X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FHexagon%2FHexagonInstrInfoV4.td;h=65b0f49743679e2d4055aa9921d123cf336f5e8e;hb=083f0122f0398137c45659518e9be00748d24b2d;hp=1d81390fe10e1b146556dde4baea1dcd06c5196c;hpb=066f43435ae586a07d30f56f033ed613625be208;p=oota-llvm.git diff --git a/lib/Target/Hexagon/HexagonInstrInfoV4.td b/lib/Target/Hexagon/HexagonInstrInfoV4.td index 1d81390fe10..65b0f497436 100644 --- a/lib/Target/Hexagon/HexagonInstrInfoV4.td +++ b/lib/Target/Hexagon/HexagonInstrInfoV4.td @@ -11,6 +11,28 @@ // //===----------------------------------------------------------------------===// +def DuplexIClass0: InstDuplex < 0 >; +def DuplexIClass1: InstDuplex < 1 >; +def DuplexIClass2: InstDuplex < 2 >; +let isExtendable = 1 in { + def DuplexIClass3: InstDuplex < 3 >; + def DuplexIClass4: InstDuplex < 4 >; + def DuplexIClass5: InstDuplex < 5 >; + def DuplexIClass6: InstDuplex < 6 >; + def DuplexIClass7: InstDuplex < 7 >; +} +def DuplexIClass8: InstDuplex < 8 >; +def DuplexIClass9: InstDuplex < 9 >; +def DuplexIClassA: InstDuplex < 0xA >; +def DuplexIClassB: InstDuplex < 0xB >; +def DuplexIClassC: InstDuplex < 0xC >; +def DuplexIClassD: InstDuplex < 0xD >; +def DuplexIClassE: InstDuplex < 0xE >; +def DuplexIClassF: InstDuplex < 0xF >; + +def addrga: PatLeaf<(i32 AddrGA:$Addr)>; +def addrgp: PatLeaf<(i32 AddrGP:$Addr)>; + let hasSideEffects = 0 in class T_Immext : EXTENDERInst<(outs), (ins ImmType:$imm), @@ -31,17 +53,14 @@ let isCodeGenOnly = 1 in { def A4_ext_g : T_Immext; } -// Fold (add (CONST32 tglobaladdr:$addr) ) into a global address. -def FoldGlobalAddr : ComplexPattern; - -// Fold (add (CONST32_GP tglobaladdr:$addr) ) into a global address. -def FoldGlobalAddrGP : ComplexPattern; - -def NumUsesBelowThresCONST32 : PatFrag<(ops node:$addr), - (HexagonCONST32 node:$addr), [{ - return hasNumUsesBelowThresGA(N->getOperand(0).getNode()); +def BITPOS32 : SDNodeXFormgetSExtValue(); + return XformMskToBitPosU5Imm(imm, SDLoc(N)); }]>; + // Hexagon V4 Architecture spec defines 8 instruction classes: // LD ST ALU32 XTYPE J JR MEMOP NV CR SYSTEM(system is not implemented in the // compiler) @@ -112,21 +131,19 @@ class T_ALU32_3op_not MajOp, bits<3> MinOp, let AsmString = "$Rd = "#mnemonic#"($Rs, ~$Rt)"; } -let BaseOpcode = "andn_rr", CextOpcode = "andn", isCodeGenOnly = 0 in +let BaseOpcode = "andn_rr", CextOpcode = "andn" in def A4_andn : T_ALU32_3op_not<"and", 0b001, 0b100, 1>; -let BaseOpcode = "orn_rr", CextOpcode = "orn", isCodeGenOnly = 0 in +let BaseOpcode = "orn_rr", CextOpcode = "orn" in def A4_orn : T_ALU32_3op_not<"or", 0b001, 0b101, 1>; -let CextOpcode = "rcmp.eq", isCodeGenOnly = 0 in +let CextOpcode = "rcmp.eq" in def A4_rcmpeq : T_ALU32_3op<"cmp.eq", 0b011, 0b010, 0, 1>; -let CextOpcode = "!rcmp.eq", isCodeGenOnly = 0 in +let CextOpcode = "!rcmp.eq" in def A4_rcmpneq : T_ALU32_3op<"!cmp.eq", 0b011, 0b011, 0, 1>; -let isCodeGenOnly = 0 in { def C4_cmpneq : T_ALU32_3op_cmp<"!cmp.eq", 0b00, 1, 1>; def C4_cmplte : T_ALU32_3op_cmp<"!cmp.gt", 0b10, 1, 0>; def C4_cmplteu : T_ALU32_3op_cmp<"!cmp.gtu", 0b11, 1, 0>; -} // Pats for instruction selection. @@ -139,11 +156,15 @@ class CmpInReg def: T_cmp32_rr_pat, i32>; def: T_cmp32_rr_pat, i32>; +def: T_cmp32_rr_pat; +def: T_cmp32_rr_pat; + +def: T_cmp32_rr_pat, i1>; + class T_CMP_rrbh MinOp, bit IsComm> : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt), "$Pd = "#mnemonic#"($Rs, $Rt)", [], "", S_3op_tc_2early_SLOT23>, ImmRegRel { - let validSubTargets = HasV4SubT; let InputType = "reg"; let CextOpcode = mnemonic; let isCompare = 1; @@ -162,13 +183,26 @@ class T_CMP_rrbh MinOp, bit IsComm> let Inst{1-0} = Pd; } -let isCodeGenOnly = 0 in { def A4_cmpbeq : T_CMP_rrbh<"cmpb.eq", 0b110, 1>; def A4_cmpbgt : T_CMP_rrbh<"cmpb.gt", 0b010, 0>; def A4_cmpbgtu : T_CMP_rrbh<"cmpb.gtu", 0b111, 0>; def A4_cmpheq : T_CMP_rrbh<"cmph.eq", 0b011, 1>; def A4_cmphgt : T_CMP_rrbh<"cmph.gt", 0b100, 0>; def A4_cmphgtu : T_CMP_rrbh<"cmph.gtu", 0b101, 0>; + +let AddedComplexity = 100 in { + def: Pat<(i1 (seteq (and (xor (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), + 255), 0)), + (A4_cmpbeq IntRegs:$Rs, IntRegs:$Rt)>; + def: Pat<(i1 (setne (and (xor (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), + 255), 0)), + (C2_not (A4_cmpbeq IntRegs:$Rs, IntRegs:$Rt))>; + def: Pat<(i1 (seteq (and (xor (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), + 65535), 0)), + (A4_cmpheq IntRegs:$Rs, IntRegs:$Rt)>; + def: Pat<(i1 (setne (and (xor (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), + 65535), 0)), + (C2_not (A4_cmpheq IntRegs:$Rs, IntRegs:$Rt))>; } class T_CMP_ribh MajOp, bit IsHalf, bit IsComm, @@ -176,7 +210,6 @@ class T_CMP_ribh MajOp, bit IsHalf, bit IsComm, : ALU64Inst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, ImmType:$Imm), "$Pd = "#mnemonic#"($Rs, #$Imm)", [], "", ALU64_tc_2early_SLOT23>, ImmRegRel { - let validSubTargets = HasV4SubT; let InputType = "imm"; let CextOpcode = mnemonic; let isCompare = 1; @@ -201,19 +234,17 @@ class T_CMP_ribh MajOp, bit IsHalf, bit IsComm, let Inst{1-0} = Pd; } -let isCodeGenOnly = 0 in { def A4_cmpbeqi : T_CMP_ribh<"cmpb.eq", 0b00, 0, 1, u8Imm, 0, 0, 8>; def A4_cmpbgti : T_CMP_ribh<"cmpb.gt", 0b01, 0, 0, s8Imm, 0, 1, 8>; def A4_cmpbgtui : T_CMP_ribh<"cmpb.gtu", 0b10, 0, 0, u7Ext, 1, 0, 7>; def A4_cmpheqi : T_CMP_ribh<"cmph.eq", 0b00, 1, 1, s8Ext, 1, 1, 8>; def A4_cmphgti : T_CMP_ribh<"cmph.gt", 0b01, 1, 0, s8Ext, 1, 1, 8>; def A4_cmphgtui : T_CMP_ribh<"cmph.gtu", 0b10, 1, 0, u7Ext, 1, 0, 7>; -} + class T_RCMP_EQ_ri : ALU32_ri<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s8Ext:$s8), "$Rd = "#mnemonic#"($Rs, #$s8)", [], "", ALU32_2op_tc_1_SLOT0123>, ImmRegRel { - let validSubTargets = HasV4SubT; let InputType = "imm"; let CextOpcode = !if (IsNeg, "!rcmp.eq", "rcmp.eq"); let isExtendable = 1; @@ -236,71 +267,24 @@ class T_RCMP_EQ_ri let Inst{4-0} = Rd; } -let isCodeGenOnly = 0 in { def A4_rcmpeqi : T_RCMP_EQ_ri<"cmp.eq", 0>; def A4_rcmpneqi : T_RCMP_EQ_ri<"!cmp.eq", 1>; -} -def: Pat<(i32 (zext (i1 (seteq (i32 IntRegs:$Rs), s8ExtPred:$s8)))), - (A4_rcmpeqi IntRegs:$Rs, s8ExtPred:$s8)>; -def: Pat<(i32 (zext (i1 (setne (i32 IntRegs:$Rs), s8ExtPred:$s8)))), - (A4_rcmpneqi IntRegs:$Rs, s8ExtPred:$s8)>; +def: Pat<(i32 (zext (i1 (seteq (i32 IntRegs:$Rs), s32ImmPred:$s8)))), + (A4_rcmpeqi IntRegs:$Rs, s32ImmPred:$s8)>; +def: Pat<(i32 (zext (i1 (setne (i32 IntRegs:$Rs), s32ImmPred:$s8)))), + (A4_rcmpneqi IntRegs:$Rs, s32ImmPred:$s8)>; // Preserve the S2_tstbit_r generation def: Pat<(i32 (zext (i1 (setne (i32 (and (i32 (shl 1, (i32 IntRegs:$src2))), (i32 IntRegs:$src1))), 0)))), (C2_muxii (S2_tstbit_r IntRegs:$src1, IntRegs:$src2), 1, 0)>; - //===----------------------------------------------------------------------===// // ALU32 - //===----------------------------------------------------------------------===// -//===----------------------------------------------------------------------===// -// ALU32/PERM + -//===----------------------------------------------------------------------===// - -// Combine -// Rdd=combine(Rs, #s8) -let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8, - hasSideEffects = 0, validSubTargets = HasV4SubT in -def COMBINE_rI_V4 : ALU32_ri<(outs DoubleRegs:$dst), - (ins IntRegs:$src1, s8Ext:$src2), - "$dst = combine($src1, #$src2)", - []>, - Requires<[HasV4T]>; - -// Rdd=combine(#s8, Rs) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 8, - hasSideEffects = 0, validSubTargets = HasV4SubT in -def COMBINE_Ir_V4 : ALU32_ir<(outs DoubleRegs:$dst), - (ins s8Ext:$src1, IntRegs:$src2), - "$dst = combine(#$src1, $src2)", - []>, - Requires<[HasV4T]>; - -def HexagonWrapperCombineRI_V4 : - SDNode<"HexagonISD::WrapperCombineRI_V4", SDTHexagonI64I32I32>; -def HexagonWrapperCombineIR_V4 : - SDNode<"HexagonISD::WrapperCombineIR_V4", SDTHexagonI64I32I32>; - -def : Pat <(HexagonWrapperCombineRI_V4 IntRegs:$r, s8ExtPred:$i), - (COMBINE_rI_V4 IntRegs:$r, s8ExtPred:$i)>, - Requires<[HasV4T]>; - -def : Pat <(HexagonWrapperCombineIR_V4 s8ExtPred:$i, IntRegs:$r), - (COMBINE_Ir_V4 s8ExtPred:$i, IntRegs:$r)>, - Requires<[HasV4T]>; - -let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 6, - hasSideEffects = 0, validSubTargets = HasV4SubT in -def COMBINE_iI_V4 : ALU32_ii<(outs DoubleRegs:$dst), - (ins s8Imm:$src1, u6Ext:$src2), - "$dst = combine(#$src1, #$src2)", - []>, - Requires<[HasV4T]>; - //===----------------------------------------------------------------------===// // ALU32/PERM + //===----------------------------------------------------------------------===// @@ -323,263 +307,370 @@ class T_Combine1 MajOp, dag ins, string AsmStr> let Inst{4-0} = Rdd; } -let opExtendable = 2, isCodeGenOnly = 0 in +let opExtendable = 2 in def A4_combineri : T_Combine1<0b00, (ins IntRegs:$Rs, s8Ext:$s8), "$Rdd = combine($Rs, #$s8)">; -let opExtendable = 1, isCodeGenOnly = 0 in +let opExtendable = 1 in def A4_combineir : T_Combine1<0b01, (ins s8Ext:$s8, IntRegs:$Rs), "$Rdd = combine(#$s8, $Rs)">; +// The complexity of the combines involving immediates should be greater +// than the complexity of the combine with two registers. +let AddedComplexity = 50 in { +def: Pat<(HexagonCOMBINE IntRegs:$r, s32ImmPred:$i), + (A4_combineri IntRegs:$r, s32ImmPred:$i)>; + +def: Pat<(HexagonCOMBINE s32ImmPred:$i, IntRegs:$r), + (A4_combineir s32ImmPred:$i, IntRegs:$r)>; +} + +// A4_combineii: Set two small immediates. +let hasSideEffects = 0, isExtendable = 1, opExtentBits = 6, opExtendable = 2 in +def A4_combineii: ALU32Inst<(outs DoubleRegs:$Rdd), (ins s8Imm:$s8, u6Ext:$U6), + "$Rdd = combine(#$s8, #$U6)"> { + bits<5> Rdd; + bits<8> s8; + bits<6> U6; + + let IClass = 0b0111; + let Inst{27-23} = 0b11001; + let Inst{20-16} = U6{5-1}; + let Inst{13} = U6{0}; + let Inst{12-5} = s8; + let Inst{4-0} = Rdd; + } + +// The complexity of the combine with two immediates should be greater than +// the complexity of a combine involving a register. +let AddedComplexity = 75 in +def: Pat<(HexagonCOMBINE s8ImmPred:$s8, u32ImmPred:$u6), + (A4_combineii imm:$s8, imm:$u6)>; + +//===----------------------------------------------------------------------===// +// ALU32/PERM - +//===----------------------------------------------------------------------===// + //===----------------------------------------------------------------------===// // LD + //===----------------------------------------------------------------------===// + +def Zext64: OutPatFrag<(ops node:$Rs), + (i64 (A4_combineir 0, (i32 $Rs)))>; +def Sext64: OutPatFrag<(ops node:$Rs), + (i64 (A2_sxtw (i32 $Rs)))>; + +// Patterns to generate indexed loads with different forms of the address: +// - frameindex, +// - base + offset, +// - base (without offset). +multiclass Loadxm_pat { + def: Pat<(VT (Load AddrFI:$fi)), + (VT (ValueMod (MI AddrFI:$fi, 0)))>; + def: Pat<(VT (Load (add AddrFI:$fi, ImmPred:$Off))), + (VT (ValueMod (MI AddrFI:$fi, imm:$Off)))>; + def: Pat<(VT (Load (add IntRegs:$Rs, ImmPred:$Off))), + (VT (ValueMod (MI IntRegs:$Rs, imm:$Off)))>; + def: Pat<(VT (Load (i32 IntRegs:$Rs))), + (VT (ValueMod (MI IntRegs:$Rs, 0)))>; +} + +defm: Loadxm_pat; +defm: Loadxm_pat; +defm: Loadxm_pat; +defm: Loadxm_pat; +defm: Loadxm_pat; +defm: Loadxm_pat; +defm: Loadxm_pat; +defm: Loadxm_pat; + +// Map Rdd = anyext(Rs) -> Rdd = combine(#0, Rs). +def: Pat<(i64 (anyext (i32 IntRegs:$src1))), (Zext64 IntRegs:$src1)>; + //===----------------------------------------------------------------------===// // Template class for load instructions with Absolute set addressing mode. //===----------------------------------------------------------------------===// -let isExtended = 1, opExtendable = 2, hasSideEffects = 0, -validSubTargets = HasV4SubT, addrMode = AbsoluteSet in -class T_LD_abs_set: - LDInst2<(outs RC:$dst1, IntRegs:$dst2), - (ins u0AlwaysExt:$addr), - "$dst1 = "#mnemonic#"($dst2=##$addr)", - []>, - Requires<[HasV4T]>; +let isExtended = 1, opExtendable = 2, opExtentBits = 6, addrMode = AbsoluteSet, + hasSideEffects = 0 in +class T_LD_abs_setMajOp>: + LDInst<(outs RC:$dst1, IntRegs:$dst2), + (ins u6Ext:$addr), + "$dst1 = "#mnemonic#"($dst2 = #$addr)", + []> { + bits<7> name; + bits<5> dst1; + bits<5> dst2; + bits<6> addr; + + let IClass = 0b1001; + let Inst{27-25} = 0b101; + let Inst{24-21} = MajOp; + let Inst{13-12} = 0b01; + let Inst{4-0} = dst1; + let Inst{20-16} = dst2; + let Inst{11-8} = addr{5-2}; + let Inst{6-5} = addr{1-0}; +} + +let accessSize = ByteAccess, hasNewValue = 1 in { + def L4_loadrb_ap : T_LD_abs_set <"memb", IntRegs, 0b1000>; + def L4_loadrub_ap : T_LD_abs_set <"memub", IntRegs, 0b1001>; +} -def LDrid_abs_set_V4 : T_LD_abs_set <"memd", DoubleRegs>; -def LDrib_abs_set_V4 : T_LD_abs_set <"memb", IntRegs>; -def LDriub_abs_set_V4 : T_LD_abs_set <"memub", IntRegs>; -def LDrih_abs_set_V4 : T_LD_abs_set <"memh", IntRegs>; -def LDriw_abs_set_V4 : T_LD_abs_set <"memw", IntRegs>; -def LDriuh_abs_set_V4 : T_LD_abs_set <"memuh", IntRegs>; +let accessSize = HalfWordAccess, hasNewValue = 1 in { + def L4_loadrh_ap : T_LD_abs_set <"memh", IntRegs, 0b1010>; + def L4_loadruh_ap : T_LD_abs_set <"memuh", IntRegs, 0b1011>; + def L4_loadbsw2_ap : T_LD_abs_set <"membh", IntRegs, 0b0001>; + def L4_loadbzw2_ap : T_LD_abs_set <"memubh", IntRegs, 0b0011>; +} +let accessSize = WordAccess, hasNewValue = 1 in + def L4_loadri_ap : T_LD_abs_set <"memw", IntRegs, 0b1100>; -// multiclass for load instructions with base + register offset -// addressing mode -multiclass ld_idxd_shl_pbase { - let isPredicatedNew = isPredNew in - def NAME : LDInst2<(outs RC:$dst), - (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$offset), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$offset)", - []>, Requires<[HasV4T]>; +let accessSize = WordAccess in { + def L4_loadbzw4_ap : T_LD_abs_set <"memubh", DoubleRegs, 0b0101>; + def L4_loadbsw4_ap : T_LD_abs_set <"membh", DoubleRegs, 0b0111>; } -multiclass ld_idxd_shl_pred { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ld_idxd_shl_pbase; - // Predicate new - defm _cdn#NAME : ld_idxd_shl_pbase; +let accessSize = DoubleWordAccess in +def L4_loadrd_ap : T_LD_abs_set <"memd", DoubleRegs, 0b1110>; + +let accessSize = ByteAccess in + def L4_loadalignb_ap : T_LD_abs_set <"memb_fifo", DoubleRegs, 0b0100>; + +let accessSize = HalfWordAccess in +def L4_loadalignh_ap : T_LD_abs_set <"memh_fifo", DoubleRegs, 0b0010>; + +// Load - Indirect with long offset +let InputType = "imm", addrMode = BaseLongOffset, isExtended = 1, +opExtentBits = 6, opExtendable = 3 in +class T_LoadAbsReg MajOp> + : LDInst <(outs RC:$dst), (ins IntRegs:$src1, u2Imm:$src2, u6Ext:$src3), + "$dst = "#mnemonic#"($src1<<#$src2 + #$src3)", + [] >, ImmRegShl { + bits<5> dst; + bits<5> src1; + bits<2> src2; + bits<6> src3; + let CextOpcode = CextOp; + let hasNewValue = !if (!eq(!cast(RC), "DoubleRegs"), 0, 1); + + let IClass = 0b1001; + let Inst{27-25} = 0b110; + let Inst{24-21} = MajOp; + let Inst{20-16} = src1; + let Inst{13} = src2{1}; + let Inst{12} = 0b1; + let Inst{11-8} = src3{5-2}; + let Inst{7} = src2{0}; + let Inst{6-5} = src3{1-0}; + let Inst{4-0} = dst; } + +let accessSize = ByteAccess in { + def L4_loadrb_ur : T_LoadAbsReg<"memb", "LDrib", IntRegs, 0b1000>; + def L4_loadrub_ur : T_LoadAbsReg<"memub", "LDriub", IntRegs, 0b1001>; + def L4_loadalignb_ur : T_LoadAbsReg<"memb_fifo", "LDrib_fifo", + DoubleRegs, 0b0100>; } -let hasSideEffects = 0 in -multiclass ld_idxd_shl { - let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in { - let isPredicable = 1 in - def NAME#_V4 : LDInst2<(outs RC:$dst), - (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), - "$dst = "#mnemonic#"($src1+$src2<<#$offset)", - []>, Requires<[HasV4T]>; - - let isPredicated = 1 in { - defm Pt_V4 : ld_idxd_shl_pred; - defm NotPt_V4 : ld_idxd_shl_pred; - } - } +let accessSize = HalfWordAccess in { + def L4_loadrh_ur : T_LoadAbsReg<"memh", "LDrih", IntRegs, 0b1010>; + def L4_loadruh_ur : T_LoadAbsReg<"memuh", "LDriuh", IntRegs, 0b1011>; + def L4_loadbsw2_ur : T_LoadAbsReg<"membh", "LDribh2", IntRegs, 0b0001>; + def L4_loadbzw2_ur : T_LoadAbsReg<"memubh", "LDriubh2", IntRegs, 0b0011>; + def L4_loadalignh_ur : T_LoadAbsReg<"memh_fifo", "LDrih_fifo", + DoubleRegs, 0b0010>; } -let addrMode = BaseRegOffset in { - let accessSize = ByteAccess in { - defm LDrib_indexed_shl: ld_idxd_shl<"memb", "LDrib", IntRegs>, - AddrModeRel; - defm LDriub_indexed_shl: ld_idxd_shl<"memub", "LDriub", IntRegs>, - AddrModeRel; - } - let accessSize = HalfWordAccess in { - defm LDrih_indexed_shl: ld_idxd_shl<"memh", "LDrih", IntRegs>, AddrModeRel; - defm LDriuh_indexed_shl: ld_idxd_shl<"memuh", "LDriuh", IntRegs>, - AddrModeRel; - } - let accessSize = WordAccess in - defm LDriw_indexed_shl: ld_idxd_shl<"memw", "LDriw", IntRegs>, AddrModeRel; +let accessSize = WordAccess in { + def L4_loadri_ur : T_LoadAbsReg<"memw", "LDriw", IntRegs, 0b1100>; + def L4_loadbsw4_ur : T_LoadAbsReg<"membh", "LDribh4", DoubleRegs, 0b0111>; + def L4_loadbzw4_ur : T_LoadAbsReg<"memubh", "LDriubh4", DoubleRegs, 0b0101>; +} - let accessSize = DoubleWordAccess in - defm LDrid_indexed_shl: ld_idxd_shl<"memd", "LDrid", DoubleRegs>, - AddrModeRel; +let accessSize = DoubleWordAccess in +def L4_loadrd_ur : T_LoadAbsReg<"memd", "LDrid", DoubleRegs, 0b1110>; + + +multiclass T_LoadAbsReg_Pat { + def : Pat <(VT (ldOp (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tglobaladdr:$src3)))), + (MI IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3)>; + def : Pat <(VT (ldOp (add IntRegs:$src1, + (HexagonCONST32 tglobaladdr:$src2)))), + (MI IntRegs:$src1, 0, tglobaladdr:$src2)>; + + def : Pat <(VT (ldOp (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tconstpool:$src3)))), + (MI IntRegs:$src1, u2ImmPred:$src2, tconstpool:$src3)>; + def : Pat <(VT (ldOp (add IntRegs:$src1, + (HexagonCONST32 tconstpool:$src2)))), + (MI IntRegs:$src1, 0, tconstpool:$src2)>; + + def : Pat <(VT (ldOp (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tjumptable:$src3)))), + (MI IntRegs:$src1, u2ImmPred:$src2, tjumptable:$src3)>; + def : Pat <(VT (ldOp (add IntRegs:$src1, + (HexagonCONST32 tjumptable:$src2)))), + (MI IntRegs:$src1, 0, tjumptable:$src2)>; } -// 'def pats' for load instructions with base + register offset and non-zero -// immediate value. Immediate value is used to left-shift the second -// register operand. -let AddedComplexity = 40 in { -def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDrib_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; - -def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDriub_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; - -def : Pat <(i32 (extloadi8 (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDriub_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; - -def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDrih_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; - -def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDriuh_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; - -def : Pat <(i32 (extloadi16 (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDriuh_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; - -def : Pat <(i32 (load (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDriw_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; - -def : Pat <(i64 (load (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$offset)))), - (LDrid_indexed_shl_V4 IntRegs:$src1, - IntRegs:$src2, u2ImmPred:$offset)>, - Requires<[HasV4T]>; +let AddedComplexity = 60 in { +defm : T_LoadAbsReg_Pat ; +defm : T_LoadAbsReg_Pat ; +defm : T_LoadAbsReg_Pat ; + +defm : T_LoadAbsReg_Pat ; +defm : T_LoadAbsReg_Pat ; +defm : T_LoadAbsReg_Pat ; + +defm : T_LoadAbsReg_Pat ; +defm : T_LoadAbsReg_Pat ; } +//===----------------------------------------------------------------------===// +// Template classes for the non-predicated load instructions with +// base + register offset addressing mode +//===----------------------------------------------------------------------===// +class T_load_rr MajOp>: + LDInst<(outs RC:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$u2), + "$dst = "#mnemonic#"($src1 + $src2<<#$u2)", + [], "", V4LDST_tc_ld_SLOT01>, ImmRegShl, AddrModeRel { + bits<5> dst; + bits<5> src1; + bits<5> src2; + bits<2> u2; -// 'def pats' for load instruction base + register offset and -// zero immediate value. -let AddedComplexity = 10 in { -def : Pat <(i64 (load (add IntRegs:$src1, IntRegs:$src2))), - (LDrid_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; + let IClass = 0b0011; -def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, IntRegs:$src2))), - (LDrib_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; + let Inst{27-24} = 0b1010; + let Inst{23-21} = MajOp; + let Inst{20-16} = src1; + let Inst{12-8} = src2; + let Inst{13} = u2{1}; + let Inst{7} = u2{0}; + let Inst{4-0} = dst; + } -def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, IntRegs:$src2))), - (LDriub_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// Template classes for the predicated load instructions with +// base + register offset addressing mode +//===----------------------------------------------------------------------===// +let isPredicated = 1 in +class T_pload_rr MajOp, + bit isNot, bit isPredNew>: + LDInst <(outs RC:$dst), + (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$u2), + !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", + ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$u2)", + [], "", V4LDST_tc_ld_SLOT01>, AddrModeRel { + bits<5> dst; + bits<2> src1; + bits<5> src2; + bits<5> src3; + bits<2> u2; + + let isPredicatedFalse = isNot; + let isPredicatedNew = isPredNew; -def : Pat <(i32 (extloadi8 (add IntRegs:$src1, IntRegs:$src2))), - (LDriub_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; + let IClass = 0b0011; -def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, IntRegs:$src2))), - (LDrih_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; + let Inst{27-26} = 0b00; + let Inst{25} = isPredNew; + let Inst{24} = isNot; + let Inst{23-21} = MajOp; + let Inst{20-16} = src2; + let Inst{12-8} = src3; + let Inst{13} = u2{1}; + let Inst{7} = u2{0}; + let Inst{6-5} = src1; + let Inst{4-0} = dst; + } -def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, IntRegs:$src2))), - (LDriuh_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// multiclass for load instructions with base + register offset +// addressing mode +//===----------------------------------------------------------------------===// +let hasSideEffects = 0, addrMode = BaseRegOffset in +multiclass ld_idxd_shl MajOp > { + let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl, + InputType = "reg" in { + let isPredicable = 1 in + def L4_#NAME#_rr : T_load_rr ; -def : Pat <(i32 (extloadi16 (add IntRegs:$src1, IntRegs:$src2))), - (LDriuh_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; + // Predicated + def L4_p#NAME#t_rr : T_pload_rr ; + def L4_p#NAME#f_rr : T_pload_rr ; -def : Pat <(i32 (load (add IntRegs:$src1, IntRegs:$src2))), - (LDriw_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>, - Requires<[HasV4T]>; + // Predicated new + def L4_p#NAME#tnew_rr : T_pload_rr ; + def L4_p#NAME#fnew_rr : T_pload_rr ; + } } -// zext i1->i64 -def : Pat <(i64 (zext (i1 PredRegs:$src1))), - (i64 (COMBINE_Ir_V4 0, (C2_muxii (i1 PredRegs:$src1), 1, 0)))>, - Requires<[HasV4T]>; +let hasNewValue = 1, accessSize = ByteAccess in { + defm loadrb : ld_idxd_shl<"memb", "LDrib", IntRegs, 0b000>; + defm loadrub : ld_idxd_shl<"memub", "LDriub", IntRegs, 0b001>; +} -// zext i32->i64 -def : Pat <(i64 (zext (i32 IntRegs:$src1))), - (i64 (COMBINE_Ir_V4 0, (i32 IntRegs:$src1)))>, - Requires<[HasV4T]>; -// zext i8->i64 -def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)), - (i64 (COMBINE_Ir_V4 0, (L2_loadrub_io AddrFI:$src1, 0)))>, - Requires<[HasV4T]>; - -let AddedComplexity = 20 in -def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1), - s11_0ExtPred:$offset))), - (i64 (COMBINE_Ir_V4 0, (L2_loadrub_io IntRegs:$src1, - s11_0ExtPred:$offset)))>, - Requires<[HasV4T]>; +let hasNewValue = 1, accessSize = HalfWordAccess in { + defm loadrh : ld_idxd_shl<"memh", "LDrih", IntRegs, 0b010>; + defm loadruh : ld_idxd_shl<"memuh", "LDriuh", IntRegs, 0b011>; +} -// zext i1->i64 -def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)), - (i64 (COMBINE_Ir_V4 0, (L2_loadrub_io AddrFI:$src1, 0)))>, - Requires<[HasV4T]>; - -let AddedComplexity = 20 in -def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1), - s11_0ExtPred:$offset))), - (i64 (COMBINE_Ir_V4 0, (L2_loadrub_io IntRegs:$src1, - s11_0ExtPred:$offset)))>, - Requires<[HasV4T]>; - -// zext i16->i64 -def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)), - (i64 (COMBINE_Ir_V4 0, (L2_loadruh_io AddrFI:$src1, 0)))>, - Requires<[HasV4T]>; - -let AddedComplexity = 20 in -def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1), - s11_1ExtPred:$offset))), - (i64 (COMBINE_Ir_V4 0, (L2_loadruh_io IntRegs:$src1, - s11_1ExtPred:$offset)))>, - Requires<[HasV4T]>; - -// anyext i16->i64 -def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)), - (i64 (COMBINE_Ir_V4 0, (L2_loadrh_io AddrFI:$src1, 