[Hexagon] Handle long-running simplification loop in idiom recognition

The initial assumption was that the simplification would converge to a
fixed point relatvely quickly. Turns out that there are legitimate situa-
tions where the complexity of the code causes it to take a large number
of iterations.

Two main changes:
- Instead of aborting upon hitting the limit, simply return nullptr.
- Reduce the limit to 10,000 from 100,000.

llvm-svn: 304441

lib/Target/Hexagon/HexagonLoopIdiomRecognition.cpp

@@ -59,6 +59,9 @@ cl::opt<bool> HexagonVolatileMemcpy("disable-hexagon-volatile-memcpy",
   cl::Hidden, cl::init(false),
   cl::desc("Enable Hexagon-specific memcpy for volatile destination."));
 
+static cl::opt<unsigned> SimplifyLimit("hlir-simplify-limit", cl::init(10000),
+  cl::Hidden, cl::desc("Maximum number of simplification steps in HLIR"));
+
 static const char *HexagonVolatileMemcpyName
   = "hexagon_memcpy_forward_vp4cp4n2";
 
@@ -477,7 +480,7 @@ Value *Simplifier::simplify(Context &C) {
   WorkListType Q;
   Q.push_back(C.Root);
   unsigned Count = 0;
-  const unsigned Limit = 100000;
+  const unsigned Limit = SimplifyLimit;
 
   while (!Q.empty()) {
     if (Count++ >= Limit)
@@ -501,8 +504,7 @@ Value *Simplifier::simplify(Context &C) {
         Q.push_back(Op);
     }
   }
-  assert(Count < Limit && "Infinite loop in HLIR/simplify?");
+  return Count < Limit ? C.Root : nullptr;
-  return C.Root;
 }
 
 

test/CodeGen/Hexagon/loop-idiom/pmpy-long-loop.ll

@@ -0,0 +1,62 @@
+; RUN: opt -march=hexagon -hexagon-loop-idiom -S < %s | FileCheck %s
+;
+; The number of nested selects caused the simplification loop to take
+; more than the maximum number of iterations. This caused the compiler
+; to crash under suspicion of an infinite loop. This (still reduced)
+; testcase shows a legitimate case where this limit was exceeded.
+; Instead of crashing, gracefully abort the simplification.
+;
+; Check for sane output.
+; CHECK: define void @fred
+
+target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+target triple = "hexagon"
+
+define void @fred() unnamed_addr #0 {
+b0:
+  %v1 = select i1 false, i32 undef, i32 2
+  br label %b2
+
+b2:                                               ; preds = %b2, %b0
+  %v3 = sext i16 undef to i32
+  %v4 = add nsw i32 %v1, %v3
+  %v5 = select i1 undef, i32 undef, i32 %v4
+  %v6 = icmp slt i32 %v5, undef
+  %v7 = select i1 %v6, i32 %v5, i32 undef
+  %v8 = icmp slt i32 %v7, 0
+  %v9 = select i1 %v8, i32 %v7, i32 0
+  %v10 = sub i32 undef, undef
+  %v11 = add i32 %v10, %v9
+  %v12 = sext i16 undef to i32
+  %v13 = sext i16 undef to i32
+  %v14 = add nsw i32 %v1, %v13
+  %v15 = select i1 undef, i32 undef, i32 %v14
+  %v16 = icmp slt i32 %v15, undef
+  %v17 = select i1 %v16, i32 %v15, i32 undef
+  %v18 = select i1 undef, i32 %v17, i32 %v12
+  %v19 = add i32 undef, %v18
+  %v20 = sext i16 undef to i32
+  %v21 = sext i16 0 to i32
+  %v22 = add nsw i32 %v1, %v21
+  %v23 = sext i16 undef to i32
+  %v24 = add nsw i32 %v1, %v23
+  %v25 = select i1 undef, i32 undef, i32 %v24
+  %v26 = icmp slt i32 %v25, %v22
+  %v27 = select i1 %v26, i32 %v25, i32 %v22
+  %v28 = icmp slt i32 %v27, %v20
+  %v29 = select i1 %v28, i32 %v27, i32 %v20
+  %v30 = add i32 undef, %v29
+  %v31 = add i32 %v11, undef
+  %v32 = add i32 %v31, undef
+  %v33 = add i32 %v32, %v19
+  %v34 = add i32 %v33, %v30
+  %v35 = add nsw i32 %v34, 32768
+  %v36 = icmp ult i32 %v35, 65536
+  %v37 = select i1 %v36, i32 %v34, i32 undef
+  br i1 undef, label %b2, label %b38
+
+b38:                                              ; preds = %b2
+  unreachable
+}
+
+attributes #0 = { "target-cpu"="hexagonv60" }