// 64-bit value.
def LoReg: OutPatFrag<(ops node:$Rs),
(EXTRACT_SUBREG (i64 $Rs), subreg_loreg)>;
+def HiReg: OutPatFrag<(ops node:$Rs),
+ (EXTRACT_SUBREG (i64 $Rs), subreg_hireg)>;
+
+// SDNode for converting immediate C to C-1.
+def DEC_CONST_SIGNED : SDNodeXForm<imm, [{
+ // Return the byte immediate const-1 as an SDNode.
+ int32_t imm = N->getSExtValue();
+ return XformSToSM1Imm(imm);
+}]>;
+
+// SDNode for converting immediate C to C-2.
+def DEC2_CONST_SIGNED : SDNodeXForm<imm, [{
+ // Return the byte immediate const-2 as an SDNode.
+ int32_t imm = N->getSExtValue();
+ return XformSToSM2Imm(imm);
+}]>;
+
+// SDNode for converting immediate C to C-3.
+def DEC3_CONST_SIGNED : SDNodeXForm<imm, [{
+ // Return the byte immediate const-3 as an SDNode.
+ int32_t imm = N->getSExtValue();
+ return XformSToSM3Imm(imm);
+}]>;
+
+// SDNode for converting immediate C to C-1.
+def DEC_CONST_UNSIGNED : SDNodeXForm<imm, [{
+ // Return the byte immediate const-1 as an SDNode.
+ uint32_t imm = N->getZExtValue();
+ return XformUToUM1Imm(imm);
+}]>;
//===----------------------------------------------------------------------===//
let Inst{1-0} = dst;
}
+let isCodeGenOnly = 0 in {
def C2_cmpeqi : T_CMP <"cmp.eq", 0b00, 0, s10Ext>;
def C2_cmpgti : T_CMP <"cmp.gt", 0b01, 0, s10Ext>;
def C2_cmpgtui : T_CMP <"cmp.gtu", 0b10, 0, u9Ext>;
+}
class T_CMP_pat <InstHexagon MI, PatFrag OpNode, PatLeaf ImmPred>
: Pat<(i1 (OpNode (i32 IntRegs:$src1), ImmPred:$src2)),
let AsmString = "$Rd = "#mnemonic#"($Rs, $Rt)"#suffix;
}
+let isCodeGenOnly = 0 in {
+def A2_svaddh : T_ALU32_3op<"vaddh", 0b110, 0b000, 0, 1>;
+def A2_svsubh : T_ALU32_3op<"vsubh", 0b110, 0b100, 1, 0>;
+}
+
let Defs = [USR_OVF], Itinerary = ALU32_3op_tc_2_SLOT0123,
isCodeGenOnly = 0 in {
+ def A2_svaddhs : T_ALU32_3op_sfx<"vaddh", ":sat", 0b110, 0b001, 0, 1>;
def A2_addsat : T_ALU32_3op_sfx<"add", ":sat", 0b110, 0b010, 0, 1>;
+ def A2_svadduhs : T_ALU32_3op_sfx<"vadduh", ":sat", 0b110, 0b011, 0, 1>;
+ def A2_svsubhs : T_ALU32_3op_sfx<"vsubh", ":sat", 0b110, 0b101, 1, 0>;
def A2_subsat : T_ALU32_3op_sfx<"sub", ":sat", 0b110, 0b110, 1, 0>;
+ def A2_svsubuhs : T_ALU32_3op_sfx<"vsubuh", ":sat", 0b110, 0b111, 1, 0>;
+}
+
+let Itinerary = ALU32_3op_tc_2_SLOT0123, isCodeGenOnly = 0 in
+def A2_svavghs : T_ALU32_3op_sfx<"vavgh", ":rnd", 0b111, 0b001, 0, 1>;
+
+let isCodeGenOnly = 0 in {
+def A2_svavgh : T_ALU32_3op<"vavgh", 0b111, 0b000, 0, 1>;
+def A2_svnavgh : T_ALU32_3op<"vnavgh", 0b111, 0b011, 1, 0>;
}
multiclass T_ALU32_3op_p<string mnemonic, bits<3> MajOp, bits<3> MinOp,
//===----------------------------------------------------------------------===//
// Template class for predicated ADD of a reg and an Immediate value.
//===----------------------------------------------------------------------===//
-let hasNewValue = 1 in
+let hasNewValue = 1, hasSideEffects = 0 in
class T_Addri_Pred <bit PredNot, bit PredNew>
: ALU32_ri <(outs IntRegs:$Rd),
(ins PredRegs:$Pu, IntRegs:$Rs, s8Ext:$s8),
//===----------------------------------------------------------------------===//
// A2_addi: Add a signed immediate to a register.
//===----------------------------------------------------------------------===//
-let hasNewValue = 1 in
+let hasNewValue = 1, hasSideEffects = 0 in
class T_Addri <Operand immOp, list<dag> pattern = [] >
: ALU32_ri <(outs IntRegs:$Rd),
(ins IntRegs:$Rs, immOp:$s16),
def TFRI64_V4 : ALU64_rr<(outs DoubleRegs:$dst), (ins u6Ext:$src1),
"$dst = #$src1">;
-// Transfer control register.
-let hasSideEffects = 0 in
-def TFCR : CRInst<(outs CRRegs:$dst), (ins IntRegs:$src1),
- "$dst = $src1",
- []>;
//===----------------------------------------------------------------------===//
// ALU32/ALU -
//===----------------------------------------------------------------------===//
def: Pat<(sext_inreg I32:$src1, i8), (A2_sxtb I32:$src1)>;
def: Pat<(sext_inreg I32:$src1, i16), (A2_sxth I32:$src1)>;
-// Mux.
-def VMUX_prr64 : ALU64_rr<(outs DoubleRegs:$dst), (ins PredRegs:$src1,
- DoubleRegs:$src2,
- DoubleRegs:$src3),
- "$dst = vmux($src1, $src2, $src3)",
- []>;
-
-
//===----------------------------------------------------------------------===//
-// ALU32/PERM -
+// Template class for vector add and avg
//===----------------------------------------------------------------------===//
+class T_VectALU_64 <string opc, bits<3> majOp, bits<3> minOp,
+ bit isSat, bit isRnd, bit isCrnd, bit SwapOps >
+ : ALU64_rr < (outs DoubleRegs:$Rdd),
+ (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Rdd = "#opc#"($Rss, $Rtt)"#!if(isRnd, ":rnd", "")
+ #!if(isCrnd,":crnd","")
+ #!if(isSat, ":sat", ""),
+ [], "", ALU64_tc_2_SLOT23 > {
+ bits<5> Rdd;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1101;
+
+ let Inst{27-24} = 0b0011;
+ let Inst{23-21} = majOp;
+ let Inst{20-16} = !if (SwapOps, Rtt, Rss);
+ let Inst{12-8} = !if (SwapOps, Rss, Rtt);
+ let Inst{7-5} = minOp;
+ let Inst{4-0} = Rdd;
+ }
+
+// ALU64 - Vector add
+// Rdd=vadd[u][bhw](Rss,Rtt)
+let Itinerary = ALU64_tc_1_SLOT23, isCodeGenOnly = 0 in {
+ def A2_vaddub : T_VectALU_64 < "vaddub", 0b000, 0b000, 0, 0, 0, 0>;
+ def A2_vaddh : T_VectALU_64 < "vaddh", 0b000, 0b010, 0, 0, 0, 0>;
+ def A2_vaddw : T_VectALU_64 < "vaddw", 0b000, 0b101, 0, 0, 0, 0>;
+}
+
+// Rdd=vadd[u][bhw](Rss,Rtt):sat
+let Defs = [USR_OVF], isCodeGenOnly = 0 in {
+ def A2_vaddubs : T_VectALU_64 < "vaddub", 0b000, 0b001, 1, 0, 0, 0>;
+ def A2_vaddhs : T_VectALU_64 < "vaddh", 0b000, 0b011, 1, 0, 0, 0>;
+ def A2_vadduhs : T_VectALU_64 < "vadduh", 0b000, 0b100, 1, 0, 0, 0>;
+ def A2_vaddws : T_VectALU_64 < "vaddw", 0b000, 0b110, 1, 0, 0, 0>;
+}
+
+// ALU64 - Vector average
+// Rdd=vavg[u][bhw](Rss,Rtt)
+let Itinerary = ALU64_tc_1_SLOT23, isCodeGenOnly = 0 in {
+ def A2_vavgub : T_VectALU_64 < "vavgub", 0b010, 0b000, 0, 0, 0, 0>;
+ def A2_vavgh : T_VectALU_64 < "vavgh", 0b010, 0b010, 0, 0, 0, 0>;
+ def A2_vavguh : T_VectALU_64 < "vavguh", 0b010, 0b101, 0, 0, 0, 0>;
+ def A2_vavgw : T_VectALU_64 < "vavgw", 0b011, 0b000, 0, 0, 0, 0>;
+ def A2_vavguw : T_VectALU_64 < "vavguw", 0b011, 0b011, 0, 0, 0, 0>;
+}
+
+// Rdd=vavg[u][bhw](Rss,Rtt)[:rnd|:crnd]
+let isCodeGenOnly = 0 in {
+def A2_vavgubr : T_VectALU_64 < "vavgub", 0b010, 0b001, 0, 1, 0, 0>;
+def A2_vavghr : T_VectALU_64 < "vavgh", 0b010, 0b011, 0, 1, 0, 0>;
+def A2_vavghcr : T_VectALU_64 < "vavgh", 0b010, 0b100, 0, 0, 1, 0>;
+def A2_vavguhr : T_VectALU_64 < "vavguh", 0b010, 0b110, 0, 1, 0, 0>;
+}
+
+let isCodeGenOnly = 0 in {
+def A2_vavgwr : T_VectALU_64 < "vavgw", 0b011, 0b001, 0, 1, 0, 0>;
+def A2_vavgwcr : T_VectALU_64 < "vavgw", 0b011, 0b010, 0, 0, 1, 0>;
+def A2_vavguwr : T_VectALU_64 < "vavguw", 0b011, 0b100, 0, 1, 0, 0>;
+}
+
+// Rdd=vnavg[bh](Rss,Rtt)
+let Itinerary = ALU64_tc_1_SLOT23, isCodeGenOnly = 0 in {
+ def A2_vnavgh : T_VectALU_64 < "vnavgh", 0b100, 0b000, 0, 0, 0, 1>;
+ def A2_vnavgw : T_VectALU_64 < "vnavgw", 0b100, 0b011, 0, 0, 0, 1>;
+}
+
+// Rdd=vnavg[bh](Rss,Rtt)[:rnd|:crnd]:sat
+let Defs = [USR_OVF], isCodeGenOnly = 0 in {
+ def A2_vnavghr : T_VectALU_64 < "vnavgh", 0b100, 0b001, 1, 1, 0, 1>;
+ def A2_vnavghcr : T_VectALU_64 < "vnavgh", 0b100, 0b010, 1, 0, 1, 1>;
+ def A2_vnavgwr : T_VectALU_64 < "vnavgw", 0b100, 0b100, 1, 1, 0, 1>;
+ def A2_vnavgwcr : T_VectALU_64 < "vnavgw", 0b100, 0b110, 1, 0, 1, 1>;
+}
+
+// Rdd=vsub[u][bh](Rss,Rtt)
+let Itinerary = ALU64_tc_1_SLOT23, isCodeGenOnly = 0 in {
+ def A2_vsubub : T_VectALU_64 < "vsubub", 0b001, 0b000, 0, 0, 0, 1>;
+ def A2_vsubh : T_VectALU_64 < "vsubh", 0b001, 0b010, 0, 0, 0, 1>;
+ def A2_vsubw : T_VectALU_64 < "vsubw", 0b001, 0b101, 0, 0, 0, 1>;
+}
+
+// Rdd=vsub[u][bh](Rss,Rtt):sat
+let Defs = [USR_OVF], isCodeGenOnly = 0 in {
+ def A2_vsububs : T_VectALU_64 < "vsubub", 0b001, 0b001, 1, 0, 0, 1>;
+ def A2_vsubhs : T_VectALU_64 < "vsubh", 0b001, 0b011, 1, 0, 0, 1>;
+ def A2_vsubuhs : T_VectALU_64 < "vsubuh", 0b001, 0b100, 1, 0, 0, 1>;
+ def A2_vsubws : T_VectALU_64 < "vsubw", 0b001, 0b110, 1, 0, 0, 1>;
+}
+
+// Rdd=vmax[u][bhw](Rss,Rtt)
+let isCodeGenOnly = 0 in {
+def A2_vmaxb : T_VectALU_64 < "vmaxb", 0b110, 0b110, 0, 0, 0, 1>;
+def A2_vmaxub : T_VectALU_64 < "vmaxub", 0b110, 0b000, 0, 0, 0, 1>;
+def A2_vmaxh : T_VectALU_64 < "vmaxh", 0b110, 0b001, 0, 0, 0, 1>;
+def A2_vmaxuh : T_VectALU_64 < "vmaxuh", 0b110, 0b010, 0, 0, 0, 1>;
+def A2_vmaxw : T_VectALU_64 < "vmaxw", 0b110, 0b011, 0, 0, 0, 1>;
+def A2_vmaxuw : T_VectALU_64 < "vmaxuw", 0b101, 0b101, 0, 0, 0, 1>;
+}
+
+// Rdd=vmin[u][bhw](Rss,Rtt)
+let isCodeGenOnly = 0 in {
+def A2_vminb : T_VectALU_64 < "vminb", 0b110, 0b111, 0, 0, 0, 1>;
+def A2_vminub : T_VectALU_64 < "vminub", 0b101, 0b000, 0, 0, 0, 1>;
+def A2_vminh : T_VectALU_64 < "vminh", 0b101, 0b001, 0, 0, 0, 1>;
+def A2_vminuh : T_VectALU_64 < "vminuh", 0b101, 0b010, 0, 0, 0, 1>;
+def A2_vminw : T_VectALU_64 < "vminw", 0b101, 0b011, 0, 0, 0, 1>;
+def A2_vminuw : T_VectALU_64 < "vminuw", 0b101, 0b100, 0, 0, 0, 1>;
+}
//===----------------------------------------------------------------------===//
-// ALU32/PRED +
+// Template class for vector compare
//===----------------------------------------------------------------------===//
+let hasSideEffects = 0 in
+class T_vcmp <string Str, bits<4> minOp>
+ : ALU64_rr <(outs PredRegs:$Pd),
+ (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Pd = "#Str#"($Rss, $Rtt)", [],
+ "", ALU64_tc_2early_SLOT23> {
+ bits<2> Pd;
+ bits<5> Rss;
+ bits<5> Rtt;
-// SDNode for converting immediate C to C-1.
-def DEC_CONST_SIGNED : SDNodeXForm<imm, [{
- // Return the byte immediate const-1 as an SDNode.
- int32_t imm = N->getSExtValue();
- return XformSToSM1Imm(imm);
-}]>;
+ let IClass = 0b1101;
-// SDNode for converting immediate C to C-1.
-def DEC_CONST_UNSIGNED : SDNodeXForm<imm, [{
- // Return the byte immediate const-1 as an SDNode.
- uint32_t imm = N->getZExtValue();
- return XformUToUM1Imm(imm);
-}]>;
+ let Inst{27-23} = 0b00100;
+ let Inst{13} = minOp{3};
+ let Inst{7-5} = minOp{2-0};
+ let Inst{1-0} = Pd;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+class T_vcmp_pat<InstHexagon MI, PatFrag Op, ValueType T>
+ : Pat<(i1 (Op (T DoubleRegs:$Rss), (T DoubleRegs:$Rtt))),
+ (i1 (MI DoubleRegs:$Rss, DoubleRegs:$Rtt))>;
-def CTLZ_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1),
- "$dst = cl0($src1)",
- [(set (i32 IntRegs:$dst), (ctlz (i32 IntRegs:$src1)))]>;
+// Vector compare bytes
+def A2_vcmpbeq : T_vcmp <"vcmpb.eq", 0b0110>;
+def A2_vcmpbgtu : T_vcmp <"vcmpb.gtu", 0b0111>;
-def CTTZ_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1),
- "$dst = ct0($src1)",
- [(set (i32 IntRegs:$dst), (cttz (i32 IntRegs:$src1)))]>;
+// Vector compare halfwords
+def A2_vcmpheq : T_vcmp <"vcmph.eq", 0b0011>;
+def A2_vcmphgt : T_vcmp <"vcmph.gt", 0b0100>;
+def A2_vcmphgtu : T_vcmp <"vcmph.gtu", 0b0101>;
-def CTLZ64_rr : SInst<(outs IntRegs:$dst), (ins DoubleRegs:$src1),
- "$dst = cl0($src1)",
- [(set (i32 IntRegs:$dst), (i32 (trunc (ctlz (i64 DoubleRegs:$src1)))))]>;
+// Vector compare words
+def A2_vcmpweq : T_vcmp <"vcmpw.eq", 0b0000>;
+def A2_vcmpwgt : T_vcmp <"vcmpw.gt", 0b0001>;
+def A2_vcmpwgtu : T_vcmp <"vcmpw.gtu", 0b0010>;
-def CTTZ64_rr : SInst<(outs IntRegs:$dst), (ins DoubleRegs:$src1),
- "$dst = ct0($src1)",
- [(set (i32 IntRegs:$dst), (i32 (trunc (cttz (i64 DoubleRegs:$src1)))))]>;
+def: T_vcmp_pat<A2_vcmpbeq, seteq, v8i8>;
+def: T_vcmp_pat<A2_vcmpbgtu, setugt, v8i8>;
+def: T_vcmp_pat<A2_vcmpheq, seteq, v4i16>;
+def: T_vcmp_pat<A2_vcmphgt, setgt, v4i16>;
+def: T_vcmp_pat<A2_vcmphgtu, setugt, v4i16>;
+def: T_vcmp_pat<A2_vcmpweq, seteq, v2i32>;
+def: T_vcmp_pat<A2_vcmpwgt, setgt, v2i32>;
+def: T_vcmp_pat<A2_vcmpwgtu, setugt, v2i32>;
-def TSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = tstbit($src1, $src2)",
- [(set (i1 PredRegs:$dst),
- (setne (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>;
+//===----------------------------------------------------------------------===//
+// ALU32/PERM -
+//===----------------------------------------------------------------------===//
+
+
+//===----------------------------------------------------------------------===//
+// ALU32/PRED +
+//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// ALU32/PRED -
def: T_cmp64_rr_pat<C2_cmpgtp, RevCmp<setlt>>;
def: T_cmp64_rr_pat<C2_cmpgtup, RevCmp<setult>>;
+let isCodeGenOnly = 0 in
+def C2_vmux : ALU64_rr<(outs DoubleRegs:$Rd),
+ (ins PredRegs:$Pu, DoubleRegs:$Rs, DoubleRegs:$Rt),
+ "$Rd = vmux($Pu, $Rs, $Rt)", [], "", ALU64_tc_1_SLOT23> {
+ let hasSideEffects = 0;
+
+ bits<5> Rd;
+ bits<2> Pu;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1101;
+ let Inst{27-24} = 0b0001;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ let Inst{6-5} = Pu;
+ let Inst{4-0} = Rd;
+}
+
class T_ALU64_rr<string mnemonic, string suffix, bits<4> RegType,
bits<3> MajOp, bits<3> MinOp, bit OpsRev, bit IsComm,
string Op2Pfx>
let Inst{4-0} = Rd;
}
-def VALIGN_rrp : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
- DoubleRegs:$src2,
- PredRegs:$src3),
- "$dst = valignb($src1, $src2, $src3)",
- []>;
-
-def VSPLICE_rrp : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
- DoubleRegs:$src2,
- PredRegs:$src3),
- "$dst = vspliceb($src1, $src2, $src3)",
- []>;
-
// User control register transfer.
