test/Transforms/LoopStrengthReduce/AArch64/prefer-all.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -mtriple=aarch64-none-elf -loop-reduce -lsr-preferred-addressing-mode=all < %s | FileCheck %s

define i32 @postindex_loop(ptr %p, i64 %n) {
; CHECK-LABEL: define i32 @postindex_loop(
; CHECK-SAME: ptr [[P:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
; CHECK:       [[FOR_BODY]]:
; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], %[[FOR_BODY]] ], [ [[P]], %[[ENTRY]] ]
; CHECK-NEXT:    [[LSR_IV:%.*]] = phi i64 [ [[LSR_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[N]], %[[ENTRY]] ]
; CHECK-NEXT:    [[RET:%.*]] = phi i32 [ [[ADD:%.*]], %[[FOR_BODY]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT:    [[VAL:%.*]] = load i32, ptr [[LSR_IV1]], align 4
; CHECK-NEXT:    [[ADD]] = add i32 [[RET]], [[VAL]]
; CHECK-NEXT:    [[LSR_IV_NEXT]] = add i64 [[LSR_IV]], -1
; CHECK-NEXT:    [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i64 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[LSR_IV_NEXT]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[EXIT:.*]], label %[[FOR_BODY]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret i32 [[ADD]]
;
entry:
  br label %for.body

for.body:
  %idx = phi i64 [ %idx.next, %for.body ], [ 0, %entry ]
  %ret = phi i32 [ %add, %for.body ], [ 0, %entry ]
  %arrayidx = getelementptr inbounds nuw i32, ptr %p, i64 %idx
  %val = load i32, ptr %arrayidx, align 4
  %add = add i32 %ret, %val
  %idx.next = add nuw nsw i64 %idx, 1
  %exitcond = icmp eq i64 %idx.next, %n
  br i1 %exitcond, label %exit, label %for.body

exit:
  ret i32 %add
}

; Preindex saves a setup instruction compared to postindex
; FIXME: We currently don't recognize that preindex is possible here
define i32 @preindex_loop(ptr %p, i64 %n) {
; CHECK-LABEL: define i32 @preindex_loop(
; CHECK-SAME: ptr [[P:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr nuw i8, ptr [[P]], i64 4
; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
; CHECK:       [[FOR_BODY]]:
; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP2:%.*]], %[[FOR_BODY]] ], [ [[SCEVGEP]], %[[ENTRY]] ]
; CHECK-NEXT:    [[LSR_IV:%.*]] = phi i64 [ [[LSR_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[N]], %[[ENTRY]] ]
; CHECK-NEXT:    [[RET:%.*]] = phi i32 [ [[ADD:%.*]], %[[FOR_BODY]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT:    [[VAL:%.*]] = load i32, ptr [[LSR_IV1]], align 4
; CHECK-NEXT:    [[ADD]] = add i32 [[RET]], [[VAL]]
; CHECK-NEXT:    [[LSR_IV_NEXT]] = add i64 [[LSR_IV]], -1
; CHECK-NEXT:    [[SCEVGEP2]] = getelementptr i8, ptr [[LSR_IV1]], i64 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[LSR_IV_NEXT]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[EXIT:.*]], label %[[FOR_BODY]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret i32 [[ADD]]
;
entry:
  br label %for.body

for.body:
  %idx = phi i64 [ %idx.next, %for.body ], [ 0, %entry ]
  %ret = phi i32 [ %add, %for.body ], [ 0, %entry ]
  %idx.next = add nuw nsw i64 %idx, 1
  %arrayidx = getelementptr inbounds nuw i32, ptr %p, i64 %idx.next
  %val = load i32, ptr %arrayidx, align 4
  %add = add i32 %ret, %val
  %exitcond = icmp eq i64 %idx.next, %n
  br i1 %exitcond, label %exit, label %for.body

exit:
  ret i32 %add
}

; We should use offset addressing here as postindex uses an extra register.
; FIXME: We currently use postindex as we don't realize the load of val2 is also
; a use of p that needs it to be live in the loop.
define i64 @offset_loop(ptr %p, i64 %n) {
; CHECK-LABEL: define i64 @offset_loop(
; CHECK-SAME: ptr [[P:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
; CHECK:       [[FOR_BODY]]:
; CHECK-NEXT:    [[LSR_IV:%.*]] = phi ptr [ [[SCEVGEP:%.*]], %[[FOR_BODY]] ], [ [[P]], %[[ENTRY]] ]
; CHECK-NEXT:    [[RET:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[ADD:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT:    [[IDX:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IDX_NEXT:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT:    [[VAL1:%.*]] = load i64, ptr [[LSR_IV]], align 4
; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds nuw i64, ptr [[P]], i64 [[VAL1]]
; CHECK-NEXT:    [[VAL2:%.*]] = load i64, ptr [[ARRAYIDX2]], align 4
; CHECK-NEXT:    [[ADD]] = add i64 [[VAL2]], [[RET]]
; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i64 [[IDX]], 1
; CHECK-NEXT:    [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV]], i64 8
; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[IDX_NEXT]], [[VAL1]]
; CHECK-NEXT:    br i1 [[CMP]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
; CHECK:       [[FOR_END]]:
; CHECK-NEXT:    ret i64 [[ADD]]
;
entry:
  br label %for.body