0)))>, - Requires<[HasV4T]>; - -let AddedComplexity = 20 in -def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1), - s11_1ExtPred:$offset))), - (i64 (COMBINE_Ir_V4 0, (L2_loadrh_io IntRegs:$src1, - s11_1ExtPred:$offset)))>, - Requires<[HasV4T]>; +let hasNewValue = 1, accessSize = WordAccess in +defm loadri : ld_idxd_shl<"memw", "LDriw", IntRegs, 0b100>; -// zext i32->i64 -def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)), - (i64 (COMBINE_Ir_V4 0, (L2_loadri_io AddrFI:$src1, 0)))>, - Requires<[HasV4T]>; +let accessSize = DoubleWordAccess in +defm loadrd : ld_idxd_shl<"memd", "LDrid", DoubleRegs, 0b110>; -let AddedComplexity = 100 in -def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))), - (i64 (COMBINE_Ir_V4 0, (L2_loadri_io IntRegs:$src1, - s11_2ExtPred:$offset)))>, - Requires<[HasV4T]>; +// 'def pats' for load instructions with base + register offset and non-zero +// immediate value. Immediate value is used to left-shift the second +// register operand. +class Loadxs_pat + : Pat<(VT (Load (add (i32 IntRegs:$Rs), + (i32 (shl (i32 IntRegs:$Rt), u2ImmPred:$u2))))), + (VT (MI IntRegs:$Rs, IntRegs:$Rt, imm:$u2))>; -// anyext i32->i64 -def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)), - (i64 (COMBINE_Ir_V4 0, (L2_loadri_io AddrFI:$src1, 0)))>, - Requires<[HasV4T]>; +let AddedComplexity = 40 in { + def: Loadxs_pat; + def: Loadxs_pat; + def: Loadxs_pat; + def: Loadxs_pat; + def: Loadxs_pat; + def: Loadxs_pat; + def: Loadxs_pat; + def: Loadxs_pat; +} -let AddedComplexity = 100 in -def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))), - (i64 (COMBINE_Ir_V4 0, (L2_loadri_io IntRegs:$src1, - s11_2ExtPred:$offset)))>, - Requires<[HasV4T]>; +// 'def pats' for load instruction base + register offset and +// zero immediate value. +class Loadxs_simple_pat + : Pat<(VT (Load (add (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)))), + (VT (MI IntRegs:$Rs, IntRegs:$Rt, 0))>; + +let AddedComplexity = 20 in { + def: Loadxs_simple_pat; + def: Loadxs_simple_pat; + def: Loadxs_simple_pat; + def: Loadxs_simple_pat; + def: Loadxs_simple_pat; + def: Loadxs_simple_pat; + def: Loadxs_simple_pat; + def: Loadxs_simple_pat; +} +// zext i1->i64 +def: Pat<(i64 (zext (i1 PredRegs:$src1))), + (Zext64 (C2_muxii PredRegs:$src1, 1, 0))>; +// zext i32->i64 +def: Pat<(i64 (zext (i32 IntRegs:$src1))), + (Zext64 IntRegs:$src1)>; //===----------------------------------------------------------------------===// // LD - @@ -592,194 +683,357 @@ def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))), //===----------------------------------------------------------------------===// // Template class for store instructions with Absolute set addressing mode. //===----------------------------------------------------------------------===// -let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT, -addrMode = AbsoluteSet in -class T_ST_abs_set: - STInst2<(outs IntRegs:$dst1), - (ins RC:$src1, u0AlwaysExt:$src2), - mnemonic#"($dst1=##$src2) = $src1", - []>, - Requires<[HasV4T]>; +let isExtended = 1, opExtendable = 1, opExtentBits = 6, + addrMode = AbsoluteSet, isNVStorable = 1 in +class T_ST_absset MajOp, MemAccessSize AccessSz, bit isHalf = 0> + : STInst<(outs IntRegs:$dst), + (ins u6Ext:$addr, RC:$src), + mnemonic#"($dst = #$addr) = $src"#!if(isHalf, ".h","")>, NewValueRel { + bits<5> dst; + bits<6> addr; + bits<5> src; + let accessSize = AccessSz; + let BaseOpcode = BaseOp#"_AbsSet"; + + let IClass = 0b1010; + + let Inst{27-24} = 0b1011; + let Inst{23-21} = MajOp; + let Inst{20-16} = dst; + let Inst{13} = 0b0; + let Inst{12-8} = src; + let Inst{7} = 0b1; + let Inst{5-0} = addr; + } -def STrid_abs_set_V4 : T_ST_abs_set <"memd", DoubleRegs>; -def STrib_abs_set_V4 : T_ST_abs_set <"memb", IntRegs>; -def STrih_abs_set_V4 : T_ST_abs_set <"memh", IntRegs>; -def STriw_abs_set_V4 : T_ST_abs_set <"memw", IntRegs>; +def S4_storerb_ap : T_ST_absset <"memb", "STrib", IntRegs, 0b000, ByteAccess>; +def S4_storerh_ap : T_ST_absset <"memh", "STrih", IntRegs, 0b010, + HalfWordAccess>; +def S4_storeri_ap : T_ST_absset <"memw", "STriw", IntRegs, 0b100, WordAccess>; -//===----------------------------------------------------------------------===// -// multiclass for store instructions with base + register offset addressing -// mode -//===----------------------------------------------------------------------===// -multiclass ST_Idxd_shl_Pbase { - let isPredicatedNew = isPredNew in - def NAME : STInst2<(outs), - (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, - RC:$src5), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"($src2+$src3<<#$src4) = $src5", - []>, - Requires<[HasV4T]>; +let isNVStorable = 0 in { + def S4_storerf_ap : T_ST_absset <"memh", "STrif", IntRegs, + 0b011, HalfWordAccess, 1>; + def S4_storerd_ap : T_ST_absset <"memd", "STrid", DoubleRegs, + 0b110, DoubleWordAccess>; } -multiclass ST_Idxd_shl_Pred { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_Idxd_shl_Pbase; - // Predicate new - defm _cdn#NAME : ST_Idxd_shl_Pbase; +let opExtendable = 1, isNewValue = 1, isNVStore = 1, opNewValue = 2, +isExtended = 1, opExtentBits= 6 in +class T_ST_absset_nv MajOp, + MemAccessSize AccessSz > + : NVInst <(outs IntRegs:$dst), + (ins u6Ext:$addr, IntRegs:$src), + mnemonic#"($dst = #$addr) = $src.new">, NewValueRel { + bits<5> dst; + bits<6> addr; + bits<3> src; + let accessSize = AccessSz; + let BaseOpcode = BaseOp#"_AbsSet"; + + let IClass = 0b1010; + + let Inst{27-21} = 0b1011101; + let Inst{20-16} = dst; + let Inst{13-11} = 0b000; + let Inst{12-11} = MajOp; + let Inst{10-8} = src; + let Inst{7} = 0b1; + let Inst{5-0} = addr; } + +let mayStore = 1, addrMode = AbsoluteSet in { + def S4_storerbnew_ap : T_ST_absset_nv <"memb", "STrib", 0b00, ByteAccess>; + def S4_storerhnew_ap : T_ST_absset_nv <"memh", "STrih", 0b01, HalfWordAccess>; + def S4_storerinew_ap : T_ST_absset_nv <"memw", "STriw", 0b10, WordAccess>; } -let isNVStorable = 1 in -multiclass ST_Idxd_shl { - let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in { - let isPredicable = 1 in - def NAME#_V4 : STInst2<(outs), - (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, RC:$src4), - mnemonic#"($src1+$src2<<#$src3) = $src4", - []>, - Requires<[HasV4T]>; - - let isPredicated = 1 in { - defm Pt_V4 : ST_Idxd_shl_Pred; - defm NotPt_V4 : ST_Idxd_shl_Pred; - } - } +let isExtended = 1, opExtendable = 2, opExtentBits = 6, InputType = "imm", +addrMode = BaseLongOffset, AddedComplexity = 40 in +class T_StoreAbsReg MajOp, MemAccessSize AccessSz, bit isHalf = 0> + : STInst<(outs), + (ins IntRegs:$src1, u2Imm:$src2, u6Ext:$src3, RC:$src4), + mnemonic#"($src1<<#$src2 + #$src3) = $src4"#!if(isHalf, ".h",""), + []>, ImmRegShl, NewValueRel { + + bits<5> src1; + bits<2> src2; + bits<6> src3; + bits<5> src4; + + let accessSize = AccessSz; + let CextOpcode = CextOp; + let BaseOpcode = CextOp#"_shl"; + let IClass = 0b1010; + + let Inst{27-24} =0b1101; + let Inst{23-21} = MajOp; + let Inst{20-16} = src1; + let Inst{13} = src2{1}; + let Inst{12-8} = src4; + let Inst{7} = 0b1; + let Inst{6} = src2{0}; + let Inst{5-0} = src3; } -// multiclass for new-value store instructions with base + register offset -// addressing mode. -multiclass ST_Idxd_shl_Pbase_nv { - let isPredicatedNew = isPredNew in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, - RC:$src5), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"($src2+$src3<<#$src4) = $src5.new", - []>, - Requires<[HasV4T]>; +def S4_storerb_ur : T_StoreAbsReg <"memb", "STrib", IntRegs, 0b000, ByteAccess>; +def S4_storerh_ur : T_StoreAbsReg <"memh", "STrih", IntRegs, 0b010, + HalfWordAccess>; +def S4_storerf_ur : T_StoreAbsReg <"memh", "STrif", IntRegs, 0b011, + HalfWordAccess, 1>; +def S4_storeri_ur : T_StoreAbsReg <"memw", "STriw", IntRegs, 0b100, WordAccess>; +def S4_storerd_ur : T_StoreAbsReg <"memd", "STrid", DoubleRegs, 0b110, + DoubleWordAccess>; + +let AddedComplexity = 40 in +multiclass T_StoreAbsReg_Pats { + def : Pat<(stOp (VT RC:$src4), + (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2), + u32ImmPred:$src3)), + (MI IntRegs:$src1, u2ImmPred:$src2, u32ImmPred:$src3, RC:$src4)>; + + def : Pat<(stOp (VT RC:$src4), + (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tglobaladdr:$src3))), + (MI IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3, RC:$src4)>; + + def : Pat<(stOp (VT RC:$src4), + (add IntRegs:$src1, (HexagonCONST32 tglobaladdr:$src3))), + (MI IntRegs:$src1, 0, tglobaladdr:$src3, RC:$src4)>; } -multiclass ST_Idxd_shl_Pred_nv { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_Idxd_shl_Pbase_nv; - // Predicate new - defm _cdn#NAME : ST_Idxd_shl_Pbase_nv; +defm : T_StoreAbsReg_Pats ; +defm : T_StoreAbsReg_Pats ; +defm : T_StoreAbsReg_Pats ; +defm : T_StoreAbsReg_Pats ; + +let mayStore = 1, isNVStore = 1, isExtended = 1, addrMode = BaseLongOffset, + opExtentBits = 6, isNewValue = 1, opNewValue = 3, opExtendable = 2 in +class T_StoreAbsRegNV MajOp, + MemAccessSize AccessSz> + : NVInst <(outs ), + (ins IntRegs:$src1, u2Imm:$src2, u6Ext:$src3, IntRegs:$src4), + mnemonic#"($src1<<#$src2 + #$src3) = $src4.new">, NewValueRel { + bits<5> src1; + bits<2> src2; + bits<6> src3; + bits<3> src4; + + let CextOpcode = CextOp; + let BaseOpcode = CextOp#"_shl"; + let IClass = 0b1010; + + let Inst{27-21} = 0b1101101; + let Inst{12-11} = 0b00; + let Inst{7} = 0b1; + let Inst{20-16} = src1; + let Inst{13} = src2{1}; + let Inst{12-11} = MajOp; + let Inst{10-8} = src4; + let Inst{6} = src2{0}; + let Inst{5-0} = src3; } -} -let mayStore = 1, isNVStore = 1 in -multiclass ST_Idxd_shl_nv { - let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in { - let isPredicable = 1 in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, RC:$src4), - mnemonic#"($src1+$src2<<#$src3) = $src4.new", - []>, - Requires<[HasV4T]>; - - let isPredicated = 1 in { - defm Pt : ST_Idxd_shl_Pred_nv; - defm NotPt : ST_Idxd_shl_Pred_nv; - } +def S4_storerbnew_ur : T_StoreAbsRegNV <"memb", "STrib", 0b00, ByteAccess>; +def S4_storerhnew_ur : T_StoreAbsRegNV <"memh", "STrih", 0b01, HalfWordAccess>; +def S4_storerinew_ur : T_StoreAbsRegNV <"memw", "STriw", 0b10, WordAccess>; + +//===----------------------------------------------------------------------===// +// Template classes for the non-predicated store instructions with +// base + register offset addressing mode +//===----------------------------------------------------------------------===// +let isPredicable = 1 in +class T_store_rr MajOp, bit isH> + : STInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt), + mnemonic#"($Rs + $Ru<<#$u2) = $Rt"#!if(isH, ".h",""), + [],"",V4LDST_tc_st_SLOT01>, ImmRegShl, AddrModeRel { + + bits<5> Rs; + bits<5> Ru; + bits<2> u2; + bits<5> Rt; + + let IClass = 0b0011; + + let Inst{27-24} = 0b1011; + let Inst{23-21} = MajOp; + let Inst{20-16} = Rs; + let Inst{12-8} = Ru; + let Inst{13} = u2{1}; + let Inst{7} = u2{0}; + let Inst{4-0} = Rt; } -} -let addrMode = BaseRegOffset, hasSideEffects = 0, -validSubTargets = HasV4SubT in { - let accessSize = ByteAccess in - defm STrib_indexed_shl: ST_Idxd_shl<"memb", "STrib", IntRegs>, - ST_Idxd_shl_nv<"memb", "STrib", IntRegs>, AddrModeRel; +//===----------------------------------------------------------------------===// +// Template classes for the predicated store instructions with +// base + register offset addressing mode +//===----------------------------------------------------------------------===// +let isPredicated = 1 in +class T_pstore_rr MajOp, + bit isNot, bit isPredNew, bit isH> + : STInst <(outs), + (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt), + + !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ", + ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Rt"#!if(isH, ".h",""), + [], "", V4LDST_tc_st_SLOT01> , AddrModeRel{ + bits<2> Pv; + bits<5> Rs; + bits<5> Ru; + bits<2> u2; + bits<5> Rt; - let accessSize = HalfWordAccess in - defm STrih_indexed_shl: ST_Idxd_shl<"memh", "STrih", IntRegs>, - ST_Idxd_shl_nv<"memh", "STrih", IntRegs>, AddrModeRel; + let isPredicatedFalse = isNot; + let isPredicatedNew = isPredNew; - let accessSize = WordAccess in - defm STriw_indexed_shl: ST_Idxd_shl<"memw", "STriw", IntRegs>, - ST_Idxd_shl_nv<"memw", "STriw", IntRegs>, AddrModeRel; + let IClass = 0b0011; - let isNVStorable = 0, accessSize = DoubleWordAccess in - defm STrid_indexed_shl: ST_Idxd_shl<"memd", "STrid", DoubleRegs>, AddrModeRel; -} + let Inst{27-26} = 0b01; + let Inst{25} = isPredNew; + let Inst{24} = isNot; + let Inst{23-21} = MajOp; + let Inst{20-16} = Rs; + let Inst{12-8} = Ru; + let Inst{13} = u2{1}; + let Inst{7} = u2{0}; + let Inst{6-5} = Pv; + let Inst{4-0} = Rt; + } -let Predicates = [HasV4T], AddedComplexity = 10 in { -def : Pat<(truncstorei8 (i32 IntRegs:$src4), - (add IntRegs:$src1, (shl IntRegs:$src2, - u2ImmPred:$src3))), - (STrib_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, - u2ImmPred:$src3, IntRegs:$src4)>; +//===----------------------------------------------------------------------===// +// Template classes for the new-value store instructions with +// base + register offset addressing mode +//===----------------------------------------------------------------------===// +let isPredicable = 1, isNewValue = 1, opNewValue = 3 in +class T_store_new_rr MajOp> : + NVInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt), + mnemonic#"($Rs + $Ru<<#$u2) = $Nt.new", + [],"",V4LDST_tc_st_SLOT0>, ImmRegShl, AddrModeRel { + + bits<5> Rs; + bits<5> Ru; + bits<2> u2; + bits<3> Nt; -def : Pat<(truncstorei16 (i32 IntRegs:$src4), - (add IntRegs:$src1, (shl IntRegs:$src2, - u2ImmPred:$src3))), - (STrih_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, - u2ImmPred:$src3, IntRegs:$src4)>; + let IClass = 0b0011; -def : Pat<(store (i32 IntRegs:$src4), - (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))), - (STriw_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, - u2ImmPred:$src3, IntRegs:$src4)>; + let Inst{27-21} = 0b1011101; + let Inst{20-16} = Rs; + let Inst{12-8} = Ru; + let Inst{13} = u2{1}; + let Inst{7} = u2{0}; + let Inst{4-3} = MajOp; + let Inst{2-0} = Nt; + } -def : Pat<(store (i64 DoubleRegs:$src4), - (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))), - (STrid_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, - u2ImmPred:$src3, DoubleRegs:$src4)>; -} +//===----------------------------------------------------------------------===// +// Template classes for the predicated new-value store instructions with +// base + register offset addressing mode +//===----------------------------------------------------------------------===// +let isPredicated = 1, isNewValue = 1, opNewValue = 4 in +class T_pstore_new_rr MajOp, bit isNot, bit isPredNew> + : NVInst<(outs), + (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt), + !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ", + ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Nt.new", + [], "", V4LDST_tc_st_SLOT0>, AddrModeRel { + bits<2> Pv; + bits<5> Rs; + bits<5> Ru; + bits<2> u2; + bits<3> Nt; + + let isPredicatedFalse = isNot; + let isPredicatedNew = isPredNew; -let isExtended = 1, opExtendable = 2 in -class T_ST_LongOff : - STInst<(outs), - (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, RC:$src4), - mnemonic#"($src1<<#$src2+##$src3) = $src4", - [(stOp (VT RC:$src4), - (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2), - u0AlwaysExtPred:$src3))]>, - Requires<[HasV4T]>; + let IClass = 0b0011; + let Inst{27-26} = 0b01; + let Inst{25} = isPredNew; + let Inst{24} = isNot; + let Inst{23-21} = 0b101; + let Inst{20-16} = Rs; + let Inst{12-8} = Ru; + let Inst{13} = u2{1}; + let Inst{7} = u2{0}; + let Inst{6-5} = Pv; + let Inst{4-3} = MajOp; + let Inst{2-0} = Nt; + } -let isExtended = 1, opExtendable = 2, mayStore = 1, isNVStore = 1 in -class T_ST_LongOff_nv : - NVInst_V4<(outs), - (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4), - mnemonic#"($src1<<#$src2+##$src3) = $src4.new", - []>, - Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// multiclass for store instructions with base + register offset addressing +// mode +//===----------------------------------------------------------------------===// +let isNVStorable = 1 in +multiclass ST_Idxd_shl MajOp, bit isH = 0> { + let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in { + def S4_#NAME#_rr : T_store_rr ; -multiclass ST_LongOff { - let BaseOpcode = BaseOp#"_shl" in { - let isNVStorable = 1 in - def NAME#_V4 : T_ST_LongOff; + // Predicated + def S4_p#NAME#t_rr : T_pstore_rr ; + def S4_p#NAME#f_rr : T_pstore_rr ; - def NAME#_nv_V4 : T_ST_LongOff_nv; + // Predicated new + def S4_p#NAME#tnew_rr : T_pstore_rr ; + def S4_p#NAME#fnew_rr : T_pstore_rr ; } } -let AddedComplexity = 10, validSubTargets = HasV4SubT in { - def STrid_shl_V4 : T_ST_LongOff<"memd", store, DoubleRegs, i64>; - defm STrib_shl : ST_LongOff <"memb", "STrib", truncstorei8>, NewValueRel; - defm STrih_shl : ST_LongOff <"memh", "Strih", truncstorei16>, NewValueRel; - defm STriw_shl : ST_LongOff <"memw", "STriw", store>, NewValueRel; +//===----------------------------------------------------------------------===// +// multiclass for new-value store instructions with base + register offset +// addressing mode. +//===----------------------------------------------------------------------===// +let mayStore = 1, isNVStore = 1 in +multiclass ST_Idxd_shl_nv MajOp> { + let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in { + def S4_#NAME#new_rr : T_store_new_rr; + + // Predicated + def S4_p#NAME#newt_rr : T_pstore_new_rr ; + def S4_p#NAME#newf_rr : T_pstore_new_rr ; + + // Predicated new + def S4_p#NAME#newtnew_rr : T_pstore_new_rr ; + def S4_p#NAME#newfnew_rr : T_pstore_new_rr ; + } } -let AddedComplexity = 40 in -multiclass T_ST_LOff_Pats { - def : Pat<(stOp (VT RC:$src4), - (add (shl IntRegs:$src1, u2ImmPred:$src2), - (NumUsesBelowThresCONST32 tglobaladdr:$src3))), - (I IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3, RC:$src4)>; +let addrMode = BaseRegOffset, InputType = "reg", hasSideEffects = 0 in { + let accessSize = ByteAccess in + defm storerb: ST_Idxd_shl<"memb", "STrib", IntRegs, 0b000>, + ST_Idxd_shl_nv<"memb", "STrib", IntRegs, 0b00>; - def : Pat<(stOp (VT RC:$src4), - (add IntRegs:$src1, - (NumUsesBelowThresCONST32 tglobaladdr:$src3))), - (I IntRegs:$src1, 0, tglobaladdr:$src3, RC:$src4)>; + let accessSize = HalfWordAccess in + defm storerh: ST_Idxd_shl<"memh", "STrih", IntRegs, 0b010>, + ST_Idxd_shl_nv<"memh", "STrih", IntRegs, 0b01>; + + let accessSize = WordAccess in + defm storeri: ST_Idxd_shl<"memw", "STriw", IntRegs, 0b100>, + ST_Idxd_shl_nv<"memw", "STriw", IntRegs, 0b10>; + + let isNVStorable = 0, accessSize = DoubleWordAccess in + defm storerd: ST_Idxd_shl<"memd", "STrid", DoubleRegs, 0b110>; + + let isNVStorable = 0, accessSize = HalfWordAccess in + defm storerf: ST_Idxd_shl<"memh", "STrif", IntRegs, 0b011, 1>; } -defm : T_ST_LOff_Pats; -defm : T_ST_LOff_Pats; -defm : T_ST_LOff_Pats; -defm : T_ST_LOff_Pats; +class Storexs_pat + : Pat<(Store Value:$Ru, (add (i32 IntRegs:$Rs), + (i32 (shl (i32 IntRegs:$Rt), u2ImmPred:$u2)))), + (MI IntRegs:$Rs, IntRegs:$Rt, imm:$u2, Value:$Ru)>; + +let AddedComplexity = 40 in { + def: Storexs_pat; + def: Storexs_pat; + def: Storexs_pat; + def: Storexs_pat; +} // memd(Rx++#s4:3)=Rtt // memd(Rx++#s4:3:circ(Mu))=Rtt @@ -792,76 +1046,152 @@ defm : T_ST_LOff_Pats; // if ([!]Pv[.new]) memd(#u6)=Rtt // TODO: needs to be implemented. +//===----------------------------------------------------------------------===// +// Template class +//===----------------------------------------------------------------------===// +let isPredicable = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 8, + opExtendable = 2 in +class T_StoreImm MajOp > + : STInst <(outs ), (ins IntRegs:$Rs, OffsetOp:$offset, s8Ext:$S8), + mnemonic#"($Rs+#$offset)=#$S8", + [], "", V4LDST_tc_st_SLOT01>, + ImmRegRel, PredNewRel { + bits<5> Rs; + bits<8> S8; + bits<8> offset; + bits<6> offsetBits; + + string OffsetOpStr = !cast(OffsetOp); + let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2}, + !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1}, + /* u6_0Imm */ offset{5-0})); + + let IClass = 0b0011; + + let Inst{27-25} = 0b110; + let Inst{22-21} = MajOp; + let Inst{20-16} = Rs; + let Inst{12-7} = offsetBits; + let Inst{13} = S8{7}; + let Inst{6-0} = S8{6-0}; + } + +let isPredicated = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 6, + opExtendable = 3 in +class T_StoreImm_pred MajOp, + bit isPredNot, bit isPredNew > + : STInst <(outs ), + (ins PredRegs:$Pv, IntRegs:$Rs, OffsetOp:$offset, s6Ext:$S6), + !if(isPredNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ", + ") ")#mnemonic#"($Rs+#$offset)=#$S6", + [], "", V4LDST_tc_st_SLOT01>, + ImmRegRel, PredNewRel { + bits<2> Pv; + bits<5> Rs; + bits<6> S6; + bits<8> offset; + bits<6> offsetBits; + + string OffsetOpStr = !cast(OffsetOp); + let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2}, + !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1}, + /* u6_0Imm */ offset{5-0})); + let isPredicatedNew = isPredNew; + let isPredicatedFalse = isPredNot; + + let IClass = 0b0011; + + let Inst{27-25} = 0b100; + let Inst{24} = isPredNew; + let Inst{23} = isPredNot; + let Inst{22-21} = MajOp; + let Inst{20-16} = Rs; + let Inst{13} = S6{5}; + let Inst{12-7} = offsetBits; + let Inst{6-5} = Pv; + let Inst{4-0} = S6{4-0}; + } + + //===----------------------------------------------------------------------===// // multiclass for store instructions with base + immediate offset // addressing mode and immediate stored value. // mem[bhw](Rx++#s4:3)=#s8 // if ([!]Pv[.new]) mem[bhw](Rx++#s4:3)=#s6 //===----------------------------------------------------------------------===// -multiclass ST_Imm_Pbase { - let isPredicatedNew = isPredNew in - def NAME : STInst2<(outs), - (ins PredRegs:$src1, IntRegs:$src2, OffsetOp:$src3, s6Ext:$src4), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"($src2+#$src3) = #$src4", - []>, - Requires<[HasV4T]>; -} -multiclass ST_Imm_Pred { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_Imm_Pbase; - // Predicate new - defm _cdn#NAME : ST_Imm_Pbase; - } +multiclass ST_Imm_Pred MajOp, + bit PredNot> { + def _io : T_StoreImm_pred ; + // Predicate new + def new_io : T_StoreImm_pred ; } -let isExtendable = 1, isExtentSigned = 1, hasSideEffects = 0 in -multiclass ST_Imm { +multiclass ST_Imm MajOp> { let CextOpcode = CextOp, BaseOpcode = CextOp#_imm in { - let opExtendable = 2, opExtentBits = 8, isPredicable = 1 in - def NAME#_V4 : STInst2<(outs), - (ins IntRegs:$src1, OffsetOp:$src2, s8Ext:$src3), - mnemonic#"($src1+#$src2) = #$src3", - []>, - Requires<[HasV4T]>; - - let opExtendable = 3, opExtentBits = 6, isPredicated = 1 in { - defm Pt_V4 : ST_Imm_Pred; - defm NotPt_V4 : ST_Imm_Pred; - } + def _io : T_StoreImm ; + + defm t : ST_Imm_Pred ; + defm f : ST_Imm_Pred ; } } -let addrMode = BaseImmOffset, InputType = "imm", -validSubTargets = HasV4SubT in { +let hasSideEffects = 0, addrMode = BaseImmOffset, + InputType = "imm" in { let accessSize = ByteAccess in - defm STrib_imm : ST_Imm<"memb", "STrib", u6_0Imm>, ImmRegRel, PredNewRel; + defm S4_storeirb : ST_Imm<"memb", "STrib", u6_0Imm, 0b00>; let accessSize = HalfWordAccess in - defm STrih_imm : ST_Imm<"memh", "STrih", u6_1Imm>, ImmRegRel, PredNewRel; + defm S4_storeirh : ST_Imm<"memh", "STrih", u6_1Imm, 0b01>; let accessSize = WordAccess in - defm STriw_imm : ST_Imm<"memw", "STriw", u6_2Imm>, ImmRegRel, PredNewRel; + defm S4_storeiri : ST_Imm<"memw", "STriw", u6_2Imm, 0b10>; } -let Predicates = [HasV4T], AddedComplexity = 10 in { -def: Pat<(truncstorei8 s8ExtPred:$src3, (add IntRegs:$src1, u6_0ImmPred:$src2)), - (STrib_imm_V4 IntRegs:$src1, u6_0ImmPred:$src2, s8ExtPred:$src3)>; +def IMM_BYTE : SDNodeXFormgetSExtValue(); + return CurDAG->getTargetConstant(imm, SDLoc(N), MVT::i32); +}]>; + +def IMM_HALF : SDNodeXFormgetSExtValue(); + return CurDAG->getTargetConstant(imm, SDLoc(N), MVT::i32); +}]>; + +def IMM_WORD : SDNodeXFormgetSExtValue(); + return CurDAG->getTargetConstant(imm, SDLoc(N), MVT::i32); +}]>; -def: Pat<(truncstorei16 s8ExtPred:$src3, (add IntRegs:$src1, - u6_1ImmPred:$src2)), - (STrih_imm_V4 IntRegs:$src1, u6_1ImmPred:$src2, s8ExtPred:$src3)>; +def ToImmByte : OutPatFrag<(ops node:$R), (IMM_BYTE $R)>; +def ToImmHalf : OutPatFrag<(ops node:$R), (IMM_HALF $R)>; +def ToImmWord : OutPatFrag<(ops node:$R), (IMM_WORD $R)>; -def: Pat<(store s8ExtPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2)), - (STriw_imm_V4 IntRegs:$src1, u6_2ImmPred:$src2, s8ExtPred:$src3)>; +let AddedComplexity = 40 in { + // Not using frameindex patterns for these stores, because the offset + // is not extendable. This could cause problems during removing the frame + // indices, since the offset with respect to R29/R30 may not fit in the + // u6 field. + def: Storexm_add_pat; + def: Storexm_add_pat; + def: Storexm_add_pat; } -let AddedComplexity = 6 in -def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)), - (STrib_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>, - Requires<[HasV4T]>; +def: Storexm_simple_pat; +def: Storexm_simple_pat; +def: Storexm_simple_pat; // memb(Rx++#s4:0:circ(Mu))=Rt // memb(Rx++I:circ(Mu))=Rt @@ -869,16 +1199,10 @@ def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)), // memb(Rx++Mu:brev)=Rt // memb(gp+#u16:0)=Rt - // Store halfword. // TODO: needs to be implemented // memh(Re=#U6)=Rt.H // memh(Rs+#s11:1)=Rt.H -let AddedComplexity = 6 in -def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)), - (STrih_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>, - Requires<[HasV4T]>; - // memh(Rs+Ru<<#u2)=Rt.H // TODO: needs to be implemented. @@ -895,7 +1219,6 @@ def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)), // if ([!]Pv[.new]) memh(#u6)=Rt.H // if ([!]Pv[.new]) memh(#u6)=Rt - // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt.H // TODO: needs to be implemented. @@ -905,20 +1228,6 @@ def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)), // Store word. // memw(Re=#U6)=Rt // TODO: Needs to be implemented. - -// Store predicate: -let hasSideEffects = 0 in -def STriw_pred_V4 : STInst2<(outs), - (ins MEMri:$addr, PredRegs:$src1), - "Error; should not emit", - []>, - Requires<[HasV4T]>; - -let AddedComplexity = 6 in -def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)), - (STriw_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>, - Requires<[HasV4T]>; - // memw(Rx++#s4:2)=Rt // memw(Rx++#s4:2:circ(Mu))=Rt // memw(Rx++I:circ(Mu))=Rt @@ -934,175 +1243,285 @@ def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)), // NV/ST + //===----------------------------------------------------------------------===// -// multiclass for new-value store instructions with base + immediate offset. -// -multiclass ST_Idxd_Pbase_nv { - let isPredicatedNew = isPredNew in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC: $src4), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"($src2+#$src3) = $src4.new", - []>, - Requires<[HasV4T]>; -} +let opNewValue = 2, opExtendable = 1, isExtentSigned = 1, isPredicable = 1 in +class T_store_io_nv MajOp> + : NVInst_V4 <(outs), + (ins IntRegs:$src1, ImmOp:$src2, RC:$src3), + mnemonic#"($src1+#$src2) = $src3.new", + [],"",ST_tc_st_SLOT0> { + bits<5> src1; + bits<13> src2; // Actual address offset + bits<3> src3; + bits<11> offsetBits; // Represents offset encoding -multiclass ST_Idxd_Pred_nv { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_Idxd_Pbase_nv; - // Predicate new - defm _cdn#NAME : ST_Idxd_Pbase_nv; - } -} + let opExtentBits = !if (!eq(mnemonic, "memb"), 11, + !if (!eq(mnemonic, "memh"), 12, + !if (!eq(mnemonic, "memw"), 13, 0))); -let mayStore = 1, isNVStore = 1, hasSideEffects = 0, isExtendable = 1 in -multiclass ST_Idxd_nv ImmBits, - bits<5> PredImmBits> { + let opExtentAlign = !if (!eq(mnemonic, "memb"), 0, + !if (!eq(mnemonic, "memh"), 1, + !if (!eq(mnemonic, "memw"), 2, 0))); - let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in { - let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits, - isPredicable = 1 in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins IntRegs:$src1, ImmOp:$src2, RC:$src3), - mnemonic#"($src1+#$src2) = $src3.new", - []>, - Requires<[HasV4T]>; - - let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits, - isPredicated = 1 in { - defm Pt : ST_Idxd_Pred_nv; - defm NotPt : ST_Idxd_Pred_nv; - } - } -} + let offsetBits = !if (!eq(mnemonic, "memb"), src2{10-0}, + !if (!eq(mnemonic, "memh"), src2{11-1}, + !if (!eq(mnemonic, "memw"), src2{12-2}, 0))); -let addrMode = BaseImmOffset, validSubTargets = HasV4SubT in { - let accessSize = ByteAccess in - defm STrib_indexed: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext, - u6_0Ext, 11, 6>, AddrModeRel; + let IClass = 0b1010; - let accessSize = HalfWordAccess in - defm STrih_indexed: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext, - u6_1Ext, 12, 7>, AddrModeRel; + let Inst{27} = 0b0; + let Inst{26-25} = offsetBits{10-9}; + let Inst{24-21} = 0b1101; + let Inst{20-16} = src1; + let Inst{13} = offsetBits{8}; + let Inst{12-11} = MajOp; + let Inst{10-8} = src3; + let Inst{7-0} = offsetBits{7-0}; + } - let accessSize = WordAccess in - defm STriw_indexed: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext, - u6_2Ext, 13, 8>, AddrModeRel; -} +let opExtendable = 2, opNewValue = 3, isPredicated = 1 in +class T_pstore_io_nv MajOp, bit PredNot, bit isPredNew> + : NVInst_V4 <(outs), + (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC:$src4), + !if(PredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", + ") ")#mnemonic#"($src2+#$src3) = $src4.new", + [],"",V2LDST_tc_st_SLOT0> { + bits<2> src1; + bits<5> src2; + bits<9> src3; + bits<3> src4; + bits<6> offsetBits; // Represents offset encoding + + let isPredicatedNew = isPredNew; + let isPredicatedFalse = PredNot; + let opExtentBits = !if (!eq(mnemonic, "memb"), 6, + !if (!eq(mnemonic, "memh"), 7, + !if (!eq(mnemonic, "memw"), 8, 0))); + + let opExtentAlign = !if (!eq(mnemonic, "memb"), 0, + !if (!eq(mnemonic, "memh"), 1, + !if (!eq(mnemonic, "memw"), 2, 0))); + + let offsetBits = !if (!eq(mnemonic, "memb"), src3{5-0}, + !if (!eq(mnemonic, "memh"), src3{6-1}, + !if (!eq(mnemonic, "memw"), src3{7-2}, 0))); + + let IClass = 0b0100; + + let Inst{27} = 0b0; + let Inst{26} = PredNot; + let Inst{25} = isPredNew; + let Inst{24-21} = 0b0101; + let Inst{20-16} = src2; + let Inst{13} = offsetBits{5}; + let Inst{12-11} = MajOp; + let Inst{10-8} = src4; + let Inst{7-3} = offsetBits{4-0}; + let Inst{2} = 0b0; + let Inst{1-0} = src1; + } // multiclass for new-value store instructions with base + immediate offset. -// and MEMri operand. -multiclass ST_MEMri_Pbase_nv { - let isPredicatedNew = isPredNew in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins PredRegs:$src1, MEMri:$addr, RC: $src2), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"($addr) = $src2.new", - []>, - Requires<[HasV4T]>; -} - -multiclass ST_MEMri_Pred_nv { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_MEMri_Pbase_nv; - - // Predicate new - defm _cdn#NAME : ST_MEMri_Pbase_nv; - } -} - -let mayStore = 1, isNVStore = 1, isExtendable = 1, hasSideEffects = 0 in -multiclass ST_MEMri_nv ImmBits, bits<5> PredImmBits> { - - let CextOpcode = CextOp, BaseOpcode = CextOp in { - let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits, - isPredicable = 1 in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins MEMri:$addr, RC:$src), - mnemonic#"($addr) = $src.new", - []>, - Requires<[HasV4T]>; - - let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits, - hasSideEffects = 0, isPredicated = 1 in { - defm Pt : ST_MEMri_Pred_nv; - defm NotPt : ST_MEMri_Pred_nv; - } +// +let mayStore = 1, isNVStore = 1, isNewValue = 1, hasSideEffects = 0, + isExtendable = 1 in +multiclass ST_Idxd_nv MajOp> { + + let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in { + def S2_#NAME#new_io : T_store_io_nv ; + // Predicated + def S2_p#NAME#newt_io :T_pstore_io_nv ; + def S2_p#NAME#newf_io :T_pstore_io_nv ; + // Predicated new + def S4_p#NAME#newtnew_io :T_pstore_io_nv ; + def S4_p#NAME#newfnew_io :T_pstore_io_nv ; } } -let addrMode = BaseImmOffset, isMEMri = "true", validSubTargets = HasV4SubT, -mayStore = 1 in { +let addrMode = BaseImmOffset, InputType = "imm" in { let accessSize = ByteAccess in - defm STrib: ST_MEMri_nv<"memb", "STrib", IntRegs, 11, 6>, AddrModeRel; + defm storerb: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext, + u6_0Ext, 0b00>, AddrModeRel; - let accessSize = HalfWordAccess in - defm STrih: ST_MEMri_nv<"memh", "STrih", IntRegs, 12, 7>, AddrModeRel; + let accessSize = HalfWordAccess, opExtentAlign = 1 in + defm storerh: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext, + u6_1Ext, 0b01>, AddrModeRel; - let accessSize = WordAccess in - defm STriw: ST_MEMri_nv<"memw", "STriw", IntRegs, 13, 8>, AddrModeRel; + let accessSize = WordAccess, opExtentAlign = 2 in + defm storeri: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext, + u6_2Ext, 0b10>, AddrModeRel; } //===----------------------------------------------------------------------===// -// Post increment store +// Post increment loads with register offset. +//===----------------------------------------------------------------------===// + +let hasNewValue = 1 in +def L2_loadbsw2_pr : T_load_pr <"membh", IntRegs, 0b0001, HalfWordAccess>; + +def L2_loadbsw4_pr : T_load_pr <"membh", DoubleRegs, 0b0111, WordAccess>; + +let hasSideEffects = 0, addrMode = PostInc in +class T_loadalign_pr MajOp, MemAccessSize AccessSz> + : LDInstPI <(outs DoubleRegs:$dst, IntRegs:$_dst_), + (ins DoubleRegs:$src1, IntRegs:$src2, ModRegs:$src3), + "$dst = "#mnemonic#"($src2++$src3)", [], + "$src1 = $dst, $src2 = $_dst_"> { + bits<5> dst; + bits<5> src2; + bits<1> src3; + + let accessSize = AccessSz; + let IClass = 0b1001; + + let Inst{27-25} = 0b110; + let Inst{24-21} = MajOp; + let Inst{20-16} = src2; + let Inst{13} = src3; + let Inst{12} = 0b0; + let Inst{7} = 0b0; + let Inst{4-0} = dst; + } + +def L2_loadalignb_pr : T_loadalign_pr <"memb_fifo", 0b0100, ByteAccess>; +def L2_loadalignh_pr : T_loadalign_pr <"memh_fifo", 0b0010, HalfWordAccess>; + +//===----------------------------------------------------------------------===// +// Template class for non-predicated post increment .new stores // mem[bhwd](Rx++#s4:[0123])=Nt.new //===----------------------------------------------------------------------===// +let isPredicable = 1, hasSideEffects = 0, addrMode = PostInc, isNVStore = 1, + isNewValue = 1, opNewValue = 3 in +class T_StorePI_nv MajOp > + : NVInstPI_V4 <(outs IntRegs:$_dst_), + (ins IntRegs:$src1, ImmOp:$offset, IntRegs:$src2), + mnemonic#"($src1++#$offset) = $src2.new", + [], "$src1 = $_dst_">, + AddrModeRel { + bits<5> src1; + bits<3> src2; + bits<7> offset; + bits<4> offsetBits; -multiclass ST_PostInc_Pbase_nv { - let isPredicatedNew = isPredNew in - def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), - (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"($src2++#$offset) = $src3.new", - [], - "$src2 = $dst">, - Requires<[HasV4T]>; -} + string ImmOpStr = !cast(ImmOp); + let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2}, + !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1}, + /* s4_0Imm */ offset{3-0})); + let IClass = 0b1010; + + let Inst{27-21} = 0b1011101; + let Inst{20-16} = src1; + let Inst{13} = 0b0; + let Inst{12-11} = MajOp; + let Inst{10-8} = src2; + let Inst{7} = 0b0; + let Inst{6-3} = offsetBits; + let Inst{1} = 0b0; + } -multiclass ST_PostInc_Pred_nv { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_PostInc_Pbase_nv; - // Predicate new - let Predicates = [HasV4T], validSubTargets = HasV4SubT in - defm _cdn#NAME : ST_PostInc_Pbase_nv; +//===----------------------------------------------------------------------===// +// Template class for predicated post increment .new stores +// if([!]Pv[.new]) mem[bhwd](Rx++#s4:[0123])=Nt.new +//===----------------------------------------------------------------------===// +let isPredicated = 1, hasSideEffects = 0, addrMode = PostInc, isNVStore = 1, + isNewValue = 1, opNewValue = 4 in +class T_StorePI_nv_pred MajOp, bit isPredNot, bit isPredNew > + : NVInstPI_V4 <(outs IntRegs:$_dst_), + (ins PredRegs:$src1, IntRegs:$src2, + ImmOp:$offset, IntRegs:$src3), + !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", + ") ")#mnemonic#"($src2++#$offset) = $src3.new", + [], "$src2 = $_dst_">, + AddrModeRel { + bits<2> src1; + bits<5> src2; + bits<3> src3; + bits<7> offset; + bits<4> offsetBits; + + string ImmOpStr = !cast(ImmOp); + let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2}, + !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1}, + /* s4_0Imm */ offset{3-0})); + let isPredicatedNew = isPredNew; + let isPredicatedFalse = isPredNot; + + let IClass = 0b1010; + + let Inst{27-21} = 0b1011101; + let Inst{20-16} = src2; + let Inst{13} = 0b1; + let Inst{12-11} = MajOp; + let Inst{10-8} = src3; + let Inst{7} = isPredNew; + let Inst{6-3} = offsetBits; + let Inst{2} = isPredNot; + let Inst{1-0} = src1; } -} -let hasCtrlDep = 1, isNVStore = 1, hasSideEffects = 0 in -multiclass ST_PostInc_nv { +multiclass ST_PostInc_Pred_nv MajOp, bit PredNot> { + def _pi : T_StorePI_nv_pred ; + // Predicate new + def new_pi : T_StorePI_nv_pred ; +} + +multiclass ST_PostInc_nv MajOp> { let BaseOpcode = "POST_"#BaseOp in { - let isPredicable = 1 in - def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), - (ins IntRegs:$src1, ImmOp:$offset, RC:$src2), - mnemonic#"($src1++#$offset) = $src2.new", - [], - "$src1 = $dst">, - Requires<[HasV4T]>; - - let isPredicated = 1 in { - defm Pt : ST_PostInc_Pred_nv; - defm NotPt : ST_PostInc_Pred_nv; - } + def S2_#NAME#_pi : T_StorePI_nv ; + + // Predicated + defm S2_p#NAME#t : ST_PostInc_Pred_nv ; + defm S2_p#NAME#f : ST_PostInc_Pred_nv ; } } -let addrMode = PostInc, validSubTargets = HasV4SubT in { -defm POST_STbri: ST_PostInc_nv <"memb", "STrib", IntRegs, s4_0Imm>, AddrModeRel; -defm POST_SThri: ST_PostInc_nv <"memh", "STrih", IntRegs, s4_1Imm>, AddrModeRel; -defm POST_STwri: ST_PostInc_nv <"memw", "STriw", IntRegs, s4_2Imm>, AddrModeRel; -} +let accessSize = ByteAccess in +defm storerbnew: ST_PostInc_nv <"memb", "STrib", s4_0Imm, 0b00>; + +let accessSize = HalfWordAccess in +defm storerhnew: ST_PostInc_nv <"memh", "STrih", s4_1Imm, 0b01>; + +let accessSize = WordAccess in +defm storerinew: ST_PostInc_nv <"memw", "STriw", s4_2Imm, 0b10>; + +//===----------------------------------------------------------------------===// +// Template class for post increment .new stores with register offset +//===----------------------------------------------------------------------===// +let isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3 in +class T_StorePI_RegNV MajOp, MemAccessSize AccessSz> + : NVInstPI_V4 <(outs IntRegs:$_dst_), + (ins IntRegs:$src1, ModRegs:$src2, IntRegs:$src3), + #mnemonic#"($src1++$src2) = $src3.new", + [], "$src1 = $_dst_"> { + bits<5> src1; + bits<1> src2; + bits<3> src3; + let accessSize = AccessSz; + + let IClass = 0b1010; + + let Inst{27-21} = 0b1101101; + let Inst{20-16} = src1; + let Inst{13} = src2; + let Inst{12-11} = MajOp; + let Inst{10-8} = src3; + let Inst{7} = 0b0; + } + +def S2_storerbnew_pr : T_StorePI_RegNV<"memb", 0b00, ByteAccess>; +def S2_storerhnew_pr : T_StorePI_RegNV<"memh", 0b01, HalfWordAccess>; +def S2_storerinew_pr : T_StorePI_RegNV<"memw", 0b10, WordAccess>; // memb(Rx++#s4:0:circ(Mu))=Nt.new // memb(Rx++I:circ(Mu))=Nt.new -// memb(Rx++Mu)=Nt.new // memb(Rx++Mu:brev)=Nt.new // memh(Rx++#s4:1:circ(Mu))=Nt.new // memh(Rx++I:circ(Mu))=Nt.new @@ -1127,7 +1546,8 @@ defm POST_STwri: ST_PostInc_nv <"memw", "STriw", IntRegs, s4_2Imm>, AddrModeRel; // operands. //===----------------------------------------------------------------------===// -let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11 in +let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11, + opExtentAlign = 2 in class NVJrr_template majOp, bit NvOpNum, bit isNegCond, bit isTak> : NVInst_V4<(outs), @@ -1135,8 +1555,7 @@ class NVJrr_template majOp, bit NvOpNum, "if ("#!if(isNegCond, "!","")#mnemonic# "($src1"#!if(!eq(NvOpNum, 0),".new, ",", ")# "$src2"#!if(!eq(NvOpNum, 1),".new))","))")#" jump:" - #!if(isTak, "t","nt")#" $offset", - []>, Requires<[HasV4T]> { + #!if(isTak, "t","nt")#" $offset", []> { bits<5> src1; bits<5> src2; @@ -1145,14 +1564,14 @@ class NVJrr_template majOp, bit NvOpNum, bits<11> offset; let isTaken = isTak; - let isBrTaken = !if(isTaken, "true", "false"); let isPredicatedFalse = isNegCond; + let opNewValue{0} = NvOpNum; let Ns = !if(!eq(NvOpNum, 0), src1{2-0}, src2{2-0}); let RegOp = !if(!eq(NvOpNum, 0), src2, src1); let IClass = 0b0010; - let Inst{26} = 0b0; + let Inst{27-26} = 0b00; let Inst{25-23} = majOp; let Inst{22} = isNegCond; let Inst{18-16} = Ns; @@ -1166,9 +1585,9 @@ class NVJrr_template majOp, bit NvOpNum, multiclass NVJrr_cond majOp, bit NvOpNum, bit isNegCond> { // Branch not taken: - def _nt_V4: NVJrr_template; + def _nt: NVJrr_template; // Branch taken: - def _t_V4: NVJrr_template; + def _t : NVJrr_template; } // NvOpNum = 0 -> First Operand is a new-value Register @@ -1177,8 +1596,8 @@ multiclass NVJrr_cond majOp, bit NvOpNum, multiclass NVJrr_base majOp, bit NvOpNum> { let BaseOpcode = BaseOp#_NVJ in { - defm _t_Jumpnv : NVJrr_cond; // True cond - defm _f_Jumpnv : NVJrr_cond; // False cond + defm _t_jumpnv : NVJrr_cond; // True cond + defm _f_jumpnv : NVJrr_cond; // False cond } } @@ -1189,12 +1608,12 @@ multiclass NVJrr_base majOp, // if ([!]cmp.gtu(Rt,Ns.new)) jump:[n]t #r9:2 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1, - Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT in { - defm CMPEQrr : NVJrr_base<"cmp.eq", "CMPEQ", 0b000, 0>, PredRel; - defm CMPGTrr : NVJrr_base<"cmp.gt", "CMPGT", 0b001, 0>, PredRel; - defm CMPGTUrr : NVJrr_base<"cmp.gtu", "CMPGTU", 0b010, 0>, PredRel; - defm CMPLTrr : NVJrr_base<"cmp.gt", "CMPLT", 0b011, 1>, PredRel; - defm CMPLTUrr : NVJrr_base<"cmp.gtu", "CMPLTU", 0b100, 1>, PredRel; + Defs = [PC], hasSideEffects = 0 in { + defm J4_cmpeq : NVJrr_base<"cmp.eq", "CMPEQ", 0b000, 0>, PredRel; + defm J4_cmpgt : NVJrr_base<"cmp.gt", "CMPGT", 0b001, 0>, PredRel; + defm J4_cmpgtu : NVJrr_base<"cmp.gtu", "CMPGTU", 0b010, 0>, PredRel; + defm J4_cmplt : NVJrr_base<"cmp.gt", "CMPLT", 0b011, 1>, PredRel; + defm J4_cmpltu : NVJrr_base<"cmp.gtu", "CMPLTU", 0b100, 1>, PredRel; } //===----------------------------------------------------------------------===// @@ -1202,18 +1621,18 @@ let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1, // with a register and an unsigned immediate (U5) operand. //===----------------------------------------------------------------------===// -let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11 in +let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11, + opExtentAlign = 2 in class NVJri_template majOp, bit isNegCond, bit isTak> : NVInst_V4<(outs), (ins IntRegs:$src1, u5Imm:$src2, brtarget:$offset), "if ("#!if(isNegCond, "!","")#mnemonic#"($src1.new, #$src2)) jump:" - #!if(isTak, "t","nt")#" $offset", - []>, Requires<[HasV4T]> { + #!if(isTak, "t","nt")#" $offset", []> { let isTaken = isTak; let isPredicatedFalse = isNegCond; - let isBrTaken = !if(isTaken, "true", "false"); + let isTaken = isTak; bits<3> src1; bits<5> src2; @@ -1232,15 +1651,15 @@ class NVJri_template majOp, bit isNegCond, multiclass NVJri_cond majOp, bit isNegCond> { // Branch not taken: - def _nt_V4: NVJri_template; + def _nt: NVJri_template; // Branch taken: - def _t_V4: NVJri_template; + def _t : NVJri_template; } multiclass NVJri_base majOp> { let BaseOpcode = BaseOp#_NVJri in { - defm _t_Jumpnv : NVJri_cond; // True Cond - defm _f_Jumpnv : NVJri_cond; // False cond + defm _t_jumpnv : NVJri_cond; // True Cond + defm _f_jumpnv : NVJri_cond; // False cond } } @@ -1249,10 +1668,10 @@ multiclass NVJri_base majOp> { // if ([!]cmp.gtu(Ns.new,#U5)) jump:[n]t #r9:2 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1, - Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT in { - defm CMPEQri : NVJri_base<"cmp.eq", "CMPEQ", 0b000>, PredRel; - defm CMPGTri : NVJri_base<"cmp.gt", "CMPGT", 0b001>, PredRel; - defm CMPGTUri : NVJri_base<"cmp.gtu", "CMPGTU", 0b010>, PredRel; + Defs = [PC], hasSideEffects = 0 in { + defm J4_cmpeqi : NVJri_base<"cmp.eq", "CMPEQ", 0b000>, PredRel; + defm J4_cmpgti : NVJri_base<"cmp.gt", "CMPGT", 0b001>, PredRel; + defm J4_cmpgtui : NVJri_base<"cmp.gtu", "CMPGTU", 0b010>, PredRel; } //===----------------------------------------------------------------------===// @@ -1260,19 +1679,19 @@ let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1, // with a register and an hardcoded 0/-1 immediate value. //===----------------------------------------------------------------------===// -let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 11 in +let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 11, + opExtentAlign = 2 in class NVJ_ConstImm_template majOp, string ImmVal, bit isNegCond, bit isTak> : NVInst_V4<(outs), (ins IntRegs:$src1, brtarget:$offset), "if ("#!if(isNegCond, "!","")#mnemonic #"($src1.new, #"#ImmVal#")) jump:" - #!if(isTak, "t","nt")#" $offset", - []>, Requires<[HasV4T]> { + #!if(isTak, "t","nt")#" $offset", []> { let isTaken = isTak; let isPredicatedFalse = isNegCond; - let isBrTaken = !if(isTaken, "true", "false"); + let isTaken = isTak; bits<3> src1; bits<11> offset; @@ -1289,16 +1708,16 @@ class NVJ_ConstImm_template majOp, string ImmVal, multiclass NVJ_ConstImm_cond majOp, string ImmVal, bit isNegCond> { // Branch not taken: - def _nt_V4: NVJ_ConstImm_template; + def _nt: NVJ_ConstImm_template; // Branch taken: - def _t_V4: NVJ_ConstImm_template; + def _t : NVJ_ConstImm_template; } multiclass NVJ_ConstImm_base majOp, string ImmVal> { let BaseOpcode = BaseOp#_NVJ_ConstImm in { - defm _t_Jumpnv : NVJ_ConstImm_cond; // True cond - defm _f_Jumpnv : NVJ_ConstImm_cond; // False Cond + defm _t_jumpnv : NVJ_ConstImm_cond; // True + defm _f_jumpnv : NVJ_ConstImm_cond; // False } } @@ -1307,51 +1726,237 @@ multiclass NVJ_ConstImm_base majOp, // if ([!]cmp.gt(Ns.new,#-1)) jump:[n]t #r9:2 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator=1, - Defs = [PC], hasSideEffects = 0 in { - defm TSTBIT0 : NVJ_ConstImm_base<"tstbit", "TSTBIT", 0b011, "0">, PredRel; - defm CMPEQn1 : NVJ_ConstImm_base<"cmp.eq", "CMPEQ", 0b100, "-1">, PredRel; - defm CMPGTn1 : NVJ_ConstImm_base<"cmp.gt", "CMPGT", 0b101, "-1">, PredRel; + Defs = [PC], hasSideEffects = 0 in { + defm J4_tstbit0 : NVJ_ConstImm_base<"tstbit", "TSTBIT", 0b011, "0">, PredRel; + defm J4_cmpeqn1 : NVJ_ConstImm_base<"cmp.eq", "CMPEQ", 0b100, "-1">, PredRel; + defm J4_cmpgtn1 : NVJ_ConstImm_base<"cmp.gt", "CMPGT", 0b101, "-1">, PredRel; } +// J4_hintjumpr: Hint indirect conditional jump. +let isBranch = 1, isIndirectBranch = 1, hasSideEffects = 0 in +def J4_hintjumpr: JRInst < + (outs), + (ins IntRegs:$Rs), + "hintjr($Rs)"> { + bits<5> Rs; + let IClass = 0b0101; + let Inst{27-21} = 0b0010101; + let Inst{20-16} = Rs; + } + +//===----------------------------------------------------------------------===// +// NV/J - +//===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// CR + +//===----------------------------------------------------------------------===// + +// PC-relative add +let hasNewValue = 1, isExtendable = 1, opExtendable = 1, + isExtentSigned = 0, opExtentBits = 6, hasSideEffects = 0, Uses = [PC] in +def C4_addipc : CRInst <(outs IntRegs:$Rd), (ins u6Ext:$u6), + "$Rd = add(pc, #$u6)", [], "", CR_tc_2_SLOT3 > { + bits<5> Rd; + bits<6> u6; + + let IClass = 0b0110; + let Inst{27-16} = 0b101001001001; + let Inst{12-7} = u6; + let Inst{4-0} = Rd; + } + + + +let hasSideEffects = 0 in +class T_LOGICAL_3OP OpBits, bit IsNeg> + : CRInst<(outs PredRegs:$Pd), + (ins PredRegs:$Ps, PredRegs:$Pt, PredRegs:$Pu), + "$Pd = " # MnOp1 # "($Ps, " # MnOp2 # "($Pt, " # + !if (IsNeg,"!","") # "$Pu))", + [], "", CR_tc_2early_SLOT23> { + bits<2> Pd; + bits<2> Ps; + bits<2> Pt; + bits<2> Pu; + + let IClass = 0b0110; + let Inst{27-24} = 0b1011; + let Inst{23} = IsNeg; + let Inst{22-21} = OpBits; + let Inst{20} = 0b1; + let Inst{17-16} = Ps; + let Inst{13} = 0b0; + let Inst{9-8} = Pt; + let Inst{7-6} = Pu; + let Inst{1-0} = Pd; +} + +def C4_and_and : T_LOGICAL_3OP<"and", "and", 0b00, 0>; +def C4_and_or : T_LOGICAL_3OP<"and", "or", 0b01, 0>; +def C4_or_and : T_LOGICAL_3OP<"or", "and", 0b10, 0>; +def C4_or_or : T_LOGICAL_3OP<"or", "or", 0b11, 0>; +def C4_and_andn : T_LOGICAL_3OP<"and", "and", 0b00, 1>; +def C4_and_orn : T_LOGICAL_3OP<"and", "or", 0b01, 1>; +def C4_or_andn : T_LOGICAL_3OP<"or", "and", 0b10, 1>; +def C4_or_orn : T_LOGICAL_3OP<"or", "or", 0b11, 1>; + +// op(Ps, op(Pt, Pu)) +class LogLog_pat + : Pat<(i1 (Op1 I1:$Ps, (Op2 I1:$Pt, I1:$Pu))), + (MI I1:$Ps, I1:$Pt, I1:$Pu)>; + +// op(Ps, op(Pt, ~Pu)) +class LogLogNot_pat + : Pat<(i1 (Op1 I1:$Ps, (Op2 I1:$Pt, (not I1:$Pu)))), + (MI I1:$Ps, I1:$Pt, I1:$Pu)>; + +def: LogLog_pat; +def: LogLog_pat; +def: LogLog_pat; +def: LogLog_pat; + +def: LogLogNot_pat; +def: LogLogNot_pat; +def: LogLogNot_pat; +def: LogLogNot_pat; + +//===----------------------------------------------------------------------===// +// PIC: Support for PIC compilations. The patterns and SD nodes defined +// below are needed to support code generation for PIC +//===----------------------------------------------------------------------===// + +def SDT_HexagonPICAdd + : SDTypeProfile<1, 1, [SDTCisVT<0, i32>, SDTCisVT<1, i32>]>; +def SDT_HexagonGOTAdd + : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisVT<1, i32>]>; + +def SDT_HexagonGOTAddInternal : SDTypeProfile<1, 1, [SDTCisVT<0, i32>]>; +def SDT_HexagonGOTAddInternalJT : SDTypeProfile<1, 1, [SDTCisVT<0, i32>]>; +def SDT_HexagonGOTAddInternalBA : SDTypeProfile<1, 1, [SDTCisVT<0, i32>]>; + +def Hexagonpic_add : SDNode<"HexagonISD::PIC_ADD", SDT_HexagonPICAdd>; +def Hexagonat_got : SDNode<"HexagonISD::AT_GOT", SDT_HexagonGOTAdd>; +def Hexagongat_pcrel : SDNode<"HexagonISD::AT_PCREL", + SDT_HexagonGOTAddInternal>; +def Hexagongat_pcrel_jt : SDNode<"HexagonISD::AT_PCREL", + SDT_HexagonGOTAddInternalJT>; +def Hexagongat_pcrel_ba : SDNode<"HexagonISD::AT_PCREL", + SDT_HexagonGOTAddInternalBA>; + +// PIC: Map from a block address computation to a PC-relative add +def: Pat<(Hexagongat_pcrel_ba tblockaddress:$src1), + (C4_addipc u32ImmPred:$src1)>; + +// PIC: Map from the computation to generate a GOT pointer to a PC-relative add +def: Pat<(Hexagonpic_add texternalsym:$src1), + (C4_addipc u32ImmPred:$src1)>; + +// PIC: Map from a jump table address computation to a PC-relative add +def: Pat<(Hexagongat_pcrel_jt tjumptable:$src1), + (C4_addipc u32ImmPred:$src1)>; + +// PIC: Map from a GOT-relative symbol reference to a load +def: Pat<(Hexagonat_got (i32 IntRegs:$src1), tglobaladdr:$src2), + (L2_loadri_io IntRegs:$src1, s30_2ImmPred:$src2)>; + +// PIC: Map from a static symbol reference to a PC-relative add +def: Pat<(Hexagongat_pcrel tglobaladdr:$src1), + (C4_addipc u32ImmPred:$src1)>; + +//===----------------------------------------------------------------------===// +// CR - +//===----------------------------------------------------------------------===// + //===----------------------------------------------------------------------===// // XTYPE/ALU + //===----------------------------------------------------------------------===// +// Logical with-not instructions. +def A4_andnp : T_ALU64_logical<"and", 0b001, 1, 0, 1>; +def A4_ornp : T_ALU64_logical<"or", 0b011, 1, 0, 1>; + +def: Pat<(i64 (and (i64 DoubleRegs:$Rs), (i64 (not (i64 DoubleRegs:$Rt))))), + (A4_andnp DoubleRegs:$Rs, DoubleRegs:$Rt)>; +def: Pat<(i64 (or (i64 DoubleRegs:$Rs), (i64 (not (i64 DoubleRegs:$Rt))))), + (A4_ornp DoubleRegs:$Rs, DoubleRegs:$Rt)>; + +let hasNewValue = 1, hasSideEffects = 0 in +def S4_parity: ALU64Inst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt), + "$Rd = parity($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<5> Rt; + + let IClass = 0b1101; + let Inst{27-21} = 0b0101111; + let Inst{20-16} = Rs; + let Inst{12-8} = Rt; + let Inst{4-0} = Rd; +} + // Add and accumulate. // Rd=add(Rs,add(Ru,#s6)) -let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 6, -validSubTargets = HasV4SubT in -def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2, s6Ext:$src3), - "$dst = add($src1, add($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2), - s6_16ExtPred:$src3)))]>, - Requires<[HasV4T]>; - -// Rd=add(Rs,sub(#s6,Ru)) -let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6, -validSubTargets = HasV4SubT in -def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst), - (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3), - "$dst = add($src1, sub(#$src2, $src3))", - [(set (i32 IntRegs:$dst), - (add (i32 IntRegs:$src1), (sub s6_10ExtPred:$src2, - (i32 IntRegs:$src3))))]>, - Requires<[HasV4T]>; - -// Generates the same instruction as ADDr_SUBri_V4 but matches different -// pattern. -// Rd=add(Rs,sub(#s6,Ru)) -let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6, -validSubTargets = HasV4SubT in -def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst), - (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3), - "$dst = add($src1, sub(#$src2, $src3))", - [(set (i32 IntRegs:$dst), - (sub (add (i32 IntRegs:$src1), s6_10ExtPred:$src2), - (i32 IntRegs:$src3)))]>, - Requires<[HasV4T]>; +let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1, opExtentBits = 6, + opExtendable = 3 in +def S4_addaddi : ALU64Inst <(outs IntRegs:$Rd), + (ins IntRegs:$Rs, IntRegs:$Ru, s6Ext:$s6), + "$Rd = add($Rs, add($Ru, #$s6))" , + [(set (i32 IntRegs:$Rd), (add (i32 IntRegs:$Rs), + (add (i32 IntRegs:$Ru), s16_16ImmPred:$s6)))], + "", ALU64_tc_2_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<5> Ru; + bits<6> s6; + + let IClass = 0b1101; + + let Inst{27-23} = 0b10110; + let Inst{22-21} = s6{5-4}; + let Inst{20-16} = Rs; + let Inst{13} = s6{3}; + let Inst{12-8} = Rd; + let Inst{7-5} = s6{2-0}; + let Inst{4-0} = Ru; + } + +let isExtentSigned = 1, hasSideEffects = 0, hasNewValue = 1, isExtendable = 1, + opExtentBits = 6, opExtendable = 2 in +def S4_subaddi: ALU64Inst <(outs IntRegs:$Rd), + (ins IntRegs:$Rs, s6Ext:$s6, IntRegs:$Ru), + "$Rd = add($Rs, sub(#$s6, $Ru))", + [], "", ALU64_tc_2_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<6> s6; + bits<5> Ru; + + let IClass = 0b1101; + + let Inst{27-23} = 0b10111; + let Inst{22-21} = s6{5-4}; + let Inst{20-16} = Rs; + let Inst{13} = s6{3}; + let Inst{12-8} = Rd; + let Inst{7-5} = s6{2-0}; + let Inst{4-0} = Ru; + } + +// Rd=add(Rs,sub(#s6,Ru)) +def: Pat<(add (i32 IntRegs:$src1), (sub s32ImmPred:$src2, + (i32 IntRegs:$src3))), + (S4_subaddi IntRegs:$src1, s32ImmPred:$src2, IntRegs:$src3)>; + +// Rd=sub(add(Rs,#s6),Ru) +def: Pat<(sub (add (i32 IntRegs:$src1), s32ImmPred:$src2), + (i32 IntRegs:$src3)), + (S4_subaddi IntRegs:$src1, s32ImmPred:$src2, IntRegs:$src3)>; + +// Rd=add(sub(Rs,Ru),#s6) +def: Pat<(add (sub (i32 IntRegs:$src1), (i32 IntRegs:$src3)), + (s32ImmPred:$src2)), + (S4_subaddi IntRegs:$src1, s32ImmPred:$src2, IntRegs:$src3)>; // Add or subtract doublewords with carry. @@ -1360,213 +1965,316 @@ def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst), //TODO: // Rdd=sub(Rss,Rtt,Px):carry +// Extract bitfield +// Rdd=extract(Rss,#u6,#U6) +// Rdd=extract(Rss,Rtt) +// Rd=extract(Rs,Rtt) +// Rd=extract(Rs,#u5,#U5) + +def S4_extractp_rp : T_S3op_64 < "extract", 0b11, 0b100, 0>; +def S4_extractp : T_S2op_extract <"extract", 0b1010, DoubleRegs, u6Imm>; + +let hasNewValue = 1 in { + def S4_extract_rp : T_S3op_extract<"extract", 0b01>; + def S4_extract : T_S2op_extract <"extract", 0b1101, IntRegs, u5Imm>; +} + +// Complex add/sub halfwords/words +let Defs = [USR_OVF] in { + def S4_vxaddsubh : T_S3op_64 < "vxaddsubh", 0b01, 0b100, 0, 1>; + def S4_vxaddsubw : T_S3op_64 < "vxaddsubw", 0b01, 0b000, 0, 1>; + def S4_vxsubaddh : T_S3op_64 < "vxsubaddh", 0b01, 0b110, 0, 1>; + def S4_vxsubaddw : T_S3op_64 < "vxsubaddw", 0b01, 0b010, 0, 1>; +} + +let Defs = [USR_OVF] in { + def S4_vxaddsubhr : T_S3op_64 < "vxaddsubh", 0b11, 0b000, 0, 1, 1, 1>; + def S4_vxsubaddhr : T_S3op_64 < "vxsubaddh", 0b11, 0b010, 0, 1, 1, 1>; +} + +let Itinerary = M_tc_3x_SLOT23, Defs = [USR_OVF] in { + def M4_mac_up_s1_sat: T_MType_acc_rr<"+= mpy", 0b011, 0b000, 0, [], 0, 1, 1>; + def M4_nac_up_s1_sat: T_MType_acc_rr<"-= mpy", 0b011, 0b001, 0, [], 0, 1, 1>; +} + +// Logical xor with xor accumulation. +// Rxx^=xor(Rss,Rtt) +let hasSideEffects = 0 in +def M4_xor_xacc + : SInst <(outs DoubleRegs:$Rxx), + (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt), + "$Rxx ^= xor($Rss, $Rtt)", + [(set (i64 DoubleRegs:$Rxx), + (xor (i64 DoubleRegs:$dst2), (xor (i64 DoubleRegs:$Rss), + (i64 DoubleRegs:$Rtt))))], + "$dst2 = $Rxx", S_3op_tc_1_SLOT23> { + bits<5> Rxx; + bits<5> Rss; + bits<5> Rtt; + + let IClass = 0b1100; + + let Inst{27-22} = 0b101010; + let Inst{20-16} = Rss; + let Inst{12-8} = Rtt; + let Inst{7-5} = 0b000; + let Inst{4-0} = Rxx; + } + +// Rotate and reduce bytes +// Rdd=vrcrotate(Rss,Rt,#u2) +let hasSideEffects = 0 in +def S4_vrcrotate + : SInst <(outs DoubleRegs:$Rdd), + (ins DoubleRegs:$Rss, IntRegs:$Rt, u2Imm:$u2), + "$Rdd = vrcrotate($Rss, $Rt, #$u2)", + [], "", S_3op_tc_3x_SLOT23> { + bits<5> Rdd; + bits<5> Rss; + bits<5> Rt; + bits<2> u2; + + let IClass = 0b1100; + + let Inst{27-22} = 0b001111; + let Inst{20-16} = Rss; + let Inst{13} = u2{1}; + let Inst{12-8} = Rt; + let Inst{7-6} = 0b11; + let Inst{5} = u2{0}; + let Inst{4-0} = Rdd; + } + +// Rotate and reduce bytes with accumulation +// Rxx+=vrcrotate(Rss,Rt,#u2) +let hasSideEffects = 0 in +def S4_vrcrotate_acc + : SInst <(outs DoubleRegs:$Rxx), + (ins DoubleRegs:$dst2, DoubleRegs:$Rss, IntRegs:$Rt, u2Imm:$u2), + "$Rxx += vrcrotate($Rss, $Rt, #$u2)", [], + "$dst2 = $Rxx", S_3op_tc_3x_SLOT23> { + bits<5> Rxx; + bits<5> Rss; + bits<5> Rt; + bits<2> u2; + + let IClass = 0b1100; + + let Inst{27-21} = 0b1011101; + let Inst{20-16} = Rss; + let Inst{13} = u2{1}; + let Inst{12-8} = Rt; + let Inst{5} = u2{0}; + let Inst{4-0} = Rxx; + } + +// Vector reduce conditional negate halfwords +let hasSideEffects = 0 in +def S2_vrcnegh + : SInst <(outs DoubleRegs:$Rxx), + (ins DoubleRegs:$dst2, DoubleRegs:$Rss, IntRegs:$Rt), + "$Rxx += vrcnegh($Rss, $Rt)", [], + "$dst2 = $Rxx", S_3op_tc_3x_SLOT23> { + bits<5> Rxx; + bits<5> Rss; + bits<5> Rt; + + let IClass = 0b1100; + + let Inst{27-21} = 0b1011001; + let Inst{20-16} = Rss; + let Inst{13} = 0b1; + let Inst{12-8} = Rt; + let Inst{7-5} = 0b111; + let Inst{4-0} = Rxx; + } + +// Split bitfield +def A4_bitspliti : T_S2op_2_di <"bitsplit", 0b110, 0b100>; + +// Arithmetic/Convergent round +def A4_cround_ri : T_S2op_2_ii <"cround", 0b111, 0b000>; -// Logical doublewords. -// Rdd=and(Rtt,~Rss) -let validSubTargets = HasV4SubT in -def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst), - (ins DoubleRegs:$src1, DoubleRegs:$src2), - "$dst = and($src1, ~$src2)", - [(set (i64 DoubleRegs:$dst), (and (i64 DoubleRegs:$src1), - (not (i64 DoubleRegs:$src2))))]>, - Requires<[HasV4T]>; - -// Rdd=or(Rtt,~Rss) -let validSubTargets = HasV4SubT in -def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst), - (ins DoubleRegs:$src1, DoubleRegs:$src2), - "$dst = or($src1, ~$src2)", - [(set (i64 DoubleRegs:$dst), - (or (i64 DoubleRegs:$src1), (not (i64 DoubleRegs:$src2))))]>, - Requires<[HasV4T]>; - - -// Logical-logical doublewords. -// Rxx^=xor(Rss,Rtt) -let validSubTargets = HasV4SubT in -def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst), - (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3), - "$dst ^= xor($src2, $src3)", - [(set (i64 DoubleRegs:$dst), - (xor (i64 DoubleRegs:$src1), (xor (i64 DoubleRegs:$src2), - (i64 DoubleRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; +def A4_round_ri : T_S2op_2_ii <"round", 0b111, 0b100>; +let Defs = [USR_OVF] in +def A4_round_ri_sat : T_S2op_2_ii <"round", 0b111, 0b110, 1>; // Logical-logical words. -// Rx=or(Ru,and(Rx,#s10)) -let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10, -validSubTargets = HasV4SubT in -def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3), - "$dst = or($src1, and($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - s10ExtPred:$src3)))], - "$src2 = $dst">, - Requires<[HasV4T]>; +// Compound or-and -- Rx=or(Ru,and(Rx,#s10)) +let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1, opExtentBits = 10, + opExtendable = 3 in +def S4_or_andix: + ALU64Inst<(outs IntRegs:$Rx), + (ins IntRegs:$Ru, IntRegs:$_src_, s10Ext:$s10), + "$Rx = or($Ru, and($_src_, #$s10))" , + [(set (i32 IntRegs:$Rx), + (or (i32 IntRegs:$Ru), (and (i32 IntRegs:$_src_), s32ImmPred:$s10)))] , + "$_src_ = $Rx", ALU64_tc_2_SLOT23> { + bits<5> Rx; + bits<5> Ru; + bits<10> s10; + + let IClass = 0b1101; + + let Inst{27-22} = 0b101001; + let Inst{20-16} = Rx; + let Inst{21} = s10{9}; + let Inst{13-5} = s10{8-0}; + let Inst{4-0} = Ru; + } + +// Miscellaneous ALU64 instructions. +// +let hasNewValue = 1, hasSideEffects = 0 in +def A4_modwrapu: ALU64Inst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt), + "$Rd = modwrap($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<5> Rt; + + let IClass = 0b1101; + let Inst{27-21} = 0b0011111; + let Inst{20-16} = Rs; + let Inst{12-8} = Rt; + let Inst{7-5} = 0b111; + let Inst{4-0} = Rd; +} + +let hasSideEffects = 0 in +def A4_bitsplit: ALU64Inst<(outs DoubleRegs:$Rd), + (ins IntRegs:$Rs, IntRegs:$Rt), + "$Rd = bitsplit($Rs, $Rt)", [], "", ALU64_tc_1_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<5> Rt; + + let IClass = 0b1101; + let Inst{27-24} = 0b0100; + let Inst{21} = 0b1; + let Inst{20-16} = Rs; + let Inst{12-8} = Rt; + let Inst{4-0} = Rd; +} + +let hasSideEffects = 0 in +def dep_S2_packhl: ALU64Inst<(outs DoubleRegs:$Rd), + (ins IntRegs:$Rs, IntRegs:$Rt), + "$Rd = packhl($Rs, $Rt):deprecated", [], "", ALU64_tc_1_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<5> Rt; + + let IClass = 0b1101; + let Inst{27-24} = 0b0100; + let Inst{21} = 0b0; + let Inst{20-16} = Rs; + let Inst{12-8} = Rt; + let Inst{4-0} = Rd; +} + +let hasNewValue = 1, hasSideEffects = 0 in +def dep_A2_addsat: ALU64Inst<(outs IntRegs:$Rd), + (ins IntRegs:$Rs, IntRegs:$Rt), + "$Rd = add($Rs, $Rt):sat:deprecated", [], "", ALU64_tc_2_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<5> Rt; + + let IClass = 0b1101; + let Inst{27-21} = 0b0101100; + let Inst{20-16} = Rs; + let Inst{12-8} = Rt; + let Inst{7} = 0b0; + let Inst{4-0} = Rd; +} + +let hasNewValue = 1, hasSideEffects = 0 in +def dep_A2_subsat: ALU64Inst<(outs IntRegs:$Rd), + (ins IntRegs:$Rs, IntRegs:$Rt), + "$Rd = sub($Rs, $Rt):sat:deprecated", [], "", ALU64_tc_2_SLOT23> { + bits<5> Rd; + bits<5> Rs; + bits<5> Rt; + + let IClass = 0b1101; + let Inst{27-21} = 0b0101100; + let Inst{20-16} = Rt; + let Inst{12-8} = Rs; + let Inst{7} = 0b1; + let Inst{4-0} = Rd; +} + +// Rx[&|]=xor(Rs,Rt) +def M4_or_xor : T_MType_acc_rr < "|= xor", 0b110, 0b001, 0>; +def M4_and_xor : T_MType_acc_rr < "&= xor", 0b010, 0b010, 0>; + +// Rx[&|^]=or(Rs,Rt) +def M4_xor_or : T_MType_acc_rr < "^= or", 0b110, 0b011, 0>; + +let CextOpcode = "ORr_ORr" in +def M4_or_or : T_MType_acc_rr < "|= or", 0b110, 0b000, 0>; +def M4_and_or : T_MType_acc_rr < "&= or", 0b010, 0b001, 0>; // Rx[&|^]=and(Rs,Rt) -// Rx&=and(Rs,Rt) -let validSubTargets = HasV4SubT in -def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst &= and($src2, $src3)", - [(set (i32 IntRegs:$dst), - (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx|=and(Rs,Rt) -let validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "reg" in -def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst |= and($src2, $src3)", - [(set (i32 IntRegs:$dst), - (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>, ImmRegRel; - -// Rx^=and(Rs,Rt) -let validSubTargets = HasV4SubT in -def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst ^= and($src2, $src3)", - [(set (i32 IntRegs:$dst), - (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; +def M4_xor_and : T_MType_acc_rr < "^= and", 0b110, 0b010, 0>; + +let CextOpcode = "ORr_ANDr" in +def M4_or_and : T_MType_acc_rr < "|= and", 0b010, 0b011, 0>; +def M4_and_and : T_MType_acc_rr < "&= and", 0b010, 0b000, 0>; // Rx[&|^]=and(Rs,~Rt) -// Rx&=and(Rs,~Rt) -let validSubTargets = HasV4SubT in -def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst &= and($src2, ~$src3)", - [(set (i32 IntRegs:$dst), - (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - (not (i32 IntRegs:$src3)))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx|=and(Rs,~Rt) -let validSubTargets = HasV4SubT in -def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst |= and($src2, ~$src3)", - [(set (i32 IntRegs:$dst), - (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - (not (i32 IntRegs:$src3)))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx^=and(Rs,~Rt) -let validSubTargets = HasV4SubT in -def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst ^= and($src2, ~$src3)", - [(set (i32 IntRegs:$dst), - (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - (not (i32 IntRegs:$src3)))))], - "$src1 = $dst">, - Requires<[HasV4T]>; +def M4_xor_andn : T_MType_acc_rr < "^= and", 0b001, 0b010, 0, [], 1>; +def M4_or_andn : T_MType_acc_rr < "|= and", 0b001, 0b000, 0, [], 1>; +def M4_and_andn : T_MType_acc_rr < "&= and", 0b001, 0b001, 0, [], 1>; + +def: T_MType_acc_pat2 ; +def: T_MType_acc_pat2 ; +def: T_MType_acc_pat2 ; +def: T_MType_acc_pat2 ; +def: T_MType_acc_pat2 ; +def: T_MType_acc_pat2 ; +def: T_MType_acc_pat2 ; +def: T_MType_acc_pat2 ; + +class T_MType_acc_pat3 + : Pat <(i32 (secOp IntRegs:$src1, (firstOp IntRegs:$src2, + (not IntRegs:$src3)))), + (i32 (MI IntRegs:$src1, IntRegs:$src2, IntRegs:$src3))>; + +def: T_MType_acc_pat3 ; +def: T_MType_acc_pat3 ; +def: T_MType_acc_pat3 ; + +// Compound or-or and or-and +let isExtentSigned = 1, InputType = "imm", hasNewValue = 1, isExtendable = 1, + opExtentBits = 10, opExtendable = 3 in +class T_CompOR MajOp, SDNode OpNode> + : MInst_acc <(outs IntRegs:$Rx), + (ins IntRegs:$src1, IntRegs:$Rs, s10Ext:$s10), + "$Rx |= "#mnemonic#"($Rs, #$s10)", + [(set (i32 IntRegs:$Rx), (or (i32 IntRegs:$src1), + (OpNode (i32 IntRegs:$Rs), s32ImmPred:$s10)))], + "$src1 = $Rx", ALU64_tc_2_SLOT23>, ImmRegRel { + bits<5> Rx; + bits<5> Rs; + bits<10> s10; -// Rx[&|^]=or(Rs,Rt) -// Rx&=or(Rs,Rt) -let validSubTargets = HasV4SubT in -def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst &= or($src2, $src3)", - [(set (i32 IntRegs:$dst), - (and (i32 IntRegs:$src1), (or (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx|=or(Rs,Rt) -let validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "reg" in -def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst |= or($src2, $src3)", - [(set (i32 IntRegs:$dst), - (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>, ImmRegRel; - -// Rx^=or(Rs,Rt) -let validSubTargets = HasV4SubT in -def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst ^= or($src2, $src3)", - [(set (i32 IntRegs:$dst), - (xor (i32 IntRegs:$src1), (or (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx[&|^]=xor(Rs,Rt) -// Rx&=xor(Rs,Rt) -let validSubTargets = HasV4SubT in -def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst &= xor($src2, $src3)", - [(set (i32 IntRegs:$dst), - (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx|=xor(Rs,Rt) -let validSubTargets = HasV4SubT in -def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst |= xor($src2, $src3)", - [(set (i32 IntRegs:$dst), - (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx^=xor(Rs,Rt) -let validSubTargets = HasV4SubT in -def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), - "$dst ^= xor($src2, $src3)", - [(set (i32 IntRegs:$dst), - (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -// Rx|=and(Rs,#s10) -let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10, -validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "imm" in -def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3), - "$dst |= and($src2, #$src3)", - [(set (i32 IntRegs:$dst), - (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - s10ExtPred:$src3)))], - "$src1 = $dst">, - Requires<[HasV4T]>, ImmRegRel; - -// Rx|=or(Rs,#s10) -let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10, -validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "imm" in -def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3), - "$dst |= or($src2, #$src3)", - [(set (i32 IntRegs:$dst), - (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), - s10ExtPred:$src3)))], - "$src1 = $dst">, - Requires<[HasV4T]>, ImmRegRel; + let IClass = 0b1101; + + let Inst{27-24} = 0b1010; + let Inst{23-22} = MajOp; + let Inst{20-16} = Rs; + let Inst{21} = s10{9}; + let Inst{13-5} = s10{8-0}; + let Inst{4-0} = Rx; + } + +let CextOpcode = "ORr_ANDr" in +def S4_or_andi : T_CompOR <"and", 0b00, and>; +let CextOpcode = "ORr_ORr" in +def S4_or_ori : T_CompOR <"or", 0b10, or>; // Modulo wrap // Rd=modwrap(Rs,Rt) @@ -1605,269 +2313,486 @@ def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst), // XTYPE/ALU - //===----------------------------------------------------------------------===// +//===----------------------------------------------------------------------===// +// XTYPE/BIT + +//===----------------------------------------------------------------------===// + +// Bit reverse +def S2_brevp : T_S2op_3 <"brev", 0b11, 0b110>; + +// Bit count +def S2_ct0p : T_COUNT_LEADING_64<"ct0", 0b111, 0b010>; +def S2_ct1p : T_COUNT_LEADING_64<"ct1", 0b111, 0b100>; +def S4_clbpnorm : T_COUNT_LEADING_64<"normamt", 0b011, 0b000>; + +// Count trailing zeros: 64-bit. +def: Pat<(i32 (trunc (cttz I64:$Rss))), (S2_ct0p I64:$Rss)>; +def: Pat<(i32 (trunc (cttz_zero_undef I64:$Rss))), (S2_ct0p I64:$Rss)>; + +// Count trailing ones: 64-bit. +def: Pat<(i32 (trunc (cttz (not I64:$Rss)))), (S2_ct1p I64:$Rss)>; +def: Pat<(i32 (trunc (cttz_zero_undef (not I64:$Rss)))), (S2_ct1p I64:$Rss)>; + +// Define leading/trailing patterns that require zero-extensions to 64 bits. +def: Pat<(i64 (ctlz I64:$Rss)), (Zext64 (S2_cl0p I64:$Rss))>; +def: Pat<(i64 (ctlz_zero_undef I64:$Rss)), (Zext64 (S2_cl0p I64:$Rss))>; +def: Pat<(i64 (cttz I64:$Rss)), (Zext64 (S2_ct0p I64:$Rss))>; +def: Pat<(i64 (cttz_zero_undef I64:$Rss)), (Zext64 (S2_ct0p I64:$Rss))>; +def: Pat<(i64 (ctlz (not I64:$Rss))), (Zext64 (S2_cl1p I64:$Rss))>; +def: Pat<(i64 (ctlz_zero_undef (not I64:$Rss))), (Zext64 (S2_cl1p I64:$Rss))>; +def: Pat<(i64 (cttz (not I64:$Rss))), (Zext64 (S2_ct1p I64:$Rss))>; +def: Pat<(i64 (cttz_zero_undef (not I64:$Rss))), (Zext64 (S2_ct1p I64:$Rss))>; + + +let hasSideEffects = 0, hasNewValue = 1 in +def S4_clbaddi : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s6Imm:$s6), + "$Rd = add(clb($Rs), #$s6)", [], "", S_2op_tc_2_SLOT23> { + bits<5> Rs; + bits<5> Rd; + bits<6> s6; + let IClass = 0b1000; + let Inst{27-24} = 0b1100; + let Inst{23-21} = 0b001; + let Inst{20-16} = Rs; + let Inst{13-8} = s6; + let Inst{7-5} = 0b000; + let Inst{4-0} = Rd; +} + +let hasSideEffects = 0, hasNewValue = 1 in +def S4_clbpaddi : SInst<(outs IntRegs:$Rd), (ins DoubleRegs:$Rs, s6Imm:$s6), + "$Rd = add(clb($Rs), #$s6)", [], "", S_2op_tc_2_SLOT23> { + bits<5> Rs; + bits<5> Rd; + bits<6> s6; + let IClass = 0b1000; + let Inst{27-24} = 0b1000; + let Inst{23-21} = 0b011; + let Inst{20-16} = Rs; + let Inst{13-8} = s6; + let Inst{7-5} = 0b010; + let Inst{4-0} = Rd; +} + + +// Bit test/set/clear +def S4_ntstbit_i : T_TEST_BIT_IMM<"!tstbit", 0b001>; +def S4_ntstbit_r : T_TEST_BIT_REG<"!tstbit", 1>; + +let AddedComplexity = 20 in { // Complexity greater than cmp reg-imm. + def: Pat<(i1 (seteq (and (shl 1, u5ImmPred:$u5), (i32 IntRegs:$Rs)), 0)), + (S4_ntstbit_i (i32 IntRegs:$Rs), u5ImmPred:$u5)>; + def: Pat<(i1 (seteq (and (shl 1, (i32 IntRegs:$Rt)), (i32 IntRegs:$Rs)), 0)), + (S4_ntstbit_r (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))>; +} + +// Add extra complexity to prefer these instructions over bitsset/bitsclr. +// The reason is that tstbit/ntstbit can be folded into a compound instruction: +// if ([!]tstbit(...)) jump ... +let AddedComplexity = 100 in +def: Pat<(i1 (setne (and (i32 IntRegs:$Rs), (i32 Set5ImmPred:$u5)), (i32 0))), + (S2_tstbit_i (i32 IntRegs:$Rs), (BITPOS32 Set5ImmPred:$u5))>; + +let AddedComplexity = 100 in +def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 Set5ImmPred:$u5)), (i32 0))), + (S4_ntstbit_i (i32 IntRegs:$Rs), (BITPOS32 Set5ImmPred:$u5))>; + +def C4_nbitsset : T_TEST_BITS_REG<"!bitsset", 0b01, 1>; +def C4_nbitsclr : T_TEST_BITS_REG<"!bitsclr", 0b10, 1>; +def C4_nbitsclri : T_TEST_BITS_IMM<"!bitsclr", 0b10, 1>; + +// Do not increase complexity of these patterns. In the DAG, "cmp i8" may be +// represented as a compare against "value & 0xFF", which is an exact match +// for cmpb (same for cmph). The patterns below do not contain any additional +// complexity that would make them preferable, and if they were actually used +// instead of cmpb/cmph, they would result in a compare against register that +// is loaded with the byte/half mask (i.e. 0xFF or 0xFFFF). +def: Pat<(i1 (setne (and I32:$Rs, u6ImmPred:$u6), 0)), + (C4_nbitsclri I32:$Rs, u6ImmPred:$u6)>; +def: Pat<(i1 (setne (and I32:$Rs, I32:$Rt), 0)), + (C4_nbitsclr I32:$Rs, I32:$Rt)>; +def: Pat<(i1 (setne (and I32:$Rs, I32:$Rt), I32:$Rt)), + (C4_nbitsset I32:$Rs, I32:$Rt)>; + +//===----------------------------------------------------------------------===// +// XTYPE/BIT - +//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===// // XTYPE/MPY + //===----------------------------------------------------------------------===// -// Multiply and user lower result. -// Rd=add(#u6,mpyi(Rs,#U6)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6, -validSubTargets = HasV4SubT in -def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst), - (ins u6Ext:$src1, IntRegs:$src2, u6Imm:$src3), - "$dst = add(#$src1, mpyi($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3), - u6ExtPred:$src1))]>, - Requires<[HasV4T]>; - -// Rd=add(##,mpyi(Rs,#U6)) -def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3), - (HexagonCONST32 tglobaladdr:$src1)), - (i32 (ADDi_MPYri_V4 tglobaladdr:$src1, IntRegs:$src2, - u6ImmPred:$src3))>; +// Rd=add(#u6,mpyi(Rs,#U6)) -- Multiply by immed and add immed. + +let hasNewValue = 1, isExtendable = 1, opExtentBits = 6, opExtendable = 1 in +def M4_mpyri_addi : MInst<(outs IntRegs:$Rd), + (ins u6Ext:$u6, IntRegs:$Rs, u6Imm:$U6), + "$Rd = add(#$u6, mpyi($Rs, #$U6))" , + [(set (i32 IntRegs:$Rd), + (add (mul (i32 IntRegs:$Rs), u6ImmPred:$U6), + u32ImmPred:$u6))] ,"",ALU64_tc_3x_SLOT23> { + bits<5> Rd; + bits<6> u6; + bits<5> Rs; + bits<6> U6; + + let IClass = 0b1101; + + let Inst{27-24} = 0b1000; + let Inst{23} = U6{5}; + let Inst{22-21} = u6{5-4}; + let Inst{20-16} = Rs; + let Inst{13} = u6{3}; + let Inst{12-8} = Rd; + let Inst{7-5} = u6{2-0}; + let Inst{4-0} = U6{4-0}; + } // Rd=add(#u6,mpyi(Rs,Rt)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6, -validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in -def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst), - (ins u6Ext:$src1, IntRegs:$src2, IntRegs:$src3), - "$dst = add(#$src1, mpyi($src2, $src3))", - [(set (i32 