//===----------------------------------------------------------------------===//
// CR -
//===----------------------------------------------------------------------===//
// LD +
//===----------------------------------------------------------------------===//
-///
-// Load -- MEMri operand
-multiclass LD_MEMri_Pbase<string mnemonic, RegisterClass RC,
- bit isNot, bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME : LDInst2<(outs RC:$dst),
- (ins PredRegs:$src1, MEMri:$addr),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#"$dst = "#mnemonic#"($addr)",
- []>;
-}
-
-multiclass LD_MEMri_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : LD_MEMri_Pbase<mnemonic, RC, PredNot, 0>;
- // Predicate new
- defm _cdn#NAME : LD_MEMri_Pbase<mnemonic, RC, PredNot, 1>;
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, AddedComplexity = 20 in
+class T_load_io <string mnemonic, RegisterClass RC, bits<4> MajOp,
+ Operand ImmOp>
+ : LDInst<(outs RC:$dst), (ins IntRegs:$src1, ImmOp:$offset),
+ "$dst = "#mnemonic#"($src1 + #$offset)", []>, AddrModeRel {
+ bits<4> name;
+ bits<5> dst;
+ bits<5> src1;
+ bits<14> offset;
+ bits<11> offsetBits;
+
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "s11_3Ext"), offset{13-3},
+ !if (!eq(ImmOpStr, "s11_2Ext"), offset{12-2},
+ !if (!eq(ImmOpStr, "s11_1Ext"), offset{11-1},
+ /* s11_0Ext */ offset{10-0})));
+ let opExtentBits = !if (!eq(ImmOpStr, "s11_3Ext"), 14,
+ !if (!eq(ImmOpStr, "s11_2Ext"), 13,
+ !if (!eq(ImmOpStr, "s11_1Ext"), 12,
+ /* s11_0Ext */ 11)));
+ let hasNewValue = !if (!eq(ImmOpStr, "s11_3Ext"), 0, 1);
+
+ let IClass = 0b1001;
+
+ let Inst{27} = 0b0;
+ let Inst{26-25} = offsetBits{10-9};
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{13-5} = offsetBits{8-0};
+ let Inst{4-0} = dst;
}
-}
-let isExtendable = 1, hasSideEffects = 0 in
-multiclass LD_MEMri<string mnemonic, string CextOp, RegisterClass RC,
- bits<5> ImmBits, bits<5> PredImmBits> {
+let opExtendable = 3, isExtentSigned = 0, isPredicated = 1 in
+class T_pload_io <string mnemonic, RegisterClass RC, bits<4>MajOp,
+ Operand ImmOp, bit isNot, bit isPredNew>
+ : LDInst<(outs RC:$dst),
+ (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
+ "if ("#!if(isNot, "!$src1", "$src1")
+ #!if(isPredNew, ".new", "")
+ #") $dst = "#mnemonic#"($src2 + #$offset)",
+ [],"", V2LDST_tc_ld_SLOT01> , AddrModeRel {
+ bits<5> dst;
+ bits<2> src1;
+ bits<5> src2;
+ bits<9> offset;
+ bits<6> offsetBits;
+ string ImmOpStr = !cast<string>(ImmOp);
+
+ let offsetBits = !if (!eq(ImmOpStr, "u6_3Ext"), offset{8-3},
+ !if (!eq(ImmOpStr, "u6_2Ext"), offset{7-2},
+ !if (!eq(ImmOpStr, "u6_1Ext"), offset{6-1},
+ /* u6_0Ext */ offset{5-0})));
+ let opExtentBits = !if (!eq(ImmOpStr, "u6_3Ext"), 9,
+ !if (!eq(ImmOpStr, "u6_2Ext"), 8,
+ !if (!eq(ImmOpStr, "u6_1Ext"), 7,
+ /* u6_0Ext */ 6)));
+ let hasNewValue = !if (!eq(ImmOpStr, "u6_3Ext"), 0, 1);
+ let isPredicatedNew = isPredNew;
+ let isPredicatedFalse = isNot;
+
+ let IClass = 0b0100;
- let CextOpcode = CextOp, BaseOpcode = CextOp in {
- let opExtendable = 2, isExtentSigned = 1, opExtentBits = ImmBits,
- isPredicable = 1 in
- def NAME : LDInst2<(outs RC:$dst), (ins MEMri:$addr),
- "$dst = "#mnemonic#"($addr)",
- []>;
-
- let opExtendable = 3, isExtentSigned = 0, opExtentBits = PredImmBits,
- isPredicated = 1 in {
- defm Pt : LD_MEMri_Pred<mnemonic, RC, 0 >;
- defm NotPt : LD_MEMri_Pred<mnemonic, RC, 1 >;
- }
+ let Inst{27} = 0b0;
+ let Inst{27} = 0b0;
+ let Inst{26} = isNot;
+ let Inst{25} = isPredNew;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = src2;
+ let Inst{13} = 0b0;
+ let Inst{12-11} = src1;
+ let Inst{10-5} = offsetBits;
+ let Inst{4-0} = dst;
}
-}
-
-let addrMode = BaseImmOffset, isMEMri = "true" in {
- let accessSize = ByteAccess in {
- defm LDrib: LD_MEMri < "memb", "LDrib", IntRegs, 11, 6>, AddrModeRel;
- defm LDriub: LD_MEMri < "memub" , "LDriub", IntRegs, 11, 6>, AddrModeRel;
- }
- let accessSize = HalfWordAccess in {
- defm LDrih: LD_MEMri < "memh", "LDrih", IntRegs, 12, 7>, AddrModeRel;
- defm LDriuh: LD_MEMri < "memuh", "LDriuh", IntRegs, 12, 7>, AddrModeRel;
- }
+let isExtendable = 1, hasSideEffects = 0, addrMode = BaseImmOffset in
+multiclass LD_Idxd<string mnemonic, string CextOp, RegisterClass RC,
+ Operand ImmOp, Operand predImmOp, bits<4>MajOp> {
+ let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
+ let isPredicable = 1 in
+ def L2_#NAME#_io : T_load_io <mnemonic, RC, MajOp, ImmOp>;
- let accessSize = WordAccess in
- defm LDriw: LD_MEMri < "memw", "LDriw", IntRegs, 13, 8>, AddrModeRel;
+ // Predicated
+ def L2_p#NAME#t_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 0, 0>;
+ def L2_p#NAME#f_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 1, 0>;
- let accessSize = DoubleWordAccess in
- defm LDrid: LD_MEMri < "memd", "LDrid", DoubleRegs, 14, 9>, AddrModeRel;
+ // Predicated new
+ def L2_p#NAME#tnew_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 0, 1>;
+ def L2_p#NAME#fnew_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 1, 1>;
+ }
}
-def : Pat < (i32 (sextloadi8 ADDRriS11_0:$addr)),
- (LDrib ADDRriS11_0:$addr) >;
-
-def : Pat < (i32 (zextloadi8 ADDRriS11_0:$addr)),
- (LDriub ADDRriS11_0:$addr) >;
-
-def : Pat < (i32 (sextloadi16 ADDRriS11_1:$addr)),
- (LDrih ADDRriS11_1:$addr) >;
-
-def : Pat < (i32 (zextloadi16 ADDRriS11_1:$addr)),
- (LDriuh ADDRriS11_1:$addr) >;
-
-def : Pat < (i32 (load ADDRriS11_2:$addr)),
- (LDriw ADDRriS11_2:$addr) >;
-
-def : Pat < (i64 (load ADDRriS11_3:$addr)),
- (LDrid ADDRriS11_3:$addr) >;
-
-
-// Load - Base with Immediate offset addressing mode
-multiclass LD_Idxd_Pbase2<string mnemonic, RegisterClass RC, Operand predImmOp,
- bit isNot, bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME : LDInst2<(outs RC:$dst),
- (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#"$dst = "#mnemonic#"($src2+#$src3)",
- []>;
+let accessSize = ByteAccess, isCodeGenOnly = 0 in {
+ defm loadrb: LD_Idxd <"memb", "LDrib", IntRegs, s11_0Ext, u6_0Ext, 0b1000>;
+ defm loadrub: LD_Idxd <"memub", "LDriub", IntRegs, s11_0Ext, u6_0Ext, 0b1001>;
}
-multiclass LD_Idxd_Pred2<string mnemonic, RegisterClass RC, Operand predImmOp,
- bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : LD_Idxd_Pbase2<mnemonic, RC, predImmOp, PredNot, 0>;
- // Predicate new
- defm _cdn#NAME : LD_Idxd_Pbase2<mnemonic, RC, predImmOp, PredNot, 1>;
- }
+let accessSize = HalfWordAccess, opExtentAlign = 1, isCodeGenOnly = 0 in {
+ defm loadrh: LD_Idxd <"memh", "LDrih", IntRegs, s11_1Ext, u6_1Ext, 0b1010>;
+ defm loadruh: LD_Idxd <"memuh", "LDriuh", IntRegs, s11_1Ext, u6_1Ext, 0b1011>;
}
-let isExtendable = 1, hasSideEffects = 0 in
-multiclass LD_Idxd2<string mnemonic, string CextOp, RegisterClass RC,
- Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
- bits<5> PredImmBits> {
+let accessSize = WordAccess, opExtentAlign = 2, isCodeGenOnly = 0 in
+defm loadri: LD_Idxd <"memw", "LDriw", IntRegs, s11_2Ext, u6_2Ext, 0b1100>;
- let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
- let opExtendable = 2, isExtentSigned = 1, opExtentBits = ImmBits,
- isPredicable = 1, AddedComplexity = 20 in
- def NAME : LDInst2<(outs RC:$dst), (ins IntRegs:$src1, ImmOp:$offset),
- "$dst = "#mnemonic#"($src1+#$offset)",
- []>;
-
- let opExtendable = 3, isExtentSigned = 0, opExtentBits = PredImmBits,
- isPredicated = 1 in {
- defm Pt : LD_Idxd_Pred2<mnemonic, RC, predImmOp, 0 >;
- defm NotPt : LD_Idxd_Pred2<mnemonic, RC, predImmOp, 1 >;
- }
- }
+let accessSize = DoubleWordAccess, opExtentAlign = 3, isCodeGenOnly = 0 in
+defm loadrd: LD_Idxd <"memd", "LDrid", DoubleRegs, s11_3Ext, u6_3Ext, 0b1110>;
+
+let accessSize = HalfWordAccess, opExtentAlign = 1, isCodeGenOnly = 0 in {
+ def L2_loadbsw2_io: T_load_io<"membh", IntRegs, 0b0001, s11_1Ext>;
+ def L2_loadbzw2_io: T_load_io<"memubh", IntRegs, 0b0011, s11_1Ext>;
}
-let addrMode = BaseImmOffset in {
- let accessSize = ByteAccess in {
- defm LDrib_indexed: LD_Idxd2 <"memb", "LDrib", IntRegs, s11_0Ext, u6_0Ext,
- 11, 6>, AddrModeRel;
- defm LDriub_indexed: LD_Idxd2 <"memub" , "LDriub", IntRegs, s11_0Ext, u6_0Ext,
- 11, 6>, AddrModeRel;
- }
- let accessSize = HalfWordAccess in {
- defm LDrih_indexed: LD_Idxd2 <"memh", "LDrih", IntRegs, s11_1Ext, u6_1Ext,
- 12, 7>, AddrModeRel;
- defm LDriuh_indexed: LD_Idxd2 <"memuh", "LDriuh", IntRegs, s11_1Ext, u6_1Ext,
- 12, 7>, AddrModeRel;
- }
- let accessSize = WordAccess in
- defm LDriw_indexed: LD_Idxd2 <"memw", "LDriw", IntRegs, s11_2Ext, u6_2Ext,
- 13, 8>, AddrModeRel;
+let accessSize = WordAccess, opExtentAlign = 2, isCodeGenOnly = 0 in {
+ def L2_loadbzw4_io: T_load_io<"memubh", DoubleRegs, 0b0101, s11_2Ext>;
+ def L2_loadbsw4_io: T_load_io<"membh", DoubleRegs, 0b0111, s11_2Ext>;
+}
- let accessSize = DoubleWordAccess in
- defm LDrid_indexed: LD_Idxd2 <"memd", "LDrid", DoubleRegs, s11_3Ext, u6_3Ext,
- 14, 9>, AddrModeRel;
+// Patterns to select load-indexed (i.e. load from base+offset).
+multiclass Loadx_pat<PatFrag Load, ValueType VT, PatLeaf ImmPred,
+ InstHexagon MI> {
+ def: Pat<(VT (Load AddrFI:$fi)), (VT (MI AddrFI:$fi, 0))>;
+ def: Pat<(VT (Load (add (i32 IntRegs:$Rs), ImmPred:$Off))),
+ (VT (MI IntRegs:$Rs, imm:$Off))>;
+ def: Pat<(VT (Load (i32 IntRegs:$Rs))), (VT (MI IntRegs:$Rs, 0))>;
}
let AddedComplexity = 20 in {
-def : Pat < (i32 (sextloadi8 (add IntRegs:$src1, s11_0ExtPred:$offset))),
- (LDrib_indexed IntRegs:$src1, s11_0ExtPred:$offset) >;
-
-def : Pat < (i32 (zextloadi8 (add IntRegs:$src1, s11_0ExtPred:$offset))),
- (LDriub_indexed IntRegs:$src1, s11_0ExtPred:$offset) >;
+ defm: Loadx_pat<load, i32, s11_2ExtPred, L2_loadri_io>;
+ defm: Loadx_pat<load, i64, s11_3ExtPred, L2_loadrd_io>;
+ defm: Loadx_pat<atomic_load_8 , i32, s11_0ExtPred, L2_loadrub_io>;
+ defm: Loadx_pat<atomic_load_16, i32, s11_1ExtPred, L2_loadruh_io>;
+ defm: Loadx_pat<atomic_load_32, i32, s11_2ExtPred, L2_loadri_io>;
+ defm: Loadx_pat<atomic_load_64, i64, s11_3ExtPred, L2_loadrd_io>;
+
+ defm: Loadx_pat<extloadi1, i32, s11_0ExtPred, L2_loadrub_io>;
+ defm: Loadx_pat<extloadi8, i32, s11_0ExtPred, L2_loadrub_io>;
+ defm: Loadx_pat<extloadi16, i32, s11_1ExtPred, L2_loadruh_io>;
+ defm: Loadx_pat<sextloadi8, i32, s11_0ExtPred, L2_loadrb_io>;
+ defm: Loadx_pat<sextloadi16, i32, s11_1ExtPred, L2_loadrh_io>;
+ defm: Loadx_pat<zextloadi1, i32, s11_0ExtPred, L2_loadrub_io>;
+ defm: Loadx_pat<zextloadi8, i32, s11_0ExtPred, L2_loadrub_io>;
+ defm: Loadx_pat<zextloadi16, i32, s11_1ExtPred, L2_loadruh_io>;
+ // No sextloadi1.
+}
+
+// Sign-extending loads of i1 need to replicate the lowest bit throughout
+// the 32-bit value. Since the loaded value can only be 0 or 1, 0-v should
+// do the trick.
+let AddedComplexity = 20 in
+def: Pat<(i32 (sextloadi1 (i32 IntRegs:$Rs))),
+ (SUB_ri 0, (L2_loadrub_io IntRegs:$Rs, 0))>;
-def : Pat < (i32 (sextloadi16 (add IntRegs:$src1, s11_1ExtPred:$offset))),
- (LDrih_indexed IntRegs:$src1, s11_1ExtPred:$offset) >;
+//===----------------------------------------------------------------------===//
+// Post increment load
+//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// Template class for non-predicated post increment loads with immediate offset.
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0, addrMode = PostInc in
+class T_load_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<4> MajOp >
+ : LDInstPI <(outs RC:$dst, IntRegs:$dst2),
+ (ins IntRegs:$src1, ImmOp:$offset),
+ "$dst = "#mnemonic#"($src1++#$offset)" ,
+ [],
+ "$src1 = $dst2" > ,
+ PredNewRel {
+ bits<5> dst;
+ bits<5> src1;
+ bits<7> offset;
+ bits<4> offsetBits;
-def : Pat < (i32 (zextloadi16 (add IntRegs:$src1, s11_1ExtPred:$offset))),
- (LDriuh_indexed IntRegs:$src1, s11_1ExtPred:$offset) >;
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
+ !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
+ !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
+ /* s4_0Imm */ offset{3-0})));
+ let hasNewValue = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
-def : Pat < (i32 (load (add IntRegs:$src1, s11_2ExtPred:$offset))),
- (LDriw_indexed IntRegs:$src1, s11_2ExtPred:$offset) >;
+ let IClass = 0b1001;
-def : Pat < (i64 (load (add IntRegs:$src1, s11_3ExtPred:$offset))),
- (LDrid_indexed IntRegs:$src1, s11_3ExtPred:$offset) >;
-}
+ let Inst{27-25} = 0b101;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{13-12} = 0b00;
+ let Inst{8-5} = offsetBits;
+ let Inst{4-0} = dst;
+ }
//===----------------------------------------------------------------------===//
-// Post increment load
+// Template class for predicated post increment loads with immediate offset.
//===----------------------------------------------------------------------===//
+let isPredicated = 1, hasSideEffects = 0, addrMode = PostInc in
+class T_pload_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<4> MajOp, bit isPredNot, bit isPredNew >
+ : LDInst <(outs RC:$dst, IntRegs:$dst2),
+ (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
+ !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+ ") ")#"$dst = "#mnemonic#"($src2++#$offset)",
+ [] ,
+ "$src2 = $dst2" > ,
+ PredNewRel {
+ bits<5> dst;
+ bits<2> src1;
+ bits<5> src2;
+ bits<7> offset;
+ bits<4> offsetBits;
-multiclass LD_PostInc_Pbase<string mnemonic, RegisterClass RC, Operand ImmOp,
- bit isNot, bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME : LDInst2PI<(outs RC:$dst, IntRegs:$dst2),
- (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#"$dst = "#mnemonic#"($src2++#$offset)",
- [],
- "$src2 = $dst2">;
-}
+ let isPredicatedNew = isPredNew;
+ let isPredicatedFalse = isPredNot;
-multiclass LD_PostInc_Pred<string mnemonic, RegisterClass RC,
- Operand ImmOp, bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : LD_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 0>;
- // Predicate new
- let Predicates = [HasV4T], validSubTargets = HasV4SubT in
- defm _cdn#NAME#_V4 : LD_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 1>;
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
+ !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
+ !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
+ /* s4_0Imm */ offset{3-0})));
+ let hasNewValue = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
+
+ let IClass = 0b1001;
+
+ let Inst{27-25} = 0b101;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = src2;
+ let Inst{13} = 0b1;
+ let Inst{12} = isPredNew;
+ let Inst{11} = isPredNot;
+ let Inst{10-9} = src1;
+ let Inst{8-5} = offsetBits;
+ let Inst{4-0} = dst;
}
-}
-multiclass LD_PostInc<string mnemonic, string BaseOp, RegisterClass RC,
- Operand ImmOp> {
+//===----------------------------------------------------------------------===//
+// Multiclass for post increment loads with immediate offset.
