xref: /freebsd/contrib/llvm-project/llvm/lib/Target/AArch64/SMEInstrFormats.td (revision 2e3507c25e42292b45a5482e116d278f5515d04d)
1//=-- SMEInstrFormats.td -  AArch64 SME Instruction classes -*- tablegen -*--=//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// AArch64 Scalable Matrix Extension (SME) Instruction Class Definitions.
10//
11//===----------------------------------------------------------------------===//
12
13def imm_to_tile8   : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAB0>", []>;
14def imm_to_tile16  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAH0>", []>;
15def imm_to_tile32  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAS0>", []>;
16def imm_to_tile64  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAD0>", []>;
17def imm_to_tile128 : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAQ0>", []>;
18
19def tileslice8   : ComplexPattern<i32 , 2, "SelectSMETileSlice<15, 1>", []>;
20def tileslice16  : ComplexPattern<i32 , 2, "SelectSMETileSlice<7,  1>", []>;
21def tileslice32  : ComplexPattern<i32 , 2, "SelectSMETileSlice<3,  1>", []>;
22def tileslice64  : ComplexPattern<i32 , 2, "SelectSMETileSlice<1,  1>", []>;
23def tileslice128 : ComplexPattern<i32 , 2, "SelectSMETileSlice<0,  1>", []>; // nop
24
25def tileslicerange3s2 : ComplexPattern<i32, 2, "SelectSMETileSlice<14, 2>", []>;
26def tileslicerange2s2 : ComplexPattern<i32, 2, "SelectSMETileSlice<6,  2>", []>;
27def tileslicerange1s2 : ComplexPattern<i32, 2, "SelectSMETileSlice<2,  2>", []>;
28def tileslicerange0s2 : ComplexPattern<i32, 2, "SelectSMETileSlice<0,  2>", []>;
29
30def tileslicerange2s4 : ComplexPattern<i32, 2, "SelectSMETileSlice<12, 4>", []>;
31def tileslicerange1s4 : ComplexPattern<i32, 2, "SelectSMETileSlice<4,  4>", []>;
32def tileslicerange0s4 : ComplexPattern<i32, 2, "SelectSMETileSlice<0,  4>", []>;
33
34def am_sme_indexed_b4 :ComplexPattern<iPTR, 2, "SelectAddrModeIndexedSVE<0,15>", [], [SDNPWantRoot]>;
35
36//===----------------------------------------------------------------------===//
37// SME Pseudo Classes
38//===----------------------------------------------------------------------===//
39
40def getSMEPseudoMap : InstrMapping {
41  let FilterClass = "SMEPseudo2Instr";
42  let RowFields = ["PseudoName"];
43  let ColFields = ["IsInstr"];
44  let KeyCol = ["0"];
45  let ValueCols = [["1"]];
46}
47
48class SMEPseudo2Instr<string name, bit instr> {
49  string PseudoName = name;
50  bit IsInstr = instr;
51}
52
53class sme_outer_product_pseudo<ZPRRegOp zpr_ty, SMEMatrixTypeEnum za_flag>
54    : Pseudo<(outs), (ins i32imm:$tile, PPR3bAny:$pn, PPR3bAny:$pm,
55                          zpr_ty:$zn, zpr_ty:$zm), []>,
56      Sched<[]> {
57  // Translated to the actual instructions in AArch64ISelLowering.cpp
58  let SMEMatrixType = za_flag;
59  let usesCustomInserter = 1;
60}
61
62class sme2_za_array_2op_multi_single_pseudo<string name, Operand index_ty, RegisterOperand multi_vector_ty,
63                                            ZPRRegOp zpr_ty, SMEMatrixTypeEnum za_flag>
64    : SMEPseudo2Instr<name, 0>,
65      Pseudo<(outs), (ins MatrixIndexGPR32Op8_11:$Rv, index_ty:$imm3, multi_vector_ty:$Zn, zpr_ty:$Zm), []> {
66  let SMEMatrixType = za_flag;
67  let usesCustomInserter = 1;
68}
69
70class sme2_za_array_2op_multi_multi_pseudo<string name, Operand index_ty, RegisterOperand multi_vector_ty,
71                                           SMEMatrixTypeEnum za_flag>
72    : SMEPseudo2Instr<name, 0>,
73      Pseudo<(outs), (ins MatrixIndexGPR32Op8_11:$Rv, index_ty:$imm3, multi_vector_ty:$Zn, multi_vector_ty:$Zm), []> {
74  let SMEMatrixType = za_flag;
75  let usesCustomInserter = 1;
76}
77
78class sme2_za_array_2op_multi_index_pseudo<string name, Operand index_ty, RegisterOperand multi_vector_ty,
79                                           ZPRRegOp zpr_ty, Operand imm_ty, SMEMatrixTypeEnum za_flag>
80    : SMEPseudo2Instr<name, 0>,
81      Pseudo<(outs), (ins MatrixIndexGPR32Op8_11:$Rv, index_ty:$imm3, multi_vector_ty:$Zn, zpr_ty:$Zm, imm_ty:$i), []> {
82  let SMEMatrixType = za_flag;
83  let usesCustomInserter = 1;
84}
85
86class sme2_move_to_za_pseudo<string name, Operand imm_ty, RegisterOperand multi_vector_ty, SMEMatrixTypeEnum za_flag>
87    : SMEPseudo2Instr<name, 0>,
88      Pseudo<(outs), (ins MatrixIndexGPR32Op8_11:$Rs, imm_ty:$imm, multi_vector_ty:$Zn), []> {
89  let SMEMatrixType = za_flag;
90  let usesCustomInserter = 1;
91}
92
93class sme2_move_to_tile_pseudo<string name, Operand tile_imm, Operand imm_ty, RegisterOperand multi_vector_ty, SMEMatrixTypeEnum za_flag>
94    : SMEPseudo2Instr<name, 0>,
95      Pseudo<(outs), (ins tile_imm:$tile, MatrixIndexGPR32Op12_15:$Rs, imm_ty:$imm, multi_vector_ty:$Zn), []> {
96  let SMEMatrixType = za_flag;
97  let usesCustomInserter = 1;
98}
99
100//===----------------------------------------------------------------------===//
101// SME pattern match helpers.
