], [ %t7, %loop.latch ]
  %cmp = icmp eq i32 %iv, 0
  br i1 %cmp, label %loop.latch, label %exit

loop.latch:
  %t0 = add i32 %v0, 1
  %t1 = add i32 %v1, 1
  %t2 = add i32 %v2, 1
  %t3 = add i32 %v3, 1
  %t4 = add i32 %v4, 1
  %t5 = add i32 %v5, 1
  %t6 = add i32 %v6, 1
  %t7 = add i32 %v7, 1
  %iv.next = add nuw nsw i32 %iv, 1
  br label %loop.header

exit:
  %sum = add i32 %v0, %v1
  store volatile i32 %sum, ptr addrspace(1) %out
  ret void
}

attributes #0 = { nounwind "amdgpu-num-sgpr"="24" "amdgpu-num-vgpr"="64" }