mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-10-19 02:52:53 +02:00
[ARM] Add MVE vector load/store instructions.
This adds the rest of the vector memory access instructions. It includes contiguous loads/stores, with an ordinary addressing mode such as [r0,#offset] (plus writeback variants); gather loads and scatter stores with a scalar base address register and a vector of offsets from it (written [r0,q1] or similar); and gather/scatters with a vector of base addresses (written [q0,#offset], again with writeback). Additionally, some of the loads can widen each loaded value into a larger vector lane, and the corresponding stores narrow them again. To implement these, we also have to add the addressing modes they need. Also, in AsmParser, the `isMem` query function now has subqueries `isGPRMem` and `isMVEMem`, according to which kind of base register is used by a given memory access operand. I've also had to add an extra check in `checkTargetMatchPredicate` in the AsmParser, without which our last-minute check of `rGPR` register operands against SP and PC was failing an assertion because Tablegen had inserted an immediate 0 in place of one of a pair of tied register operands. (This matches the way the corresponding check for `MCK_rGPR` in `validateTargetOperandClass` is guarded.) Apparently the MVE load instructions were the first to have ever triggered this assertion, but I think only because they were the first to have a combination of the usual Arm pre/post writeback system and the `rGPR` class in particular. Reviewers: dmgreen, samparker, SjoerdMeijer, t.p.northover Subscribers: javed.absar, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D62680 llvm-svn: 364291
This commit is contained in:
Simon Tatham
parent
263a203b09
commit
84fe6af65f
@ -2477,11 +2477,14 @@ bool llvm::rewriteARMFrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
|
||||
NumBits = 8;
|
||||
Scale = 2;
|
||||
break;
|
||||
case ARMII::AddrModeT2_i7:
|
||||
case ARMII::AddrModeT2_i7s2:
|
||||
case ARMII::AddrModeT2_i7s4:
|
||||
ImmIdx = FrameRegIdx+1;
|
||||
InstrOffs = MI.getOperand(ImmIdx).getImm();
|
||||
NumBits = 7;
|
||||
Scale = 4;
|
||||
Scale = (AddrMode == ARMII::AddrModeT2_i7s2 ? 2 :
|
||||
AddrMode == ARMII::AddrModeT2_i7s4 ? 4 : 1);
|
||||
break;
|
||||
default:
|
||||
llvm_unreachable("Unsupported addressing mode!");
|
||||
|
||||
@ -110,6 +110,8 @@ def AddrMode_i12 : AddrMode<16>;
|
||||
def AddrMode5FP16 : AddrMode<17>;
|
||||
def AddrModeT2_ldrex : AddrMode<18>;
|
||||
def AddrModeT2_i7s4 : AddrMode<19>;
|
||||
def AddrModeT2_i7s2 : AddrMode<20>;
|
||||
def AddrModeT2_i7 : AddrMode<21>;
|
||||
|
||||
// Load / store index mode.
|
||||
class IndexMode<bits<2> val> {
|
||||
|
||||
@ -1234,6 +1234,15 @@ def addr_offset_none : MemOperand,
|
||||
let MIOperandInfo = (ops GPR:$base);
|
||||
}
|
||||
|
||||
// t_addr_offset_none := reg [r0-r7]
|
||||
def MemNoOffsetTAsmOperand : AsmOperandClass { let Name = "MemNoOffsetT"; }
|
||||
def t_addr_offset_none : MemOperand {
|
||||
let PrintMethod = "printAddrMode7Operand";
|
||||
let DecoderMethod = "DecodetGPRRegisterClass";
|
||||
let ParserMatchClass = MemNoOffsetTAsmOperand;
|
||||
let MIOperandInfo = (ops tGPR:$base);
|
||||
}
|
||||
|
||||
def nohash_imm : Operand<i32> {
|
||||
let PrintMethod = "printNoHashImmediate";
|
||||
}
|
||||
|
||||
@ -153,6 +153,127 @@ def VecList4Q : RegisterOperand<QQQQPR, "printMVEVectorListFourQ"> {
|
||||
let PrintMethod = "printMVEVectorList<4>";
|
||||
}
|
||||
|
||||
// taddrmode_imm7 := reg[r0-r7] +/- (imm7 << shift)
|
||||
class TMemImm7ShiftOffsetAsmOperand<int shift> : AsmOperandClass {
|
||||
let Name = "TMemImm7Shift"#shift#"Offset";
|
||||
let PredicateMethod = "isMemImm7ShiftedOffset<"#shift#",ARM::tGPRRegClassID>";
|
||||
let RenderMethod = "addMemImmOffsetOperands";
|
||||
}
|
||||
|
||||
class taddrmode_imm7<int shift> : MemOperand {
|
||||
let ParserMatchClass = TMemImm7ShiftOffsetAsmOperand<shift>;
|
||||
// They are printed the same way as the T2 imm8 version
|
||||
let PrintMethod = "printT2AddrModeImm8Operand<false>";
|
||||
// This can also be the same as the T2 version.
|
||||
let EncoderMethod = "getT2AddrModeImmOpValue<7,"#shift#">";
|
||||
let DecoderMethod = "DecodeTAddrModeImm7<"#shift#">";
|
||||
let MIOperandInfo = (ops tGPR:$base, i32imm:$offsimm);
|
||||
}
|
||||
|
||||
// t2addrmode_imm7 := reg +/- (imm7)
|
||||
class MemImm7ShiftOffsetAsmOperand<int shift> : AsmOperandClass {
|
||||
let Name = "MemImm7Shift"#shift#"Offset";
|
||||
let PredicateMethod = "isMemImm7ShiftedOffset<" # shift #
|
||||
",ARM::GPRnopcRegClassID>";
|
||||
let RenderMethod = "addMemImmOffsetOperands";
|
||||
}
|
||||
|
||||
def MemImm7Shift0OffsetAsmOperand : MemImm7ShiftOffsetAsmOperand<0>;
|
||||
def MemImm7Shift1OffsetAsmOperand : MemImm7ShiftOffsetAsmOperand<1>;
|
||||
def MemImm7Shift2OffsetAsmOperand : MemImm7ShiftOffsetAsmOperand<2>;
|
||||
class T2AddrMode_Imm7<int shift> : MemOperand,
|
||||
ComplexPattern<i32, 2, "SelectT2AddrModeImm7<"#shift#">", []> {
|
||||
let EncoderMethod = "getT2AddrModeImmOpValue<7,"#shift#">";
|
||||
let DecoderMethod = "DecodeT2AddrModeImm7<"#shift#", 0>";
|
||||
let ParserMatchClass =
|
||||
!cast<AsmOperandClass>("MemImm7Shift"#shift#"OffsetAsmOperand");
|
||||
let MIOperandInfo = (ops GPRnopc:$base, i32imm:$offsimm);
|
||||
}
|
||||
|
||||
class t2addrmode_imm7<int shift> : T2AddrMode_Imm7<shift> {
|
||||
// They are printed the same way as the imm8 version
|
||||
let PrintMethod = "printT2AddrModeImm8Operand<false>";
|
||||
}
|
||||
|
||||
class MemImm7ShiftOffsetWBAsmOperand<int shift> : AsmOperandClass {
|
||||
let Name = "MemImm7Shift"#shift#"OffsetWB";
|
||||
let PredicateMethod = "isMemImm7ShiftedOffset<" # shift #
|
||||
",ARM::rGPRRegClassID>";
|
||||
let RenderMethod = "addMemImmOffsetOperands";
|
||||
}
|
||||
|
||||
def MemImm7Shift0OffsetWBAsmOperand : MemImm7ShiftOffsetWBAsmOperand<0>;
|
||||
def MemImm7Shift1OffsetWBAsmOperand : MemImm7ShiftOffsetWBAsmOperand<1>;
|
||||
def MemImm7Shift2OffsetWBAsmOperand : MemImm7ShiftOffsetWBAsmOperand<2>;
|
||||
|
||||
class t2addrmode_imm7_pre<int shift> : T2AddrMode_Imm7<shift> {
|
||||
// They are printed the same way as the imm8 version
|
||||
let PrintMethod = "printT2AddrModeImm8Operand<true>";
|
||||
let ParserMatchClass =
|
||||
!cast<AsmOperandClass>("MemImm7Shift"#shift#"OffsetWBAsmOperand");
|
||||
let DecoderMethod = "DecodeT2AddrModeImm7<"#shift#", 1>";
|
||||
let MIOperandInfo = (ops rGPR:$base, i32imm:$offsim);
|
||||
}
|
||||
|
||||
class t2am_imm7shiftOffsetAsmOperand<int shift>
|
||||
: AsmOperandClass { let Name = "Imm7Shift"#shift; }
|
||||
def t2am_imm7shift0OffsetAsmOperand : t2am_imm7shiftOffsetAsmOperand<0>;
|
||||
def t2am_imm7shift1OffsetAsmOperand : t2am_imm7shiftOffsetAsmOperand<1>;
|
||||
def t2am_imm7shift2OffsetAsmOperand : t2am_imm7shiftOffsetAsmOperand<2>;
|
||||
|
||||
class t2am_imm7_offset<int shift> : MemOperand {
|
||||
// They are printed the same way as the imm8 version
|
||||
let PrintMethod = "printT2AddrModeImm8OffsetOperand";
|
||||
let ParserMatchClass =
|
||||
!cast<AsmOperandClass>("t2am_imm7shift"#shift#"OffsetAsmOperand");
|
||||
let EncoderMethod = "getT2ScaledImmOpValue<7,"#shift#">";
|
||||
let DecoderMethod = "DecodeT2Imm7<"#shift#">";
|
||||
}
|
||||
|
||||
// Operands for gather/scatter loads of the form [Rbase, Qoffsets]
|
||||
class MemRegRQOffsetAsmOperand<int shift> : AsmOperandClass {
|
||||
let Name = "MemRegRQS"#shift#"Offset";
|
||||
let PredicateMethod = "isMemRegRQOffset<"#shift#">";
|
||||
let RenderMethod = "addMemRegRQOffsetOperands";
|
||||
}
|
||||
|
||||
def MemRegRQS0OffsetAsmOperand : MemRegRQOffsetAsmOperand<0>;
|
||||
def MemRegRQS1OffsetAsmOperand : MemRegRQOffsetAsmOperand<1>;
|
||||
def MemRegRQS2OffsetAsmOperand : MemRegRQOffsetAsmOperand<2>;
|
||||
def MemRegRQS3OffsetAsmOperand : MemRegRQOffsetAsmOperand<3>;
|
||||
|
||||
// mve_addr_rq_shift := reg + vreg{ << UXTW #shift}
|
||||
class mve_addr_rq_shift<int shift> : MemOperand {
|
||||
let EncoderMethod = "getMveAddrModeRQOpValue";
|
||||
let PrintMethod = "printMveAddrModeRQOperand<"#shift#">";
|
||||
let ParserMatchClass =
|
||||
!cast<AsmOperandClass>("MemRegRQS"#shift#"OffsetAsmOperand");
|
||||
let DecoderMethod = "DecodeMveAddrModeRQ";
|
||||
let MIOperandInfo = (ops GPRnopc:$base, MQPR:$offsreg);
|
||||
}
|
||||
|
||||
class MemRegQOffsetAsmOperand<int shift> : AsmOperandClass {
|
||||
let Name = "MemRegQS"#shift#"Offset";
|
||||
let PredicateMethod = "isMemRegQOffset<"#shift#">";
|
||||
let RenderMethod = "addMemImmOffsetOperands";
|
||||
}
|
||||
|
||||
def MemRegQS2OffsetAsmOperand : MemRegQOffsetAsmOperand<2>;
|
||||
def MemRegQS3OffsetAsmOperand : MemRegQOffsetAsmOperand<3>;
|
||||
|
||||
// mve_addr_q_shift := vreg {+ #imm7s2/4}
|
||||
class mve_addr_q_shift<int shift> : MemOperand {
|
||||
let EncoderMethod = "getMveAddrModeQOpValue<"#shift#">";
|
||||
// Can be printed same way as other reg + imm operands
|
||||
let PrintMethod = "printT2AddrModeImm8Operand<false>";
|
||||
let ParserMatchClass =
|
||||
!cast<AsmOperandClass>("MemRegQS"#shift#"OffsetAsmOperand");
|
||||
let DecoderMethod = "DecodeMveAddrModeQ<"#shift#">";
|
||||
let MIOperandInfo = (ops MQPR:$base, i32imm:$imm);
|
||||
}
|
||||
|
||||
// --------- Start of base classes for the instructions themselves
|
||||
|
||||
class MVE_MI<dag oops, dag iops, InstrItinClass itin, string asm,
|
||||
string ops, string cstr, list<dag> pattern>
|
||||
: Thumb2XI<oops, iops, AddrModeNone, 4, itin, !strconcat(asm, "\t", ops), cstr,
|
||||
@ -3245,6 +3366,359 @@ foreach wb = [MVE_vldst24_writeback<
|
||||
|
||||
// end of MVE interleaving load/store
|
||||
|
||||
// start of MVE predicable load/store
|
||||
|
||||
// A parameter class for the direction of transfer.
