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

[Nikita Popov via llvm-commits]

================
@@ -0,0 +1,465 @@
+; 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"
+
+; Function Attrs: mustprogress uwtable
+define dso_local void @can_optimize_trap(i32 noundef %block_size) local_unnamed_addr #0 {
+; CHECK-LABEL: define dso_local void @can_optimize_trap(
+; CHECK-SAME: i32 noundef [[BLOCK_SIZE:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[FOO_ARR:%.*]] = alloca [2 x i8], align 16
+; CHECK-NEXT:    [[BAR_ARR:%.*]] = alloca [2 x i8], align 16
+; 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:    [[TMP1:%.*]] = add i32 [[BLOCK_SIZE]], -1
+; CHECK-NEXT:    [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP1]], i32 3)
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 3, [[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:    ret void
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[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, i32 [[I_015]]
+; CHECK-NEXT:    [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1, !tbaa [[TBAA4:![0-9]+]]
+; CHECK-NEXT:    [[TMP4:%.*]] = xor i8 [[TMP3]], 54
+; CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]]
+; CHECK-NEXT:    store i8 [[TMP4]], ptr [[ARRAYIDX7]], align 1, !tbaa [[TBAA4]]
+; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I_015]], 1
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[INC]], [[BLOCK_SIZE]]
+; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP7:![0-9]+]]
+;
+entry:
+  %foo_arr = alloca [2 x i8], align 16
+  %bar_arr = alloca [2 x i8], align 16
+  call void @x(ptr noundef nonnull %foo_arr)
+  %cmp14.not = icmp eq i32 %block_size, 0
+  br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader
+
+for.body.preheader:                               ; preds = %entry
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %if.end4
+  br label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  call void @x(ptr noundef nonnull %bar_arr)
+  ret void
+
+for.body:                                         ; preds = %for.body.preheader, %if.end4
+  %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ]
+  %cmp1 = icmp samesign ugt i32 %i.015, 2
+  br i1 %cmp1, label %if.then, label %if.end4
+
+if.then:                                          ; preds = %for.body
+  call void @llvm.trap()
+  unreachable
+
+if.end4:                                          ; preds = %for.body
+  %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015
+  %0 = load i8, ptr %arrayidx, align 1, !tbaa !5
+  %1 = xor i8 %0, 54
+  %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015
+  store i8 %1, ptr %arrayidx7, align 1, !tbaa !5
+  %inc = add nuw nsw i32 %i.015, 1
+  %cmp = icmp ult i32 %inc, %block_size
+  br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit, !llvm.loop !8
----------------
nikic wrote:

Please drop any unnecessary metadata like tbaa and loop -- unless these tests actually depend on mustprogress metadata?

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