[llvm] r357408 - [LoopPred] Be uniform about proving generated conditions

[Philip Reames via llvm-commits]

Author: reames
Date: Mon Apr  1 09:26:08 2019
New Revision: 357408

URL: http://llvm.org/viewvc/llvm-project?rev=357408&view=rev
Log:
[LoopPred] Be uniform about proving generated conditions

We'd been optimizing the case where the predicate was obviously true, do the same for the false case.  Mostly just for completeness sake, but also may improve compile time in loops which will exit through the guard.  Such loops are presumed rare in fastpath code, but may be present down untaken paths, so optimizing for them is still useful.


Modified:
    llvm/trunk/lib/Transforms/Scalar/LoopPredication.cpp
    llvm/trunk/test/Transforms/LoopPredication/basic.ll

Modified: llvm/trunk/lib/Transforms/Scalar/LoopPredication.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopPredication.cpp?rev=357408&r1=357407&r2=357408&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/LoopPredication.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/LoopPredication.cpp Mon Apr  1 09:26:08 2019
@@ -396,6 +396,9 @@ Value *LoopPredication::expandCheck(SCEV
 
   if (SE->isLoopEntryGuardedByCond(L, Pred, LHS, RHS))
     return Builder.getTrue();
+  if (SE->isLoopEntryGuardedByCond(L, ICmpInst::getInversePredicate(Pred),
+                                   LHS, RHS))
+    return Builder.getFalse();
 
   Instruction *InsertAt = &*Builder.GetInsertPoint();
   Value *LHSV = Expander.expandCodeFor(LHS, Ty, InsertAt);

Modified: llvm/trunk/test/Transforms/LoopPredication/basic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopPredication/basic.ll?rev=357408&r1=357407&r2=357408&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/LoopPredication/basic.ll (original)
+++ llvm/trunk/test/Transforms/LoopPredication/basic.ll Mon Apr  1 09:26:08 2019
@@ -1544,3 +1544,52 @@ exit:
   %result = phi i32 [ 0, %entry ], [ %loop.acc.next, %loop ]
   ret i32 %result
 }
+
+
+; This is a case where the length information tells us that the guard
+; must trigger on some iteration.
+define i32 @provably_taken(i32* %array, i32* %length.ptr) {
+; CHECK-LABEL: @provably_taken(
+; CHECK-NEXT:  loop.preheader:
+; CHECK-NEXT:    [[LENGTH:%.*]] = load i32, i32* [[LENGTH_PTR:%.*]], !range !1
+; CHECK-NEXT:    [[TMP0:%.*]] = icmp ult i32 0, [[LENGTH]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i1 [[TMP0]], false
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[LOOP_ACC:%.*]] = phi i32 [ [[LOOP_ACC_NEXT:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER:%.*]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[I_NEXT:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[TMP1]], i32 9) [ "deopt"() ]
+; CHECK-NEXT:    [[I_I64:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT:    [[ARRAY_I_PTR:%.*]] = getelementptr inbounds i32, i32* [[ARRAY:%.*]], i64 [[I_I64]]
+; CHECK-NEXT:    [[ARRAY_I:%.*]] = load i32, i32* [[ARRAY_I_PTR]], align 4
+; CHECK-NEXT:    [[LOOP_ACC_NEXT]] = add i32 [[LOOP_ACC]], [[ARRAY_I]]
+; CHECK-NEXT:    [[I_NEXT]] = add nuw i32 [[I]], 1
+; CHECK-NEXT:    [[CONTINUE:%.*]] = icmp slt i32 [[I_NEXT]], 200
+; CHECK-NEXT:    br i1 [[CONTINUE]], label [[LOOP]], label [[EXIT:%.*]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[RESULT:%.*]] = phi i32 [ [[LOOP_ACC_NEXT]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[RESULT]]
+;
+loop.preheader:
+  %length = load i32, i32* %length.ptr, !range !{i32 0, i32 50}
+  br label %loop
+
+loop:
+  %loop.acc = phi i32 [ %loop.acc.next, %loop ], [ 0, %loop.preheader ]
+  %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.i64 = zext i32 %i to i64
+  %array.i.ptr = getelementptr inbounds i32, i32* %array, i64 %i.i64
+  %array.i = load i32, i32* %array.i.ptr, align 4
+  %loop.acc.next = add i32 %loop.acc, %array.i
+
+  %i.next = add nuw i32 %i, 1
+  %continue = icmp slt i32 %i.next, 200
+  br i1 %continue, label %loop, label %exit
+
+exit:
+  %result = phi i32 [ %loop.acc.next, %loop ]
+  ret i32 %result
+}