[mips] Support SELECT nodes for targets that don't have conditional-move instructions.
Summary:
For Mips targets that do not have conditional-move instructions, ie. targets
before MIPS32 and MIPS-IV, we have to insert a diamond control-flow
pattern in order to support SELECT nodes. In order to do that, we add
pseudo-instructions with a custom inserter that emits the necessary
control-flow that selects the correct value.
With this patch we add complete support for code generation of Mips-II targets
based on the LLVM test-suite.
Reviewers: dsanders
Reviewed By: dsanders
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D6212
llvm-svn: 224124
diff --git a/llvm/lib/Target/Mips/MipsCondMov.td b/llvm/lib/Target/Mips/MipsCondMov.td
index 690f626..aac5413 100644
--- a/llvm/lib/Target/Mips/MipsCondMov.td
+++ b/llvm/lib/Target/Mips/MipsCondMov.td
@@ -263,3 +263,34 @@
FGR_64;
defm : MovnPats<GPR64, FGR64, MOVN_I64_D64, XOR64>, INSN_MIPS4_32_NOT_32R6_64R6,
FGR_64;
+
+// For targets that don't have conditional-move instructions
+// we have to match SELECT nodes with pseudo instructions.
+let usesCustomInserter = 1 in {
+ class Select_Pseudo<RegisterOperand RC> :
+ PseudoSE<(outs RC:$dst), (ins GPR32Opnd:$cond, RC:$T, RC:$F),
+ [(set RC:$dst, (select GPR32Opnd:$cond, RC:$T, RC:$F))]>,
+ ISA_MIPS1_NOT_4_32;
+
+ class SelectFP_Pseudo_T<RegisterOperand RC> :
+ PseudoSE<(outs RC:$dst), (ins GPR32Opnd:$cond, RC:$T, RC:$F),
+ [(set RC:$dst, (MipsCMovFP_T RC:$T, GPR32Opnd:$cond, RC:$F))]>,
+ ISA_MIPS1_NOT_4_32;
+
+ class SelectFP_Pseudo_F<RegisterOperand RC> :
+ PseudoSE<(outs RC:$dst), (ins GPR32Opnd:$cond, RC:$T, RC:$F),
+ [(set RC:$dst, (MipsCMovFP_F RC:$T, GPR32Opnd:$cond, RC:$F))]>,
+ ISA_MIPS1_NOT_4_32;
+}
+
+def PseudoSELECT_I : Select_Pseudo<GPR32Opnd>;
+def PseudoSELECT_S : Select_Pseudo<FGR32Opnd>;
+def PseudoSELECT_D32 : Select_Pseudo<AFGR64Opnd>;
+
+def PseudoSELECTFP_T_I : SelectFP_Pseudo_T<GPR32Opnd>;
+def PseudoSELECTFP_T_S : SelectFP_Pseudo_T<FGR32Opnd>;
+def PseudoSELECTFP_T_D32 : SelectFP_Pseudo_T<AFGR64Opnd>;
+
+def PseudoSELECTFP_F_I : SelectFP_Pseudo_F<GPR32Opnd>;
+def PseudoSELECTFP_F_S : SelectFP_Pseudo_F<FGR32Opnd>;
+def PseudoSELECTFP_F_D32 : SelectFP_Pseudo_F<AFGR64Opnd>;
diff --git a/llvm/lib/Target/Mips/MipsISelLowering.cpp b/llvm/lib/Target/Mips/MipsISelLowering.cpp
index 8e904eb..7fc504e 100644
--- a/llvm/lib/Target/Mips/MipsISelLowering.cpp
+++ b/llvm/lib/Target/Mips/MipsISelLowering.cpp
@@ -945,6 +945,19 @@
MI, *BB, *getTargetMachine().getSubtargetImpl()->getInstrInfo(), true);
case Mips::SEL_D:
return emitSEL_D(MI, BB);
+
+ case Mips::PseudoSELECT_I:
+ case Mips::PseudoSELECT_S:
+ case Mips::PseudoSELECT_D32:
+ return emitPseudoSELECT(MI, BB, false, Mips::BNE);
+ case Mips::PseudoSELECTFP_F_I:
+ case Mips::PseudoSELECTFP_F_S:
+ case Mips::PseudoSELECTFP_F_D32:
+ return emitPseudoSELECT(MI, BB, true, Mips::BC1F);
