[llvm] [IndVarSimplify] Allow predicateLoopExit on some loops with local writes (PR #155901)

[Florian Mayer via llvm-commits]

================
@@ -0,0 +1,536 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -passes=indvars < %s | FileCheck %s
+
+source_filename = "/usr/local/google/home/fmayer/loop/src3.cc"
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Function Attrs: mustprogress uwtable
+define dso_local void @foo(i32 noundef %block_size) local_unnamed_addr #0 {
+; CHECK-LABEL: define dso_local void @foo(
+; CHECK-SAME: i32 noundef [[BLOCK_SIZE:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[FOO_ARR:%.*]] = alloca [1024 x i8], align 16
+; CHECK-NEXT:    [[BAR_ARR:%.*]] = alloca [1025 x i8], align 16
+; CHECK-NEXT:    call void @llvm.lifetime.start.p0(ptr nonnull [[FOO_ARR]]) #[[ATTR6:[0-9]+]]
+; CHECK-NEXT:    call void @llvm.lifetime.start.p0(ptr nonnull [[BAR_ARR]]) #[[ATTR6]]
+; CHECK-NEXT:    call void @x(ptr noundef nonnull [[FOO_ARR]])
+; CHECK-NEXT:    [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0
+; CHECK-NEXT:    br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK:       [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[BLOCK_SIZE]] to i64
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[BLOCK_SIZE]], -1
+; CHECK-NEXT:    [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP1]], i32 1024)
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 1024, [[UMIN]]
+; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
+; CHECK:       [[FOR_COND_CLEANUP_LOOPEXIT:.*]]:
+; CHECK-NEXT:    br label %[[FOR_COND_CLEANUP]]
+; CHECK:       [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT:    call void @x(ptr noundef nonnull [[BAR_ARR]])
+; CHECK-NEXT:    call void @llvm.lifetime.end.p0(ptr nonnull [[BAR_ARR]]) #[[ATTR6]]
+; CHECK-NEXT:    call void @llvm.lifetime.end.p0(ptr nonnull [[FOO_ARR]]) #[[ATTR6]]
+; CHECK-NEXT:    ret void
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    br i1 [[TMP2]], label %[[IF_THEN:.*]], label %[[IF_END4]]
+; CHECK:       [[IF_THEN]]:
+; CHECK-NEXT:    call void @llvm.trap()
+; CHECK-NEXT:    unreachable
+; CHECK:       [[IF_END4]]:
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1, !tbaa [[TBAA5:![0-9]+]]
+; CHECK-NEXT:    [[TMP4:%.*]] = xor i8 [[TMP3]], 54
+; CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i64 [[INDVARS_IV]]
+; CHECK-NEXT:    store i8 [[TMP4]], ptr [[ARRAYIDX7]], align 1, !tbaa [[TBAA5]]
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[TMP0]]
+; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP8:![0-9]+]]
+;
+entry:
+  %foo_arr = alloca [1024 x i8], align 16
+  %bar_arr = alloca [1025 x i8], align 16
+  call void @llvm.lifetime.start.p0(ptr nonnull %foo_arr) #4
+  call void @llvm.lifetime.start.p0(ptr nonnull %bar_arr) #4
----------------
fmayer wrote:

Removed some unnecessary stuff. Left the AA metadata in case we go down that route later.

https://github.com/llvm/llvm-project/pull/155901