Do not assume the value passed to memset is an i32.

[jobnoorman]

Summary

The code in SelectionDAG::getMemset for some reason assumes the value passed to memset is an i32. This breaks the generated code for targets that only have registers smaller than 32 bits because the value might get split into multiple registers by the calling convention. See the test for the MSP430 target included in the patch for an example.

This patch ensures that nothing is assumed about the type of the value. Instead, the type is taken from the selected overload of the llvm.memset intrinsic.

[resistor]

LGTM.

—Owen

[jobnoorman]

Thanks for the reviews! r216716

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

SDValue SelectionDAG::getMemset(SDValue Chain, SDLoc dl, SDValue Dst,
 
   // Emit a library call.
   const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering();
   Type *IntPtrTy = TLI->getDataLayout()->getIntPtrType(*getContext());
   TargetLowering::ArgListTy Args;
   TargetLowering::ArgListEntry Entry;
   Entry.Node = Dst; Entry.Ty = IntPtrTy;
   Args.push_back(Entry);
-  // Extend or truncate the argument to be an i32 value for the call.
-  if (Src.getValueType().bitsGT(MVT::i32))
-    Src = getNode(ISD::TRUNCATE, dl, MVT::i32, Src);
-  else
-    Src = getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Src);
   Entry.Node = Src;
-  Entry.Ty = Type::getInt32Ty(*getContext());
-  Entry.isSExt = true;
+  Entry.Ty = Src.getValueType().getTypeForEVT(*getContext());
   Args.push_back(Entry);
   Entry.Node = Size;
   Entry.Ty = IntPtrTy;
-  Entry.isSExt = false;
   Args.push_back(Entry);
 
   // FIXME: pass in SDLoc
   TargetLowering::CallLoweringInfo CLI(*this);
   CLI.setDebugLoc(dl).setChain(Chain)
     .setCallee(TLI->getLibcallCallingConv(RTLIB::MEMSET),
                Type::getVoidTy(*getContext()),
                getExternalSymbol(TLI->getLibcallName(RTLIB::MEMSET),

test/CodeGen/MSP430/memset.ll

+; RUN: llc < %s | FileCheck %s
+target datalayout = "e-m:e-p:16:16-i32:16:32-a:16-n8:16"
+target triple = "msp430---elf"
+
+@buf = external global i8*
+
+; Function Attrs: nounwind
+define void @test() nounwind {
+entry:
+; CHECK-LABEL: test:
+  %0 = load i8** @buf, align 2
+; CHECK: mov.w &buf, r15
+; CHECK-NEXT: mov.w #5, r14
+; CHECK-NEXT: mov.w #128, r13
+; CHECK-NEXT: call #memset
+  call void @llvm.memset.p0i8.i16(i8* %0, i8 5, i16 128, i32 1, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind
+declare void @llvm.memset.p0i8.i16(i8* nocapture, i8, i16, i32, i1) nounwind
+