IntRegs:$dst), - (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)), - u6ExtPred:$src1))]>, - Requires<[HasV4T]>, ImmRegRel; - -// Rd=add(##,mpyi(Rs,Rt)) -def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)), - (HexagonCONST32 tglobaladdr:$src1)), - (i32 (ADDi_MPYrr_V4 tglobaladdr:$src1, IntRegs:$src2, - IntRegs:$src3))>; - -// Rd=add(Ru,mpyi(#u6:2,Rs)) -let validSubTargets = HasV4SubT in -def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst), - (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3), - "$dst = add($src1, mpyi(#$src2, $src3))", - [(set (i32 IntRegs:$dst), - (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3), - u6_2ImmPred:$src2)))]>, - Requires<[HasV4T]>; - -// Rd=add(Ru,mpyi(Rs,#u6)) -let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 6, -validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in -def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2, u6Ext:$src3), - "$dst = add($src1, mpyi($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2), - u6ExtPred:$src3)))]>, - Requires<[HasV4T]>, ImmRegRel; +let CextOpcode = "ADD_MPY", InputType = "imm", hasNewValue = 1, + isExtendable = 1, opExtentBits = 6, opExtendable = 1 in +def M4_mpyrr_addi : MInst <(outs IntRegs:$Rd), + (ins u6Ext:$u6, IntRegs:$Rs, IntRegs:$Rt), + "$Rd = add(#$u6, mpyi($Rs, $Rt))" , + [(set (i32 IntRegs:$Rd), + (add (mul (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), u32ImmPred:$u6))], + "", ALU64_tc_3x_SLOT23>, ImmRegRel { + bits<5> Rd; + bits<6> u6; + bits<5> Rs; + bits<5> Rt; + + let IClass = 0b1101; + + let Inst{27-23} = 0b01110; + let Inst{22-21} = u6{5-4}; + let Inst{20-16} = Rs; + let Inst{13} = u6{3}; + let Inst{12-8} = Rt; + let Inst{7-5} = u6{2-0}; + let Inst{4-0} = Rd; + } + +let hasNewValue = 1 in +class T_AddMpy + : ALU64Inst <(outs IntRegs:$dst), ins, + "$dst = add($src1, mpyi("#!if(MajOp,"$src3, #$src2))", + "#$src2, $src3))"), + [(set (i32 IntRegs:$dst), + (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3), ImmPred:$src2)))], + "", ALU64_tc_3x_SLOT23> { + bits<5> dst; + bits<5> src1; + bits<8> src2; + bits<5> src3; + + let IClass = 0b1101; + + bits<6> ImmValue = !if(MajOp, src2{5-0}, src2{7-2}); + + let Inst{27-24} = 0b1111; + let Inst{23} = MajOp; + let Inst{22-21} = ImmValue{5-4}; + let Inst{20-16} = src3; + let Inst{13} = ImmValue{3}; + let Inst{12-8} = dst; + let Inst{7-5} = ImmValue{2-0}; + let Inst{4-0} = src1; + } + +def M4_mpyri_addr_u2 : T_AddMpy<0b0, u6_2ImmPred, + (ins IntRegs:$src1, u6_2Imm:$src2, IntRegs:$src3)>; + +let isExtendable = 1, opExtentBits = 6, opExtendable = 3, + CextOpcode = "ADD_MPY", InputType = "imm" in +def M4_mpyri_addr : T_AddMpy<0b1, u32ImmPred, + (ins IntRegs:$src1, IntRegs:$src3, u6Ext:$src2)>, ImmRegRel; // Rx=add(Ru,mpyi(Rx,Rs)) -let validSubTargets = HasV4SubT, InputType = "reg", CextOpcode = "ADD_MPY" in -def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3), - "$dst = add($src1, mpyi($src2, $src3))", - [(set (i32 IntRegs:$dst), - (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2), - (i32 IntRegs:$src3))))], - "$src2 = $dst">, - Requires<[HasV4T]>, ImmRegRel; +let CextOpcode = "ADD_MPY", InputType = "reg", hasNewValue = 1 in +def M4_mpyrr_addr: MInst_acc <(outs IntRegs:$Rx), + (ins IntRegs:$Ru, IntRegs:$_src_, IntRegs:$Rs), + "$Rx = add($Ru, mpyi($_src_, $Rs))", + [(set (i32 IntRegs:$Rx), (add (i32 IntRegs:$Ru), + (mul (i32 IntRegs:$_src_), (i32 IntRegs:$Rs))))], + "$_src_ = $Rx", M_tc_3x_SLOT23>, ImmRegRel { + bits<5> Rx; + bits<5> Ru; + bits<5> Rs; + let IClass = 0b1110; + + let Inst{27-21} = 0b0011000; + let Inst{12-8} = Rx; + let Inst{4-0} = Ru; + let Inst{20-16} = Rs; + } -// Polynomial multiply words -// Rdd=pmpyw(Rs,Rt) -// Rxx^=pmpyw(Rs,Rt) // Vector reduce multiply word by signed half (32x16) -// Rdd=vrmpyweh(Rss,Rtt)[:<<1] -// Rdd=vrmpywoh(Rss,Rtt)[:<<1] -// Rxx+=vrmpyweh(Rss,Rtt)[:<<1] -// Rxx+=vrmpywoh(Rss,Rtt)[:<<1] - -// Multiply and use upper result -// Rd=mpy(Rs,Rt.H):<<1:sat -// Rd=mpy(Rs,Rt.L):<<1:sat -// Rd=mpy(Rs,Rt):<<1 -// Rd=mpy(Rs,Rt):<<1:sat -// Rd=mpysu(Rs,Rt) -// Rx+=mpy(Rs,Rt):<<1:sat -// Rx-=mpy(Rs,Rt):<<1:sat - -// Vector multiply bytes -// Rdd=vmpybsu(Rs,Rt) -// Rdd=vmpybu(Rs,Rt) -// Rxx+=vmpybsu(Rs,Rt) -// Rxx+=vmpybu(Rs,Rt) +//Rdd=vrmpyweh(Rss,Rtt)[:<<1] +def M4_vrmpyeh_s0 : T_M2_vmpy<"vrmpyweh", 0b010, 0b100, 0, 0, 0>; +def M4_vrmpyeh_s1 : T_M2_vmpy<"vrmpyweh", 0b110, 0b100, 1, 0, 0>; + +//Rdd=vrmpywoh(Rss,Rtt)[:<<1] +def M4_vrmpyoh_s0 : T_M2_vmpy<"vrmpywoh", 0b001, 0b010, 0, 0, 0>; +def M4_vrmpyoh_s1 : T_M2_vmpy<"vrmpywoh", 0b101, 0b010, 1, 0, 0>; + +//Rdd+=vrmpyweh(Rss,Rtt)[:<<1] +def M4_vrmpyeh_acc_s0: T_M2_vmpy_acc<"vrmpyweh", 0b001, 0b110, 0, 0>; +def M4_vrmpyeh_acc_s1: T_M2_vmpy_acc<"vrmpyweh", 0b101, 0b110, 1, 0>; + +//Rdd=vrmpywoh(Rss,Rtt)[:<<1] +def M4_vrmpyoh_acc_s0: T_M2_vmpy_acc<"vrmpywoh", 0b011, 0b110, 0, 0>; +def M4_vrmpyoh_acc_s1: T_M2_vmpy_acc<"vrmpywoh", 0b111, 0b110, 1, 0>; + +// Vector multiply halfwords, signed by unsigned +// Rdd=vmpyhsu(Rs,Rt)[:<<]:sat +def M2_vmpy2su_s0 : T_XTYPE_mpy64 < "vmpyhsu", 0b000, 0b111, 1, 0, 0>; +def M2_vmpy2su_s1 : T_XTYPE_mpy64 < "vmpyhsu", 0b100, 0b111, 1, 1, 0>; + +// Rxx+=vmpyhsu(Rs,Rt)[:<<1]:sat +def M2_vmac2su_s0 : T_XTYPE_mpy64_acc < "vmpyhsu", "+", 0b011, 0b101, 1, 0, 0>; +def M2_vmac2su_s1 : T_XTYPE_mpy64_acc < "vmpyhsu", "+", 0b111, 0b101, 1, 1, 0>; // Vector polynomial multiply halfwords // Rdd=vpmpyh(Rs,Rt) +def M4_vpmpyh : T_XTYPE_mpy64 < "vpmpyh", 0b110, 0b111, 0, 0, 0>; + // Rxx^=vpmpyh(Rs,Rt) +def M4_vpmpyh_acc : T_XTYPE_mpy64_acc < "vpmpyh", "^", 0b101, 0b111, 0, 0, 0>; + +// Polynomial multiply words +// Rdd=pmpyw(Rs,Rt) +def M4_pmpyw : T_XTYPE_mpy64 < "pmpyw", 0b010, 0b111, 0, 0, 0>; + +// Rxx^=pmpyw(Rs,Rt) +def M4_pmpyw_acc : T_XTYPE_mpy64_acc < "pmpyw", "^", 0b001, 0b111, 0, 0, 0>; //===----------------------------------------------------------------------===// // XTYPE/MPY - //===----------------------------------------------------------------------===// +//===----------------------------------------------------------------------===// +// ALU64/Vector compare +//===----------------------------------------------------------------------===// +//===----------------------------------------------------------------------===// +// Template class for vector compare +//===----------------------------------------------------------------------===// + +let hasSideEffects = 0 in +class T_vcmpImm cmpOp, bits<2> minOp, Operand ImmOprnd> + : ALU64_rr <(outs PredRegs:$Pd), + (ins DoubleRegs:$Rss, ImmOprnd:$Imm), + "$Pd = "#Str#"($Rss, #$Imm)", + [], "", ALU64_tc_2early_SLOT23> { + bits<2> Pd; + bits<5> Rss; + bits<32> Imm; + bits<8> ImmBits; + let ImmBits{6-0} = Imm{6-0}; + let ImmBits{7} = !if (!eq(cmpOp,0b10), 0b0, Imm{7}); // 0 for vcmp[bhw].gtu + + let IClass = 0b1101; + + let Inst{27-24} = 0b1100; + let Inst{22-21} = cmpOp; + let Inst{20-16} = Rss; + let Inst{12-5} = ImmBits; + let Inst{4-3} = minOp; + let Inst{1-0} = Pd; + } + +// Vector compare bytes +def A4_vcmpbgt : T_vcmp <"vcmpb.gt", 0b1010>; +def: T_vcmp_pat; + +let AsmString = "$Pd = any8(vcmpb.eq($Rss, $Rtt))" in +def A4_vcmpbeq_any : T_vcmp <"any8(vcmpb.gt", 0b1000>; + +def A4_vcmpbeqi : T_vcmpImm <"vcmpb.eq", 0b00, 0b00, u8Imm>; +def A4_vcmpbgti : T_vcmpImm <"vcmpb.gt", 0b01, 0b00, s8Imm>; +def A4_vcmpbgtui : T_vcmpImm <"vcmpb.gtu", 0b10, 0b00, u7Imm>; + +// Vector compare halfwords +def A4_vcmpheqi : T_vcmpImm <"vcmph.eq", 0b00, 0b01, s8Imm>; +def A4_vcmphgti : T_vcmpImm <"vcmph.gt", 0b01, 0b01, s8Imm>; +def A4_vcmphgtui : T_vcmpImm <"vcmph.gtu", 0b10, 0b01, u7Imm>; + +// Vector compare words +def A4_vcmpweqi : T_vcmpImm <"vcmpw.eq", 0b00, 0b10, s8Imm>; +def A4_vcmpwgti : T_vcmpImm <"vcmpw.gt", 0b01, 0b10, s8Imm>; +def A4_vcmpwgtui : T_vcmpImm <"vcmpw.gtu", 0b10, 0b10, u7Imm>; //===----------------------------------------------------------------------===// // XTYPE/SHIFT + //===----------------------------------------------------------------------===// - -// Shift by immediate and accumulate. -// Rx=add(#u8,asl(Rx,#U5)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -validSubTargets = HasV4SubT in -def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = add(#$src1, asl($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - -// Rx=add(#u8,lsr(Rx,#U5)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -validSubTargets = HasV4SubT in -def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = add(#$src1, lsr($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - -// Rx=sub(#u8,asl(Rx,#U5)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -validSubTargets = HasV4SubT in -def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = sub(#$src1, asl($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - -// Rx=sub(#u8,lsr(Rx,#U5)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -validSubTargets = HasV4SubT in -def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = sub(#$src1, lsr($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - - -//Shift by immediate and logical. -//Rx=and(#u8,asl(Rx,#U5)) +// Shift by immediate and accumulate/logical. +// Rx=add(#u8,asl(Rx,#U5)) Rx=add(#u8,lsr(Rx,#U5)) +// Rx=sub(#u8,asl(Rx,#U5)) Rx=sub(#u8,lsr(Rx,#U5)) +// Rx=and(#u8,asl(Rx,#U5)) Rx=and(#u8,lsr(Rx,#U5)) +// Rx=or(#u8,asl(Rx,#U5)) Rx=or(#u8,lsr(Rx,#U5)) let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -validSubTargets = HasV4SubT in -def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = and(#$src1, asl($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - -//Rx=and(#u8,lsr(Rx,#U5)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -validSubTargets = HasV4SubT in -def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = and(#$src1, lsr($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - -//Rx=or(#u8,asl(Rx,#U5)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -AddedComplexity = 30, validSubTargets = HasV4SubT in -def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = or(#$src1, asl($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - -//Rx=or(#u8,lsr(Rx,#U5)) -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8, -AddedComplexity = 30, validSubTargets = HasV4SubT in -def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), - (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3), - "$dst = or(#$src1, lsr($src2, #$src3))", - [(set (i32 IntRegs:$dst), - (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3), - u8ExtPred:$src1))], - "$src2 = $dst">, - Requires<[HasV4T]>; - - -//Shift by register. -//Rd=lsl(#s6,Rt) -let validSubTargets = HasV4SubT in { -def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2), - "$dst = lsl(#$src1, $src2)", - [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1, - (i32 IntRegs:$src2)))]>, - Requires<[HasV4T]>; - - -//Shift by register and logical. -//Rxx^=asl(Rss,Rt) -def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), - (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), - "$dst ^= asl($src2, $src3)", - [(set (i64 DoubleRegs:$dst), - (xor (i64 DoubleRegs:$src1), (shl (i64 DoubleRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -//Rxx^=asr(Rss,Rt) -def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), - (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), - "$dst ^= asr($src2, $src3)", - [(set (i64 DoubleRegs:$dst), - (xor (i64 DoubleRegs:$src1), (sra (i64 DoubleRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -//Rxx^=lsl(Rss,Rt) -def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), - (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), - "$dst ^= lsl($src2, $src3)", - [(set (i64 DoubleRegs:$dst), (xor (i64 DoubleRegs:$src1), - (shl (i64 DoubleRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; - -//Rxx^=lsr(Rss,Rt) -def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), - (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), - "$dst ^= lsr($src2, $src3)", - [(set (i64 DoubleRegs:$dst), - (xor (i64 DoubleRegs:$src1), (srl (i64 DoubleRegs:$src2), - (i32 IntRegs:$src3))))], - "$src1 = $dst">, - Requires<[HasV4T]>; + hasNewValue = 1, opNewValue = 0 in +class T_S4_ShiftOperate MajOp, InstrItinClass Itin> + : MInst_acc<(outs IntRegs:$Rd), (ins u8Ext:$u8, IntRegs:$Rx, u5Imm:$U5), + "$Rd = "#MnOp#"(#$u8, "#MnSh#"($Rx, #$U5))", + [(set (i32 IntRegs:$Rd), + (Op (Sh I32:$Rx, u5ImmPred:$U5), u32ImmPred:$u8))], + "$Rd = $Rx", Itin> { + + bits<5> Rd; + bits<8> u8; + bits<5> Rx; + bits<5> U5; + + let IClass = 0b1101; + let Inst{27-24} = 0b1110; + let Inst{23-21} = u8{7-5}; + let Inst{20-16} = Rd; + let Inst{13} = u8{4}; + let Inst{12-8} = U5; + let Inst{7-5} = u8{3-1}; + let Inst{4} = asl_lsr; + let Inst{3} = u8{0}; + let Inst{2-1} = MajOp; } +multiclass T_ShiftOperate MajOp, + InstrItinClass Itin> { + def _asl_ri : T_S4_ShiftOperate; + def _lsr_ri : T_S4_ShiftOperate; +} + +let AddedComplexity = 200 in { + defm S4_addi : T_ShiftOperate<"add", add, 0b10, ALU64_tc_2_SLOT23>; + defm S4_andi : T_ShiftOperate<"and", and, 0b00, ALU64_tc_2_SLOT23>; +} + +let AddedComplexity = 30 in +defm S4_ori : T_ShiftOperate<"or", or, 0b01, ALU64_tc_1_SLOT23>; + +defm S4_subi : T_ShiftOperate<"sub", sub, 0b11, ALU64_tc_1_SLOT23>; + +let AddedComplexity = 200 in { + def: Pat<(add addrga:$addr, (shl I32:$src2, u5ImmPred:$src3)), + (S4_addi_asl_ri addrga:$addr, IntRegs:$src2, u5ImmPred:$src3)>; + def: Pat<(add addrga:$addr, (srl I32:$src2, u5ImmPred:$src3)), + (S4_addi_lsr_ri addrga:$addr, IntRegs:$src2, u5ImmPred:$src3)>; + def: Pat<(sub addrga:$addr, (shl I32:$src2, u5ImmPred:$src3)), + (S4_subi_asl_ri addrga:$addr, IntRegs:$src2, u5ImmPred:$src3)>; + def: Pat<(sub addrga:$addr, (srl I32:$src2, u5ImmPred:$src3)), + (S4_subi_lsr_ri addrga:$addr, IntRegs:$src2, u5ImmPred:$src3)>; +} + +// Vector conditional negate +// Rdd=vcnegh(Rss,Rt) +let Defs = [USR_OVF], Itinerary = S_3op_tc_2_SLOT23 in +def S2_vcnegh : T_S3op_shiftVect < "vcnegh", 0b11, 0b01>; + +// Rd=[cround|round](Rs,Rt) +let hasNewValue = 1, Itinerary = S_3op_tc_2_SLOT23 in { + def A4_cround_rr : T_S3op_3 < "cround", IntRegs, 0b11, 0b00>; + def A4_round_rr : T_S3op_3 < "round", IntRegs, 0b11, 0b10>; +} + +// Rd=round(Rs,Rt):sat +let hasNewValue = 1, Defs = [USR_OVF], Itinerary = S_3op_tc_2_SLOT23 in +def A4_round_rr_sat : T_S3op_3 < "round", IntRegs, 0b11, 0b11, 1>; + +// Rd=[cmpyiwh|cmpyrwh](Rss,Rt):<<1:rnd:sat +let Defs = [USR_OVF], Itinerary = S_3op_tc_3x_SLOT23 in { + def M4_cmpyi_wh : T_S3op_8<"cmpyiwh", 0b100, 1, 1, 1>; + def M4_cmpyr_wh : T_S3op_8<"cmpyrwh", 0b110, 1, 1, 1>; +} + +// Rdd=[add|sub](Rss,Rtt,Px):carry +let isPredicateLate = 1, hasSideEffects = 0 in +class T_S3op_carry MajOp> + : SInst < (outs DoubleRegs:$Rdd, PredRegs:$Px), + (ins DoubleRegs:$Rss, DoubleRegs:$Rtt, PredRegs:$Pu), + "$Rdd = "#mnemonic#"($Rss, $Rtt, $Pu):carry", + [], "$Px = $Pu", S_3op_tc_1_SLOT23 > { + bits<5> Rdd; + bits<5> Rss; + bits<5> Rtt; + bits<2> Pu; + + let IClass = 0b1100; + + let Inst{27-24} = 0b0010; + let Inst{23-21} = MajOp; + let Inst{20-16} = Rss; + let Inst{12-8} = Rtt; + let Inst{6-5} = Pu; + let Inst{4-0} = Rdd; + } + +def A4_addp_c : T_S3op_carry < "add", 0b110 >; +def A4_subp_c : T_S3op_carry < "sub", 0b111 >; + +let Itinerary = S_3op_tc_3_SLOT23, hasSideEffects = 0 in +class T_S3op_6 MinOp, bit isUnsigned> + : SInst <(outs DoubleRegs:$Rxx), + (ins DoubleRegs:$dst2, DoubleRegs:$Rss, IntRegs:$Ru), + "$Rxx = "#mnemonic#"($Rss, $Ru)" , + [] , "$dst2 = $Rxx"> { + bits<5> Rxx; + bits<5> Rss; + bits<5> Ru; + + let IClass = 0b1100; + + let Inst{27-21} = 0b1011001; + let Inst{20-16} = Rss; + let Inst{13} = isUnsigned; + let Inst{12-8} = Rxx; + let Inst{7-5} = MinOp; + let Inst{4-0} = Ru; + } + +// Vector reduce maximum halfwords +// Rxx=vrmax[u]h(Rss,Ru) +def A4_vrmaxh : T_S3op_6 < "vrmaxh", 0b001, 0>; +def A4_vrmaxuh : T_S3op_6 < "vrmaxuh", 0b001, 1>; + +// Vector reduce maximum words +// Rxx=vrmax[u]w(Rss,Ru) +def A4_vrmaxw : T_S3op_6 < "vrmaxw", 0b010, 0>; +def A4_vrmaxuw : T_S3op_6 < "vrmaxuw", 0b010, 1>; + +// Vector reduce minimum halfwords +// Rxx=vrmin[u]h(Rss,Ru) +def A4_vrminh : T_S3op_6 < "vrminh", 0b101, 0>; +def A4_vrminuh : T_S3op_6 < "vrminuh", 0b101, 1>; + +// Vector reduce minimum words +// Rxx=vrmin[u]w(Rss,Ru) +def A4_vrminw : T_S3op_6 < "vrminw", 0b110, 0>; +def A4_vrminuw : T_S3op_6 < "vrminuw", 0b110, 1>; + +// Shift an immediate left by register amount. +let hasNewValue = 1, hasSideEffects = 0 in +def S4_lsli: SInst <(outs IntRegs:$Rd), (ins s6Imm:$s6, IntRegs:$Rt), + "$Rd = lsl(#$s6, $Rt)" , + [(set (i32 IntRegs:$Rd), (shl s6ImmPred:$s6, + (i32 IntRegs:$Rt)))], + "", S_3op_tc_1_SLOT23> { + bits<5> Rd; + bits<6> s6; + bits<5> Rt; + + let IClass = 0b1100; + + let Inst{27-22} = 0b011010; + let Inst{20-16} = s6{5-1}; + let Inst{12-8} = Rt; + let Inst{7-6} = 0b11; + let Inst{4-0} = Rd; + let Inst{5} = s6{0}; + } + //===----------------------------------------------------------------------===// // XTYPE/SHIFT - //===----------------------------------------------------------------------===// @@ -1880,7 +2805,7 @@ def MEMOPIMM : SDNodeXFormgetSExtValue(); - return XformM5ToU5Imm(imm); + return XformM5ToU5Imm(imm, SDLoc(N)); }]>; def MEMOPIMM_HALF : SDNodeXFormgetSExtValue(); - return XformM5ToU5Imm(imm); + return XformM5ToU5Imm(imm, SDLoc(N)); }]>; def MEMOPIMM_BYTE : SDNodeXFormgetSExtValue(); - return XformM5ToU5Imm(imm); + return XformM5ToU5Imm(imm, SDLoc(N)); }]>; def SETMEMIMM : SDNodeXFormgetSExtValue(); - return XformMskToBitPosU5Imm(imm); + return XformMskToBitPosU5Imm(imm, SDLoc(N)); }]>; def CLRMEMIMM : SDNodeXFormgetSExtValue()); - return XformMskToBitPosU5Imm(imm); + return XformMskToBitPosU5Imm(imm, SDLoc(N)); }]>; def SETMEMIMM_SHORT : SDNodeXFormgetSExtValue(); - return XformMskToBitPosU4Imm(imm); + return XformMskToBitPosU4Imm(imm, SDLoc(N)); }]>; def CLRMEMIMM_SHORT : SDNodeXFormgetSExtValue()); - return XformMskToBitPosU4Imm(imm); + return XformMskToBitPosU4Imm(imm, SDLoc(N)); }]>; def SETMEMIMM_BYTE : SDNodeXFormgetSExtValue(); - return XformMskToBitPosU3Imm(imm); + return XformMskToBitPosU3Imm(imm, SDLoc(N)); }]>; def CLRMEMIMM_BYTE : SDNodeXFormgetSExtValue()); - return XformMskToBitPosU3Imm(imm); + return XformMskToBitPosU3Imm(imm, SDLoc(N)); }]>; //===----------------------------------------------------------------------===// @@ -1955,7 +2880,7 @@ class MemOp_rr_base opcBits, Operand ImmOp, (ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta), opc#"($base+#$offset)"#memOp#"$delta", []>, - Requires<[HasV4T, UseMEMOP]> { + Requires<[UseMEMOP]> { bits<5> base; bits<5> delta; @@ -1966,6 +2891,7 @@ class MemOp_rr_base opcBits, Operand ImmOp, !if (!eq(opcBits, 0b01), offset{6-1}, !if (!eq(opcBits, 0b10), offset{7-2},0))); + let opExtentAlign = opcBits; let IClass = 0b0011; let Inst{27-24} = 0b1110; let Inst{22-21} = opcBits; @@ -1986,7 +2912,7 @@ class MemOp_ri_base opcBits, Operand ImmOp, opc#"($base+#$offset)"#memOp#"#$delta" #!if(memOpBits{1},")", ""), // clrbit, setbit - include ')' []>, - Requires<[HasV4T, UseMEMOP]> { + Requires<[UseMEMOP]> { bits<5> base; bits<5> delta; @@ -1997,6 +2923,7 @@ class MemOp_ri_base opcBits, Operand ImmOp, !if (!eq(opcBits, 0b01), offset{6-1}, !if (!eq(opcBits, 0b10), offset{7-2},0))); + let opExtentAlign = opcBits; let IClass = 0b0011; let Inst{27-24} = 0b1111; let Inst{22-21} = opcBits; @@ -2009,36 +2936,35 @@ class MemOp_ri_base opcBits, Operand ImmOp, // multiclass to define MemOp instructions with register operand. multiclass MemOp_rr opcBits, Operand ImmOp> { - def _ADD#NAME#_V4 : MemOp_rr_base ; - def _SETBIT#NAME#_V4 : MemOp_ri_base; + def L4_iadd#NAME : MemOp_ri_base ; + def L4_ior#NAME : MemOp_ri_base; } multiclass MemOp_base opcBits, Operand ImmOp> { - defm r : MemOp_rr ; - defm i : MemOp_ri ; + defm _#NAME : MemOp_rr ; + defm _#NAME : MemOp_ri ; } // Define MemOp instructions. -let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, -validSubTargets =HasV4SubT in { +let isExtendable = 1, opExtendable = 1, isExtentSigned = 0 in { let opExtentBits = 6, accessSize = ByteAccess in - defm MemOPb : MemOp_base <"memb", 0b00, u6_0Ext>; + defm memopb_io : MemOp_base <"memb", 0b00, u6_0Ext>; let opExtentBits = 7, accessSize = HalfWordAccess in - defm MemOPh : MemOp_base <"memh", 0b01, u6_1Ext>; + defm memoph_io : MemOp_base <"memh", 0b01, u6_1Ext>; let opExtentBits = 8, accessSize = WordAccess in - defm MemOPw : MemOp_base <"memw", 0b10, u6_2Ext>; + defm memopw_io : MemOp_base <"memw", 0b10, u6_2Ext>; } //===----------------------------------------------------------------------===// @@ -2048,43 +2974,43 @@ validSubTargets =HasV4SubT in { // mem[bh](Rs+#u6) += #U5 //===----------------------------------------------------------------------===// -multiclass MemOpi_u5Pats { let AddedComplexity = 180 in - def : Pat < (stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend), - IntRegs:$addr), - (MI IntRegs:$addr, #0, u5ImmPred:$addend )>; + def: Pat<(stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend), + IntRegs:$addr), + (MI IntRegs:$addr, 0, u5ImmPred:$addend)>; let AddedComplexity = 190 in - def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, ExtPred:$offset)), - u5ImmPred:$addend), - (add IntRegs:$base, ExtPred:$offset)), - (MI IntRegs:$base, ExtPred:$offset, u5ImmPred:$addend)>; + def: Pat<(stOp (OpNode (ldOp (add IntRegs:$base, ImmPred:$offset)), + u5ImmPred:$addend), + (add IntRegs:$base, ImmPred:$offset)), + (MI IntRegs:$base, ImmPred:$offset, u5ImmPred:$addend)>; } -multiclass MemOpi_u5ALUOp { - defm : MemOpi_u5Pats; - defm : MemOpi_u5Pats; + defm: MemOpi_u5Pats; + defm: MemOpi_u5Pats; } multiclass MemOpi_u5ExtType { // Half Word - defm : MemOpi_u5ALUOp ; + defm: MemOpi_u5ALUOp ; // Byte - defm : MemOpi_u5ALUOp ; + defm: MemOpi_u5ALUOp ; } -let Predicates = [HasV4T, UseMEMOP] in { - defm : MemOpi_u5ExtType; // zero extend - defm : MemOpi_u5ExtType; // sign extend - defm : MemOpi_u5ExtType; // any extend +let Predicates = [UseMEMOP] in { + defm: MemOpi_u5ExtType; // zero extend + defm: MemOpi_u5ExtType; // sign extend + defm: MemOpi_u5ExtType; // any extend // Word - defm : MemOpi_u5ALUOp ; + defm: MemOpi_u5ALUOp ; } //===----------------------------------------------------------------------===// @@ -2094,38 +3020,37 @@ let Predicates = [HasV4T, UseMEMOP] in { // mem[bh](Rs+#u6) += #m5 //===----------------------------------------------------------------------===// -multiclass MemOpi_m5Pats { +multiclass MemOpi_m5Pats { let AddedComplexity = 190 in - def : Pat <(stOp (add (ldOp IntRegs:$addr), immPred:$subend), - IntRegs:$addr), - (MI IntRegs:$addr, #0, (xformFunc immPred:$subend) )>; + def: Pat<(stOp (add (ldOp IntRegs:$addr), immPred:$subend), IntRegs:$addr), + (MI IntRegs:$addr, 0, (xformFunc immPred:$subend))>; let AddedComplexity = 195 in - def : Pat<(stOp (add (ldOp (add IntRegs:$base, extPred:$offset)), - immPred:$subend), - (add IntRegs:$base, extPred:$offset)), - (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$subend))>; + def: Pat<(stOp (add (ldOp (add IntRegs:$base, ImmPred:$offset)), + immPred:$subend), + (add IntRegs:$base, ImmPred:$offset)), + (MI IntRegs:$base, ImmPred:$offset, (xformFunc immPred:$subend))>; } multiclass MemOpi_m5ExtType { // Half Word - defm : MemOpi_m5Pats ; + defm: MemOpi_m5Pats ; // Byte - defm : MemOpi_m5Pats ; + defm: MemOpi_m5Pats ; } -let Predicates = [HasV4T, UseMEMOP] in { - defm : MemOpi_m5ExtType; // zero extend - defm : MemOpi_m5ExtType; // sign extend - defm : MemOpi_m5ExtType; // any extend +let Predicates = [UseMEMOP] in { + defm: MemOpi_m5ExtType; // zero extend + defm: MemOpi_m5ExtType; // sign extend + defm: MemOpi_m5ExtType; // any extend // Word - defm : MemOpi_m5Pats ; + defm: MemOpi_m5Pats ; } //===----------------------------------------------------------------------===// @@ -2135,52 +3060,50 @@ let Predicates = [HasV4T, UseMEMOP] in { //===----------------------------------------------------------------------===// multiclass MemOpi_bitPats { + PatLeaf extPred, SDNodeXForm xformFunc, InstHexagon MI, + SDNode OpNode> { // mem[bhw](Rs+#u6:[012]) = [clrbit|setbit](#U5) let AddedComplexity = 250 in - def : Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)), - immPred:$bitend), - (add IntRegs:$base, extPred:$offset)), - (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>; + def: Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)), + immPred:$bitend), + (add IntRegs:$base, extPred:$offset)), + (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>; // mem[bhw](Rs+#0) = [clrbit|setbit](#U5) let AddedComplexity = 225 in - def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)), - immPred:$bitend), - (addrPred (i32 IntRegs:$addr), extPred:$offset)), - (MI IntRegs:$addr, extPred:$offset, (xformFunc immPred:$bitend))>; + def: Pat<(stOp (OpNode (ldOp IntRegs:$addr), immPred:$bitend), IntRegs:$addr), + (MI IntRegs:$addr, 0, (xformFunc immPred:$bitend))>; } -multiclass MemOpi_bitExtType { +multiclass MemOpi_bitExtType { // Byte - clrbit - defm : MemOpi_bitPats; + defm: MemOpi_bitPats; // Byte - setbit - defm : MemOpi_bitPats; + defm: MemOpi_bitPats; // Half Word - clrbit - defm : MemOpi_bitPats; + defm: MemOpi_bitPats; // Half Word - setbit - defm : MemOpi_bitPats; + defm: MemOpi_bitPats; } -let Predicates = [HasV4T, UseMEMOP] in { +let Predicates = [UseMEMOP] in { // mem[bh](Rs+#0) = [clrbit|setbit](#U5) // mem[bh](Rs+#u6:[01]) = [clrbit|setbit](#U5) - defm : MemOpi_bitExtType; // zero extend - defm : MemOpi_bitExtType; // sign extend - defm : MemOpi_bitExtType; // any extend + defm: MemOpi_bitExtType; // zero extend + defm: MemOpi_bitExtType; // sign extend + defm: MemOpi_bitExtType; // any extend // memw(Rs+#0) = [clrbit|setbit](#U5) // memw(Rs+#u6:2) = [clrbit|setbit](#U5) - defm : MemOpi_bitPats; - defm : MemOpi_bitPats; + defm: MemOpi_bitPats; + defm: MemOpi_bitPats; } //===----------------------------------------------------------------------===// @@ -2190,54 +3113,51 @@ let Predicates = [HasV4T, UseMEMOP] in { // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt //===----------------------------------------------------------------------===// -multiclass MemOpr_Pats { +multiclass MemOpr_Pats { let AddedComplexity = 141 in // mem[bhw](Rs+#0) [+-&|]= Rt - def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)), - (i32 IntRegs:$addend)), - (addrPred (i32 IntRegs:$addr), extPred:$offset)), - (MI IntRegs:$addr, extPred:$offset, (i32 IntRegs:$addend) )>; + def: Pat<(stOp (OpNode (ldOp IntRegs:$addr), (i32 IntRegs:$addend)), + IntRegs:$addr), + (MI IntRegs:$addr, 0, (i32 IntRegs:$addend))>; // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt let AddedComplexity = 150 in - def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)), - (i32 IntRegs:$orend)), - (add IntRegs:$base, extPred:$offset)), - (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend) )>; + def: Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)), + (i32 IntRegs:$orend)), + (add IntRegs:$base, extPred:$offset)), + (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend))>; } -multiclass MemOPr_ALUOp { - - defm : MemOpr_Pats ; - defm : MemOpr_Pats ; - defm : MemOpr_Pats ; - defm : MemOpr_Pats ; + InstHexagon andMI, InstHexagon orMI> { + defm: MemOpr_Pats ; + defm: MemOpr_Pats ; + defm: MemOpr_Pats ; + defm: MemOpr_Pats ; } multiclass MemOPr_ExtType { // Half Word - defm : MemOPr_ALUOp ; + defm: MemOPr_ALUOp ; // Byte - defm : MemOPr_ALUOp ; + defm: MemOPr_ALUOp ; } // Define 'def Pats' for MemOps with register addend. -let Predicates = [HasV4T, UseMEMOP] in { +let Predicates = [UseMEMOP] in { // Byte, Half Word - defm : MemOPr_ExtType; // zero extend - defm : MemOPr_ExtType; // sign extend - defm : MemOPr_ExtType; // any extend + defm: MemOPr_ExtType; // zero extend + defm: MemOPr_ExtType; // sign extend + defm: MemOPr_ExtType; // any extend // Word - defm : MemOPr_ALUOp ; + defm: MemOPr_ALUOp ; } //===----------------------------------------------------------------------===// @@ -2255,311 +3175,41 @@ let Predicates = [HasV4T, UseMEMOP] in { // incorrect code for negative numbers. // Pd=cmpb.eq(Rs,#u8) -let isCompare = 1, isExtendable = 1, opExtendable = 2, hasSideEffects = 0, - validSubTargets = HasV4SubT in -class CMP_NOT_REG_IMM op, Operand ImmOp, - list Pattern> - : ALU32Inst <(outs PredRegs:$dst), (ins IntRegs:$src1, ImmOp:$src2), - "$dst = !cmp."