llvm/test/Transforms/LoopVectorize/VPlan/dissolve-replicate-regions.ll

; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 6

; RUN: opt -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=1 -vplan-print-after="VPlanTransforms::replicateByVF$" -disable-output %s 2>&1 | FileCheck --check-prefix=SCALAR %s
; RUN: opt -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=2 -vplan-print-after="VPlanTransforms::replicateByVF$" -disable-output %s 2>&1 | FileCheck --check-prefix=VECTOR %s

define void @predicated_load(i1 %c, ptr %ptr, ptr %dst) {
; SCALAR-LABEL: VPlan for loop in 'predicated_load'
; SCALAR:  VPlan 'Initial VPlan for VF={1},UF={2}' {
; SCALAR-NEXT:  Live-in vp<[[VP0:%[0-9]+]]> = VF
; SCALAR-NEXT:  Live-in vp<[[VP1:%[0-9]+]]> = VF * UF
; SCALAR-NEXT:  Live-in ir<1024> = vector-trip-count
; SCALAR-NEXT:  Live-in ir<1024> = original trip-count
; SCALAR-EMPTY:
; SCALAR-NEXT:  ir-bb<entry>:
; SCALAR-NEXT:    EMIT branch-on-cond ir<false>
; SCALAR-NEXT:  Successor(s): scalar.ph, vector.ph
; SCALAR-EMPTY:
; SCALAR-NEXT:  vector.ph:
; SCALAR-NEXT:  Successor(s): vector loop
; SCALAR-EMPTY:
; SCALAR-NEXT:  <x1> vector loop: {
; SCALAR-NEXT:    vector.body:
; SCALAR-NEXT:      EMIT vp<[[VP3:%[0-9]+]]> = CANONICAL-INDUCTION ir<0>, vp<%index.next>
; SCALAR-NEXT:      vp<[[VP4:%[0-9]+]]> = SCALAR-STEPS vp<[[VP3]]>, ir<1>, vp<[[VP0]]>
; SCALAR-NEXT:      vp<[[VP5:%[0-9]+]]> = SCALAR-STEPS vp<[[VP3]]>, ir<1>, vp<[[VP0]]>, vp<[[VP0]]>
; SCALAR-NEXT:      EMIT branch-on-cond ir<%c>
; SCALAR-NEXT:    Successor(s): pred.load.if, pred.load.continue
; SCALAR-EMPTY:
; SCALAR-NEXT:    pred.load.if:
; SCALAR-NEXT:      CLONE ir<%gep> = getelementptr ir<%ptr>, vp<[[VP4]]>
; SCALAR-NEXT:      CLONE ir<%lv> = load ir<%gep>
; SCALAR-NEXT:    Successor(s): pred.load.continue
; SCALAR-EMPTY:
; SCALAR-NEXT:    pred.load.continue:
; SCALAR-NEXT:      EMIT-SCALAR vp<[[VP7:%[0-9]+]]> = phi [ ir<poison>, vector.body ], [ ir<%lv>, pred.load.if ]
; SCALAR-NEXT:      EMIT branch-on-cond ir<%c>
; SCALAR-NEXT:    Successor(s): pred.load.if, pred.load.continue
; SCALAR-EMPTY:
; SCALAR-NEXT:    pred.load.if:
; SCALAR-NEXT:      CLONE ir<%gep>.1 = getelementptr ir<%ptr>, vp<[[VP5]]>
; SCALAR-NEXT:      CLONE ir<%lv>.1 = load ir<%gep>.1
; SCALAR-NEXT:    Successor(s): pred.load.continue
; SCALAR-EMPTY:
; SCALAR-NEXT:    pred.load.continue:
; SCALAR-NEXT:      EMIT-SCALAR vp<[[VP9:%[0-9]+]]> = phi [ ir<poison>, pred.load.continue ], [ ir<%lv>.1, pred.load.if ]
; SCALAR-NEXT:      BLEND ir<%pred.val> = ir<0> vp<%7>/ir<%c>
; SCALAR-NEXT:      BLEND ir<%pred.val>.1 = ir<0> vp<%9>/ir<%c>
; SCALAR-NEXT:      CLONE ir<%gep.dst> = getelementptr ir<%dst>, vp<[[VP4]]>
; SCALAR-NEXT:      CLONE ir<%gep.dst>.1 = getelementptr ir<%dst>, vp<[[VP5]]>
; SCALAR-NEXT:      CLONE store ir<%pred.val>, ir<%gep.dst>
; SCALAR-NEXT:      CLONE store ir<%pred.val>.1, ir<%gep.dst>.1
