[PPC] Avoid non-simple MVT in STBRX optimization

[Carrot]

Summary

PR35402 triggered this case. It bswap and stores a 48bit value, current STBRX optimization transforms it into STBRX. Unfortunately 48bit is not a simple MVT, there is no PPC instruction to support it, and it can't be automatically expanded by llvm, so caused a crash.

This patch detects the non-simple MVT and returns early.

[echristo]

LGTM.

Thanks!

lib/Target/PowerPC/PPCISelLowering.cpp

case ISD::STORE: {
     // Turn STORE (BSWAP) -> sthbrx/stwbrx.
     if (cast<StoreSDNode>(N)->isUnindexed() &&
         N->getOperand(1).getOpcode() == ISD::BSWAP &&
         N->getOperand(1).getNode()->hasOneUse() &&
         (N->getOperand(1).getValueType() == MVT::i32 ||
          N->getOperand(1).getValueType() == MVT::i16 ||
          (Subtarget.hasLDBRX() && Subtarget.isPPC64() &&
           N->getOperand(1).getValueType() == MVT::i64))) {
+      // STBRX can only handle simple types.
+      EVT mVT = cast<StoreSDNode>(N)->getMemoryVT();
+      if (mVT.isExtended())
+        break;
+
       SDValue BSwapOp = N->getOperand(1).getOperand(0);
       // Do an any-extend to 32-bits if this is a half-word input.
       if (BSwapOp.getValueType() == MVT::i16)
         BSwapOp = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i32, BSwapOp);
 
       // If the type of BSWAP operand is wider than stored memory width
       // it need to be shifted to the right side before STBRX.
-      EVT mVT = cast<StoreSDNode>(N)->getMemoryVT();
       if (Op1VT.bitsGT(mVT)) {
         int Shift = Op1VT.getSizeInBits() - mVT.getSizeInBits();
         BSwapOp = DAG.getNode(ISD::SRL, dl, Op1VT, BSwapOp,
                               DAG.getConstant(Shift, dl, MVT::i32));
         // Need to truncate if this is a bswap of i64 stored as i32/i16.
         if (Op1VT == MVT::i64)
           BSwapOp = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, BSwapOp);
       }

test/CodeGen/PowerPC/pr35402.ll

+; RUN: llc -O2 < %s | FileCheck %s
+target triple = "powerpc64le-linux-gnu"
+
+define void @test(i8* %p, i64 %data) {
+entry:
+  %0 = tail call i64 @llvm.bswap.i64(i64 %data)
+  %ptr = bitcast i8* %p to i48*
+  %val = trunc i64 %0 to i48
+  store i48 %val, i48* %ptr, align 1
+  ret void
+
+; CHECK:     sth
+; CHECK:     stw
+; CHECK-NOT: stdbrx
+
+}
+
+declare i64 @llvm.bswap.i64(i64)