|
||||
class MVE_ldst_direction<bit b, dag Oo, dag Io, string c=""> {
|
||||
bit load = b;
|
||||
dag Oops = Oo;
|
||||
dag Iops = Io;
|
||||
string cstr = c;
|
||||
}
|
||||
def MVE_ld: MVE_ldst_direction<1, (outs MQPR:$Qd), (ins), ",@earlyclobber $Qd">;
|
||||
def MVE_st: MVE_ldst_direction<0, (outs), (ins MQPR:$Qd)>;
|
||||
|
||||
// A parameter class for the size of memory access in a load.
|
||||
class MVE_memsz<bits<2> e, int s, AddrMode m, string mn, list<string> types> {
|
||||
bits<2> encoding = e; // opcode bit(s) for encoding
|
||||
int shift = s; // shift applied to immediate load offset
|
||||
AddrMode AM = m;
|
||||
|
||||
// For instruction aliases: define the complete list of type
|
||||
// suffixes at this size, and the canonical ones for loads and
|
||||
// stores.
|
||||
string MnemonicLetter = mn;
|
||||
int TypeBits = !shl(8, s);
|
||||
string CanonLoadSuffix = ".u" # TypeBits;
|
||||
string CanonStoreSuffix = "." # TypeBits;
|
||||
list<string> suffixes = !foreach(letter, types, "." # letter # TypeBits);
|
||||
}
|
||||
|
||||
// Instances of MVE_memsz.
|
||||
//
|
||||
// (memD doesn't need an AddrMode, because those are only for
|
||||
// contiguous loads, and memD is only used by gather/scatters.)
|
||||
def MVE_memB: MVE_memsz<0b00, 0, AddrModeT2_i7, "b", ["", "u", "s"]>;
|
||||
def MVE_memH: MVE_memsz<0b01, 1, AddrModeT2_i7s2, "h", ["", "u", "s", "f"]>;
|
||||
def MVE_memW: MVE_memsz<0b10, 2, AddrModeT2_i7s4, "w", ["", "u", "s", "f"]>;
|
||||
def MVE_memD: MVE_memsz<0b11, 3, ?, "d", ["", "u", "s", "f"]>;
|
||||
|
||||
// This is the base class for all the MVE loads and stores other than
|
||||
// the interleaving ones. All the non-interleaving loads/stores share
|
||||
// the characteristic that they operate on just one vector register,
|
||||
// so they are VPT-predicable.
|
||||
//
|
||||
// The predication operand is vpred_n, for both loads and stores. For
|
||||
// store instructions, the reason is obvious: if there is no output
|
||||
// register, there can't be a need for an input parameter giving the
|
||||
// output register's previous value. Load instructions also don't need
|
||||
// that input parameter, because unlike MVE data processing
|
||||
// instructions, predicated loads are defined to set the inactive
|
||||
// lanes of the output register to zero, instead of preserving their
|
||||
// input values.
|
||||
class MVE_VLDRSTR_base<MVE_ldst_direction dir, bit U, bit P, bit W, bit opc,
|
||||
dag oops, dag iops, string asm, string suffix,
|
||||
string ops, string cstr, list<dag> pattern=[]>
|
||||
: MVE_p<oops, iops, NoItinerary, asm, suffix, ops, vpred_n, cstr, pattern> {
|
||||
bits<3> Qd;
|
||||
|
||||
let Inst{28} = U;
|
||||
let Inst{25} = 0b0;
|
||||
let Inst{24} = P;
|
||||
let Inst{22} = 0b0;
|
||||
let Inst{21} = W;
|
||||
let Inst{20} = dir.load;
|
||||
let Inst{15-13} = Qd{2-0};
|
||||
let Inst{12} = opc;
|
||||
let Inst{11-9} = 0b111;
|
||||
|
||||
let mayLoad = dir.load;
|
||||
let mayStore = !eq(dir.load,0);
|
||||
}
|
||||
|
||||
// Contiguous load and store instructions. These come in two main
|
||||
// categories: same-size loads/stores in which 128 bits of vector
|
||||
// register is transferred to or from 128 bits of memory in the most
|
||||
// obvious way, and widening loads / narrowing stores, in which the
|
||||
// size of memory accessed is less than the size of a vector register,
|
||||
// so the load instructions sign- or zero-extend each memory value
|
||||
// into a wider vector lane, and the store instructions truncate
|
||||
// correspondingly.
|
||||
//
|
||||
// The instruction mnemonics for these two classes look reasonably
|
||||
// similar, but the actual encodings are different enough to need two
|
||||
// separate base classes.
|
||||
|
||||
// Contiguous, same size
|
||||
class MVE_VLDRSTR_cs<MVE_ldst_direction dir, MVE_memsz memsz, bit P, bit W,
|
||||
dag oops, dag iops, string asm, string suffix,
|
||||
IndexMode im, string ops, string cstr>
|
||||
: MVE_VLDRSTR_base<dir, 0, P, W, 1, oops, iops, asm, suffix, ops, cstr> {
|
||||
bits<12> addr;
|
||||
let Inst{23} = addr{7};
|
||||
let Inst{19-16} = addr{11-8};
|
||||
let Inst{8-7} = memsz.encoding;
|
||||
let Inst{6-0} = addr{6-0};
|
||||
}
|
||||
|
||||
// Contiguous, widening/narrowing
|
||||
class MVE_VLDRSTR_cw<MVE_ldst_direction dir, MVE_memsz memsz, bit U,
|
||||
bit P, bit W, bits<2> size, dag oops, dag iops,
|
||||
string asm, string suffix, IndexMode im,
|
||||
string ops, string cstr>
|
||||
: MVE_VLDRSTR_base<dir, U, P, W, 0, oops, iops, asm, suffix, ops, cstr> {
|
||||
bits<11> addr;
|
||||
let Inst{23} = addr{7};
|
||||
let Inst{19} = memsz.encoding{0}; // enough to tell 16- from 32-bit
|
||||
let Inst{18-16} = addr{10-8};
|
||||
let Inst{8-7} = size;
|
||||
let Inst{6-0} = addr{6-0};
|
||||
|
||||
let IM = im;
|
||||
}
|
||||
|
||||
// Multiclass wrapper on each of the _cw and _cs base classes, to
|
||||
// generate three writeback modes (none, preindex, postindex).
|
||||
|
||||
multiclass MVE_VLDRSTR_cw_m<MVE_ldst_direction dir, MVE_memsz memsz,
|
||||
string asm, string suffix, bit U, bits<2> size> {
|
||||
let AM = memsz.AM in {
|
||||
def "" : MVE_VLDRSTR_cw<
|
||||
dir, memsz, U, 1, 0, size,
|
||||
dir.Oops, !con(dir.Iops, (ins taddrmode_imm7<memsz.shift>:$addr)),
|
||||
asm, suffix, IndexModeNone, "$Qd, $addr", "">;
|
||||
|
||||
def _pre : MVE_VLDRSTR_cw<
|
||||
dir, memsz, U, 1, 1, size,
|
||||
!con((outs tGPR:$wb), dir.Oops),
|
||||
!con(dir.Iops, (ins taddrmode_imm7<memsz.shift>:$addr)),
|
||||
asm, suffix, IndexModePre, "$Qd, $addr!", "$addr.base = $wb"> {
|
||||
let DecoderMethod = "DecodeMVE_MEM_1_pre<"#memsz.shift#">";
|
||||
}
|
||||
|
||||
def _post : MVE_VLDRSTR_cw<
|
||||
dir, memsz, U, 0, 1, size,
|
||||
!con((outs tGPR:$wb), dir.Oops),
|
||||
!con(dir.Iops, (ins t_addr_offset_none:$Rn,
|
||||
t2am_imm7_offset<memsz.shift>:$addr)),
|
||||
asm, suffix, IndexModePost, "$Qd, $Rn$addr", "$Rn.base = $wb"> {
|
||||
bits<4> Rn;
|
||||
let Inst{18-16} = Rn{2-0};
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
multiclass MVE_VLDRSTR_cs_m<MVE_ldst_direction dir, MVE_memsz memsz,
|
||||
string asm, string suffix> {
|
||||
let AM = memsz.AM in {
|
||||
def "" : MVE_VLDRSTR_cs<
|
||||
dir, memsz, 1, 0,
|
||||
dir.Oops, !con(dir.Iops, (ins t2addrmode_imm7<memsz.shift>:$addr)),
|
||||
asm, suffix, IndexModeNone, "$Qd, $addr", "">;
|
||||
|
||||
def _pre : MVE_VLDRSTR_cs<
|
||||
dir, memsz, 1, 1,
|
||||
!con((outs rGPR:$wb), dir.Oops),
|
||||
!con(dir.Iops, (ins t2addrmode_imm7_pre<memsz.shift>:$addr)),
|
||||
asm, suffix, IndexModePre, "$Qd, $addr!", "$addr.base = $wb"> {
|
||||
let DecoderMethod = "DecodeMVE_MEM_2_pre<"#memsz.shift#">";
|
||||
}
|
||||
|
||||
def _post : MVE_VLDRSTR_cs<
|
||||
dir, memsz, 0, 1,
|
||||
!con((outs rGPR:$wb), dir.Oops),
|
||||
// We need an !if here to select the base register class,
|
||||
// because it's legal to write back to SP in a load of this
|
||||
// type, but not in a store.