; SCALAR-NEXT:      EMIT vp<%index.next> = add nuw vp<[[VP3]]>, vp<[[VP1]]>
; SCALAR-NEXT:      EMIT branch-on-count vp<%index.next>, ir<1024>
; SCALAR-NEXT:    No successors
; SCALAR-NEXT:  }
; SCALAR-NEXT:  Successor(s): middle.block
; SCALAR-EMPTY:
; SCALAR-NEXT:  middle.block:
; SCALAR-NEXT:    EMIT vp<%cmp.n> = icmp eq ir<1024>, ir<1024>
; SCALAR-NEXT:    EMIT branch-on-cond vp<%cmp.n>
; SCALAR-NEXT:  Successor(s): ir-bb<exit>, scalar.ph
; SCALAR-EMPTY:
; SCALAR-NEXT:  ir-bb<exit>:
; SCALAR-NEXT:  No successors
; SCALAR-EMPTY:
; SCALAR-NEXT:  scalar.ph:
; SCALAR-NEXT:    EMIT-SCALAR vp<%bc.resume.val> = phi [ ir<1024>, middle.block ], [ ir<0>, ir-bb<entry> ]
; SCALAR-NEXT:  Successor(s): ir-bb<loop.header>
; SCALAR-EMPTY:
; SCALAR-NEXT:  ir-bb<loop.header>:
; SCALAR-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ] (extra operand: vp<%bc.resume.val> from scalar.ph)
; SCALAR-NEXT:    IR   %gep = getelementptr i8, ptr %ptr, i64 %iv
; SCALAR-NEXT:  No successors
; SCALAR-NEXT:  }
;
; VECTOR-LABEL: VPlan for loop in 'predicated_load'
; VECTOR:  VPlan 'Initial VPlan for VF={2},UF={2}' {
; VECTOR-NEXT:  Live-in vp<[[VP0:%[0-9]+]]> = VF
; VECTOR-NEXT:  Live-in vp<[[VP1:%[0-9]+]]> = VF * UF
; VECTOR-NEXT:  Live-in ir<1024> = vector-trip-count
; VECTOR-NEXT:  Live-in ir<1024> = original trip-count
; VECTOR-EMPTY:
; VECTOR-NEXT:  ir-bb<entry>:
; VECTOR-NEXT:    EMIT branch-on-cond ir<false>
; VECTOR-NEXT:  Successor(s): scalar.ph, vector.ph
; VECTOR-EMPTY:
; VECTOR-NEXT:  vector.ph:
; VECTOR-NEXT:    EMIT vp<[[VP4:%[0-9]+]]> = broadcast ir<%c>
; VECTOR-NEXT:  Successor(s): vector loop
; VECTOR-EMPTY:
; VECTOR-NEXT:  <x1> vector loop: {
; VECTOR-NEXT:    vector.body:
; VECTOR-NEXT:      EMIT vp<[[VP5:%[0-9]+]]> = CANONICAL-INDUCTION ir<0>, vp<%index.next>
; VECTOR-NEXT:      vp<[[VP6:%[0-9]+]]> = SCALAR-STEPS vp<[[VP5]]>, ir<1>, vp<[[VP0]]>
; VECTOR-NEXT:      vp<[[VP7:%[0-9]+]]> = SCALAR-STEPS vp<[[VP5]]>, ir<1>, vp<[[VP0]]>, ir<1>
; VECTOR-NEXT:    Successor(s): pred.load
; VECTOR-EMPTY:
; VECTOR-NEXT:    <xVFxUF> pred.load: {
; VECTOR-NEXT:      pred.load.entry:
; VECTOR-NEXT:        BRANCH-ON-MASK ir<%c>
; VECTOR-NEXT:      Successor(s): pred.load.if, pred.load.continue
; VECTOR-EMPTY:
; VECTOR-NEXT:      pred.load.if:
; VECTOR-NEXT:        vp<[[VP8:%[0-9]+]]> = SCALAR-STEPS vp<[[VP5]]>, ir<1>, vp<[[VP0]]>
; VECTOR-NEXT:        REPLICATE ir<%gep> = getelementptr ir<%ptr>, vp<[[VP8]]>
; VECTOR-NEXT:        REPLICATE ir<%lv> = load ir<%gep> (S->V)
; VECTOR-NEXT:      Successor(s): pred.load.continue
; VECTOR-EMPTY:
; VECTOR-NEXT:      pred.load.continue:
; VECTOR-NEXT:        PHI-PREDICATED-INSTRUCTION vp<[[VP9:%[0-9]+]]> = ir<%lv>
; VECTOR-NEXT:      No successors
; VECTOR-NEXT:    }
; VECTOR-NEXT:    Successor(s): pred.load
; VECTOR-EMPTY:
; VECTOR-NEXT:    <xVFxUF> pred.load: {
