X86/AMX/amx-fastconfig-phi2.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=x86_64-- -mattr=+amx-tile -run-pass=fastpretileconfig -o - %s | FileCheck %s
#
#             bb.0
#            def %0
#          /        \
#       bb.1         bb.2   <---------
#      def %1       %2=phi(%0, %3)   |
#           \      /                 |
#             bb.3 -------------------
#         %3=phi(%1, %2)
#
# This case test tile PHIs depend each other, and the its def block is
# not visited yet.
---
name:            foo
alignment:       16
tracksRegLiveness: true
registers:
  - { id: 1, class: tile }
  - { id: 2, class: tile }
  - { id: 3, class: tile }
  - { id: 4, class: tile }
  - { id: 12, class: gr32 }
  - { id: 13, class: gr32 }
  - { id: 16, class: gr8 }
  - { id: 17, class: gr16 }
  - { id: 18, class: gr16 }
  - { id: 19, class: gr64_nosp }
  - { id: 22, class: gr32 }
  - { id: 23, class: gr16 }
  - { id: 24, class: gr16 }
liveins:
  - { reg: '$edi', virtual-reg: '%12' }
frameInfo:
  maxAlignment:    1
machineFunctionInfo:
  amxProgModel: ManagedRA
body:             |
  ; CHECK-LABEL: name: foo
  ; CHECK: bb.0.entry:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT:   liveins: $edi
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[V_SET0_:%[0-9]+]]:vr128 = V_SET0
  ; CHECK-NEXT:   MOVUPSmr %stack.1, 1, $noreg, 0, $noreg, [[V_SET0_]] :: (store (s512) into %stack.1, align 4)
  ; CHECK-NEXT:   MOVUPSmr %stack.1, 1, $noreg, 16, $noreg, [[V_SET0_]] :: (store (s512) into %stack.1 + 16, align 4)
  ; CHECK-NEXT:   MOVUPSmr %stack.1, 1, $noreg, 32, $noreg, [[V_SET0_]] :: (store (s512) into %stack.1 + 32, align 4)
  ; CHECK-NEXT:   MOVUPSmr %stack.1, 1, $noreg, 48, $noreg, [[V_SET0_]] :: (store (s512) into %stack.1 + 48, align 4)
  ; CHECK-NEXT:   MOV8mi %stack.1, 1, $noreg, 0, $noreg, 1 :: (store (s512) into %stack.1, align 4)
  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:gr32 = COPY $edi
  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:gr32 = COPY killed [[COPY]]
  ; CHECK-NEXT:   %r0:gr16 = MOV16ri 64
  ; CHECK-NEXT:   %c0:gr16 = MOV16ri 16
  ; CHECK-NEXT:   PLDTILECFGV %stack.1, 1, $noreg, 0, $noreg, implicit-def $tmm0, implicit-def $tmm1, implicit-def $tmm2, implicit-def $tmm3, implicit-def $tmm4, implicit-def $tmm5, implicit-def $tmm6, implicit-def $tmm7 :: (load (s512) from %stack.1, align 4)
  ; CHECK-NEXT:   [[LEA64r:%[0-9]+]]:gr64_nosp = LEA64r %stack.0, 1, $noreg, 0, $noreg
  ; CHECK-NEXT:   %t0:tile = PTILEZEROV %r0, %c0
  ; CHECK-NEXT:   [[MOV64ri:%[0-9]+]]:gr64_nosp = MOV64ri 64
  ; CHECK-NEXT:   TILESTORED %stack.0, 1, killed [[MOV64ri]], 0, $noreg, %t0 :: (store (s8192) into %stack.0)
  ; CHECK-NEXT:   CMP32ri8 [[COPY1]], 0, implicit-def $eflags
  ; CHECK-NEXT:   [[SETCCr:%[0-9]+]]:gr8 = SETCCr 4, implicit $eflags
  ; CHECK-NEXT:   TEST8ri [[SETCCr]], 1, implicit-def $eflags
  ; CHECK-NEXT:   JCC_1 %bb.2, 5, implicit $eflags
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[MOV16ri:%[0-9]+]]:gr16 = MOV16ri 64
  ; CHECK-NEXT:   [[MOV16ri1:%[0-9]+]]:gr16 = MOV16ri 16
  ; CHECK-NEXT:   PLDTILECFGV %stack.1, 1, $noreg, 0, $noreg, implicit-def $tmm0, implicit-def $tmm1, implicit-def $tmm2, implicit-def $tmm3, implicit-def $tmm4, implicit-def $tmm5, implicit-def $tmm6, implicit-def $tmm7 :: (load (s512) from %stack.1, align 4)
