def F32 : PatLeaf<(f32 IntRegs:$R)>;
def F64 : PatLeaf<(f64 DoubleRegs:$R)>;
+// Pattern fragments to extract the low and high subregisters from a
+// 64-bit value.
+def LoReg: OutPatFrag<(ops node:$Rs),
+ (EXTRACT_SUBREG (i64 $Rs), subreg_loreg)>;
+
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Transfer control register.
let hasSideEffects = 0 in
-def TFCR : CRInst<(outs CRRegs:$dst), (ins IntRegs:$src1),
+def TFCR : CRInst<(outs CtrRegs:$dst), (ins IntRegs:$src1),
"$dst = $src1",
[]>;
//===----------------------------------------------------------------------===//
// CR -
//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// JR +
+//===----------------------------------------------------------------------===//
+
def retflag : SDNode<"HexagonISD::RET_FLAG", SDTNone,
[SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
-def eh_return: SDNode<"HexagonISD::EH_RETURN", SDTNone,
- [SDNPHasChain]>;
+def eh_return: SDNode<"HexagonISD::EH_RETURN", SDTNone, [SDNPHasChain]>;
def SDHexagonBR_JT: SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
def HexagonBR_JT: SDNode<"HexagonISD::BR_JT", SDHexagonBR_JT, [SDNPHasChain]>;
-let InputType = "imm", isBarrier = 1, isPredicable = 1,
-Defs = [PC], isExtendable = 1, opExtendable = 0, isExtentSigned = 1,
-opExtentBits = 24, isCodeGenOnly = 0 in
-class T_JMP <dag InsDag, list<dag> JumpList = []>
- : JInst<(outs), InsDag,
- "jump $dst" , JumpList> {
- bits<24> dst;
+class CondStr<string CReg, bit True, bit New> {
+ string S = "if (" # !if(True,"","!") # CReg # !if(New,".new","") # ") ";
+}
+class JumpOpcStr<string Mnemonic, bit New, bit Taken> {
+ string S = Mnemonic # !if(New, !if(Taken,":t",":nt"), "");
+}
+let isBranch = 1, isBarrier = 1, Defs = [PC], hasSideEffects = 0,
+ isPredicable = 1,
+ isExtendable = 1, opExtendable = 0, isExtentSigned = 1,
+ opExtentBits = 24, opExtentAlign = 2, InputType = "imm" in
+class T_JMP<string ExtStr>
+ : JInst<(outs), (ins brtarget:$dst),
+ "jump " # ExtStr # "$dst",
+ [], "", J_tc_2early_SLOT23> {
+ bits<24> dst;
let IClass = 0b0101;
let Inst{27-25} = 0b100;
let Inst{13-1} = dst{14-2};
}
-let InputType = "imm", isExtendable = 1, opExtendable = 1, isExtentSigned = 1,
-Defs = [PC], isPredicated = 1, opExtentBits = 17 in
-class T_JMP_c <bit PredNot, bit isPredNew, bit isTak>:
- JInst<(outs ), (ins PredRegs:$src, brtarget:$dst),
- !if(PredNot, "if (!$src", "if ($src")#
- !if(isPredNew, ".new) ", ") ")#"jump"#
- !if(isPredNew, !if(isTak, ":t ", ":nt "), " ")#"$dst"> {
-
+let isBranch = 1, Defs = [PC], hasSideEffects = 0, isPredicated = 1,
+ isExtendable = 1, opExtendable = 1, isExtentSigned = 1,
+ opExtentBits = 17, opExtentAlign = 2, InputType = "imm" in
+class T_JMP_c<bit PredNot, bit isPredNew, bit isTak, string ExtStr>
+ : JInst<(outs), (ins PredRegs:$src, brtarget:$dst),
+ CondStr<"$src", !if(PredNot,0,1), isPredNew>.S #
+ JumpOpcStr<"jump", isPredNew, isTak>.S # " " #
+ ExtStr # "$dst",
+ [], "", J_tc_2early_SLOT23>, ImmRegRel {
let isTaken = isTak;
- let isBrTaken = !if(isPredNew, !if(isTaken, "true", "false"), "");
let isPredicatedFalse = PredNot;
let isPredicatedNew = isPredNew;
bits<2> src;
let Inst{7-1} = dst{8-2};
}
-let isBarrier = 1, Defs = [PC], isPredicable = 1, InputType = "reg" in
-class T_JMPr<dag InsDag = (ins IntRegs:$dst)>
- : JRInst<(outs ), InsDag,
- "jumpr $dst" ,
- []> {
+multiclass JMP_Pred<bit PredNot, string ExtStr> {
+ def NAME : T_JMP_c<PredNot, 0, 0, ExtStr>;
+ // Predicate new
+ def NAME#newpt : T_JMP_c<PredNot, 1, 1, ExtStr>; // taken
+ def NAME#new : T_JMP_c<PredNot, 1, 0, ExtStr>; // not taken
+}
+
+multiclass JMP_base<string BaseOp, string ExtStr> {
+ let BaseOpcode = BaseOp in {
+ def NAME : T_JMP<ExtStr>;
+ defm t : JMP_Pred<0, ExtStr>;
+ defm f : JMP_Pred<1, ExtStr>;
+ }
+}
+
+// Jumps to address stored in a register, JUMPR_MISC
+// if ([[!]P[.new]]) jumpr[:t/nt] Rs
+let isBranch = 1, isIndirectBranch = 1, isBarrier = 1, Defs = [PC],
+ isPredicable = 1, hasSideEffects = 0, InputType = "reg" in
+class T_JMPr
+ : JRInst<(outs), (ins IntRegs:$dst),
+ "jumpr $dst", [], "", J_tc_2early_SLOT2> {
bits<5> dst;
let IClass = 0b0101;
let Inst{20-16} = dst;
}
-let Defs = [PC], isPredicated = 1, InputType = "reg" in
-class T_JMPr_c <bit PredNot, bit isPredNew, bit isTak>:
- JRInst <(outs ), (ins PredRegs:$src, IntRegs:$dst),
- !if(PredNot, "if (!$src", "if ($src")#
- !if(isPredNew, ".new) ", ") ")#"jumpr"#
- !if(isPredNew, !if(isTak, ":t ", ":nt "), " ")#"$dst"> {
+let isBranch = 1, isIndirectBranch = 1, Defs = [PC], isPredicated = 1,
+ hasSideEffects = 0, InputType = "reg" in
+class T_JMPr_c <bit PredNot, bit isPredNew, bit isTak>
+ : JRInst <(outs), (ins PredRegs:$src, IntRegs:$dst),
+ CondStr<"$src", !if(PredNot,0,1), isPredNew>.S #
+ JumpOpcStr<"jumpr", isPredNew, isTak>.S # " $dst", [],
+ "", J_tc_2early_SLOT2> {
let isTaken = isTak;
- let isBrTaken = !if(isPredNew, !if(isTaken, "true", "false"), "");
let isPredicatedFalse = PredNot;
let isPredicatedNew = isPredNew;
bits<2> src;
let Inst{12} = !if(isPredNew, isTak, zero);
let Inst{11} = isPredNew;
let Inst{9-8} = src;
- let Predicates = !if(isPredNew, [HasV3T], [HasV2T]);
- let validSubTargets = !if(isPredNew, HasV3SubT, HasV2SubT);
-}
-
-multiclass JMP_Pred<bit PredNot> {
- def _#NAME : T_JMP_c<PredNot, 0, 0>;
- // Predicate new
- def _#NAME#new_t : T_JMP_c<PredNot, 1, 1>; // taken
- def _#NAME#new_nt : T_JMP_c<PredNot, 1, 0>; // not taken
-}
-
-multiclass JMP_base<string BaseOp> {
- let BaseOpcode = BaseOp in {
- def NAME : T_JMP<(ins brtarget:$dst), [(br bb:$dst)]>;
- defm t : JMP_Pred<0>;
- defm f : JMP_Pred<1>;
- }
}
multiclass JMPR_Pred<bit PredNot> {
def NAME: T_JMPr_c<PredNot, 0, 0>;
// Predicate new
- def NAME#new
_tV3 : T_JMPr_c<PredNot, 1, 1>; // taken
- def NAME#new
_ntV3 : T_JMPr_c<PredNot, 1, 0>; // not taken
+ def NAME#new
pt : T_JMPr_c<PredNot, 1, 1>; // taken
+ def NAME#new
: T_JMPr_c<PredNot, 1, 0>; // not taken
}
multiclass JMPR_base<string BaseOp> {
let BaseOpcode = BaseOp in {
def NAME : T_JMPr;
- defm _t : JMPR_Pred<0>;
- defm _f : JMPR_Pred<1>;
+ defm t : JMPR_Pred<0>;
+ defm f : JMPR_Pred<1>;
}
}
-let isTerminator = 1, hasSideEffects = 0 in {
-let isBranch = 1 in
-defm JMP : JMP_base<"JMP">, PredNewRel;
+let isCall = 1, hasSideEffects = 1 in
+class JUMPR_MISC_CALLR<bit isPred, bit isPredNot,
+ dag InputDag = (ins IntRegs:$Rs)>
+ : JRInst<(outs), InputDag,
+ !if(isPred, !if(isPredNot, "if (!$Pu) callr $Rs",
+ "if ($Pu) callr $Rs"),
+ "callr $Rs"),
+ [], "", J_tc_2early_SLOT2> {
+ bits<5> Rs;
+ bits<2> Pu;
+ let isPredicated = isPred;
+ let isPredicatedFalse = isPredNot;
+
+ let IClass = 0b0101;
+ let Inst{27-25} = 0b000;
+ let Inst{24-23} = !if (isPred, 0b10, 0b01);
+ let Inst{22} = 0;
+ let Inst{21} = isPredNot;
+ let Inst{9-8} = !if (isPred, Pu, 0b00);
+ let Inst{20-16} = Rs;
-let isBranch = 1, isIndirectBranch = 1 in
-defm JMPR : JMPR_base<"JMPr">, PredNewRel;
+ }
-let isReturn = 1, isCodeGenOnly = 1 in
-defm JMPret : JMPR_base<"JMPret">, PredNewRel;
+let Defs = VolatileV3.Regs, isCodeGenOnly = 0 in {
+ def J2_callrt : JUMPR_MISC_CALLR<1, 0, (ins PredRegs:$Pu, IntRegs:$Rs)>;
+ def J2_callrf : JUMPR_MISC_CALLR<1, 1, (ins PredRegs:$Pu, IntRegs:$Rs)>;
}
-def : Pat<(retflag),
- (JMPret (i32 R31))>;
+let isTerminator = 1, hasSideEffects = 0, isCodeGenOnly = 0 in {
+ defm J2_jump : JMP_base<"JMP", "">, PredNewRel;
+
+ // Deal with explicit assembly
+ // - never extened a jump #, always extend a jump ##
+ let isAsmParserOnly = 1 in {
+ defm J2_jump_ext : JMP_base<"JMP", "##">;
+ defm J2_jump_noext : JMP_base<"JMP", "#">;
+ }
+
+ defm J2_jumpr : JMPR_base<"JMPr">, PredNewRel;
+
+ let isReturn = 1, isCodeGenOnly = 1 in
+ defm JMPret : JMPR_base<"JMPret">, PredNewRel;
+}
-def : Pat <(brcond (i1 PredRegs:$src1), bb:$offset),
- (JMP_t (i1 PredRegs:$src1), bb:$offset)>;
+def: Pat<(br bb:$dst),
+ (J2_jump brtarget:$dst)>;
+def: Pat<(retflag),
+ (JMPret (i32 R31))>;
+def: Pat<(brcond (i1 PredRegs:$src1), bb:$offset),
+ (J2_jumpt PredRegs:$src1, bb:$offset)>;
// A return through builtin_eh_return.
