[RISCV] Guard against out of bound shifts in expandMul.

Spotted while reviewing llvm#150211. If we're multiplying by -3 in i32
MulAmt contains 4,294,967,293 since we zero extend to uint64_t.
Adding 3 to this gives 0x100000000 which is a power of 2 and the log2
of that is 32, but we can't shift left by 32 in an i32.

We should use 0 instead of the shl in this case.

Normally we don't hit this case because decomeMulByConstant handles
it, but that's disabled by Qciac. And after llvm#150211 the path in
expandMul will also be unreachable. So I didn't add a test to avoid
messing with that review.

Author: topperc

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -16184,10 +16184,15 @@ static SDValue expandMul(SDNode *N, SelectionDAG &DAG,
     // 2^N - 3/5/9 --> (sub (shl X, C1), (shXadd X, x))
     for (uint64_t Offset : {3, 5, 9}) {
       if (isPowerOf2_64(MulAmt + Offset)) {
+        unsigned ShAmt = Log2_64(MulAmt + Offset);
         SDLoc DL(N);
-        SDValue Shift1 =
+        SDValue Shift1;
+        if (ShAmt >= VT.getSizeInBits())
+          Shift1 = DAG.getConstant(0, DL, VT);
+        else
+          Shift1 =
             DAG.getNode(ISD::SHL, DL, VT, X,
-                        DAG.getConstant(Log2_64(MulAmt + Offset), DL, VT));
+                        DAG.getConstant(ShAmt, DL, VT));
         SDValue Mul359 =
             DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
                         DAG.getConstant(Log2_64(Offset - 1), DL, VT), X);