|
||||
!con(dir.Iops, (ins !if(dir.load, t2_addr_offset_none,
|
||||
t2_nosp_addr_offset_none):$Rn,
|
||||
t2am_imm7_offset<memsz.shift>:$addr)),
|
||||
asm, suffix, IndexModePost, "$Qd, $Rn$addr", "$Rn.base = $wb"> {
|
||||
bits<4> Rn;
|
||||
let Inst{19-16} = Rn{3-0};
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Now actually declare all the contiguous load/stores, via those
|
||||
// multiclasses. The instruction ids coming out of this are the bare
|
||||
// names shown in the defm, with _pre or _post appended for writeback,
|
||||
// e.g. MVE_VLDRBS16, MVE_VSTRB16_pre, MVE_VSTRHU16_post.
|
||||
|
||||
defm MVE_VLDRBS16: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "s16", 0, 0b01>;
|
||||
defm MVE_VLDRBS32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "s32", 0, 0b10>;
|
||||
defm MVE_VLDRBU16: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "u16", 1, 0b01>;
|
||||
defm MVE_VLDRBU32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "u32", 1, 0b10>;
|
||||
defm MVE_VLDRHS32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memH, "vldrh", "s32", 0, 0b10>;
|
||||
defm MVE_VLDRHU32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memH, "vldrh", "u32", 1, 0b10>;
|
||||
|
||||
defm MVE_VLDRBU8: MVE_VLDRSTR_cs_m<MVE_ld, MVE_memB, "vldrb", "u8">;
|
||||
defm MVE_VLDRHU16: MVE_VLDRSTR_cs_m<MVE_ld, MVE_memH, "vldrh", "u16">;
|
||||
defm MVE_VLDRWU32: MVE_VLDRSTR_cs_m<MVE_ld, MVE_memW, "vldrw", "u32">;
|
||||
|
||||
defm MVE_VSTRB16: MVE_VLDRSTR_cw_m<MVE_st, MVE_memB, "vstrb", "16", 0, 0b01>;
|
||||
defm MVE_VSTRB32: MVE_VLDRSTR_cw_m<MVE_st, MVE_memB, "vstrb", "32", 0, 0b10>;
|
||||
defm MVE_VSTRH32: MVE_VLDRSTR_cw_m<MVE_st, MVE_memH, "vstrh", "32", 0, 0b10>;
|
||||
|
||||
defm MVE_VSTRBU8 : MVE_VLDRSTR_cs_m<MVE_st, MVE_memB, "vstrb", "8">;
|
||||
defm MVE_VSTRHU16: MVE_VLDRSTR_cs_m<MVE_st, MVE_memH, "vstrh", "16">;
|
||||
defm MVE_VSTRWU32: MVE_VLDRSTR_cs_m<MVE_st, MVE_memW, "vstrw", "32">;
|
||||
|
||||
// Gather loads / scatter stores whose address operand is of the form
|
||||
// [Rn,Qm], i.e. a single GPR as the common base address, plus a
|
||||
// vector of offset from it. ('Load/store this sequence of elements of
|
||||
// the same array.')
|
||||
//
|
||||
// Like the contiguous family, these loads and stores can widen the
|
||||
// loaded values / truncate the stored ones, or they can just
|
||||
// load/store the same size of memory and vector lane. But unlike the
|
||||
// contiguous family, there's no particular difference in encoding
|
||||
// between those two cases.
|
||||
//
|
||||
// This family also comes with the option to scale the offset values
|
||||
// in Qm by the size of the loaded memory (i.e. to treat them as array
|
||||
// indices), or not to scale them (to treat them as plain byte offsets
|
||||
// in memory, so that perhaps the loaded values are unaligned). The
|
||||
// scaled instructions' address operand in assembly looks like
|
||||
// [Rn,Qm,UXTW #2] or similar.
|
||||
|
||||
// Base class.
|
||||
class MVE_VLDRSTR_rq<MVE_ldst_direction dir, MVE_memsz memsz, bit U,
|
||||
bits<2> size, bit os, string asm, string suffix, int shift>
|
||||
: MVE_VLDRSTR_base<dir, U, 0b0, 0b0, 0, dir.Oops,
|
||||
!con(dir.Iops, (ins mve_addr_rq_shift<shift>:$addr)),
|
||||
asm, suffix, "$Qd, $addr", dir.cstr> {
|
||||
bits<7> addr;
|
||||
let Inst{23} = 0b1;
|
||||
let Inst{19-16} = addr{6-3};
|
||||
let Inst{8-7} = size;
|
||||
let Inst{6} = memsz.encoding{1};
|
||||
let Inst{5} = 0;
|
||||
let Inst{4} = memsz.encoding{0};
|
||||
let Inst{3-1} = addr{2-0};
|
||||
let Inst{0} = os;
|
||||
}
|
||||
|
||||
// Multiclass that defines the scaled and unscaled versions of an
|
||||
// instruction, when the memory size is wider than a byte. The scaled
|
||||
// version gets the default name like MVE_VLDRBU16_rq; the unscaled /
|
||||
// potentially unaligned version gets a "_u" suffix, e.g.
|
||||
// MVE_VLDRBU16_rq_u.
|
||||
multiclass MVE_VLDRSTR_rq_w<MVE_ldst_direction dir, MVE_memsz memsz,
|
||||
string asm, string suffix, bit U, bits<2> size> {
|
||||
def _u : MVE_VLDRSTR_rq<dir, memsz, U, size, 0, asm, suffix, 0>;
|
||||
def "" : MVE_VLDRSTR_rq<dir, memsz, U, size, 1, asm, suffix, memsz.shift>;
|
||||
}
|
||||
|
||||
// Subclass of MVE_VLDRSTR_rq with the same API as that multiclass,
|
||||
// for use when the memory size is one byte, so there's no 'scaled'
|
||||
// version of the instruction at all. (This is encoded as if it were
|
||||
// unscaled, but named in the default way with no _u suffix.)
|
||||
class MVE_VLDRSTR_rq_b<MVE_ldst_direction dir, MVE_memsz memsz,
|
||||
string asm, string suffix, bit U, bits<2> size>
|
||||
: MVE_VLDRSTR_rq<dir, memsz, U, size, 0, asm, suffix, 0>;
|
||||
|
||||
// Actually define all the loads and stores in this family.
|
||||
|
||||
def MVE_VLDRBU8_rq : MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","u8", 1,0b00>;
|
||||
def MVE_VLDRBU16_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","u16", 1,0b01>;
|
||||
def MVE_VLDRBS16_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","s16", 0,0b01>;
|
||||
def MVE_VLDRBU32_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","u32", 1,0b10>;
|
||||
def MVE_VLDRBS32_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","s32", 0,0b10>;
|
||||
|
||||
defm MVE_VLDRHU16_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memH, "vldrh","u16", 1,0b01>;
|
||||
defm MVE_VLDRHU32_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memH, "vldrh","u32", 1,0b10>;
|
||||
defm MVE_VLDRHS32_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memH, "vldrh","s32", 0,0b10>;
|
||||
defm MVE_VLDRWU32_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memW, "vldrw","u32", 1,0b10>;
|
||||
defm MVE_VLDRDU64_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memD, "vldrd","u64", 1,0b11>;
|
||||
|
||||
def MVE_VSTRB8_rq : MVE_VLDRSTR_rq_b<MVE_st, MVE_memB, "vstrb","8", 0,0b00>;
|
||||
def MVE_VSTRB16_rq : MVE_VLDRSTR_rq_b<MVE_st, MVE_memB, "vstrb","16", 0,0b01>;
|
||||
def MVE_VSTRB32_rq : MVE_VLDRSTR_rq_b<MVE_st, MVE_memB, "vstrb","32", 0,0b10>;
|
||||
|
||||
defm MVE_VSTRH16_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memH, "vstrh","16", 0,0b01>;
|
||||
defm MVE_VSTRH32_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memH, "vstrh","32", 0,0b10>;
|
||||
defm MVE_VSTRW32_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memW, "vstrw","32", 0,0b10>;
|
||||
defm MVE_VSTRD64_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memD, "vstrd","64", 0,0b11>;
|
||||
|
||||
// Gather loads / scatter stores whose address operand is of the form
|
||||
// [Qm,#imm], i.e. a vector containing a full base address for each
|
||||
// loaded item, plus an immediate offset applied consistently to all
|
||||
// of them. ('Load/store the same field from this vector of pointers
|
||||
// to a structure type.')
|
||||
//
|
||||
// This family requires the vector lane size to be at least 32 bits
|
||||
// (so there's room for an address in each lane at all). It has no
|
||||
// widening/narrowing variants. But it does support preindex
|
||||
// writeback, in which the address vector is updated to hold the
|
||||
// addresses actually loaded from.
|
||||
|
||||
// Base class.
|
||||
class MVE_VLDRSTR_qi<MVE_ldst_direction dir, MVE_memsz memsz, bit W, dag wbops,
|
||||
string asm, string wbAsm, string suffix, string cstr = "">
|
||||
: MVE_VLDRSTR_base<dir, 1, 1, W, 1, !con(wbops, dir.Oops),
|
||||
!con(dir.Iops, (ins mve_addr_q_shift<memsz.shift>:$addr)),
|
||||
asm, suffix, "$Qd, $addr" # wbAsm, cstr # dir.cstr> {
|
||||
bits<11> addr;
|
||||
let Inst{23} = addr{7};
|
||||
let Inst{19-17} = addr{10-8};
|
||||
let Inst{16} = 0;
|
||||
let Inst{8} = memsz.encoding{0}; // enough to distinguish 32- from 64-bit
|
||||
let Inst{7} = 0;
|
||||
let Inst{6-0} = addr{6-0};
|
||||
}
|
||||
|
||||
// Multiclass that generates the non-writeback and writeback variants.
