test/Analysis/LoopAccessAnalysis/runtime-checks-read-write-merge.ll

; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 6
; RUN: opt -passes='print<access-info>' -memory-check-merge-threshold=0 -disable-output < %s 2>&1 | FileCheck %s

; The read and write entries for %a0 must not exhaust the merge budget by
; processing the same runtime pointer twice. Both accesses to %a then fit in
; one checking group, leaving one runtime check against %b instead of two.

define void @merge_read_write_pointer(ptr %a, ptr %b) {
; CHECK-LABEL: 'merge_read_write_pointer'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe with run-time checks
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %v = load i32, ptr %a0, align 4 ->
; CHECK-NEXT:            store i32 %sum, ptr %a0, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Check 0:
; CHECK-NEXT:        Comparing group GRP0:
; CHECK-NEXT:          %a1 = getelementptr i32, ptr %a, i64 %offset
; CHECK-NEXT:          %a0 = getelementptr i32, ptr %a, i64 %iv
; CHECK-NEXT:        Against group GRP1:
; CHECK-NEXT:          %bp = getelementptr i32, ptr %b, i64 %iv
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (384 + %a))
; CHECK-NEXT:            Member: {(256 + %a),+,4}<nw><%loop>
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-NEXT:        Group GRP1:
; CHECK-NEXT:          (Low: %b High: (128 + %b))
; CHECK-NEXT:            Member: {%b,+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
;
entry:
  br label %loop

loop:
  %iv = phi i64 [0, %entry], [%next, %loop]
  %a0 = getelementptr i32, ptr %a, i64 %iv
  %v = load i32, ptr %a0
  %bp = getelementptr i32, ptr %b, i64 %iv
  %w = load i32, ptr %bp
  %sum = add i32 %v, %w
  store i32 %sum, ptr %a0
  %offset = add nuw nsw i64 %iv, 64
  %a1 = getelementptr i32, ptr %a, i64 %offset
  store i32 %w, ptr %a1
  %next = add nuw nsw i64 %iv, 1
  %done = icmp eq i64 %next, 32
  br i1 %done, label %exit, label %loop

exit:
  ret void
}