102//===----------------------------------------------------------------------===//
103
104class SME2_ZA_TwoOp_Multi_Single_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty,
105                                     ValueType vt, ComplexPattern tileslice>
106    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn, vt:$Zm),
107          (!cast<Instruction>(name # _PSEUDO) $base, $offset, vt:$Zn, zpr_ty:$Zm)>;
108
109
110class SME2_ZA_TwoOp_VG2_Multi_Single_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty,
111                                         ValueType vt, ComplexPattern tileslice>
112    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2, vt:$Zm),
113          (!cast<Instruction>(name # _PSEUDO) $base, $offset, (REG_SEQUENCE ZPR2, vt:$Zn1, zsub0, vt:$Zn2, zsub1),
114                                              zpr_ty:$Zm)>;
115class SME2_ZA_TwoOp_VG4_Multi_Single_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty,
116                                         ValueType vt, ComplexPattern tileslice>
117    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)),
118                     vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4, vt:$Zm),
119          (!cast<Instruction>(name # _PSEUDO) $base, $offset,
120                                              (REG_SEQUENCE ZPR4, vt:$Zn1, zsub0, vt:$Zn2, zsub1, vt:$Zn3, zsub2, vt:$Zn4, zsub3),
121                                              zpr_ty:$Zm)>;
122
123class SME2_ZA_TwoOp_VG2_Multi_Multi_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ValueType vt, ComplexPattern tileslice>
124    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2, vt:$Zm1, vt:$Zm2),
125          (!cast<Instruction>(name # _PSEUDO) $base, $offset,
126                                              (REG_SEQUENCE ZPR2Mul2, vt:$Zn1, zsub0, vt:$Zn2, zsub1),
127                                              (REG_SEQUENCE ZPR2Mul2, vt:$Zm1, zsub0, vt:$Zm2, zsub1))>;
128
129class SME2_ZA_TwoOp_VG4_Multi_Multi_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ValueType vt, ComplexPattern tileslice>
130    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)),
131                     vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4, vt:$Zm1, vt:$Zm2, vt:$Zm3, vt:$Zm4),
132          (!cast<Instruction>(name # _PSEUDO) $base, $offset,
133                                              (REG_SEQUENCE ZPR4Mul4, vt:$Zn1, zsub0, vt:$Zn2, zsub1, vt:$Zn3, zsub2, vt:$Zn4, zsub3),
134                                              (REG_SEQUENCE ZPR4Mul4, vt:$Zm1, zsub0, vt:$Zm2, zsub1, vt:$Zm3, zsub2, vt:$Zm4, zsub3))>;
135
136class SME2_ZA_TwoOp_Multi_Index_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty, ValueType vt,
137                                    Operand imm_ty, ComplexPattern tileslice>
138   : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn, vt:$Zm, (i32 imm_ty:$i)),
139         (!cast<Instruction>(name # _PSEUDO) $base, $offset, vt:$Zn, zpr_ty:$Zm, (i32 imm_ty:$i))>;
140
141
142class SME2_ZA_TwoOp_VG2_Multi_Index_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty, ValueType vt,
143                                        Operand imm_ty, ComplexPattern tileslice>
144    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2, vt:$Zm, (i32 imm_ty:$i)),
145          (!cast<Instruction>(name # _PSEUDO) $base, $offset,
146                                              (REG_SEQUENCE ZPR2Mul2, vt:$Zn1, zsub0, vt:$Zn2, zsub1), zpr_ty:$Zm, imm_ty:$i)>;
147
148class SME2_ZA_TwoOp_VG4_Multi_Index_Pat<string name, SDPatternOperator intrinsic, Operand index_ty, ZPRRegOp zpr_ty, ValueType vt,
149                                        Operand imm_ty, ComplexPattern tileslice>
150    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)),
151                     vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4, vt:$Zm, (i32 imm_ty:$i)),
152          (!cast<Instruction>(name # _PSEUDO) $base, $offset,
153                                              (REG_SEQUENCE ZPR4Mul4, vt:$Zn1, zsub0, vt:$Zn2, zsub1, vt:$Zn3, zsub2, vt:$Zn4, zsub3),
154                                              zpr_ty:$Zm, imm_ty:$i)>;
155
156class SME2_Sat_Shift_VG2_Pat<string name, SDPatternOperator intrinsic, ValueType out_vt, ValueType in_vt, Operand imm_ty>
157    : Pat<(out_vt (intrinsic in_vt:$Zn1, in_vt:$Zn2, (i32 imm_ty:$i))),
158                  (!cast<Instruction>(name) (REG_SEQUENCE ZPR2Mul2, in_vt:$Zn1, zsub0, in_vt:$Zn2, zsub1), imm_ty:$i)>;