|
||||
multiclass MVE_VLDRSTR_qi_m<MVE_ldst_direction dir, MVE_memsz memsz,
|
||||
string asm, string suffix> {
|
||||
def "" : MVE_VLDRSTR_qi<dir, memsz, 0, (outs), asm, "", suffix>;
|
||||
def _pre : MVE_VLDRSTR_qi<dir, memsz, 1, (outs MQPR:$wb), asm, "!", suffix,
|
||||
"$addr.base = $wb"> {
|
||||
let DecoderMethod="DecodeMVE_MEM_3_pre<"#memsz.shift#">";
|
||||
}
|
||||
}
|
||||
|
||||
// Actual instruction definitions.
|
||||
defm MVE_VLDRWU32_qi: MVE_VLDRSTR_qi_m<MVE_ld, MVE_memW, "vldrw", "u32">;
|
||||
defm MVE_VLDRDU64_qi: MVE_VLDRSTR_qi_m<MVE_ld, MVE_memD, "vldrd", "u64">;
|
||||
defm MVE_VSTRW32_qi: MVE_VLDRSTR_qi_m<MVE_st, MVE_memW, "vstrw", "32">;
|
||||
defm MVE_VSTRD64_qi: MVE_VLDRSTR_qi_m<MVE_st, MVE_memD, "vstrd", "64">;
|
||||
|
||||
// Define aliases for all the instructions where memory size and
|
||||
// vector lane size are the same. These are mnemonic aliases, so they
|
||||
// apply consistently across all of the above families - contiguous
|
||||
// loads, and both the rq and qi types of gather/scatter.
|
||||
//
|
||||
// Rationale: As long as you're loading (for example) 16-bit memory
|
||||
// values into 16-bit vector lanes, you can think of them as signed or
|
||||
// unsigned integers, fp16 or just raw 16-bit blobs and it makes no
|
||||
// difference. So we permit all of vldrh.16, vldrh.u16, vldrh.s16,
|
||||
// vldrh.f16 and treat them all as equivalent to the canonical
|
||||
// spelling (which happens to be .u16 for loads, and just .16 for
|
||||
// stores).
|
||||
|
||||
foreach vpt_cond = ["", "t", "e"] in
|
||||
foreach memsz = [MVE_memB, MVE_memH, MVE_memW, MVE_memD] in
|
||||
foreach suffix = memsz.suffixes in {
|
||||
|
||||
// These foreaches are conceptually ifs, implemented by iterating a
|
||||
// dummy variable over a list with 0 or 1 elements depending on the
|
||||
// condition. The idea is to iterate over _nearly_ all the suffixes
|
||||
// in memsz.suffixes, but omit the one we want all the others to alias.
|
||||
|
||||
foreach _ = !if(!ne(suffix, memsz.CanonLoadSuffix), [1], []<int>) in
|
||||
def : MnemonicAlias<
|
||||
"vldr" # memsz.MnemonicLetter # vpt_cond # suffix,
|
||||
"vldr" # memsz.MnemonicLetter # vpt_cond # memsz.CanonLoadSuffix>;
|
||||
|
||||
foreach _ = !if(!ne(suffix, memsz.CanonStoreSuffix), [1], []<int>) in
|
||||
def : MnemonicAlias<
|
||||
"vstr" # memsz.MnemonicLetter # vpt_cond # suffix,
|
||||
"vstr" # memsz.MnemonicLetter # vpt_cond # memsz.CanonStoreSuffix>;
|
||||
}
|
||||
|
||||
// end of MVE predicable load/store
|
||||
|
||||
class MVE_VPT<string suffix, bits<2> size, dag iops, string asm, list<dag> pattern=[]>
|
||||
: MVE_MI<(outs ), iops, NoItinerary, !strconcat("vpt", "${Mk}", ".", suffix), asm, "", pattern> {
|
||||
bits<3> fc;
|
||||
|
||||
@ -212,31 +212,40 @@ def t2adrlabel : Operand<i32> {
|
||||
}
|
||||
|
||||
// t2addrmode_posimm8 := reg + imm8
|
||||
def MemPosImm8OffsetAsmOperand : AsmOperandClass {let Name="MemPosImm8Offset";}
|
||||
def MemPosImm8OffsetAsmOperand : AsmOperandClass {
|
||||
let Name="MemPosImm8Offset";
|
||||
let RenderMethod = "addMemImmOffsetOperands";
|
||||
}
|
||||
def t2addrmode_posimm8 : MemOperand {
|
||||
let PrintMethod = "printT2AddrModeImm8Operand<false>";
|
||||
let EncoderMethod = "getT2AddrModeImm8OpValue";
|
||||
let EncoderMethod = "getT2AddrModeImmOpValue<8,0>";
|
||||
let DecoderMethod = "DecodeT2AddrModeImm8";
|
||||
let ParserMatchClass = MemPosImm8OffsetAsmOperand;
|
||||
let MIOperandInfo = (ops GPR:$base, i32imm:$offsimm);
|
||||
}
|
||||
|
||||
// t2addrmode_negimm8 := reg - imm8
|
||||
def MemNegImm8OffsetAsmOperand : AsmOperandClass {let Name="MemNegImm8Offset";}
|
||||
def MemNegImm8OffsetAsmOperand : AsmOperandClass {
|
||||
let Name="MemNegImm8Offset";
|
||||
let RenderMethod = "addMemImmOffsetOperands";
|
||||
}
|
||||
def t2addrmode_negimm8 : MemOperand,
|
||||
ComplexPattern<i32, 2, "SelectT2AddrModeImm8", []> {
|
||||
let PrintMethod = "printT2AddrModeImm8Operand<false>";
|
||||
let EncoderMethod = "getT2AddrModeImm8OpValue";
|
||||
let EncoderMethod = "getT2AddrModeImmOpValue<8,0>";
|
||||
let DecoderMethod = "DecodeT2AddrModeImm8";
|
||||
let ParserMatchClass = MemNegImm8OffsetAsmOperand;
|
||||
let MIOperandInfo = (ops GPR:$base, i32imm:$offsimm);
|
||||
}
|
||||
|
||||
// t2addrmode_imm8 := reg +/- imm8
|
||||
def MemImm8OffsetAsmOperand : AsmOperandClass { let Name = "MemImm8Offset"; }
|
||||
def MemImm8OffsetAsmOperand : AsmOperandClass {
|
||||
let Name = "MemImm8Offset";
|
||||
let RenderMethod = "addMemImmOffsetOperands";
|
||||
}
|
||||
class T2AddrMode_Imm8 : MemOperand,
|
||||
ComplexPattern<i32, 2, "SelectT2AddrModeImm8", []> {
|
||||
let EncoderMethod = "getT2AddrModeImm8OpValue";
|
||||
let EncoderMethod = "getT2AddrModeImmOpValue<8,0>";
|
||||
let DecoderMethod = "DecodeT2AddrModeImm8";
|
||||
let ParserMatchClass = MemImm8OffsetAsmOperand;
|
||||
let MIOperandInfo = (ops GPR:$base, i32imm:$offsimm);
|
||||
|
||||
@ -1025,7 +1025,7 @@ public:
|
||||
if (!CE) return false;
|
||||
Val = CE->getValue();
|
||||
}
|
||||
else if (isMem()) {
|
||||
else if (isGPRMem()) {
|
||||
if(!Memory.OffsetImm || Memory.OffsetRegNum) return false;
|
||||
if(Memory.BaseRegNum != ARM::PC) return false;
|
||||
Val = Memory.OffsetImm->getValue();
|
||||
@ -1059,7 +1059,14 @@ public:
|
||||
int64_t Value = CE->getValue();
|
||||
return ((Value & 3) == 0) && Value >= N && Value <= M;
|
||||
}
|
||||
|
||||
template<int64_t N, int64_t M>
|
||||
bool isImmediateS2() const {
|
||||
if (!isImm()) return false;
|
||||
const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
|
||||
if (!CE) return false;
|
||||
int64_t Value = CE->getValue();
|
||||
return ((Value & 1) == 0) && Value >= N && Value <= M;
|
||||
}
|
||||
bool isFBits16() const {
|
||||
return isImmediate<0, 17>();
|
||||
}
|
||||
@ -1072,6 +1079,18 @@ public:
|
||||
bool isImm7s4() const {
|
||||
return isImmediateS4<-508, 508>();
|
||||
}
|
||||
bool isImm7Shift0() const {
|
||||
return isImmediate<-127, 127>();
|
||||
}
|
||||
bool isImm7Shift1() const {
|
||||
return isImmediateS2<-255, 255>();
|
||||
}
|
||||
bool isImm7Shift2() const {
|
||||
return isImmediateS4<-511, 511>();
|
||||
}
|
||||
bool isImm7() const {
|
||||
return isImmediate<-127, 127>();
|
||||
}
|
||||
bool isImm0_1020s4() const {
|
||||
return isImmediateS4<0, 1020>();
|
||||
}
|
||||
@ -1253,6 +1272,22 @@ public:
|
||||
bool isInstSyncBarrierOpt() const { return Kind == k_InstSyncBarrierOpt; }
|
||||
bool isTraceSyncBarrierOpt() const { return Kind == k_TraceSyncBarrierOpt; }
|
||||
bool isMem() const override {
|
||||
return isGPRMem() || isMVEMem();
|
||||
}
|
||||
bool isMVEMem() const {
|
||||
if (Kind != k_Memory)
|
||||
return false;
|
||||
if (Memory.BaseRegNum &&
|
||||
!ARMMCRegisterClasses[ARM::GPRRegClassID].contains(Memory.BaseRegNum) &&
|
||||
!ARMMCRegisterClasses[ARM::MQPRRegClassID].contains(Memory.BaseRegNum))
|
||||
return false;
|
||||
if (Memory.OffsetRegNum &&
|
||||
!ARMMCRegisterClasses[ARM::MQPRRegClassID].contains(
|
||||
Memory.OffsetRegNum))
|
||||
return false;
|
||||
return true;
|
||||
}
|
||||
bool isGPRMem() const {
|
||||
if (Kind != k_Memory)
|
||||
return false;
|
||||
if (Memory.BaseRegNum &&
|
||||
@ -1332,14 +1367,14 @@ public:
|
||||
return isPostIdxRegShifted() && PostIdxReg.ShiftTy == ARM_AM::no_shift;
|
||||
}
|
||||
bool isMemNoOffset(bool alignOK = false, unsigned Alignment = 0) const {
|
||||
if (!isMem())
|
||||
if (!isGPRMem())
|
||||
return false;
|
||||
// No offset of any kind.