let isReturn = 1, isTerminator = 1, isBarrier = 1, hasSideEffects = 0,
-isCodeGenOnly = 1, Defs = [PC], Uses = [R28], isPredicable = 0 in
+ isCodeGenOnly = 1, Defs = [PC], Uses = [R28], isPredicable = 0 in
def EH_RETURN_JMPR : T_JMPr;
-def : Pat<(eh_return),
- (EH_RETURN_JMPR (i32 R31))>;
-
-def : Pat<(HexagonBR_JT (i32 IntRegs:$dst)),
- (JMPR (i32 IntRegs:$dst))>;
-
-def : Pat<(brind (i32 IntRegs:$dst)),
- (JMPR (i32 IntRegs:$dst))>;
+def: Pat<(eh_return),
+ (EH_RETURN_JMPR (i32 R31))>;
+def: Pat<(HexagonBR_JT (i32 IntRegs:$dst)),
+ (J2_jumpr IntRegs:$dst)>;
+def: Pat<(brind (i32 IntRegs:$dst)),
+ (J2_jumpr IntRegs:$dst)>;
//===----------------------------------------------------------------------===//
// JR -
// Load - Base with Immediate offset addressing mode
-multiclass LD_Idxd_Pbase<string mnemonic, RegisterClass RC, Operand predImmOp,
+multiclass LD_Idxd_Pbase2<string mnemonic, RegisterClass RC, Operand predImmOp,
bit isNot, bit isPredNew> {
let isPredicatedNew = isPredNew in
def NAME : LDInst2<(outs RC:$dst),
[]>;
}
-multiclass LD_Idxd_Pred<string mnemonic, RegisterClass RC, Operand predImmOp,
+multiclass LD_Idxd_Pred2<string mnemonic, RegisterClass RC, Operand predImmOp,
bit PredNot> {
let isPredicatedFalse = PredNot in {
- defm _c#NAME : LD_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 0>;
+ defm _c#NAME : LD_Idxd_Pbase2<mnemonic, RC, predImmOp, PredNot, 0>;
// Predicate new
- defm _cdn#NAME : LD_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 1>;
+ defm _cdn#NAME : LD_Idxd_Pbase2<mnemonic, RC, predImmOp, PredNot, 1>;
}
}
let isExtendable = 1, hasSideEffects = 0 in
-multiclass LD_Idxd<string mnemonic, string CextOp, RegisterClass RC,
+multiclass LD_Idxd2<string mnemonic, string CextOp, RegisterClass RC,
Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
bits<5> PredImmBits> {
let opExtendable = 3, isExtentSigned = 0, opExtentBits = PredImmBits,
isPredicated = 1 in {
- defm Pt : LD_Idxd_Pred<mnemonic, RC, predImmOp, 0 >;
- defm NotPt : LD_Idxd_Pred<mnemonic, RC, predImmOp, 1 >;
+ defm Pt : LD_Idxd_Pred2<mnemonic, RC, predImmOp, 0 >;
+ defm NotPt : LD_Idxd_Pred2<mnemonic, RC, predImmOp, 1 >;
}
}
}
let addrMode = BaseImmOffset in {
let accessSize = ByteAccess in {
- defm LDrib_indexed: LD_Idxd <"memb", "LDrib", IntRegs, s11_0Ext, u6_0Ext,
+ defm LDrib_indexed: LD_Idxd2 <"memb", "LDrib", IntRegs, s11_0Ext, u6_0Ext,
11, 6>, AddrModeRel;
- defm LDriub_indexed: LD_Idxd <"memub" , "LDriub", IntRegs, s11_0Ext, u6_0Ext,
+ defm LDriub_indexed: LD_Idxd2 <"memub" , "LDriub", IntRegs, s11_0Ext, u6_0Ext,
11, 6>, AddrModeRel;
}
let accessSize = HalfWordAccess in {
- defm LDrih_indexed: LD_Idxd <"memh", "LDrih", IntRegs, s11_1Ext, u6_1Ext,
+ defm LDrih_indexed: LD_Idxd2 <"memh", "LDrih", IntRegs, s11_1Ext, u6_1Ext,
12, 7>, AddrModeRel;
- defm LDriuh_indexed: LD_Idxd <"memuh", "LDriuh", IntRegs, s11_1Ext, u6_1Ext,
+ defm LDriuh_indexed: LD_Idxd2 <"memuh", "LDriuh", IntRegs, s11_1Ext, u6_1Ext,
12, 7>, AddrModeRel;
}
let accessSize = WordAccess in
- defm LDriw_indexed: LD_Idxd <"memw", "LDriw", IntRegs, s11_2Ext, u6_2Ext,
+ defm LDriw_indexed: LD_Idxd2 <"memw", "LDriw", IntRegs, s11_2Ext, u6_2Ext,
13, 8>, AddrModeRel;
let accessSize = DoubleWordAccess in
- defm LDrid_indexed: LD_Idxd <"memd", "LDrid", DoubleRegs, s11_3Ext, u6_3Ext,
+ defm LDrid_indexed: LD_Idxd2 <"memd", "LDrid", DoubleRegs, s11_3Ext, u6_3Ext,
14, 9>, AddrModeRel;
}
//===----------------------------------------------------------------------===//
// MTYPE/MPYH +
//===----------------------------------------------------------------------===//
-// Multiply and use lower result.
-// Rd=+mpyi(Rs,#u8)
-let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 8 in
-def MPYI_riu : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u8Ext:$src2),
- "$dst =+ mpyi($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
- u8ExtPred:$src2))]>;
-// Rd=-mpyi(Rs,#u8)
-def MPYI_rin : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u8Imm:$src2),
- "$dst =- mpyi($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (ineg (mul (i32 IntRegs:$src1),
- u8ImmPred:$src2)))]>;
+//===----------------------------------------------------------------------===//
+// Template Class
+// MPYS / Multipy signed/unsigned halfwords
+//Rd=mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:rnd][:sat]
+//===----------------------------------------------------------------------===//
+
+let hasNewValue = 1, opNewValue = 0 in
+class T_M2_mpy < bits<2> LHbits, bit isSat, bit isRnd,
+ bit hasShift, bit isUnsigned>
+ : MInst < (outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
+ "$Rd = "#!if(isUnsigned,"mpyu","mpy")#"($Rs."#!if(LHbits{1},"h","l")
+ #", $Rt."#!if(LHbits{0},"h)","l)")
+ #!if(hasShift,":<<1","")
+ #!if(isRnd,":rnd","")
+ #!if(isSat,":sat",""),
+ [], "", M_tc_3x_SLOT23 > {
+ bits<5> Rd;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b1100;
+ let Inst{23} = hasShift;
+ let Inst{22} = isUnsigned;
+ let Inst{21} = isRnd;
+ let Inst{7} = isSat;
+ let Inst{6-5} = LHbits;
+ let Inst{4-0} = Rd;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ }
+
+//Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1]
+let isCodeGenOnly = 0 in {
+def M2_mpy_ll_s1: T_M2_mpy<0b00, 0, 0, 1, 0>;
+def M2_mpy_ll_s0: T_M2_mpy<0b00, 0, 0, 0, 0>;
+def M2_mpy_lh_s1: T_M2_mpy<0b01, 0, 0, 1, 0>;
+def M2_mpy_lh_s0: T_M2_mpy<0b01, 0, 0, 0, 0>;
+def M2_mpy_hl_s1: T_M2_mpy<0b10, 0, 0, 1, 0>;
+def M2_mpy_hl_s0: T_M2_mpy<0b10, 0, 0, 0, 0>;
+def M2_mpy_hh_s1: T_M2_mpy<0b11, 0, 0, 1, 0>;
+def M2_mpy_hh_s0: T_M2_mpy<0b11, 0, 0, 0, 0>;
+}
+
+//Rd=mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
+let isCodeGenOnly = 0 in {
+def M2_mpyu_ll_s1: T_M2_mpy<0b00, 0, 0, 1, 1>;
+def M2_mpyu_ll_s0: T_M2_mpy<0b00, 0, 0, 0, 1>;
+def M2_mpyu_lh_s1: T_M2_mpy<0b01, 0, 0, 1, 1>;
+def M2_mpyu_lh_s0: T_M2_mpy<0b01, 0, 0, 0, 1>;
+def M2_mpyu_hl_s1: T_M2_mpy<0b10, 0, 0, 1, 1>;
+def M2_mpyu_hl_s0: T_M2_mpy<0b10, 0, 0, 0, 1>;
+def M2_mpyu_hh_s1: T_M2_mpy<0b11, 0, 0, 1, 1>;
+def M2_mpyu_hh_s0: T_M2_mpy<0b11, 0, 0, 0, 1>;
+}
+
+//Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1]:rnd
+let isCodeGenOnly = 0 in {
+def M2_mpy_rnd_ll_s1: T_M2_mpy <0b00, 0, 1, 1, 0>;
+def M2_mpy_rnd_ll_s0: T_M2_mpy <0b00, 0, 1, 0, 0>;
+def M2_mpy_rnd_lh_s1: T_M2_mpy <0b01, 0, 1, 1, 0>;
+def M2_mpy_rnd_lh_s0: T_M2_mpy <0b01, 0, 1, 0, 0>;
+def M2_mpy_rnd_hl_s1: T_M2_mpy <0b10, 0, 1, 1, 0>;
+def M2_mpy_rnd_hl_s0: T_M2_mpy <0b10, 0, 1, 0, 0>;
+def M2_mpy_rnd_hh_s1: T_M2_mpy <0b11, 0, 1, 1, 0>;
+def M2_mpy_rnd_hh_s0: T_M2_mpy <0b11, 0, 1, 0, 0>;
+}
+
+//Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1][:sat]
+//Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1][:rnd][:sat]
+let Defs = [USR_OVF], isCodeGenOnly = 0 in {
+ def M2_mpy_sat_ll_s1: T_M2_mpy <0b00, 1, 0, 1, 0>;
+ def M2_mpy_sat_ll_s0: T_M2_mpy <0b00, 1, 0, 0, 0>;
+ def M2_mpy_sat_lh_s1: T_M2_mpy <0b01, 1, 0, 1, 0>;
+ def M2_mpy_sat_lh_s0: T_M2_mpy <0b01, 1, 0, 0, 0>;
+ def M2_mpy_sat_hl_s1: T_M2_mpy <0b10, 1, 0, 1, 0>;
+ def M2_mpy_sat_hl_s0: T_M2_mpy <0b10, 1, 0, 0, 0>;
+ def M2_mpy_sat_hh_s1: T_M2_mpy <0b11, 1, 0, 1, 0>;
+ def M2_mpy_sat_hh_s0: T_M2_mpy <0b11, 1, 0, 0, 0>;
+
+ def M2_mpy_sat_rnd_ll_s1: T_M2_mpy <0b00, 1, 1, 1, 0>;
+ def M2_mpy_sat_rnd_ll_s0: T_M2_mpy <0b00, 1, 1, 0, 0>;
+ def M2_mpy_sat_rnd_lh_s1: T_M2_mpy <0b01, 1, 1, 1, 0>;
+ def M2_mpy_sat_rnd_lh_s0: T_M2_mpy <0b01, 1, 1, 0, 0>;
+ def M2_mpy_sat_rnd_hl_s1: T_M2_mpy <0b10, 1, 1, 1, 0>;
+ def M2_mpy_sat_rnd_hl_s0: T_M2_mpy <0b10, 1, 1, 0, 0>;
+ def M2_mpy_sat_rnd_hh_s1: T_M2_mpy <0b11, 1, 1, 1, 0>;
+ def M2_mpy_sat_rnd_hh_s0: T_M2_mpy <0b11, 1, 1, 0, 0>;
+}
+
+//===----------------------------------------------------------------------===//
+// Template Class
+// MPYS / Multipy signed/unsigned halfwords and add/subtract the
+// result from the accumulator.