+//===----------------------------------------------------------------------===//
+multiclass LD_PostInc <string mnemonic, string BaseOp, RegisterClass RC,
+ Operand ImmOp, bits<4> MajOp> {
let BaseOpcode = "POST_"#BaseOp in {
let isPredicable = 1 in
- def NAME : LDInst2PI<(outs RC:$dst, IntRegs:$dst2),
- (ins IntRegs:$src1, ImmOp:$offset),
- "$dst = "#mnemonic#"($src1++#$offset)",
- [],
- "$src1 = $dst2">;
-
- let isPredicated = 1 in {
- defm Pt : LD_PostInc_Pred<mnemonic, RC, ImmOp, 0 >;
- defm NotPt : LD_PostInc_Pred<mnemonic, RC, ImmOp, 1 >;
- }
+ def L2_#NAME#_pi : T_load_pi < mnemonic, RC, ImmOp, MajOp>;
+
+ // Predicated
+ def L2_p#NAME#t_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 0, 0>;
+ def L2_p#NAME#f_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 1, 0>;
+
+ // Predicated new
+ def L2_p#NAME#tnew_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 0, 1>;
+ def L2_p#NAME#fnew_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 1, 1>;
}
}
-let hasCtrlDep = 1, hasSideEffects = 0, addrMode = PostInc in {
- defm POST_LDrib : LD_PostInc<"memb", "LDrib", IntRegs, s4_0Imm>,
- PredNewRel;
- defm POST_LDriub : LD_PostInc<"memub", "LDriub", IntRegs, s4_0Imm>,
- PredNewRel;
- defm POST_LDrih : LD_PostInc<"memh", "LDrih", IntRegs, s4_1Imm>,
- PredNewRel;
- defm POST_LDriuh : LD_PostInc<"memuh", "LDriuh", IntRegs, s4_1Imm>,
- PredNewRel;
- defm POST_LDriw : LD_PostInc<"memw", "LDriw", IntRegs, s4_2Imm>,
- PredNewRel;
- defm POST_LDrid : LD_PostInc<"memd", "LDrid", DoubleRegs, s4_3Imm>,
- PredNewRel;
+// post increment byte loads with immediate offset
+let accessSize = ByteAccess, isCodeGenOnly = 0 in {
+ defm loadrb : LD_PostInc <"memb", "LDrib", IntRegs, s4_0Imm, 0b1000>;
+ defm loadrub : LD_PostInc <"memub", "LDriub", IntRegs, s4_0Imm, 0b1001>;
}
-def : Pat< (i32 (extloadi1 ADDRriS11_0:$addr)),
- (i32 (LDrib ADDRriS11_0:$addr)) >;
+// post increment halfword loads with immediate offset
+let accessSize = HalfWordAccess, opExtentAlign = 1, isCodeGenOnly = 0 in {
+ defm loadrh : LD_PostInc <"memh", "LDrih", IntRegs, s4_1Imm, 0b1010>;
+ defm loadruh : LD_PostInc <"memuh", "LDriuh", IntRegs, s4_1Imm, 0b1011>;
+}
-// Load byte any-extend.
-def : Pat < (i32 (extloadi8 ADDRriS11_0:$addr)),
- (i32 (LDrib ADDRriS11_0:$addr)) >;
+// post increment word loads with immediate offset
+let accessSize = WordAccess, opExtentAlign = 2, isCodeGenOnly = 0 in
+defm loadri : LD_PostInc <"memw", "LDriw", IntRegs, s4_2Imm, 0b1100>;
-// Indexed load byte any-extend.
-let AddedComplexity = 20 in
-def : Pat < (i32 (extloadi8 (add IntRegs:$src1, s11_0ImmPred:$offset))),
- (i32 (LDrib_indexed IntRegs:$src1, s11_0ImmPred:$offset)) >;
+// post increment doubleword loads with immediate offset
+let accessSize = DoubleWordAccess, opExtentAlign = 3, isCodeGenOnly = 0 in
+defm loadrd : LD_PostInc <"memd", "LDrid", DoubleRegs, s4_3Imm, 0b1110>;
-def : Pat < (i32 (extloadi16 ADDRriS11_1:$addr)),
- (i32 (LDrih ADDRriS11_1:$addr))>;
+// Rd=memb[u]h(Rx++#s4:1)
+// Rdd=memb[u]h(Rx++#s4:2)
+let accessSize = HalfWordAccess, opExtentAlign = 1, isCodeGenOnly = 0 in {
+ def L2_loadbsw2_pi : T_load_pi <"membh", IntRegs, s4_1Imm, 0b0001>;
+ def L2_loadbzw2_pi : T_load_pi <"memubh", IntRegs, s4_1Imm, 0b0011>;
+}
+let accessSize = WordAccess, opExtentAlign = 2, hasNewValue = 0,
+ isCodeGenOnly = 0 in {
+ def L2_loadbsw4_pi : T_load_pi <"membh", DoubleRegs, s4_2Imm, 0b0111>;
+ def L2_loadbzw4_pi : T_load_pi <"memubh", DoubleRegs, s4_2Imm, 0b0101>;
+}
-let AddedComplexity = 20 in
-def : Pat < (i32 (extloadi16 (add IntRegs:$src1, s11_1ImmPred:$offset))),
- (i32 (LDrih_indexed IntRegs:$src1, s11_1ImmPred:$offset)) >;
+//===----------------------------------------------------------------------===//
+// Template class for post increment loads with register offset.
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0, addrMode = PostInc in
+class T_load_pr <string mnemonic, RegisterClass RC, bits<4> MajOp,
+ MemAccessSize AccessSz>
+ : LDInstPI <(outs RC:$dst, IntRegs:$_dst_),
+ (ins IntRegs:$src1, ModRegs:$src2),
+ "$dst = "#mnemonic#"($src1++$src2)" ,
+ [], "$src1 = $_dst_" > {
+ bits<5> dst;
+ bits<5> src1;
+ bits<1> src2;
-let AddedComplexity = 10 in
-def : Pat < (i32 (zextloadi1 ADDRriS11_0:$addr)),
- (i32 (LDriub ADDRriS11_0:$addr))>;
+ let accessSize = AccessSz;
+ let IClass = 0b1001;
-let AddedComplexity = 20 in
-def : Pat < (i32 (zextloadi1 (add IntRegs:$src1, s11_0ImmPred:$offset))),
- (i32 (LDriub_indexed IntRegs:$src1, s11_0ImmPred:$offset))>;
+ let Inst{27-25} = 0b110;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{13} = src2;
+ let Inst{12} = 0b0;
+ let Inst{7} = 0b0;
+ let Inst{4-0} = dst;
+ }
+
+let hasNewValue = 1, isCodeGenOnly = 0 in {
+ def L2_loadrb_pr : T_load_pr <"memb", IntRegs, 0b1000, ByteAccess>;
+ def L2_loadrub_pr : T_load_pr <"memub", IntRegs, 0b1001, ByteAccess>;
+ def L2_loadrh_pr : T_load_pr <"memh", IntRegs, 0b1010, HalfWordAccess>;
+ def L2_loadruh_pr : T_load_pr <"memuh", IntRegs, 0b1011, HalfWordAccess>;
+ def L2_loadri_pr : T_load_pr <"memw", IntRegs, 0b1100, WordAccess>;
+
+ def L2_loadbzw2_pr : T_load_pr <"memubh", IntRegs, 0b0011, HalfWordAccess>;
+}
+
+let isCodeGenOnly = 0 in {
+def L2_loadrd_pr : T_load_pr <"memd", DoubleRegs, 0b1110, DoubleWordAccess>;
+def L2_loadbzw4_pr : T_load_pr <"memubh", DoubleRegs, 0b0101, WordAccess>;
+}
// Load predicate.
let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 13,
"Error; should not emit",
[]>;
-// Deallocate stack frame.
-let Defs = [R29, R30, R31], Uses = [R29], hasSideEffects = 0 in {
- def DEALLOCFRAME : LDInst2<(outs), (ins),
+let Defs = [R29, R30, R31], Uses = [R30], hasSideEffects = 0, isCodeGenOnly = 0 in
+ def L2_deallocframe : LDInst<(outs), (ins),
"deallocframe",
- []>;
+ []> {
+ let IClass = 0b1001;
+
+ let Inst{27-16} = 0b000000011110;
+ let Inst{13} = 0b0;
+ let Inst{4-0} = 0b11110;
}
-// Load and unpack bytes to halfwords.
-//===----------------------------------------------------------------------===//
-// LD -
-//===----------------------------------------------------------------------===//
+// Load / Post increment circular addressing mode.
+let Uses = [CS], hasSideEffects = 0, hasNewValue = 1, opNewValue = 0 in
+class T_load_pcr<string mnemonic, RegisterClass RC, bits<4> MajOp>
+ : LDInst <(outs RC:$dst, IntRegs:$_dst_),
+ (ins IntRegs:$Rz, ModRegs:$Mu),
+ "$dst = "#mnemonic#"($Rz ++ I:circ($Mu))", [],
+ "$Rz = $_dst_" > {
+ bits<5> dst;
+ bits<5> Rz;
+ bit Mu;
-//===----------------------------------------------------------------------===//
-// MTYPE/ALU +
-//===----------------------------------------------------------------------===//
-//===----------------------------------------------------------------------===//
-// MTYPE/ALU -
-//===----------------------------------------------------------------------===//
+ let IClass = 0b1001;
-//===----------------------------------------------------------------------===//
-// MTYPE/COMPLEX +
-//===----------------------------------------------------------------------===//
-//===----------------------------------------------------------------------===//
-// MTYPE/COMPLEX -
-//===----------------------------------------------------------------------===//
+ let Inst{27-25} = 0b100;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12} = 0b0;
+ let Inst{9} = 0b1;
+ let Inst{7} = 0b0;
+ let Inst{4-0} = dst;
+ }
-//===----------------------------------------------------------------------===//
-// MTYPE/MPYH +
-//===----------------------------------------------------------------------===//
+let accessSize = ByteAccess, isCodeGenOnly = 0 in {
+ def L2_loadrb_pcr : T_load_pcr <"memb", IntRegs, 0b1000>;
+ def L2_loadrub_pcr : T_load_pcr <"memub", IntRegs, 0b1001>;
+}
+
+let accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
+ def L2_loadrh_pcr : T_load_pcr <"memh", IntRegs, 0b1010>;
+ def L2_loadruh_pcr : T_load_pcr <"memuh", IntRegs, 0b1011>;
+ def L2_loadbsw2_pcr : T_load_pcr <"membh", IntRegs, 0b0001>;
+ def L2_loadbzw2_pcr : T_load_pcr <"memubh", IntRegs, 0b0011>;
+}
+
+let accessSize = WordAccess, isCodeGenOnly = 0 in {
+ def L2_loadri_pcr : T_load_pcr <"memw", IntRegs, 0b1100>;
+ let hasNewValue = 0 in {
+ def L2_loadbzw4_pcr : T_load_pcr <"memubh", DoubleRegs, 0b0101>;
+ def L2_loadbsw4_pcr : T_load_pcr <"membh", DoubleRegs, 0b0111>;
+ }
+}
+
+let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
+def L2_loadrd_pcr : T_load_pcr <"memd", DoubleRegs, 0b1110>;
//===----------------------------------------------------------------------===//
-// Template Class
-// MPYS / Multipy signed/unsigned halfwords
-//Rd=mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:rnd][:sat]
+// Circular loads with immediate offset.
//===----------------------------------------------------------------------===//
+let Uses = [CS], mayLoad = 1, hasSideEffects = 0, hasNewValue = 1 in
+class T_load_pci <string mnemonic, RegisterClass RC,
+ Operand ImmOp, bits<4> MajOp>
+ : LDInstPI<(outs RC:$dst, IntRegs:$_dst_),
+ (ins IntRegs:$Rz, ImmOp:$offset, ModRegs:$Mu),
+ "$dst = "#mnemonic#"($Rz ++ #$offset:circ($Mu))", [],
+ "$Rz = $_dst_"> {
+ bits<5> dst;
+ bits<5> Rz;
+ bits<1> Mu;
+ bits<7> offset;
+ bits<4> offsetBits;
+
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
+ !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
+ !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
+ /* s4_0Imm */ offset{3-0})));
+ let IClass = 0b1001;
+ let Inst{27-25} = 0b100;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12} = 0b0;
+ let Inst{9} = 0b0;
+ let Inst{8-5} = offsetBits;
+ let Inst{4-0} = dst;
+ }
-let hasNewValue = 1, opNewValue = 0 in
-class T_M2_mpy < bits<2> LHbits, bit isSat, bit isRnd,
- bit hasShift, bit isUnsigned>
- : MInst < (outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
- "$Rd = "#!if(isUnsigned,"mpyu","mpy")#"($Rs."#!if(LHbits{1},"h","l")
+// Byte variants of circ load
+let accessSize = ByteAccess, isCodeGenOnly = 0 in {
+ def L2_loadrb_pci : T_load_pci <"memb", IntRegs, s4_0Imm, 0b1000>;
+ def L2_loadrub_pci : T_load_pci <"memub", IntRegs, s4_0Imm, 0b1001>;
+}
+
+// Half word variants of circ load
+let accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
+ def L2_loadrh_pci : T_load_pci <"memh", IntRegs, s4_1Imm, 0b1010>;
+ def L2_loadruh_pci : T_load_pci <"memuh", IntRegs, s4_1Imm, 0b1011>;
+ def L2_loadbzw2_pci : T_load_pci <"memubh", IntRegs, s4_1Imm, 0b0011>;
+ def L2_loadbsw2_pci : T_load_pci <"membh", IntRegs, s4_1Imm, 0b0001>;
+}
+
+// Word variants of circ load
+let accessSize = WordAccess, isCodeGenOnly = 0 in
+def L2_loadri_pci : T_load_pci <"memw", IntRegs, s4_2Imm, 0b1100>;
+
+let accessSize = WordAccess, hasNewValue = 0, isCodeGenOnly = 0 in {
+ def L2_loadbzw4_pci : T_load_pci <"memubh", DoubleRegs, s4_2Imm, 0b0101>;
+ def L2_loadbsw4_pci : T_load_pci <"membh", DoubleRegs, s4_2Imm, 0b0111>;
+}
+
+let accessSize = DoubleWordAccess, hasNewValue = 0, isCodeGenOnly = 0 in
+def L2_loadrd_pci : T_load_pci <"memd", DoubleRegs, s4_3Imm, 0b1110>;
+
+// L[24]_load[wd]_locked: Load word/double with lock.
+let isSoloAX = 1 in
+class T_load_locked <string mnemonic, RegisterClass RC>
+ : LD0Inst <(outs RC:$dst),
+ (ins IntRegs:$src),
+ "$dst = "#mnemonic#"($src)"> {
+ bits<5> dst;
+ bits<5> src;
+ let IClass = 0b1001;
+ let Inst{27-21} = 0b0010000;
+ let Inst{20-16} = src;
+ let Inst{13-12} = !if (!eq(mnemonic, "memd_locked"), 0b01, 0b00);
+ let Inst{4-0} = dst;
+}
+let hasNewValue = 1, accessSize = WordAccess, opNewValue = 0, isCodeGenOnly = 0 in
+ def L2_loadw_locked : T_load_locked <"memw_locked", IntRegs>;
+let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
+ def L4_loadd_locked : T_load_locked <"memd_locked", DoubleRegs>;
+
+// S[24]_store[wd]_locked: Store word/double conditionally.