#OpName#"($src1, #$src2)", - Pattern, - "", ALU32_2op_tc_2early_SLOT0123> { - bits<2> dst; - bits<5> src1; - bits<10> src2; +// p=!cmp.eq(r1,#s10) +def C4_cmpneqi : T_CMP <"cmp.eq", 0b00, 1, s10Ext>; +def C4_cmpltei : T_CMP <"cmp.gt", 0b01, 1, s10Ext>; +def C4_cmplteui : T_CMP <"cmp.gtu", 0b10, 1, u9Ext>; - let IClass = 0b0111; - let Inst{27-24} = 0b0101; - let Inst{23-22} = op; - let Inst{20-16} = src1; - let Inst{21} = !if (!eq(OpName, "gtu"), 0b0, src2{9}); - let Inst{13-5} = src2{8-0}; - let Inst{4-2} = 0b100; - let Inst{1-0} = dst; -} - -let opExtentBits = 10, isExtentSigned = 1 in { -def C4_cmpneqi : CMP_NOT_REG_IMM <"eq", 0b00, s10Ext, [(set (i1 PredRegs:$dst), - (setne (i32 IntRegs:$src1), s10ExtPred:$src2))]>; - -def C4_cmpltei : CMP_NOT_REG_IMM <"gt", 0b01, s10Ext, [(set (i1 PredRegs:$dst), - (not (setgt (i32 IntRegs:$src1), s10ExtPred:$src2)))]>; - -} -let opExtentBits = 9 in -def C4_cmplteui : CMP_NOT_REG_IMM <"gtu", 0b10, u9Ext, [(set (i1 PredRegs:$dst), - (not (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)))]>; - - - -// p=!cmp.eq(r1,r2) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPnotEQ_rr : ALU32_rr<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = !cmp.eq($src1, $src2)", - [(set (i1 PredRegs:$dst), - (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2)))]>, - Requires<[HasV4T]>; - -// p=!cmp.gt(r1,r2) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPnotGT_rr : ALU32_rr<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = !cmp.gt($src1, $src2)", - [(set (i1 PredRegs:$dst), - (not (setgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>, - Requires<[HasV4T]>; - - -// p=!cmp.gtu(r1,r2) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPnotGTU_rr : ALU32_rr<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = !cmp.gtu($src1, $src2)", - [(set (i1 PredRegs:$dst), - (not (setugt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>, - Requires<[HasV4T]>; - -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPbEQri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, u8Imm:$src2), - "$dst = cmpb.eq($src1, #$src2)", - [(set (i1 PredRegs:$dst), - (seteq (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2))]>, - Requires<[HasV4T]>; - -def : Pat <(brcond (i1 (setne (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2)), - bb:$offset), - (J2_jumpf (CMPbEQri_V4 (i32 IntRegs:$src1), u8ImmPred:$src2), - bb:$offset)>, - Requires<[HasV4T]>; - -// Pd=cmpb.eq(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmpb.eq($src1, $src2)", - [(set (i1 PredRegs:$dst), - (seteq (and (xor (i32 IntRegs:$src1), - (i32 IntRegs:$src2)), 255), 0))]>, - Requires<[HasV4T]>; - -// Pd=cmpb.eq(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmpb.eq($src1, $src2)", - [(set (i1 PredRegs:$dst), - (seteq (shl (i32 IntRegs:$src1), (i32 24)), - (shl (i32 IntRegs:$src2), (i32 24))))]>, - Requires<[HasV4T]>; - -// Pd=cmpb.gt(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmpb.gt($src1, $src2)", - [(set (i1 PredRegs:$dst), - (setgt (shl (i32 IntRegs:$src1), (i32 24)), - (shl (i32 IntRegs:$src2), (i32 24))))]>, - Requires<[HasV4T]>; - -// Pd=cmpb.gtu(Rs,#u7) -let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7, -isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", InputType = "imm" in -def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, u7Ext:$src2), - "$dst = cmpb.gtu($src1, #$src2)", - [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255), - u7ExtPred:$src2))]>, - Requires<[HasV4T]>, ImmRegRel; +def : T_CMP_pat ; +def : T_CMP_pat ; +def : T_CMP_pat ; + +// rs <= rt -> !(rs > rt). +/* +def: Pat<(i1 (setle (i32 IntRegs:$src1), s32ImmPred:$src2)), + (C2_not (C2_cmpgti IntRegs:$src1, s32ImmPred:$src2))>; +// (C4_cmpltei IntRegs:$src1, s32ImmPred:$src2)>; +*/ +// Map cmplt(Rs, Imm) -> !cmpgt(Rs, Imm-1). +def: Pat<(i1 (setlt (i32 IntRegs:$src1), s32ImmPred:$src2)), + (C4_cmpltei IntRegs:$src1, (DEC_CONST_SIGNED s32ImmPred:$src2))>; + +// rs != rt -> !(rs == rt). +def: Pat<(i1 (setne (i32 IntRegs:$src1), s32ImmPred:$src2)), + (C4_cmpneqi IntRegs:$src1, s32ImmPred:$src2)>; // SDNode for converting immediate C to C-1. def DEC_CONST_BYTE : SDNodeXFormgetSExtValue(); - return XformU7ToU7M1Imm(imm); + return XformU7ToU7M1Imm(imm, SDLoc(N)); }]>; -// For the sequence -// zext( seteq ( and(Rs, 255), u8)) -// Generate -// Pd=cmpb.eq(Rs, #u8) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 -def : Pat <(i32 (zext (i1 (seteq (i32 (and (i32 IntRegs:$Rs), 255)), - u8ExtPred:$u8)))), - (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs), - (u8ExtPred:$u8))), - 1, 0))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setne ( and(Rs, 255), u8)) -// Generate -// Pd=cmpb.eq(Rs, #u8) -// if (Pd.new) Rd=#0 -// if (!Pd.new) Rd=#1 -def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 IntRegs:$Rs), 255)), - u8ExtPred:$u8)))), - (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs), - (u8ExtPred:$u8))), - 0, 1))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( seteq (Rs, and(Rt, 255))) -// Generate -// Pd=cmpb.eq(Rs, Rt) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 -def : Pat <(i32 (zext (i1 (seteq (i32 IntRegs:$Rt), - (i32 (and (i32 IntRegs:$Rs), 255)))))), - (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs), - (i32 IntRegs:$Rt))), - 1, 0))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setne (Rs, and(Rt, 255))) -// Generate -// Pd=cmpb.eq(Rs, Rt) -// if (Pd.new) Rd=#0 -// if (!Pd.new) Rd=#1 -def : Pat <(i32 (zext (i1 (setne (i32 IntRegs:$Rt), - (i32 (and (i32 IntRegs:$Rs), 255)))))), - (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs), - (i32 IntRegs:$Rt))), - 0, 1))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setugt ( and(Rs, 255), u8)) -// Generate -// Pd=cmpb.gtu(Rs, #u8) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 -def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 255)), - u8ExtPred:$u8)))), - (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs), - (u8ExtPred:$u8))), - 1, 0))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setugt ( and(Rs, 254), u8)) -// Generate -// Pd=cmpb.gtu(Rs, #u8) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 -def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 254)), - u8ExtPred:$u8)))), - (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs), - (u8ExtPred:$u8))), - 1, 0))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setult ( Rs, Rt)) -// Generate -// Pd=cmp.ltu(Rs, Rt) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 -// cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs) -def : Pat <(i32 (zext (i1 (setult (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), - (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt), - (i32 IntRegs:$Rs))), - 1, 0))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setlt ( Rs, Rt)) -// Generate -// Pd=cmp.lt(Rs, Rt) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 -// cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs) -def : Pat <(i32 (zext (i1 (setlt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), - (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt), - (i32 IntRegs:$Rs))), - 1, 0))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setugt ( Rs, Rt)) -// Generate -// Pd=cmp.gtu(Rs, Rt) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 -def : Pat <(i32 (zext (i1 (setugt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), - (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs), - (i32 IntRegs:$Rt))), - 1, 0))>, - Requires<[HasV4T]>; - -// This pattern interefers with coremark performance, not implementing at this -// time. -// For the sequence -// zext( setgt ( Rs, Rt)) -// Generate -// Pd=cmp.gt(Rs, Rt) -// if (Pd.new) Rd=#1 -// if (!Pd.new) Rd=#0 - -// For the sequence -// zext( setuge ( Rs, Rt)) -// Generate -// Pd=cmp.ltu(Rs, Rt) -// if (Pd.new) Rd=#0 -// if (!Pd.new) Rd=#1 -// cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs) -def : Pat <(i32 (zext (i1 (setuge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), - (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt), - (i32 IntRegs:$Rs))), - 0, 1))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setge ( Rs, Rt)) -// Generate -// Pd=cmp.lt(Rs, Rt) -// if (Pd.new) Rd=#0 -// if (!Pd.new) Rd=#1 -// cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs) -def : Pat <(i32 (zext (i1 (setge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), - (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt), - (i32 IntRegs:$Rs))), - 0, 1))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setule ( Rs, Rt)) -// Generate -// Pd=cmp.gtu(Rs, Rt) -// if (Pd.new) Rd=#0 -// if (!Pd.new) Rd=#1 -def : Pat <(i32 (zext (i1 (setule (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), - (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs), - (i32 IntRegs:$Rt))), - 0, 1))>, - Requires<[HasV4T]>; - -// For the sequence -// zext( setle ( Rs, Rt)) -// Generate -// Pd=cmp.gt(Rs, Rt) -// if (Pd.new) Rd=#0 -// if (!Pd.new) Rd=#1 -def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), - (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rs), - (i32 IntRegs:$Rt))), - 0, 1))>, - Requires<[HasV4T]>; - // For the sequence // zext( setult ( and(Rs, 255), u8)) // Use the isdigit transformation below -// Generate code of the form 'mux_ii(cmpbgtu(Rdd, C-1),0,1)' +// Generate code of the form 'C2_muxii(cmpbgtui(Rdd, C-1),0,1)' // for C code of the form r = ((c>='0') & (c<='9')) ? 1 : 0;. // The isdigit transformation relies on two 'clever' aspects: // 1) The data type is unsigned which allows us to eliminate a zero test after @@ -2572,961 +3222,1048 @@ def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))), // The code is transformed upstream of llvm into // retval = (c-48) < 10 ? 1 : 0; let AddedComplexity = 139 in -def : Pat <(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)), - u7StrictPosImmPred:$src2)))), - (i32 (C2_muxii (i1 (CMPbGTUri_V4 (i32 IntRegs:$src1), - (DEC_CONST_BYTE u7StrictPosImmPred:$src2))), - 0, 1))>, - Requires<[HasV4T]>; - -// Pd=cmpb.gtu(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", -InputType = "reg" in -def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmpb.gtu($src1, $src2)", - [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255), - (and (i32 IntRegs:$src2), 255)))]>, - Requires<[HasV4T]>, ImmRegRel; - -// Following instruction is not being extended as it results into the incorrect -// code for negative numbers. - -// Signed half compare(.eq) ri. -// Pd=cmph.eq(Rs,#s8) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPhEQri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, s8Imm:$src2), - "$dst = cmph.eq($src1, #$src2)", - [(set (i1 PredRegs:$dst), (seteq (and (i32 IntRegs:$src1), 65535), - s8ImmPred:$src2))]>, - Requires<[HasV4T]>; - -// Signed half compare(.eq) rr. -// Case 1: xor + and, then compare: -// r0=xor(r0,r1) -// r0=and(r0,#0xffff) -// p0=cmp.eq(r0,#0) -// Pd=cmph.eq(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmph.eq($src1, $src2)", - [(set (i1 PredRegs:$dst), (seteq (and (xor (i32 IntRegs:$src1), - (i32 IntRegs:$src2)), - 65535), 0))]>, - Requires<[HasV4T]>; - -// Signed half compare(.eq) rr. -// Case 2: shift left 16 bits then compare: -// r0=asl(r0,16) -// r1=asl(r1,16) -// p0=cmp.eq(r0,r1) -// Pd=cmph.eq(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmph.eq($src1, $src2)", - [(set (i1 PredRegs:$dst), - (seteq (shl (i32 IntRegs:$src1), (i32 16)), - (shl (i32 IntRegs:$src2), (i32 16))))]>, - Requires<[HasV4T]>; - -/* Incorrect Pattern -- immediate should be right shifted before being -used in the cmph.gt instruction. -// Signed half compare(.gt) ri. -// Pd=cmph.gt(Rs,#s8) - -let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8, -isCompare = 1, validSubTargets = HasV4SubT in -def CMPhGTri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, s8Ext:$src2), - "$dst = cmph.gt($src1, #$src2)", - [(set (i1 PredRegs:$dst), - (setgt (shl (i32 IntRegs:$src1), (i32 16)), - s8ExtPred:$src2))]>, - Requires<[HasV4T]>; -*/ - -// Signed half compare(.gt) rr. -// Pd=cmph.gt(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT in -def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmph.gt($src1, $src2)", - [(set (i1 PredRegs:$dst), - (setgt (shl (i32 IntRegs:$src1), (i32 16)), - (shl (i32 IntRegs:$src2), (i32 16))))]>, - Requires<[HasV4T]>; - -// Unsigned half compare rr (.gtu). -// Pd=cmph.gtu(Rs,Rt) -let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU", -InputType = "reg" in -def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, IntRegs:$src2), - "$dst = cmph.gtu($src1, $src2)", - [(set (i1 PredRegs:$dst), - (setugt (and (i32 IntRegs:$src1), 65535), - (and (i32 IntRegs:$src2), 65535)))]>, - Requires<[HasV4T]>, ImmRegRel; - -// Unsigned half compare ri (.gtu). -// Pd=cmph.gtu(Rs,#u7) -let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7, -isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU", -InputType = "imm" in -def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, u7Ext:$src2), - "$dst = cmph.gtu($src1, #$src2)", - [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535), - u7ExtPred:$src2))]>, - Requires<[HasV4T]>, ImmRegRel; - -let validSubTargets = HasV4SubT in -def NTSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), - "$dst = !tstbit($src1, $src2)", - [(set (i1 PredRegs:$dst), - (seteq (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>, - Requires<[HasV4T]>; - -let validSubTargets = HasV4SubT in -def NTSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2), - "$dst = !tstbit($src1, $src2)", - [(set (i1 PredRegs:$dst), - (seteq (and (shl 1, u5ImmPred:$src2), (i32 IntRegs:$src1)), 0))]>, - Requires<[HasV4T]>; +def: Pat<(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)), + u7StrictPosImmPred:$src2)))), + (C2_muxii (A4_cmpbgtui IntRegs:$src1, + (DEC_CONST_BYTE u7StrictPosImmPred:$src2)), + 0, 1)>; //===----------------------------------------------------------------------===// // XTYPE/PRED - //===----------------------------------------------------------------------===// -//Deallocate frame and return. -// dealloc_return -let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicable = 1, - Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0 in { -let validSubTargets = HasV4SubT in - def DEALLOC_RET_V4 : LD0Inst<(outs), (ins), - "dealloc_return", - []>, - Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// Multiclass for DeallocReturn +//===----------------------------------------------------------------------===// +class L4_RETURN + : LD0Inst<(outs), (ins PredRegs:$src), + !if(isNot, "if (!$src", "if ($src")# + !if(isPredNew, ".new) ", ") ")#mnemonic# + !if(isPredNew, #!if(isTak,":t", ":nt"),""), + [], "", LD_tc_3or4stall_SLOT0> { + + bits<2> src; + let BaseOpcode = "L4_RETURN"; + let isPredicatedFalse = isNot; + let isPredicatedNew = isPredNew; + let isTaken = isTak; + let IClass = 0b1001; + + let Inst{27-16} = 0b011000011110; + + let Inst{13} = isNot; + let Inst{12} = isTak; + let Inst{11} = isPredNew; + let Inst{10} = 0b0; + let Inst{9-8} = src; + let Inst{4-0} = 0b11110; + } + +// Produce all predicated forms, p, !p, p.new, !p.new, :t, :nt +multiclass L4_RETURN_PRED { + let isPredicated = 1 in { + def _#NAME# : L4_RETURN ; + def _#NAME#new_pnt : L4_RETURN ; + def _#NAME#new_pt : L4_RETURN ; + } +} + +multiclass LD_MISC_L4_RETURN { + let isBarrier = 1, isPredicable = 1 in + def NAME : LD0Inst <(outs), (ins), mnemonic, [], "", + LD_tc_3or4stall_SLOT0> { + let BaseOpcode = "L4_RETURN"; + let IClass = 0b1001; + let Inst{27-16} = 0b011000011110; + let Inst{13-10} = 0b0000; + let Inst{4-0} = 0b11110; + } + defm t : L4_RETURN_PRED; + defm f : L4_RETURN_PRED; } +let isReturn = 1, isTerminator = 1, + Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0 in +defm L4_return: LD_MISC_L4_RETURN <"dealloc_return">, PredNewRel; + // Restore registers and dealloc return function call. let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC] in { -let validSubTargets = HasV4SubT in - def RESTORE_DEALLOC_RET_JMP_V4 : JInst<(outs), - (ins calltarget:$dst), - "jump $dst", - []>, - Requires<[HasV4T]>; + Defs = [R29, R30, R31, PC], isPredicable = 0, isAsmParserOnly = 1 in { + def RESTORE_DEALLOC_RET_JMP_V4 : T_JMP<"">; } // Restore registers and dealloc frame before a tail call. -let isCall = 1, isBarrier = 1, - Defs = [R29, R30, R31, PC] in { -let validSubTargets = HasV4SubT in - def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : JInst<(outs), - (ins calltarget:$dst), - "call $dst", - []>, - Requires<[HasV4T]>; +let isCall = 1, Defs = [R29, R30, R31, PC], isAsmParserOnly = 1 in { + def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : T_Call<"">, PredRel; } // Save registers function call. -let isCall = 1, isBarrier = 1, - Uses = [R29, R31] in { - def SAVE_REGISTERS_CALL_V4 : JInst<(outs), - (ins calltarget:$dst), - "call $dst // Save_calle_saved_registers", - []>, - Requires<[HasV4T]>; +let isCall = 1, Uses = [R29, R31], isAsmParserOnly = 1 in { + def SAVE_REGISTERS_CALL_V4 : T_Call<"">, PredRel; } -// if (Ps) dealloc_return -let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0, - isPredicated = 1 in { -let validSubTargets = HasV4SubT in - def DEALLOC_RET_cPt_V4 : LD0Inst<(outs), - (ins PredRegs:$src1), - "if ($src1) dealloc_return", - []>, - Requires<[HasV4T]>; -} - -// if (!Ps) dealloc_return -let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0, - isPredicated = 1, isPredicatedFalse = 1 in { -let validSubTargets = HasV4SubT in - def DEALLOC_RET_cNotPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1), - "if (!$src1) dealloc_return", - []>, - Requires<[HasV4T]>; -} - -// if (Ps.new) dealloc_return:nt -let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0, - isPredicated = 1 in { -let validSubTargets = HasV4SubT in - def DEALLOC_RET_cdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1), - "if ($src1.new) dealloc_return:nt", - []>, - Requires<[HasV4T]>; -} +//===----------------------------------------------------------------------===// +// Template class for non predicated store instructions with +// GP-Relative or absolute addressing. +//===----------------------------------------------------------------------===// +let hasSideEffects = 0, isPredicable = 1, isNVStorable = 1 in +class T_StoreAbsGP MajOp, Operand AddrOp, bit isAbs, bit isHalf> + : STInst<(outs), (ins AddrOp:$addr, RC:$src), + mnemonic # !if(isAbs, "(##", "(#")#"$addr) = $src"#!if(isHalf, ".h",""), + [], "", V2LDST_tc_st_SLOT01> { + bits<19> addr; + bits<5> src; + bits<16> offsetBits; + + string ImmOpStr = !cast(ImmOp); + let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3}, + !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2}, + !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1}, + /* u16_0Imm */ addr{15-0}))); + let IClass = 0b0100; + let Inst{27} = 1; + let Inst{26-25} = offsetBits{15-14}; + let Inst{24} = 0b0; + let Inst{23-22} = MajOp; + let Inst{21} = isHalf; + let Inst{20-16} = offsetBits{13-9}; + let Inst{13} = offsetBits{8}; + let Inst{12-8} = src; + let Inst{7-0} = offsetBits{7-0}; + } -// if (!Ps.new) dealloc_return:nt -let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0, - isPredicated = 1, isPredicatedFalse = 1 in { -let validSubTargets = HasV4SubT in - def DEALLOC_RET_cNotdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1), - "if (!$src1.new) dealloc_return:nt", - []>, - Requires<[HasV4T]>; -} +//===----------------------------------------------------------------------===// +// Template class for predicated store instructions with +// GP-Relative or absolute addressing. +//===----------------------------------------------------------------------===// +let hasSideEffects = 0, isPredicated = 1, isNVStorable = 1, opExtentBits = 6, + opExtendable = 1 in +class T_StoreAbs_Pred MajOp, + bit isHalf, bit isNot, bit isNew> + : STInst<(outs), (ins PredRegs:$src1, u6Ext:$absaddr, RC: $src2), + !if(isNot, "if (!$src1", "if ($src1")#!if(isNew, ".new) ", + ") ")#mnemonic#"(#$absaddr) = $src2"#!if(isHalf, ".h",""), + [], "", ST_tc_st_SLOT01>, AddrModeRel { + bits<2> src1; + bits<6> absaddr; + bits<5> src2; + + let isPredicatedNew = isNew; + let isPredicatedFalse = isNot; + + let IClass = 0b1010; -// if (Ps.new) dealloc_return:t -let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0, - isPredicated = 1 in { -let validSubTargets = HasV4SubT in - def DEALLOC_RET_cdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1), - "if ($src1.new) dealloc_return:t", - []>, - Requires<[HasV4T]>; -} + let Inst{27-24} = 0b1111; + let Inst{23-22} = MajOp; + let Inst{21} = isHalf; + let Inst{17-16} = absaddr{5-4}; + let Inst{13} = isNew; + let Inst{12-8} = src2; + let Inst{7} = 0b1; + let Inst{6-3} = absaddr{3-0}; + let Inst{2} = isNot; + let Inst{1-0} = src1; + } -// if (!Ps.new) dealloc_return:nt -let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0, - isPredicated = 1, isPredicatedFalse = 1 in { -let validSubTargets = HasV4SubT in - def DEALLOC_RET_cNotdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1), - "if (!