|
||||
return Memory.OffsetRegNum == 0 && Memory.OffsetImm == nullptr &&
|
||||
(alignOK || Memory.Alignment == Alignment);
|
||||
}
|
||||
bool isMemNoOffsetT2(bool alignOK = false, unsigned Alignment = 0) const {
|
||||
if (!isMem())
|
||||
if (!isGPRMem())
|
||||
return false;
|
||||
|
||||
if (!ARMMCRegisterClasses[ARM::GPRnopcRegClassID].contains(
|
||||
@ -1351,7 +1386,7 @@ public:
|
||||
(alignOK || Memory.Alignment == Alignment);
|
||||
}
|
||||
bool isMemNoOffsetT2NoSp(bool alignOK = false, unsigned Alignment = 0) const {
|
||||
if (!isMem())
|
||||
if (!isGPRMem())
|
||||
return false;
|
||||
|
||||
if (!ARMMCRegisterClasses[ARM::rGPRRegClassID].contains(
|
||||
@ -1362,8 +1397,20 @@ public:
|
||||
return Memory.OffsetRegNum == 0 && Memory.OffsetImm == nullptr &&
|
||||
(alignOK || Memory.Alignment == Alignment);
|
||||
}
|
||||
bool isMemNoOffsetT(bool alignOK = false, unsigned Alignment = 0) const {
|
||||
if (!isGPRMem())
|
||||
return false;
|
||||
|
||||
if (!ARMMCRegisterClasses[ARM::tGPRRegClassID].contains(
|
||||
Memory.BaseRegNum))
|
||||
return false;
|
||||
|
||||
// No offset of any kind.
|
||||
return Memory.OffsetRegNum == 0 && Memory.OffsetImm == nullptr &&
|
||||
(alignOK || Memory.Alignment == Alignment);
|
||||
}
|
||||
bool isMemPCRelImm12() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Base register must be PC.
|
||||
if (Memory.BaseRegNum != ARM::PC)
|
||||
@ -1450,7 +1497,7 @@ public:
|
||||
}
|
||||
|
||||
bool isAddrMode2() const {
|
||||
if (!isMem() || Memory.Alignment != 0) return false;
|
||||
if (!isGPRMem() || Memory.Alignment != 0) return false;
|
||||
// Check for register offset.
|
||||
if (Memory.OffsetRegNum) return true;
|
||||
// Immediate offset in range [-4095, 4095].
|
||||
@ -1475,7 +1522,7 @@ public:
|
||||
// and we reject it.
|
||||
if (isImm() && !isa<MCConstantExpr>(getImm()))
|
||||
return true;
|
||||
if (!isMem() || Memory.Alignment != 0) return false;
|
||||
if (!isGPRMem() || Memory.Alignment != 0) return false;
|
||||
// No shifts are legal for AM3.
|
||||
if (Memory.ShiftType != ARM_AM::no_shift) return false;
|
||||
// Check for register offset.
|
||||
@ -1509,7 +1556,7 @@ public:
|
||||
// and we reject it.
|
||||
if (isImm() && !isa<MCConstantExpr>(getImm()))
|
||||
return true;
|
||||
if (!isMem() || Memory.Alignment != 0) return false;
|
||||
if (!isGPRMem() || Memory.Alignment != 0) return false;
|
||||
// Check for register offset.
|
||||
if (Memory.OffsetRegNum) return false;
|
||||
// Immediate offset in range [-1020, 1020] and a multiple of 4.
|
||||
@ -1525,7 +1572,7 @@ public:
|
||||
// and we reject it.
|
||||
if (isImm() && !isa<MCConstantExpr>(getImm()))
|
||||
return true;
|
||||
if (!isMem() || Memory.Alignment != 0) return false;
|
||||
if (!isGPRMem() || Memory.Alignment != 0) return false;
|
||||
// Check for register offset.
|
||||
if (Memory.OffsetRegNum) return false;
|
||||
// Immediate offset in range [-510, 510] and a multiple of 2.
|
||||
@ -1536,14 +1583,14 @@ public:
|
||||
}
|
||||
|
||||
bool isMemTBB() const {
|
||||
if (!isMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
if (!isGPRMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
Memory.ShiftType != ARM_AM::no_shift || Memory.Alignment != 0)
|
||||
return false;
|
||||
return true;
|
||||
}
|
||||
|
||||
bool isMemTBH() const {
|
||||
if (!isMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
if (!isGPRMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
Memory.ShiftType != ARM_AM::lsl || Memory.ShiftImm != 1 ||
|
||||
Memory.Alignment != 0 )
|
||||
return false;
|
||||
@ -1551,13 +1598,13 @@ public:
|
||||
}
|
||||
|
||||
bool isMemRegOffset() const {
|
||||
if (!isMem() || !Memory.OffsetRegNum || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || !Memory.OffsetRegNum || Memory.Alignment != 0)
|
||||
return false;
|
||||
return true;
|
||||
}
|
||||
|
||||
bool isT2MemRegOffset() const {
|
||||
if (!isMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
if (!isGPRMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
Memory.Alignment != 0 || Memory.BaseRegNum == ARM::PC)
|
||||
return false;
|
||||
// Only lsl #{0, 1, 2, 3} allowed.
|
||||
@ -1571,7 +1618,7 @@ public:
|
||||
bool isMemThumbRR() const {
|
||||
// Thumb reg+reg addressing is simple. Just two registers, a base and
|
||||
// an offset. No shifts, negations or any other complicating factors.
|
||||
if (!isMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
if (!isGPRMem() || !Memory.OffsetRegNum || Memory.isNegative ||
|
||||
Memory.ShiftType != ARM_AM::no_shift || Memory.Alignment != 0)
|
||||
return false;
|
||||
return isARMLowRegister(Memory.BaseRegNum) &&
|
||||
@ -1579,7 +1626,7 @@ public:
|
||||
}
|
||||
|
||||
bool isMemThumbRIs4() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 ||
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 ||
|
||||
!isARMLowRegister(Memory.BaseRegNum) || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset, multiple of 4 in range [0, 124].
|
||||
@ -1589,7 +1636,7 @@ public:
|
||||
}
|
||||
|
||||
bool isMemThumbRIs2() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 ||
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 ||
|
||||
!isARMLowRegister(Memory.BaseRegNum) || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset, multiple of 4 in range [0, 62].
|
||||
@ -1599,7 +1646,7 @@ public:
|
||||
}
|
||||
|
||||
bool isMemThumbRIs1() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 ||
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 ||
|
||||
!isARMLowRegister(Memory.BaseRegNum) || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset in range [0, 31].
|
||||
@ -1609,7 +1656,7 @@ public:
|
||||
}
|
||||
|
||||
bool isMemThumbSPI() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 ||
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 ||
|
||||
Memory.BaseRegNum != ARM::SP || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset, multiple of 4 in range [0, 1020].
|
||||
@ -1624,7 +1671,7 @@ public:
|
||||
// and we reject it.
|
||||
if (isImm() && !isa<MCConstantExpr>(getImm()))
|
||||
return true;
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset a multiple of 4 in range [-1020, 1020].
|
||||
if (!Memory.OffsetImm) return true;
|
||||
@ -1639,7 +1686,7 @@ public:
|
||||
// and we reject it.
|
||||
if (isImm() && !isa<MCConstantExpr>(getImm()))
|
||||
return true;
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0 ||
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0 ||
|
||||
!ARMMCRegisterClasses[ARM::GPRnopcRegClassID].contains(
|
||||
Memory.BaseRegNum))
|
||||
return false;
|
||||
@ -1650,7 +1697,7 @@ public:
|
||||
return (Val >= -508 && Val <= 508 && (Val & 3) == 0) || Val == INT32_MIN;
|
||||
}
|
||||
bool isMemImm0_1020s4Offset() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset a multiple of 4 in range [0, 1020].
|
||||
if (!Memory.OffsetImm) return true;
|
||||
@ -1659,7 +1706,7 @@ public:
|
||||
}
|
||||
|
||||
bool isMemImm8Offset() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Base reg of PC isn't allowed for these encodings.
|
||||
if (Memory.BaseRegNum == ARM::PC) return false;
|
||||
@ -1670,8 +1717,81 @@ public:
|
||||
(Val > -256 && Val < 256);
|
||||
}
|
||||
|
||||
template<unsigned Bits, unsigned RegClassID>
|
||||
bool isMemImm7ShiftedOffset() const {
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0 ||
|
||||
!ARMMCRegisterClasses[RegClassID].contains(Memory.BaseRegNum))
|
||||
return false;
|
||||
|
||||
// Expect an immediate offset equal to an element of the range
|
||||
// [-127, 127], shifted left by Bits.
|
||||
|
||||
if (!Memory.OffsetImm) return true;
|
||||
int64_t Val = Memory.OffsetImm->getValue();
|
||||
|
||||
// INT32_MIN is a special-case value (indicating the encoding with
|
||||
// zero offset and the subtract bit set)
|
||||
if (Val == INT32_MIN)
|
||||
return true;
|
||||
|
||||
unsigned Divisor = 1U << Bits;
|
||||
|
||||
// Check that the low bits are zero
|
||||
if (Val % Divisor != 0)
|
||||
return false;
|
||||
|
||||
// Check that the remaining offset is within range.
|
||||
Val /= Divisor;
|
||||
return (Val >= -127 && Val <= 127);
|
||||
}
|
||||
|
||||
template <int shift> bool isMemRegRQOffset() const {
|
||||
if (!isMVEMem() || Memory.OffsetImm != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
|
||||
if (!ARMMCRegisterClasses[ARM::GPRnopcRegClassID].contains(
|
||||
Memory.BaseRegNum))
|
||||
return false;
|
||||
if (!ARMMCRegisterClasses[ARM::MQPRRegClassID].contains(
|
||||
Memory.OffsetRegNum))
|
||||
return false;
|
||||
|
||||
if (shift == 0 && Memory.ShiftType != ARM_AM::no_shift)
|
||||
return false;
|
||||
|
||||
if (shift > 0 &&
|
||||
(Memory.ShiftType != ARM_AM::uxtw || Memory.ShiftImm != shift))
|
||||
return false;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
template <int shift> bool isMemRegQOffset() const {
|
||||
if (!isMVEMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
|
||||
if (!ARMMCRegisterClasses[ARM::MQPRRegClassID].contains(
|
||||
Memory.BaseRegNum))
|
||||
return false;
|
||||
|
||||
if(!Memory.OffsetImm) return true;
|
||||
static_assert(shift < 56,
|
||||
"Such that we dont shift by a value higher than 62");
|
||||
int64_t Val = Memory.OffsetImm->getValue();
|
||||
|
||||
// The value must be a multiple of (1 << shift)
|
||||
if ((Val & ((1U << shift) - 1)) != 0)
|
||||
return false;
|
||||
|
||||
// And be in the right range, depending on the amount that it is shifted
|
||||
// by. Shift 0, is equal to 7 unsigned bits, the sign bit is set
|
||||
// separately.