+//Rx [-+]= mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:sat]
+//===----------------------------------------------------------------------===//
+
+let hasNewValue = 1, opNewValue = 0 in
+class T_M2_mpy_acc < bits<2> LHbits, bit isSat, bit isNac,
+ bit hasShift, bit isUnsigned >
+ : MInst_acc<(outs IntRegs:$Rx), (ins IntRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
+ "$Rx "#!if(isNac,"-= ","+= ")#!if(isUnsigned,"mpyu","mpy")
+ #"($Rs."#!if(LHbits{1},"h","l")
+ #", $Rt."#!if(LHbits{0},"h)","l)")
+ #!if(hasShift,":<<1","")
+ #!if(isSat,":sat",""),
+ [], "$dst2 = $Rx", M_tc_3x_SLOT23 > {
+ bits<5> Rx;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1110;
+ let Inst{27-24} = 0b1110;
+ let Inst{23} = hasShift;
+ let Inst{22} = isUnsigned;
+ let Inst{21} = isNac;
+ let Inst{7} = isSat;
+ let Inst{6-5} = LHbits;
+ let Inst{4-0} = Rx;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ }
+
+//Rx += mpy(Rs.[H|L],Rt.[H|L])[:<<1]
+let isCodeGenOnly = 0 in {
+def M2_mpy_acc_ll_s1: T_M2_mpy_acc <0b00, 0, 0, 1, 0>;
+def M2_mpy_acc_ll_s0: T_M2_mpy_acc <0b00, 0, 0, 0, 0>;
+def M2_mpy_acc_lh_s1: T_M2_mpy_acc <0b01, 0, 0, 1, 0>;
+def M2_mpy_acc_lh_s0: T_M2_mpy_acc <0b01, 0, 0, 0, 0>;
+def M2_mpy_acc_hl_s1: T_M2_mpy_acc <0b10, 0, 0, 1, 0>;
+def M2_mpy_acc_hl_s0: T_M2_mpy_acc <0b10, 0, 0, 0, 0>;
+def M2_mpy_acc_hh_s1: T_M2_mpy_acc <0b11, 0, 0, 1, 0>;
+def M2_mpy_acc_hh_s0: T_M2_mpy_acc <0b11, 0, 0, 0, 0>;
+}
+
+//Rx += mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
+let isCodeGenOnly = 0 in {
+def M2_mpyu_acc_ll_s1: T_M2_mpy_acc <0b00, 0, 0, 1, 1>;
+def M2_mpyu_acc_ll_s0: T_M2_mpy_acc <0b00, 0, 0, 0, 1>;
+def M2_mpyu_acc_lh_s1: T_M2_mpy_acc <0b01, 0, 0, 1, 1>;
+def M2_mpyu_acc_lh_s0: T_M2_mpy_acc <0b01, 0, 0, 0, 1>;
+def M2_mpyu_acc_hl_s1: T_M2_mpy_acc <0b10, 0, 0, 1, 1>;
+def M2_mpyu_acc_hl_s0: T_M2_mpy_acc <0b10, 0, 0, 0, 1>;
+def M2_mpyu_acc_hh_s1: T_M2_mpy_acc <0b11, 0, 0, 1, 1>;
+def M2_mpyu_acc_hh_s0: T_M2_mpy_acc <0b11, 0, 0, 0, 1>;
+}
+
+//Rx -= mpy(Rs.[H|L],Rt.[H|L])[:<<1]
+let isCodeGenOnly = 0 in {
+def M2_mpy_nac_ll_s1: T_M2_mpy_acc <0b00, 0, 1, 1, 0>;
+def M2_mpy_nac_ll_s0: T_M2_mpy_acc <0b00, 0, 1, 0, 0>;
+def M2_mpy_nac_lh_s1: T_M2_mpy_acc <0b01, 0, 1, 1, 0>;
+def M2_mpy_nac_lh_s0: T_M2_mpy_acc <0b01, 0, 1, 0, 0>;
+def M2_mpy_nac_hl_s1: T_M2_mpy_acc <0b10, 0, 1, 1, 0>;
+def M2_mpy_nac_hl_s0: T_M2_mpy_acc <0b10, 0, 1, 0, 0>;
+def M2_mpy_nac_hh_s1: T_M2_mpy_acc <0b11, 0, 1, 1, 0>;
+def M2_mpy_nac_hh_s0: T_M2_mpy_acc <0b11, 0, 1, 0, 0>;
+}
+
+//Rx -= mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
+let isCodeGenOnly = 0 in {
+def M2_mpyu_nac_ll_s1: T_M2_mpy_acc <0b00, 0, 1, 1, 1>;
+def M2_mpyu_nac_ll_s0: T_M2_mpy_acc <0b00, 0, 1, 0, 1>;
+def M2_mpyu_nac_lh_s1: T_M2_mpy_acc <0b01, 0, 1, 1, 1>;
+def M2_mpyu_nac_lh_s0: T_M2_mpy_acc <0b01, 0, 1, 0, 1>;
+def M2_mpyu_nac_hl_s1: T_M2_mpy_acc <0b10, 0, 1, 1, 1>;
+def M2_mpyu_nac_hl_s0: T_M2_mpy_acc <0b10, 0, 1, 0, 1>;
+def M2_mpyu_nac_hh_s1: T_M2_mpy_acc <0b11, 0, 1, 1, 1>;
+def M2_mpyu_nac_hh_s0: T_M2_mpy_acc <0b11, 0, 1, 0, 1>;
+}
+
+//Rx += mpy(Rs.[H|L],Rt.[H|L])[:<<1]:sat
+let isCodeGenOnly = 0 in {
+def M2_mpy_acc_sat_ll_s1: T_M2_mpy_acc <0b00, 1, 0, 1, 0>;
+def M2_mpy_acc_sat_ll_s0: T_M2_mpy_acc <0b00, 1, 0, 0, 0>;
+def M2_mpy_acc_sat_lh_s1: T_M2_mpy_acc <0b01, 1, 0, 1, 0>;
+def M2_mpy_acc_sat_lh_s0: T_M2_mpy_acc <0b01, 1, 0, 0, 0>;
+def M2_mpy_acc_sat_hl_s1: T_M2_mpy_acc <0b10, 1, 0, 1, 0>;
+def M2_mpy_acc_sat_hl_s0: T_M2_mpy_acc <0b10, 1, 0, 0, 0>;
+def M2_mpy_acc_sat_hh_s1: T_M2_mpy_acc <0b11, 1, 0, 1, 0>;
+def M2_mpy_acc_sat_hh_s0: T_M2_mpy_acc <0b11, 1, 0, 0, 0>;
+}
+
+//Rx -= mpy(Rs.[H|L],Rt.[H|L])[:<<1]:sat
+let isCodeGenOnly = 0 in {
+def M2_mpy_nac_sat_ll_s1: T_M2_mpy_acc <0b00, 1, 1, 1, 0>;
+def M2_mpy_nac_sat_ll_s0: T_M2_mpy_acc <0b00, 1, 1, 0, 0>;
+def M2_mpy_nac_sat_lh_s1: T_M2_mpy_acc <0b01, 1, 1, 1, 0>;
+def M2_mpy_nac_sat_lh_s0: T_M2_mpy_acc <0b01, 1, 1, 0, 0>;
+def M2_mpy_nac_sat_hl_s1: T_M2_mpy_acc <0b10, 1, 1, 1, 0>;
+def M2_mpy_nac_sat_hl_s0: T_M2_mpy_acc <0b10, 1, 1, 0, 0>;
+def M2_mpy_nac_sat_hh_s1: T_M2_mpy_acc <0b11, 1, 1, 1, 0>;
+def M2_mpy_nac_sat_hh_s0: T_M2_mpy_acc <0b11, 1, 1, 0, 0>;
+}
+
+//===----------------------------------------------------------------------===//
+// Template Class
+// MPYS / Multipy signed/unsigned halfwords and add/subtract the
+// result from the 64-bit destination register.
+//Rxx [-+]= mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:sat]
+//===----------------------------------------------------------------------===//
+
+class T_M2_mpyd_acc < bits<2> LHbits, bit isNac, bit hasShift, bit isUnsigned>
+ : MInst_acc<(outs DoubleRegs:$Rxx),
+ (ins DoubleRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
+ "$Rxx "#!if(isNac,"-= ","+= ")#!if(isUnsigned,"mpyu","mpy")
+ #"($Rs."#!if(LHbits{1},"h","l")
+ #", $Rt."#!if(LHbits{0},"h)","l)")
+ #!if(hasShift,":<<1",""),
+ [], "$dst2 = $Rxx", M_tc_3x_SLOT23 > {
+ bits<5> Rxx;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b0110;
+ let Inst{23} = hasShift;
+ let Inst{22} = isUnsigned;
+ let Inst{21} = isNac;
+ let Inst{7} = 0;
+ let Inst{6-5} = LHbits;
+ let Inst{4-0} = Rxx;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ }
+
+let isCodeGenOnly = 0 in {
+def M2_mpyd_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 0>;
+def M2_mpyd_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 0>;
+def M2_mpyd_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 0>;
+def M2_mpyd_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 0>;
+
+def M2_mpyd_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 0>;
+def M2_mpyd_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 0>;
+def M2_mpyd_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 0>;
+def M2_mpyd_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 0>;
+
+def M2_mpyd_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 0>;
+def M2_mpyd_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 0>;
+def M2_mpyd_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 0>;
+def M2_mpyd_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 0>;
+
+def M2_mpyd_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 0>;
+def M2_mpyd_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 0>;
+def M2_mpyd_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 0>;
+def M2_mpyd_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 0>;
+
+def M2_mpyud_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 1>;
+def M2_mpyud_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 1>;
+def M2_mpyud_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 1>;
+def M2_mpyud_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 1>;
+
+def M2_mpyud_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 1>;
+def M2_mpyud_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 1>;
+def M2_mpyud_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 1>;
+def M2_mpyud_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 1>;
+
+def M2_mpyud_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 1>;
+def M2_mpyud_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 1>;
+def M2_mpyud_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 1>;
+def M2_mpyud_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 1>;
+
+def M2_mpyud_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 1>;
+def M2_mpyud_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 1>;
+def M2_mpyud_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 1>;
+def M2_mpyud_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 1>;
+}
+
+let hasNewValue = 1, opNewValue = 0 in
+class T_MType_mpy <string mnemonic, bits<4> RegTyBits, RegisterClass RC,
+ bits<3> MajOp, bits<3> MinOp, bit isSat = 0, bit isRnd = 0,
+ string op2Suffix = "", bit isRaw = 0, bit isHi = 0 >
+ : MInst <(outs IntRegs:$dst), (ins RC:$src1, RC:$src2),
+ "$dst = "#mnemonic
+ #"($src1, $src2"#op2Suffix#")"
+ #!if(MajOp{2}, ":<<1", "")
+ #!if(isRnd, ":rnd", "")
+ #!if(isSat, ":sat", "")
+ #!if(isRaw, !if(isHi, ":raw:hi", ":raw:lo"), ""), [] > {
+ bits<5> dst;
+ bits<5> src1;
+ bits<5> src2;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = RegTyBits;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{13} = 0b0;
+ let Inst{12-8} = src2;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = dst;
+ }
+
+class T_MType_dd <string mnemonic, bits<3> MajOp, bits<3> MinOp,
+ bit isSat = 0, bit isRnd = 0 >
+ : T_MType_mpy <mnemonic, 0b1001, DoubleRegs, MajOp, MinOp, isSat, isRnd>;
+
+class T_MType_rr1 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
+ bit isSat = 0, bit isRnd = 0 >
+ : T_MType_mpy<mnemonic, 0b1101, IntRegs, MajOp, MinOp, isSat, isRnd>;
+
+class T_MType_rr2 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
+ bit isSat = 0, bit isRnd = 0, string op2str = "" >
+ : T_MType_mpy<mnemonic, 0b1101, IntRegs, MajOp, MinOp, isSat, isRnd, op2str>;
+
+let CextOpcode = "mpyi", InputType = "reg", isCodeGenOnly = 0 in
+def M2_mpyi : T_MType_rr1 <"mpyi", 0b000, 0b000>, ImmRegRel;
+
+let isCodeGenOnly = 0 in {
+def M2_mpy_up : T_MType_rr1 <"mpy", 0b000, 0b001>;
+def M2_mpyu_up : T_MType_rr1 <"mpyu", 0b010, 0b001>;
+}
+
+let isCodeGenOnly = 0 in
+def M2_dpmpyss_rnd_s0 : T_MType_rr1 <"mpy", 0b001, 0b001, 0, 1>;
+
+let isCodeGenOnly = 0 in {
+def M2_hmmpyh_rs1 : T_MType_rr2 <"mpy", 0b101, 0b100, 1, 1, ".h">;
+def M2_hmmpyl_rs1 : T_MType_rr2 <"mpy", 0b111, 0b100, 1, 1, ".l">;
+}
+
+// V4 Instructions
+let isCodeGenOnly = 0 in {
+def M2_mpysu_up : T_MType_rr1 <"mpysu", 0b011, 0b001, 0>;
+def M2_mpy_up_s1_sat : T_MType_rr1 <"mpy", 0b111, 0b000, 1>;
+
+def M2_hmmpyh_s1 : T_MType_rr2 <"mpy", 0b101, 0b000, 1, 0, ".h">;
+def M2_hmmpyl_s1 : T_MType_rr2 <"mpy", 0b101, 0b001, 1, 0, ".l">;
+}
+
+def: Pat<(i32 (mul I32:$src1, I32:$src2)), (M2_mpyi I32:$src1, I32:$src2)>;
+def: Pat<(i32 (mulhs I32:$src1, I32:$src2)), (M2_mpy_up I32:$src1, I32:$src2)>;
+def: Pat<(i32 (mulhu I32:$src1, I32:$src2)), (M2_mpyu_up I32:$src1, I32:$src2)>;
+
+let hasNewValue = 1, opNewValue = 0 in
+class T_MType_mpy_ri <bit isNeg, Operand ImmOp, list<dag> pattern>
+ : MInst < (outs IntRegs:$Rd), (ins IntRegs:$Rs, ImmOp:$u8),
+ "$Rd ="#!if(isNeg, "- ", "+ ")#"mpyi($Rs, #$u8)" ,
+ pattern, "", M_tc_3x_SLOT23> {
+ bits<5> Rd;
+ bits<5> Rs;
+ bits<8> u8;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b0000;
+ let Inst{23} = isNeg;
+ let Inst{13} = 0b0;
+ let Inst{4-0} = Rd;
+ let Inst{20-16} = Rs;
+ let Inst{12-5} = u8;
+ }
+
+let isExtendable = 1, opExtentBits = 8, opExtendable = 2, isCodeGenOnly = 0 in
+def M2_mpysip : T_MType_mpy_ri <0, u8Ext,
+ [(set (i32 IntRegs:$Rd), (mul IntRegs:$Rs, u8ExtPred:$u8))]>;
+
+let isCodeGenOnly = 0 in
+def M2_mpysin : T_MType_mpy_ri <1, u8Imm,
+ [(set (i32 IntRegs:$Rd), (ineg (mul IntRegs:$Rs,
+ u8ImmPred:$u8)))]>;
+
+// Assember mapped to M2_mpyi
+let isAsmParserOnly = 1 in
+def M2_mpyui : MInst<(outs IntRegs:$dst),
+ (ins IntRegs:$src1, IntRegs:$src2),
+ "$dst = mpyui($src1, $src2)">;
// Rd=mpyi(Rs,#m9)
// s9 is NOT the same as m9 - but it works.. so far.
