From 5a637cbdc7d8a59d5d56f329871cb3834c1f5095 Mon Sep 17 00:00:00 2001
From: Philip Reames <listmail@philipreames.com>
Date: Tue, 9 Jul 2019 01:27:45 +0000
Subject: [PATCH] [LoopPred] Extend LFTR normalization to the inverse EQ case

A while back, I added support for NE latches formed by LFTR.  I didn't think that quite through, as LFTR will also produce the inverse EQ form for some loops and I hadn't handled that.  This change just adds handling for that case as well.

llvm-svn: 365419
---
 .../lib/Transforms/Scalar/LoopPredication.cpp |  5 +++
 llvm/test/Transforms/LoopPredication/basic.ll | 43 +++++++++++++++++++
 2 files changed, 48 insertions(+)

diff --git a/llvm/lib/Transforms/Scalar/LoopPredication.cpp b/llvm/lib/Transforms/Scalar/LoopPredication.cpp
index d3cec7568b6a..4ee3d0079c78 100644
--- a/llvm/lib/Transforms/Scalar/LoopPredication.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopPredication.cpp
@@ -653,6 +653,11 @@ static void normalizePredicate(ScalarEvolution *SE, Loop *L,
       RC.IV->getStepRecurrence(*SE)->isOne() &&
       SE->isKnownPredicate(ICmpInst::ICMP_ULE, RC.IV->getStart(), RC.Limit))
     RC.Pred = ICmpInst::ICMP_ULT;
+  if (RC.Pred == ICmpInst::ICMP_EQ &&
+      RC.IV->getStepRecurrence(*SE)->isOne() &&
+      SE->isKnownPredicate(ICmpInst::ICMP_ULE, RC.IV->getStart(), RC.Limit))
+    RC.Pred = ICmpInst::ICMP_UGE;
+
 }
 
 
diff --git a/llvm/test/Transforms/LoopPredication/basic.ll b/llvm/test/Transforms/LoopPredication/basic.ll
index 88fa1bb95b2a..4a9cf6713193 100644
--- a/llvm/test/Transforms/LoopPredication/basic.ll
+++ b/llvm/test/Transforms/LoopPredication/basic.ll
@@ -1756,6 +1756,49 @@ exit:
   ret i32 0
 }
 
+; Same as previous, except swapped br/cmp
+define i32 @eq_latch_dom_check_preinc(i32* %array, i32 %length, i32 %n) {
+; CHECK-LABEL: @eq_latch_dom_check_preinc(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp sle i32 [[N:%.*]], 0
+; CHECK-NEXT:    br i1 [[TMP5]], label [[EXIT:%.*]], label [[LOOP_PREHEADER:%.*]]
+; CHECK:       loop.preheader:
+; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[LENGTH:%.*]], -1
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ule i32 [[N]], [[TMP0]]
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ult i32 0, [[LENGTH]]
+; CHECK-NEXT:    [[TMP3:%.*]] = and i1 [[TMP2]], [[TMP1]]
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[I_NEXT:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[TMP3]], i32 9) [ "deopt"() ]
+; CHECK-NEXT:    [[I_NEXT]] = add nuw i32 [[I]], 1
+; CHECK-NEXT:    [[DONE:%.*]] = icmp eq i32 [[I]], [[N]]
+; CHECK-NEXT:    br i1 [[DONE]], label [[EXIT_LOOPEXIT:%.*]], label [[LOOP]]
+; CHECK:       exit.loopexit:
+; CHECK-NEXT:    br label [[EXIT]]
+; CHECK:       exit:
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %tmp5 = icmp sle i32 %n, 0
+  br i1 %tmp5, label %exit, label %loop.preheader
+
+loop.preheader:
+  br label %loop
+
+loop:
+  %i = phi i32 [ %i.next, %loop ], [ 0, %loop.preheader ]
+  %within.bounds = icmp ult i32 %i, %length
+  call void (i1, ...) @llvm.experimental.guard(i1 %within.bounds, i32 9) [ "deopt"() ]
+
+  %i.next = add nuw i32 %i, 1
+  %done = icmp eq i32 %i, %n
+  br i1 %done, label %exit, label %loop
+
+exit:
+  ret i32 0
+}
+
 
 ; NE latch - can't prove (end-start) mod step == 0 (i.e. might wrap
 ; around several times or even be infinite)