for.body:
  %ret = phi i64 [ 0, %entry ], [ %add, %for.body ]
  %idx = phi i64 [ 0, %entry ], [ %idx.next, %for.body ]
  %arrayidx1 = getelementptr inbounds nuw i64, ptr %p, i64 %idx
  %val1 = load i64, ptr %arrayidx1, align 4
  %arrayidx2 = getelementptr inbounds nuw i64, ptr %p, i64 %val1
  %val2 = load i64, ptr %arrayidx2, align 4
  %add = add i64 %val2, %ret
  %idx.next = add nuw nsw i64 %idx, 1
  %cmp = icmp eq i64 %idx.next, %val1
  br i1 %cmp, label %for.end, label %for.body

for.end:
  ret i64 %add
}

; We can't use postindex addressing on the conditional load of qval and can't
; convert the loop condition to a compare with zero, so we should instead use
; offset addressing.
; FIXME: Currently we don't notice the load of qval is conditional, and attempt
; postindex addressing anyway.
define i32 @conditional_load(ptr %p, ptr %q, ptr %n) {
; CHECK-LABEL: define i32 @conditional_load(
; CHECK-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], ptr [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
; CHECK:       [[FOR_BODY]]:
; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP2:%.*]], %[[FOR_INC:.*]] ], [ [[P]], %[[ENTRY]] ]
; CHECK-NEXT:    [[LSR_IV:%.*]] = phi ptr [ [[SCEVGEP:%.*]], %[[FOR_INC]] ], [ [[Q]], %[[ENTRY]] ]
; CHECK-NEXT:    [[IDX:%.*]] = phi i64 [ [[IDX_NEXT:%.*]], %[[FOR_INC]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT:    [[RET:%.*]] = phi i32 [ [[RET_NEXT:%.*]], %[[FOR_INC]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT:    [[PVAL:%.*]] = load i32, ptr [[LSR_IV1]], align 4
; CHECK-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[PVAL]], 0
; CHECK-NEXT:    [[SCEVGEP2]] = getelementptr i8, ptr [[LSR_IV1]], i64 4
; CHECK-NEXT:    br i1 [[TOBOOL_NOT]], label %[[FOR_INC]], label %[[IF_THEN:.*]]
; CHECK:       [[IF_THEN]]:
; CHECK-NEXT:    [[QVAL:%.*]] = load i32, ptr [[LSR_IV]], align 4
; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[RET]], [[QVAL]]
; CHECK-NEXT:    br label %[[FOR_INC]]
; CHECK:       [[FOR_INC]]:
; CHECK-NEXT:    [[RET_NEXT]] = phi i32 [ [[ADD]], %[[IF_THEN]] ], [ [[RET]], %[[FOR_BODY]] ]
; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i64 [[IDX]], 1
; CHECK-NEXT:    [[NVAL:%.*]] = load volatile i64, ptr [[N]], align 8
; CHECK-NEXT:    [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV]], i64 4
; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i64 [[IDX_NEXT]], [[NVAL]]
; CHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY]], label %[[EXIT:.*]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret i32 [[RET_NEXT]]
;
entry:
  br label %for.body

for.body:
  %idx = phi i64 [ %idx.next, %for.inc ], [ 0, %entry ]
  %ret = phi i32 [ %ret.next, %for.inc ], [ 0, %entry ]
  %arrayidx = getelementptr inbounds nuw i32, ptr %p, i64 %idx
  %pval = load i32, ptr %arrayidx, align 4
  %tobool.not = icmp eq i32 %pval, 0
  br i1 %tobool.not, label %for.inc, label %if.then

if.then:
  %arrayidx1 = getelementptr inbounds nuw i32, ptr %q, i64 %idx
  %qval = load i32, ptr %arrayidx1, align 4
  %add = add i32 %ret, %qval
  br label %for.inc

for.inc:
  %ret.next = phi i32 [ %add, %if.then ], [ %ret, %for.body ]
  %idx.next = add nuw nsw i64 %idx, 1
  %nval = load volatile i64, ptr %n, align 8
  %cmp = icmp slt i64 %idx.next, %nval
  br i1 %cmp, label %for.body, label %exit

exit:
  ret i32 %ret.next
}