+let isSoloAX = 1, isPredicateLate = 1 in
+class T_store_locked <string mnemonic, RegisterClass RC>
+ : ST0Inst <(outs PredRegs:$Pd), (ins IntRegs:$Rs, RC:$Rt),
+ mnemonic#"($Rs, $Pd) = $Rt"> {
+ bits<2> Pd;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1010;
+ let Inst{27-23} = 0b00001;
+ let Inst{22} = !if (!eq(mnemonic, "memw_locked"), 0b0, 0b1);
+ let Inst{21} = 0b1;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ let Inst{1-0} = Pd;
+}
+
+let accessSize = WordAccess, isCodeGenOnly = 0 in
+def S2_storew_locked : T_store_locked <"memw_locked", IntRegs>;
+
+let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
+def S4_stored_locked : T_store_locked <"memd_locked", DoubleRegs>;
+
+//===----------------------------------------------------------------------===//
+// Bit-reversed loads with auto-increment register
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0 in
+class T_load_pbr<string mnemonic, RegisterClass RC,
+ MemAccessSize addrSize, bits<4> majOp>
+ : LDInst
+ <(outs RC:$dst, IntRegs:$_dst_),
+ (ins IntRegs:$Rz, ModRegs:$Mu),
+ "$dst = "#mnemonic#"($Rz ++ $Mu:brev)" ,
+ [] , "$Rz = $_dst_" > {
+
+ let accessSize = addrSize;
+
+ bits<5> dst;
+ bits<5> Rz;
+ bits<1> Mu;
+
+ let IClass = 0b1001;
+
+ let Inst{27-25} = 0b111;
+ let Inst{24-21} = majOp;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12} = 0b0;
+ let Inst{7} = 0b0;
+ let Inst{4-0} = dst;
+ }
+
+let hasNewValue =1, opNewValue = 0, isCodeGenOnly = 0 in {
+ def L2_loadrb_pbr : T_load_pbr <"memb", IntRegs, ByteAccess, 0b1000>;
+ def L2_loadrub_pbr : T_load_pbr <"memub", IntRegs, ByteAccess, 0b1001>;
+ def L2_loadrh_pbr : T_load_pbr <"memh", IntRegs, HalfWordAccess, 0b1010>;
+ def L2_loadruh_pbr : T_load_pbr <"memuh", IntRegs, HalfWordAccess, 0b1011>;
+ def L2_loadbsw2_pbr : T_load_pbr <"membh", IntRegs, HalfWordAccess, 0b0001>;
+ def L2_loadbzw2_pbr : T_load_pbr <"memubh", IntRegs, HalfWordAccess, 0b0011>;
+ def L2_loadri_pbr : T_load_pbr <"memw", IntRegs, WordAccess, 0b1100>;
+}
+
+let isCodeGenOnly = 0 in {
+def L2_loadbzw4_pbr : T_load_pbr <"memubh", DoubleRegs, WordAccess, 0b0101>;
+def L2_loadbsw4_pbr : T_load_pbr <"membh", DoubleRegs, WordAccess, 0b0111>;
+def L2_loadrd_pbr : T_load_pbr <"memd", DoubleRegs, DoubleWordAccess, 0b1110>;
+}
+
+//===----------------------------------------------------------------------===//
+// LD -
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// MTYPE/ALU +
+//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// MTYPE/ALU -
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// MTYPE/COMPLEX +
+//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// MTYPE/COMPLEX -
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// MTYPE/MPYH +
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// 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")
#", $Rt."#!if(LHbits{0},"h)","l)")
#!if(hasShift,":<<1","")
#!if(isRnd,":rnd","")
def M2_mpyud_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 1>;
}
+//===----------------------------------------------------------------------===//
+// Template Class -- Vector Multipy
+// Used for complex multiply real or imaginary, dual multiply and even halfwords
+//===----------------------------------------------------------------------===//
+class T_M2_vmpy < string opc, bits<3> MajOp, bits<3> MinOp, bit hasShift,
+ bit isRnd, bit isSat >
+ : MInst <(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Rdd = "#opc#"($Rss, $Rtt)"#!if(hasShift,":<<1","")
+ #!if(isRnd,":rnd","")
+ #!if(isSat,":sat",""),
+ [] > {
+ bits<5> Rdd;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b1000;
+ let Inst{23-21} = MajOp;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rdd;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+// Vector complex multiply imaginary: Rdd=vcmpyi(Rss,Rtt)[:<<1]:sat
+let Defs = [USR_OVF], isCodeGenOnly = 0 in {
+def M2_vcmpy_s1_sat_i: T_M2_vmpy <"vcmpyi", 0b110, 0b110, 1, 0, 1>;
+def M2_vcmpy_s0_sat_i: T_M2_vmpy <"vcmpyi", 0b010, 0b110, 0, 0, 1>;
+}
+
+// Vector complex multiply real: Rdd=vcmpyr(Rss,Rtt)[:<<1]:sat
+let isCodeGenOnly = 0 in {
+def M2_vcmpy_s1_sat_r: T_M2_vmpy <"vcmpyr", 0b101, 0b110, 1, 0, 1>;
+def M2_vcmpy_s0_sat_r: T_M2_vmpy <"vcmpyr", 0b001, 0b110, 0, 0, 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,
let Inst{4-0} = dst;
}
+class T_MType_vrcmpy <string mnemonic, bits<3> MajOp, bits<3> MinOp, bit isHi>
+ : T_MType_mpy <mnemonic, 0b1001, DoubleRegs, MajOp, MinOp, 1, 1, "", 1, isHi>;
+
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>;
bit isSat = 0, bit isRnd = 0, string op2str = "" >
: T_MType_mpy<mnemonic, 0b1101, IntRegs, MajOp, MinOp, isSat, isRnd, op2str>;
+let isCodeGenOnly = 0 in
+def M2_vradduh : T_MType_dd <"vradduh", 0b000, 0b001, 0, 0>;
+
let CextOpcode = "mpyi", InputType = "reg", isCodeGenOnly = 0 in
def M2_mpyi : T_MType_rr1 <"mpyi", 0b000, 0b000>, ImmRegRel;
def M2_hmmpyl_rs1 : T_MType_rr2 <"mpy", 0b111, 0b100, 1, 1, ".l">;
}
+let isCodeGenOnly = 0 in {
+def M2_cmpyrs_s0 : T_MType_rr2 <"cmpy", 0b001, 0b110, 1, 1>;
+def M2_cmpyrs_s1 : T_MType_rr2 <"cmpy", 0b101, 0b110, 1, 1>;
+def M2_cmpyrsc_s0 : T_MType_rr2 <"cmpy", 0b011, 0b110, 1, 1, "*">;
+def M2_cmpyrsc_s1 : T_MType_rr2 <"cmpy", 0b111, 0b110, 1, 1, "*">;
+}
+
// V4 Instructions
let isCodeGenOnly = 0 in {
+def M2_vraddh : T_MType_dd <"vraddh", 0b001, 0b111, 0>;
def M2_mpysu_up : T_MType_rr1 <"mpysu", 0b011, 0b001, 0>;
+def M2_mpy_up_s1 : T_MType_rr1 <"mpy", 0b101, 0b010, 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 : T_MType_acc_pat1 <M2_naccii, add, sub, s8_16ExtPred>;
def : T_MType_acc_pat2 <M2_nacci, add, sub>;
+
+//===----------------------------------------------------------------------===//
+// Template Class -- XType Vector Instructions
+//===----------------------------------------------------------------------===//
+class T_XTYPE_Vect < string opc, bits<3> MajOp, bits<3> MinOp, bit isConj >
+ : MInst <(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Rdd = "#opc#"($Rss, $Rtt"#!if(isConj,"*)",")"),
+ [] > {
+ bits<5> Rdd;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b1000;
+ let Inst{23-21} = MajOp;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rdd;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+class T_XTYPE_Vect_acc < string opc, bits<3> MajOp, bits<3> MinOp, bit isConj >
+ : MInst <(outs DoubleRegs:$Rdd),
+ (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Rdd += "#opc#"($Rss, $Rtt"#!if(isConj,"*)",")"),
+ [], "$dst2 = $Rdd",M_tc_3x_SLOT23 > {
+ bits<5> Rdd;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b1010;
+ let Inst{23-21} = MajOp;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rdd;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+class T_XTYPE_Vect_diff < bits<3> MajOp, string opc >
+ : MInst <(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rtt, DoubleRegs:$Rss),
+ "$Rdd = "#opc#"($Rtt, $Rss)",
+ [], "",M_tc_2_SLOT23 > {
+ bits<5> Rdd;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b1000;
+ let Inst{23-21} = MajOp;
+ let Inst{7-5} = 0b000;
+ let Inst{4-0} = Rdd;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+// Vector reduce add unsigned bytes: Rdd32=vrmpybu(Rss32,Rtt32)
+let isCodeGenOnly = 0 in {
+def A2_vraddub: T_XTYPE_Vect <"vraddub", 0b010, 0b001, 0>;
+def A2_vraddub_acc: T_XTYPE_Vect_acc <"vraddub", 0b010, 0b001, 0>;
+}
+
+// Vector sum of absolute differences unsigned bytes: Rdd=vrsadub(Rss,Rtt)
+let isCodeGenOnly = 0 in {
+def A2_vrsadub: T_XTYPE_Vect <"vrsadub", 0b010, 0b010, 0>;
+def A2_vrsadub_acc: T_XTYPE_Vect_acc <"vrsadub", 0b010, 0b010, 0>;
+}
+
+// Vector absolute difference: Rdd=vabsdiffh(Rtt,Rss)
+let isCodeGenOnly = 0 in
+def M2_vabsdiffh: T_XTYPE_Vect_diff<0b011, "vabsdiffh">;
+
+// Vector reduce complex multiply real or imaginary:
+// Rdd[+]=vrcmpy[ir](Rss,Rtt[*])
+let isCodeGenOnly = 0 in {
+def M2_vrcmpyi_s0: T_XTYPE_Vect <"vrcmpyi", 0b000, 0b000, 0>;
+def M2_vrcmpyi_s0c: T_XTYPE_Vect <"vrcmpyi", 0b010, 0b000, 1>;
+def M2_vrcmaci_s0: T_XTYPE_Vect_acc <"vrcmpyi", 0b000, 0b000, 0>;
+def M2_vrcmaci_s0c: T_XTYPE_Vect_acc <"vrcmpyi", 0b010, 0b000, 1>;
+}
+
+let isCodeGenOnly = 0 in {
+def M2_vrcmpyr_s0: T_XTYPE_Vect <"vrcmpyr", 0b000, 0b001, 0>;
+def M2_vrcmpyr_s0c: T_XTYPE_Vect <"vrcmpyr", 0b011, 0b001, 1>;
+def M2_vrcmacr_s0: T_XTYPE_Vect_acc <"vrcmpyr", 0b000, 0b001, 0>;
+def M2_vrcmacr_s0c: T_XTYPE_Vect_acc <"vrcmpyr", 0b011, 0b001, 1>;
+}
+
+//===----------------------------------------------------------------------===//
+// Template Class -- Vector Multipy with accumulation.
+// Used for complex multiply real or imaginary, dual multiply and even halfwords
+//===----------------------------------------------------------------------===//
+let Defs = [USR_OVF] in
+class T_M2_vmpy_acc_sat < string opc, bits<3> MajOp, bits<3> MinOp,
+ bit hasShift, bit isRnd >
+ : MInst <(outs DoubleRegs:$Rxx),
+ (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Rxx += "#opc#"($Rss, $Rtt)"#!if(hasShift,":<<1","")
+ #!if(isRnd,":rnd","")#":sat",
+ [], "$dst2 = $Rxx",M_tc_3x_SLOT23 > {
+ bits<5> Rxx;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1110;
+
+ let Inst{27-24} = 0b1010;
+ let Inst{23-21} = MajOp;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rxx;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+// Vector complex multiply real or imaginary with accumulation
+// Rxx+=vcmpy[ir](Rss,Rtt):sat
+let isCodeGenOnly = 0 in {
+def M2_vcmac_s0_sat_r: T_M2_vmpy_acc_sat <"vcmpyr", 0b001, 0b100, 0, 0>;
+def M2_vcmac_s0_sat_i: T_M2_vmpy_acc_sat <"vcmpyi", 0b010, 0b100, 0, 0>;
+}
+
//===----------------------------------------------------------------------===//
// Template Class -- Multiply signed/unsigned halfwords with and without
// saturation and rounding
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>;
}
+// Complex multiply real or imaginary
+// Rxx=cmpy[ir](Rs,Rt)
+let isCodeGenOnly = 0 in {
+def M2_cmpyi_s0 : T_XTYPE_mpy64 < "cmpyi", 0b000, 0b001, 0, 0, 0>;
+def M2_cmpyr_s0 : T_XTYPE_mpy64 < "cmpyr", 0b000, 0b010, 0, 0, 0>;
+}
+
+// Rxx+=cmpy[ir](Rs,Rt)
+let isCodeGenOnly = 0 in {
+def M2_cmaci_s0 : T_XTYPE_mpy64_acc < "cmpyi", "+", 0b000, 0b001, 0, 0, 0>;
+def M2_cmacr_s0 : T_XTYPE_mpy64_acc < "cmpyr", "+", 0b000, 0b010, 0, 0, 0>;
+}
+
+// Complex multiply
+// Rdd=cmpy(Rs,Rt)[:<<]:sat
+let isCodeGenOnly = 0 in {
+def M2_cmpys_s0 : T_XTYPE_mpy64 < "cmpy", 0b000, 0b110, 1, 0, 0>;
+def M2_cmpys_s1 : T_XTYPE_mpy64 < "cmpy", 0b100, 0b110, 1, 1, 0>;
+}
+
+// Rdd=cmpy(Rs,Rt*)[:<<]:sat
+let isCodeGenOnly = 0 in {
+def M2_cmpysc_s0 : T_XTYPE_mpy64 < "cmpy", 0b010, 0b110, 1, 0, 1>;
+def M2_cmpysc_s1 : T_XTYPE_mpy64 < "cmpy", 0b110, 0b110, 1, 1, 1>;
+}
+
+// Rxx[-+]=cmpy(Rs,Rt)[:<<1]:sat
+let isCodeGenOnly = 0 in {
+def M2_cmacs_s0 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b000, 0b110, 1, 0, 0>;
+def M2_cnacs_s0 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b000, 0b111, 1, 0, 0>;
+def M2_cmacs_s1 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b100, 0b110, 1, 1, 0>;
+def M2_cnacs_s1 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b100, 0b111, 1, 1, 0>;
+}
+
+// Rxx[-+]=cmpy(Rs,Rt*)[:<<1]:sat
+let isCodeGenOnly = 0 in {
+def M2_cmacsc_s0 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b010, 0b110, 1, 0, 1>;
+def M2_cnacsc_s0 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b010, 0b111, 1, 0, 1>;
+def M2_cmacsc_s1 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b110, 0b110, 1, 1, 1>;
+def M2_cnacsc_s1 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b110, 0b111, 1, 1, 1>;
+}
def: Pat<(i64 (mul (i64 (anyext (i32 IntRegs:$src1))),
(i64 (anyext (i32 IntRegs:$src2))))),
//===----------------------------------------------------------------------===//
///
// Store doubleword.
-
//===----------------------------------------------------------------------===//
-// Post increment store
+// Template class for non-predicated post increment stores with immediate offset
//===----------------------------------------------------------------------===//
+let isPredicable = 1, hasSideEffects = 0, addrMode = PostInc in
+class T_store_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<4> MajOp, bit isHalf >
+ : STInst <(outs IntRegs:$_dst_),
+ (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
+ mnemonic#"($src1++#$offset) = $src2"#!if(isHalf, ".h", ""),
+ [], "$src1 = $_dst_" >,
+ AddrModeRel {
+ bits<5> src1;
+ bits<5> src2;
+ bits<7> offset;
+ bits<4> offsetBits;
+
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
+ !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
+ !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
+ /* s4_0Imm */ offset{3-0})));
+ let isNVStorable = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
+
+ let IClass = 0b1010;
+
+ let Inst{27-25} = 0b101;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{13} = 0b0;
+ let Inst{12-8} = src2;
+ let Inst{7} = 0b0;
+ let Inst{6-3} = offsetBits;
+ let Inst{1} = 0b0;
+ }
-multiclass ST_PostInc_Pbase<string mnemonic, RegisterClass RC, Operand ImmOp,
- bit isNot, bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME : STInst2PI<(outs IntRegs:$dst),
+//===----------------------------------------------------------------------===//
+// Template class for predicated post increment stores with immediate offset
+//===----------------------------------------------------------------------===//
+let isPredicated = 1, hasSideEffects = 0, addrMode = PostInc in
+class T_pstore_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<4> MajOp, bit isHalf, bit isPredNot, bit isPredNew >
+ : STInst <(outs IntRegs:$_dst_),
(ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#mnemonic#"($src2++#$offset) = $src3",
- [],
- "$src2 = $dst">;
-}
+ !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+ ") ")#mnemonic#"($src2++#$offset) = $src3"#!if(isHalf, ".h", ""),
+ [], "$src2 = $_dst_" >,
+ AddrModeRel {
+ bits<2> src1;
+ bits<5> src2;
+ bits<7> offset;
+ bits<5> src3;
+ bits<4> offsetBits;
-multiclass ST_PostInc_Pred<string mnemonic, RegisterClass RC,
- Operand ImmOp, bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : ST_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 0>;
- // Predicate new
- let Predicates = [HasV4T], validSubTargets = HasV4SubT in
- defm _cdn#NAME#_V4 : ST_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 1>;
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
+ !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
+ !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
+ /* s4_0Imm */ offset{3-0})));
+
+ let isNVStorable = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
+ let isPredicatedNew = isPredNew;
+ let isPredicatedFalse = isPredNot;
+
+ let IClass = 0b1010;
+
+ let Inst{27-25} = 0b101;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = src2;
+ let Inst{13} = 0b1;
+ let Inst{12-8} = src3;
+ let Inst{7} = isPredNew;
+ let Inst{6-3} = offsetBits;
+ let Inst{2} = isPredNot;
+ let Inst{1-0} = src1;
}
-}
-let hasCtrlDep = 1, isNVStorable = 1, hasSideEffects = 0 in
multiclass ST_PostInc<string mnemonic, string BaseOp, RegisterClass RC,
- Operand ImmOp> {
+ Operand ImmOp, bits<4> MajOp, bit isHalf = 0 > {
- let hasCtrlDep = 1, BaseOpcode = "POST_"#BaseOp in {
- let isPredicable = 1 in
- def NAME : STInst2PI<(outs IntRegs:$dst),
- (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
- mnemonic#"($src1++#$offset) = $src2",
- [],
- "$src1 = $dst">;
-
- let isPredicated = 1 in {
- defm Pt : ST_PostInc_Pred<mnemonic, RC, ImmOp, 0 >;
- defm NotPt : ST_PostInc_Pred<mnemonic, RC, ImmOp, 1 >;
- }
+ let BaseOpcode = "POST_"#BaseOp in {
+ def S2_#NAME#_pi : T_store_pi <mnemonic, RC, ImmOp, MajOp, isHalf>;
+
+ // Predicated
+ def S2_p#NAME#t_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp, isHalf, 0, 0>;
+ def S2_p#NAME#f_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp, isHalf, 1, 0>;
+
+ // Predicated new
+ def S2_p#NAME#tnew_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp,
+ isHalf, 0, 1>;
+ def S2_p#NAME#fnew_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp,
+ isHalf, 1, 1>;
}
}
-defm POST_STbri: ST_PostInc <"memb", "STrib", IntRegs, s4_0Imm>, AddrModeRel;
-defm POST_SThri: ST_PostInc <"memh", "STrih", IntRegs, s4_1Imm>, AddrModeRel;
-defm POST_STwri: ST_PostInc <"memw", "STriw", IntRegs, s4_2Imm>, AddrModeRel;
+let accessSize = ByteAccess, isCodeGenOnly = 0 in
+defm storerb: ST_PostInc <"memb", "STrib", IntRegs, s4_0Imm, 0b1000>;
+
+let accessSize = HalfWordAccess, isCodeGenOnly = 0 in
+defm storerh: ST_PostInc <"memh", "STrih", IntRegs, s4_1Imm, 0b1010>;
-let isNVStorable = 0 in
-defm POST_STdri: ST_PostInc <"memd", "STrid", DoubleRegs, s4_3Imm>, AddrModeRel;
+let accessSize = WordAccess, isCodeGenOnly = 0 in
+defm storeri: ST_PostInc <"memw", "STriw", IntRegs, s4_2Imm, 0b1100>;
+
+let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
+defm storerd: ST_PostInc <"memd", "STrid", DoubleRegs, s4_3Imm, 0b1110>;
+
+let accessSize = HalfWordAccess, isNVStorable = 0, isCodeGenOnly = 0 in
+defm storerf: ST_PostInc <"memh", "STrih_H", IntRegs, s4_1Imm, 0b1011, 1>;
+
+// Patterns for generating stores, where the address takes different forms:
+// - frameindex,,
+// - base + offset,
+// - simple (base address without offset).
+// These would usually be used together (via Storex_pat defined below), but
+// in some cases one may want to apply different properties (such as
+// AddedComplexity) to the individual patterns.
+class Storex_fi_pat<PatFrag Store, PatFrag Value, InstHexagon MI>
+ : Pat<(Store Value:$Rs, AddrFI:$fi), (MI AddrFI:$fi, 0, Value:$Rs)>;
+class Storex_add_pat<PatFrag Store, PatFrag Value, PatFrag ImmPred,
+ InstHexagon MI>
+ : Pat<(Store Value:$Rt, (add (i32 IntRegs:$Rs), ImmPred:$Off)),
+ (MI IntRegs:$Rs, imm:$Off, Value:$Rt)>;
+
+multiclass Storex_pat<PatFrag Store, PatFrag Value, PatLeaf ImmPred,
+ InstHexagon MI> {
+ def: Storex_fi_pat <Store, Value, MI>;
+ def: Storex_add_pat <Store, Value, ImmPred, MI>;
+}
def : Pat<(post_truncsti8 (i32 IntRegs:$src1), IntRegs:$src2,
s4_3ImmPred:$offset),
- (POST_STbri IntRegs:$src2, s4_0ImmPred:$offset, IntRegs:$src1)>;
+ (S2_storerb_pi IntRegs:$src2, s4_0ImmPred:$offset, IntRegs:$src1)>;
def : Pat<(post_truncsti16 (i32 IntRegs:$src1), IntRegs:$src2,
s4_3ImmPred:$offset),
- (POST_SThri IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
+ (S2_storerh_pi IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
def : Pat<(post_store (i32 IntRegs:$src1), IntRegs:$src2, s4_2ImmPred:$offset),
- (POST_STwri IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
+ (S2_storeri_pi IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
def : Pat<(post_store (i64 DoubleRegs:$src1), IntRegs:$src2,
s4_3ImmPred:$offset),
- (POST_STdri IntRegs:$src2, s4_3ImmPred:$offset, DoubleRegs:$src1)>;
+ (S2_storerd_pi IntRegs:$src2, s4_3ImmPred:$offset, DoubleRegs:$src1)>;
//===----------------------------------------------------------------------===//
-// multiclass for the store instructions with MEMri operand.
+// Template class for post increment stores with register offset.