159
160class SME2_Sat_Shift_VG4_Pat<string name, SDPatternOperator intrinsic, ValueType out_vt, ValueType in_vt, Operand imm_ty>
161    : Pat<(out_vt (intrinsic in_vt:$Zn1, in_vt:$Zn2, in_vt:$Zn3, in_vt:$Zn4, (i32 imm_ty:$i))),
162                  (!cast<Instruction>(name) (REG_SEQUENCE ZPR4Mul4, in_vt:$Zn1, zsub0, in_vt:$Zn2, zsub1, in_vt:$Zn3, zsub2, in_vt:$Zn4, zsub3),
163                                            imm_ty:$i)>;
164
165class SME2_Cvt_VG4_Pat<string name, SDPatternOperator intrinsic, ValueType out_vt, ValueType in_vt>
166    : Pat<(out_vt (intrinsic in_vt:$Zn1, in_vt:$Zn2, in_vt:$Zn3, in_vt:$Zn4)),
167                  (!cast<Instruction>(name) (REG_SEQUENCE ZPR4Mul4, in_vt:$Zn1, zsub0, in_vt:$Zn2, zsub1, in_vt:$Zn3, zsub2, in_vt:$Zn4, zsub3))>;
168
169class SME2_ZA_VG1x2_Multi_Pat<string name, SDPatternOperator intrinsic, ValueType vt, Operand index_ty, ComplexPattern tileslice>
170    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2),
171          (!cast<Instruction>(name # _PSEUDO) $base, $offset, (REG_SEQUENCE ZPR2Mul2, vt:$Zn1, zsub0, vt:$Zn2, zsub1))>;
172
173class SME2_ZA_VG1x4_Multi_Pat<string name, SDPatternOperator intrinsic, ValueType vt, Operand index_ty, ComplexPattern tileslice>
174    : Pat<(intrinsic (i32 (tileslice MatrixIndexGPR32Op8_11:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4),
175          (!cast<Instruction>(name # _PSEUDO) $base, $offset, (REG_SEQUENCE ZPR4Mul4, vt:$Zn1, zsub0, vt:$Zn2, zsub1, vt:$Zn3, zsub2, vt:$Zn4, zsub3))>;
176
177class SME2_Tile_VG2_Multi_Pat<string name, SDPatternOperator intrinsic, Operand tile_imm, ValueType vt, Operand index_ty, ComplexPattern tileslice>
178    : Pat<(intrinsic tile_imm:$tile, (i32 (tileslice MatrixIndexGPR32Op12_15:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2),
179          (!cast<Instruction>(name # _PSEUDO) $tile, $base, $offset, (REG_SEQUENCE ZPR2Mul2, vt:$Zn1, zsub0, vt:$Zn2, zsub1))>;
180
181class SME2_Tile_VG4_Multi_Pat<string name, SDPatternOperator intrinsic, Operand tile_imm, ValueType vt, Operand index_ty, ComplexPattern tileslice>
182    : Pat<(intrinsic tile_imm:$tile, (i32 (tileslice MatrixIndexGPR32Op12_15:$base, index_ty:$offset)), vt:$Zn1, vt:$Zn2, vt:$Zn3, vt:$Zn4),
183          (!cast<Instruction>(name # _PSEUDO) $tile, $base, $offset, (REG_SEQUENCE ZPR4Mul4, vt:$Zn1, zsub0, vt:$Zn2, zsub1, vt:$Zn3, zsub2, vt:$Zn4, zsub3))>;
184
185//===----------------------------------------------------------------------===//
186// SME pattern match helpers.
187//===----------------------------------------------------------------------===//
188
189class SME_ZA_Tile_TwoPred_TwoVec_Pat<string name, SDPatternOperator intrinsic, Operand imm_ty, ValueType pg_ty, ValueType vt>
190    : Pat<(intrinsic imm_ty:$tile, (pg_ty PPR3bAny:$Pn), (pg_ty PPR3bAny:$Pm), vt:$Zn, vt:$Zm),
191          (!cast<Instruction>(name # _PSEUDO) $tile, $Pn, $Pm, $Zn, $Zm)>;
192
193//===----------------------------------------------------------------------===//
194// SME Outer Products
195//===----------------------------------------------------------------------===//
196
197class sme_fp_outer_product_inst<bit S, bits<2> sz, bit op, MatrixTileOperand za_ty,
198                                ZPRRegOp zpr_ty, string mnemonic>
199    : I<(outs za_ty:$ZAda),
200      (ins za_ty:$_ZAda, PPR3bAny:$Pn, PPR3bAny:$Pm, zpr_ty:$Zn, zpr_ty:$Zm),
201        mnemonic, "\t$ZAda, $Pn/m, $Pm/m, $Zn, $Zm",
202        "", []>,
203      Sched<[]> {
204  bits<5> Zm;
205  bits<3> Pm;
206  bits<3> Pn;
207  bits<5> Zn;
208  let Inst{31-25} = 0b1000000;
209  let Inst{24}    = op;
210  let Inst{23}    = 0b1;
211  let Inst{22-21} = sz;
212  let Inst{20-16} = Zm;
213  let Inst{15-13} = Pm;
214  let Inst{12-10} = Pn;
215  let Inst{9-5}   = Zn;
216  let Inst{4}     = S;
217  let Inst{3}     = op;
218
219  let Constraints = "$ZAda = $_ZAda";
220}
221
222multiclass sme_outer_product_fp32<bit S, string mnemonic, SDPatternOperator op> {
223  def NAME : sme_fp_outer_product_inst<S, 0b00, 0b0, TileOp32, ZPR32, mnemonic>, SMEPseudo2Instr<NAME, 1> {
224    bits<2> ZAda;
225    let Inst{1-0} = ZAda;
226    let Inst{2}   = 0b0;
227  }
228
229  def NAME # _PSEUDO : sme_outer_product_pseudo<ZPR32, SMEMatrixTileS>, SMEPseudo2Instr<NAME, 0>;
230
231  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, op, timm32_0_3, nxv4i1, nxv4f32>;
232}
233
234multiclass sme_outer_product_fp64<bit S, string mnemonic, SDPatternOperator op> {