$src1.new) dealloc_return:t", - []>, - Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// Template class for predicated store instructions with absolute addressing. +//===----------------------------------------------------------------------===// +class T_StoreAbs MajOp, bit isHalf> + : T_StoreAbsGP , + AddrModeRel { + string ImmOpStr = !cast(ImmOp); + let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19, + !if (!eq(ImmOpStr, "u16_2Imm"), 18, + !if (!eq(ImmOpStr, "u16_1Imm"), 17, + /* u16_0Imm */ 16))); + + let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3, + !if (!eq(ImmOpStr, "u16_2Imm"), 2, + !if (!eq(ImmOpStr, "u16_1Imm"), 1, + /* u16_0Imm */ 0))); } -// Load/Store with absolute addressing mode -// memw(#u6)=Rt +//===----------------------------------------------------------------------===// +// Multiclass for store instructions with absolute addressing. +//===----------------------------------------------------------------------===// +let addrMode = Absolute, isExtended = 1 in +multiclass ST_Abs MajOp, bit isHalf = 0> { + let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in { + let opExtendable = 0, isPredicable = 1 in + def S2_#NAME#abs : T_StoreAbs ; -multiclass ST_Abs_Predbase { - let isPredicatedNew = isPredNew in - def NAME#_V4 : STInst2<(outs), - (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"(##$absaddr) = $src2", - []>, - Requires<[HasV4T]>; -} + // Predicated + def S4_p#NAME#t_abs : T_StoreAbs_Pred; + def S4_p#NAME#f_abs : T_StoreAbs_Pred; -multiclass ST_Abs_Pred { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_Abs_Predbase; - // Predicate new - defm _cdn#NAME : ST_Abs_Predbase; + // .new Predicated + def S4_p#NAME#tnew_abs : T_StoreAbs_Pred; + def S4_p#NAME#fnew_abs : T_StoreAbs_Pred; } } -let isNVStorable = 1, isExtended = 1, hasSideEffects = 0 in -multiclass ST_Abs { - let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in { - let opExtendable = 0, isPredicable = 1 in - def NAME#_V4 : STInst2<(outs), - (ins u0AlwaysExt:$absaddr, RC:$src), - mnemonic#"(##$absaddr) = $src", - []>, - Requires<[HasV4T]>; - - let opExtendable = 1, isPredicated = 1 in { - defm Pt : ST_Abs_Pred; - defm NotPt : ST_Abs_Pred; - } +//===----------------------------------------------------------------------===// +// Template class for non predicated new-value store instructions with +// GP-Relative or absolute addressing. +//===----------------------------------------------------------------------===// +let hasSideEffects = 0, isPredicable = 1, mayStore = 1, isNVStore = 1, + isNewValue = 1, opNewValue = 1 in +class T_StoreAbsGP_NV MajOp, bit isAbs> + : NVInst_V4<(outs), (ins u32Imm:$addr, IntRegs:$src), + mnemonic # !if(isAbs, "(##", "(#")#"$addr) = $src.new", + [], "", V2LDST_tc_st_SLOT0> { + bits<19> addr; + bits<3> src; + bits<16> offsetBits; + + string ImmOpStr = !cast(ImmOp); + let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3}, + !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2}, + !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1}, + /* u16_0Imm */ addr{15-0}))); + let IClass = 0b0100; + + let Inst{27} = 1; + let Inst{26-25} = offsetBits{15-14}; + let Inst{24-21} = 0b0101; + let Inst{20-16} = offsetBits{13-9}; + let Inst{13} = offsetBits{8}; + let Inst{12-11} = MajOp; + let Inst{10-8} = src; + let Inst{7-0} = offsetBits{7-0}; } -} -multiclass ST_Abs_Predbase_nv { - let isPredicatedNew = isPredNew in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#mnemonic#"(##$absaddr) = $src2.new", - []>, - Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// Template class for predicated new-value store instructions with +// absolute addressing. +//===----------------------------------------------------------------------===// +let hasSideEffects = 0, isPredicated = 1, mayStore = 1, isNVStore = 1, + isNewValue = 1, opNewValue = 2, opExtentBits = 6, opExtendable = 1 in +class T_StoreAbs_NV_Pred MajOp, bit isNot, bit isNew> + : NVInst_V4<(outs), (ins PredRegs:$src1, u6Ext:$absaddr, IntRegs:$src2), + !if(isNot, "if (!$src1", "if ($src1")#!if(isNew, ".new) ", + ") ")#mnemonic#"(#$absaddr) = $src2.new", + [], "", ST_tc_st_SLOT0>, AddrModeRel { + bits<2> src1; + bits<6> absaddr; + bits<3> src2; + + let isPredicatedNew = isNew; + let isPredicatedFalse = isNot; + + let IClass = 0b1010; + + let Inst{27-24} = 0b1111; + let Inst{23-21} = 0b101; + let Inst{17-16} = absaddr{5-4}; + let Inst{13} = isNew; + let Inst{12-11} = MajOp; + let Inst{10-8} = src2; + let Inst{7} = 0b1; + let Inst{6-3} = absaddr{3-0}; + let Inst{2} = isNot; + let Inst{1-0} = src1; } -multiclass ST_Abs_Pred_nv { - let isPredicatedFalse = PredNot in { - defm _c#NAME : ST_Abs_Predbase_nv; - // Predicate new - defm _cdn#NAME : ST_Abs_Predbase_nv; - } +//===----------------------------------------------------------------------===// +// Template class for non-predicated new-value store instructions with +// absolute addressing. +//===----------------------------------------------------------------------===// +class T_StoreAbs_NV MajOp> + : T_StoreAbsGP_NV , AddrModeRel { + + string ImmOpStr = !cast(ImmOp); + let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19, + !if (!eq(ImmOpStr, "u16_2Imm"), 18, + !if (!eq(ImmOpStr, "u16_1Imm"), 17, + /* u16_0Imm */ 16))); + + let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3, + !if (!eq(ImmOpStr, "u16_2Imm"), 2, + !if (!eq(ImmOpStr, "u16_1Imm"), 1, + /* u16_0Imm */ 0))); } -let mayStore = 1, isNVStore = 1, isExtended = 1, hasSideEffects = 0 in -multiclass ST_Abs_nv { +//===----------------------------------------------------------------------===// +// Multiclass for new-value store instructions with absolute addressing. +//===----------------------------------------------------------------------===// +let addrMode = Absolute, isExtended = 1 in +multiclass ST_Abs_NV MajOp> { let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in { let opExtendable = 0, isPredicable = 1 in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins u0AlwaysExt:$absaddr, RC:$src), - mnemonic#"(##$absaddr) = $src.new", - []>, - Requires<[HasV4T]>; - - let opExtendable = 1, isPredicated = 1 in { - defm Pt : ST_Abs_Pred_nv; - defm NotPt : ST_Abs_Pred_nv; - } - } -} - -let addrMode = Absolute in { - let accessSize = ByteAccess in - defm STrib_abs : ST_Abs<"memb", "STrib", IntRegs>, - ST_Abs_nv<"memb", "STrib", IntRegs>, AddrModeRel; - - let accessSize = HalfWordAccess in - defm STrih_abs : ST_Abs<"memh", "STrih", IntRegs>, - ST_Abs_nv<"memh", "STrih", IntRegs>, AddrModeRel; + def S2_#NAME#newabs : T_StoreAbs_NV ; - let accessSize = WordAccess in - defm STriw_abs : ST_Abs<"memw", "STriw", IntRegs>, - ST_Abs_nv<"memw", "STriw", IntRegs>, AddrModeRel; + // Predicated + def S4_p#NAME#newt_abs : T_StoreAbs_NV_Pred ; + def S4_p#NAME#newf_abs : T_StoreAbs_NV_Pred ; - let accessSize = DoubleWordAccess, isNVStorable = 0 in - defm STrid_abs : ST_Abs<"memd", "STrid", DoubleRegs>, AddrModeRel; + // .new Predicated + def S4_p#NAME#newtnew_abs : T_StoreAbs_NV_Pred ; + def S4_p#NAME#newfnew_abs : T_StoreAbs_NV_Pred ; + } } -let Predicates = [HasV4T], AddedComplexity = 30 in { -def : Pat<(truncstorei8 (i32 IntRegs:$src1), - (HexagonCONST32 tglobaladdr:$absaddr)), - (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>; +//===----------------------------------------------------------------------===// +// Stores with absolute addressing +//===----------------------------------------------------------------------===// +let accessSize = ByteAccess in +defm storerb : ST_Abs <"memb", "STrib", IntRegs, u16_0Imm, 0b00>, + ST_Abs_NV <"memb", "STrib", u16_0Imm, 0b00>; + +let accessSize = HalfWordAccess in +defm storerh : ST_Abs <"memh", "STrih", IntRegs, u16_1Imm, 0b01>, + ST_Abs_NV <"memh", "STrih", u16_1Imm, 0b01>; -def : Pat<(truncstorei16 (i32 IntRegs:$src1), - (HexagonCONST32 tglobaladdr:$absaddr)), - (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>; +let accessSize = WordAccess in +defm storeri : ST_Abs <"memw", "STriw", IntRegs, u16_2Imm, 0b10>, + ST_Abs_NV <"memw", "STriw", u16_2Imm, 0b10>; -def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)), - (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>; +let isNVStorable = 0, accessSize = DoubleWordAccess in +defm storerd : ST_Abs <"memd", "STrid", DoubleRegs, u16_3Imm, 0b11>; -def : Pat<(store (i64 DoubleRegs:$src1), - (HexagonCONST32 tglobaladdr:$absaddr)), - (STrid_abs_V4 tglobaladdr: $absaddr, DoubleRegs: $src1)>; -} +let isNVStorable = 0, accessSize = HalfWordAccess in +defm storerf : ST_Abs <"memh", "STrif", IntRegs, u16_1Imm, 0b01, 1>; //===----------------------------------------------------------------------===// -// multiclass for store instructions with GP-relative addressing mode. +// GP-relative stores. // mem[bhwd](#global)=Rt -// if ([!]Pv[.new]) mem[bhwd](##global) = Rt +// Once predicated, these instructions map to absolute addressing mode. +// if ([!]Pv[.new]) mem[bhwd](##global)=Rt //===----------------------------------------------------------------------===// -let mayStore = 1, isNVStorable = 1 in -multiclass ST_GP { - let BaseOpcode = BaseOp, isPredicable = 1 in - def NAME#_V4 : STInst2<(outs), - (ins globaladdress:$global, RC:$src), - mnemonic#"(#$global) = $src", - []>; - // When GP-relative instructions are predicated, their addressing mode is - // changed to absolute and they are always constant extended. - let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1, - isPredicated = 1 in { - defm Pt : ST_Abs_Pred ; - defm NotPt : ST_Abs_Pred ; +let isAsmParserOnly = 1 in +class T_StoreGP MajOp, bit isHalf = 0> + : T_StoreAbsGP { + // Set BaseOpcode same as absolute addressing instructions so that + // non-predicated GP-Rel instructions can have relate with predicated + // Absolute instruction. + let BaseOpcode = BaseOp#_abs; + } + +let isAsmParserOnly = 1 in +multiclass ST_GP MajOp, bit isHalf = 0> { + // Set BaseOpcode same as absolute addressing instructions so that + // non-predicated GP-Rel instructions can have relate with predicated + // Absolute instruction. + let BaseOpcode = BaseOp#_abs in { + def NAME#gp : T_StoreAbsGP ; + // New-value store + def NAME#newgp : T_StoreAbsGP_NV ; } } -let mayStore = 1, isNVStore = 1 in -multiclass ST_GP_nv { - let BaseOpcode = BaseOp, isPredicable = 1 in - def NAME#_nv_V4 : NVInst_V4<(outs), - (ins u0AlwaysExt:$global, RC:$src), - mnemonic#"(#$global) = $src.new", - []>, - Requires<[HasV4T]>; - - // When GP-relative instructions are predicated, their addressing mode is - // changed to absolute and they are always constant extended. - let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1, - isPredicated = 1 in { - defm Pt : ST_Abs_Pred_nv; - defm NotPt : ST_Abs_Pred_nv; - } -} - -let validSubTargets = HasV4SubT, hasSideEffects = 0 in { - let isNVStorable = 0 in - defm STd_GP : ST_GP <"memd", "STd_GP", DoubleRegs>, PredNewRel; - - defm STb_GP : ST_GP<"memb", "STb_GP", IntRegs>, - ST_GP_nv<"memb", "STb_GP", IntRegs>, NewValueRel; - defm STh_GP : ST_GP<"memh", "STh_GP", IntRegs>, - ST_GP_nv<"memh", "STh_GP", IntRegs>, NewValueRel; - defm STw_GP : ST_GP<"memw", "STw_GP", IntRegs>, - ST_GP_nv<"memw", "STw_GP", IntRegs>, NewValueRel; -} - -// 64 bit atomic store -def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global), - (i64 DoubleRegs:$src1)), - (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>, - Requires<[HasV4T]>; - -// Map from store(globaladdress) -> memd(#foo) -let AddedComplexity = 100 in -def : Pat <(store (i64 DoubleRegs:$src1), - (HexagonCONST32_GP tglobaladdr:$global)), - (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>; +let accessSize = ByteAccess in +defm S2_storerb : ST_GP<"memb", "STrib", u16_0Imm, 0b00>, NewValueRel; -// 8 bit atomic store -def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global), - (i32 IntRegs:$src1)), - (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>; +let accessSize = HalfWordAccess in +defm S2_storerh : ST_GP<"memh", "STrih", u16_1Imm, 0b01>, NewValueRel; -// Map from store(globaladdress) -> memb(#foo) -let AddedComplexity = 100 in -def : Pat<(truncstorei8 (i32 IntRegs:$src1), - (HexagonCONST32_GP tglobaladdr:$global)), - (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>; +let accessSize = WordAccess in +defm S2_storeri : ST_GP<"memw", "STriw", u16_2Imm, 0b10>, NewValueRel; -// Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1" -// to "r0 = 1; memw(#foo) = r0" -let AddedComplexity = 100 in -def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)), - (STb_GP_V4 tglobaladdr:$global, (A2_tfrsi 1))>; +let isNVStorable = 0, accessSize = DoubleWordAccess in +def S2_storerdgp : T_StoreGP <"memd", "STrid", DoubleRegs, + u16_3Imm, 0b11>, PredNewRel; -def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global), - (i32 IntRegs:$src1)), - (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>; +let isNVStorable = 0, accessSize = HalfWordAccess in +def S2_storerfgp : T_StoreGP <"memh", "STrif", IntRegs, + u16_1Imm, 0b01, 1>, PredNewRel; -// Map from store(globaladdress) -> memh(#foo) -let AddedComplexity = 100 in -def : Pat<(truncstorei16 (i32 IntRegs:$src1), - (HexagonCONST32_GP tglobaladdr:$global)), - (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>; +class Loada_pat + : Pat<(VT (Load Addr:$addr)), (MI Addr:$addr)>; -// 32 bit atomic store -def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global), - (i32 IntRegs:$src1)), - (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>; +class Loadam_pat + : Pat<(VT (Load Addr:$addr)), (ValueMod (MI Addr:$addr))>; -// Map from store(globaladdress) -> memw(#foo) -let AddedComplexity = 100 in -def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)), - (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>; +class Storea_pat + : Pat<(Store Value:$val, Addr:$addr), (MI Addr:$addr, Value:$val)>; -//===----------------------------------------------------------------------===// -// Multiclass for the load instructions with absolute addressing mode. -//===----------------------------------------------------------------------===// -multiclass LD_Abs_Predbase { - let isPredicatedNew = isPredNew in - def NAME : LDInst2<(outs RC:$dst), - (ins PredRegs:$src1, u0AlwaysExt:$absaddr), - !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", - ") ")#"$dst = "#mnemonic#"(##$absaddr)", - []>, - Requires<[HasV4T]>; -} +class Stoream_pat + : Pat<(Store Value:$val, Addr:$addr), + (MI Addr:$addr, (ValueMod Value:$val))>; -multiclass LD_Abs_Pred { - let isPredicatedFalse = PredNot in { - defm _c#NAME : LD_Abs_Predbase; - // Predicate new - defm _cdn#NAME : LD_Abs_Predbase; - } +def: Storea_pat, I32, addrgp, S2_storerbgp>; +def: Storea_pat, I32, addrgp, S2_storerhgp>; +def: Storea_pat, I32, addrgp, S2_storerigp>; +def: Storea_pat, I64, addrgp, S2_storerdgp>; + +let AddedComplexity = 100 in { + def: Storea_pat; + def: Storea_pat; + def: Storea_pat; + def: Storea_pat; + + // Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1" + // to "r0 = 1; memw(#foo) = r0" + let AddedComplexity = 100 in + def: Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)), + (S2_storerbgp tglobaladdr:$global, (A2_tfrsi 1))>; } -let isExtended = 1, hasSideEffects = 0 in -multiclass LD_Abs { - let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in { - let opExtendable = 1, isPredicable = 1 in - def NAME#_V4 : LDInst2<(outs RC:$dst), - (ins u0AlwaysExt:$absaddr), - "$dst = "#mnemonic#"(##$absaddr)", - []>, - Requires<[HasV4T]>; - - let opExtendable = 2, isPredicated = 1 in { - defm Pt_V4 : LD_Abs_Pred; - defm NotPt_V4 : LD_Abs_Pred; - } +//===----------------------------------------------------------------------===// +// Template class for non predicated load instructions with +// absolute addressing mode. +//===----------------------------------------------------------------------===// +let isPredicable = 1, hasSideEffects = 0 in +class T_LoadAbsGP MajOp, Operand AddrOp, bit isAbs> + : LDInst <(outs RC:$dst), (ins AddrOp:$addr), + "$dst = "#mnemonic# !if(isAbs, "(##", "(#")#"$addr)", + [], "", V2LDST_tc_ld_SLOT01> { + bits<5> dst; + bits<19> addr; + bits<16> offsetBits; + + string ImmOpStr = !cast(ImmOp); + let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3}, + !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2}, + !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1}, + /* u16_0Imm */ addr{15-0}))); + + let IClass = 0b0100; + + let Inst{27} = 0b1; + let Inst{26-25} = offsetBits{15-14}; + let Inst{24} = 0b1; + let Inst{23-21} = MajOp; + let Inst{20-16} = offsetBits{13-9}; + let Inst{13-5} = offsetBits{8-0}; + let Inst{4-0} = dst; } -} -let addrMode = Absolute in { - let accessSize = ByteAccess in { - defm LDrib_abs : LD_Abs<"memb", "LDrib", IntRegs>, AddrModeRel; - defm LDriub_abs : LD_Abs<"memub", "LDriub", IntRegs>, AddrModeRel; +class T_LoadAbs MajOp> + : T_LoadAbsGP , AddrModeRel { + + string ImmOpStr = !cast(ImmOp); + let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19, + !if (!eq(ImmOpStr, "u16_2Imm"), 18, + !if (!eq(ImmOpStr, "u16_1Imm"), 17, + /* u16_0Imm */ 16))); + + let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3, + !if (!eq(ImmOpStr, "u16_2Imm"), 2, + !if (!eq(ImmOpStr, "u16_1Imm"), 1, + /* u16_0Imm */ 0))); } - let accessSize = HalfWordAccess in { - defm LDrih_abs : LD_Abs<"memh", "LDrih", IntRegs>, AddrModeRel; - defm LDriuh_abs : LD_Abs<"memuh", "LDriuh", IntRegs>, AddrModeRel; + +//===----------------------------------------------------------------------===// +// Template class for predicated load instructions with +// absolute addressing mode. +//===----------------------------------------------------------------------===// +let isPredicated = 1, opExtentBits = 6, opExtendable = 2 in +class T_LoadAbs_Pred MajOp, + bit isPredNot, bit isPredNew> + : LDInst <(outs RC:$dst), (ins PredRegs:$src1, u6Ext:$absaddr), + !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ", + ") ")#"$dst = "#mnemonic#"(#$absaddr)">, AddrModeRel { + bits<5> dst; + bits<2> src1; + bits<6> absaddr; + + let isPredicatedNew = isPredNew; + let isPredicatedFalse = isPredNot; + let hasNewValue = !if (!eq(!cast(RC), "DoubleRegs"), 0, 1); + + let IClass = 0b1001; + + let Inst{27-24} = 0b1111; + let Inst{23-21} = MajOp; + let Inst{20-16} = absaddr{5-1}; + let Inst{13} = 0b1; + let Inst{12} = isPredNew; + let Inst{11} = isPredNot; + let Inst{10-9} = src1; + let Inst{8} = absaddr{0}; + let Inst{7} = 0b1; + let Inst{4-0} = dst; } - let accessSize = WordAccess in - defm LDriw_abs : LD_Abs<"memw", "LDriw", IntRegs>, AddrModeRel; - let accessSize = DoubleWordAccess in - defm LDrid_abs : LD_Abs<"memd", "LDrid", DoubleRegs>, AddrModeRel; +//===----------------------------------------------------------------------===// +// Multiclass for the load instructions with absolute addressing mode. +//===----------------------------------------------------------------------===// +multiclass LD_Abs_Pred MajOp, + bit PredNot> { + def _abs : T_LoadAbs_Pred ; + // Predicate new + def new_abs : T_LoadAbs_Pred ; } -let Predicates = [HasV4T], AddedComplexity = 30 in { -def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))), - (LDriw_abs_V4 tglobaladdr: $absaddr)>; - -def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))), - (LDrib_abs_V4 tglobaladdr:$absaddr)>; +let addrMode = Absolute, isExtended = 1 in +multiclass LD_Abs MajOp> { + let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in { + let opExtendable = 1, isPredicable = 1 in + def L4_#NAME#_abs: T_LoadAbs ; -def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))), - (LDriub_abs_V4 tglobaladdr:$absaddr)>; + // Predicated + defm L4_p#NAME#t : LD_Abs_Pred; + defm L4_p#NAME#f : LD_Abs_Pred; + } +} -def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))), - (LDrih_abs_V4 tglobaladdr:$absaddr)>; +let accessSize = ByteAccess, hasNewValue = 1 in { + defm loadrb : LD_Abs<"memb", "LDrib", IntRegs, u16_0Imm, 0b000>; + defm loadrub : LD_Abs<"memub", "LDriub", IntRegs, u16_0Imm, 0b001>; +} -def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))), - (LDriuh_abs_V4 tglobaladdr:$absaddr)>; +let accessSize = HalfWordAccess, hasNewValue = 1 in { + defm loadrh : LD_Abs<"memh", "LDrih", IntRegs, u16_1Imm, 0b010>; + defm loadruh : LD_Abs<"memuh", "LDriuh", IntRegs, u16_1Imm, 0b011>; } +let accessSize = WordAccess, hasNewValue = 1 in +defm loadri : LD_Abs<"memw", "LDriw", IntRegs, u16_2Imm, 0b100>; + +let accessSize = DoubleWordAccess in +defm loadrd : LD_Abs<"memd", "LDrid", DoubleRegs, u16_3Imm, 0b110>; + //===----------------------------------------------------------------------===// // multiclass for load instructions with GP-relative addressing mode. // Rx=mem[bhwd](##global) +// Once predicated, these instructions map to absolute addressing mode. // if ([!]Pv[.new]) Rx=mem[bhwd](##global) //===----------------------------------------------------------------------===// -let hasSideEffects = 0, validSubTargets = HasV4SubT in -multiclass LD_GP { - let BaseOpcode = BaseOp in { - let isPredicable = 1 in - def NAME#_V4 : LDInst2<(outs RC:$dst), - (ins globaladdress:$global), - "$dst = "#mnemonic#"(#$global)", - []>; - - let isExtended = 1, opExtendable = 2, isPredicated = 1 in { - defm Pt_V4 : LD_Abs_Pred; - defm NotPt_V4 : LD_Abs_Pred; - } - } -} -defm LDd_GP : LD_GP<"memd", "LDd_GP", DoubleRegs>, PredNewRel; -defm LDb_GP : LD_GP<"memb", "LDb_GP", IntRegs>, PredNewRel; -defm LDub_GP : LD_GP<"memub", "LDub_GP", IntRegs>, PredNewRel; -defm LDh_GP : LD_GP<"memh", "LDh_GP", IntRegs>, PredNewRel; -defm LDuh_GP : LD_GP<"memuh", "LDuh_GP", IntRegs>, PredNewRel; -defm LDw_GP : LD_GP<"memw", "LDw_GP", IntRegs>, PredNewRel; +let isAsmParserOnly = 1 in +class T_LoadGP MajOp> + : T_LoadAbsGP , PredNewRel { + let BaseOpcode = BaseOp#_abs; + } -def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)), - (i64 (LDd_GP_V4 tglobaladdr:$global))>; +let accessSize = ByteAccess, hasNewValue = 1 in { + def L2_loadrbgp : T_LoadGP<"memb", "LDrib", IntRegs, u16_0Imm, 0b000>; + def L2_loadrubgp : T_LoadGP<"memub", "LDriub", IntRegs, u16_0Imm, 0b001>; +} -def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)), - (i32 (LDw_GP_V4 tglobaladdr:$global))>; +let accessSize = HalfWordAccess, hasNewValue = 1 in { + def L2_loadrhgp : T_LoadGP<"memh", "LDrih", IntRegs, u16_1Imm, 0b010>; + def L2_loadruhgp : T_LoadGP<"memuh", "LDriuh", IntRegs, u16_1Imm, 0b011>; +} -def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)), - (i32 (LDuh_GP_V4 tglobaladdr:$global))>; +let accessSize = WordAccess, hasNewValue = 1 in +def L2_loadrigp : T_LoadGP<"memw", "LDriw", IntRegs, u16_2Imm, 0b100>; -def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)), - (i32 (LDub_GP_V4 tglobaladdr:$global))>; +let accessSize = DoubleWordAccess in +def L2_loadrdgp : T_LoadGP<"memd", "LDrid", DoubleRegs, u16_3Imm, 0b110>; -// Map from load(globaladdress) -> memw(#foo + 0) -let AddedComplexity = 100 in -def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))), - (i64 (LDd_GP_V4 tglobaladdr:$global))>; +def: Loada_pat; +def: Loada_pat; +def: Loada_pat; +def: Loada_pat; // Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd -let AddedComplexity = 100 in -def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))), - (i1 (C2_tfrrp (i32 (LDb_GP_V4 tglobaladdr:$global))))>; +def: Loadam_pat; +def: Loadam_pat; + +def: Stoream_pat; +def: Stoream_pat; + +// Map from load(globaladdress) -> mem[u][bhwd](#foo) +class LoadGP_pats + : Pat <(VT (ldOp (HexagonCONST32_GP tglobaladdr:$global))), + (VT (MI tglobaladdr:$global))>; + +let AddedComplexity = 100 in { + def: LoadGP_pats ; + def: LoadGP_pats ; + def: LoadGP_pats ; + def: LoadGP_pats ; + def: LoadGP_pats ; + def: LoadGP_pats ; + def: LoadGP_pats ; + def: LoadGP_pats ; +} // When the Interprocedural Global Variable optimizer realizes that a certain // global variable takes only two constant values, it shrinks the global to // a boolean. Catch those loads here in the following 3 patterns. -let AddedComplexity = 100 in -def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDb_GP_V4 tglobaladdr:$global))>; +let AddedComplexity = 100 in { + def: LoadGP_pats ; + def: LoadGP_pats ; +} -let AddedComplexity = 100 in -def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDb_GP_V4 tglobaladdr:$global))>; +// Transfer global address into a register +def: Pat<(HexagonCONST32 tglobaladdr:$Rs), (A2_tfrsi s16Ext:$Rs)>; +def: Pat<(HexagonCONST32_GP tblockaddress:$Rs), (A2_tfrsi s16Ext:$Rs)>; +def: Pat<(HexagonCONST32_GP tglobaladdr:$Rs), (A2_tfrsi s16Ext:$Rs)>; + +let AddedComplexity = 30 in { + def: Storea_pat; + def: Storea_pat; + def: Storea_pat; +} -// Map from load(globaladdress) -> memb(#foo) -let AddedComplexity = 100 in -def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDb_GP_V4 tglobaladdr:$global))>; +let AddedComplexity = 30 in { + def: Loada_pat; + def: Loada_pat; + def: Loada_pat; + def: Loada_pat; + def: Loada_pat; +} -// Map from load(globaladdress) -> memb(#foo) +// Indexed store word - global address. +// memw(Rs+#u6:2)=#S8 let AddedComplexity = 100 in -def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDb_GP_V4 tglobaladdr:$global))>; +def: Storex_add_pat; -let AddedComplexity = 100 in -def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDub_GP_V4 tglobaladdr:$global))>; +// Load from a global address that has only one use in the current basic block. +let AddedComplexity = 100 in { + def: Loada_pat; + def: Loada_pat; + def: Loada_pat; -// Map from load(globaladdress) -> memub(#foo) -let AddedComplexity = 100 in -def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDub_GP_V4 tglobaladdr:$global))>; + def: Loada_pat; + def: Loada_pat; + def: Loada_pat; -// Map from load(globaladdress) -> memh(#foo) -let AddedComplexity = 100 in -def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDh_GP_V4 tglobaladdr:$global))>; + def: Loada_pat; + def: Loada_pat; +} -// Map from load(globaladdress) -> memh(#foo) -let AddedComplexity = 100 in -def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDh_GP_V4 tglobaladdr:$global))>; +// Store to a global address that has only one use in the current basic block. +let AddedComplexity = 100 in { + def: Storea_pat; + def: Storea_pat; + def: Storea_pat; + def: Storea_pat; -// Map from load(globaladdress) -> memuh(#foo) -let AddedComplexity = 100 in -def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDuh_GP_V4 tglobaladdr:$global))>; + def: Stoream_pat; +} -// Map from load(globaladdress) -> memw(#foo) +// Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd let AddedComplexity = 100 in -def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))), - (i32 (LDw_GP_V4 tglobaladdr:$global))>; - +def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))), + (i1 (C2_tfrrp (i32 (L2_loadrbgp tglobaladdr:$global))))>; // Transfer global address into a register let isExtended = 1, opExtendable = 1, AddedComplexity=50, isMoveImm = 1, -isAsCheapAsAMove = 1, isReMaterializable = 1, validSubTargets = HasV4SubT in +isAsCheapAsAMove = 1, isReMaterializable = 1, isCodeGenOnly = 1 in def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins s16Ext:$src1), "$dst = #$src1", - [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>, - Requires<[HasV4T]>; + [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>; // Transfer a block address into a register def : Pat<(HexagonCONST32_GP tblockaddress:$src1), - (TFRI_V4 tblockaddress:$src1)>, - Requires<[HasV4T]>; - -let isExtended = 1, opExtendable = 2, AddedComplexity=50, -hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in -def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst), - (ins PredRegs:$src1, s16Ext:$src2), - "if($src1) $dst = #$src2", - []>, - Requires<[HasV4T]>; - -let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1, -hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in -def TFRI_cNotPt_V4 : ALU32_ri<(outs IntRegs:$dst), - (ins PredRegs:$src1, s16Ext:$src2), - "if(!