+ case Mips::PseudoSELECTFP_T_I:
+ case Mips::PseudoSELECTFP_T_S:
+ case Mips::PseudoSELECTFP_T_D32:
+ return emitPseudoSELECT(MI, BB, true, Mips::BC1T);
}
}
@@ -3736,3 +3749,80 @@
State->addInRegsParamInfo(FirstReg, FirstReg + NumRegs);
}
+
+MachineBasicBlock *
+MipsTargetLowering::emitPseudoSELECT(MachineInstr *MI, MachineBasicBlock *BB,
+ bool isFPCmp, unsigned Opc) const {
+ assert(!(Subtarget.hasMips4() || Subtarget.hasMips32()) &&
+ "Subtarget already supports SELECT nodes with the use of"
+ "conditional-move instructions.");
+
+ const TargetInstrInfo *TII =
+ getTargetMachine().getSubtargetImpl()->getInstrInfo();
+ DebugLoc DL = MI->getDebugLoc();
+
+ // To "insert" a SELECT 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,
+ std::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);
+
+ if (isFPCmp) {
+ // bc1[tf] cc, sinkMBB
+ BuildMI(BB, DL, TII->get(Opc))
+ .addReg(MI->getOperand(1).getReg())
+ .addMBB(sinkMBB);
+ } else {
+ // bne rs, $0, sinkMBB
+ BuildMI(BB, DL, TII->get(Opc))
+ .addReg(MI->getOperand(1).getReg())
+ .addReg(Mips::ZERO)
+ .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(2).getReg()).addMBB(thisMBB)
+ .addReg(MI->getOperand(3).getReg()).addMBB(copy0MBB);
+
+ MI->eraseFromParent(); // The pseudo instruction is gone now.
+
+ return BB;
+}
diff --git a/llvm/lib/Target/Mips/MipsISelLowering.h b/llvm/lib/Target/Mips/MipsISelLowering.h
index 60e53da..3254f1f 100644
--- a/llvm/lib/Target/Mips/MipsISelLowering.h
+++ b/llvm/lib/Target/Mips/MipsISelLowering.h
@@ -534,6 +534,9 @@
MachineBasicBlock *emitAtomicCmpSwapPartword(MachineInstr *MI,
MachineBasicBlock *BB, unsigned Size) const;
MachineBasicBlock *emitSEL_D(MachineInstr *MI, MachineBasicBlock *BB) const;
+ MachineBasicBlock *emitPseudoSELECT(MachineInstr *MI,
+ MachineBasicBlock *BB, bool isFPCmp,
+ unsigned Opc) const;
};
/// Create MipsTargetLowering objects.
diff --git a/llvm/lib/Target/Mips/MipsInstrInfo.td b/llvm/lib/Target/Mips/MipsInstrInfo.td
index e5e6c9e..18f774a 100644
--- a/llvm/lib/Target/Mips/MipsInstrInfo.td
+++ b/llvm/lib/Target/Mips/MipsInstrInfo.td
@@ -156,6 +156,8 @@
AssemblerPredicate<"FeatureMips3">;
def HasMips4_32 : Predicate<"Subtarget->hasMips4_32()">,
AssemblerPredicate<"FeatureMips4_32">;
+def NotMips4_32 : Predicate<"!Subtarget->hasMips4_32()">,
+ AssemblerPredicate<"FeatureMips4_32">;
def HasMips4_32r2 : Predicate<"Subtarget->hasMips4_32r2()">,
AssemblerPredicate<"FeatureMips4_32r2">;
def HasMips5_32r2 : Predicate<"Subtarget->hasMips5_32r2()">,
@@ -220,6 +222,9 @@
// subtractive predicate will hopefully keep us under the 32 predicate
// limit long enough to develop an alternative way to handle P1||P2
// predicates.
+class ISA_MIPS1_NOT_4_32 {
+ list<Predicate> InsnPredicates = [NotMips4_32];
+}
class ISA_MIPS1_NOT_32R6_64R6 {
list<Predicate> InsnPredicates = [NotMips32r6, NotMips64r6];
}