[IRCE] Add NSW flag to main loop's indvar base

We have guarantees that induction variable will not overflow in the main
loop after the loop constrained. Therefore we can add no wrap flags on
its base in order not to miss info that loop is countable.

Add NSW flag now, since adding NUW flag requires a bit more complicated
analysis.

Reviewed By: skatkov

Differential Revision: https://reviews.llvm.org/D154954

llvm/lib/Transforms/Scalar/InductiveRangeCheckElimination.cpp

@@ -1717,6 +1717,27 @@ bool LoopConstrainer::run() {
     CanonicalizeLoop(PostL, false);
   CanonicalizeLoop(&OriginalLoop, true);
 
+  /// At this point:
+  /// - We've broken a "main loop" out of the loop in a way that the "main loop"
+  /// runs with the induction variable in a subset of [Begin, End).
+  /// - There is no overflow when computing "main loop" exit limit.
+  /// - Max latch taken count of the loop is limited.
+  /// It guarantees that induction variable will not overflow iterating in the
+  /// "main loop".
+  if (auto BO = dyn_cast<BinaryOperator>(MainLoopStructure.IndVarBase))
+    if (IsSignedPredicate)
+      BO->setHasNoSignedWrap(true);
+  /// TODO: support unsigned predicate.
+  /// To add NUW flag we need to prove that both operands of BO are
+  /// non-negative. E.g:
+  /// ...
+  /// %iv.next = add nsw i32 %iv, -1
+  /// %cmp = icmp ult i32 %iv.next, %n
+  /// br i1 %cmp, label %loopexit, label %loop
+  ///
+  /// -1 is MAX_UINT in terms of unsigned int. Adding anything but zero will
+  /// overflow, therefore NUW flag is not legal here.
+
   return true;
 }
 

llvm/test/Transforms/IRCE/conjunctive-checks.ll

@@ -18,7 +18,7 @@ define void @f_0(ptr %arr, ptr %a_len_ptr, i32 %n, ptr %cond_buf) {
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IDX:%.*]] = phi i32 [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ], [ 0, [[LOOP_PREHEADER1]] ]
-; CHECK-NEXT:    [[IDX_NEXT]] = add i32 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nsw i32 [[IDX]], 1
 ; CHECK-NEXT:    [[IDX_FOR_ABC:%.*]] = add i32 [[IDX]], 4
 ; CHECK-NEXT:    [[ABC_ACTUAL:%.*]] = icmp slt i32 [[IDX_FOR_ABC]], [[LEN]]
 ; CHECK-NEXT:    [[COND:%.*]] = load volatile i1, ptr [[COND_BUF:%.*]], align 1
@@ -64,7 +64,7 @@ define void @f_0(ptr %arr, ptr %a_len_ptr, i32 %n, ptr %cond_buf) {
 ; CHECK-NEXT:    [[ADDR_POSTLOOP:%.*]] = getelementptr i32, ptr [[ARR]], i32 [[IDX_FOR_ABC_POSTLOOP]]
 ; CHECK-NEXT:    store i32 0, ptr [[ADDR_POSTLOOP]], align 4
 ; CHECK-NEXT:    [[NEXT_POSTLOOP:%.*]] = icmp slt i32 [[IDX_NEXT_POSTLOOP]], [[N]]
-; CHECK-NEXT:    br i1 [[NEXT_POSTLOOP]], label [[LOOP_POSTLOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP2:![0-9]+]], !irce.loop.clone !7
+; CHECK-NEXT:    br i1 [[NEXT_POSTLOOP]], label [[LOOP_POSTLOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP2:![0-9]+]], !irce.loop.clone [[META7:![0-9]+]]
 ;
 entry:
   %len = load i32, ptr %a_len_ptr, !range !0
@@ -110,7 +110,7 @@ define void @f_1(
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IDX:%.*]] = phi i32 [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ], [ 0, [[LOOP_PREHEADER2]] ]
-; CHECK-NEXT:    [[IDX_NEXT]] = add i32 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nsw i32 [[IDX]], 1
 ; CHECK-NEXT:    [[ABC_A:%.*]] = icmp slt i32 [[IDX]], [[LEN_A]]
 ; CHECK-NEXT:    [[ABC_B:%.*]] = icmp slt i32 [[IDX]], [[LEN_B]]
 ; CHECK-NEXT:    [[ABC:%.*]] = and i1 true, true
@@ -158,7 +158,7 @@ define void @f_1(
 ; CHECK-NEXT:    [[ADDR_B_POSTLOOP:%.*]] = getelementptr i32, ptr [[ARR_B]], i32 [[IDX_POSTLOOP]]
 ; CHECK-NEXT:    store i32 -1, ptr [[ADDR_B_POSTLOOP]], align 4
 ; CHECK-NEXT:    [[NEXT_POSTLOOP:%.*]] = icmp slt i32 [[IDX_NEXT_POSTLOOP]], [[N]]
-; CHECK-NEXT:    br i1 [[NEXT_POSTLOOP]], label [[LOOP_POSTLOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP8:![0-9]+]], !irce.loop.clone !7
+; CHECK-NEXT:    br i1 [[NEXT_POSTLOOP]], label [[LOOP_POSTLOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP8:![0-9]+]], !irce.loop.clone [[META7]]
 ;
   ptr %arr_a, ptr %a_len_ptr, ptr %arr_b, ptr %b_len_ptr, i32 %n) {
 