-// Assembler maps to either Rd=+mpyi(Rs,#u8 or Rd=-mpyi(Rs,#u8)
+// Assembler maps to either Rd=+mpyi(Rs,#u8) or Rd=-mpyi(Rs,#u8)
// depending on the value of m9. See Arch Spec.
let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 9,
-CextOpcode = "MPYI", InputType = "imm" in
-def MPYI_ri : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s9Ext:$src2),
- "$dst = mpyi($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
- s9ExtPred:$src2))]>, ImmRegRel;
-
-// Rd=mpyi(Rs,Rt)
-let CextOpcode = "MPYI", InputType = "reg" in
-def MPYI : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = mpyi($src1, $src2)",
- [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
- (i32 IntRegs:$src2)))]>, ImmRegRel;
-
-// Rx+=mpyi(Rs,#u8)
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 8,
-CextOpcode = "MPYI_acc", InputType = "imm" in
-def MPYI_acc_ri : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs:$src2, u8Ext:$src3),
- "$dst += mpyi($src2, #$src3)",
- [(set (i32 IntRegs:$dst),
- (add (mul (i32 IntRegs:$src2), u8ExtPred:$src3),
- (i32 IntRegs:$src1)))],
- "$src1 = $dst">, ImmRegRel;
+ CextOpcode = "mpyi", InputType = "imm", hasNewValue = 1 in
+def M2_mpysmi : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s9Ext:$src2),
+ "$dst = mpyi($src1, #$src2)",
+ [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
+ s9ExtPred:$src2))]>, ImmRegRel;
+
+let hasNewValue = 1, isExtendable = 1, opExtentBits = 8, opExtendable = 3,
+ InputType = "imm" in
+class T_MType_acc_ri <string mnemonic, bits<3> MajOp, Operand ImmOp,
+ list<dag> pattern = []>
+ : MInst < (outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, ImmOp:$src3),
+ "$dst "#mnemonic#"($src2, #$src3)",
+ pattern, "$src1 = $dst", M_tc_2_SLOT23> {
+ bits<5> dst;
+ bits<5> src2;
+ bits<8> src3;
+
+ let IClass = 0b1110;
+
+ let Inst{27-26} = 0b00;
+ let Inst{25-23} = MajOp;
+ let Inst{20-16} = src2;
+ let Inst{13} = 0b0;
+ let Inst{12-5} = src3;
+ let Inst{4-0} = dst;
+ }
-// Rx+=mpyi(Rs,Rt)
-let CextOpcode = "MPYI_acc", InputType = "reg" in
-def MPYI_acc_rr : MInst_acc<(outs IntRegs:$dst),
+let InputType = "reg", hasNewValue = 1 in
+class T_MType_acc_rr <string mnemonic, bits<3> MajOp, bits<3> MinOp,
+ bit isSwap = 0, list<dag> pattern = [], bit hasNot = 0,
+ bit isSat = 0, bit isShift = 0>
+ : MInst < (outs IntRegs:$dst),
(ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
- "$dst += mpyi($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
- (i32 IntRegs:$src1)))],
- "$src1 = $dst">, ImmRegRel;
-
-// Rx-=mpyi(Rs,#u8)
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 8 in
-def MPYI_sub_ri : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs:$src2, u8Ext:$src3),
- "$dst -= mpyi($src2, #$src3)",
- [(set (i32 IntRegs:$dst),
- (sub (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
- u8ExtPred:$src3)))],
- "$src1 = $dst">;
+ "$dst "#mnemonic#"($src2, "#!if(hasNot, "~$src3)","$src3)")
+ #!if(isShift, ":<<1", "")
+ #!if(isSat, ":sat", ""),
+ pattern, "$src1 = $dst", M_tc_2_SLOT23 > {
+ bits<5> dst;
+ bits<5> src2;
+ bits<5> src3;
-// Multiply and use upper result.
-// Rd=mpy(Rs,Rt.H):<<1:rnd:sat
-// Rd=mpy(Rs,Rt.L):<<1:rnd:sat
-// Rd=mpy(Rs,Rt)
-def MPY : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = mpy($src1, $src2)",
- [(set (i32 IntRegs:$dst), (mulhs (i32 IntRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-// Rd=mpy(Rs,Rt):rnd
-// Rd=mpyu(Rs,Rt)
-def MPYU : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = mpyu($src1, $src2)",
- [(set (i32 IntRegs:$dst), (mulhu (i32 IntRegs:$src1),
- (i32 IntRegs:$src2)))]>;
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b1111;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = !if(isSwap, src3, src2);
+ let Inst{13} = 0b0;
+ let Inst{12-8} = !if(isSwap, src2, src3);
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = dst;
+ }
+
+let CextOpcode = "MPYI_acc", Itinerary = M_tc_3x_SLOT23, isCodeGenOnly = 0 in {
+ def M2_macsip : T_MType_acc_ri <"+= mpyi", 0b010, u8Ext,
+ [(set (i32 IntRegs:$dst),
+ (add (mul IntRegs:$src2, u8ExtPred:$src3),
+ IntRegs:$src1))]>, ImmRegRel;
+
+ def M2_maci : T_MType_acc_rr <"+= mpyi", 0b000, 0b000, 0,
+ [(set (i32 IntRegs:$dst),
+ (add (mul IntRegs:$src2, IntRegs:$src3),
+ IntRegs:$src1))]>, ImmRegRel;
+}
+
+let CextOpcode = "ADD_acc", isCodeGenOnly = 0 in {
+ let isExtentSigned = 1 in
+ def M2_accii : T_MType_acc_ri <"+= add", 0b100, s8Ext,
+ [(set (i32 IntRegs:$dst),
+ (add (add (i32 IntRegs:$src2), s8_16ExtPred:$src3),
+ (i32 IntRegs:$src1)))]>, ImmRegRel;
+
+ def M2_acci : T_MType_acc_rr <"+= add", 0b000, 0b001, 0,
+ [(set (i32 IntRegs:$dst),
+ (add (add (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
+ (i32 IntRegs:$src1)))]>, ImmRegRel;
+}
+
+let CextOpcode = "SUB_acc", isCodeGenOnly = 0 in {
+ let isExtentSigned = 1 in
+ def M2_naccii : T_MType_acc_ri <"-= add", 0b101, s8Ext>, ImmRegRel;
+
+ def M2_nacci : T_MType_acc_rr <"-= add", 0b100, 0b001, 0>, ImmRegRel;
+}
+
+let Itinerary = M_tc_3x_SLOT23, isCodeGenOnly = 0 in
+def M2_macsin : T_MType_acc_ri <"-= mpyi", 0b011, u8Ext>;
+
+let isCodeGenOnly = 0 in {
+def M2_xor_xacc : T_MType_acc_rr < "^= xor", 0b100, 0b011, 0>;
+def M2_subacc : T_MType_acc_rr <"+= sub", 0b000, 0b011, 1>;
+}
+
+class T_MType_acc_pat1 <InstHexagon MI, SDNode firstOp, SDNode secOp,
+ PatLeaf ImmPred>
+ : Pat <(secOp IntRegs:$src1, (firstOp IntRegs:$src2, ImmPred:$src3)),
+ (MI IntRegs:$src1, IntRegs:$src2, ImmPred:$src3)>;
+
+class T_MType_acc_pat2 <InstHexagon MI, SDNode firstOp, SDNode secOp>
+ : Pat <(i32 (secOp IntRegs:$src1, (firstOp IntRegs:$src2, IntRegs:$src3))),
+ (MI IntRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
+
+def : T_MType_acc_pat2 <M2_xor_xacc, xor, xor>;
+def : T_MType_acc_pat1 <M2_macsin, mul, sub, u8ExtPred>;
+
+def : T_MType_acc_pat1 <M2_naccii, add, sub, s8_16ExtPred>;
+def : T_MType_acc_pat2 <M2_nacci, add, sub>;
+//===----------------------------------------------------------------------===//
+// Template Class -- Multiply signed/unsigned halfwords with and without
+// saturation and rounding
+//===----------------------------------------------------------------------===//
+class T_M2_mpyd < bits<2> LHbits, bit isRnd, bit hasShift, bit isUnsigned >
+ : MInst < (outs DoubleRegs:$Rdd), (ins IntRegs:$Rs, IntRegs:$Rt),
+ "$Rdd = "#!if(isUnsigned,"mpyu","mpy")#"($Rs."#!if(LHbits{1},"h","l")
+ #", $Rt."#!if(LHbits{0},"h)","l)")
+ #!if(hasShift,":<<1","")
+ #!if(isRnd,":rnd",""),
+ [] > {
+ bits<5> Rdd;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b0100;
+ let Inst{23} = hasShift;
+ let Inst{22} = isUnsigned;
+ let Inst{21} = isRnd;
+ let Inst{6-5} = LHbits;
+ let Inst{4-0} = Rdd;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+}
+
+let isCodeGenOnly = 0 in {
+def M2_mpyd_hh_s0: T_M2_mpyd<0b11, 0, 0, 0>;
+def M2_mpyd_hl_s0: T_M2_mpyd<0b10, 0, 0, 0>;
+def M2_mpyd_lh_s0: T_M2_mpyd<0b01, 0, 0, 0>;
+def M2_mpyd_ll_s0: T_M2_mpyd<0b00, 0, 0, 0>;
+
+def M2_mpyd_hh_s1: T_M2_mpyd<0b11, 0, 1, 0>;
+def M2_mpyd_hl_s1: T_M2_mpyd<0b10, 0, 1, 0>;
+def M2_mpyd_lh_s1: T_M2_mpyd<0b01, 0, 1, 0>;
+def M2_mpyd_ll_s1: T_M2_mpyd<0b00, 0, 1, 0>;
+
+def M2_mpyd_rnd_hh_s0: T_M2_mpyd<0b11, 1, 0, 0>;
+def M2_mpyd_rnd_hl_s0: T_M2_mpyd<0b10, 1, 0, 0>;
+def M2_mpyd_rnd_lh_s0: T_M2_mpyd<0b01, 1, 0, 0>;
+def M2_mpyd_rnd_ll_s0: T_M2_mpyd<0b00, 1, 0, 0>;
+
+def M2_mpyd_rnd_hh_s1: T_M2_mpyd<0b11, 1, 1, 0>;
+def M2_mpyd_rnd_hl_s1: T_M2_mpyd<0b10, 1, 1, 0>;
+def M2_mpyd_rnd_lh_s1: T_M2_mpyd<0b01, 1, 1, 0>;
+def M2_mpyd_rnd_ll_s1: T_M2_mpyd<0b00, 1, 1, 0>;
+
+//Rdd=mpyu(Rs.[HL],Rt.[HL])[:<<1]
+def M2_mpyud_hh_s0: T_M2_mpyd<0b11, 0, 0, 1>;
+def M2_mpyud_hl_s0: T_M2_mpyd<0b10, 0, 0, 1>;
+def M2_mpyud_lh_s0: T_M2_mpyd<0b01, 0, 0, 1>;
+def M2_mpyud_ll_s0: T_M2_mpyd<0b00, 0, 0, 1>;
+
+def M2_mpyud_hh_s1: T_M2_mpyd<0b11, 0, 1, 1>;
+def M2_mpyud_hl_s1: T_M2_mpyd<0b10, 0, 1, 1>;
+def M2_mpyud_lh_s1: T_M2_mpyd<0b01, 0, 1, 1>;
+def M2_mpyud_ll_s1: T_M2_mpyd<0b00, 0, 1, 1>;
+}
+//===----------------------------------------------------------------------===//
+// Template Class for xtype mpy:
+// Vector multiply
+// Complex multiply
+// multiply 32X32 and use full result
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0 in
+class T_XTYPE_mpy64 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
+ bit isSat, bit hasShift, bit isConj>
+ : MInst <(outs DoubleRegs:$Rdd),
+ (ins IntRegs:$Rs, IntRegs:$Rt),
+ "$Rdd = "#mnemonic#"($Rs, $Rt"#!if(isConj,"*)",")")
+ #!if(hasShift,":<<1","")
+ #!if(isSat,":sat",""),
+ [] > {
+ bits<5> Rdd;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b0101;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rdd;
+ }
+
+//===----------------------------------------------------------------------===//
+// Template Class for xtype mpy with accumulation into 64-bit:
+// Vector multiply
+// Complex multiply
+// multiply 32X32 and use full result
+//===----------------------------------------------------------------------===//
+class T_XTYPE_mpy64_acc <string op1, string op2, bits<3> MajOp, bits<3> MinOp,
+ bit isSat, bit hasShift, bit isConj>
+ : MInst <(outs DoubleRegs:$Rxx),
+ (ins DoubleRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
+ "$Rxx "#op2#"= "#op1#"($Rs, $Rt"#!if(isConj,"*)",")")
+ #!if(hasShift,":<<1","")
+ #!if(isSat,":sat",""),
+
+ [] , "$dst2 = $Rxx" > {
+ bits<5> Rxx;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b0111;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rxx;
+ }
+
+// MPY - Multiply and use full result
+// Rdd = mpy[u](Rs,Rt)
+let isCodeGenOnly = 0 in {
+def M2_dpmpyss_s0 : T_XTYPE_mpy64 < "mpy", 0b000, 0b000, 0, 0, 0>;
+def M2_dpmpyuu_s0 : T_XTYPE_mpy64 < "mpyu", 0b010, 0b000, 0, 0, 0>;
+
+// Rxx[+-]= mpy[u](Rs,Rt)
+def M2_dpmpyss_acc_s0 : T_XTYPE_mpy64_acc < "mpy", "+", 0b000, 0b000, 0, 0, 0>;
+def M2_dpmpyss_nac_s0 : T_XTYPE_mpy64_acc < "mpy", "-", 0b001, 0b000, 0, 0, 0>;
+def M2_dpmpyuu_acc_s0 : T_XTYPE_mpy64_acc < "mpyu", "+", 0b010, 0b000, 0, 0, 0>;
+def M2_dpmpyuu_nac_s0 : T_XTYPE_mpy64_acc < "mpyu", "-", 0b011, 0b000, 0, 0, 0>;
+}
+
+def: Pat<(i64 (mul (i64 (anyext (i32 IntRegs:$src1))),
+ (i64 (anyext (i32 IntRegs:$src2))))),
+ (M2_dpmpyuu_s0 IntRegs:$src1, IntRegs:$src2)>;
+
+def: Pat<(i64 (mul (i64 (sext (i32 IntRegs:$src1))),
+ (i64 (sext (i32 IntRegs:$src2))))),
+ (M2_dpmpyss_s0 IntRegs:$src1, IntRegs:$src2)>;
-// Multiply and use full result.