235  def NAME : sme_fp_outer_product_inst<S, 0b10, 0b0, TileOp64, ZPR64, mnemonic>, SMEPseudo2Instr<NAME, 1> {
236    bits<3> ZAda;
237    let Inst{2-0} = ZAda;
238  }
239
240  def NAME # _PSEUDO : sme_outer_product_pseudo<ZPR64, SMEMatrixTileD>, SMEPseudo2Instr<NAME, 0>;
241
242  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, op, timm32_0_7, nxv2i1, nxv2f64>;
243}
244
245multiclass sme2p1_fmop_tile_fp16<string mnemonic, bit bf, bit s>{
246  def NAME : sme_fp_outer_product_inst<s, {0,bf}, 0b1, TileOp16, ZPR16, mnemonic> {
247    bits<1> ZAda;
248    let Inst{2-1} = 0b00;
249    let Inst{0}   = ZAda;
250  }
251}
252
253class sme_int_outer_product_inst<bits<3> opc, bit sz, bit sme2,
254                                 MatrixTileOperand za_ty, ZPRRegOp zpr_ty,
255                                 string mnemonic>
256    : I<(outs za_ty:$ZAda),
257        (ins za_ty:$_ZAda, PPR3bAny:$Pn, PPR3bAny:$Pm, zpr_ty:$Zn, zpr_ty:$Zm),
258        mnemonic, "\t$ZAda, $Pn/m, $Pm/m, $Zn, $Zm",
259        "", []>,
260      Sched<[]> {
261  bits<5> Zm;
262  bits<3> Pm;
263  bits<3> Pn;
264  bits<5> Zn;
265  let Inst{31-25} = 0b1010000;
266  let Inst{24}    = opc{2}; // u0
267  let Inst{23}    = 0b1;
268  let Inst{22}    = sz;
269  let Inst{21}    = opc{1}; // u1
270  let Inst{20-16} = Zm;
271  let Inst{15-13} = Pm;
272  let Inst{12-10} = Pn;
273  let Inst{9-5}   = Zn;
274  let Inst{4}     = opc{0};  //S;
275  let Inst{3}     = sme2;
276
277  let Constraints = "$ZAda = $_ZAda";
278}
279
280multiclass sme_int_outer_product_i32<bits<3> opc, string mnemonic,
281                                     SDPatternOperator op> {
282  def NAME : sme_int_outer_product_inst<opc, 0b0, 0b0,  TileOp32,
283                                        ZPR8, mnemonic>, SMEPseudo2Instr<NAME, 1> {
284    bits<2> ZAda;
285    let Inst{1-0} = ZAda;
286    let Inst{2}   = 0b0;
287  }
288
289  def NAME # _PSEUDO : sme_outer_product_pseudo<ZPR8, SMEMatrixTileS>, SMEPseudo2Instr<NAME, 0>;
290
291  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, op, timm32_0_3, nxv16i1, nxv16i8>;
292}
293
294multiclass sme_int_outer_product_i64<bits<3> opc, string mnemonic,
295                                     SDPatternOperator op> {
296  def NAME : sme_int_outer_product_inst<opc, 0b1, 0b0, TileOp64,
297                                        ZPR16, mnemonic>, SMEPseudo2Instr<NAME, 1> {
298    bits<3> ZAda;
299    let Inst{2-0} = ZAda;
300  }
301
302  def NAME # _PSEUDO : sme_outer_product_pseudo<ZPR16, SMEMatrixTileD>, SMEPseudo2Instr<NAME, 0>;
303
304  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, op, timm32_0_7, nxv8i1, nxv8i16>;
305}
306
307class sme_outer_product_widening_inst<bits<3> opc, ZPRRegOp zpr_ty, string mnemonic>
308    : I<(outs TileOp32:$ZAda),
309        (ins  TileOp32:$_ZAda, PPR3bAny:$Pn, PPR3bAny:$Pm, zpr_ty:$Zn, zpr_ty:$Zm),
310        mnemonic, "\t$ZAda, $Pn/m, $Pm/m, $Zn, $Zm",
311        "", []>,
312      Sched<[]> {
313  bits<5> Zm;
314  bits<3> Pm;
315  bits<3> Pn;
316  bits<5> Zn;
317  bits<2> ZAda;
318  let Inst{31-25} = 0b1000000;
319  let Inst{24}    = !if(opc{2}, 0, 1);
320  let Inst{23-22} = 0b10;
321  let Inst{21}    = opc{1};
322  let Inst{20-16} = Zm;
323  let Inst{15-13} = Pm;
324  let Inst{12-10} = Pn;
325  let Inst{9-5}   = Zn;
326  let Inst{4}     = opc{0};
327  let Inst{3}     = opc{2};
328  let Inst{2}     = 0b0;
329  let Inst{1-0}   = ZAda;
330
331  let Constraints = "$ZAda = $_ZAda";
332}
333
334multiclass sme_bf16_outer_product<bits<3> opc, string mnemonic, SDPatternOperator op> {
335  def NAME : sme_outer_product_widening_inst<opc, ZPR16, mnemonic>, SMEPseudo2Instr<NAME, 1>;
336
337  def NAME # _PSEUDO : sme_outer_product_pseudo<ZPR16, SMEMatrixTileS>, SMEPseudo2Instr<NAME, 0>;
338
339  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, op, timm32_0_3, nxv8i1, nxv8bf16>;
340}
341
342multiclass sme_f16_outer_product<bits<3> opc, string mnemonic, SDPatternOperator op> {
343  def NAME : sme_outer_product_widening_inst<opc, ZPR16, mnemonic>, SMEPseudo2Instr<NAME, 1>;
344
345  def NAME # _PSEUDO : sme_outer_product_pseudo<ZPR16, SMEMatrixTileS>, SMEPseudo2Instr<NAME, 0>;
346
347  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, op, timm32_0_3, nxv8i1, nxv8f16>;
348}
349
350//===----------------------------------------------------------------------===//
351// SME Add Vector to Tile
352//===----------------------------------------------------------------------===//
353
