[llvm] [ConstraintElim] Use constraints from bounded memory accesses (PR #155253)

[Nikita Popov via llvm-commits]

================
@@ -0,0 +1,272 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=constraint-elimination -S %s | FileCheck %s
+
+ at g = private unnamed_addr constant [5 x i8] c"test\00"
+
+declare void @free(ptr allocptr noundef captures(none)) mustprogress nounwind willreturn allockind("free") memory(argmem: readwrite, inaccessiblemem: readwrite) "alloc-family"="malloc"
+declare ptr @malloc(i64) mustprogress nofree nounwind willreturn allockind("alloc,uninitialized") allocsize(0) memory(inaccessiblemem: readwrite) "alloc-family"="malloc"
+declare void @callee(i1)
+
+define i8 @load_global(i64 %idx) {
+; CHECK-LABEL: define i8 @load_global(
+; CHECK-SAME: i64 [[IDX:%.*]]) {
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr @g, i64 [[IDX]]
+; CHECK-NEXT:    [[LOAD:%.*]] = load i8, ptr [[GEP]], align 1
+; CHECK-NEXT:    [[ZEXT:%.*]] = zext i1 true to i8
+; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[LOAD]], [[ZEXT]]
+; CHECK-NEXT:    ret i8 [[ADD]]
+;
+  %gep = getelementptr inbounds i8, ptr @g, i64 %idx
+  %load = load i8, ptr %gep
+  %cmp = icmp ult i64 %idx, 5
+  %zext = zext i1 %cmp to i8
+  %add = add i8 %load, %zext
+  ret i8 %add
+}
+
+define i1 @store_global(i64 %idx) {
+; CHECK-LABEL: define i1 @store_global(
+; CHECK-SAME: i64 [[IDX:%.*]]) {
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr @g, i64 [[IDX]]
+; CHECK-NEXT:    store i8 0, ptr [[GEP]], align 1
+; CHECK-NEXT:    ret i1 true
+;
+  %gep = getelementptr inbounds i8, ptr @g, i64 %idx
+  store i8 0, ptr %gep
+  %cmp = icmp ult i64 %idx, 5
+  ret i1 %cmp
+}
+
+define i8 @load_byval(ptr byval([5 x i8]) %p, i64 %idx) {
+; CHECK-LABEL: define i8 @load_byval(
+; CHECK-SAME: ptr byval([5 x i8]) [[P:%.*]], i64 [[IDX:%.*]]) {
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 [[IDX]]
+; CHECK-NEXT:    [[LOAD:%.*]] = load i8, ptr [[GEP]], align 1
+; CHECK-NEXT:    [[ZEXT:%.*]] = zext i1 true to i8
+; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[LOAD]], [[ZEXT]]
+; CHECK-NEXT:    ret i8 [[ADD]]
+;
+  %gep = getelementptr inbounds i8, ptr %p, i64 %idx
+  %load = load i8, ptr %gep
+  %cmp = icmp ult i64 %idx, 5
+  %zext = zext i1 %cmp to i8
+  %add = add i8 %load, %zext
+  ret i8 %add
+}
+
+define i8 @load_alloca(i64 %idx) {
+; CHECK-LABEL: define i8 @load_alloca(
+; CHECK-SAME: i64 [[IDX:%.*]]) {
+; CHECK-NEXT:    [[ALLOC:%.*]] = alloca [5 x i8], align 1
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[ALLOC]], ptr @g, i64 5, i1 false)
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr [[ALLOC]], i64 [[IDX]]
+; CHECK-NEXT:    [[LOAD:%.*]] = load i8, ptr [[GEP]], align 1
+; CHECK-NEXT:    [[ZEXT:%.*]] = zext i1 true to i8
+; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[LOAD]], [[ZEXT]]
+; CHECK-NEXT:    ret i8 [[ADD]]
+;