-// Rdd=mpyu(Rs,Rt)
-def MPYU64 : MInst<(outs DoubleRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = mpyu($src1, $src2)",
- [(set (i64 DoubleRegs:$dst),
- (mul (i64 (anyext (i32 IntRegs:$src1))),
- (i64 (anyext (i32 IntRegs:$src2)))))]>;
-
-// Rdd=mpy(Rs,Rt)
-def MPY64 : MInst<(outs DoubleRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = mpy($src1, $src2)",
- [(set (i64 DoubleRegs:$dst),
- (mul (i64 (sext (i32 IntRegs:$src1))),
- (i64 (sext (i32 IntRegs:$src2)))))]>;
+def: Pat<(i64 (mul (is_sext_i32:$src1),
+ (is_sext_i32:$src2))),
+ (M2_dpmpyss_s0 (LoReg DoubleRegs:$src1), (LoReg DoubleRegs:$src2))>;
// Multiply and accumulate, use full result.
// Rxx[+-]=mpy(Rs,Rt)
-// Rxx+=mpy(Rs,Rt)
-def MPY64_acc : MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3),
- "$dst += mpy($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (add (mul (i64 (sext (i32 IntRegs:$src2))),
- (i64 (sext (i32 IntRegs:$src3)))),
- (i64 DoubleRegs:$src1)))],
- "$src1 = $dst">;
-// Rxx-=mpy(Rs,Rt)
-def MPY64_sub : MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3),
- "$dst -= mpy($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (sub (i64 DoubleRegs:$src1),
- (mul (i64 (sext (i32 IntRegs:$src2))),
- (i64 (sext (i32 IntRegs:$src3))))))],
- "$src1 = $dst">;
+def: Pat<(i64 (add (i64 DoubleRegs:$src1),
+ (mul (i64 (sext (i32 IntRegs:$src2))),
+ (i64 (sext (i32 IntRegs:$src3)))))),
+ (M2_dpmpyss_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
-// Rxx[+-]=mpyu(Rs,Rt)
-// Rxx+=mpyu(Rs,Rt)
-def MPYU64_acc : MInst_acc<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
- IntRegs:$src2, IntRegs:$src3),
- "$dst += mpyu($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (add (mul (i64 (anyext (i32 IntRegs:$src2))),
- (i64 (anyext (i32 IntRegs:$src3)))),
- (i64 DoubleRegs:$src1)))], "$src1 = $dst">;
-
-// Rxx-=mpyu(Rs,Rt)
-def MPYU64_sub : MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3),
- "$dst -= mpyu($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (sub (i64 DoubleRegs:$src1),
- (mul (i64 (anyext (i32 IntRegs:$src2))),
- (i64 (anyext (i32 IntRegs:$src3))))))],
- "$src1 = $dst">;
+def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
+ (mul (i64 (sext (i32 IntRegs:$src2))),
+ (i64 (sext (i32 IntRegs:$src3)))))),
+ (M2_dpmpyss_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
+
+def: Pat<(i64 (add (i64 DoubleRegs:$src1),
+ (mul (i64 (anyext (i32 IntRegs:$src2))),
+ (i64 (anyext (i32 IntRegs:$src3)))))),
+ (M2_dpmpyuu_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
+
+def: Pat<(i64 (add (i64 DoubleRegs:$src1),
+ (mul (i64 (zext (i32 IntRegs:$src2))),
+ (i64 (zext (i32 IntRegs:$src3)))))),
+ (M2_dpmpyuu_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
+def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
+ (mul (i64 (anyext (i32 IntRegs:$src2))),
+ (i64 (anyext (i32 IntRegs:$src3)))))),
+ (M2_dpmpyuu_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
-let InputType = "reg", CextOpcode = "ADD_acc" in
-def ADDrr_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
- IntRegs:$src2, IntRegs:$src3),
- "$dst += add($src2, $src3)",
- [(set (i32 IntRegs:$dst), (add (add (i32 IntRegs:$src2),
- (i32 IntRegs:$src3)),
- (i32 IntRegs:$src1)))],
- "$src1 = $dst">, ImmRegRel;
-
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 8,
-InputType = "imm", CextOpcode = "ADD_acc" in
-def ADDri_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
- IntRegs:$src2, s8Ext:$src3),
- "$dst += add($src2, #$src3)",
- [(set (i32 IntRegs:$dst), (add (add (i32 IntRegs:$src2),
- s8_16ExtPred:$src3),
- (i32 IntRegs:$src1)))],
- "$src1 = $dst">, ImmRegRel;
-
-let CextOpcode = "SUB_acc", InputType = "reg" in
-def SUBrr_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
- IntRegs:$src2, IntRegs:$src3),
- "$dst -= add($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (sub (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">, ImmRegRel;
-
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 8,
-CextOpcode = "SUB_acc", InputType = "imm" in
-def SUBri_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
- IntRegs:$src2, s8Ext:$src3),
- "$dst -= add($src2, #$src3)",
- [(set (i32 IntRegs:$dst), (sub (i32 IntRegs:$src1),
- (add (i32 IntRegs:$src2),
- s8_16ExtPred:$src3)))],
- "$src1 = $dst">, ImmRegRel;
+def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
+ (mul (i64 (zext (i32 IntRegs:$src2))),
+ (i64 (zext (i32 IntRegs:$src3)))))),
+ (M2_dpmpyuu_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
//===----------------------------------------------------------------------===//
// MTYPE/MPYH -
[(set (i64 DoubleRegs:$dst), (not (i64 DoubleRegs:$src1)))]>;
-// Sign extend word to doubleword.
-def SXTW : ALU64_rr<(outs DoubleRegs:$dst), (ins IntRegs:$src1),
- "$dst = sxtw($src1)",
- [(set (i64 DoubleRegs:$dst), (sext (i32 IntRegs:$src1)))]>;
//===----------------------------------------------------------------------===//
// STYPE/ALU -
//===----------------------------------------------------------------------===//
-//===----------------------------------------------------------------------===//
-// STYPE/BIT +
-//===----------------------------------------------------------------------===//
+let hasSideEffects = 0 in
+class T_S2op_1 <string mnemonic, bits<4> RegTyBits, RegisterClass RCOut,
+ RegisterClass RCIn, bits<2> MajOp, bits<3> MinOp, bit isSat>
+ : SInst <(outs RCOut:$dst), (ins RCIn:$src),
+ "$dst = "#mnemonic#"($src)"#!if(isSat, ":sat", ""),
+ [], "", S_2op_tc_1_SLOT23 > {
+ bits<5> dst;
+ bits<5> src;
-// clrbit.
-def CLRBIT : ALU64_rr<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = clrbit($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (and (i32 IntRegs:$src1),
- (not
- (shl 1, u5ImmPred:$src2))))]>;
+ let IClass = 0b1000;
-def CLRBIT_31 : ALU64_rr<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = clrbit($src1, #$src2)",
- []>;
+ let Inst{27-24} = RegTyBits;
+ let Inst{23-22} = MajOp;
+ let Inst{21} = 0b0;
+ let Inst{20-16} = src;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = dst;
+ }
-// Map from r0 = and(r1, 2147483647) to r0 = clrbit(r1, #31).
-def : Pat <(and (i32 IntRegs:$src1), 2147483647),
- (CLRBIT_31 (i32 IntRegs:$src1), 31)>;
+class T_S2op_1_di <string mnemonic, bits<2> MajOp, bits<3> MinOp>
+ : T_S2op_1 <mnemonic, 0b0100, DoubleRegs, IntRegs, MajOp, MinOp, 0>;
-// setbit.
-def SETBIT : ALU64_rr<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = setbit($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (or (i32 IntRegs:$src1),
- (shl 1, u5ImmPred:$src2)))]>;
+let hasNewValue = 1 in
+class T_S2op_1_id <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit isSat = 0>
+ : T_S2op_1 <mnemonic, 0b1000, IntRegs, DoubleRegs, MajOp, MinOp, isSat>;
-// Map from r0 = or(r1, -2147483648) to r0 = setbit(r1, #31).
-def SETBIT_31 : ALU64_rr<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = setbit($src1, #$src2)",
- []>;
+let hasNewValue = 1 in
+class T_S2op_1_ii <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit isSat = 0>
+ : T_S2op_1 <mnemonic, 0b1100, IntRegs, IntRegs, MajOp, MinOp, isSat>;
-def : Pat <(or (i32 IntRegs:$src1), -2147483648),
- (SETBIT_31 (i32 IntRegs:$src1), 31)>;
+// Sign extend word to doubleword
+let isCodeGenOnly = 0 in
+def A2_sxtw : T_S2op_1_di <"sxtw", 0b01, 0b000>;
-// togglebit.
-def TOGBIT : ALU64_rr<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = setbit($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (xor (i32 IntRegs:$src1),
- (shl 1, u5ImmPred:$src2)))]>;
+def: Pat <(i64 (sext I32:$src)), (A2_sxtw I32:$src)>;
-// Map from r0 = xor(r1, -2147483648) to r0 = togglebit(r1, #31).