//===----------------------------------------------------------------------===//
-multiclass ST_MEMri_Pbase<string mnemonic, RegisterClass RC, bit isNot,
- bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME : STInst2<(outs),
- (ins PredRegs:$src1, MEMri:$addr, RC: $src2),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#mnemonic#"($addr) = $src2",
- []>;
-}
+let isNVStorable = 1 in
+class T_store_pr <string mnemonic, RegisterClass RC, bits<3> MajOp,
+ MemAccessSize AccessSz, bit isHalf = 0>
+ : STInst <(outs IntRegs:$_dst_),
+ (ins IntRegs:$src1, ModRegs:$src2, RC:$src3),
+ mnemonic#"($src1++$src2) = $src3"#!if(isHalf, ".h", ""),
+ [], "$src1 = $_dst_" > {
+ bits<5> src1;
+ bits<1> src2;
+ bits<5> src3;
+ let accessSize = AccessSz;
-multiclass ST_MEMri_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : ST_MEMri_Pbase<mnemonic, RC, PredNot, 0>;
+ let IClass = 0b1010;
- // Predicate new
- let validSubTargets = HasV4SubT, Predicates = [HasV4T] in
- defm _cdn#NAME#_V4 : ST_MEMri_Pbase<mnemonic, RC, PredNot, 1>;
+ let Inst{27-24} = 0b1101;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{13} = src2;
+ let Inst{12-8} = src3;
+ let Inst{7} = 0b0;
}
+
+let isCodeGenOnly = 0 in {
+def S2_storerb_pr : T_store_pr<"memb", IntRegs, 0b000, ByteAccess>;
+def S2_storerh_pr : T_store_pr<"memh", IntRegs, 0b010, HalfWordAccess>;
+def S2_storeri_pr : T_store_pr<"memw", IntRegs, 0b100, WordAccess>;
+def S2_storerd_pr : T_store_pr<"memd", DoubleRegs, 0b110, DoubleWordAccess>;
+
+def S2_storerf_pr : T_store_pr<"memh", IntRegs, 0b011, HalfWordAccess, 1>;
}
+let opExtendable = 1, isExtentSigned = 1, isPredicable = 1 in
+class T_store_io <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<3>MajOp, bit isH = 0>
+ : STInst <(outs),
+ (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
+ mnemonic#"($src1+#$src2) = $src3"#!if(isH,".h","")>,
+ AddrModeRel, ImmRegRel {
+ bits<5> src1;
+ bits<14> src2; // Actual address offset
+ bits<5> src3;
+ bits<11> offsetBits; // Represents offset encoding
+
+ string ImmOpStr = !cast<string>(ImmOp);
+
+ let opExtentBits = !if (!eq(ImmOpStr, "s11_3Ext"), 14,
+ !if (!eq(ImmOpStr, "s11_2Ext"), 13,
+ !if (!eq(ImmOpStr, "s11_1Ext"), 12,
+ /* s11_0Ext */ 11)));
+ let offsetBits = !if (!eq(ImmOpStr, "s11_3Ext"), src2{13-3},
+ !if (!eq(ImmOpStr, "s11_2Ext"), src2{12-2},
+ !if (!eq(ImmOpStr, "s11_1Ext"), src2{11-1},
+ /* s11_0Ext */ src2{10-0})));
+ let IClass = 0b1010;
+
+ let Inst{27} = 0b0;
+ let Inst{26-25} = offsetBits{10-9};
+ let Inst{24} = 0b1;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{13} = offsetBits{8};
+ let Inst{12-8} = src3;
+ let Inst{7-0} = offsetBits{7-0};
+ }
+
+let opExtendable = 2, isPredicated = 1 in
+class T_pstore_io <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<3>MajOp, bit PredNot, bit isPredNew, bit isH = 0>
+ : STInst <(outs),
+ (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$src3, RC:$src4),
+ !if(PredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+ ") ")#mnemonic#"($src2+#$src3) = $src4"#!if(isH,".h",""),
+ [],"",V2LDST_tc_st_SLOT01 >,
+ AddrModeRel, ImmRegRel {
+ bits<2> src1;
+ bits<5> src2;
+ bits<9> src3; // Actual address offset
+ bits<5> src4;
+ bits<6> offsetBits; // Represents offset encoding
+
+ let isPredicatedNew = isPredNew;
+ let isPredicatedFalse = PredNot;
+
+ string ImmOpStr = !cast<string>(ImmOp);
+ let opExtentBits = !if (!eq(ImmOpStr, "u6_3Ext"), 9,
+ !if (!eq(ImmOpStr, "u6_2Ext"), 8,
+ !if (!eq(ImmOpStr, "u6_1Ext"), 7,
+ /* u6_0Ext */ 6)));
+ let offsetBits = !if (!eq(ImmOpStr, "u6_3Ext"), src3{8-3},
+ !if (!eq(ImmOpStr, "u6_2Ext"), src3{7-2},
+ !if (!eq(ImmOpStr, "u6_1Ext"), src3{6-1},
+ /* u6_0Ext */ src3{5-0})));
+ let IClass = 0b0100;
+
+ let Inst{27} = 0b0;
+ let Inst{26} = PredNot;
+ let Inst{25} = isPredNew;
+ let Inst{24} = 0b0;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = src2;
+ let Inst{13} = offsetBits{5};
+ let Inst{12-8} = src4;
+ let Inst{7-3} = offsetBits{4-0};
+ let Inst{1-0} = src1;
+ }
let isExtendable = 1, isNVStorable = 1, hasSideEffects = 0 in
-multiclass ST_MEMri<string mnemonic, string CextOp, RegisterClass RC,
- bits<5> ImmBits, bits<5> PredImmBits> {
+multiclass ST_Idxd<string mnemonic, string CextOp, RegisterClass RC,
+ Operand ImmOp, Operand predImmOp, bits<3> MajOp, bit isH = 0> {
+ let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
+ def S2_#NAME#_io : T_store_io <mnemonic, RC, ImmOp, MajOp, isH>;
- let CextOpcode = CextOp, BaseOpcode = CextOp in {
- let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
- isPredicable = 1 in
- def NAME : STInst2<(outs),
- (ins MEMri:$addr, RC:$src),
- mnemonic#"($addr) = $src",
- []>;
+ // Predicated
+ def S2_p#NAME#t_io : T_pstore_io<mnemonic, RC, predImmOp, MajOp, 0, 0, isH>;
+ def S2_p#NAME#f_io : T_pstore_io<mnemonic, RC, predImmOp, MajOp, 1, 0, isH>;
- let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
- isPredicated = 1 in {
- defm Pt : ST_MEMri_Pred<mnemonic, RC, 0>;
- defm NotPt : ST_MEMri_Pred<mnemonic, RC, 1>;
- }
+ // Predicated new
+ def S4_p#NAME#tnew_io : T_pstore_io <mnemonic, RC, predImmOp,
+ MajOp, 0, 1, isH>;
+ def S4_p#NAME#fnew_io : T_pstore_io <mnemonic, RC, predImmOp,
+ MajOp, 1, 1, isH>;
}
}
-let addrMode = BaseImmOffset, isMEMri = "true" in {
+let addrMode = BaseImmOffset, InputType = "imm", isCodeGenOnly = 0 in {
let accessSize = ByteAccess in
- defm STrib: ST_MEMri < "memb", "STrib", IntRegs, 11, 6>, AddrModeRel;
+ defm storerb: ST_Idxd < "memb", "STrib", IntRegs, s11_0Ext, u6_0Ext, 0b000>;
+
+ let accessSize = HalfWordAccess, opExtentAlign = 1 in
+ defm storerh: ST_Idxd < "memh", "STrih", IntRegs, s11_1Ext, u6_1Ext, 0b010>;
- let accessSize = HalfWordAccess in
- defm STrih: ST_MEMri < "memh", "STrih", IntRegs, 12, 7>, AddrModeRel;
+ let accessSize = WordAccess, opExtentAlign = 2 in
+ defm storeri: ST_Idxd < "memw", "STriw", IntRegs, s11_2Ext, u6_2Ext, 0b100>;
- let accessSize = WordAccess in
- defm STriw: ST_MEMri < "memw", "STriw", IntRegs, 13, 8>, AddrModeRel;
+ let accessSize = DoubleWordAccess, isNVStorable = 0, opExtentAlign = 3 in
+ defm storerd: ST_Idxd < "memd", "STrid", DoubleRegs, s11_3Ext,
+ u6_3Ext, 0b110>;
- let accessSize = DoubleWordAccess, isNVStorable = 0 in
- defm STrid: ST_MEMri < "memd", "STrid", DoubleRegs, 14, 9>, AddrModeRel;
+ let accessSize = HalfWordAccess, opExtentAlign = 1 in
+ defm storerf: ST_Idxd < "memh", "STrif", IntRegs, s11_1Ext,
+ u6_1Ext, 0b011, 1>;
}
+class Storex_simple_pat<PatFrag Store, PatFrag Value, InstHexagon MI>
+ : Pat<(Store Value:$Rt, (i32 IntRegs:$Rs)),
+ (MI IntRegs:$Rs, 0, Value:$Rt)>;
+
+// Regular stores in the DAG have two operands: value and address.
+// Atomic stores also have two, but they are reversed: address, value.
+// To use atomic stores with the patterns, they need to have their operands
+// swapped. This relies on the knowledge that the F.Fragment uses names
+// "ptr" and "val".
+class SwapSt<PatFrag F>
+ : PatFrag<(ops node:$val, node:$ptr), F.Fragment>;
+
+def: Storex_simple_pat<SwapSt<atomic_store_8>, I32, S2_storerb_io>;
+def: Storex_simple_pat<SwapSt<atomic_store_16>, I32, S2_storerh_io>;
+def: Storex_simple_pat<SwapSt<atomic_store_32>, I32, S2_storeri_io>;
+def: Storex_simple_pat<SwapSt<atomic_store_64>, I64, S2_storerd_io>;
+
def : Pat<(truncstorei8 (i32 IntRegs:$src1), ADDRriS11_0:$addr),
- (STrib ADDRriS11_0:$addr, (i32 IntRegs:$src1))>;
+ (S2_storerb_io AddrFI:$addr, 0, (i32 IntRegs:$src1))>;
def : Pat<(truncstorei16 (i32 IntRegs:$src1), ADDRriS11_1:$addr),
- (STrih ADDRriS11_1:$addr, (i32 IntRegs:$src1))>;
+ (S2_storerh_io AddrFI:$addr, 0, (i32 IntRegs:$src1))>;
def : Pat<(store (i32 IntRegs:$src1), ADDRriS11_2:$addr),
- (STriw ADDRriS11_2:$addr, (i32 IntRegs:$src1))>;
+ (S2_storeri_io AddrFI:$addr, 0, (i32 IntRegs:$src1))>;
def : Pat<(store (i64 DoubleRegs:$src1), ADDRriS11_3:$addr),
- (STrid ADDRriS11_3:$addr, (i64 DoubleRegs:$src1))>;
+ (S2_storerd_io AddrFI:$addr, 0, (i64 DoubleRegs:$src1))>;
-//===----------------------------------------------------------------------===//
-// multiclass for the store instructions with base+immediate offset
-// addressing mode
-//===----------------------------------------------------------------------===//
-multiclass ST_Idxd_Pbase<string mnemonic, RegisterClass RC, Operand predImmOp,
- bit isNot, bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME : STInst2<(outs),
- (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC: $src4),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#mnemonic#"($src2+#$src3) = $src4",
- []>;
-}
-
-multiclass ST_Idxd_Pred<string mnemonic, RegisterClass RC, Operand predImmOp,
- bit PredNot> {
- let isPredicatedFalse = PredNot, isPredicated = 1 in {
- defm _c#NAME : ST_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 0>;
+let AddedComplexity = 10 in {
+def : Pat<(truncstorei8 (i32 IntRegs:$src1), (add IntRegs:$src2,
+ s11_0ExtPred:$offset)),
+ (S2_storerb_io IntRegs:$src2, s11_0ImmPred:$offset,
+ (i32 IntRegs:$src1))>;
- // Predicate new
- let validSubTargets = HasV4SubT, Predicates = [HasV4T] in
- defm _cdn#NAME#_V4 : ST_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 1>;
- }
-}
+def : Pat<(truncstorei16 (i32 IntRegs:$src1), (add IntRegs:$src2,
+ s11_1ExtPred:$offset)),
+ (S2_storerh_io IntRegs:$src2, s11_1ImmPred:$offset,
+ (i32 IntRegs:$src1))>;
-let isExtendable = 1, isNVStorable = 1, hasSideEffects = 0 in
-multiclass ST_Idxd<string mnemonic, string CextOp, RegisterClass RC,
- Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
- bits<5> PredImmBits> {
+def : Pat<(store (i32 IntRegs:$src1), (add IntRegs:$src2,
+ s11_2ExtPred:$offset)),
+ (S2_storeri_io IntRegs:$src2, s11_2ImmPred:$offset,
+ (i32 IntRegs:$src1))>;
- let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
- let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
- isPredicable = 1 in
- def NAME : STInst2<(outs),
- (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
- mnemonic#"($src1+#$src2) = $src3",
- []>;
+def : Pat<(store (i64 DoubleRegs:$src1), (add IntRegs:$src2,
+ s11_3ExtPred:$offset)),
+ (S2_storerd_io IntRegs:$src2, s11_3ImmPred:$offset,
+ (i64 DoubleRegs:$src1))>;
+}
- let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits in {
- defm Pt : ST_Idxd_Pred<mnemonic, RC, predImmOp, 0>;
- defm NotPt : ST_Idxd_Pred<mnemonic, RC, predImmOp, 1>;
- }
+// memh(Rx++#s4:1)=Rt.H
+
+// Store predicate.
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 13,
+ isCodeGenOnly = 1, isPseudo = 1, hasSideEffects = 0 in
+def STriw_pred : STInst<(outs),
+ (ins IntRegs:$addr, s11_2Ext:$off, PredRegs:$src1),
+ ".error \"should not emit\"", []>;
+
+// S2_allocframe: Allocate stack frame.
+let Defs = [R29, R30], Uses = [R29, R31, R30],
+ hasSideEffects = 0, accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
+def S2_allocframe: ST0Inst <
+ (outs), (ins u11_3Imm:$u11_3),
+ "allocframe(#$u11_3)" > {
+ bits<14> u11_3;
+
+ let IClass = 0b1010;
+ let Inst{27-16} = 0b000010011101;
+ let Inst{13-11} = 0b000;
+ let Inst{10-0} = u11_3{13-3};
+ }
+
+// S2_storer[bhwdf]_pci: Store byte/half/word/double.
+// S2_storer[bhwdf]_pci -> S2_storerbnew_pci
+let Uses = [CS], isNVStorable = 1 in
+class T_store_pci <string mnemonic, RegisterClass RC,
+ Operand Imm, bits<4>MajOp,
+ MemAccessSize AlignSize, string RegSrc = "Rt">
+ : STInst <(outs IntRegs:$_dst_),
+ (ins IntRegs:$Rz, Imm:$offset, ModRegs:$Mu, RC:$Rt),
+ #mnemonic#"($Rz ++ #$offset:circ($Mu)) = $"#RegSrc#"",
+ [] ,
+ "$Rz = $_dst_" > {
+ bits<5> Rz;
+ bits<7> offset;
+ bits<1> Mu;
+ bits<5> Rt;
+ let accessSize = AlignSize;
+
+ let IClass = 0b1010;
+ let Inst{27-25} = 0b100;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12-8} = Rt;
+ let Inst{7} = 0b0;
+ let Inst{6-3} =
+ !if (!eq(!cast<string>(AlignSize), "DoubleWordAccess"), offset{6-3},
+ !if (!eq(!cast<string>(AlignSize), "WordAccess"), offset{5-2},
+ !if (!eq(!cast<string>(AlignSize), "HalfWordAccess"), offset{4-1},
+ /* ByteAccess */ offset{3-0})));
+ let Inst{1} = 0b0;
}
+
+let isCodeGenOnly = 0 in {
+def S2_storerb_pci : T_store_pci<"memb", IntRegs, s4_0Imm, 0b1000,
+ ByteAccess>;
+def S2_storerh_pci : T_store_pci<"memh", IntRegs, s4_1Imm, 0b1010,
+ HalfWordAccess>;
+def S2_storerf_pci : T_store_pci<"memh", IntRegs, s4_1Imm, 0b1011,
+ HalfWordAccess, "Rt.h">;
+def S2_storeri_pci : T_store_pci<"memw", IntRegs, s4_2Imm, 0b1100,
+ WordAccess>;
+def S2_storerd_pci : T_store_pci<"memd", DoubleRegs, s4_3Imm, 0b1110,
+ DoubleWordAccess>;
+}
+
+let Uses = [CS], isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 4 in
+class T_storenew_pci <string mnemonic, Operand Imm,
+ bits<2>MajOp, MemAccessSize AlignSize>
+ : NVInst < (outs IntRegs:$_dst_),
+ (ins IntRegs:$Rz, Imm:$offset, ModRegs:$Mu, IntRegs:$Nt),
+ #mnemonic#"($Rz ++ #$offset:circ($Mu)) = $Nt.new",
+ [],
+ "$Rz = $_dst_"> {
+ bits<5> Rz;
+ bits<6> offset;
+ bits<1> Mu;
+ bits<3> Nt;
+
+ let accessSize = AlignSize;
+
+ let IClass = 0b1010;
+ let Inst{27-21} = 0b1001101;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12-11} = MajOp;
+ let Inst{10-8} = Nt;
+ let Inst{7} = 0b0;
+ let Inst{6-3} =
+ !if (!eq(!cast<string>(AlignSize), "WordAccess"), offset{5-2},
+ !if (!eq(!cast<string>(AlignSize), "HalfWordAccess"), offset{4-1},
+ /* ByteAccess */ offset{3-0}));
+ let Inst{1} = 0b0;
+ }
+let isCodeGenOnly = 0 in {
+def S2_storerbnew_pci : T_storenew_pci <"memb", s4_0Imm, 0b00, ByteAccess>;
+def S2_storerhnew_pci : T_storenew_pci <"memh", s4_1Imm, 0b01, HalfWordAccess>;
+def S2_storerinew_pci : T_storenew_pci <"memw", s4_2Imm, 0b10, WordAccess>;
}
-let addrMode = BaseImmOffset, InputType = "reg" in {
- let accessSize = ByteAccess in
- defm STrib_indexed: ST_Idxd < "memb", "STrib", IntRegs, s11_0Ext,
- u6_0Ext, 11, 6>, AddrModeRel, ImmRegRel;
+//===----------------------------------------------------------------------===//
+// Circular stores with auto-increment register
+//===----------------------------------------------------------------------===//
+let Uses = [CS], isNVStorable = 1, isCodeGenOnly = 0 in
+class T_store_pcr <string mnemonic, RegisterClass RC, bits<4>MajOp,
+ MemAccessSize AlignSize, string RegSrc = "Rt">
+ : STInst <(outs IntRegs:$_dst_),
+ (ins IntRegs:$Rz, ModRegs:$Mu, RC:$Rt),
+ #mnemonic#"($Rz ++ I:circ($Mu)) = $"#RegSrc#"",
+ [],
+ "$Rz = $_dst_" > {
+ bits<5> Rz;
+ bits<1> Mu;
+ bits<5> Rt;
- let accessSize = HalfWordAccess in
- defm STrih_indexed: ST_Idxd < "memh", "STrih", IntRegs, s11_1Ext,
- u6_1Ext, 12, 7>, AddrModeRel, ImmRegRel;
+ let accessSize = AlignSize;
- let accessSize = WordAccess in
- defm STriw_indexed: ST_Idxd < "memw", "STriw", IntRegs, s11_2Ext,
- u6_2Ext, 13, 8>, AddrModeRel, ImmRegRel;
+ let IClass = 0b1010;
+ let Inst{27-25} = 0b100;
+ let Inst{24-21} = MajOp;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12-8} = Rt;
+ let Inst{7} = 0b0;
+ let Inst{1} = 0b1;
+ }
- let accessSize = DoubleWordAccess, isNVStorable = 0 in
- defm STrid_indexed: ST_Idxd < "memd", "STrid", DoubleRegs, s11_3Ext,
- u6_3Ext, 14, 9>, AddrModeRel;
+let isCodeGenOnly = 0 in {
+def S2_storerb_pcr : T_store_pcr<"memb", IntRegs, 0b1000, ByteAccess>;
+def S2_storerh_pcr : T_store_pcr<"memh", IntRegs, 0b1010, HalfWordAccess>;
+def S2_storeri_pcr : T_store_pcr<"memw", IntRegs, 0b1100, WordAccess>;
+def S2_storerd_pcr : T_store_pcr<"memd", DoubleRegs, 0b1110, DoubleWordAccess>;
+def S2_storerf_pcr : T_store_pcr<"memh", IntRegs, 0b1011,
+ HalfWordAccess, "Rt.h">;
+}
+
+//===----------------------------------------------------------------------===//
+// Circular .new stores with auto-increment register
+//===----------------------------------------------------------------------===//
+let Uses = [CS], isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3 in
+class T_storenew_pcr <string mnemonic, bits<2>MajOp,
+ MemAccessSize AlignSize>
+ : NVInst <(outs IntRegs:$_dst_),
+ (ins IntRegs:$Rz, ModRegs:$Mu, IntRegs:$Nt),
+ #mnemonic#"($Rz ++ I:circ($Mu)) = $Nt.new" ,
+ [] ,
+ "$Rz = $_dst_"> {
+ bits<5> Rz;
+ bits<1> Mu;
+ bits<3> Nt;
+
+ let accessSize = AlignSize;
+
+ let IClass = 0b1010;
+ let Inst{27-21} = 0b1001101;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12-11} = MajOp;
+ let Inst{10-8} = Nt;
+ let Inst{7} = 0b0;
+ let Inst{1} = 0b1;
+ }
+
+let isCodeGenOnly = 0 in {
+def S2_storerbnew_pcr : T_storenew_pcr <"memb", 0b00, ByteAccess>;
+def S2_storerhnew_pcr : T_storenew_pcr <"memh", 0b01, HalfWordAccess>;
+def S2_storerinew_pcr : T_storenew_pcr <"memw", 0b10, WordAccess>;
}
-let AddedComplexity = 10 in {
-def : Pat<(truncstorei8 (i32 IntRegs:$src1), (add IntRegs:$src2,
- s11_0ExtPred:$offset)),
- (STrib_indexed IntRegs:$src2, s11_0ImmPred:$offset,
- (i32 IntRegs:$src1))>;
+//===----------------------------------------------------------------------===//
+// Bit-reversed stores with auto-increment register
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0 in
+class T_store_pbr<string mnemonic, RegisterClass RC,
+ MemAccessSize addrSize, bits<3> majOp,
+ bit isHalf = 0>
+ : STInst
+ <(outs IntRegs:$_dst_),
+ (ins IntRegs:$Rz, ModRegs:$Mu, RC:$src),
+ #mnemonic#"($Rz ++ $Mu:brev) = $src"#!if (!eq(isHalf, 1), ".h", ""),
+ [], "$Rz = $_dst_" > {
+
+ let accessSize = addrSize;
+
+ bits<5> Rz;
+ bits<1> Mu;
+ bits<5> src;
+
+ let IClass = 0b1010;
+
+ let Inst{27-24} = 0b1111;
+ let Inst{23-21} = majOp;
+ let Inst{7} = 0b0;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{12-8} = src;
+ }
-def : Pat<(truncstorei16 (i32 IntRegs:$src1), (add IntRegs:$src2,
- s11_1ExtPred:$offset)),
- (STrih_indexed IntRegs:$src2, s11_1ImmPred:$offset,
- (i32 IntRegs:$src1))>;
+let isNVStorable = 1, isCodeGenOnly = 0 in {
+ let BaseOpcode = "S2_storerb_pbr" in
+ def S2_storerb_pbr : T_store_pbr<"memb", IntRegs, ByteAccess,
+ 0b000>, NewValueRel;
+ let BaseOpcode = "S2_storerh_pbr" in
+ def S2_storerh_pbr : T_store_pbr<"memh", IntRegs, HalfWordAccess,
+ 0b010>, NewValueRel;
+ let BaseOpcode = "S2_storeri_pbr" in
+ def S2_storeri_pbr : T_store_pbr<"memw", IntRegs, WordAccess,
+ 0b100>, NewValueRel;
+}
+let isCodeGenOnly = 0 in {
+def S2_storerf_pbr : T_store_pbr<"memh", IntRegs, HalfWordAccess, 0b011, 1>;
+def S2_storerd_pbr : T_store_pbr<"memd", DoubleRegs, DoubleWordAccess, 0b110>;
+}
-def : Pat<(store (i32 IntRegs:$src1), (add IntRegs:$src2,
- s11_2ExtPred:$offset)),
- (STriw_indexed IntRegs:$src2, s11_2ImmPred:$offset,
- (i32 IntRegs:$src1))>;
+//===----------------------------------------------------------------------===//
+// Bit-reversed .new stores with auto-increment register
+//===----------------------------------------------------------------------===//
+let isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3,
+ hasSideEffects = 0 in
+class T_storenew_pbr<string mnemonic, MemAccessSize addrSize, bits<2> majOp>
+ : NVInst <(outs IntRegs:$_dst_),
+ (ins IntRegs:$Rz, ModRegs:$Mu, IntRegs:$Nt),
+ #mnemonic#"($Rz ++ $Mu:brev) = $Nt.new", [],
+ "$Rz = $_dst_">, NewValueRel {
+ let accessSize = addrSize;
+ bits<5> Rz;
+ bits<1> Mu;
+ bits<3> Nt;
-def : Pat<(store (i64 DoubleRegs:$src1), (add IntRegs:$src2,
- s11_3ExtPred:$offset)),
- (STrid_indexed IntRegs:$src2, s11_3ImmPred:$offset,
- (i64 DoubleRegs:$src1))>;
-}
+ let IClass = 0b1010;
-// memh(Rx++#s4:1)=Rt.H
+ let Inst{27-21} = 0b1111101;
+ let Inst{12-11} = majOp;
+ let Inst{7} = 0b0;
+ let Inst{20-16} = Rz;
+ let Inst{13} = Mu;
+ let Inst{10-8} = Nt;
+ }
-// Store word.