354class sme_add_vector_to_tile_inst<bit op, bit V, MatrixTileOperand tile_ty,
355                                  ZPRRegOp zpr_ty, string mnemonic>
356    : I<(outs tile_ty:$ZAda),
357        (ins tile_ty:$_ZAda, PPR3bAny:$Pn, PPR3bAny:$Pm, zpr_ty:$Zn),
358        mnemonic, "\t$ZAda, $Pn/m, $Pm/m, $Zn",
359        "", []>, Sched<[]> {
360  bits<3> Pm;
361  bits<3> Pn;
362  bits<5> Zn;
363  let Inst{31-23} = 0b110000001;
364  let Inst{22}    = op;
365  let Inst{21-17} = 0b01000;
366  let Inst{16}    = V;
367  let Inst{15-13} = Pm;
368  let Inst{12-10} = Pn;
369  let Inst{9-5}   = Zn;
370  let Inst{4-3}   = 0b00;
371
372  let Constraints = "$ZAda = $_ZAda";
373}
374
375class sme_add_vector_to_tile_pseudo<ZPRRegOp zpr_ty, SMEMatrixTypeEnum za_flag>
376    : Pseudo<(outs),
377             (ins i32imm:$tile, PPR3bAny:$Pn, PPR3bAny:$Pm, zpr_ty:$Zn), []>,
378      Sched<[]> {
379  // Translated to the actual instructions in AArch64ISelLowering.cpp
380  let SMEMatrixType = za_flag;
381  let usesCustomInserter = 1;
382}
383
384multiclass sme_add_vector_to_tile_u32<bit V, string mnemonic, SDPatternOperator op> {
385    def NAME : sme_add_vector_to_tile_inst<0b0, V, TileOp32, ZPR32, mnemonic>, SMEPseudo2Instr<NAME, 1> {
386  bits<2> ZAda;
387  let Inst{2}   = 0b0;
388  let Inst{1-0} = ZAda;
389  }
390
391  def _PSEUDO_S : sme_add_vector_to_tile_pseudo<ZPR32, SMEMatrixTileS>, SMEPseudo2Instr<NAME, 0>;
392
393  def : Pat<(op timm32_0_3:$tile, (nxv4i1 PPR3bAny:$pn), (nxv4i1 PPR3bAny:$pm),
394            (nxv4i32 ZPR32:$zn)),
395          (!cast<Instruction>(NAME # _PSEUDO_S) timm32_0_3:$tile, $pn, $pm, $zn)>;
396}
397
398multiclass sme_add_vector_to_tile_u64<bit V, string mnemonic, SDPatternOperator op> {
399    def NAME : sme_add_vector_to_tile_inst<0b1, V, TileOp64, ZPR64, mnemonic>, SMEPseudo2Instr<NAME, 1> {
400  bits<3> ZAda;
401  let Inst{2-0} = ZAda;
402  }
403
404  def _PSEUDO_D : sme_add_vector_to_tile_pseudo<ZPR64, SMEMatrixTileD>, SMEPseudo2Instr<NAME, 0>;
405
406  let Predicates = [HasSMEI16I64] in {
407  def : Pat<(op timm32_0_7:$tile, (nxv2i1 PPR3bAny:$pn), (nxv2i1 PPR3bAny:$pm),
408                (nxv2i64 ZPR64:$zn)),
409            (!cast<Instruction>(NAME # _PSEUDO_D) timm32_0_7:$tile, $pn, $pm, $zn)>;
410  }
411}
412
413//===----------------------------------------------------------------------===//
414// SME Contiguous Loads
415//===----------------------------------------------------------------------===//
416
417class sme_mem_ld_ss_base<bit Q, bit V, bits<2> msz, dag outs, dag ins,
418                         string mnemonic, string argstr>
419    : I<outs, ins, mnemonic, argstr, "", []>, Sched<[]> {
420  bits<5> Rm;
421  bits<2> Rv;
422  bits<3> Pg;
423  bits<5> Rn;
424  let Inst{31-25} = 0b1110000;
425  let Inst{24}    = Q;
426  let Inst{23-22} = msz;
427  let Inst{21}    = 0b0;
428  let Inst{20-16} = Rm;
429  let Inst{15}    = V;
430  let Inst{14-13} = Rv;
431  let Inst{12-10} = Pg;
432  let Inst{9-5}   = Rn;
433  let Inst{4}     = 0b0;
434
435  let mayLoad = 1;
436}
437
438class sme_mem_ld_ss_inst<bit Q, bits<2> msz, string mnemonic,
439                         MatrixTileVectorOperand tile_ty, bit is_col,
440                         Operand imm_ty, RegisterOperand gpr_ty>
441    : sme_mem_ld_ss_base<
442        Q, is_col, msz, (outs tile_ty:$ZAt),
443        (ins MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm, PPR3bAny:$Pg, GPR64sp:$Rn,
444             gpr_ty:$Rm),
445        mnemonic, "\t\\{$ZAt[$Rv, $imm]\\}, $Pg/z, [$Rn, $Rm]">;
446
447multiclass sme_mem_ss_aliases_base<string mnemonic, Instruction inst,
448                                   MatrixTileVectorOperand tile_ty,
449                                   Operand imm_ty, RegisterOperand gpr_ty,
450                                   string pg_suffix=""> {
451  def : InstAlias<mnemonic # "\t$ZAt[$Rv, $imm], $Pg" # pg_suffix # ", [$Rn, $Rm]",
452                  (inst tile_ty:$ZAt, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm, PPR3bAny:$Pg, GPR64sp:$Rn, gpr_ty:$Rm), 0>;
453  // Default XZR offset aliases
454  def : InstAlias<mnemonic # "\t\\{$ZAt[$Rv, $imm]\\}, $Pg" # pg_suffix # ", [$Rn]",
455                  (inst tile_ty:$ZAt, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm, PPR3bAny:$Pg, GPR64sp:$Rn, XZR), 1>;
456  def : InstAlias<mnemonic # "\t$ZAt[$Rv, $imm], $Pg" # pg_suffix # ", [$Rn]",
457                  (inst tile_ty:$ZAt, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm, PPR3bAny:$Pg, GPR64sp:$Rn, XZR), 0>;
458}
