[LV] Fix an issue where forming LCSSA in the place that we did would

change the set of uniform instructions in the loop causing an assert
failure.

The problem is that the legalization checking also builds data
structures mapping various facts about the loop body. The immediate
cause was the set of uniform instructions. If these then change when
LCSSA is formed, the data structures would already have been built and
become stale. The included test case triggered an assert in loop
vectorize that was reduced out of the new PM's pipeline.

The solution is to form LCSSA early enough that no information is cached
across the changes made. The only really obvious position is outside of
the main logic to vectorize the loop. This also has the advantage of
removing one case where forming LCSSA could mutate the loop but we
wouldn't track that as a "Changed" state.

If it is significantly advantageous to do some legalization checking
prior to this, we can do a more careful positioning but it seemed best
to just back off to a safe position first.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293168 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index ce3f3b0..4c2de93 100644 (file)
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7522,8 +7522,6 @@ bool LoopVectorizePass::processLoop(Loop *L) {
     DEBUG(dbgs() << "LV: Interleave Count is " << IC << '\n');
   }
 
-  formLCSSARecursively(*L, *DT, LI, SE);
-
   using namespace ore;
   if (!VectorizeLoop) {
     assert(IC > 1 && "interleave count should not be 1 or 0");
@@ -7620,8 +7618,15 @@ bool LoopVectorizePass::runImpl(
   LoopsAnalyzed += Worklist.size();
 
   // Now walk the identified inner loops.
-  while (!Worklist.empty())
-    Changed |= processLoop(Worklist.pop_back_val());
+  while (!Worklist.empty()) {
+    Loop *L = Worklist.pop_back_val();
+
+    // For the inner loops we actually process, form LCSSA to simplify the
+    // transform.
+    Changed |= formLCSSARecursively(*L, *DT, LI, SE);
+
+    Changed |= processLoop(L);
+  }
 
   // Process each loop nest in the function.
   return Changed;

diff --git a/test/Transforms/LoopVectorize/lcssa-crash.ll b/test/Transforms/LoopVectorize/lcssa-crash.ll
index e6bd6ed..3d3ef9e 100644 (file)
--- a/test/Transforms/LoopVectorize/lcssa-crash.ll
+++ b/test/Transforms/LoopVectorize/lcssa-crash.ll
@@ -37,3 +37,26 @@ L0:
 L1:
   ret void
 }
+
+; This loop has different uniform instructions before and after LCSSA.
+define void @test3() {
+entry:
+  %add41 = add i32 undef, undef
+  %idxprom4736 = zext i32 %add41 to i64
+  br label %while.body
+
+while.body:
+  %idxprom4738 = phi i64 [ %idxprom47, %while.body ], [ %idxprom4736, %entry ]
+  %pos.337 = phi i32 [ %inc46, %while.body ], [ %add41, %entry ]
+  %inc46 = add i32 %pos.337, 1
+  %arrayidx48 = getelementptr inbounds [1024 x i8], [1024 x i8]* undef, i64 0, i64 %idxprom4738
+  store i8 0, i8* %arrayidx48, align 1
+  %and43 = and i32 %inc46, 3
+  %cmp44 = icmp eq i32 %and43, 0
+  %idxprom47 = zext i32 %inc46 to i64
+  br i1 %cmp44, label %while.end, label %while.body
+
+while.end:
+  %add58 = add i32 %inc46, 4
+  ret void
+}