|
||||
int64_t Range = (1U << (7+shift)) - 1;
|
||||
return (Val == INT32_MIN) || (Val > -Range && Val < Range);
|
||||
}
|
||||
|
||||
bool isMemPosImm8Offset() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset in range [0, 255].
|
||||
if (!Memory.OffsetImm) return true;
|
||||
@ -1680,7 +1800,7 @@ public:
|
||||
}
|
||||
|
||||
bool isMemNegImm8Offset() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Base reg of PC isn't allowed for these encodings.
|
||||
if (Memory.BaseRegNum == ARM::PC) return false;
|
||||
@ -1692,7 +1812,7 @@ public:
|
||||
}
|
||||
|
||||
bool isMemUImm12Offset() const {
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset in range [0, 4095].
|
||||
if (!Memory.OffsetImm) return true;
|
||||
@ -1708,7 +1828,7 @@ public:
|
||||
if (isImm() && !isa<MCConstantExpr>(getImm()))
|
||||
return true;
|
||||
|
||||
if (!isMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
if (!isGPRMem() || Memory.OffsetRegNum != 0 || Memory.Alignment != 0)
|
||||
return false;
|
||||
// Immediate offset in range [-4095, 4095].
|
||||
if (!Memory.OffsetImm) return true;
|
||||
@ -2423,6 +2543,34 @@ public:
|
||||
Inst.addOperand(MCOperand::createImm(CE->getValue()));
|
||||
}
|
||||
|
||||
void addImm7Shift0Operands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 1 && "Invalid number of operands!");
|
||||
const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
|
||||
assert(CE != nullptr && "Invalid operand type!");
|
||||
Inst.addOperand(MCOperand::createImm(CE->getValue()));
|
||||
}
|
||||
|
||||
void addImm7Shift1Operands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 1 && "Invalid number of operands!");
|
||||
const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
|
||||
assert(CE != nullptr && "Invalid operand type!");
|
||||
Inst.addOperand(MCOperand::createImm(CE->getValue()));
|
||||
}
|
||||
|
||||
void addImm7Shift2Operands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 1 && "Invalid number of operands!");
|
||||
const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
|
||||
assert(CE != nullptr && "Invalid operand type!");
|
||||
Inst.addOperand(MCOperand::createImm(CE->getValue()));
|
||||
}
|
||||
|
||||
void addImm7Operands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 1 && "Invalid number of operands!");
|
||||
const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
|
||||
assert(CE != nullptr && "Invalid operand type!");
|
||||
Inst.addOperand(MCOperand::createImm(CE->getValue()));
|
||||
}
|
||||
|
||||
void addImm0_1020s4Operands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 1 && "Invalid number of operands!");
|
||||
// The immediate is scaled by four in the encoding and is stored
|
||||
@ -2532,7 +2680,7 @@ public:
|
||||
return;
|
||||
}
|
||||
|
||||
assert(isMem() && "Unknown value type!");
|
||||
assert(isGPRMem() && "Unknown value type!");
|
||||
assert(isa<MCConstantExpr>(Memory.OffsetImm) && "Unknown value type!");
|
||||
Inst.addOperand(MCOperand::createImm(Memory.OffsetImm->getValue()));
|
||||
}
|
||||
@ -2567,6 +2715,11 @@ public:
|
||||
Inst.addOperand(MCOperand::createReg(Memory.BaseRegNum));
|
||||
}
|
||||
|
||||
void addMemNoOffsetTOperands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 1 && "Invalid number of operands!");
|
||||
Inst.addOperand(MCOperand::createReg(Memory.BaseRegNum));
|
||||
}
|
||||
|
||||
void addMemPCRelImm12Operands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 1 && "Invalid number of operands!");
|
||||
int32_t Imm = Memory.OffsetImm->getValue();
|
||||
@ -2808,19 +2961,17 @@ public:
|
||||
Inst.addOperand(MCOperand::createImm(Val));
|
||||
}
|
||||
|
||||
void addMemImm8OffsetOperands(MCInst &Inst, unsigned N) const {
|
||||
void addMemImmOffsetOperands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 2 && "Invalid number of operands!");
|
||||
int64_t Val = Memory.OffsetImm ? Memory.OffsetImm->getValue() : 0;
|
||||
Inst.addOperand(MCOperand::createReg(Memory.BaseRegNum));
|
||||
Inst.addOperand(MCOperand::createImm(Val));
|
||||
}
|
||||
|
||||
void addMemPosImm8OffsetOperands(MCInst &Inst, unsigned N) const {
|
||||
addMemImm8OffsetOperands(Inst, N);
|
||||
}
|
||||
|
||||
void addMemNegImm8OffsetOperands(MCInst &Inst, unsigned N) const {
|
||||
addMemImm8OffsetOperands(Inst, N);
|
||||
void addMemRegRQOffsetOperands(MCInst &Inst, unsigned N) const {
|
||||
assert(N == 2 && "Invalid number of operands!");
|
||||
Inst.addOperand(MCOperand::createReg(Memory.BaseRegNum));
|
||||
Inst.addOperand(MCOperand::createReg(Memory.OffsetRegNum));
|
||||
}
|
||||
|
||||
void addMemUImm12OffsetOperands(MCInst &Inst, unsigned N) const {
|
||||
@ -5657,6 +5808,8 @@ bool ARMAsmParser::parseMemRegOffsetShift(ARM_AM::ShiftOpc &St,
|
||||
St = ARM_AM::ror;
|
||||
else if (ShiftName == "rrx" || ShiftName == "RRX")
|
||||
St = ARM_AM::rrx;
|
||||
else if (ShiftName == "uxtw" || ShiftName == "UXTW")
|
||||
St = ARM_AM::uxtw;
|
||||
else
|
||||
return Error(Loc, "illegal shift operator");
|
||||
Parser.Lex(); // Eat shift type token.
|
||||
@ -6518,7 +6671,7 @@ void ARMAsmParser::fixupGNULDRDAlias(StringRef Mnemonic,
|
||||
|
||||
if (!Op2.isReg())
|
||||
return;
|
||||
if (!Op3.isMem())
|
||||
if (!Op3.isGPRMem())
|
||||
return;
|
||||
|
||||
const MCRegisterClass &GPR = MRI->getRegClass(ARM::GPRRegClassID);
|
||||
@ -7291,6 +7444,54 @@ bool ARMAsmParser::validateInstruction(MCInst &Inst,
|
||||
"destination register and base register can't be identical");
|
||||
return false;
|
||||
}
|
||||
|
||||
case ARM::MVE_VLDRBU8_rq:
|
||||
case ARM::MVE_VLDRBU16_rq:
|
||||
case ARM::MVE_VLDRBS16_rq:
|
||||
case ARM::MVE_VLDRBU32_rq:
|
||||
case ARM::MVE_VLDRBS32_rq:
|
||||
case ARM::MVE_VLDRHU16_rq:
|
||||
case ARM::MVE_VLDRHU16_rq_u:
|
||||
case ARM::MVE_VLDRHU32_rq:
|
||||
case ARM::MVE_VLDRHU32_rq_u:
|
||||
case ARM::MVE_VLDRHS32_rq:
|
||||
case ARM::MVE_VLDRHS32_rq_u:
|
||||
case ARM::MVE_VLDRWU32_rq:
|
||||
case ARM::MVE_VLDRWU32_rq_u:
|
||||
case ARM::MVE_VLDRDU64_rq:
|
||||
case ARM::MVE_VLDRDU64_rq_u:
|
||||
case ARM::MVE_VLDRWU32_qi:
|
||||
case ARM::MVE_VLDRWU32_qi_pre:
|
||||
case ARM::MVE_VLDRDU64_qi:
|
||||
case ARM::MVE_VLDRDU64_qi_pre: {
|
||||
// Qd must be different from Qm.
|
||||
unsigned QdIdx = 0, QmIdx = 2;
|
||||
bool QmIsPointer = false;
|
||||
switch (Opcode) {
|
||||
case ARM::MVE_VLDRWU32_qi:
|
||||
case ARM::MVE_VLDRDU64_qi:
|
||||
QmIdx = 1;
|
||||
QmIsPointer = true;
|
||||
break;
|
||||
case ARM::MVE_VLDRWU32_qi_pre:
|
||||
case ARM::MVE_VLDRDU64_qi_pre:
|
||||
QdIdx = 1;
|
||||
QmIsPointer = true;
|
||||
break;
|
||||
}
|
||||
|
||||
const unsigned Qd = MRI->getEncodingValue(Inst.getOperand(QdIdx).getReg());
|
||||
const unsigned Qm = MRI->getEncodingValue(Inst.getOperand(QmIdx).getReg());
|
||||
|
||||
if (Qd == Qm) {
|
||||
return Error(Operands[3]->getStartLoc(),
|
||||
Twine("destination vector register and vector ") +
|
||||
(QmIsPointer ? "pointer" : "offset") +
|
||||
" register can't be identical");
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
case ARM::SBFX:
|
||||
case ARM::t2SBFX:
|
||||
case ARM::UBFX:
|
||||
@ -10042,9 +10243,23 @@ unsigned ARMAsmParser::checkTargetMatchPredicate(MCInst &Inst) {
|
||||
for (unsigned I = 0; I < MCID.NumOperands; ++I)
|
||||
if (MCID.OpInfo[I].RegClass == ARM::rGPRRegClassID) {
|
||||
// rGPRRegClass excludes PC, and also excluded SP before ARMv8
|
||||
if ((Inst.getOperand(I).getReg() == ARM::SP) && !hasV8Ops())
|
||||
const auto &Op = Inst.getOperand(I);
|
||||
if (!Op.isReg()) {
|
||||
// This can happen in awkward cases with tied operands, e.g. a
|
||||
// writeback load/store with a complex addressing mode in
|
||||
// which there's an output operand corresponding to the
|
||||
// updated written-back base register: the Tablegen-generated
|
||||
// AsmMatcher will have written a placeholder operand to that
|
||||
// slot in the form of an immediate 0, because it can't
|
||||
// generate the register part of the complex addressing-mode
|
||||
// operand ahead of time.