+  %alloc = alloca [5 x i8], align 1
+  call void @llvm.memcpy.p0.p0.i64(ptr %alloc, ptr @g, i64 5, i1 false)
+  %gep = getelementptr inbounds i8, ptr %alloc, i64 %idx
+  %load = load i8, ptr %gep
+  %cmp = icmp ult i64 %idx, 5
+  %zext = zext i1 %cmp to i8
+  %add = add i8 %load, %zext
+  ret i8 %add
+}
+
+define i8 @load_malloc(i64 %idx) {
+; CHECK-LABEL: define i8 @load_malloc(
+; CHECK-SAME: i64 [[IDX:%.*]]) {
+; CHECK-NEXT:    [[ALLOC:%.*]] = call ptr @malloc(i64 5)
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[ALLOC]], ptr @g, i64 5, i1 false)
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr [[ALLOC]], i64 [[IDX]]
+; CHECK-NEXT:    [[LOAD:%.*]] = load i8, ptr [[GEP]], align 1
+; CHECK-NEXT:    [[ZEXT:%.*]] = zext i1 true to i8
+; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[LOAD]], [[ZEXT]]
+; CHECK-NEXT:    call void @free(ptr [[ALLOC]])
+; CHECK-NEXT:    ret i8 [[ADD]]
+;
+  %alloc = call ptr @malloc(i64 5)
+  call void @llvm.memcpy.p0.p0.i64(ptr %alloc, ptr @g, i64 5, i1 false)
+  %gep = getelementptr inbounds i8, ptr %alloc, i64 %idx
+  %load = load i8, ptr %gep
+  %cmp = icmp ult i64 %idx, 5
+  %zext = zext i1 %cmp to i8
+  %add = add i8 %load, %zext
+  call void @free(ptr %alloc)
+  ret i8 %add
+}
+
+define i32 @load_byval_i32(ptr byval([10 x i8]) %p, i64 %idx) {
+; CHECK-LABEL: define i32 @load_byval_i32(
+; CHECK-SAME: ptr byval([10 x i8]) [[P:%.*]], i64 [[IDX:%.*]]) {
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 [[IDX]]
+; CHECK-NEXT:    [[LOAD:%.*]] = load i32, ptr [[GEP]], align 4
+; CHECK-NEXT:    [[ZEXT:%.*]] = zext i1 true to i32
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[LOAD]], [[ZEXT]]
+; CHECK-NEXT:    ret i32 [[ADD]]
+;
+  %gep = getelementptr inbounds i8, ptr %p, i64 %idx
+  %load = load i32, ptr %gep
+  %cmp = icmp ult i64 %idx, 7
+  %zext = zext i1 %cmp to i32
+  %add = add i32 %load, %zext
+  ret i32 %add
+}
+
+define i8 @load_global_may_noreturn_dom_bb(i64 %idx) {
+; CHECK-LABEL: define i8 @load_global_may_noreturn_dom_bb(
+; CHECK-SAME: i64 [[IDX:%.*]]) {
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr @g, i64 [[IDX]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult i64 [[IDX]], 5
+; CHECK-NEXT:    call void @callee(i1 [[CMP1]])
+; CHECK-NEXT:    [[LOAD:%.*]] = load i8, ptr [[GEP]], align 1
+; CHECK-NEXT:    br label %[[NEXT:.*]]
+; CHECK:       [[NEXT]]:
+; CHECK-NEXT:    [[ZEXT:%.*]] = zext i1 true to i8
+; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[LOAD]], [[ZEXT]]
+; CHECK-NEXT:    ret i8 [[ADD]]
+;
+  %gep = getelementptr inbounds i8, ptr @g, i64 %idx
+  %cmp1 = icmp ult i64 %idx, 5
+  call void @callee(i1 %cmp1) ; %cmp1 should not be simplified.
----------------
nikic wrote:

```suggestion
  call void @may_not_return(i1 %cmp1) ; %cmp1 should not be simplified.
```
Clarify that this is not use a use function.

https://github.com/llvm/llvm-project/pull/155253