-def TOGBIT_31 : ALU64_rr<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = togglebit($src1, #$src2)",
- []>;
+// Swizzle the bytes of a word
+let isCodeGenOnly = 0 in
+def A2_swiz : T_S2op_1_ii <"swiz", 0b10, 0b111>;
+
+// Saturate
+let Defs = [USR_OVF], isCodeGenOnly = 0 in {
+ def A2_sat : T_S2op_1_id <"sat", 0b11, 0b000>;
+ def A2_satb : T_S2op_1_ii <"satb", 0b11, 0b111>;
+ def A2_satub : T_S2op_1_ii <"satub", 0b11, 0b110>;
+ def A2_sath : T_S2op_1_ii <"sath", 0b11, 0b100>;
+ def A2_satuh : T_S2op_1_ii <"satuh", 0b11, 0b101>;
+}
-def : Pat <(xor (i32 IntRegs:$src1), -2147483648),
- (TOGBIT_31 (i32 IntRegs:$src1), 31)>;
+let Itinerary = S_2op_tc_2_SLOT23, isCodeGenOnly = 0 in {
+ // Absolute value word
+ def A2_abs : T_S2op_1_ii <"abs", 0b10, 0b100>;
-//===----------------------------------------------------------------------===//
-// STYPE/BIT -
-//===----------------------------------------------------------------------===//
+ let Defs = [USR_OVF] in
+ def A2_abssat : T_S2op_1_ii <"abs", 0b10, 0b101, 1>;
+
+ // Negate with saturation
+ let Defs = [USR_OVF] in
+ def A2_negsat : T_S2op_1_ii <"neg", 0b10, 0b110, 1>;
+}
+
+def: Pat<(i32 (select (i1 (setlt (i32 IntRegs:$src), 0)),
+ (i32 (sub 0, (i32 IntRegs:$src))),
+ (i32 IntRegs:$src))),
+ (A2_abs IntRegs:$src)>;
+
+let AddedComplexity = 50 in
+def: Pat<(i32 (xor (add (sra (i32 IntRegs:$src), (i32 31)),
+ (i32 IntRegs:$src)),
+ (sra (i32 IntRegs:$src), (i32 31)))),
+ (A2_abs IntRegs:$src)>;
+
+class T_S2op_2 <string mnemonic, bits<4> RegTyBits, RegisterClass RCOut,
+ RegisterClass RCIn, bits<3> MajOp, bits<3> MinOp,
+ bit isSat, bit isRnd, list<dag> pattern = []>
+ : SInst <(outs RCOut:$dst),
+ (ins RCIn:$src, u5Imm:$u5),
+ "$dst = "#mnemonic#"($src, #$u5)"#!if(isSat, ":sat", "")
+ #!if(isRnd, ":rnd", ""),
+ pattern, "", S_2op_tc_2_SLOT23> {
+ bits<5> dst;
+ bits<5> src;
+ bits<5> u5;
+
+ let IClass = 0b1000;
+
+ let Inst{27-24} = RegTyBits;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = src;
+ let Inst{13} = 0b0;
+ let Inst{12-8} = u5;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = dst;
+ }
+
+let hasNewValue = 1 in
+class T_S2op_2_ii <string mnemonic, bits<3> MajOp, bits<3> MinOp,
+ bit isSat = 0, bit isRnd = 0, list<dag> pattern = []>
+ : T_S2op_2 <mnemonic, 0b1100, IntRegs, IntRegs, MajOp, MinOp,
+ isSat, isRnd, pattern>;
+
+class T_S2op_shift <string mnemonic, bits<3> MajOp, bits<3> MinOp, SDNode OpNd>
+ : T_S2op_2_ii <mnemonic, MajOp, MinOp, 0, 0,
+ [(set (i32 IntRegs:$dst), (OpNd (i32 IntRegs:$src),
+ (u5ImmPred:$u5)))]>;
+
+// Arithmetic/logical shift right/left by immediate
+let Itinerary = S_2op_tc_1_SLOT23, isCodeGenOnly = 0 in {
+ def S2_asr_i_r : T_S2op_shift <"asr", 0b000, 0b000, sra>;
+ def S2_lsr_i_r : T_S2op_shift <"lsr", 0b000, 0b001, srl>;
+ def S2_asl_i_r : T_S2op_shift <"asl", 0b000, 0b010, shl>;
+}
+
+// Shift left by immediate with saturation
+let Defs = [USR_OVF], isCodeGenOnly = 0 in
+def S2_asl_i_r_sat : T_S2op_2_ii <"asl", 0b010, 0b010, 1>;
+
+// Shift right with round
+let isCodeGenOnly = 0 in
+def S2_asr_i_r_rnd : T_S2op_2_ii <"asr", 0b010, 0b000, 0, 1>;
+
+def: Pat<(i32 (sra (i32 (add (i32 (sra I32:$src1, u5ImmPred:$src2)),
+ (i32 1))),
+ (i32 1))),
+ (S2_asr_i_r_rnd IntRegs:$src1, u5ImmPred:$src2)>;
+
+class T_S2op_3<string opc, bits<2>MajOp, bits<3>minOp, bits<1> sat = 0>
+ : SInst<(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rss),
+ "$Rdd = "#opc#"($Rss)"#!if(!eq(sat, 1),":sat","")> {
+ bits<5> Rss;
+ bits<5> Rdd;
+ let IClass = 0b1000;
+ let Inst{27-24} = 0;
+ let Inst{23-22} = MajOp;
+ let Inst{20-16} = Rss;
+ let Inst{7-5} = minOp;
+ let Inst{4-0} = Rdd;
+}
+
+let isCodeGenOnly = 0 in {
+def A2_absp : T_S2op_3 <"abs", 0b10, 0b110>;
+def A2_negp : T_S2op_3 <"neg", 0b10, 0b101>;
+def A2_notp : T_S2op_3 <"not", 0b10, 0b100>;
+}
+
+// Innterleave/deinterleave
+let isCodeGenOnly = 0 in {
+def S2_interleave : T_S2op_3 <"interleave", 0b11, 0b101>;
+def S2_deinterleave : T_S2op_3 <"deinterleave", 0b11, 0b100>;
+}
//===----------------------------------------------------------------------===//
-// STYPE/PRED +
+// STYPE/BIT +
//===----------------------------------------------------------------------===//
+// Bit count
-// Predicate transfer.
-let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
-def C2_tfrpr : SInst<(outs IntRegs:$Rd), (ins PredRegs:$Ps),
- "$Rd = $Ps", [], "", S_2op_tc_1_SLOT23> {
+let hasSideEffects = 0, hasNewValue = 1 in
+class T_COUNT_LEADING<string MnOp, bits<3> MajOp, bits<3> MinOp, bit Is32,
+ dag Out, dag Inp>
+ : SInst<Out, Inp, "$Rd = "#MnOp#"($Rs)", [], "", S_2op_tc_1_SLOT23> {
+ bits<5> Rs;
bits<5> Rd;
- bits<2> Ps;
-
let IClass = 0b1000;
- let Inst{27-24} = 0b1001;
- let Inst{22} = 0b1;
- let Inst{17-16} = Ps;
+ let Inst{27} = 0b1;
+ let Inst{26} = Is32;
+ let Inst{25-24} = 0b00;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = Rs;
+ let Inst{7-5} = MinOp;
let Inst{4-0} = Rd;
}
-// Transfer general register to predicate.
-let hasSideEffects = 0, isCodeGenOnly = 0 in
-def C2_tfrrp: SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs),
- "$Pd = $Rs", [], "", S_2op_tc_2early_SLOT23> {
- bits<2> Pd;
- bits<5> Rs;
+class T_COUNT_LEADING_32<string MnOp, bits<3> MajOp, bits<3> MinOp>
+ : T_COUNT_LEADING<MnOp, MajOp, MinOp, 0b1,
+ (outs IntRegs:$Rd), (ins IntRegs:$Rs)>;
+
+class T_COUNT_LEADING_64<string MnOp, bits<3> MajOp, bits<3> MinOp>
+ : T_COUNT_LEADING<MnOp, MajOp, MinOp, 0b0,
+ (outs IntRegs:$Rd), (ins DoubleRegs:$Rs)>;
+
+let isCodeGenOnly = 0 in {
+def S2_cl0 : T_COUNT_LEADING_32<"cl0", 0b000, 0b101>;
+def S2_cl1 : T_COUNT_LEADING_32<"cl1", 0b000, 0b110>;
+def S2_ct0 : T_COUNT_LEADING_32<"ct0", 0b010, 0b100>;
+def S2_ct1 : T_COUNT_LEADING_32<"ct1", 0b010, 0b101>;
+def S2_cl0p : T_COUNT_LEADING_64<"cl0", 0b010, 0b010>;
+def S2_cl1p : T_COUNT_LEADING_64<"cl1", 0b010, 0b100>;
+def S2_clb : T_COUNT_LEADING_32<"clb", 0b000, 0b100>;
+def S2_clbp : T_COUNT_LEADING_64<"clb", 0b010, 0b000>;
+def S2_clbnorm : T_COUNT_LEADING_32<"normamt", 0b000, 0b111>;
+}
+
+def: Pat<(i32 (ctlz I32:$Rs)), (S2_cl0 I32:$Rs)>;
+def: Pat<(i32 (ctlz (not I32:$Rs))), (S2_cl1 I32:$Rs)>;
+def: Pat<(i32 (cttz I32:$Rs)), (S2_ct0 I32:$Rs)>;
+def: Pat<(i32 (cttz (not I32:$Rs))), (S2_ct1 I32:$Rs)>;
+def: Pat<(i32 (trunc (ctlz I64:$Rss))), (S2_cl0p I64:$Rss)>;
+def: Pat<(i32 (trunc (ctlz (not I64:$Rss)))), (S2_cl1p I64:$Rss)>;
+// Bit set/clear/toggle
+
+let hasSideEffects = 0, hasNewValue = 1 in
+class T_SCT_BIT_IMM<string MnOp, bits<3> MinOp>
+ : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, u5Imm:$u5),
+ "$Rd = "#MnOp#"($Rs, #$u5)", [], "", S_2op_tc_1_SLOT23> {
+ bits<5> Rd;
+ bits<5> Rs;
+ bits<5> u5;
let IClass = 0b1000;
- let Inst{27-21} = 0b0101010;
+ let Inst{27-21} = 0b1100110;
let Inst{20-16} = Rs;
- let Inst{1-0} = Pd;
+ let Inst{13} = 0b0;
+ let Inst{12-8} = u5;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rd;
}
+let hasSideEffects = 0, hasNewValue = 1 in
+class T_SCT_BIT_REG<string MnOp, bits<2> MinOp>
+ : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
+ "$Rd = "#MnOp#"($Rs, $Rt)", [], "", S_3op_tc_1_SLOT23> {
+ bits<5> Rd;
+ bits<5> Rs;
+ bits<5> Rt;
+ let IClass = 0b1100;
+ let Inst{27-22} = 0b011010;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ let Inst{7-6} = MinOp;
+ let Inst{4-0} = Rd;
+}
+
+let isCodeGenOnly = 0 in {
+def S2_clrbit_i : T_SCT_BIT_IMM<"clrbit", 0b001>;
+def S2_setbit_i : T_SCT_BIT_IMM<"setbit", 0b000>;
+def S2_togglebit_i : T_SCT_BIT_IMM<"togglebit", 0b010>;
+def S2_clrbit_r : T_SCT_BIT_REG<"clrbit", 0b01>;
+def S2_setbit_r : T_SCT_BIT_REG<"setbit", 0b00>;
+def S2_togglebit_r : T_SCT_BIT_REG<"togglebit", 0b10>;
+}
+
+def: Pat<(i32 (and (i32 IntRegs:$Rs), (not (shl 1, u5ImmPred:$u5)))),
+ (S2_clrbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
+def: Pat<(i32 (or (i32 IntRegs:$Rs), (shl 1, u5ImmPred:$u5))),
+ (S2_setbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
+def: Pat<(i32 (xor (i32 IntRegs:$Rs), (shl 1, u5ImmPred:$u5))),
+ (S2_togglebit_i IntRegs:$Rs, u5ImmPred:$u5)>;
+def: Pat<(i32 (and (i32 IntRegs:$Rs), (not (shl 1, (i32 IntRegs:$Rt))))),
+ (S2_clrbit_r IntRegs:$Rs, IntRegs:$Rt)>;
+def: Pat<(i32 (or (i32 IntRegs:$Rs), (shl 1, (i32 IntRegs:$Rt)))),
+ (S2_setbit_r IntRegs:$Rs, IntRegs:$Rt)>;
+def: Pat<(i32 (xor (i32 IntRegs:$Rs), (shl 1, (i32 IntRegs:$Rt)))),
+ (S2_togglebit_r IntRegs:$Rs, IntRegs:$Rt)>;
+
+// Bit test
+
let hasSideEffects = 0 in
class T_TEST_BIT_IMM<string MnOp, bits<3> MajOp>
: SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, u5Imm:$u5),
let Inst{1-0} = Pd;
}
+let hasSideEffects = 0 in
+class T_TEST_BIT_REG<string MnOp, bit IsNeg>
+ : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
+ "$Pd = "#MnOp#"($Rs, $Rt)",
+ [], "", S_3op_tc_2early_SLOT23> {
+ bits<2> Pd;
+ bits<5> Rs;
+ bits<5> Rt;
+ let IClass = 0b1100;
+ let Inst{27-22} = 0b011100;
+ let Inst{21} = IsNeg;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ let Inst{1-0} = Pd;
+}
+
+let isCodeGenOnly = 0 in {
def S2_tstbit_i : T_TEST_BIT_IMM<"tstbit", 0b000>;
+def S2_tstbit_r : T_TEST_BIT_REG<"tstbit", 0>;
+}
let AddedComplexity = 20 in { // Complexity greater than cmp reg-imm.
+ def: Pat<(i1 (setne (and (shl 1, u5ImmPred:$u5), (i32 IntRegs:$Rs)), 0)),
+ (S2_tstbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
+ def: Pat<(i1 (setne (and (shl 1, (i32 IntRegs:$Rt)), (i32 IntRegs:$Rs)), 0)),
+ (S2_tstbit_r IntRegs:$Rs, IntRegs:$Rt)>;
def: Pat<(i1 (trunc (i32 IntRegs:$Rs))),
(S2_tstbit_i IntRegs:$Rs, 0)>;
+ def: Pat<(i1 (trunc (i64 DoubleRegs:$Rs))),
+ (S2_tstbit_i (LoReg DoubleRegs:$Rs), 0)>;
+}
+let hasSideEffects = 0 in
+class T_TEST_BITS_IMM<string MnOp, bits<2> MajOp, bit IsNeg>
+ : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, u6Imm:$u6),
+ "$Pd = "#MnOp#"($Rs, #$u6)",
+ [], "", S_2op_tc_2early_SLOT23> {
+ bits<2> Pd;
+ bits<5> Rs;
+ bits<6> u6;
+ let IClass = 0b1000;
+ let Inst{27-24} = 0b0101;
+ let Inst{23-22} = MajOp;
+ let Inst{21} = IsNeg;
+ let Inst{20-16} = Rs;
+ let Inst{13-8} = u6;
+ let Inst{1-0} = Pd;
+}
+
+let hasSideEffects = 0 in
+class T_TEST_BITS_REG<string MnOp, bits<2> MajOp, bit IsNeg>
+ : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
+ "$Pd = "#MnOp#"($Rs, $Rt)",
+ [], "", S_3op_tc_2early_SLOT23> {
+ bits<2> Pd;
+ bits<5> Rs;
+ bits<5> Rt;
+ let IClass = 0b1100;
+ let Inst{27-24} = 0b0111;
+ let Inst{23-22} = MajOp;
+ let Inst{21} = IsNeg;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ let Inst{1-0} = Pd;
+}
+
+let isCodeGenOnly = 0 in {
+def C2_bitsclri : T_TEST_BITS_IMM<"bitsclr", 0b10, 0>;
+def C2_bitsclr : T_TEST_BITS_REG<"bitsclr", 0b10, 0>;
+def C2_bitsset : T_TEST_BITS_REG<"bitsset", 0b01, 0>;
+}
+
+let AddedComplexity = 20 in { // Complexity greater than compare reg-imm.
