Loop predication expand both sides of the widened condition

This is a fix for a loop predication bug which resulted in malformed IR generation.

Loop invariant side of the widened condition is not guaranteed to be available in the preheader as is, so we need to expand it as well. See added unsigned_loop_0_to_n_hoist_length test for example.

Reviewed By: sanjoy, mkazantsev

Differential Revision: https://reviews.llvm.org/D30099

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296345 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/LoopPredication.cpp b/lib/Transforms/Scalar/LoopPredication.cpp
index fac2b8aecc8f0ea9717bd630b86f53b3984857e5..0ce60442932615cdb6f8ee6190dacf10b9efc9b0 100644 (file)
--- a/lib/Transforms/Scalar/LoopPredication.cpp
+++ b/lib/Transforms/Scalar/LoopPredication.cpp
@@ -141,10 +141,9 @@ Optional<Value *> LoopPredication::widenICmpRangeCheck(ICmpInst *ICI,
     std::swap(LHSS, RHSS);
     Pred = ICmpInst::getSwappedPredicate(Pred);
   }
-  if (!SE->isLoopInvariant(RHSS, L))
+  if (!SE->isLoopInvariant(RHSS, L) || !isSafeToExpand(RHSS, *SE))
     return None;
 
-  Value *Bound = RHS;
   const SCEVAddRecExpr *IndexAR = dyn_cast<SCEVAddRecExpr>(LHSS);
   if (!IndexAR || IndexAR->getLoop() != L)
     return None;
@@ -176,9 +175,12 @@ Optional<Value *> LoopPredication::widenICmpRangeCheck(ICmpInst *ICI,
 
   DEBUG(dbgs() << "NewLHSS is loop invariant and safe to expand. Expand!\n");
 
-  Value *NewLHS = Expander.expandCodeFor(NewLHSS, Bound->getType(),
-                                         Preheader->getTerminator());
-  return Builder.CreateICmp(Pred, NewLHS, Bound);
+  Type *Ty = LHS->getType();
+  Instruction *InsertAt = Preheader->getTerminator();
+  assert(Ty == RHS->getType() && "icmp operands have different types?");
+  Value *NewLHS = Expander.expandCodeFor(NewLHSS, Ty, InsertAt);
+  Value *NewRHS = Expander.expandCodeFor(RHSS, Ty, InsertAt);
+  return Builder.CreateICmp(Pred, NewLHS, NewRHS);
 }
 
 bool LoopPredication::widenGuardConditions(IntrinsicInst *Guard,

diff --git a/test/Transforms/LoopPredication/basic.ll b/test/Transforms/LoopPredication/basic.ll
index a347e6192bd91473db21d575332f76ad8c775ce6..6ce07819cb0398c455da5e0bdcbc19a3c86e8d03 100644 (file)
--- a/test/Transforms/LoopPredication/basic.ll
+++ b/test/Transforms/LoopPredication/basic.ll
@@ -493,4 +493,79 @@ loop:
 exit:
   %result = phi i32 [ 0, %entry ], [ %loop.acc.next, %loop ]
   ret i32 %result
-}
\ No newline at end of file
+}
+
+define i32 @unsigned_loop_0_to_n_hoist_length(i32* %array, i16 %length.i16, i32 %n) {
+; CHECK-LABEL: @unsigned_loop_0_to_n_hoist_length
+entry:
+  %tmp5 = icmp eq i32 %n, 0
+  br i1 %tmp5, label %exit, label %loop.preheader
+
+loop.preheader:
+; CHECK: loop.preheader:
+; CHECK: [[max_index:[^ ]+]] = add i32 %n, -1
+; CHECK-NEXT: [[length:[^ ]+]] = zext i16 %length.i16 to i32
+; CHECK-NEXT: [[wide_cond:[^ ]+]] = icmp ult i32 [[max_index]], [[length]]
+; CHECK-NEXT: br label %loop
+  br label %loop
+
+loop:
+; CHECK: loop:
+; CHECK: call void (i1, ...) @llvm.experimental.guard(i1 [[wide_cond]], i32 9) [ "deopt"() ]
+  %loop.acc = phi i32 [ %loop.acc.next, %loop ], [ 0, %loop.preheader ]
+  %i = phi i32 [ %i.next, %loop ], [ 0, %loop.preheader ]
+  %length = zext i16 %length.i16 to i32
+  %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 ult i32 %i.next, %n
+  br i1 %continue, label %loop, label %exit
+
+exit:
+  %result = phi i32 [ 0, %entry ], [ %loop.acc.next, %loop ]
+  ret i32 %result
+}
+
+define i32 @unsigned_loop_0_to_n_cant_hoist_length(i32* %array, i32 %length, i32 %divider, i32 %n) {
+; CHECK-LABEL: @unsigned_loop_0_to_n_cant_hoist_length
+entry:
+  %tmp5 = icmp eq i32 %n, 0
+  br i1 %tmp5, label %exit, label %loop.preheader
+
+loop.preheader:
+; CHECK: loop.preheader:
+; CHECK-NEXT: br label %loop
+  br label %loop
+
+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: %length.udiv = udiv i32 %length, %divider
+; CHECK-NEXT: %within.bounds = icmp ult i32 %i, %length.udiv
+; CHECK-NEXT: call void (i1, ...) @llvm.experimental.guard(i1 %within.bounds, i32 9) [ "deopt"() ]
+  %loop.acc = phi i32 [ %loop.acc.next, %loop ], [ 0, %loop.preheader ]
+  %i = phi i32 [ %i.next, %loop ], [ 0, %loop.preheader ]
+  %length.udiv = udiv i32 %length, %divider
+  %within.bounds = icmp ult i32 %i, %length.udiv
+  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 ult i32 %i.next, %n
+  br i1 %continue, label %loop, label %exit
+
+exit:
+  %result = phi i32 [ 0, %entry ], [ %loop.acc.next, %loop ]
+  ret i32 %result
+}