llvm/test/Transforms/IRCE/correct-loop-info.ll

@@ -34,7 +34,7 @@ define void @baz() personality ptr @ham {
 ; CHECK-NEXT:    invoke void @pluto()
 ; CHECK-NEXT:    to label [[BB5:%.*]] unwind label [[OUTER_EXITING_LOOPEXIT_SPLIT_LP_LOOPEXIT_SPLIT_LP:%.*]]
 ; CHECK:       bb5:
-; CHECK-NEXT:    [[TMP6]] = add i32 [[TMP4]], 1
+; CHECK-NEXT:    [[TMP6]] = add nsw i32 [[TMP4]], 1
 ; CHECK-NEXT:    [[TMP7:%.*]] = icmp slt i32 [[TMP6]], 1
 ; CHECK-NEXT:    br i1 true, label [[BB8]], label [[EXIT3_LOOPEXIT5:%.*]]
 ; CHECK:       bb8:
@@ -99,7 +99,7 @@ define void @baz() personality ptr @ham {
 ; CHECK:       bb8.preloop:
 ; CHECK-NEXT:    [[TMP9_PRELOOP:%.*]] = icmp slt i32 [[TMP6_PRELOOP]], 84
 ; CHECK-NEXT:    [[TMP4:%.*]] = icmp slt i32 [[TMP6_PRELOOP]], [[EXIT_PRELOOP_AT]]
-; CHECK-NEXT:    br i1 [[TMP4]], label [[INNERHEADER_PRELOOP]], label [[PRELOOP_EXIT_SELECTOR:%.*]], [[LOOP0:!llvm.loop !.*]], !irce.loop.clone !5
+; CHECK-NEXT:    br i1 [[TMP4]], label [[INNERHEADER_PRELOOP]], label [[PRELOOP_EXIT_SELECTOR:%.*]], !llvm.loop [[LOOP0:![0-9]+]], !irce.loop.clone [[META5:![0-9]+]]
 ; CHECK:       preloop.exit.selector:
 ; CHECK-NEXT:    [[TMP6_PRELOOP_LCSSA:%.*]] = phi i32 [ [[TMP6_PRELOOP]], [[BB8_PRELOOP]] ]
 ; CHECK-NEXT:    [[TMP5:%.*]] = icmp slt i32 [[TMP6_PRELOOP_LCSSA]], 84
@@ -120,7 +120,7 @@ define void @baz() personality ptr @ham {
 ; CHECK-NEXT:    br i1 [[TMP7_POSTLOOP]], label [[BB8_POSTLOOP]], label [[EXIT3_LOOPEXIT4:%.*]]
 ; CHECK:       bb8.postloop:
 ; CHECK-NEXT:    [[TMP9_POSTLOOP:%.*]] = icmp slt i32 [[TMP6_POSTLOOP]], 84
-; CHECK-NEXT:    br i1 [[TMP9_POSTLOOP]], label [[INNERHEADER_POSTLOOP]], label [[BB13_LOOPEXIT:%.*]], [[LOOP6:!llvm.loop !.*]], !irce.loop.clone !5
+; CHECK-NEXT:    br i1 [[TMP9_POSTLOOP]], label [[INNERHEADER_POSTLOOP]], label [[BB13_LOOPEXIT:%.*]], !llvm.loop [[LOOP6:![0-9]+]], !irce.loop.clone [[META5]]
 ;
 bb:
   br label %outerheader