459
460multiclass sme_mem_ss_aliases<string mnemonic, string inst, bit is_col,
461                              string pg_suffix=""> {
462  defm : sme_mem_ss_aliases_base<mnemonic # "b", !cast<Instruction>(inst # _B),
463                                 !if(is_col, TileVectorOpV8, TileVectorOpH8),
464                                 sme_elm_idx0_15, GPR64shifted8, pg_suffix>;
465  defm : sme_mem_ss_aliases_base<mnemonic # "h", !cast<Instruction>(inst # _H),
466                                 !if(is_col, TileVectorOpV16, TileVectorOpH16),
467                                 sme_elm_idx0_7, GPR64shifted16, pg_suffix>;
468  defm : sme_mem_ss_aliases_base<mnemonic # "w", !cast<Instruction>(inst # _S),
469                                 !if(is_col, TileVectorOpV32, TileVectorOpH32),
470                                 sme_elm_idx0_3, GPR64shifted32, pg_suffix>;
471  defm : sme_mem_ss_aliases_base<mnemonic # "d", !cast<Instruction>(inst # _D),
472                                 !if(is_col, TileVectorOpV64, TileVectorOpH64),
473                                 sme_elm_idx0_1, GPR64shifted64, pg_suffix>;
474  defm : sme_mem_ss_aliases_base<mnemonic # "q", !cast<Instruction>(inst # _Q),
475                                 !if(is_col, TileVectorOpV128, TileVectorOpH128),
476                                 sme_elm_idx0_0, GPR64shifted128, pg_suffix>;
477}
478
479multiclass sme_mem_ld_ss_aliases<string inst, bit is_col> {
480  defm NAME : sme_mem_ss_aliases<"ld1", inst, is_col, "/z">;
481}
482
483multiclass sme_mem_ld_ss_patterns<Instruction Inst, SDPatternOperator Load,
484                                  Operand tile_ty, Operand offset_ty,
485                                  ComplexPattern addr,
486                                  ComplexPattern tileslice> {
487  // base, tileslice
488  def : Pat<(Load PPR3bAny:$pg, GPR64sp:$base, tile_ty:$tile,
489                  (i32 (tileslice MatrixIndexGPR32Op12_15:$idx, offset_ty:$imm))),
490            (Inst tile_ty:$tile, $idx, $imm, $pg, $base, XZR)>;
491
492  // reg + reg, tileslice
493  let AddedComplexity = 1 in {
494    def : Pat<(Load PPR3bAny:$pg, (addr GPR64sp:$base, GPR64:$offset),
495                    tile_ty:$tile, (i32 (tileslice MatrixIndexGPR32Op12_15:$idx,
496                                              offset_ty:$imm))),
497              (Inst tile_ty:$tile, $idx, $imm, $pg, $base, $offset)>;
498  }
499}
500
501class sme_load_pseudo
502    : Pseudo<(outs), (ins i32imm:$tile, MatrixIndexGPR32Op12_15:$idx,
503                          i32imm:$imm, PPR3bAny:$pg, GPR64sp:$base, GPR64:$offset), []>,
504      Sched<[]> {
505  // Translated to the actual instructions in AArch64ISelLowering.cpp
506  let usesCustomInserter = 1;
507  let mayLoad = 1;
508}
509
510multiclass sme_mem_ld_v_ss<string mnemonic, bit is_col> {
511  def _B : sme_mem_ld_ss_inst<0b0, 0b00, mnemonic # "b",
512                              !if(is_col, TileVectorOpV8, TileVectorOpH8),
513                              is_col, sme_elm_idx0_15, GPR64shifted8> {
514    bits<4> imm;
515    let Inst{3-0} = imm;
516  }
517  def _H : sme_mem_ld_ss_inst<0b0, 0b01, mnemonic # "h",
518                              !if(is_col, TileVectorOpV16, TileVectorOpH16),
519                              is_col, sme_elm_idx0_7, GPR64shifted16> {
520    bits<1> ZAt;
521    bits<3> imm;
522    let Inst{3}   = ZAt;
523    let Inst{2-0} = imm;
524  }
525  def _S : sme_mem_ld_ss_inst<0b0, 0b10, mnemonic # "w",
526                              !if(is_col, TileVectorOpV32, TileVectorOpH32),
527                              is_col, sme_elm_idx0_3, GPR64shifted32> {
528    bits<2> ZAt;
529    bits<2> imm;
530    let Inst{3-2} = ZAt;
531    let Inst{1-0} = imm;
532  }
533  def _D : sme_mem_ld_ss_inst<0b0, 0b11, mnemonic # "d",
534                              !if(is_col, TileVectorOpV64, TileVectorOpH64),
535                              is_col, sme_elm_idx0_1, GPR64shifted64> {
536    bits<3> ZAt;
537    bits<1> imm;
538    let Inst{3-1} = ZAt;
539    let Inst{0}   = imm;
540  }
541  def _Q : sme_mem_ld_ss_inst<0b1, 0b11, mnemonic # "q",
542                              !if(is_col, TileVectorOpV128, TileVectorOpH128),
543                              is_col, sme_elm_idx0_0, GPR64shifted128> {
544    bits<4> ZAt;
545    let Inst{3-0} = ZAt;
546  }
547
548  defm : sme_mem_ld_ss_aliases<NAME, is_col>;
549
550  // Pseudo instructions for lowering intrinsics, using immediates instead of
551  // tile registers.