-// Store predicate.
-let Defs = [R10,R11,D5], hasSideEffects = 0 in
-def STriw_pred : STInst2<(outs),
- (ins MEMri:$addr, PredRegs:$src1),
- "Error; should not emit",
- []>;
+let BaseOpcode = "S2_storerb_pbr", isCodeGenOnly = 0 in
+def S2_storerbnew_pbr : T_storenew_pbr<"memb", ByteAccess, 0b00>;
+
+let BaseOpcode = "S2_storerh_pbr", isCodeGenOnly = 0 in
+def S2_storerhnew_pbr : T_storenew_pbr<"memh", HalfWordAccess, 0b01>;
+
+let BaseOpcode = "S2_storeri_pbr", isCodeGenOnly = 0 in
+def S2_storerinew_pbr : T_storenew_pbr<"memw", WordAccess, 0b10>;
-// Allocate stack frame.
-let Defs = [R29, R30], Uses = [R31, R30], hasSideEffects = 0 in {
- def ALLOCFRAME : STInst2<(outs),
- (ins i32imm:$amt),
- "allocframe(#$amt)",
- []>;
-}
//===----------------------------------------------------------------------===//
// ST -
//===----------------------------------------------------------------------===//
def A2_satub : T_S2op_1_ii <"satub", 0b11, 0b110>;
def A2_sath : T_S2op_1_ii <"sath", 0b11, 0b100>;
def A2_satuh : T_S2op_1_ii <"satuh", 0b11, 0b101>;
+ def A2_roundsat : T_S2op_1_id <"round", 0b11, 0b001, 0b1>;
}
let Itinerary = S_2op_tc_2_SLOT23, isCodeGenOnly = 0 in {
+ // Bit reverse
+ def S2_brev : T_S2op_1_ii <"brev", 0b01, 0b110>;
+
// Absolute value word
def A2_abs : T_S2op_1_ii <"abs", 0b10, 0b100>;
let Inst{7-5} = MinOp;
let Inst{4-0} = dst;
}
+
+class T_S2op_2_di <string mnemonic, bits<3> MajOp, bits<3> MinOp>
+ : T_S2op_2 <mnemonic, 0b1000, DoubleRegs, IntRegs, MajOp, MinOp, 0, 0>;
+
+let hasNewValue = 1 in
+class T_S2op_2_id <string mnemonic, bits<3> MajOp, bits<3> MinOp>
+ : T_S2op_2 <mnemonic, 0b1000, IntRegs, DoubleRegs, MajOp, MinOp, 0, 0>;
let hasNewValue = 1 in
class T_S2op_2_ii <string mnemonic, bits<3> MajOp, bits<3> MinOp,
let isCodeGenOnly = 0 in
def S2_asr_i_r_rnd : T_S2op_2_ii <"asr", 0b010, 0b000, 0, 1>;
+def S2_asr_i_r_rnd_goodsyntax
+ : SInst <(outs IntRegs:$dst), (ins IntRegs:$src, u5Imm:$u5),
+ "$dst = asrrnd($src, #$u5)",
+ [], "", S_2op_tc_1_SLOT23>;
+
def: Pat<(i32 (sra (i32 (add (i32 (sra I32:$src1, u5ImmPred:$src2)),
(i32 1))),
(i32 1))),
def S2_deinterleave : T_S2op_3 <"deinterleave", 0b11, 0b100>;
}
+// Vector Complex conjugate
+let isCodeGenOnly = 0 in
+def A2_vconj : T_S2op_3 <"vconj", 0b10, 0b111, 1>;
+
+// Vector absolute value halfwords with and without saturation
+// Rdd64=vabsh(Rss64)[:sat]
+let isCodeGenOnly = 0 in {
+def A2_vabsh : T_S2op_3 <"vabsh", 0b01, 0b100>;
+def A2_vabshsat : T_S2op_3 <"vabsh", 0b01, 0b101, 1>;
+}
+
+// Vector absolute value words with and without saturation
+let isCodeGenOnly = 0 in {
+def A2_vabsw : T_S2op_3 <"vabsw", 0b01, 0b110>;
+def A2_vabswsat : T_S2op_3 <"vabsw", 0b01, 0b111, 1>;
+}
+
//===----------------------------------------------------------------------===//
// STYPE/BIT +
//===----------------------------------------------------------------------===//
let Inst{4-0} = Rd;
}
-// Shift by immediate and add.
-let AddedComplexity = 100 in
-def ADDASL : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2,
- u3Imm:$src3),
- "$dst = addasl($src1, $src2, #$src3)",
- [(set (i32 IntRegs:$dst), (add (i32 IntRegs:$src1),
- (shl (i32 IntRegs:$src2),
- u3ImmPred:$src3)))]>;
-
-// Shift by register.
-def ASL_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = asl($src1, $src2)",
- [(set (i32 IntRegs:$dst), (shl (i32 IntRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-def ASR_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = asr($src1, $src2)",
- [(set (i32 IntRegs:$dst), (sra (i32 IntRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-def LSL_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = lsl($src1, $src2)",
- [(set (i32 IntRegs:$dst), (shl (i32 IntRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-def LSR_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = lsr($src1, $src2)",
- [(set (i32 IntRegs:$dst), (srl (i32 IntRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-def ASLd : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, IntRegs:$src2),
- "$dst = asl($src1, $src2)",
- [(set (i64 DoubleRegs:$dst), (shl (i64 DoubleRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-def LSLd : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, IntRegs:$src2),
- "$dst = lsl($src1, $src2)",
- [(set (i64 DoubleRegs:$dst), (shl (i64 DoubleRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-def ASRd_rr : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
- IntRegs:$src2),
- "$dst = asr($src1, $src2)",
- [(set (i64 DoubleRegs:$dst), (sra (i64 DoubleRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
-def LSRd_rr : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
- IntRegs:$src2),
- "$dst = lsr($src1, $src2)",
- [(set (i64 DoubleRegs:$dst), (srl (i64 DoubleRegs:$src1),
- (i32 IntRegs:$src2)))]>;
-
//===----------------------------------------------------------------------===//
// STYPE/SHIFT -
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// SYSTEM/USER +
//===----------------------------------------------------------------------===//
-def SDHexagonBARRIER: SDTypeProfile<0, 0, []>;
-def HexagonBARRIER: SDNode<"HexagonISD::BARRIER", SDHexagonBARRIER,
- [SDNPHasChain]>;
+def HexagonBARRIER: SDNode<"HexagonISD::BARRIER", SDTNone, [SDNPHasChain]>;
-let hasSideEffects = 1, isSolo = 1 in
+let hasSideEffects = 1, isSoloAX = 1, isCodeGenOnly = 0 in
def BARRIER : SYSInst<(outs), (ins),
"barrier",
- [(HexagonBARRIER)]>;
+ [(HexagonBARRIER)],"",ST_tc_st_SLOT0> {
+ let Inst{31-28} = 0b1010;
+ let Inst{27-21} = 0b1000000;
+}
//===----------------------------------------------------------------------===//
// 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;
-// TFRI64 - assembly mapped.
-let isReMaterializable = 1 in
-def TFRI64 : ALU64_rr<(outs DoubleRegs:$dst), (ins s8Imm64:$src1),
- "$dst = #$src1",
- [(set (i64 DoubleRegs:$dst), s8Imm64Pred:$src1)]>;
+ 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, $Rs)" > {
+ bits<9> r7_2;
+ bits<5> Rs;
+
+ let IClass = 0b0110;
+
+ let Inst{22-21} = op;
+ let Inst{27-23} = 0b00001;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = r7_2{8-4};
+ let Inst{4-3} = r7_2{3-2};
+ }
+
+multiclass SPLOOP_ri<string mnemonic, bits<2> op> {
+ def i : SPLOOP_iBase<mnemonic, op>;
+ def r : SPLOOP_rBase<mnemonic, op>;
+}
+
+let isCodeGenOnly = 0 in {
+defm J2_ploop1s : SPLOOP_ri<"1", 0b01>;
+defm J2_ploop2s : SPLOOP_ri<"2", 0b10>;
+defm J2_ploop3s : SPLOOP_ri<"3", 0b11>;
+}
+
+
+// if (Rs[!>=<]=#0) jump:[t/nt]
+let Defs = [PC], isPredicated = 1, isBranch = 1, hasSideEffects = 0,
+ hasSideEffects = 0 in
+class J2_jump_0_Base<string compare, bit isTak, bits<2> op>
+ : CRInst <(outs), (ins IntRegs:$Rs, brtarget:$r13_2),
+ "if ($Rs"#compare#"#0) jump"#!if(isTak, ":t", ":nt")#" $r13_2" > {
+ bits<5> Rs;
+ bits<15> r13_2;
+
+ let IClass = 0b0110;
+
+ let Inst{27-24} = 0b0001;
+ let Inst{23-22} = op;
+ let Inst{12} = isTak;
+ let Inst{21} = r13_2{14};
+ let Inst{20-16} = Rs;
+ let Inst{11-1} = r13_2{12-2};
+ let Inst{13} = r13_2{13};
+ }
+
+multiclass J2_jump_compare_0<string compare, bits<2> op> {
+ def NAME : J2_jump_0_Base<compare, 0, op>;
+ def NAME#pt : J2_jump_0_Base<compare, 1, op>;
+}
+let isCodeGenOnly = 0 in {
+defm J2_jumprz : J2_jump_compare_0<"!=", 0b00>;
+defm J2_jumprgtez : J2_jump_compare_0<">=", 0b01>;
+defm J2_jumprnz : J2_jump_compare_0<"==", 0b10>;
+defm J2_jumprltez : J2_jump_compare_0<"<=", 0b11>;
+}
+
+// Transfer to/from Control/GPR Guest/GPR
+let hasSideEffects = 0 in
+class TFR_CR_RS_base<RegisterClass CTRC, RegisterClass RC, bit isDouble>
+ : CRInst <(outs CTRC:$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:$Rd, C7)>;
+
+// Y4_trace: Send value to etm trace.
+let isSoloAX = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
+def Y4_trace: CRInst <(outs), (ins IntRegs:$Rs),
+ "trace($Rs)"> {
+ bits<5> Rs;
+
+ let IClass = 0b0110;
+ let Inst{27-21} = 0b0010010;
+ let Inst{20-16} = Rs;
+ }
let AddedComplexity = 100, isPredicated = 1 in
def TFR_condset_ri : ALU32_rr<(outs IntRegs:$dst),
"$dst = add($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, [
- SDTCisVT<0, i32>,
- SDTCisVT<1, i32>,
- SDTCisPtrTy<0>]>;
-def HexagonCONST32 : SDNode<"HexagonISD::CONST32", SDTHexagonCONST32>;
+def SDTHexagonCONST32 : SDTypeProfile<1, 1, [SDTCisVT<0, i32>,
+ SDTCisVT<1, i32>,
+ SDTCisPtrTy<0>]>;
+def HexagonCONST32 : SDNode<"HexagonISD::CONST32", SDTHexagonCONST32>;
def HexagonCONST32_GP : SDNode<"HexagonISD::CONST32_GP", SDTHexagonCONST32>;
// HI/LO Instructions
"$dst = xor($dst, $dst)",
[(set (i1 PredRegs:$dst), 0)]>;
-def MPY_trsext : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
- "$dst = mpy($src1, $src2)",
- [(set (i32 IntRegs:$dst),
- (trunc (i64 (srl (i64 (mul (i64 (sext (i32 IntRegs:$src1))),
- (i64 (sext (i32 IntRegs:$src2))))),
- (i32 32)))))]>;
-
// Pseudo instructions.
def SDT_SPCallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
-
-def SDT_SPCallSeqEnd : SDCallSeqEnd<[ SDTCisVT<0, i32>,
+def SDT_SPCallSeqEnd : SDCallSeqEnd<[ SDTCisVT<0, i32>,
SDTCisVT<1, i32> ]>;
-def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_SPCallSeqEnd,
- [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
-
def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_SPCallSeqStart,
[SDNPHasChain, SDNPOutGlue]>;
+def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_SPCallSeqEnd,
+ [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
-def SDT_SPCall : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
-
-def call : SDNode<"HexagonISD::CALL", SDT_SPCall,
- [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue, SDNPVariadic]>;
+def SDT_SPCall : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
// For tailcalls a HexagonTCRet SDNode has 3 SDNode Properties - a chain,
// Optional Flag and Variable Arguments.
[], "", J_tc_2early_SLOT23>;
}
-// Map call instruction.