+ def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), u6ImmPred:$u6), 0)),
+ (C2_bitsclri IntRegs:$Rs, u6ImmPred:$u6)>;
+ def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), 0)),
+ (C2_bitsclr IntRegs:$Rs, IntRegs:$Rt)>;
+}
+
+let AddedComplexity = 10 in // Complexity greater than compare reg-reg.
+def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), IntRegs:$Rt)),
+ (C2_bitsset IntRegs:$Rs, IntRegs:$Rt)>;
+
+//===----------------------------------------------------------------------===//
+// STYPE/BIT -
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// STYPE/COMPLEX +
+//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// STYPE/COMPLEX -
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// XTYPE/PERM +
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// XTYPE/PERM -
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// STYPE/PRED +
+//===----------------------------------------------------------------------===//
+
+// Predicate transfer.
+let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
+def C2_tfrpr : SInst<(outs IntRegs:$Rd), (ins PredRegs:$Ps),
+ "$Rd = $Ps", [], "", S_2op_tc_1_SLOT23> {
+ bits<5> Rd;
+ bits<2> Ps;
+
+ let IClass = 0b1000;
+ let Inst{27-24} = 0b1001;
+ let Inst{22} = 0b1;
+ let Inst{17-16} = Ps;
+ let Inst{4-0} = Rd;
+}
+
+// Transfer general register to predicate.
+let hasSideEffects = 0, isCodeGenOnly = 0 in
+def C2_tfrrp: SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs),
+ "$Pd = $Rs", [], "", S_2op_tc_2early_SLOT23> {
+ bits<2> Pd;
+ bits<5> Rs;
+
+ let IClass = 0b1000;
+ let Inst{27-21} = 0b0101010;
+ let Inst{20-16} = Rs;
+ let Inst{1-0} = Pd;
}
//===----------------------------------------------------------------------===//
// STYPE/SHIFT +
//===----------------------------------------------------------------------===//
-// Shift by immediate.
-def ASR_ri : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = asr($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (sra (i32 IntRegs:$src1),
- u5ImmPred:$src2))]>;
+class S_2OpInstImm<string Mnemonic, bits<3>MajOp, bits<3>MinOp,
+ Operand Imm, list<dag> pattern = [], bit isRnd = 0>
+ : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, Imm:$src2),
+ "$dst = "#Mnemonic#"($src1, #$src2)"#!if(isRnd, ":rnd", ""),
+ pattern> {
+ bits<5> src1;
+ bits<5> dst;
+ let IClass = 0b1000;
+ let Inst{27-24} = 0;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = dst;
+}
-def ASRd_ri : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, u6Imm:$src2),
- "$dst = asr($src1, #$src2)",
- [(set (i64 DoubleRegs:$dst), (sra (i64 DoubleRegs:$src1),
- u6ImmPred:$src2))]>;
+class S_2OpInstImmI6<string Mnemonic, SDNode OpNode, bits<3>MinOp>
+ : S_2OpInstImm<Mnemonic, 0b000, MinOp, u6Imm,
+ [(set (i64 DoubleRegs:$dst), (OpNode (i64 DoubleRegs:$src1),
+ u6ImmPred:$src2))]> {
+ bits<6> src2;
+ let Inst{13-8} = src2;
+}
-def ASL : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = asl($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (shl (i32 IntRegs:$src1),
- u5ImmPred:$src2))]>;
+// Shift by immediate.
+let isCodeGenOnly = 0 in {
+def S2_asr_i_p : S_2OpInstImmI6<"asr", sra, 0b000>;
+def S2_asl_i_p : S_2OpInstImmI6<"asl", shl, 0b010>;
+def S2_lsr_i_p : S_2OpInstImmI6<"lsr", srl, 0b001>;
+}
-def ASLd_ri : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, u6Imm:$src2),
- "$dst = asl($src1, #$src2)",
- [(set (i64 DoubleRegs:$dst), (shl (i64 DoubleRegs:$src1),
- u6ImmPred:$src2))]>;
+// Shift left by small amount and add.
+let AddedComplexity = 100, hasNewValue = 1, hasSideEffects = 0,
+ isCodeGenOnly = 0 in
+def S2_addasl_rrri: SInst <(outs IntRegs:$Rd),
+ (ins IntRegs:$Rt, IntRegs:$Rs, u3Imm:$u3),
+ "$Rd = addasl($Rt, $Rs, #$u3)" ,
+ [(set (i32 IntRegs:$Rd), (add (i32 IntRegs:$Rt),
+ (shl (i32 IntRegs:$Rs), u3ImmPred:$u3)))],
+ "", S_3op_tc_2_SLOT23> {
+ bits<5> Rd;
+ bits<5> Rt;
+ bits<5> Rs;
+ bits<3> u3;
-def LSR_ri : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
- "$dst = lsr($src1, #$src2)",
- [(set (i32 IntRegs:$dst), (srl (i32 IntRegs:$src1),
- u5ImmPred:$src2))]>;
+ let IClass = 0b1100;
-def LSRd_ri : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, u6Imm:$src2),
- "$dst = lsr($src1, #$src2)",
- [(set (i64 DoubleRegs:$dst), (srl (i64 DoubleRegs:$src1),
- u6ImmPred:$src2))]>;
+ let Inst{27-21} = 0b0100000;
+ let Inst{20-16} = Rs;
+ let Inst{13} = 0b0;
+ let Inst{12-8} = Rt;
+ let Inst{7-5} = u3;
+ let Inst{4-0} = Rd;
+ }
// Shift by immediate and add.
let AddedComplexity = 100 in
//===----------------------------------------------------------------------===//
// SYSTEM/SUPER -
//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// CRUSER - Type.
+//===----------------------------------------------------------------------===//
+// HW loop
+let isExtendable = 1, isExtentSigned = 1, opExtentBits = 9, opExtentAlign = 2,
+ opExtendable = 0, hasSideEffects = 0 in
+class LOOP_iBase<string mnemonic, Operand brOp, bit mustExtend = 0>
+ : CRInst<(outs), (ins brOp:$offset, u10Imm:$src2),
+ #mnemonic#"($offset, #$src2)",
+ [], "" , CR_tc_3x_SLOT3> {
+ bits<9> offset;
+ bits<10> src2;
+
+ let IClass = 0b0110;
+
+ let Inst{27-22} = 0b100100;
+ let Inst{21} = !if (!eq(mnemonic, "loop0"), 0b0, 0b1);
+ let Inst{20-16} = src2{9-5};
+ let Inst{12-8} = offset{8-4};
+ let Inst{7-5} = src2{4-2};
+ let Inst{4-3} = offset{3-2};
+ let Inst{1-0} = src2{1-0};
+}
+
+let isExtendable = 1, isExtentSigned = 1, opExtentBits = 9, opExtentAlign = 2,
+ opExtendable = 0, hasSideEffects = 0 in
+class LOOP_rBase<string mnemonic, Operand brOp, bit mustExtend = 0>
+ : CRInst<(outs), (ins brOp:$offset, IntRegs:$src2),
+ #mnemonic#"($offset, $src2)",
+ [], "" ,CR_tc_3x_SLOT3> {
+ bits<9> offset;
+ bits<5> src2;
+
+ let IClass = 0b0110;
+
+ let Inst{27-22} = 0b000000;
+ let Inst{21} = !if (!eq(mnemonic, "loop0"), 0b0, 0b1);
+ let Inst{20-16} = src2;
+ let Inst{12-8} = offset{8-4};
+ let Inst{4-3} = offset{3-2};
+ }
+
+multiclass LOOP_ri<string mnemonic> {
+ def i : LOOP_iBase<mnemonic, brtarget>;
+ def r : LOOP_rBase<mnemonic, brtarget>;
+}
+
+
+let Defs = [SA0, LC0, USR], isCodeGenOnly = 0 in
+defm J2_loop0 : LOOP_ri<"loop0">;
+
+// Interestingly only loop0's appear to set usr.lpcfg
+let Defs = [SA1, LC1], isCodeGenOnly = 0 in
+defm J2_loop1 : LOOP_ri<"loop1">;
+
+let isBranch = 1, isTerminator = 1, hasSideEffects = 0,
+ Defs = [PC, LC0], Uses = [SA0, LC0] in {
+def ENDLOOP0 : Endloop<(outs), (ins brtarget:$offset),
+ ":endloop0",
+ []>;
+}
+
+let isBranch = 1, isTerminator = 1, hasSideEffects = 0,
+ Defs = [PC, LC1], Uses = [SA1, LC1] in {
+def ENDLOOP1 : Endloop<(outs), (ins brtarget:$offset),
+ ":endloop1",
+ []>;
+}
+
+// Pipelined loop instructions, sp[123]loop0
+let Defs = [LC0, SA0, P3, USR], hasSideEffects = 0,
+ isExtentSigned = 1, isExtendable = 1, opExtentBits = 9, opExtentAlign = 2,
+ opExtendable = 0, isPredicateLate = 1 in
+class SPLOOP_iBase<string SP, bits<2> op>
+ : CRInst <(outs), (ins brtarget:$r7_2, u10Imm:$U10),
+ "p3 = sp"#SP#"loop0($r7_2, #$U10)" > {
+ bits<9> r7_2;
+ bits<10> U10;
+
+ let IClass = 0b0110;
+
+ let Inst{22-21} = op;
+ let Inst{27-23} = 0b10011;
+ let Inst{20-16} = U10{9-5};
+ let Inst{12-8} = r7_2{8-4};
+ let Inst{7-5} = U10{4-2};
+ let Inst{4-3} = r7_2{3-2};
+ let Inst{1-0} = U10{1-0};
+ }
+
+let Defs = [LC0, SA0, P3, USR], hasSideEffects = 0,
+ isExtentSigned = 1, isExtendable = 1, opExtentBits = 9, opExtentAlign = 2,
+ opExtendable = 0, isPredicateLate = 1 in
+class SPLOOP_rBase<string SP, bits<2> op>
+ : CRInst <(outs), (ins brtarget:$r7_2, IntRegs:$Rs),
+ "p3 = sp"#SP#"loop0($r7_2, $Rs)" > {
+ bits<9> r7_2;
+ bits<5> Rs;
+
+ let IClass = 0b0110;
+
+ let Inst{22-21} = op;
+ let Inst{27-23} = 0b00001;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = r7_2{8-4};
+ let Inst{4-3} = r7_2{3-2};
+ }
+
+multiclass SPLOOP_ri<string mnemonic, bits<2> op> {
+ def i : SPLOOP_iBase<mnemonic, op>;
+ def r : SPLOOP_rBase<mnemonic, op>;
+}
+
+let isCodeGenOnly = 0 in {
+defm J2_ploop1s : SPLOOP_ri<"1", 0b01>;
+defm J2_ploop2s : SPLOOP_ri<"2", 0b10>;
+defm J2_ploop3s : SPLOOP_ri<"3", 0b11>;
+}
+
+// Transfer to/from Control/GPR Guest/GPR
+let hasSideEffects = 0 in
+class TFR_CR_RS_base<RegisterClass CTRC, RegisterClass RC, bit isDouble>
+ : CRInst <(outs CTRC:$dst), (ins RC:$src),
+ "$dst = $src", [], "", CR_tc_3x_SLOT3> {
+ bits<5> dst;
+ bits<5> src;
+
+ let IClass = 0b0110;
+
+ let Inst{27-25} = 0b001;
+ let Inst{24} = isDouble;
+ let Inst{23-21} = 0b001;
+ let Inst{20-16} = src;
+ let Inst{4-0} = dst;
+ }
+let isCodeGenOnly = 0 in
+def A2_tfrrcr : TFR_CR_RS_base<CtrRegs, IntRegs, 0b0>;
+def : InstAlias<"m0 = $Rs", (A2_tfrrcr C6, IntRegs:$Rs)>;
+def : InstAlias<"m1 = $Rs", (A2_tfrrcr C7, IntRegs:$Rs)>;
+
+let hasSideEffects = 0 in
+class TFR_RD_CR_base<RegisterClass RC, RegisterClass CTRC, bit isSingle>
+ : CRInst <(outs RC:$dst), (ins CTRC:$src),
+ "$dst = $src", [], "", CR_tc_3x_SLOT3> {
+ bits<5> dst;
+ bits<5> src;
+
+ let IClass = 0b0110;
+
+ let Inst{27-26} = 0b10;
+ let Inst{25} = isSingle;
+ let Inst{24-21} = 0b0000;
+ let Inst{20-16} = src;
+ let Inst{4-0} = dst;
+ }
+
+let hasNewValue = 1, opNewValue = 0, isCodeGenOnly = 0 in
+def A2_tfrcrr : TFR_RD_CR_base<IntRegs, CtrRegs, 1>;
+def : InstAlias<"$Rd = m0", (A2_tfrcrr IntRegs:$Rd, C6)>;
+def : InstAlias<"$Rd = m1", (A2_tfrcrr IntRegs:$Rd, C7)>;
// TFRI64 - assembly mapped.
let isReMaterializable = 1 in
"$dst = add($src1)",
[(set (i32 IntRegs:$dst), ADDRri:$src1)]>;
-//
-// CR - Type.