552  def _PSEUDO_B : sme_load_pseudo;
553  def _PSEUDO_H : sme_load_pseudo;
554  def _PSEUDO_S : sme_load_pseudo;
555  def _PSEUDO_D : sme_load_pseudo;
556  def _PSEUDO_Q : sme_load_pseudo;
557
558  defm : sme_mem_ld_ss_patterns<!cast<Instruction>(NAME # _PSEUDO_B),
559                                !if(is_col, int_aarch64_sme_ld1b_vert,
560                                            int_aarch64_sme_ld1b_horiz),
561                                sme_elm_idx0_0, timm32_0_15, am_sve_regreg_lsl0,
562                                tileslice8>;
563  defm : sme_mem_ld_ss_patterns<!cast<Instruction>(NAME # _PSEUDO_H),
564                                !if(is_col, int_aarch64_sme_ld1h_vert,
565                                            int_aarch64_sme_ld1h_horiz),
566                                timm32_0_1, timm32_0_7, am_sve_regreg_lsl1,
567                                tileslice16>;
568  defm : sme_mem_ld_ss_patterns<!cast<Instruction>(NAME # _PSEUDO_S),
569                                !if(is_col, int_aarch64_sme_ld1w_vert,
570                                            int_aarch64_sme_ld1w_horiz),
571                                timm32_0_3, timm32_0_3, am_sve_regreg_lsl2,
572                                tileslice32>;
573  defm : sme_mem_ld_ss_patterns<!cast<Instruction>(NAME # _PSEUDO_D),
574                                !if(is_col, int_aarch64_sme_ld1d_vert,
575                                            int_aarch64_sme_ld1d_horiz),
576                                timm32_0_7, timm32_0_1, am_sve_regreg_lsl3,
577                                tileslice64>;
578  defm : sme_mem_ld_ss_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
579                                !if(is_col, int_aarch64_sme_ld1q_vert,
580                                            int_aarch64_sme_ld1q_horiz),
581                                timm32_0_15, sme_elm_idx0_0, am_sve_regreg_lsl4,
582                                tileslice128>;
583}
584
585multiclass sme_mem_ld_ss<string mnemonic> {
586  defm _H : sme_mem_ld_v_ss<mnemonic, /*is_col=*/0b0>;
587  defm _V : sme_mem_ld_v_ss<mnemonic, /*is_col=*/0b1>;
588}
589
590//===----------------------------------------------------------------------===//
591// SME Contiguous Stores
592//===----------------------------------------------------------------------===//
593
594class sme_mem_st_ss_base<bit Q, bit V, bits<2> msz, dag ins,
595                         string mnemonic, string argstr>
596    : I<(outs), ins, mnemonic, argstr, "", []>, Sched<[]> {
597  bits<5> Rm;
598  bits<2> Rv;
599  bits<3> Pg;
600  bits<5> Rn;
601  let Inst{31-25} = 0b1110000;
602  let Inst{24}    = Q;
603  let Inst{23-22} = msz;
604  let Inst{21}    = 0b1;
605  let Inst{20-16} = Rm;
606  let Inst{15}    = V;
607  let Inst{14-13} = Rv;
608  let Inst{12-10} = Pg;
609  let Inst{9-5}   = Rn;
610  let Inst{4}     = 0b0;
611
612  let mayStore = 1;
613  let hasSideEffects = 1;
614}
615
616class sme_mem_st_ss_inst<bit Q, bits<2> msz, string mnemonic,
617                         MatrixTileVectorOperand tile_ty, bit is_col,
618                         Operand imm_ty, RegisterOperand gpr_ty>
619    : sme_mem_st_ss_base<
620        Q, is_col, msz,
621        (ins tile_ty:$ZAt, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm, PPR3bAny:$Pg,
622             GPR64sp:$Rn, gpr_ty:$Rm),
623        mnemonic, "\t\\{$ZAt[$Rv, $imm]\\}, $Pg, [$Rn, $Rm]">;
624
625multiclass sme_mem_st_ss_aliases<string inst, bit is_col> {
626  defm NAME : sme_mem_ss_aliases<"st1", inst, is_col>;
627}
628
629multiclass sme_mem_st_ss_patterns<Instruction Inst, SDPatternOperator Store,
630                                  Operand offset_ty,
631                                  ComplexPattern imm2tile,
632                                  ComplexPattern addr,
633                                  ComplexPattern tileslice> {
634  // base, tileslice
635  def : Pat<(Store PPR3bAny:$pg, GPR64sp:$base, (imm2tile untyped:$tile),
636                   (i32 (tileslice MatrixIndexGPR32Op12_15:$idx, offset_ty:$imm))),
637            (Inst $tile, $idx, $imm, $pg, $base, XZR)>;
638
639  // reg + reg, tileslice
640  let AddedComplexity = 1 in {
641    def : Pat<(Store PPR3bAny:$pg, (addr GPR64sp:$base, GPR64:$offset),
642                     (imm2tile untyped:$tile),