|
||||
continue;
|
||||
}
|
||||
|
||||
unsigned Reg = Op.getReg();
|
||||
if ((Reg == ARM::SP) && !hasV8Ops())
|
||||
return Match_RequiresV8;
|
||||
else if (Inst.getOperand(I).getReg() == ARM::PC)
|
||||
else if (Reg == ARM::PC)
|
||||
return Match_InvalidOperand;
|
||||
}
|
||||
|
||||
|
||||
@ -332,6 +332,11 @@ static DecodeStatus DecodeTBLInstruction(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodePostIdxReg(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodeMveAddrModeRQ(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder);
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeMveAddrModeQ(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodeCoprocessor(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodeMemBarrierOption(MCInst &Inst, unsigned Insn,
|
||||
@ -428,8 +433,17 @@ static DecodeStatus DecodeT2AddrModeImm0_1020s4(MCInst &Inst,unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodeT2Imm8(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeT2Imm7(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodeT2AddrModeImm8(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeTAddrModeImm7(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
template<int shift, int WriteBack>
|
||||
static DecodeStatus DecodeT2AddrModeImm7(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodeThumbAddSPImm(MCInst &Inst, uint16_t Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
static DecodeStatus DecodeThumbAddSPReg(MCInst &Inst, uint16_t Insn,
|
||||
@ -509,6 +523,15 @@ template<bool Writeback>
|
||||
static DecodeStatus DecodeVSTRVLDR_SYSREG(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address,
|
||||
const void *Decoder);
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeMVE_MEM_1_pre(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeMVE_MEM_2_pre(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeMVE_MEM_3_pre(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder);
|
||||
template<unsigned MinLog, unsigned MaxLog>
|
||||
static DecodeStatus DecodePowerTwoOperand(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address,
|
||||
@ -4138,6 +4161,21 @@ static DecodeStatus DecodeT2Imm8(MCInst &Inst, unsigned Val,
|
||||
return MCDisassembler::Success;
|
||||
}
|
||||
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeT2Imm7(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
int imm = Val & 0x7F;
|
||||
if (Val == 0)
|
||||
imm = INT32_MIN;
|
||||
else if (!(Val & 0x80))
|
||||
imm *= -1;
|
||||
if (imm != INT32_MIN)
|
||||
imm <<= shift;
|
||||
Inst.addOperand(MCOperand::createImm(imm));
|
||||
|
||||
return MCDisassembler::Success;
|
||||
}
|
||||
|
||||
static DecodeStatus DecodeT2AddrModeImm8(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
DecodeStatus S = MCDisassembler::Success;
|
||||
@ -4184,6 +4222,42 @@ static DecodeStatus DecodeT2AddrModeImm8(MCInst &Inst, unsigned Val,
|
||||
return S;
|
||||
}
|
||||
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeTAddrModeImm7(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address,
|
||||
const void *Decoder) {
|
||||
DecodeStatus S = MCDisassembler::Success;
|
||||
|
||||
unsigned Rn = fieldFromInstruction(Val, 8, 3);
|
||||
unsigned imm = fieldFromInstruction(Val, 0, 8);
|
||||
|
||||
if (!Check(S, DecodetGPRRegisterClass(Inst, Rn, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
if (!Check(S, DecodeT2Imm7<shift>(Inst, imm, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
|
||||
return S;
|
||||
}
|
||||
|
||||
template<int shift, int WriteBack>
|
||||
static DecodeStatus DecodeT2AddrModeImm7(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address,
|
||||
const void *Decoder) {
|
||||
DecodeStatus S = MCDisassembler::Success;
|
||||
|
||||
unsigned Rn = fieldFromInstruction(Val, 8, 4);
|
||||
unsigned imm = fieldFromInstruction(Val, 0, 8);
|
||||
if (WriteBack) {
|
||||
if (!Check(S, DecoderGPRRegisterClass(Inst, Rn, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
} else if (!Check(S, DecodeGPRnopcRegisterClass(Inst, Rn, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
if (!Check(S, DecodeT2Imm7<shift>(Inst, imm, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
|
||||
return S;
|
||||
}
|
||||
|
||||
static DecodeStatus DecodeT2LdStPre(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
DecodeStatus S = MCDisassembler::Success;
|
||||
@ -4331,6 +4405,43 @@ static DecodeStatus DecodePostIdxReg(MCInst &Inst, unsigned Insn,
|
||||
return S;
|
||||
}
|
||||
|
||||
static DecodeStatus DecodeMveAddrModeRQ(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
DecodeStatus S = MCDisassembler::Success;
|
||||
unsigned Rn = fieldFromInstruction(Insn, 3, 4);
|
||||
unsigned Qm = fieldFromInstruction(Insn, 0, 3);
|
||||
|
||||
if (!Check(S, DecodeGPRnopcRegisterClass(Inst, Rn, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
if (!Check(S, DecodeMQPRRegisterClass(Inst, Qm, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
|
||||
return S;
|
||||
}
|
||||
|
||||
template<int shift>
|
||||
static DecodeStatus DecodeMveAddrModeQ(MCInst &Inst, unsigned Insn,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
DecodeStatus S = MCDisassembler::Success;
|
||||
unsigned Qm = fieldFromInstruction(Insn, 8, 3);
|
||||
int imm = fieldFromInstruction(Insn, 0, 7);
|
||||
|
||||
if (!Check(S, DecodeMQPRRegisterClass(Inst, Qm, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
|
||||
if(!fieldFromInstruction(Insn, 7, 1)) {
|
||||
if (imm == 0)
|
||||
imm = INT32_MIN; // indicate -0
|
||||
else
|
||||
imm *= -1;
|
||||
}
|
||||
if (imm != INT32_MIN)
|
||||
imm <<= shift;
|
||||
Inst.addOperand(MCOperand::createImm(imm));
|
||||
|
||||
return S;
|
||||
}
|
||||
|
||||
static DecodeStatus DecodeThumbBLXOffset(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
// Val is passed in as S:J1:J2:imm10H:imm10L:'0'
|
||||
@ -6175,6 +6286,52 @@ static DecodeStatus DecodeVSTRVLDR_SYSREG(MCInst &Inst, unsigned Val,
|
||||
return S;
|
||||
}
|
||||
|
||||
static inline DecodeStatus DecodeMVE_MEM_pre(
|
||||
MCInst &Inst, unsigned Val, uint64_t Address, const void *Decoder,
|
||||
unsigned Rn, OperandDecoder RnDecoder, OperandDecoder AddrDecoder) {
|
||||
DecodeStatus S = MCDisassembler::Success;
|
||||
|
||||
unsigned Qd = fieldFromInstruction(Val, 13, 3);
|
||||
unsigned addr = fieldFromInstruction(Val, 0, 7) |
|
||||
(fieldFromInstruction(Val, 23, 1) << 7) | (Rn << 8);
|
||||
|
||||
if (!Check(S, RnDecoder(Inst, Rn, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
if (!Check(S, DecodeMQPRRegisterClass(Inst, Qd, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
if (!Check(S, AddrDecoder(Inst, addr, Address, Decoder)))
|
||||
return MCDisassembler::Fail;
|
||||
|
||||
return S;
|
||||
}
|
||||
|
||||
template <int shift>
|
||||
static DecodeStatus DecodeMVE_MEM_1_pre(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
return DecodeMVE_MEM_pre(Inst, Val, Address, Decoder,
|
||||
fieldFromInstruction(Val, 16, 3),
|
||||
DecodetGPRRegisterClass,
|
||||
DecodeTAddrModeImm7<shift>);
|
||||
}
|
||||
|
||||
template <int shift>
|
||||
static DecodeStatus DecodeMVE_MEM_2_pre(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
return DecodeMVE_MEM_pre(Inst, Val, Address, Decoder,
|
||||
fieldFromInstruction(Val, 16, 4),
|
||||
DecoderGPRRegisterClass,
|
||||
DecodeT2AddrModeImm7<shift,1>);
|
||||
}
|
||||
|
||||
template <int shift>
|
||||
static DecodeStatus DecodeMVE_MEM_3_pre(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address, const void *Decoder) {
|
||||
return DecodeMVE_MEM_pre(Inst, Val, Address, Decoder,
|
||||
fieldFromInstruction(Val, 17, 3),
|
||||
DecodeMQPRRegisterClass,
|
||||
DecodeMveAddrModeQ<shift>);
|
||||
}
|
||||
|
||||
template<unsigned MinLog, unsigned MaxLog>
|
||||
static DecodeStatus DecodePowerTwoOperand(MCInst &Inst, unsigned Val,
|
||||
uint64_t Address,
|
||||
|
||||
@ -30,7 +30,8 @@ namespace ARM_AM {
|
||||
lsl,
|
||||
lsr,
|
||||
ror,
|
||||
rrx
|
||||
rrx,
|
||||
uxtw
|
||||
};
|
||||
|
||||
enum AddrOpc {
|
||||
@ -48,6 +49,7 @@ namespace ARM_AM {
|
||||
case ARM_AM::lsr: return "lsr";
|
||||
case ARM_AM::ror: return "ror";
|
||||
case ARM_AM::rrx: return "rrx";
|
||||
case ARM_AM::uxtw: return "uxtw";
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -203,6 +203,8 @@ namespace ARMII {
|
||||
AddrMode5FP16 = 17, // i8 * 2
|
||||
AddrModeT2_ldrex = 18, // i8 * 4, with unscaled offset in MCInst
|
||||
AddrModeT2_i7s4 = 19, // i7 * 4
|
||||
AddrModeT2_i7s2 = 20, // i7 * 2
|
||||
AddrModeT2_i7 = 21, // i7 * 1
|
||||
};
|
||||
|
||||
inline static const char *AddrModeToString(AddrMode addrmode) {
|
||||
@ -227,6 +229,8 @@ namespace ARMII {
|
||||
case AddrMode_i12: return "AddrMode_i12";
|
||||
case AddrModeT2_ldrex:return "AddrModeT2_ldrex";
|
||||
case AddrModeT2_i7s4: return "AddrModeT2_i7s4";
|
||||
case AddrModeT2_i7s2: return "AddrModeT2_i7s2";
|
||||
case AddrModeT2_i7: return "AddrModeT2_i7";
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -603,6 +603,40 @@ void ARMInstPrinter::printPostIdxImm8s4Operand(const MCInst *MI, unsigned OpNum,
|
||||
<< markup(">");
|
||||
}
|
||||
|
||||
template<int shift>
|
||||
void ARMInstPrinter::printMveAddrModeRQOperand(const MCInst *MI, unsigned OpNum,
|
||||
const MCSubtargetInfo &STI,
|
||||
raw_ostream &O) {
|
||||
const MCOperand &MO1 = MI->getOperand(OpNum);
|
||||
const MCOperand &MO2 = MI->getOperand(OpNum + 1);
|
||||
|
||||
O << markup("<mem:") << "[";
|
||||
printRegName(O, MO1.getReg());
|
||||
O << ", ";
|
||||
printRegName(O, MO2.getReg());
|
||||
|
||||
if (shift > 0)
|
||||
printRegImmShift(O, ARM_AM::uxtw, shift, UseMarkup);
|
||||
|
||||
O << "]" << markup(">");
|
||||
}
|
||||
|
||||
void ARMInstPrinter::printMveAddrModeQOperand(const MCInst *MI, unsigned OpNum,
|
||||
const MCSubtargetInfo &STI,
|
||||
raw_ostream &O) {
|
||||
const MCOperand &MO1 = MI->getOperand(OpNum);
|
||||
const MCOperand &MO2 = MI->getOperand(OpNum + 1);
|
||||
|
||||
O << markup("<mem:") << "[";
|
||||
printRegName(O, MO1.getReg());
|
||||
|
||||
int64_t Imm = MO2.getImm();
|
||||
if (Imm != 0)
|
||||
O << ", " << markup("<imm:") << '#' << Imm << markup(">");
|
||||
|
||||
O << "]" << markup(">");
|
||||
}
|
||||
|
||||
void ARMInstPrinter::printLdStmModeOperand(const MCInst *MI, unsigned OpNum,
|
||||
const MCSubtargetInfo &STI,
|
||||
raw_ostream &O) {
|
||||
|
||||
@ -255,6 +255,11 @@ public:
|
||||
raw_ostream &O);
|
||||
void printVPTMask(const MCInst *MI, unsigned OpNum,
|
||||
const MCSubtargetInfo &STI, raw_ostream &O);
|
||||
template<int shift>
|
||||
void printMveAddrModeRQOperand(const MCInst *MI, unsigned OpNum,
|
||||
const MCSubtargetInfo &STI, raw_ostream &O);
|
||||
void printMveAddrModeQOperand(const MCInst *MI, unsigned OpNum,
|
||||
const MCSubtargetInfo &STI, raw_ostream &O);
|
||||
void printExpandedImmOperand(const MCInst *MI, unsigned OpNum,
|
||||
const MCSubtargetInfo &STI, raw_ostream &O);
|
||||
|
||||
|
||||
@ -208,6 +208,19 @@ public:
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const;
|
||||
|
||||
/// getMveAddrModeRQOpValue - Return encoding info for 'reg, vreg'
|
||||
/// operand.
|
||||
uint32_t getMveAddrModeRQOpValue(const MCInst &MI, unsigned OpIdx,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const;
|
||||
|
||||
/// getMveAddrModeQOpValue - Return encoding info for 'reg +/- imm7<<{shift}'
|
||||
/// operand.
|
||||
template<int shift>
|
||||
uint32_t getMveAddrModeQOpValue(const MCInst &MI, unsigned OpIdx,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const;
|
||||
|
||||
/// getLdStSORegOpValue - Return encoding info for 'reg +/- reg shop imm'
|
||||
/// operand as needed by load/store instructions.
|
||||
uint32_t getLdStSORegOpValue(const MCInst &MI, unsigned OpIdx,
|
||||
@ -238,8 +251,10 @@ public:
|
||||
case ARM_AM::asr: return 2;
|
||||
case ARM_AM::ror:
|
||||
case ARM_AM::rrx: return 3;
|
||||
case ARM_AM::uxtw:
|
||||
default:
|
||||
llvm_unreachable("Invalid ShiftOpc!");
|
||||
}
|
||||
llvm_unreachable("Invalid ShiftOpc!");
|
||||
}
|
||||
|
||||
/// getAddrMode2OffsetOpValue - Return encoding for am2offset operands.
|
||||
@ -338,7 +353,8 @@ public:
|
||||
unsigned getT2AddrModeSORegOpValue(const MCInst &MI, unsigned OpNum,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const;
|
||||
unsigned getT2AddrModeImm8OpValue(const MCInst &MI, unsigned OpNum,
|
||||
template<unsigned Bits, unsigned Shift>
|
||||
unsigned getT2AddrModeImmOpValue(const MCInst &MI, unsigned OpNum,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const;
|
||||
unsigned getT2AddrModeImm8OffsetOpValue(const MCInst &MI, unsigned OpNum,
|
||||
@ -1040,6 +1056,58 @@ getT2ScaledImmOpValue(const MCInst &MI, unsigned OpIdx,
|
||||
Binary |= (1U << Bits);
|
||||
return Binary;
|
||||
}
|
||||
|
||||
/// getMveAddrModeRQOpValue - Return encoding info for 'reg, vreg'
|
||||
/// operand.
|
||||
uint32_t ARMMCCodeEmitter::
|
||||
getMveAddrModeRQOpValue(const MCInst &MI, unsigned OpIdx,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const {
|
||||
// {6-3} Rn
|
||||
// {2-0} Qm
|
||||
const MCOperand &M0 = MI.getOperand(OpIdx);
|
||||
const MCOperand &M1 = MI.getOperand(OpIdx + 1);
|
||||
|
||||
unsigned Rn = CTX.getRegisterInfo()->getEncodingValue(M0.getReg());
|
||||
unsigned Qm = CTX.getRegisterInfo()->getEncodingValue(M1.getReg());
|
||||
|
||||
assert(Qm < 8 && "Qm is supposed to be encodable in 3 bits");
|
||||
|
||||
return (Rn << 3) | Qm;
|
||||
}
|
||||
|
||||
/// getMveAddrModeRQOpValue - Return encoding info for 'reg, vreg'
|
||||
/// operand.
|
||||
template<int shift>
|
||||
uint32_t ARMMCCodeEmitter::
|
||||
getMveAddrModeQOpValue(const MCInst &MI, unsigned OpIdx,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const {
|
||||
// {10-8} Qm
|
||||
// {7-0} Imm
|
||||
const MCOperand &M0 = MI.getOperand(OpIdx);
|
||||
const MCOperand &M1 = MI.getOperand(OpIdx + 1);
|
||||
|
||||
unsigned Qm = CTX.getRegisterInfo()->getEncodingValue(M0.getReg());
|
||||
int32_t Imm = M1.getImm();
|
||||
|
||||
bool isAdd = Imm >= 0;
|
||||
|
||||
Imm >>= shift;
|
||||
|
||||
if (!isAdd)
|
||||
Imm = -(uint32_t)Imm;
|
||||
|
||||
Imm &= 0x7f;
|
||||
|
||||
if (isAdd)
|
||||
Imm |= 0x80;
|
||||
|
||||
assert(Qm < 8 && "Qm is supposed to be encodable in 3 bits");
|
||||
|
||||
return (Qm << 8) | Imm;
|
||||
}
|
||||
|
||||
/// getT2AddrModeImm8s4OpValue - Return encoding info for
|
||||
/// 'reg +/- imm8<<2' operand.
|
||||
uint32_t ARMMCCodeEmitter::
|
||||
@ -1540,25 +1608,26 @@ getT2AddrModeSORegOpValue(const MCInst &MI, unsigned OpNum,
|
||||
return Value;
|
||||
}
|
||||
|
||||
template<unsigned Bits, unsigned Shift>
|
||||
unsigned ARMMCCodeEmitter::
|
||||
getT2AddrModeImm8OpValue(const MCInst &MI, unsigned OpNum,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const {
|
||||
getT2AddrModeImmOpValue(const MCInst &MI, unsigned OpNum,
|
||||
SmallVectorImpl<MCFixup> &Fixups,
|
||||
const MCSubtargetInfo &STI) const {
|
||||
const MCOperand &MO1 = MI.getOperand(OpNum);
|
||||
const MCOperand &MO2 = MI.getOperand(OpNum+1);
|
||||
|
||||
// FIXME: Needs fixup support.
|
||||
unsigned Value = CTX.getRegisterInfo()->getEncodingValue(MO1.getReg());
|
||||
|
||||
// Even though the immediate is 8 bits long, we need 9 bits in order
|
||||
// If the immediate is B bits long, we need B+1 bits in order
|
||||
// to represent the (inverse of the) sign bit.
|
||||
Value <<= 9;
|
||||
int32_t tmp = (int32_t)MO2.getImm();
|
||||
Value <<= (Bits + 1);
|
||||
int32_t tmp = (int32_t)MO2.getImm() >> Shift;
|
||||
if (tmp < 0)
|
||||
tmp = abs(tmp);
|
||||
else
|
||||
Value |= 256; // Set the ADD bit
|
||||
Value |= tmp & 255;
|
||||
Value |= (1U << Bits); // Set the ADD bit
|
||||
Value |= tmp & ((1U << Bits) - 1);
|
||||
return Value;
|
||||
}
|
||||
|
||||
|
||||
@ -610,17 +610,23 @@ bool llvm::rewriteT2FrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
|
||||
Offset = -Offset;
|
||||
isSub = true;
|
||||
}
|
||||
} else if (AddrMode == ARMII::AddrModeT2_i7s4) {
|
||||
} else if (AddrMode == ARMII::AddrModeT2_i7s4 ||
|
||||
AddrMode == ARMII::AddrModeT2_i7s2 ||
|
||||
AddrMode == ARMII::AddrModeT2_i7) {
|
||||
Offset += MI.getOperand(FrameRegIdx + 1).getImm();
|
||||
NumBits = 9; // 7 bits scaled by 4
|
||||
unsigned OffsetMask = 0x3;
|
||||
unsigned OffsetMask;
|
||||
switch (AddrMode) {
|
||||
case ARMII::AddrModeT2_i7s4: NumBits = 9; OffsetMask = 0x3; break;
|
||||
case ARMII::AddrModeT2_i7s2: NumBits = 8; OffsetMask = 0x1; break;
|
||||
default: NumBits = 7; OffsetMask = 0x0; break;
|
||||
}
|
||||
// MCInst operand expects already scaled value.
|
||||
Scale = 1;
|
||||
assert((Offset & OffsetMask) == 0 && "Can't encode this offset!");
|
||||
(void)OffsetMask; // squash unused-variable warning at -NDEBUG
|
||||
} else if (AddrMode == ARMII::AddrModeT2_i8s4) {
|
||||
Offset += MI.getOperand(FrameRegIdx + 1).getImm() * 4;
|
||||
NumBits = 10; // 8 bits scaled by 4
|
||||
NumBits = 8 + 2;
|
||||
// MCInst operand expects already scaled value.
|
||||
Scale = 1;
|
||||
assert((Offset & 3) == 0 && "Can't encode this offset!");
|
||||
|
||||
1944
test/MC/ARM/mve-load-store.s
Normal file
1944
test/MC/ARM/mve-load-store.s
Normal file
File diff suppressed because it is too large
Load Diff
1378
test/MC/Disassembler/ARM/mve-load-store.txt
Normal file
1378
test/MC/Disassembler/ARM/mve-load-store.txt
Normal file
File diff suppressed because it is too large
Load Diff
Loading…
Reference in New Issue
Block a user