[PATCH] D34879: [LoopInterchange] Skip zext instructions when looking for induction var.

[Florian Hahn via Phabricator via llvm-commits]

fhahn created this revision.
Herald added a subscriber: mzolotukhin.

SimplifyIndVar may introduce zext instructions to widen arguments of the
loop exit check. They should not prevent us from splitting the loop at
the induction variable, but maybe the check should be more conservative,
e.g. making sure it only extends arguments used by a comparison?


https://reviews.llvm.org/D34879

Files:
  lib/Transforms/Scalar/LoopInterchange.cpp
  test/Transforms/LoopInterchange/interchange.ll


Index: test/Transforms/LoopInterchange/interchange.ll
===================================================================
--- test/Transforms/LoopInterchange/interchange.ll
+++ test/Transforms/LoopInterchange/interchange.ll
@@ -747,3 +747,81 @@
 ; CHECK:   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
 ; CHECK:   %exitcond = icmp ne i64 %indvars.iv.next, 3
 ; CHECK:   br i1 %exitcond, label %for.body4, label %for.cond.cleanup
+
+;;-----------------------------------Test case 11-------------------------------
+;; Test to make sure we can handle zext intructions introduced by
+;; IndVarSimplify.
+;;
+;;  for (int i = 0; i < 2; ++i)
+;;    for(int j = 0; j < n; ++j) {
+;;      A[j][i] = i;
+;;    }
+
+ at A11 = local_unnamed_addr global [3 x [3 x i32]] zeroinitializer, align 16
+
+define void @interchange_11(i32 %n) {
+entry:
+  br label %for.cond1.preheader
+
+for.cond.loopexit:                                ; preds = %for.body4
+  %exitcond28 = icmp ne i64 %indvars.iv.next27, 2
+  br i1 %exitcond28, label %for.cond1.preheader, label %for.cond.cleanup
+
+for.cond1.preheader:                              ; preds = %for.cond.loopexit, %entry
+  %indvars.iv26 = phi i64 [ 0, %entry ], [ %indvars.iv.next27, %for.cond.loopexit ]
+  %indvars.iv.next27 = add nuw nsw i64 %indvars.iv26, 1
+  br label %for.body4
+
+for.cond.cleanup:                                 ; preds = %for.cond.loopexit
+  ret void
+
+for.body4:                                        ; preds = %for.body4, %for.cond1.preheader
+  %indvars.iv = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next, %for.body4 ]
+  %arrayidx6 = getelementptr inbounds [3 x [3 x i32]], [3 x [3 x i32]]* @A10, i64 0, i64 %indvars.iv, i64 %indvars.iv26
+  %tmp = trunc i64 %indvars.iv26 to i32
+  store i32 %tmp, i32* %arrayidx6, align 4
+  %arrayidx10 = getelementptr inbounds [3 x [3 x i32]], [3 x [3 x i32]]* @A10, i64 0, i64 %indvars.iv, i64 %indvars.iv.next27
+  %tmp1 = trunc i64 %indvars.iv to i32
+  store i32 %tmp1, i32* %arrayidx10, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %n.wide = zext i32 %n to i64
+  %exitcond = icmp ne i64 %indvars.iv.next, %n.wide
+  br i1 %exitcond, label %for.body4, label %for.cond.loopexit
+}
+
+; CHECK-LABEL: @interchange_11
+; CHECK: entry:
+; CHECK:   br label %for.body4.preheader
+
+; CHECK: for.cond1.preheader.preheader:
+; CHECK:   br label %for.cond1.preheader
+
+; CHECK: for.cond.loopexit:
+; CHECK:   %exitcond28 = icmp ne i64 %indvars.iv.next27, 2
+; CHECK:   br i1 %exitcond28, label %for.cond1.preheader, label %for.body4.split
+
+; CHECK: for.cond1.preheader:
+; CHECK:   %indvars.iv26 = phi i64 [ %indvars.iv.next27, %for.cond.loopexit ], [ 0, %for.cond1.preheader.preheader ]
+; CHECK:   %indvars.iv.next27 = add nuw nsw i64 %indvars.iv26, 1
+; CHECK:   br label %for.body4.split1
+
+; CHECK: for.body4.preheader:
+; CHECK:   br label %for.body4
+
+; CHECK: for.cond.cleanup:
+; CHECK:   ret void
+
+; CHECK: for.body4:
+; CHECK:   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body4.split ], [ 0, %for.body4.preheader ]
+; CHECK:   br label %for.cond1.preheader.preheader
+
+; CHECK: for.body4.split1:
+; CHECK:   %arrayidx6 = getelementptr inbounds [3 x [3 x i32]], [3 x [3 x i32]]* @A10, i64 0, i64 %indvars.iv, i64 %indvars.iv26
+; CHECK:   %tmp = trunc i64 %indvars.iv26 to i32
+; CHECK:   store i32 %tmp, i32* %arrayidx6, align 4
+; CHECK:   br label %for.cond.loopexit
+
+; CHECK: for.body4.split:
+; CHECK:   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+; CHECK:   %exitcond = icmp ne i64 %indvars.iv.next, %n.wide
+; CHECK:   br i1 %exitcond, label %for.body4, label %for.cond.cleanup
Index: lib/Transforms/Scalar/LoopInterchange.cpp
===================================================================
--- lib/Transforms/Scalar/LoopInterchange.cpp
+++ lib/Transforms/Scalar/LoopInterchange.cpp
@@ -825,7 +825,8 @@
 
   bool FoundInduction = false;
   for (const Instruction &I : reverse(*InnerLoopLatch)) {
-    if (isa<BranchInst>(I) || isa<CmpInst>(I) || isa<TruncInst>(I))
+    if (isa<BranchInst>(I) || isa<CmpInst>(I) || isa<TruncInst>(I) ||
+        isa<ZExtInst>(I))
       continue;
     // We found an instruction. If this is not induction variable then it is not
     // safe to split this loop latch.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D34879.104851.patch
Type: text/x-patch
Size: 4277 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170630/1a6437f3/attachment.bin>