  ; CHECK-NEXT:   [[LEA64r1:%[0-9]+]]:gr64_nosp = LEA64r %stack.2, 1, $noreg, 0, $noreg
  ; CHECK-NEXT:   [[PTILEZEROV:%[0-9]+]]:tile = PTILEZEROV [[MOV16ri1]], [[MOV16ri]]
  ; CHECK-NEXT:   [[MOV64ri1:%[0-9]+]]:gr64_nosp = MOV64ri 64
  ; CHECK-NEXT:   TILESTORED %stack.2, 1, killed [[MOV64ri1]], 0, $noreg, [[PTILEZEROV]] :: (store (s8192) into %stack.2)
  ; CHECK-NEXT:   JMP_1 %bb.3
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[PHI:%[0-9]+]]:gr16 = PHI %c0, %bb.0, %24, %bb.3
  ; CHECK-NEXT:   [[PHI1:%[0-9]+]]:gr16 = PHI %r0, %bb.0, %23, %bb.3
  ; CHECK-NEXT:   [[PHI2:%[0-9]+]]:gr64_nosp = PHI [[LEA64r]], %bb.0, %22, %bb.3
  ; CHECK-NEXT:   PLDTILECFGV %stack.1, 1, $noreg, 0, $noreg, implicit-def $tmm0, implicit-def $tmm1, implicit-def $tmm2, implicit-def $tmm3, implicit-def $tmm4, implicit-def $tmm5, implicit-def $tmm6, implicit-def $tmm7 :: (load (s512) from %stack.1, align 4)
  ; CHECK-NEXT:   [[MOV64ri2:%[0-9]+]]:gr64_nosp = MOV64ri 64
  ; CHECK-NEXT:   [[PTILELOADDV:%[0-9]+]]:tile = PTILELOADDV [[PHI1]], [[PHI]], [[PHI2]], 1, killed [[MOV64ri2]], 0, $noreg
  ; CHECK-NEXT:   [[MOV64ri3:%[0-9]+]]:gr64_nosp = MOV64ri 64
  ; CHECK-NEXT:   TILESTORED %stack.3, 1, killed [[MOV64ri3]], 0, $noreg, [[PTILELOADDV]] :: (store (s8192) into %stack.3)
  ; CHECK-NEXT:   [[MOV32ri64_:%[0-9]+]]:gr64_nosp = MOV32ri64 32
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.3:
  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[PHI3:%[0-9]+]]:gr16 = PHI [[MOV16ri]], %bb.1, [[PHI]], %bb.2
  ; CHECK-NEXT:   [[PHI4:%[0-9]+]]:gr16 = PHI [[MOV16ri1]], %bb.1, [[PHI1]], %bb.2
  ; CHECK-NEXT:   [[PHI5:%[0-9]+]]:gr64_nosp = PHI [[LEA64r1]], %bb.1, [[PHI2]], %bb.2
  ; CHECK-NEXT:   PLDTILECFGV %stack.1, 1, $noreg, 0, $noreg, implicit-def $tmm0, implicit-def $tmm1, implicit-def $tmm2, implicit-def $tmm3, implicit-def $tmm4, implicit-def $tmm5, implicit-def $tmm6, implicit-def $tmm7 :: (load (s512) from %stack.1, align 4)
  ; CHECK-NEXT:   [[MOV64ri4:%[0-9]+]]:gr64_nosp = MOV64ri 64
  ; CHECK-NEXT:   [[PTILELOADDV1:%[0-9]+]]:tile = PTILELOADDV [[PHI4]], [[PHI3]], [[PHI5]], 1, killed [[MOV64ri4]], 0, $noreg
  ; CHECK-NEXT:   [[MOV64ri5:%[0-9]+]]:gr64_nosp = MOV64ri 64
  ; CHECK-NEXT:   TILESTORED %stack.4, 1, killed [[MOV64ri5]], 0, $noreg, [[PTILELOADDV1]] :: (store (s8192) into %stack.4)
  ; CHECK-NEXT:   [[MOV16ri2:%[0-9]+]]:gr16 = MOV16ri 64
  ; CHECK-NEXT:   [[MOV16ri3:%[0-9]+]]:gr16 = MOV16ri 16
  ; CHECK-NEXT:   JMP_1 %bb.2
  bb.0.entry:
    liveins: $edi

    %12:gr32 = COPY $edi
    %13:gr32 = COPY killed %12
    %r0:gr16 = MOV16ri 64
    %c0:gr16 = MOV16ri 16
    %t0:tile = PTILEZEROV killed %r0, killed %c0
    CMP32ri8 %13, 0, implicit-def $eflags
    %16:gr8 = SETCCr 4, implicit $eflags
    TEST8ri %16, 1, implicit-def $eflags
    JCC_1 %bb.2, 5, implicit $eflags

  bb.1:
    %17:gr16 = MOV16ri 64
    %18:gr16 = MOV16ri 16
    %1:tile = PTILEZEROV killed %18, killed %17
    JMP_1 %bb.3

  bb.2:
    %2:tile = PHI %t0, %bb.0, %3, %bb.3
    %19:gr64_nosp = MOV32ri64 32

  bb.3:
    %3:tile = PHI %1, %bb.1, %2, %bb.2
    %23:gr16 = MOV16ri 64
    %24:gr16 = MOV16ri 16
    JMP_1 %bb.2