[SelectionDAG] Don't pass on DemandedElts when handling SCALAR_TO_VECTOR

[pendingchaos]

Summary

Fixes an assertion:

llc: lib/CodeGen/SelectionDAG/SelectionDAG.cpp:2200: llvm::KnownBits llvm::SelectionDAG::computeKnownBits(llvm::SDValue, const llvm::APInt&, unsigned int) const: Assertion `(!Op.getValueType().isVector() || NumElts == Op.getValueType().getVectorNumElements()) && "Unexpected vector size"' failed.

[RKSimon]

test/CodeGen/AMDGPU/computeKnownBits-scalar-to-vector-crash.ll

1 ↗

(On Diff #176410)

Even though this is a crash test its probably worth adding at least some basic FileCheck CHECKS (or just run update_llc_test_checks)

[pendingchaos]

Makes the test case use FileCheck.

I don't have commit access.

[RKSimon]

LGTM - I'll commit shortly but please make sure you've requested commit access

llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

KnownBits SelectionDAG::computeKnownBits(SDValue Op, const APInt &DemandedElts,
   }
   case ISD::SCALAR_TO_VECTOR: {
     // We know about scalar_to_vector as much as we know about it source,
     // which becomes the first element of otherwise unknown vector.
     if (DemandedElts != 1)
       break;
 
     SDValue N0 = Op.getOperand(0);
-    Known = computeKnownBits(N0, DemandedElts, Depth + 1);
+    Known = computeKnownBits(N0, Depth + 1);
     if (N0.getValueSizeInBits() != BitWidth)
       Known = Known.trunc(BitWidth);
 
     break;
   }
   case ISD::BITCAST: {
     SDValue N0 = Op.getOperand(0);
     EVT SubVT = N0.getValueType();

llvm/trunk/test/CodeGen/AMDGPU/computeKnownBits-scalar-to-vector-crash.ll

+; RUN: llc -march=amdgcn -mcpu=gfx802 -verify-machineinstrs < %s | FileCheck %s
+
+; CHECK: s_waitcnt
+define <2 x i16> @main(<2 x float>) #0 {
+  %2 = bitcast <2 x float> %0 to <4 x i16>
+  %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <2 x i32> <i32 0, i32 undef>
+  %4 = extractelement <4 x i16> %2, i32 0
+  %5 = insertelement <2 x i16> %3, i16 %4, i32 0
+  ret <2 x i16> %5
+}
+