[X86] Guard against leaving a dangling node in combineTruncateWithSat.

When handling the packus pattern for i32->i8 we do a two step
process using a packss to i16 followed by a packus to i8. If the
final i8 step is a type with less than 64-bits the packus step
will return SDValue(), but the i32->i16 step might have succeeded.
This leaves the nodes from the middle step dangling.

Guard against this by pre-checking that the number of elements is
at least 8 before doing the middle step.

With that check in place this should mean the only other
case the middle step itself can fail is when SSE2 is disabled. So
add an early SSE2 check then just assert that neither the middle
or final step ever fail.

llvm-svn: 374460

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 2033825..b9cb85e 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -39830,6 +39830,9 @@ static SDValue detectSSatPattern(SDValue In, EVT VT, bool MatchPackUS = false) {
 static SDValue combineTruncateWithSat(SDValue In, EVT VT, const SDLoc &DL,
                                       SelectionDAG &DAG,
                                       const X86Subtarget &Subtarget) {
+  if (!Subtarget.hasSSE2())
+    return SDValue();
+
   EVT SVT = VT.getScalarType();
   EVT InVT = In.getValueType();
   EVT InSVT = InVT.getScalarType();
@@ -39852,6 +39855,7 @@ static SDValue combineTruncateWithSat(SDValue In, EVT VT, const SDLoc &DL,
       // Emit a VPACKUSDW+VPERMQ followed by a VPMOVUSWB.
       SDValue Mid = truncateVectorWithPACK(X86ISD::PACKUS, MVT::v16i16, USatVal,
                                            DL, DAG, Subtarget);
+      assert(Mid && "Failed to pack!");
       return DAG.getNode(X86ISD::VTRUNCUS, DL, VT, Mid);
     }
   }
@@ -39863,14 +39867,19 @@ static SDValue combineTruncateWithSat(SDValue In, EVT VT, const SDLoc &DL,
       (InSVT == MVT::i16 || InSVT == MVT::i32)) {
     if (auto USatVal = detectSSatPattern(In, VT, true)) {
       // vXi32 -> vXi8 must be performed as PACKUSWB(PACKSSDW,PACKSSDW).
-      if (SVT == MVT::i8 && InSVT == MVT::i32) {
+      // Only do this when the result is at least 64 bits or we'll leaving
+      // dangling PACKSSDW nodes.
+      if (SVT == MVT::i8 && InSVT == MVT::i32 &&
+          VT.getVectorNumElements() >= 8) {
         EVT MidVT = EVT::getVectorVT(*DAG.getContext(), MVT::i16,
                                      VT.getVectorNumElements());
         SDValue Mid = truncateVectorWithPACK(X86ISD::PACKSS, MidVT, USatVal, DL,
                                              DAG, Subtarget);
-        if (Mid)
-          return truncateVectorWithPACK(X86ISD::PACKUS, VT, Mid, DL, DAG,
-                                        Subtarget);
+        assert(Mid && "Failed to pack!");
+        SDValue V = truncateVectorWithPACK(X86ISD::PACKUS, VT, Mid, DL, DAG,
+                                           Subtarget);
+        assert(V && "Failed to pack!");
+        return V;
       } else if (SVT == MVT::i8 || Subtarget.hasSSE41())
         return truncateVectorWithPACK(X86ISD::PACKUS, VT, USatVal, DL, DAG,
                                       Subtarget);