; VECTOR-NEXT:      pred.load.entry:
; VECTOR-NEXT:        BRANCH-ON-MASK ir<%c>
; VECTOR-NEXT:      Successor(s): pred.load.if, pred.load.continue
; VECTOR-EMPTY:
; VECTOR-NEXT:      pred.load.if:
; VECTOR-NEXT:        vp<[[VP10:%[0-9]+]]> = SCALAR-STEPS vp<[[VP5]]>, ir<1>, vp<[[VP0]]>, vp<[[VP0]]>
; VECTOR-NEXT:        REPLICATE ir<%gep>.1 = getelementptr ir<%ptr>, vp<[[VP10]]>
; VECTOR-NEXT:        REPLICATE ir<%lv>.1 = load ir<%gep>.1 (S->V)
; VECTOR-NEXT:      Successor(s): pred.load.continue
; VECTOR-EMPTY:
; VECTOR-NEXT:      pred.load.continue:
; VECTOR-NEXT:        PHI-PREDICATED-INSTRUCTION vp<[[VP11:%[0-9]+]]> = ir<%lv>.1
; VECTOR-NEXT:      No successors
; VECTOR-NEXT:    }
; VECTOR-NEXT:    Successor(s): if.then.0
; VECTOR-EMPTY:
; VECTOR-NEXT:    if.then.0:
; VECTOR-NEXT:      BLEND ir<%pred.val> = ir<0> vp<%9>/vp<[[VP4]]>
; VECTOR-NEXT:      BLEND ir<%pred.val>.1 = ir<0> vp<%11>/vp<[[VP4]]>
; VECTOR-NEXT:      CLONE ir<%gep.dst> = getelementptr ir<%dst>, vp<[[VP6]]>
; VECTOR-NEXT:      EMIT vp<[[VP12:%[0-9]+]]> = mul nuw nsw vp<[[VP0]]>, ir<1>
; VECTOR-NEXT:      vp<[[VP13:%[0-9]+]]> = vector-pointer ir<%gep.dst>
; VECTOR-NEXT:      vp<[[VP14:%[0-9]+]]> = vector-pointer ir<%gep.dst>, vp<[[VP12]]>
; VECTOR-NEXT:      WIDEN store vp<[[VP13]]>, ir<%pred.val>
; VECTOR-NEXT:      WIDEN store vp<[[VP14]]>, ir<%pred.val>.1
; VECTOR-NEXT:      EMIT vp<%index.next> = add nuw vp<[[VP5]]>, vp<[[VP1]]>
; VECTOR-NEXT:      EMIT branch-on-count vp<%index.next>, ir<1024>
; VECTOR-NEXT:    No successors
; VECTOR-NEXT:  }
; VECTOR-NEXT:  Successor(s): middle.block
; VECTOR-EMPTY:
; VECTOR-NEXT:  middle.block:
; VECTOR-NEXT:    EMIT vp<%cmp.n> = icmp eq ir<1024>, ir<1024>
; VECTOR-NEXT:    EMIT branch-on-cond vp<%cmp.n>
; VECTOR-NEXT:  Successor(s): ir-bb<exit>, scalar.ph
; VECTOR-EMPTY:
; VECTOR-NEXT:  ir-bb<exit>:
; VECTOR-NEXT:  No successors
; VECTOR-EMPTY:
; VECTOR-NEXT:  scalar.ph:
; VECTOR-NEXT:    EMIT-SCALAR vp<%bc.resume.val> = phi [ ir<1024>, middle.block ], [ ir<0>, ir-bb<entry> ]
; VECTOR-NEXT:  Successor(s): ir-bb<loop.header>
; VECTOR-EMPTY:
; VECTOR-NEXT:  ir-bb<loop.header>:
; VECTOR-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ] (extra operand: vp<%bc.resume.val> from scalar.ph)
; VECTOR-NEXT:    IR   %gep = getelementptr i8, ptr %ptr, i64 %iv
; VECTOR-NEXT:  No successors
; VECTOR-NEXT:  }
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %gep = getelementptr i8, ptr %ptr, i64 %iv
  br i1 %c, label %if.then, label %loop.latch

if.then:
  %lv = load i8, ptr %gep, align 1
  br label %loop.latch

loop.latch:
  %pred.val = phi i8 [ 0, %loop.header ], [ %lv, %if.then ]
  %gep.dst = getelementptr i8, ptr %dst, i64 %iv
  store i8 %pred.val, ptr %gep.dst, align 1
  %iv.next = add nuw nsw i64 %iv, 1
  %cmp = icmp eq i64 %iv.next, 1024
  br i1 %cmp, label %exit, label %loop.header

exit:
  ret void
}