test/Transforms/LoopVectorize/predicated-load-access-safety.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph" --version 6
; RUN: opt -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S %s | FileCheck %s

; A byte load does not make a wider load at the same pointer safe.
define void @wider_predicated_load(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @wider_predicated_load(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT:    [[TMP8:%.*]] = load i8, ptr [[TMP4]], align 1
; CHECK-NEXT:    [[TMP9:%.*]] = load i8, ptr [[TMP5]], align 1
; CHECK-NEXT:    [[TMP10:%.*]] = load i8, ptr [[TMP6]], align 1
; CHECK-NEXT:    [[TMP11:%.*]] = load i8, ptr [[TMP7]], align 1
; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <4 x i8> poison, i8 [[TMP8]], i64 0
; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <4 x i8> [[TMP12]], i8 [[TMP9]], i64 1
; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <4 x i8> [[TMP13]], i8 [[TMP10]], i64 2
; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <4 x i8> [[TMP14]], i8 [[TMP11]], i64 3
; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i8> [[TMP15]], zeroinitializer
; CHECK-NEXT:    [[TMP32:%.*]] = load <4 x i64>, ptr [[TMP4]], align 1
; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP16]], <4 x i64> [[TMP32]], <4 x i64> zeroinitializer
; CHECK-NEXT:    [[TMP33:%.*]] = getelementptr i64, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT:    store <4 x i64> [[PREDPHI]], ptr [[TMP33]], align 1
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT:    [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP34]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %p = getelementptr i64, ptr %src, i64 %iv
  %guard = load i8, ptr %p, align 1
  %keep = icmp ne i8 %guard, 0
  br i1 %keep, label %read, label %loop.latch

read:
  %value = load i64, ptr %p, align 1
  br label %loop.latch

loop.latch:
  %result = phi i64 [ %value, %read ], [ 0, %loop.header ]
  %out = getelementptr i64, ptr %dst, i64 %iv
  store i64 %result, ptr %out, align 1
  %iv.next = add nuw i64 %iv, 1
  %done = icmp eq i64 %iv.next, %n
  br i1 %done, label %exit, label %loop.header

exit:
  ret void
}

; A weaker alignment does not justify the predicated load alignment.
define void @stronger_predicated_alignment(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @stronger_predicated_alignment(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP4]], align 4
; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
; CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP6]], align 4
; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP7]], align 4
; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <4 x i32> poison, i32 [[TMP8]], i64 0
; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <4 x i32> [[TMP12]], i32 [[TMP9]], i64 1
; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <4 x i32> [[TMP13]], i32 [[TMP10]], i64 2
; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <4 x i32> [[TMP14]], i32 [[TMP11]], i64 3
; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
; CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[TMP4]], align 16
; CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP5]], align 16
; CHECK-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP6]], align 16
; CHECK-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP7]], align 16
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i32> poison, i32 [[TMP18]], i64 0
; CHECK-NEXT:    [[TMP23:%.*]] = insertelement <4 x i32> [[TMP21]], i32 [[TMP22]], i64 1
; CHECK-NEXT:    [[TMP28:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP26]], i64 2
; CHECK-NEXT:    [[TMP31:%.*]] = insertelement <4 x i32> [[TMP28]], i32 [[TMP30]], i64 3
; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP16]], <4 x i32> [[TMP31]], <4 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP33:%.*]] = getelementptr i32, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT:    store <4 x i32> [[PREDPHI]], ptr [[TMP33]], align 1
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT:    [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP34]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %p = getelementptr [4 x i32], ptr %src, i64 %iv
  %guard = load i32, ptr %p, align 4
  %keep = icmp ne i32 %guard, 0
  br i1 %keep, label %read, label %loop.latch

read:
  %value = load i32, ptr %p, align 16
  br label %loop.latch

loop.latch:
  %result = phi i32 [ %value, %read ], [ 0, %loop.header ]
  %out = getelementptr i32, ptr %dst, i64 %iv
  store i32 %result, ptr %out, align 1
  %iv.next = add nuw i64 %iv, 1
  %done = icmp eq i64 %iv.next, %n
  br i1 %done, label %exit, label %loop.header

exit:
  ret void
}

; Keep the strongest known alignment when several accesses share a pointer.
define void @sufficient_alignment(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @sufficient_alignment(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP4]], align 16
; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 16
; CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP6]], align 16
; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP7]], align 16
; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <4 x i32> poison, i32 [[TMP8]], i64 0
; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <4 x i32> [[TMP12]], i32 [[TMP9]], i64 1
; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <4 x i32> [[TMP13]], i32 [[TMP10]], i64 2
; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <4 x i32> [[TMP14]], i32 [[TMP11]], i64 3
; CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP4]], align 4
; CHECK-NEXT:    [[TMP17:%.*]] = load i32, ptr [[TMP5]], align 4
; CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[TMP6]], align 4
; CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[TMP7]], align 4
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i32> poison, i32 [[TMP16]], i64 0
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i32> [[TMP20]], i32 [[TMP17]], i64 1
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i32> [[TMP21]], i32 [[TMP18]], i64 2
; CHECK-NEXT:    [[TMP23:%.*]] = insertelement <4 x i32> [[TMP22]], i32 [[TMP19]], i64 3
; CHECK-NEXT:    [[TMP24:%.*]] = add <4 x i32> [[TMP15]], [[TMP23]]
; CHECK-NEXT:    [[TMP25:%.*]] = icmp ne <4 x i32> [[TMP24]], zeroinitializer
; CHECK-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP4]], align 16
; CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[TMP5]], align 16
; CHECK-NEXT:    [[TMP28:%.*]] = load i32, ptr [[TMP6]], align 16
; CHECK-NEXT:    [[TMP29:%.*]] = load i32, ptr [[TMP7]], align 16
; CHECK-NEXT:    [[TMP30:%.*]] = insertelement <4 x i32> poison, i32 [[TMP26]], i64 0
; CHECK-NEXT:    [[TMP31:%.*]] = insertelement <4 x i32> [[TMP30]], i32 [[TMP27]], i64 1
; CHECK-NEXT:    [[TMP32:%.*]] = insertelement <4 x i32> [[TMP31]], i32 [[TMP28]], i64 2
; CHECK-NEXT:    [[TMP33:%.*]] = insertelement <4 x i32> [[TMP32]], i32 [[TMP29]], i64 3
; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP25]], <4 x i32> [[TMP33]], <4 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP34:%.*]] = getelementptr i32, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT:    store <4 x i32> [[PREDPHI]], ptr [[TMP34]], align 1
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT:    [[TMP35:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP35]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %p = getelementptr [4 x i32], ptr %src, i64 %iv
  %guard = load i32, ptr %p, align 16
  %weak = load i32, ptr %p, align 4
  %sum = add i32 %guard, %weak
  %keep = icmp ne i32 %sum, 0
  br i1 %keep, label %read, label %loop.latch

read:
  %value = load i32, ptr %p, align 16
  br label %loop.latch

loop.latch:
  %result = phi i32 [ %value, %read ], [ 0, %loop.header ]
  %out = getelementptr i32, ptr %dst, i64 %iv
  store i32 %result, ptr %out, align 1
  %iv.next = add nuw i64 %iv, 1
  %done = icmp eq i64 %iv.next, %n
  br i1 %done, label %exit, label %loop.header

exit:
  ret void
}