-def : Pat<(call (i32 IntRegs:$dst)),
- (J2_callr (i32 IntRegs:$dst))>, Requires<[HasV2TOnly]>;
-def : Pat<(call tglobaladdr:$dst),
- (CALL tglobaladdr:$dst)>, Requires<[HasV2TOnly]>;
-def : Pat<(call texternalsym:$dst),
- (CALL texternalsym:$dst)>, Requires<[HasV2TOnly]>;
//Tail calls.
def : Pat<(HexagonTCRet tglobaladdr:$dst),
(TCRETURNtg tglobaladdr:$dst)>;
def : Pat<(HexagonTCRet (i32 IntRegs:$dst)),
(TCRETURNR (i32 IntRegs:$dst))>;
-// Atomic load and store support
-// 8 bit atomic load
-def : Pat<(atomic_load_8 ADDRriS11_0:$src1),
- (i32 (LDriub ADDRriS11_0:$src1))>;
-
-def : Pat<(atomic_load_8 (add (i32 IntRegs:$src1), s11_0ImmPred:$offset)),
- (i32 (LDriub_indexed (i32 IntRegs:$src1), s11_0ImmPred:$offset))>;
-
-// 16 bit atomic load
-def : Pat<(atomic_load_16 ADDRriS11_1:$src1),
- (i32 (LDriuh ADDRriS11_1:$src1))>;
-
-def : Pat<(atomic_load_16 (add (i32 IntRegs:$src1), s11_1ImmPred:$offset)),
- (i32 (LDriuh_indexed (i32 IntRegs:$src1), s11_1ImmPred:$offset))>;
-
-def : Pat<(atomic_load_32 ADDRriS11_2:$src1),
- (i32 (LDriw ADDRriS11_2:$src1))>;
-
-def : Pat<(atomic_load_32 (add (i32 IntRegs:$src1), s11_2ImmPred:$offset)),
- (i32 (LDriw_indexed (i32 IntRegs:$src1), s11_2ImmPred:$offset))>;
-
-// 64 bit atomic load
-def : Pat<(atomic_load_64 ADDRriS11_3:$src1),
- (i64 (LDrid ADDRriS11_3:$src1))>;
-
-def : Pat<(atomic_load_64 (add (i32 IntRegs:$src1), s11_3ImmPred:$offset)),
- (i64 (LDrid_indexed (i32 IntRegs:$src1), s11_3ImmPred:$offset))>;
-
-
-def : Pat<(atomic_store_8 ADDRriS11_0:$src2, (i32 IntRegs:$src1)),
- (STrib ADDRriS11_0:$src2, (i32 IntRegs:$src1))>;
-
-def : Pat<(atomic_store_8 (add (i32 IntRegs:$src2), s11_0ImmPred:$offset),
- (i32 IntRegs:$src1)),
- (STrib_indexed (i32 IntRegs:$src2), s11_0ImmPred:$offset,
- (i32 IntRegs:$src1))>;
-
-
-def : Pat<(atomic_store_16 ADDRriS11_1:$src2, (i32 IntRegs:$src1)),
- (STrih ADDRriS11_1:$src2, (i32 IntRegs:$src1))>;
-
-def : Pat<(atomic_store_16 (i32 IntRegs:$src1),
- (add (i32 IntRegs:$src2), s11_1ImmPred:$offset)),
- (STrih_indexed (i32 IntRegs:$src2), s11_1ImmPred:$offset,
- (i32 IntRegs:$src1))>;
-
-def : Pat<(atomic_store_32 ADDRriS11_2:$src2, (i32 IntRegs:$src1)),
- (STriw ADDRriS11_2:$src2, (i32 IntRegs:$src1))>;
-
-def : Pat<(atomic_store_32 (add (i32 IntRegs:$src2), s11_2ImmPred:$offset),
- (i32 IntRegs:$src1)),
- (STriw_indexed (i32 IntRegs:$src2), s11_2ImmPred:$offset,
- (i32 IntRegs:$src1))>;
-
-
-
-
-def : Pat<(atomic_store_64 ADDRriS11_3:$src2, (i64 DoubleRegs:$src1)),
- (STrid ADDRriS11_3:$src2, (i64 DoubleRegs:$src1))>;
-
-def : Pat<(atomic_store_64 (add (i32 IntRegs:$src2), s11_3ImmPred:$offset),
- (i64 DoubleRegs:$src1)),
- (STrid_indexed (i32 IntRegs:$src2), s11_3ImmPred:$offset,
- (i64 DoubleRegs:$src1))>;
-
// Map from r0 = and(r1, 65535) to r0 = zxth(r1)
def : Pat <(and (i32 IntRegs:$src1), 65535),
(A2_zxth (i32 IntRegs:$src1))>;
let AddedComplexity = 100 in
def : Pat <(i64 (zextloadi1 (HexagonCONST32 tglobaladdr:$global))),
(i64 (A2_combinew (A2_tfrsi 0),
- (LDriub_indexed (CONST32_set tglobaladdr:$global), 0)))>,
+ (L2_loadrub_io (CONST32_set tglobaladdr:$global), 0)))>,
Requires<[NoV4T]>;
// Map from i1 loads to 32 bits. This assumes that the i1* is byte aligned.
let AddedComplexity = 10 in
def : Pat <(i32 (zextloadi1 ADDRriS11_0:$addr)),
- (i32 (A2_and (i32 (LDrib ADDRriS11_0:$addr)), (A2_tfrsi 0x1)))>;
+ (i32 (A2_and (i32 (L2_loadrb_io AddrFI:$addr, 0)), (A2_tfrsi 0x1)))>;
// Map from Rdd = sign_extend_inreg(Rss, i32) -> Rdd = A2_sxtw(Rss.lo).
def : Pat <(i64 (sext_inreg (i64 DoubleRegs:$src1), i32)),
// Map Pd = load(addr) -> Rs = load(addr); Pd = Rs.
def : Pat<(i1 (load ADDRriS11_2:$addr)),
- (i1 (C2_tfrrp (i32 (LDrib ADDRriS11_2:$addr))))>;
+ (i1 (C2_tfrrp (i32 (L2_loadrb_io AddrFI:$addr, 0))))>;
// Map for truncating from 64 immediates to 32 bit immediates.
def : Pat<(i32 (trunc (i64 DoubleRegs:$src))),
// Map memb(Rs) = Rdd -> memb(Rs) = Rt.
def : Pat<(truncstorei8 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
- (STrib ADDRriS11_0:$addr, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
+ (S2_storerb_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
subreg_loreg)))>;
// Map memh(Rs) = Rdd -> memh(Rs) = Rt.
def : Pat<(truncstorei16 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
- (STrih ADDRriS11_0:$addr, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
+ (S2_storerh_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
subreg_loreg)))>;
// Map memw(Rs) = Rdd -> memw(Rs) = Rt
def : Pat<(truncstorei32 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
- (STriw ADDRriS11_0:$addr, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
+ (S2_storeri_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
subreg_loreg)))>;
// Map memw(Rs) = Rdd -> memw(Rs) = Rt.
def : Pat<(truncstorei32 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
- (STriw ADDRriS11_0:$addr, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
+ (S2_storeri_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
subreg_loreg)))>;
// Map from i1 = constant<-1>; memw(addr) = i1 -> r0 = 1; memw(addr) = r0.
def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
- (STrib ADDRriS11_2:$addr, (A2_tfrsi 1))>;
+ (S2_storerb_io AddrFI:$addr, 0, (A2_tfrsi 1))>;
// Map from i1 = constant<-1>; store i1 -> r0 = 1; store r0.
def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
- (STrib ADDRriS11_2:$addr, (A2_tfrsi 1))>;
+ (S2_storerb_io AddrFI:$addr, 0, (A2_tfrsi 1))>;
// Map from memb(Rs) = Pd -> Rt = mux(Pd, #0, #1); store Rt.
def : Pat<(store (i1 PredRegs:$src1), ADDRriS11_2:$addr),
- (STrib ADDRriS11_2:$addr, (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0)) )>;
+ (S2_storerb_io AddrFI:$addr, 0, (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0)) )>;
// Map Rdd = anyext(Rs) -> Rdd = A2_sxtw(Rs).
// Hexagon_TODO: We can probably use combine but that will cost 2 instructions.
// Convert sign-extended load back to load and sign extend.
// i8 -> i64
def: Pat <(i64 (sextloadi8 ADDRriS11_0:$src1)),
- (i64 (A2_sxtw (LDrib ADDRriS11_0:$src1)))>;
+ (i64 (A2_sxtw (L2_loadrb_io AddrFI:$src1, 0)))>;
// Convert any-extended load back to load and sign extend.
// i8 -> i64
def: Pat <(i64 (extloadi8 ADDRriS11_0:$src1)),
- (i64 (A2_sxtw (LDrib ADDRriS11_0:$src1)))>;
+ (i64 (A2_sxtw (L2_loadrb_io AddrFI:$src1, 0)))>;
// Convert sign-extended load back to load and sign extend.
// i16 -> i64
def: Pat <(i64 (sextloadi16 ADDRriS11_1:$src1)),
- (i64 (A2_sxtw (LDrih ADDRriS11_1:$src1)))>;
+ (i64 (A2_sxtw (L2_loadrh_io AddrFI:$src1, 0)))>;
// Convert sign-extended load back to load and sign extend.
// i32 -> i64
def: Pat <(i64 (sextloadi32 ADDRriS11_2:$src1)),
- (i64 (A2_sxtw (LDriw ADDRriS11_2:$src1)))>;
+ (i64 (A2_sxtw (L2_loadri_io AddrFI:$src1, 0)))>;
// Zero extends.
// i8 -> i64
def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriub ADDRriS11_0:$src1)))>,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io AddrFI:$src1, 0)))>,
Requires<[NoV4T]>;
let AddedComplexity = 20 in
def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
s11_0ExtPred:$offset))),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriub_indexed IntRegs:$src1,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io IntRegs:$src1,
s11_0ExtPred:$offset)))>,
Requires<[NoV4T]>;
// i1 -> i64
def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriub ADDRriS11_0:$src1)))>,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io AddrFI:$src1, 0)))>,
Requires<[NoV4T]>;
let AddedComplexity = 20 in
def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
s11_0ExtPred:$offset))),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriub_indexed IntRegs:$src1,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io IntRegs:$src1,
s11_0ExtPred:$offset)))>,
Requires<[NoV4T]>;
// i16 -> i64
def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriuh ADDRriS11_1:$src1)))>,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadruh_io AddrFI:$src1, 0)))>,
Requires<[NoV4T]>;
let AddedComplexity = 20 in
def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
s11_1ExtPred:$offset))),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriuh_indexed IntRegs:$src1,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadruh_io IntRegs:$src1,
s11_1ExtPred:$offset)))>,
Requires<[NoV4T]>;
// i32 -> i64
def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriw ADDRriS11_2:$src1)))>,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io AddrFI:$src1, 0)))>,
Requires<[NoV4T]>;
let AddedComplexity = 100 in
def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriw_indexed IntRegs:$src1,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io IntRegs:$src1,
s11_2ExtPred:$offset)))>,
Requires<[NoV4T]>;
let AddedComplexity = 10 in
def: Pat <(i32 (zextloadi1 ADDRriS11_0:$src1)),
- (i32 (LDriw ADDRriS11_0:$src1))>;
+ (i32 (L2_loadri_io AddrFI:$src1, 0))>;
// Map from Rs = Pd to Pd = mux(Pd, #1, #0)
def : Pat <(i32 (zext (i1 PredRegs:$src1))),
(i64 (zextloadi32 (i32 (add IntRegs:$src2,
s11_2ExtPred:$offset2)))))),
(i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
- (LDriw_indexed IntRegs:$src2,
+ (L2_loadri_io IntRegs:$src2,
s11_2ExtPred:$offset2)))>;
def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
(i32 32))),
(i64 (zextloadi32 ADDRriS11_2:$srcLow)))),
(i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
- (LDriw ADDRriS11_2:$srcLow)))>;
+ (L2_loadri_io AddrFI:$srcLow, 0)))>;
def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
(i32 32))),
(i64 (zextloadi32 (i32 (add IntRegs:$src2,
s11_2ExtPred:$offset2)))))),
(i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
- (LDriw_indexed IntRegs:$src2,
+ (L2_loadri_io IntRegs:$src2,
s11_2ExtPred:$offset2)))>;
def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
(i32 32))),
(i64 (zextloadi32 ADDRriS11_2:$srcLow)))),
(i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
- (LDriw ADDRriS11_2:$srcLow)))>;
+ (L2_loadri_io AddrFI:$srcLow, 0)))>;
def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
(i32 32))),
// Any extended 64-bit load.
// anyext i32 -> i64
def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriw ADDRriS11_2:$src1)))>,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io AddrFI:$src1, 0)))>,
Requires<[NoV4T]>;
// When there is an offset we should prefer the pattern below over the pattern above.
// ********************************************
let AddedComplexity = 100 in
def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
- (i64 (A2_combinew (A2_tfrsi 0), (LDriw_indexed IntRegs:$src1,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io IntRegs:$src1,
s11_2ExtPred:$offset)))>,
Requires<[NoV4T]>;
// anyext i16 -> i64.
def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
- (i64 (A2_combinew (A2_tfrsi 0), (LDrih ADDRriS11_2:$src1)))>,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrh_io AddrFI:$src1, 0)))>,
Requires<[NoV4T]>;
let AddedComplexity = 20 in
def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
s11_1ExtPred:$offset))),
- (i64 (A2_combinew (A2_tfrsi 0), (LDrih_indexed IntRegs:$src1,
+ (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrh_io IntRegs:$src1,
s11_1ExtPred:$offset)))>,
Requires<[NoV4T]>;
(i32 (CONST32_set_jt tjumptable:$dst))>;
// XTYPE/SHIFT
-
-// Multi-class for logical operators :
+//
+//===----------------------------------------------------------------------===//
+// Template Class
// Shift by immediate/register and accumulate/logical
-multiclass xtype_imm<string OpcStr, SDNode OpNode1, SDNode OpNode2> {
- def _ri : SInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs:$src2, u5Imm:$src3),
- !strconcat("$dst ", !strconcat(OpcStr, "($src2, #$src3)")),
- [(set (i32 IntRegs:$dst),
- (OpNode2 (i32 IntRegs:$src1),
- (OpNode1 (i32 IntRegs:$src2),
- u5ImmPred:$src3)))],
- "$src1 = $dst">;
-
- def d_ri : SInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2, u6Imm:$src3),
- !strconcat("$dst ", !strconcat(OpcStr, "($src2, #$src3)")),
- [(set (i64 DoubleRegs:$dst), (OpNode2 (i64 DoubleRegs:$src1),
- (OpNode1 (i64 DoubleRegs:$src2), u6ImmPred:$src3)))],
- "$src1 = $dst">;
-}
-
-// Multi-class for logical operators :
-// Shift by register and accumulate/logical (32/64 bits)
-multiclass xtype_reg<string OpcStr, SDNode OpNode1, SDNode OpNode2> {
- def _rr : SInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
- !strconcat("$dst ", !strconcat(OpcStr, "($src2, $src3)")),
- [(set (i32 IntRegs:$dst),
- (OpNode2 (i32 IntRegs:$src1),
- (OpNode1 (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">;
+//===----------------------------------------------------------------------===//
+
+// Rx[+-&|]=asr(Rs,#u5)
+// Rx[+-&|^]=lsr(Rs,#u5)
+// Rx[+-&|^]=asl(Rs,#u5)
- def d_rr : SInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
- !strconcat("$dst ", !strconcat(OpcStr, "($src2, $src3)")),
- [(set (i64 DoubleRegs:$dst),
- (OpNode2 (i64 DoubleRegs:$src1),
- (OpNode1 (i64 DoubleRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">;
+let hasNewValue = 1, opNewValue = 0 in
+class T_shift_imm_acc_r <string opc1, string opc2, SDNode OpNode1,
+ SDNode OpNode2, bits<3> majOp, bits<2> minOp>
+ : SInst_acc<(outs IntRegs:$Rx),
+ (ins IntRegs:$src1, IntRegs:$Rs, u5Imm:$u5),
+ "$Rx "#opc2#opc1#"($Rs, #$u5)",
+ [(set (i32 IntRegs:$Rx),
+ (OpNode2 (i32 IntRegs:$src1),
+ (OpNode1 (i32 IntRegs:$Rs), u5ImmPred:$u5)))],
+ "$src1 = $Rx", S_2op_tc_2_SLOT23> {
+ bits<5> Rx;
+ bits<5> Rs;
+ bits<5> u5;
+
+ let IClass = 0b1000;
+
+ let Inst{27-24} = 0b1110;
+ let Inst{23-22} = majOp{2-1};
+ let Inst{13} = 0b0;
+ let Inst{7} = majOp{0};
+ let Inst{6-5} = minOp;
+ let Inst{4-0} = Rx;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = u5;
+ }
+
+// Rx[+-&|]=asr(Rs,Rt)
+// Rx[+-&|^]=lsr(Rs,Rt)
+// Rx[+-&|^]=asl(Rs,Rt)
+
+let hasNewValue = 1, opNewValue = 0 in
+class T_shift_reg_acc_r <string opc1, string opc2, SDNode OpNode1,
+ SDNode OpNode2, bits<2> majOp, bits<2> minOp>
+ : SInst_acc<(outs IntRegs:$Rx),
+ (ins IntRegs:$src1, IntRegs:$Rs, IntRegs:$Rt),
+ "$Rx "#opc2#opc1#"($Rs, $Rt)",
+ [(set (i32 IntRegs:$Rx),
+ (OpNode2 (i32 IntRegs:$src1),
+ (OpNode1 (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))],
+ "$src1 = $Rx", S_3op_tc_2_SLOT23 > {
+ bits<5> Rx;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1100;
+
+ let Inst{27-24} = 0b1100;
+ let Inst{23-22} = majOp;
+ let Inst{7-6} = minOp;
+ let Inst{4-0} = Rx;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rt;
+ }
+// Rxx[+-&|]=asr(Rss,#u6)
+// Rxx[+-&|^]=lsr(Rss,#u6)
+// Rxx[+-&|^]=asl(Rss,#u6)
+
+class T_shift_imm_acc_p <string opc1, string opc2, SDNode OpNode1,
+ SDNode OpNode2, bits<3> majOp, bits<2> minOp>
+ : SInst_acc<(outs DoubleRegs:$Rxx),
+ (ins DoubleRegs:$src1, DoubleRegs:$Rss, u6Imm:$u6),
+ "$Rxx "#opc2#opc1#"($Rss, #$u6)",
+ [(set (i64 DoubleRegs:$Rxx),
+ (OpNode2 (i64 DoubleRegs:$src1),
+ (OpNode1 (i64 DoubleRegs:$Rss), u6ImmPred:$u6)))],
+ "$src1 = $Rxx", S_2op_tc_2_SLOT23> {
+ bits<5> Rxx;
+ bits<5> Rss;
+ bits<6> u6;
+
+ let IClass = 0b1000;
+
+ let Inst{27-24} = 0b0010;
+ let Inst{23-22} = majOp{2-1};
+ let Inst{7} = majOp{0};
+ let Inst{6-5} = minOp;
+ let Inst{4-0} = Rxx;
+ let Inst{20-16} = Rss;
+ let Inst{13-8} = u6;
+ }
+
+
+// Rxx[+-&|]=asr(Rss,Rt)
+// Rxx[+-&|^]=lsr(Rss,Rt)
+// Rxx[+-&|^]=asl(Rss,Rt)
+// Rxx[+-&|^]=lsl(Rss,Rt)
+
+class T_shift_reg_acc_p <string opc1, string opc2, SDNode OpNode1,
+ SDNode OpNode2, bits<3> majOp, bits<2> minOp>
+ : SInst_acc<(outs DoubleRegs:$Rxx),
+ (ins DoubleRegs:$src1, DoubleRegs:$Rss, IntRegs:$Rt),
+ "$Rxx "#opc2#opc1#"($Rss, $Rt)",
+ [(set (i64 DoubleRegs:$Rxx),
+ (OpNode2 (i64 DoubleRegs:$src1),
+ (OpNode1 (i64 DoubleRegs:$Rss), (i32 IntRegs:$Rt))))],
+ "$src1 = $Rxx", S_3op_tc_2_SLOT23> {
+ bits<5> Rxx;
+ bits<5> Rss;
+ bits<5> Rt;
+
+ let IClass = 0b1100;
+
+ let Inst{27-24} = 0b1011;
+ let Inst{23-21} = majOp;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rt;
+ let Inst{7-6} = minOp;
+ let Inst{4-0} = Rxx;
+ }
+
+//===----------------------------------------------------------------------===//
+// Multi-class for the shift instructions with logical/arithmetic operators.