$src1) $dst = #$src2", - []>, - Requires<[HasV4T]>; - -let isExtended = 1, opExtendable = 2, AddedComplexity=50, -hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in -def TFRI_cdnPt_V4 : ALU32_ri<(outs IntRegs:$dst), - (ins PredRegs:$src1, s16Ext:$src2), - "if($src1.new) $dst = #$src2", - []>, - Requires<[HasV4T]>; - -let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1, -hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in -def TFRI_cdnNotPt_V4 : ALU32_ri<(outs IntRegs:$dst), - (ins PredRegs:$src1, s16Ext:$src2), - "if(!$src1.new) $dst = #$src2", - []>, - Requires<[HasV4T]>; - -let AddedComplexity = 50, Predicates = [HasV4T] in -def : Pat<(HexagonCONST32_GP tglobaladdr:$src1), - (TFRI_V4 tglobaladdr:$src1)>, - Requires<[HasV4T]>; - - -// Load - Indirect with long offset: These instructions take global address -// as an operand -let isExtended = 1, opExtendable = 3, AddedComplexity = 40, -validSubTargets = HasV4SubT in -def LDrid_ind_lo_V4 : LDInst<(outs DoubleRegs:$dst), - (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset), - "$dst=memd($src1<<#$src2+##$offset)", - [(set (i64 DoubleRegs:$dst), - (load (add (shl IntRegs:$src1, u2ImmPred:$src2), - (HexagonCONST32 tglobaladdr:$offset))))]>, - Requires<[HasV4T]>; - -let AddedComplexity = 40 in -multiclass LD_indirect_lo { -let isExtended = 1, opExtendable = 3, validSubTargets = HasV4SubT in - def _lo_V4 : LDInst<(outs IntRegs:$dst), - (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset), - !strconcat("$dst = ", - !strconcat(OpcStr, "($src1<<#$src2+##$offset)")), - [(set IntRegs:$dst, - (i32 (OpNode (add (shl IntRegs:$src1, u2ImmPred:$src2), - (HexagonCONST32 tglobaladdr:$offset)))))]>, - Requires<[HasV4T]>; -} - -defm LDrib_ind : LD_indirect_lo<"memb", sextloadi8>; -defm LDriub_ind : LD_indirect_lo<"memub", zextloadi8>; -defm LDriub_ind_anyext : LD_indirect_lo<"memub", extloadi8>; -defm LDrih_ind : LD_indirect_lo<"memh", sextloadi16>; -defm LDriuh_ind : LD_indirect_lo<"memuh", zextloadi16>; -defm LDriuh_ind_anyext : LD_indirect_lo<"memuh", extloadi16>; -defm LDriw_ind : LD_indirect_lo<"memw", load>; - -let AddedComplexity = 40 in -def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, - (NumUsesBelowThresCONST32 tglobaladdr:$offset)))), - (i32 (LDrib_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>, - Requires<[HasV4T]>; + (TFRI_V4 tblockaddress:$src1)>; -let AddedComplexity = 40 in -def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, - (NumUsesBelowThresCONST32 tglobaladdr:$offset)))), - (i32 (LDriub_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>, - Requires<[HasV4T]>; +let AddedComplexity = 50 in +def : Pat<(HexagonCONST32_GP tglobaladdr:$src1), + (TFRI_V4 tglobaladdr:$src1)>; -let Predicates = [HasV4T], AddedComplexity = 30 in { -def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2), - (STrib_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>; +// i8/i16/i32 -> i64 loads +// We need a complexity of 120 here to override preceding handling of +// zextload. +let AddedComplexity = 120 in { + def: Loadam_pat; + def: Loadam_pat; + def: Loadam_pat; -def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2), - (STrih_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>; + def: Loadam_pat; + def: Loadam_pat; + def: Loadam_pat; -def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2), - (STriw_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>; + def: Loadam_pat; + def: Loadam_pat; + def: Loadam_pat; } -let Predicates = [HasV4T], AddedComplexity = 30 in { -def : Pat<(i32 (load u0AlwaysExtPred:$src)), - (LDriw_abs_V4 u0AlwaysExtPred:$src)>; +let AddedComplexity = 100 in { + def: Loada_pat; + def: Loada_pat; + def: Loada_pat; -def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)), - (LDrib_abs_V4 u0AlwaysExtPred:$src)>; + def: Loada_pat; + def: Loada_pat; + def: Loada_pat; -def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)), - (LDriub_abs_V4 u0AlwaysExtPred:$src)>; - -def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)), - (LDrih_abs_V4 u0AlwaysExtPred:$src)>; - -def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)), - (LDriuh_abs_V4 u0AlwaysExtPred:$src)>; + def: Loada_pat; + def: Loada_pat; } -// Indexed store word - global address. -// memw(Rs+#u6:2)=#S8 -let AddedComplexity = 10 in -def STriw_offset_ext_V4 : STInst<(outs), - (ins IntRegs:$src1, u6_2Imm:$src2, globaladdress:$src3), - "memw($src1+#$src2) = ##$src3", - [(store (HexagonCONST32 tglobaladdr:$src3), - (add IntRegs:$src1, u6_2ImmPred:$src2))]>, - Requires<[HasV4T]>; - -def : Pat<(i64 (ctlz (i64 DoubleRegs:$src1))), - (i64 (COMBINE_Ir_V4 (i32 0), (i32 (CTLZ64_rr DoubleRegs:$src1))))>, - Requires<[HasV4T]>; +let AddedComplexity = 100 in { + def: Storea_pat; + def: Storea_pat; + def: Storea_pat; + def: Storea_pat; +} -def : Pat<(i64 (cttz (i64 DoubleRegs:$src1))), - (i64 (COMBINE_Ir_V4 (i32 0), (i32 (CTTZ64_rr DoubleRegs:$src1))))>, - Requires<[HasV4T]>; +def: Loada_pat; +def: Loada_pat; +def: Loada_pat; +def: Loada_pat; + +def: Storea_pat, I32, addrgp, S2_storerbabs>; +def: Storea_pat, I32, addrgp, S2_storerhabs>; +def: Storea_pat, I32, addrgp, S2_storeriabs>; +def: Storea_pat, I64, addrgp, S2_storerdabs>; + +let Constraints = "@earlyclobber $dst" in +def Insert4 : PseudoM<(outs DoubleRegs:$dst), (ins IntRegs:$a, IntRegs:$b, + IntRegs:$c, IntRegs:$d), + ".error \"Should never try to emit Insert4\"", + [(set (i64 DoubleRegs:$dst), + (or (or (or (shl (i64 (zext (i32 (and (i32 IntRegs:$b), (i32 65535))))), + (i32 16)), + (i64 (zext (i32 (and (i32 IntRegs:$a), (i32 65535)))))), + (shl (i64 (anyext (i32 (and (i32 IntRegs:$c), (i32 65535))))), + (i32 32))), + (shl (i64 (anyext (i32 IntRegs:$d))), (i32 48))))]>; +//===----------------------------------------------------------------------===// +// :raw for of boundscheck:hi:lo insns +//===----------------------------------------------------------------------===// -// i8 -> i64 loads -// We need a complexity of 120 here to override preceding handling of -// zextloadi8. -let Predicates = [HasV4T], AddedComplexity = 120 in { -def: Pat <(i64 (extloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (COMBINE_Ir_V4 0, (LDrib_abs_V4 tglobaladdr:$addr)))>; +// A4_boundscheck_lo: Detect if a register is within bounds. +let hasSideEffects = 0 in +def A4_boundscheck_lo: ALU64Inst < + (outs PredRegs:$Pd), + (ins DoubleRegs:$Rss, DoubleRegs:$Rtt), + "$Pd = boundscheck($Rss, $Rtt):raw:lo"> { + bits<2> Pd; + bits<5> Rss; + bits<5> Rtt; + + let IClass = 0b1101; + + let Inst{27-23} = 0b00100; + let Inst{13} = 0b1; + let Inst{7-5} = 0b100; + let Inst{1-0} = Pd; + let Inst{20-16} = Rss; + let Inst{12-8} = Rtt; + } -def: Pat <(i64 (zextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (COMBINE_Ir_V4 0, (LDriub_abs_V4 tglobaladdr:$addr)))>; +// A4_boundscheck_hi: Detect if a register is within bounds. +let hasSideEffects = 0 in +def A4_boundscheck_hi: ALU64Inst < + (outs PredRegs:$Pd), + (ins DoubleRegs:$Rss, DoubleRegs:$Rtt), + "$Pd = boundscheck($Rss, $Rtt):raw:hi"> { + bits<2> Pd; + bits<5> Rss; + bits<5> Rtt; + + let IClass = 0b1101; + + let Inst{27-23} = 0b00100; + let Inst{13} = 0b1; + let Inst{7-5} = 0b101; + let Inst{1-0} = Pd; + let Inst{20-16} = Rss; + let Inst{12-8} = Rtt; + } -def: Pat <(i64 (sextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (A2_sxtw (LDrib_abs_V4 tglobaladdr:$addr)))>; +let hasSideEffects = 0, isAsmParserOnly = 1 in +def A4_boundscheck : MInst < + (outs PredRegs:$Pd), (ins IntRegs:$Rs, DoubleRegs:$Rtt), + "$Pd=boundscheck($Rs,$Rtt)">; + +// A4_tlbmatch: Detect if a VA/ASID matches a TLB entry. +let isPredicateLate = 1, hasSideEffects = 0 in +def A4_tlbmatch : ALU64Inst<(outs PredRegs:$Pd), + (ins DoubleRegs:$Rs, IntRegs:$Rt), + "$Pd = tlbmatch($Rs, $Rt)", + [], "", ALU64_tc_2early_SLOT23> { + bits<2> Pd; + bits<5> Rs; + bits<5> Rt; -def: Pat <(i64 (extloadi8 FoldGlobalAddr:$addr)), - (i64 (COMBINE_Ir_V4 0, (LDrib_abs_V4 FoldGlobalAddr:$addr)))>; + let IClass = 0b1101; + let Inst{27-23} = 0b00100; + let Inst{20-16} = Rs; + let Inst{13} = 0b1; + let Inst{12-8} = Rt; + let Inst{7-5} = 0b011; + let Inst{1-0} = Pd; + } -def: Pat <(i64 (zextloadi8 FoldGlobalAddr:$addr)), - (i64 (COMBINE_Ir_V4 0, (LDriub_abs_V4 FoldGlobalAddr:$addr)))>; +// We need custom lowering of ISD::PREFETCH into HexagonISD::DCFETCH +// because the SDNode ISD::PREFETCH has properties MayLoad and MayStore. +// We don't really want either one here. +def SDTHexagonDCFETCH : SDTypeProfile<0, 2, [SDTCisPtrTy<0>,SDTCisInt<1>]>; +def HexagonDCFETCH : SDNode<"HexagonISD::DCFETCH", SDTHexagonDCFETCH, + [SDNPHasChain]>; + +// Use LD0Inst for dcfetch, but set "mayLoad" to 0 because this doesn't +// really do a load. +let hasSideEffects = 1, mayLoad = 0 in +def Y2_dcfetchbo : LD0Inst<(outs), (ins IntRegs:$Rs, u11_3Imm:$u11_3), + "dcfetch($Rs + #$u11_3)", + [(HexagonDCFETCH IntRegs:$Rs, u11_3ImmPred:$u11_3)], + "", LD_tc_ld_SLOT0> { + bits<5> Rs; + bits<14> u11_3; -def: Pat <(i64 (sextloadi8 FoldGlobalAddr:$addr)), - (i64 (A2_sxtw (LDrib_abs_V4 FoldGlobalAddr:$addr)))>; + let IClass = 0b1001; + let Inst{27-21} = 0b0100000; + let Inst{20-16} = Rs; + let Inst{13} = 0b0; + let Inst{10-0} = u11_3{13-3}; } -// i16 -> i64 loads -// We need a complexity of 120 here to override preceding handling of -// zextloadi16. -let AddedComplexity = 120 in { -def: Pat <(i64 (extloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (COMBINE_Ir_V4 0, (LDrih_abs_V4 tglobaladdr:$addr)))>, - Requires<[HasV4T]>; -def: Pat <(i64 (zextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (COMBINE_Ir_V4 0, (LDriuh_abs_V4 tglobaladdr:$addr)))>, - Requires<[HasV4T]>; - -def: Pat <(i64 (sextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (A2_sxtw (LDrih_abs_V4 tglobaladdr:$addr)))>, - Requires<[HasV4T]>; - -def: Pat <(i64 (extloadi16 FoldGlobalAddr:$addr)), - (i64 (COMBINE_Ir_V4 0, (LDrih_abs_V4 FoldGlobalAddr:$addr)))>, - Requires<[HasV4T]>; - -def: Pat <(i64 (zextloadi16 FoldGlobalAddr:$addr)), - (i64 (COMBINE_Ir_V4 0, (LDriuh_abs_V4 FoldGlobalAddr:$addr)))>, - Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// Compound instructions +//===----------------------------------------------------------------------===// -def: Pat <(i64 (sextloadi16 FoldGlobalAddr:$addr)), - (i64 (A2_sxtw (LDrih_abs_V4 FoldGlobalAddr:$addr)))>, - Requires<[HasV4T]>; +let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1, + isPredicated = 1, isPredicatedNew = 1, isExtendable = 1, + opExtentBits = 11, opExtentAlign = 2, opExtendable = 1, + isTerminator = 1 in +class CJInst_tstbit_R0 + : InstHexagon<(outs), (ins IntRegs:$Rs, brtarget:$r9_2), + ""#px#" = tstbit($Rs, #0); if (" + #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2", + [], "", COMPOUND, TypeCOMPOUND>, OpcodeHexagon { + bits<4> Rs; + bits<11> r9_2; + + // np: !p[01] + let isPredicatedFalse = np; + // tnt: Taken/Not Taken + let isBrTaken = !if (!eq(tnt, "t"), "true", "false"); + let isTaken = !if (!eq(tnt, "t"), 1, 0); + + let IClass = 0b0001; + let Inst{27-26} = 0b00; + let Inst{25} = !if (!eq(px, "!p1"), 1, + !if (!eq(px, "p1"), 1, 0)); + let Inst{24-23} = 0b11; + let Inst{22} = np; + let Inst{21-20} = r9_2{10-9}; + let Inst{19-16} = Rs; + let Inst{13} = !if (!eq(tnt, "t"), 1, 0); + let Inst{9-8} = 0b11; + let Inst{7-1} = r9_2{8-2}; } -// i32->i64 loads -// We need a complexity of 120 here to override preceding handling of -// zextloadi32. -let AddedComplexity = 120 in { -def: Pat <(i64 (extloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (COMBINE_Ir_V4 0, (LDriw_abs_V4 tglobaladdr:$addr)))>, - Requires<[HasV4T]>; - -def: Pat <(i64 (zextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (COMBINE_Ir_V4 0, (LDriw_abs_V4 tglobaladdr:$addr)))>, - Requires<[HasV4T]>; -def: Pat <(i64 (sextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))), - (i64 (A2_sxtw (LDriw_abs_V4 tglobaladdr:$addr)))>, - Requires<[HasV4T]>; - -def: Pat <(i64 (extloadi32 FoldGlobalAddr:$addr)), - (i64 (COMBINE_Ir_V4 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>, - Requires<[HasV4T]>; - -def: Pat <(i64 (zextloadi32 FoldGlobalAddr:$addr)), - (i64 (COMBINE_Ir_V4 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>, - Requires<[HasV4T]>; - -def: Pat <(i64 (sextloadi32 FoldGlobalAddr:$addr)), - (i64 (A2_sxtw (LDriw_abs_V4 FoldGlobalAddr:$addr)))>, - Requires<[HasV4T]>; +let Defs = [PC, P0], Uses = [P0] in { + def J4_tstbit0_tp0_jump_nt : CJInst_tstbit_R0<"p0", 0, "nt">; + def J4_tstbit0_tp0_jump_t : CJInst_tstbit_R0<"p0", 0, "t">; + def J4_tstbit0_fp0_jump_nt : CJInst_tstbit_R0<"p0", 1, "nt">; + def J4_tstbit0_fp0_jump_t : CJInst_tstbit_R0<"p0", 1, "t">; } -// Indexed store double word - global address. -// memw(Rs+#u6:2)=#S8 -let AddedComplexity = 10 in -def STrih_offset_ext_V4 : STInst<(outs), - (ins IntRegs:$src1, u6_1Imm:$src2, globaladdress:$src3), - "memh($src1+#$src2) = ##$src3", - [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3), - (add IntRegs:$src1, u6_1ImmPred:$src2))]>, - Requires<[HasV4T]>; -// Map from store(globaladdress + x) -> memd(#foo + x) -let AddedComplexity = 100 in -def : Pat<(store (i64 DoubleRegs:$src1), - FoldGlobalAddrGP:$addr), - (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>, - Requires<[HasV4T]>; - -def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr, - (i64 DoubleRegs:$src1)), - (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>, - Requires<[HasV4T]>; - -// Map from store(globaladdress + x) -> memb(#foo + x) -let AddedComplexity = 100 in -def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr), - (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>, - Requires<[HasV4T]>; - -def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)), - (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>, - Requires<[HasV4T]>; - -// Map from store(globaladdress + x) -> memh(#foo + x) -let AddedComplexity = 100 in -def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr), - (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>, - Requires<[HasV4T]>; - -def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)), - (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>, - Requires<[HasV4T]>; - -// Map from store(globaladdress + x) -> memw(#foo + x) -let AddedComplexity = 100 in -def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr), - (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>, - Requires<[HasV4T]>; - -def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)), - (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>, - Requires<[HasV4T]>; - -// Map from load(globaladdress + x) -> memd(#foo + x) -let AddedComplexity = 100 in -def : Pat<(i64 (load FoldGlobalAddrGP:$addr)), - (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; - -def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr), - (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; - -// Map from load(globaladdress + x) -> memb(#foo + x) -let AddedComplexity = 100 in -def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)), - (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; - -// Map from load(globaladdress + x) -> memb(#foo + x) -let AddedComplexity = 100 in -def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)), - (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; - -//let AddedComplexity = 100 in -let AddedComplexity = 100 in -def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)), - (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; - -// Map from load(globaladdress + x) -> memh(#foo + x) -let AddedComplexity = 100 in -def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)), - (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; +let Defs = [PC, P1], Uses = [P1] in { + def J4_tstbit0_tp1_jump_nt : CJInst_tstbit_R0<"p1", 0, "nt">; + def J4_tstbit0_tp1_jump_t : CJInst_tstbit_R0<"p1", 0, "t">; + def J4_tstbit0_fp1_jump_nt : CJInst_tstbit_R0<"p1", 1, "nt">; + def J4_tstbit0_fp1_jump_t : CJInst_tstbit_R0<"p1", 1, "t">; +} -// Map from load(globaladdress + x) -> memuh(#foo + x) -let AddedComplexity = 100 in -def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)), - (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; -def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr), - (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; +let isBranch = 1, hasSideEffects = 0, + isExtentSigned = 1, isPredicated = 1, isPredicatedNew = 1, + isExtendable = 1, opExtentBits = 11, opExtentAlign = 2, + opExtendable = 2, isTerminator = 1 in +class CJInst_RR + : InstHexagon<(outs), (ins IntRegs:$Rs, IntRegs:$Rt, brtarget:$r9_2), + ""#px#" = cmp."#op#"($Rs, $Rt); if (" + #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2", + [], "", COMPOUND, TypeCOMPOUND>, OpcodeHexagon { + bits<4> Rs; + bits<4> Rt; + bits<11> r9_2; + + // np: !p[01] + let isPredicatedFalse = np; + // tnt: Taken/Not Taken + let isBrTaken = !if (!eq(tnt, "t"), "true", "false"); + let isTaken = !if (!eq(tnt, "t"), 1, 0); + + let IClass = 0b0001; + let Inst{27-23} = !if (!eq(op, "eq"), 0b01000, + !if (!eq(op, "gt"), 0b01001, + !if (!eq(op, "gtu"), 0b01010, 0))); + let Inst{22} = np; + let Inst{21-20} = r9_2{10-9}; + let Inst{19-16} = Rs; + let Inst{13} = !if (!eq(tnt, "t"), 1, 0); + // px: Predicate reg 0/1 + let Inst{12} = !if (!eq(px, "!p1"), 1, + !if (!eq(px, "p1"), 1, 0)); + let Inst{11-8} = Rt; + let Inst{7-1} = r9_2{8-2}; +} -// Map from load(globaladdress + x) -> memub(#foo + x) -let AddedComplexity = 100 in -def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)), - (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; +// P[10] taken/not taken. +multiclass T_tnt_CJInst_RR { + let Defs = [PC, P0], Uses = [P0] in { + def NAME#p0_jump_nt : CJInst_RR<"p0", op, np, "nt">; + def NAME#p0_jump_t : CJInst_RR<"p0", op, np, "t">; + } + let Defs = [PC, P1], Uses = [P1] in { + def NAME#p1_jump_nt : CJInst_RR<"p1", op, np, "nt">; + def NAME#p1_jump_t : CJInst_RR<"p1", op, np, "t">; + } +} +// Predicate / !Predicate +multiclass T_pnp_CJInst_RR{ + defm J4_cmp#NAME#_t : T_tnt_CJInst_RR; + defm J4_cmp#NAME#_f : T_tnt_CJInst_RR; +} +// TypeCJ Instructions compare RR and jump +defm eq : T_pnp_CJInst_RR<"eq">; +defm gt : T_pnp_CJInst_RR<"gt">; +defm gtu : T_pnp_CJInst_RR<"gtu">; + +let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1, + isPredicated = 1, isPredicatedNew = 1, isExtendable = 1, opExtentBits = 11, + opExtentAlign = 2, opExtendable = 2, isTerminator = 1 in +class CJInst_RU5 + : InstHexagon<(outs), (ins IntRegs:$Rs, u5Imm:$U5, brtarget:$r9_2), + ""#px#" = cmp."#op#"($Rs, #$U5); if (" + #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2", + [], "", COMPOUND, TypeCOMPOUND>, OpcodeHexagon { + bits<4> Rs; + bits<5> U5; + bits<11> r9_2; + + // np: !p[01] + let isPredicatedFalse = np; + // tnt: Taken/Not Taken + let isBrTaken = !if (!eq(tnt, "t"), "true", "false"); + let isTaken = !if (!eq(tnt, "t"), 1, 0); + + let IClass = 0b0001; + let Inst{27-26} = 0b00; + // px: Predicate reg 0/1 + let Inst{25} = !if (!eq(px, "!p1"), 1, + !if (!eq(px, "p1"), 1, 0)); + let Inst{24-23} = !if (!eq(op, "eq"), 0b00, + !if (!eq(op, "gt"), 0b01, + !if (!eq(op, "gtu"), 0b10, 0))); + let Inst{22} = np; + let Inst{21-20} = r9_2{10-9}; + let Inst{19-16} = Rs; + let Inst{13} = !if (!eq(tnt, "t"), 1, 0); + let Inst{12-8} = U5; + let Inst{7-1} = r9_2{8-2}; +} +// P[10] taken/not taken. +multiclass T_tnt_CJInst_RU5 { + let Defs = [PC, P0], Uses = [P0] in { + def NAME#p0_jump_nt : CJInst_RU5<"p0", op, np, "nt">; + def NAME#p0_jump_t : CJInst_RU5<"p0", op, np, "t">; + } + let Defs = [PC, P1], Uses = [P1] in { + def NAME#p1_jump_nt : CJInst_RU5<"p1", op, np, "nt">; + def NAME#p1_jump_t : CJInst_RU5<"p1", op, np, "t">; + } +} +// Predicate / !Predicate +multiclass T_pnp_CJInst_RU5{ + defm J4_cmp#NAME#i_t : T_tnt_CJInst_RU5; + defm J4_cmp#NAME#i_f : T_tnt_CJInst_RU5; +} +// TypeCJ Instructions compare RI and jump +defm eq : T_pnp_CJInst_RU5<"eq">; +defm gt : T_pnp_CJInst_RU5<"gt">; +defm gtu : T_pnp_CJInst_RU5<"gtu">; + +let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1, + isPredicated = 1, isPredicatedFalse = 1, isPredicatedNew = 1, + isExtendable = 1, opExtentBits = 11, opExtentAlign = 2, opExtendable = 1, + isTerminator = 1 in +class CJInst_Rn1 + : InstHexagon<(outs), (ins IntRegs:$Rs, brtarget:$r9_2), + ""#px#" = cmp."#op#"($Rs,#-1); if (" + #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2", + [], "", COMPOUND, TypeCOMPOUND>, OpcodeHexagon { + bits<4> Rs; + bits<11> r9_2; + + // np: !p[01] + let isPredicatedFalse = np; + // tnt: Taken/Not Taken + let isBrTaken = !if (!eq(tnt, "t"), "true", "false"); + let isTaken = !if (!eq(tnt, "t"), 1, 0); + + let IClass = 0b0001; + let Inst{27-26} = 0b00; + let Inst{25} = !if (!eq(px, "!p1"), 1, + !if (!eq(px, "p1"), 1, 0)); + + let Inst{24-23} = 0b11; + let Inst{22} = np; + let Inst{21-20} = r9_2{10-9}; + let Inst{19-16} = Rs; + let Inst{13} = !if (!eq(tnt, "t"), 1, 0); + let Inst{9-8} = !if (!eq(op, "eq"), 0b00, + !if (!eq(op, "gt"), 0b01, 0)); + let Inst{7-1} = r9_2{8-2}; +} -def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr), - (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; +// P[10] taken/not taken. +multiclass T_tnt_CJInst_Rn1 { + let Defs = [PC, P0], Uses = [P0] in { + def NAME#p0_jump_nt : CJInst_Rn1<"p0", op, np, "nt">; + def NAME#p0_jump_t : CJInst_Rn1<"p0", op, np, "t">; + } + let Defs = [PC, P1], Uses = [P1] in { + def NAME#p1_jump_nt : CJInst_Rn1<"p1", op, np, "nt">; + def NAME#p1_jump_t : CJInst_Rn1<"p1", op, np, "t">; + } +} +// Predicate / !Predicate +multiclass T_pnp_CJInst_Rn1{ + defm J4_cmp#NAME#n1_t : T_tnt_CJInst_Rn1; + defm J4_cmp#NAME#n1_f : T_tnt_CJInst_Rn1; +} +// TypeCJ Instructions compare -1 and jump +defm eq : T_pnp_CJInst_Rn1<"eq">; +defm gt : T_pnp_CJInst_Rn1<"gt">; + +// J4_jumpseti: Direct unconditional jump and set register to immediate. +let Defs = [PC], isBranch = 1, hasSideEffects = 0, hasNewValue = 1, + isExtentSigned = 1, opNewValue = 0, isExtendable = 1, opExtentBits = 11, + opExtentAlign = 2, opExtendable = 2 in +def J4_jumpseti: CJInst < + (outs IntRegs:$Rd), + (ins u6Imm:$U6, brtarget:$r9_2), + "$Rd = #$U6 ; jump $r9_2"> { + bits<4> Rd; + bits<6> U6; + bits<11> r9_2; + + let IClass = 0b0001; + let Inst{27-24} = 0b0110; + let Inst{21-20} = r9_2{10-9}; + let Inst{19-16} = Rd; + let Inst{13-8} = U6; + let Inst{7-1} = r9_2{8-2}; + } -// Map from load(globaladdress + x) -> memw(#foo + x) -let AddedComplexity = 100 in -def : Pat<(i32 (load FoldGlobalAddrGP:$addr)), - (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; +// J4_jumpsetr: Direct unconditional jump and transfer register. +let Defs = [PC], isBranch = 1, hasSideEffects = 0, hasNewValue = 1, + isExtentSigned = 1, opNewValue = 0, isExtendable = 1, opExtentBits = 11, + opExtentAlign = 2, opExtendable = 2 in +def J4_jumpsetr: CJInst < + (outs IntRegs:$Rd), + (ins IntRegs:$Rs, brtarget:$r9_2), + "$Rd = $Rs ; jump $r9_2"> { + bits<4> Rd; + bits<4> Rs; + bits<11> r9_2; + + let IClass = 0b0001; + let Inst{27-24} = 0b0111; + let Inst{21-20} = r9_2{10-9}; + let Inst{11-8} = Rd; + let Inst{19-16} = Rs; + let Inst{7-1} = r9_2{8-2}; + } -def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr), - (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>, - Requires<[HasV4T]>; +// Duplex instructions +//===----------------------------------------------------------------------===// +include "HexagonIsetDx.td"