llvm/test/Transforms/InferAlignment/memintrinsics.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt < %s -passes=infer-alignment -S | FileCheck %s

declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1)
declare void @llvm.memmove.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1)
declare void @llvm.memset.p0.i64(ptr nocapture, i8, i64, i1)

define void @memcpy(i64 %len) {
; CHECK-LABEL: define void @memcpy(
; CHECK-SAME: i64 [[LEN:%.*]]) {
; CHECK-NEXT:    [[DST:%.*]] = alloca [64 x i8], align 32
; CHECK-NEXT:    [[SRC:%.*]] = alloca [64 x i8], align 16
; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 32 [[DST]], ptr align 16 [[SRC]], i64 [[LEN]], i1 false)
; CHECK-NEXT:    ret void
;
  %dst = alloca [64 x i8], align 32
  %src = alloca [64 x i8], align 16
  call void @llvm.memcpy.p0.p0.i64(ptr %dst, ptr %src, i64 %len, i1 false)
  ret void
}

define void @memmove(i64 %len) {
; CHECK-LABEL: define void @memmove(
; CHECK-SAME: i64 [[LEN:%.*]]) {
; CHECK-NEXT:    [[DST:%.*]] = alloca [64 x i8], align 64
; CHECK-NEXT:    [[SRC:%.*]] = alloca [64 x i8], align 8
; CHECK-NEXT:    call void @llvm.memmove.p0.p0.i64(ptr align 64 [[DST]], ptr align 8 [[SRC]], i64 [[LEN]], i1 false)
; CHECK-NEXT:    ret void
;
  %dst = alloca [64 x i8], align 64
  %src = alloca [64 x i8], align 8
  call void @llvm.memmove.p0.p0.i64(ptr %dst, ptr %src, i64 %len, i1 false)
  ret void
}

define void @memset(i64 %len) {
; CHECK-LABEL: define void @memset(
; CHECK-SAME: i64 [[LEN:%.*]]) {
; CHECK-NEXT:    [[DST:%.*]] = alloca [64 x i8], align 32
; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr align 32 [[DST]], i8 0, i64 [[LEN]], i1 false)
; CHECK-NEXT:    ret void
;
  %dst = alloca [64 x i8], align 32
  call void @llvm.memset.p0.i64(ptr %dst, i8 0, i64 %len, i1 false)
  ret void
}

; The align attribute on a memory intrinsic is not proof of an access (the
; length may be zero and the pointer poison), so it must not be used to
; infer the alignment of other accesses to the same pointer.
define void @memset_attr_does_not_seed_store(ptr %p, i64 %len) {
; CHECK-LABEL: define void @memset_attr_does_not_seed_store(
; CHECK-SAME: ptr [[P:%.*]], i64 [[LEN:%.*]]) {
; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr align 16 [[P]], i8 0, i64 [[LEN]], i1 false)
; CHECK-NEXT:    store i32 0, ptr [[P]], align 4
; CHECK-NEXT:    ret void
;
  call void @llvm.memset.p0.i64(ptr align 16 %p, i8 0, i64 %len, i1 false)
  store i32 0, ptr %p
  ret void
}

define i32 @memcpy_attr_does_not_seed_load_store(ptr %dst, ptr %src, i64 %len) {
; CHECK-LABEL: define i32 @memcpy_attr_does_not_seed_load_store(
; CHECK-SAME: ptr [[DST:%.*]], ptr [[SRC:%.*]], i64 [[LEN:%.*]]) {
; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 16 [[DST]], ptr align 32 [[SRC]], i64 [[LEN]], i1 false)
; CHECK-NEXT:    [[V:%.*]] = load i32, ptr [[SRC]], align 4
; CHECK-NEXT:    store i32 [[V]], ptr [[DST]], align 4
; CHECK-NEXT:    ret i32 [[V]]
;
  call void @llvm.memcpy.p0.p0.i64(ptr align 16 %dst, ptr align 32 %src, i64 %len, i1 false)
  %v = load i32, ptr %src
  store i32 %v, ptr %dst
  ret i32 %v
}

define void @memset_attr_does_not_seed_dominated_store(ptr %p, i64 %len, i1 %c) {
; CHECK-LABEL: define void @memset_attr_does_not_seed_dominated_store(
; CHECK-SAME: ptr [[P:%.*]], i64 [[LEN:%.*]], i1 [[C:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr align 16 [[P]], i8 0, i64 [[LEN]], i1 false)
; CHECK-NEXT:    br i1 [[C]], label %[[THEN:.*]], label %[[EXIT:.*]]
; CHECK:       [[THEN]]:
; CHECK-NEXT:    [[Q:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 8
; CHECK-NEXT:    store i32 0, ptr [[Q]], align 4
; CHECK-NEXT:    br label %[[EXIT]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
entry:
  call void @llvm.memset.p0.i64(ptr align 16 %p, i8 0, i64 %len, i1 false)
  br i1 %c, label %then, label %exit

then:
  %q = getelementptr inbounds i8, ptr %p, i64 8
  store i32 0, ptr %q
  br label %exit

exit:
  ret void
}

; With noundef (or dereferenceable) the pointer cannot be poison, so the
; align attribute is a real guarantee and may be used.
define void @memset_noundef_attr_seeds_store(ptr %p, i64 %len) {
; CHECK-LABEL: define void @memset_noundef_attr_seeds_store(
; CHECK-SAME: ptr [[P:%.*]], i64 [[LEN:%.*]]) {
; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr noundef align 16 [[P]], i8 0, i64 [[LEN]], i1 false)
; CHECK-NEXT:    store i32 0, ptr [[P]], align 16
; CHECK-NEXT:    ret void
;
  call void @llvm.memset.p0.i64(ptr noundef align 16 %p, i8 0, i64 %len, i1 false)
  store i32 0, ptr %p
  ret void
}

define i32 @memcpy_dereferenceable_attr_seeds_load_store(ptr %dst, ptr %src, i64 %len) {
; CHECK-LABEL: define i32 @memcpy_dereferenceable_attr_seeds_load_store(
; CHECK-SAME: ptr [[DST:%.*]], ptr [[SRC:%.*]], i64 [[LEN:%.*]]) {
; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 16 dereferenceable(16) [[DST]], ptr noundef align 32 [[SRC]], i64 [[LEN]], i1 false)
; CHECK-NEXT:    [[V:%.*]] = load i32, ptr [[SRC]], align 32
; CHECK-NEXT:    store i32 [[V]], ptr [[DST]], align 16
; CHECK-NEXT:    ret i32 [[V]]
;
  call void @llvm.memcpy.p0.p0.i64(ptr align 16 dereferenceable(16) %dst, ptr noundef align 32 %src, i64 %len, i1 false)
  %v = load i32, ptr %src
  store i32 %v, ptr %dst
  ret i32 %v
}

; A real store does prove the alignment, so the memset may still use it.
define void @store_seeds_memset(ptr %p, i64 %len) {
; CHECK-LABEL: define void @store_seeds_memset(
; CHECK-SAME: ptr [[P:%.*]], i64 [[LEN:%.*]]) {
; CHECK-NEXT:    store i32 0, ptr [[P]], align 16
; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr align 16 [[P]], i8 0, i64 [[LEN]], i1 false)
; CHECK-NEXT:    ret void
;
  store i32 0, ptr %p, align 16
  call void @llvm.memset.p0.i64(ptr %p, i8 0, i64 %len, i1 false)
  ret void
}