Expand the sel pseudo/macro. This generates basic blocks where previously

there were inline br .+4 instructions. Soon everything can enjoy the
full instruction scheduling experience.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175718 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Reed Kotler 2013-02-21 04:22:38 +00:00
parent b8ebc9e244
commit ffbe432595
4 changed files with 100 additions and 4 deletions

View File

@ -76,6 +76,10 @@ class F2RI16_ins<bits<5> _op, string asmstr,
let Constraints = "$rx_ = $rx";
}
class FRI16_B_ins<bits<5> _op, string asmstr,
InstrItinClass itin>:
FRI16<_op, (outs), (ins CPU16Regs:$rx, brtarget:$imm),
!strconcat(asmstr, "\t$rx, $imm # 16 bit inst"), [], itin>;
//
// Compare a register and immediate and place result in CC
// Implicit use of T8
@ -362,6 +366,7 @@ class Sel<string op>:
!strconcat(op, "\t$rt, .+4\n\t\n\tmove $rd, $rs"), []> {
//let isCodeGenOnly=1;
let Constraints = "$rd = $rd_";
let usesCustomInserter = 1;
}
//
@ -518,6 +523,14 @@ def AdduRxRyRz16: FRRR16_ins<01, "addu", IIAlu>, ArithLogic16Defs<1>;
def AndRxRxRy16: FRxRxRy16_ins<0b01100, "and", IIAlu>, ArithLogic16Defs<1>;
//
// Format: BEQZ rx, offset MIPS16e
// Purpose: Branch on Equal to Zero
// To test a GPR then do a PC-relative conditional branch.
//
def BeqzRxImm16: FRI16_B_ins<0b00100, "beqz", IIAlu>, cbranch16;
//
// Format: BEQZ rx, offset MIPS16e
// Purpose: Branch on Equal to Zero (Extended)
@ -531,6 +544,13 @@ def BeqzRxImmX16: FEXT_RI16_B_ins<0b00100, "beqz", IIAlu>, cbranch16;
//
def BimmX16: FEXT_I16_ins<0b00010, "b", IIAlu>, branch16;
//
// Format: BNEZ rx, offset MIPS16e
// Purpose: Branch on Not Equal to Zero
// To test a GPR then do a PC-relative conditional branch.
//
def BnezRxImm16: FRI16_B_ins<0b00101, "bnez", IIAlu>, cbranch16;
//
// Format: BNEZ rx, offset MIPS16e
// Purpose: Branch on Not Equal to Zero (Extended)

View File

@ -55,6 +55,12 @@ Mips16HardFloat("mips16-hard-float", cl::NotHidden,
cl::desc("MIPS: mips16 hard float enable."),
cl::init(false));
static cl::opt<bool> DontExpandCondPseudos16(
"mips16-dont-expand-cond-pseudo",
cl::init(false),
cl::desc("Dont expand conditional move related "
"pseudos for Mips 16"),
cl::Hidden);
static const uint16_t O32IntRegs[4] = {
@ -1230,11 +1236,74 @@ MipsTargetLowering::EmitBPOSGE32(MachineInstr *MI, MachineBasicBlock *BB) const{
return Sink;
}
MachineBasicBlock *MipsTargetLowering::EmitSel16(unsigned Opc, MachineInstr *MI,
MachineBasicBlock *BB) const {
if (DontExpandCondPseudos16)
return BB;
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
// To "insert" a SELECT_CC instruction, we actually have to insert the
// diamond control-flow pattern. The incoming instruction knows the
// destination vreg to set, the condition code register to branch on, the
// true/false values to select between, and a branch opcode to use.
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator It = BB;
++It;
// thisMBB:
// ...
// TrueVal = ...
// setcc r1, r2, r3
// bNE r1, r0, copy1MBB
// fallthrough --> copy0MBB
MachineBasicBlock *thisMBB = BB;
MachineFunction *F = BB->getParent();
MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, copy0MBB);
F->insert(It, sinkMBB);
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), BB,
llvm::next(MachineBasicBlock::iterator(MI)),
BB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
// Next, add the true and fallthrough blocks as its successors.
BB->addSuccessor(copy0MBB);
BB->addSuccessor(sinkMBB);
BuildMI(BB, dl, TII->get(Opc)).addReg(MI->getOperand(3).getReg())
.addMBB(sinkMBB);
// copy0MBB:
// %FalseValue = ...
// # fallthrough to sinkMBB
BB = copy0MBB;
// Update machine-CFG edges
BB->addSuccessor(sinkMBB);
// sinkMBB:
// %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ]
// ...
BB = sinkMBB;
BuildMI(*BB, BB->begin(), dl,
TII->get(Mips::PHI), MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg()).addMBB(thisMBB)
.addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB);
MI->eraseFromParent(); // The pseudo instruction is gone now.
return BB;
}
MachineBasicBlock *
MipsTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *BB) const {
switch (MI->getOpcode()) {
default: llvm_unreachable("Unexpected instr type to insert");
default:
llvm_unreachable("Unexpected instr type to insert");
case Mips::ATOMIC_LOAD_ADD_I8:
case Mips::ATOMIC_LOAD_ADD_I8_P8:
return EmitAtomicBinaryPartword(MI, BB, 1, Mips::ADDu);
@ -1340,6 +1409,10 @@ MipsTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
return EmitAtomicCmpSwap(MI, BB, 8);
case Mips::BPOSGE32_PSEUDO:
return EmitBPOSGE32(MI, BB);
case Mips::SelBeqZ:
return EmitSel16(Mips::BeqzRxImm16, MI, BB);
case Mips::SelBneZ:
return EmitSel16(Mips::BnezRxImm16, MI, BB);
}
}

View File

@ -404,6 +404,9 @@ namespace llvm {
MachineBasicBlock *BB, unsigned Size) const;
MachineBasicBlock *EmitAtomicCmpSwapPartword(MachineInstr *MI,
MachineBasicBlock *BB, unsigned Size) const;
MachineBasicBlock *EmitSel16(unsigned Opc, MachineInstr *MI,
MachineBasicBlock *BB) const;
};
}

View File

@ -67,7 +67,7 @@ entry:
%2 = load i32* @f, align 4
%cond = select i1 %cmp, i32 %1, i32 %2
store i32 %cond, i32* @z1, align 4
; 16: beqz ${{[0-9]+}}, .+4
; 16: beqz ${{[0-9]+}}, $BB{{[0-9]+}}_{{[0-9]}}
; 16: move ${{[0-9]+}}, ${{[0-9]+}}
%3 = load i32* @b, align 4
%cmp1 = icmp eq i32 %3, 0
@ -238,7 +238,7 @@ entry:
%2 = load i32* @t, align 4
%cond = select i1 %cmp, i32 %1, i32 %2
store i32 %cond, i32* @z1, align 4
; 16: bnez ${{[0-9]+}}, .+4
; 16: bnez ${{[0-9]+}}, $BB{{[0-9]+}}_{{[0-9]}}
; 16: move ${{[0-9]+}}, ${{[0-9]+}}
%3 = load i32* @b, align 4
%cmp1 = icmp ne i32 %3, 0
@ -260,7 +260,7 @@ entry:
%2 = load i32* @t, align 4
%cond = select i1 %tobool, i32 %1, i32 %2
store i32 %cond, i32* @z1, align 4
; 16: bnez ${{[0-9]+}}, .+4
; 16: bnez ${{[0-9]+}}, $BB{{[0-9]+}}_{{[0-9]}}
; 16: move ${{[0-9]+}}, ${{[0-9]+}}
%3 = load i32* @b, align 4
%tobool1 = icmp ne i32 %3, 0