-//
-let hasSideEffects = 0, Defs = [SA0, LC0] in {
-def LOOP0_i : CRInst<(outs), (ins brtarget:$offset, u10Imm:$src2),
- "loop0($offset, #$src2)",
- []>;
-}
-
-let hasSideEffects = 0, Defs = [SA0, LC0] in {
-def LOOP0_r : CRInst<(outs), (ins brtarget:$offset, IntRegs:$src2),
- "loop0($offset, $src2)",
- []>;
-}
-
-let isBranch = 1, isTerminator = 1, hasSideEffects = 0,
- Defs = [PC, LC0], Uses = [SA0, LC0] in {
-def ENDLOOP0 : Endloop<(outs), (ins brtarget:$offset),
- ":endloop0",
- []>;
-}
-
// Support for generating global address.
// Taken from X86InstrInfo.td.
def SDTHexagonCONST32 : SDTypeProfile<1, 1, [
"call $dst", []>;
}
-// Call subroutine from register.
-let isCall = 1, hasSideEffects = 0,
- Defs = [D0, D1, D2, D3, D4, D5, D6, D7, D8, D9, D10,
- R22, R23, R28, R31, P0, P1, P2, P3, LC0, LC1, SA0, SA1] in {
- def CALLR : JRInst<(outs), (ins IntRegs:$dst),
- "callr $dst",
- []>;
- }
-
+// Call subroutine indirectly.
+let Defs = VolatileV3.Regs, isCodeGenOnly = 0 in
+def J2_callr : JUMPR_MISC_CALLR<0, 1>;
// Indirect tail-call.
let isCodeGenOnly = 1, isCall = 1, isReturn = 1 in
// Direct tail-calls.
let isCall = 1, isReturn = 1, isBarrier = 1, isPredicable = 0,
isTerminator = 1, isCodeGenOnly = 1 in {
- def TCRETURNtg : T_JMP<(ins calltarget:$dst)>;
- def TCRETURNtext : T_JMP<(ins calltarget:$dst)>;
+ def TCRETURNtg : JInst<(outs), (ins calltarget:$dst), "jump $dst",
+ [], "", J_tc_2early_SLOT23>;
+ def TCRETURNtext : JInst<(outs), (ins calltarget:$dst), "jump $dst",
+ [], "", J_tc_2early_SLOT23>;
}
// Map call instruction.
def : Pat<(call (i32 IntRegs:$dst)),
- (CALLR (i32 IntRegs:$dst))>, Requires<[HasV2TOnly]>;
+ (J2_callr (i32 IntRegs:$dst))>, Requires<[HasV2TOnly]>;
def : Pat<(call tglobaladdr:$dst),
(CALL tglobaladdr:$dst)>, Requires<[HasV2TOnly]>;
def : Pat<(call texternalsym:$dst),
// Map from p0 = pnot(p0); if (p0) jump => if (!p0) jump.
def : Pat <(brcond (not (i1 PredRegs:$src1)), bb:$offset),
- (JMP_f (i1 PredRegs:$src1), bb:$offset)>;
+ (J2_jumpf (i1 PredRegs:$src1), bb:$offset)>;
// Map from p2 = pnot(p2); p1 = and(p0, p2) => p1 = and(p0, !p2).
def : Pat <(and (i1 PredRegs:$src1), (not (i1 PredRegs:$src2))),
def : Pat <(i32 (zextloadi1 ADDRriS11_0:$addr)),
(i32 (A2_and (i32 (LDrib ADDRriS11_0:$addr)), (A2_tfrsi 0x1)))>;
-// Map from Rdd = sign_extend_inreg(Rss, i32) -> Rdd = SXTW(Rss.lo).
+// Map from Rdd = sign_extend_inreg(Rss, i32) -> Rdd = A2_sxtw(Rss.lo).
def : Pat <(i64 (sext_inreg (i64 DoubleRegs:$src1), i32)),
- (i64 (SXTW (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_loreg))))>;
+ (i64 (A2_sxtw (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_loreg))))>;
-// Map from Rdd = sign_extend_inreg(Rss, i16) -> Rdd = SXTW(SXTH(Rss.lo)).
+// Map from Rdd = sign_extend_inreg(Rss, i16) -> Rdd = A2_sxtw(SXTH(Rss.lo)).
def : Pat <(i64 (sext_inreg (i64 DoubleRegs:$src1), i16)),
- (i64 (SXTW (i32 (A2_sxth (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
+ (i64 (A2_sxtw (i32 (A2_sxth (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
subreg_loreg))))))>;
-// Map from Rdd = sign_extend_inreg(Rss, i8) -> Rdd = SXTW(SXTB(Rss.lo)).
+// Map from Rdd = sign_extend_inreg(Rss, i8) -> Rdd = A2_sxtw(SXTB(Rss.lo)).
def : Pat <(i64 (sext_inreg (i64 DoubleRegs:$src1), i8)),
- (i64 (SXTW (i32 (A2_sxtb (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
+ (i64 (A2_sxtw (i32 (A2_sxtb (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
subreg_loreg))))))>;
// We want to prevent emitting pnot's as much as possible.
-// Map brcond with an unsupported setcc to a JMP_f.
+// Map brcond with an unsupported setcc to a J2_jumpf.
def : Pat <(brcond (i1 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
bb:$offset),
- (JMP_f (C2_cmpeq (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
+ (J2_jumpf (C2_cmpeq (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
bb:$offset)>;
def : Pat <(brcond (i1 (setne (i32 IntRegs:$src1), s10ImmPred:$src2)),
bb:$offset),
- (JMP_f (C2_cmpeqi (i32 IntRegs:$src1), s10ImmPred:$src2), bb:$offset)>;
+ (J2_jumpf (C2_cmpeqi (i32 IntRegs:$src1), s10ImmPred:$src2), bb:$offset)>;
def : Pat <(brcond (i1 (setne (i1 PredRegs:$src1), (i1 -1))), bb:$offset),
- (JMP_f (i1 PredRegs:$src1), bb:$offset)>;
+ (J2_jumpf (i1 PredRegs:$src1), bb:$offset)>;
def : Pat <(brcond (i1 (setne (i1 PredRegs:$src1), (i1 0))), bb:$offset),
- (JMP_t (i1 PredRegs:$src1), bb:$offset)>;
+ (J2_jumpt (i1 PredRegs:$src1), bb:$offset)>;
// cmp.lt(Rs, Imm) -> !cmp.ge(Rs, Imm) -> !cmp.gt(Rs, Imm-1)
def : Pat <(brcond (i1 (setlt (i32 IntRegs:$src1), s8ImmPred:$src2)),
bb:$offset),
- (JMP_f (C2_cmpgti (i32 IntRegs:$src1),
+ (J2_jumpf (C2_cmpgti (i32 IntRegs:$src1),
(DEC_CONST_SIGNED s8ImmPred:$src2)), bb:$offset)>;
// cmp.lt(r0, r1) -> cmp.gt(r1, r0)
def : Pat <(brcond (i1 (setlt (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
bb:$offset),
- (JMP_t (C2_cmpgt (i32 IntRegs:$src2), (i32 IntRegs:$src1)), bb:$offset)>;
+ (J2_jumpt (C2_cmpgt (i32 IntRegs:$src2), (i32 IntRegs:$src1)), bb:$offset)>;
def : Pat <(brcond (i1 (setuge (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
bb:$offset),
- (JMP_f (C2_cmpgtup (i64 DoubleRegs:$src2), (i64 DoubleRegs:$src1)),
+ (J2_jumpf (C2_cmpgtup (i64 DoubleRegs:$src2), (i64 DoubleRegs:$src1)),
bb:$offset)>;
def : Pat <(brcond (i1 (setule (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
bb:$offset),
- (JMP_f (C2_cmpgtu (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
+ (J2_jumpf (C2_cmpgtu (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
bb:$offset)>;
def : Pat <(brcond (i1 (setule (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
bb:$offset),
- (JMP_f (C2_cmpgtup (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
+ (J2_jumpf (C2_cmpgtup (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
bb:$offset)>;
// Map from a 64-bit select to an emulated 64-bit mux.
def : Pat<(store (i1 PredRegs:$src1), ADDRriS11_2:$addr),
(STrib ADDRriS11_2:$addr, (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0)) )>;
-// Map Rdd = anyext(Rs) -> Rdd = sxtw(Rs).
+// Map Rdd = anyext(Rs) -> Rdd = A2_sxtw(Rs).
// Hexagon_TODO: We can probably use combine but that will cost 2 instructions.
// Better way to do this?
def : Pat<(i64 (anyext (i32 IntRegs:$src1))),
- (i64 (SXTW (i32 IntRegs:$src1)))>;
+ (i64 (A2_sxtw (i32 IntRegs:$src1)))>;
// Map cmple -> cmpgt.
// rs <= rt -> !(rs > rt).
// Convert sign-extended load back to load and sign extend.
// i8 -> i64
def: Pat <(i64 (sextloadi8 ADDRriS11_0:$src1)),
- (i64 (SXTW (LDrib ADDRriS11_0:$src1)))>;
+ (i64 (A2_sxtw (LDrib ADDRriS11_0:$src1)))>;
// Convert any-extended load back to load and sign extend.
// i8 -> i64
def: Pat <(i64 (extloadi8 ADDRriS11_0:$src1)),
- (i64 (SXTW (LDrib ADDRriS11_0:$src1)))>;
+ (i64 (A2_sxtw (LDrib ADDRriS11_0:$src1)))>;
// Convert sign-extended load back to load and sign extend.
// i16 -> i64
def: Pat <(i64 (sextloadi16 ADDRriS11_1:$src1)),
- (i64 (SXTW (LDrih ADDRriS11_1:$src1)))>;
+ (i64 (A2_sxtw (LDrih ADDRriS11_1:$src1)))>;
// Convert sign-extended load back to load and sign extend.
// i32 -> i64
def: Pat <(i64 (sextloadi32 ADDRriS11_2:$src1)),
- (i64 (SXTW (LDriw ADDRriS11_2:$src1)))>;
+ (i64 (A2_sxtw (LDriw ADDRriS11_2:$src1)))>;
// Zero extends.
def : Pat <(i32 (anyext (i1 PredRegs:$src1))),
(i32 (C2_muxii (i1 PredRegs:$src1), 1, 0))>;
-// Map from Rss = Pd to Rdd = sxtw (mux(Pd, #1, #0))
+// Map from Rss = Pd to Rdd = A2_sxtw (mux(Pd, #1, #0))
def : Pat <(i64 (anyext (i1 PredRegs:$src1))),
- (i64 (SXTW (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0))))>;
+ (i64 (A2_sxtw (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0))))>;
let AddedComplexity = 100 in
// Multiply 64-bit unsigned and use upper result.
def : Pat <(mulhu (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
(i64
- (MPYU64_acc
+ (M2_dpmpyuu_acc_s0
(i64
(A2_combinew
(A2_tfrsi 0),
(i32
(EXTRACT_SUBREG
(i64
- (LSRd_ri
+ (S2_lsr_i_p
(i64
- (MPYU64_acc
+ (M2_dpmpyuu_acc_s0
(i64
- (MPYU64_acc
+ (M2_dpmpyuu_acc_s0
(i64
(A2_combinew (A2_tfrsi 0),
(i32
(EXTRACT_SUBREG
(i64
- (LSRd_ri
+ (S2_lsr_i_p
(i64
- (MPYU64 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
+ (M2_dpmpyuu_s0
+ (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
subreg_loreg)),
(i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2),
subreg_loreg)))), 32)),
// Multiply 64-bit signed and use upper result.
def : Pat <(mulhs (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
(i64
- (MPY64_acc
+ (M2_dpmpyss_acc_s0
(i64
(A2_combinew (A2_tfrsi 0),
(i32
(EXTRACT_SUBREG
(i64
- (LSRd_ri
+ (S2_lsr_i_p
(i64
- (MPY64_acc
+ (M2_dpmpyss_acc_s0
(i64
- (MPY64_acc
+ (M2_dpmpyss_acc_s0
(i64
(A2_combinew (A2_tfrsi 0),
(i32
(EXTRACT_SUBREG
(i64
- (LSRd_ri
+ (S2_lsr_i_p
(i64
- (MPYU64 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
+ (M2_dpmpyuu_s0
+ (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
subreg_loreg)),
(i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2),
subreg_loreg)))), 32)),
// Change the sign of the immediate for Rd=-mpyi(Rs,#u8)
def : Pat <(mul (i32 IntRegs:$src1), (ineg n8ImmPred:$src2)),
- (i32 (MPYI_rin (i32 IntRegs:$src1), u8ImmPred:$src2))>;
+ (i32 (M2_mpysin (i32 IntRegs:$src1), u8ImmPred:$src2))>;
//===----------------------------------------------------------------------===//
// V3 Instructions +
projects / oota-llvm.git / blobdiff