llvm/test/Transforms/IRCE/iv-plus-offset-range-check.ll

@@ -56,7 +56,7 @@ define i8 @test1(i8 %limit, i8 %n) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp sge i8 [[SUB]], 2
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT8:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP11]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -150,7 +150,7 @@ define i8 @test1a(i8 %limit, ptr %p) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp sge i8 [[SUB]], 2
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT5:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP4:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP4]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -404,7 +404,7 @@ define i8 @test3a(i8 %limit, ptr %p) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp sle i8 [[ADD]], [[N]]
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT4:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP2]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -508,7 +508,7 @@ define i8 @test4(i8 %limit, i8 %n) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp sgt i8 [[SUB]], 2
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT9:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP11]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -607,7 +607,7 @@ define i8 @test4a(i8 %limit, ptr %p) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp sgt i8 [[SUB]], 2
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT7:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP7]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -812,7 +812,7 @@ define i8 @test6(i8 %limit, i8 %n) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp slt i8 [[ADD]], [[N]]
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT7:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP5]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -903,7 +903,7 @@ define i8 @test6a(i8 %limit, ptr %p) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp slt i8 [[ADD]], [[N]]
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT4:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP2]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -997,7 +997,7 @@ define i8 @test_overflow_check_compile_time(i8 %limit, ptr %p) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp sgt i8 [[SUB]], -2
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT4:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP2]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
@@ -1106,7 +1106,7 @@ define i8 @test_overflow_check_runtime(i8 %limit, ptr %p) {
 ; CHECK-NEXT:    [[CHECK:%.*]] = icmp sge i8 [[SUB]], -2
 ; CHECK-NEXT:    br i1 true, label [[INBOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT8:%.*]]
 ; CHECK:       inbounds:
-; CHECK-NEXT:    [[IDX_NEXT]] = add nuw i8 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nuw nsw i8 [[IDX]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i8 [[IDX_NEXT]], [[LIMIT]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = icmp slt i8 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP12]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]

llvm/test/Transforms/IRCE/multiple-access-no-preloop.ll

@@ -19,7 +19,7 @@ define void @multiple_access_no_preloop(
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IDX:%.*]] = phi i32 [ [[IDX_NEXT:%.*]], [[IN_BOUNDS_B:%.*]] ], [ 0, [[LOOP_PREHEADER2]] ]
-; CHECK-NEXT:    [[IDX_NEXT]] = add i32 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nsw i32 [[IDX]], 1
 ; CHECK-NEXT:    [[ABC_A:%.*]] = icmp slt i32 [[IDX]], [[LEN_A]]
 ; CHECK-NEXT:    br i1 true, label [[IN_BOUNDS_A:%.*]], label [[OUT_OF_BOUNDS_LOOPEXIT3:%.*]], !prof [[PROF1:![0-9]+]]
 ; CHECK:       in.bounds.a:
@@ -69,7 +69,7 @@ define void @multiple_access_no_preloop(
 ; CHECK-NEXT:    [[ADDR_B_POSTLOOP:%.*]] = getelementptr i32, ptr [[ARR_B]], i32 [[IDX_POSTLOOP]]
 ; CHECK-NEXT:    store i32 -1, ptr [[ADDR_B_POSTLOOP]], align 4
 ; CHECK-NEXT:    [[NEXT_POSTLOOP:%.*]] = icmp slt i32 [[IDX_NEXT_POSTLOOP]], [[N]]
-; CHECK-NEXT:    br i1 [[NEXT_POSTLOOP]], label [[LOOP_POSTLOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP2:![0-9]+]], !irce.loop.clone !7
+; CHECK-NEXT:    br i1 [[NEXT_POSTLOOP]], label [[LOOP_POSTLOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP2:![0-9]+]], !irce.loop.clone [[META7:![0-9]+]]
 ;
   ptr %arr_a, ptr %a_len_ptr, ptr %arr_b, ptr %b_len_ptr, i32 %n) {
 

llvm/test/Transforms/IRCE/pre_post_loops.ll

@@ -17,7 +17,7 @@ define void @test_01(ptr %arr, ptr %a_len_ptr) {
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IDX:%.*]] = phi i32 [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
-; CHECK-NEXT:    [[IDX_NEXT]] = add i32 [[IDX]], 1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nsw i32 [[IDX]], 1
 ; CHECK-NEXT:    [[ABC:%.*]] = icmp slt i32 [[IDX]], [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 true, label [[IN_BOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT1:%.*]]
 ; CHECK:       in.bounds:
@@ -94,7 +94,7 @@ define void @test_02(ptr %arr, ptr %a_len_ptr) {
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IDX:%.*]] = phi i32 [ [[IDX_PRELOOP_COPY:%.*]], [[MAINLOOP:%.*]] ], [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ]
-; CHECK-NEXT:    [[IDX_NEXT]] = add i32 [[IDX]], -1
+; CHECK-NEXT:    [[IDX_NEXT]] = add nsw i32 [[IDX]], -1
 ; CHECK-NEXT:    [[ABC:%.*]] = icmp slt i32 [[IDX]], [[LEN]]
 ; CHECK-NEXT:    br i1 true, label [[IN_BOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT1:%.*]]
 ; CHECK:       in.bounds: