[UNROLL] Postpone ScalarEvolution::forgetLoop after TripCountSC is expanded

when unroll runtime iteration loop.

In llvm::UnrollRuntimeLoopRemainder, if the loop to be unrolled is the inner
loop inside a loop nest, the scalar evolution needs to be dropped for its
parent loop which is done by ScalarEvolution::forgetLoop. However, we can
postpone forgetLoop to the end of UnrollRuntimeLoopRemainder so TripCountSC
expansion can still reuse existing value.

Differential Revision: https://reviews.llvm.org/D23572

llvm-svn: 279748

diff --git a/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp b/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp
index 16197a9..5758a41 100644 (file)
--- a/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp
@@ -491,11 +491,6 @@ bool llvm::UnrollRuntimeLoopRemainder(Loop *L, unsigned Count,
   if (Log2_32(Count) > BEWidth)
     return false;
 
-  // If this loop is nested, then the loop unroller changes the code in the
-  // parent loop, so the Scalar Evolution pass needs to be run again.
-  if (Loop *ParentLoop = L->getParentLoop())
-    SE->forgetLoop(ParentLoop);
-
   BasicBlock *Latch = L->getLoopLatch();
 
   // Loop structure is the following:
@@ -685,6 +680,12 @@ bool llvm::UnrollRuntimeLoopRemainder(Loop *L, unsigned Count,
     ConnectProlog(L, BECount, Count, PrologExit, PreHeader, NewPreHeader,
                   VMap, DT, LI, PreserveLCSSA);
   }
+
+  // If this loop is nested, then the loop unroller changes the code in the
+  // parent loop, so the Scalar Evolution pass needs to be run again.
+  if (Loop *ParentLoop = L->getParentLoop())
+    SE->forgetLoop(ParentLoop);
+
   NumRuntimeUnrolled++;
   return true;
 }

diff --git a/llvm/test/Analysis/ScalarEvolution/scev-expander-reuse-unroll.ll b/llvm/test/Analysis/ScalarEvolution/scev-expander-reuse-unroll.ll
new file mode 100644 (file)
index 0000000..e74b004
--- /dev/null
+++ b/llvm/test/Analysis/ScalarEvolution/scev-expander-reuse-unroll.ll
@@ -0,0 +1,35 @@
+; RUN: opt < %s -loop-unroll -unroll-runtime -unroll-count=2 -verify-scev-maps -S | FileCheck %s
+
+; Check SCEV expansion uses existing value when unrolling an inner loop with runtime trip count in a loop nest.
+; CHECK-LABEL: @foo(
+; CHECK: select
+; CHECK-NOT: select
+; CHECK: ret
+
+define void @foo(i32 %xfL, i32 %scaleL) local_unnamed_addr {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body5, %for.body, %entry
+  %xfL.addr.033 = phi i32 [ %xfL, %entry ], [ %add, %for.body5 ]
+  %add = add nsw i32 %xfL.addr.033, %scaleL
+  %shr = ashr i32 %add, 16
+  %cmp.i = icmp slt i32 0, %shr
+  %.sroa.speculated = select i1 %cmp.i, i32 0, i32 %shr
+  %cmp425 = icmp slt i32 0, %.sroa.speculated
+  br i1 %cmp425, label %for.body5.preheader, label %for.end
+
+for.body5.preheader:                              ; preds = %for.body
+  %tmp0 = sext i32 %.sroa.speculated to i64
+  br label %for.body5
+
+for.body5:                                        ; preds = %for.body5, %for.body5.preheader
+  %indvars.iv = phi i64 [ 0, %for.body5.preheader ], [ %indvars.iv.next, %for.body5 ]
+  %indvars.iv.next = add nsw i64 %indvars.iv, 1
+  %cmp4 = icmp slt i64 %indvars.iv.next, %tmp0
+  br i1 %cmp4, label %for.body5, label %for.body
+
+for.end:
+  ret void
+}
+

diff --git a/llvm/test/Analysis/ScalarEvolution/scev-expander-existing-value.ll b/llvm/test/Analysis/ScalarEvolution/scev-expander-reuse-vect.ll
similarity index 97%
rename from llvm/test/Analysis/ScalarEvolution/scev-expander-existing-value.ll
rename to llvm/test/Analysis/ScalarEvolution/scev-expander-reuse-vect.ll
index 0c1d676..a818b47 100644 (file)
--- a/llvm/test/Analysis/ScalarEvolution/scev-expander-existing-value.ll
+++ b/llvm/test/Analysis/ScalarEvolution/scev-expander-reuse-vect.ll
@@ -1,8 +1,10 @@
 ; RUN: opt < %s -loop-vectorize -force-vector-width=4 -verify-scev-maps -S |FileCheck %s
 
 ; SCEV expansion uses existing value when the SCEV has no AddRec expr.
+; CHECK-LABEL: @foo(
 ; CHECK: select
 ; CHECK-NOT: select
+; CHECK: ret
 
 @a = common global [1000 x i16] zeroinitializer, align 16