Currently, codegen may spent some time in SDISel passes even if an entire
function is successfully handled by fast-isel. That's because function
arguments are *always* handled by SDISel. Introduce FastLowerArguments to
allow each target to provide hook to handle formal argument lowering.
As a proof-of-concept, add ARMFastIsel::FastLowerArguments to handle
functions with 4 or fewer scalar integer (i8, i16, or i32) arguments. It
completely eliminates the need for SDISel for trivial functions.
rdar://13163905
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@174855 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/ARM/ARMFastISel.cpp b/lib/Target/ARM/ARMFastISel.cpp
index 4b7978a..022fe97 100644
--- a/lib/Target/ARM/ARMFastISel.cpp
+++ b/lib/Target/ARM/ARMFastISel.cpp
@@ -146,6 +146,7 @@
virtual unsigned TargetMaterializeAlloca(const AllocaInst *AI);
virtual bool TryToFoldLoad(MachineInstr *MI, unsigned OpNo,
const LoadInst *LI);
+ virtual bool FastLowerArguments();
private:
#include "ARMGenFastISel.inc"
@@ -2884,6 +2885,79 @@
return DestReg2;
}
+bool ARMFastISel::FastLowerArguments() {
+ if (!FuncInfo.CanLowerReturn)
+ return false;
+
+ const Function *F = FuncInfo.Fn;
+ if (F->isVarArg())
+ return false;
+
+ CallingConv::ID CC = F->getCallingConv();
+ switch (CC) {
+ default:
+ return false;
+ case CallingConv::Fast:
+ case CallingConv::C:
+ case CallingConv::ARM_AAPCS_VFP:
+ case CallingConv::ARM_AAPCS:
+ case CallingConv::ARM_APCS:
+ break;
+ }
+
+ // Only handle simple cases. i.e. Up to 4 i8/i16/i32 scalar arguments
+ // which are passed in r0 - r3.
+ unsigned Idx = 1;
+ for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
+ I != E; ++I, ++Idx) {
+ if (Idx > 4)
+ return false;
+
+ if (F->getAttributes().hasAttribute(Idx, Attribute::InReg) ||
+ F->getAttributes().hasAttribute(Idx, Attribute::StructRet) ||
+ F->getAttributes().hasAttribute(Idx, Attribute::ByVal))
+ return false;
+
+ Type *ArgTy = I->getType();
+ if (ArgTy->isStructTy() || ArgTy->isArrayTy() || ArgTy->isVectorTy())
+ return false;
+
+ EVT ArgVT = TLI.getValueType(ArgTy);
+ switch (ArgVT.getSimpleVT().SimpleTy) {
+ case MVT::i8:
+ case MVT::i16:
+ case MVT::i32:
+ break;
+ default:
+ return false;
+ }
+ }
+
+
+ static const uint16_t GPRArgRegs[] = {
+ ARM::R0, ARM::R1, ARM::R2, ARM::R3
+ };
+
+ const TargetRegisterClass *RC = TLI.getRegClassFor(MVT::i32);
+ Idx = 0;
+ for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
+ I != E; ++I, ++Idx) {
+ if (I->use_empty())
+ continue;
+ unsigned SrcReg = GPRArgRegs[Idx];
+ unsigned DstReg = FuncInfo.MF->addLiveIn(SrcReg, RC);
+ // FIXME: Unfortunately it's necessary to emit a copy from the livein copy.
+ // Without this, EmitLiveInCopies may eliminate the livein if its only
+ // use is a bitcast (which isn't turned into an instruction).
+ unsigned ResultReg = createResultReg(RC);
+ BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+ ResultReg).addReg(DstReg, getKillRegState(true));
+ UpdateValueMap(I, ResultReg);
+ }
+
+ return true;
+}
+
namespace llvm {
FastISel *ARM::createFastISel(FunctionLoweringInfo &funcInfo,
const TargetLibraryInfo *libInfo) {