+//===----------------------------------------------------------------------===//
+
+multiclass xtype_imm_base<string OpcStr1, string OpcStr2, SDNode OpNode1,
+ SDNode OpNode2, bits<3> majOp, bits<2> minOp > {
+ def _i_r#NAME : T_shift_imm_acc_r< OpcStr1, OpcStr2, OpNode1,
+ OpNode2, majOp, minOp >;
+ def _i_p#NAME : T_shift_imm_acc_p< OpcStr1, OpcStr2, OpNode1,
+ OpNode2, majOp, minOp >;
}
-multiclass basic_xtype_imm<string OpcStr, SDNode OpNode> {
-let AddedComplexity = 100 in
- defm _ADD : xtype_imm< !strconcat("+= ", OpcStr), OpNode, add>;
- defm _SUB : xtype_imm< !strconcat("-= ", OpcStr), OpNode, sub>;
- defm _AND : xtype_imm< !strconcat("&= ", OpcStr), OpNode, and>;
- defm _OR : xtype_imm< !strconcat("|= ", OpcStr), OpNode, or>;
+multiclass xtype_imm_acc<string opc1, SDNode OpNode, bits<2>minOp> {
+ let AddedComplexity = 100 in
+ defm _acc : xtype_imm_base< opc1, "+= ", OpNode, add, 0b001, minOp>;
+
+ defm _nac : xtype_imm_base< opc1, "-= ", OpNode, sub, 0b000, minOp>;
+ defm _and : xtype_imm_base< opc1, "&= ", OpNode, and, 0b010, minOp>;
+ defm _or : xtype_imm_base< opc1, "|= ", OpNode, or, 0b011, minOp>;
}
-multiclass basic_xtype_reg<string OpcStr, SDNode OpNode> {
+multiclass xtype_xor_imm_acc<string opc1, SDNode OpNode, bits<2>minOp> {
let AddedComplexity = 100 in
- defm _ADD : xtype_reg< !strconcat("+= ", OpcStr), OpNode, add>;
- defm _SUB : xtype_reg< !strconcat("-= ", OpcStr), OpNode, sub>;
- defm _AND : xtype_reg< !strconcat("&= ", OpcStr), OpNode, and>;
- defm _OR : xtype_reg< !strconcat("|= ", OpcStr), OpNode, or>;
+ defm _xacc : xtype_imm_base< opc1, "^= ", OpNode, xor, 0b100, minOp>;
}
-multiclass xtype_xor_imm<string OpcStr, SDNode OpNode> {
-let AddedComplexity = 100 in
- defm _XOR : xtype_imm< !strconcat("^= ", OpcStr), OpNode, xor>;
+let isCodeGenOnly = 0 in {
+defm S2_asr : xtype_imm_acc<"asr", sra, 0b00>;
+
+defm S2_lsr : xtype_imm_acc<"lsr", srl, 0b01>,
+ xtype_xor_imm_acc<"lsr", srl, 0b01>;
+
+defm S2_asl : xtype_imm_acc<"asl", shl, 0b10>,
+ xtype_xor_imm_acc<"asl", shl, 0b10>;
}
-defm ASL : basic_xtype_imm<"asl", shl>, basic_xtype_reg<"asl", shl>,
- xtype_xor_imm<"asl", shl>;
+multiclass xtype_reg_acc_r<string opc1, SDNode OpNode, bits<2>minOp> {
+ let AddedComplexity = 100 in
+ def _acc : T_shift_reg_acc_r <opc1, "+= ", OpNode, add, 0b11, minOp>;
-defm LSR : basic_xtype_imm<"lsr", srl>, basic_xtype_reg<"lsr", srl>,
- xtype_xor_imm<"lsr", srl>;
+ def _nac : T_shift_reg_acc_r <opc1, "-= ", OpNode, sub, 0b10, minOp>;
+ def _and : T_shift_reg_acc_r <opc1, "&= ", OpNode, and, 0b01, minOp>;
+ def _or : T_shift_reg_acc_r <opc1, "|= ", OpNode, or, 0b00, minOp>;
+}
+
+multiclass xtype_reg_acc_p<string opc1, SDNode OpNode, bits<2>minOp> {
+ let AddedComplexity = 100 in
+ def _acc : T_shift_reg_acc_p <opc1, "+= ", OpNode, add, 0b110, minOp>;
+
+ def _nac : T_shift_reg_acc_p <opc1, "-= ", OpNode, sub, 0b100, minOp>;
+ def _and : T_shift_reg_acc_p <opc1, "&= ", OpNode, and, 0b010, minOp>;
+ def _or : T_shift_reg_acc_p <opc1, "|= ", OpNode, or, 0b000, minOp>;
+ def _xor : T_shift_reg_acc_p <opc1, "^= ", OpNode, xor, 0b011, minOp>;
+}
+
+multiclass xtype_reg_acc<string OpcStr, SDNode OpNode, bits<2> minOp > {
+ defm _r_r : xtype_reg_acc_r <OpcStr, OpNode, minOp>;
+ defm _r_p : xtype_reg_acc_p <OpcStr, OpNode, minOp>;
+}
+
+let isCodeGenOnly = 0 in {
+defm S2_asl : xtype_reg_acc<"asl", shl, 0b10>;
+defm S2_asr : xtype_reg_acc<"asr", sra, 0b00>;
+defm S2_lsr : xtype_reg_acc<"lsr", srl, 0b01>;
+defm S2_lsl : xtype_reg_acc<"lsl", shl, 0b11>;
+}
+
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0 in
+class T_S3op_1 <string mnemonic, RegisterClass RC, bits<2> MajOp, bits<3> MinOp,
+ bit SwapOps, bit isSat = 0, bit isRnd = 0, bit hasShift = 0>
+ : SInst <(outs RC:$dst),
+ (ins DoubleRegs:$src1, DoubleRegs:$src2),
+ "$dst = "#mnemonic#"($src1, $src2)"#!if(isRnd, ":rnd", "")
+ #!if(hasShift,":>>1","")
+ #!if(isSat, ":sat", ""),
+ [], "", S_3op_tc_2_SLOT23 > {
+ bits<5> dst;
+ bits<5> src1;
+ bits<5> src2;
+
+ let IClass = 0b1100;
+
+ let Inst{27-24} = 0b0001;
+ let Inst{23-22} = MajOp;
+ let Inst{20-16} = !if (SwapOps, src2, src1);
+ let Inst{12-8} = !if (SwapOps, src1, src2);
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = dst;
+ }
+
+class T_S3op_64 <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit SwapOps,
+ bit isSat = 0, bit isRnd = 0, bit hasShift = 0 >
+ : T_S3op_1 <mnemonic, DoubleRegs, MajOp, MinOp, SwapOps,
+ isSat, isRnd, hasShift>;
+
+let isCodeGenOnly = 0 in
+def S2_lfsp : T_S3op_64 < "lfs", 0b10, 0b110, 0>;
+
+let hasSideEffects = 0 in
+class T_S3op_2 <string mnemonic, bits<3> MajOp, bit SwapOps>
+ : SInst < (outs DoubleRegs:$Rdd),
+ (ins DoubleRegs:$Rss, DoubleRegs:$Rtt, PredRegs:$Pu),
+ "$Rdd = "#mnemonic#"($Rss, $Rtt, $Pu)",
+ [], "", S_3op_tc_1_SLOT23 > {
+ bits<5> Rdd;
+ bits<5> Rss;
+ bits<5> Rtt;
+ bits<2> Pu;
+
+ let IClass = 0b1100;
+
+ let Inst{27-24} = 0b0010;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = !if (SwapOps, Rtt, Rss);
+ let Inst{12-8} = !if (SwapOps, Rss, Rtt);
+ let Inst{6-5} = Pu;
+ let Inst{4-0} = Rdd;
+ }
+
+let isCodeGenOnly = 0 in {
+def S2_valignrb : T_S3op_2 < "valignb", 0b000, 1>;
+def S2_vsplicerb : T_S3op_2 < "vspliceb", 0b100, 0>;
+}
+
+//===----------------------------------------------------------------------===//
+// Template class used by vector shift, vector rotate, vector neg,
+// 32-bit shift, 64-bit shifts, etc.
+//===----------------------------------------------------------------------===//
+
+let hasSideEffects = 0 in
+class T_S3op_3 <string mnemonic, RegisterClass RC, bits<2> MajOp,
+ bits<2> MinOp, bit isSat = 0, list<dag> pattern = [] >
+ : SInst <(outs RC:$dst),
+ (ins RC:$src1, IntRegs:$src2),
+ "$dst = "#mnemonic#"($src1, $src2)"#!if(isSat, ":sat", ""),
+ pattern, "", S_3op_tc_1_SLOT23> {
+ bits<5> dst;
+ bits<5> src1;
+ bits<5> src2;
+
+ let IClass = 0b1100;
+
+ let Inst{27-24} = !if(!eq(!cast<string>(RC), "IntRegs"), 0b0110, 0b0011);
+ let Inst{23-22} = MajOp;
+ let Inst{20-16} = src1;
+ let Inst{12-8} = src2;
+ let Inst{7-6} = MinOp;
+ let Inst{4-0} = dst;
+ }
-defm ASR : basic_xtype_imm<"asr", sra>, basic_xtype_reg<"asr", sra>;
-defm LSL : basic_xtype_reg<"lsl", shl>;
+let hasNewValue = 1 in
+class T_S3op_shift32 <string mnemonic, SDNode OpNode, bits<2> MinOp>
+ : T_S3op_3 <mnemonic, IntRegs, 0b01, MinOp, 0,
+ [(set (i32 IntRegs:$dst), (OpNode (i32 IntRegs:$src1),
+ (i32 IntRegs:$src2)))]>;
+
+let hasNewValue = 1, Itinerary = S_3op_tc_2_SLOT23 in
+class T_S3op_shift32_Sat <string mnemonic, bits<2> MinOp>
+ : T_S3op_3 <mnemonic, IntRegs, 0b00, MinOp, 1, []>;
+
+
+class T_S3op_shift64 <string mnemonic, SDNode OpNode, bits<2> MinOp>
+ : T_S3op_3 <mnemonic, DoubleRegs, 0b10, MinOp, 0,
+ [(set (i64 DoubleRegs:$dst), (OpNode (i64 DoubleRegs:$src1),
+ (i32 IntRegs:$src2)))]>;
+
+
+class T_S3op_shiftVect <string mnemonic, bits<2> MajOp, bits<2> MinOp>
+ : T_S3op_3 <mnemonic, DoubleRegs, MajOp, MinOp, 0, []>;
+
+
+// Shift by register
+// Rdd=[asr|lsr|asl|lsl](Rss,Rt)
+
+let isCodeGenOnly = 0 in {
+def S2_asr_r_p : T_S3op_shift64 < "asr", sra, 0b00>;
+def S2_lsr_r_p : T_S3op_shift64 < "lsr", srl, 0b01>;
+def S2_asl_r_p : T_S3op_shift64 < "asl", shl, 0b10>;
+def S2_lsl_r_p : T_S3op_shift64 < "lsl", shl, 0b11>;
+}
+
+// Rd=[asr|lsr|asl|lsl](Rs,Rt)
+
+let isCodeGenOnly = 0 in {
+def S2_asr_r_r : T_S3op_shift32<"asr", sra, 0b00>;
+def S2_lsr_r_r : T_S3op_shift32<"lsr", srl, 0b01>;
+def S2_asl_r_r : T_S3op_shift32<"asl", shl, 0b10>;
+def S2_lsl_r_r : T_S3op_shift32<"lsl", shl, 0b11>;
+}
+
+// Shift by register with saturation
+// Rd=asr(Rs,Rt):sat
+// Rd=asl(Rs,Rt):sat
+
+let Defs = [USR_OVF], isCodeGenOnly = 0 in {
+ def S2_asr_r_r_sat : T_S3op_shift32_Sat<"asr", 0b00>;
+ def S2_asl_r_r_sat : T_S3op_shift32_Sat<"asl", 0b10>;
+}
+
+let hasNewValue = 1, hasSideEffects = 0 in
+class T_S3op_8 <string opc, bits<3> MinOp, bit isSat, bit isRnd, bit hasShift, bit hasSplat = 0>
+ : SInst < (outs IntRegs:$Rd),
+ (ins DoubleRegs:$Rss, IntRegs:$Rt),
+ "$Rd = "#opc#"($Rss, $Rt"#!if(hasSplat, "*", "")#")"
+ #!if(hasShift, ":<<1", "")
+ #!if(isRnd, ":rnd", "")
+ #!if(isSat, ":sat", ""),
+ [], "", S_3op_tc_1_SLOT23 > {
+ bits<5> Rd;
+ bits<5> Rss;
+ bits<5> Rt;
+
+ let IClass = 0b1100;
+
+ let Inst{27-24} = 0b0101;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rt;
+ let Inst{7-5} = MinOp;
+ let Inst{4-0} = Rd;
+ }
+
+let Defs = [USR_OVF], Itinerary = S_3op_tc_2_SLOT23, isCodeGenOnly = 0 in
+def S2_vcrotate : T_S3op_shiftVect < "vcrotate", 0b11, 0b00>;
+
+//===----------------------------------------------------------------------===//
+// Template class for 'insert bitfield' instructions
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0 in
+class T_S3op_insert <string mnemonic, RegisterClass RC>
+ : SInst <(outs RC:$dst),
+ (ins RC:$src1, RC:$src2, DoubleRegs:$src3),
+ "$dst = "#mnemonic#"($src2, $src3)" ,
+ [], "$src1 = $dst", S_3op_tc_1_SLOT23 > {
+ bits<5> dst;
+ bits<5> src2;
+ bits<5> src3;
+
+ let IClass = 0b1100;
+
+ let Inst{27-26} = 0b10;
+ let Inst{25-24} = !if(!eq(!cast<string>(RC), "IntRegs"), 0b00, 0b10);
+ let Inst{23} = 0b0;
+ let Inst{20-16} = src2;
+ let Inst{12-8} = src3;
+ let Inst{4-0} = dst;
+ }
+
+let hasSideEffects = 0 in
+class T_S2op_insert <bits<4> RegTyBits, RegisterClass RC, Operand ImmOp>
+ : SInst <(outs RC:$dst), (ins RC:$dst2, RC:$src1, ImmOp:$src2, ImmOp:$src3),
+ "$dst = insert($src1, #$src2, #$src3)",
+ [], "$dst2 = $dst", S_2op_tc_2_SLOT23> {
+ bits<5> dst;
+ bits<5> src1;
+ bits<6> src2;
+ bits<6> src3;
+ bit bit23;
+ bit bit13;
+ string ImmOpStr = !cast<string>(ImmOp);
+
+ let bit23 = !if (!eq(ImmOpStr, "u6Imm"), src3{5}, 0);
+ let bit13 = !if (!eq(ImmOpStr, "u6Imm"), src2{5}, 0);
+
+ let IClass = 0b1000;
+
+ let Inst{27-24} = RegTyBits;
+ let Inst{23} = bit23;
+ let Inst{22-21} = src3{4-3};
+ let Inst{20-16} = src1;
+ let Inst{13} = bit13;
+ let Inst{12-8} = src2{4-0};
+ let Inst{7-5} = src3{2-0};
+ let Inst{4-0} = dst;
+ }
+
+// Rx=insert(Rs,Rtt)
+// Rx=insert(Rs,#u5,#U5)
+let hasNewValue = 1, isCodeGenOnly = 0 in {
+ def S2_insert_rp : T_S3op_insert <"insert", IntRegs>;
+ def S2_insert : T_S2op_insert <0b1111, IntRegs, u5Imm>;
+}
+
+// Rxx=insert(Rss,Rtt)
+// Rxx=insert(Rss,#u6,#U6)
+let isCodeGenOnly = 0 in {
+def S2_insertp_rp : T_S3op_insert<"insert", DoubleRegs>;
+def S2_insertp : T_S2op_insert <0b0011, DoubleRegs, u6Imm>;
+}
+
+//===----------------------------------------------------------------------===//
+// Template class for 'extract bitfield' instructions
+//===----------------------------------------------------------------------===//
+let hasNewValue = 1, hasSideEffects = 0 in
+class T_S3op_extract <string mnemonic, bits<2> MinOp>
+ : SInst <(outs IntRegs:$Rd), (ins IntRegs:$Rs, DoubleRegs:$Rtt),
+ "$Rd = "#mnemonic#"($Rs, $Rtt)",
+ [], "", S_3op_tc_2_SLOT23 > {
+ bits<5> Rd;
+ bits<5> Rs;
+ bits<5> Rtt;
+
+ let IClass = 0b1100;
+
+ let Inst{27-22} = 0b100100;
+ let Inst{20-16} = Rs;
+ let Inst{12-8} = Rtt;
+ let Inst{7-6} = MinOp;
+ let Inst{4-0} = Rd;
+ }
+
+let hasSideEffects = 0 in
+class T_S2op_extract <string mnemonic, bits<4> RegTyBits,
+ RegisterClass RC, Operand ImmOp>
+ : SInst <(outs RC:$dst), (ins RC:$src1, ImmOp:$src2, ImmOp:$src3),
+ "$dst = "#mnemonic#"($src1, #$src2, #$src3)",
+ [], "", S_2op_tc_2_SLOT23> {
+ bits<5> dst;
+ bits<5> src1;
+ bits<6> src2;
+ bits<6> src3;
+ bit bit23;
+ bit bit13;
+ string ImmOpStr = !cast<string>(ImmOp);
+
+ let bit23 = !if (!eq(ImmOpStr, "u6Imm"), src3{5},
+ !if (!eq(mnemonic, "extractu"), 0, 1));
+
+ let bit13 = !if (!eq(ImmOpStr, "u6Imm"), src2{5}, 0);
+
+ let IClass = 0b1000;
+
+ let Inst{27-24} = RegTyBits;
+ let Inst{23} = bit23;
+ let Inst{22-21} = src3{4-3};
+ let Inst{20-16} = src1;
+ let Inst{13} = bit13;
+ let Inst{12-8} = src2{4-0};
+ let Inst{7-5} = src3{2-0};
+ let Inst{4-0} = dst;
+ }
+
+// Extract bitfield
+
+// Rdd=extractu(Rss,Rtt)
+// Rdd=extractu(Rss,#u6,#U6)
+let isCodeGenOnly = 0 in {
+def S2_extractup_rp : T_S3op_64 < "extractu", 0b00, 0b000, 0>;
+def S2_extractup : T_S2op_extract <"extractu", 0b0001, DoubleRegs, u6Imm>;
+}
+
+// Rd=extractu(Rs,Rtt)
+// Rd=extractu(Rs,#u5,#U5)
+let hasNewValue = 1, isCodeGenOnly = 0 in {
+ def S2_extractu_rp : T_S3op_extract<"extractu", 0b00>;
+ def S2_extractu : T_S2op_extract <"extractu", 0b1101, IntRegs, u5Imm>;
+}
+
+//===----------------------------------------------------------------------===//
+// :raw for of tableindx[bdhw] insns
+//===----------------------------------------------------------------------===//
+
+let hasSideEffects = 0, hasNewValue = 1, opNewValue = 0 in
+class tableidxRaw<string OpStr, bits<2>MinOp>
+ : SInst <(outs IntRegs:$Rx),
+ (ins IntRegs:$_dst_, IntRegs:$Rs, u4Imm:$u4, s6Imm:$S6),
+ "$Rx = "#OpStr#"($Rs, #$u4, #$S6):raw",
+ [], "$Rx = $_dst_" > {
+ bits<5> Rx;
+ bits<5> Rs;
+ bits<4> u4;
+ bits<6> S6;
+
+ let IClass = 0b1000;
+
+ let Inst{27-24} = 0b0111;
+ let Inst{23-22} = MinOp;
+ let Inst{21} = u4{3};
+ let Inst{20-16} = Rs;
+ let Inst{13-8} = S6;
+ let Inst{7-5} = u4{2-0};
+ let Inst{4-0} = Rx;
+ }
+
+let isCodeGenOnly = 0 in {
+def S2_tableidxb : tableidxRaw<"tableidxb", 0b00>;
+def S2_tableidxh : tableidxRaw<"tableidxh", 0b01>;
+def S2_tableidxw : tableidxRaw<"tableidxw", 0b10>;
+def S2_tableidxd : tableidxRaw<"tableidxd", 0b11>;
+}
// Change the sign of the immediate for Rd=-mpyi(Rs,#u8)
def : Pat <(mul (i32 IntRegs:$src1), (ineg n8ImmPred:$src2)),
projects / oota-llvm.git / blobdiff