643                     (i32 (tileslice MatrixIndexGPR32Op12_15:$idx, offset_ty:$imm))),
644              (Inst $tile, $idx, $imm, $pg, $base, $offset)>;
645  }
646}
647
648multiclass sme_mem_st_v_ss<string mnemonic, bit is_col> {
649  def _B : sme_mem_st_ss_inst<0b0, 0b00, mnemonic # "b",
650                              !if(is_col, TileVectorOpV8, TileVectorOpH8),
651                              is_col, sme_elm_idx0_15, GPR64shifted8> {
652    bits<4> imm;
653    let Inst{3-0} = imm;
654  }
655  def _H : sme_mem_st_ss_inst<0b0, 0b01, mnemonic # "h",
656                              !if(is_col, TileVectorOpV16, TileVectorOpH16),
657                              is_col, sme_elm_idx0_7, GPR64shifted16> {
658    bits<1> ZAt;
659    bits<3> imm;
660    let Inst{3}   = ZAt;
661    let Inst{2-0} = imm;
662  }
663  def _S : sme_mem_st_ss_inst<0b0, 0b10, mnemonic # "w",
664                              !if(is_col, TileVectorOpV32, TileVectorOpH32),
665                              is_col, sme_elm_idx0_3, GPR64shifted32> {
666    bits<2> ZAt;
667    bits<2> imm;
668    let Inst{3-2} = ZAt;
669    let Inst{1-0} = imm;
670  }
671  def _D : sme_mem_st_ss_inst<0b0, 0b11, mnemonic # "d",
672                              !if(is_col, TileVectorOpV64, TileVectorOpH64),
673                              is_col, sme_elm_idx0_1, GPR64shifted64> {
674    bits<3> ZAt;
675    bits<1> imm;
676    let Inst{3-1} = ZAt;
677    let Inst{0}   = imm;
678  }
679  def _Q : sme_mem_st_ss_inst<0b1, 0b11, mnemonic # "q",
680                              !if(is_col, TileVectorOpV128, TileVectorOpH128),
681                              is_col, sme_elm_idx0_0, GPR64shifted128> {
682    bits<4> ZAt;
683    let Inst{3-0} = ZAt;
684  }
685
686  defm : sme_mem_st_ss_aliases<NAME, is_col>;
687
688  defm : sme_mem_st_ss_patterns<!cast<Instruction>(NAME # _B),
689                                !if(is_col, int_aarch64_sme_st1b_vert,
690                                            int_aarch64_sme_st1b_horiz),
691                                timm32_0_15, imm_to_tile8, am_sve_regreg_lsl0,
692                                tileslice8>;
693  defm : sme_mem_st_ss_patterns<!cast<Instruction>(NAME # _H),
694                                !if(is_col, int_aarch64_sme_st1h_vert,
695                                            int_aarch64_sme_st1h_horiz),
696                                timm32_0_7, imm_to_tile16, am_sve_regreg_lsl1,
697                                tileslice16>;
698  defm : sme_mem_st_ss_patterns<!cast<Instruction>(NAME # _S),
699                                !if(is_col, int_aarch64_sme_st1w_vert,
700                                            int_aarch64_sme_st1w_horiz),
701                                timm32_0_3, imm_to_tile32, am_sve_regreg_lsl2,
702                                tileslice32>;
703  defm : sme_mem_st_ss_patterns<!cast<Instruction>(NAME # _D),
704                                !if(is_col, int_aarch64_sme_st1d_vert,
705                                            int_aarch64_sme_st1d_horiz),
706                                timm32_0_1, imm_to_tile64, am_sve_regreg_lsl3,
707                                tileslice64>;
708  defm : sme_mem_st_ss_patterns<!cast<Instruction>(NAME # _Q),
709                                !if(is_col, int_aarch64_sme_st1q_vert,
710                                            int_aarch64_sme_st1q_horiz),
711                                sme_elm_idx0_0, imm_to_tile128,
712                                am_sve_regreg_lsl4, tileslice128>;
713}
714
715multiclass sme_mem_st_ss<string mnemonic> {
716  defm _H : sme_mem_st_v_ss<mnemonic, /*is_col=*/0b0>;
717  defm _V : sme_mem_st_v_ss<mnemonic, /*is_col=*/0b1>;
718}
719
720//===----------------------------------------------------------------------===//
721// SME Save and Restore Array
722//===----------------------------------------------------------------------===//
723
724class sme_spill_fill_base<bit isStore, dag outs, dag ins, string opcodestr>
725    : I<outs, ins, opcodestr, "\t$ZAt[$Rv, $imm4], [$Rn, $offset, mul vl]", "",
726        []>,
727      Sched<[]> {
728  bits<2> Rv;
729  bits<5> Rn;
730  bits<4> imm4;
731  let Inst{31-22} = 0b1110000100;
732  let Inst{21}    = isStore;
733  let Inst{20-15} = 0b000000;
734  let Inst{14-13} = Rv;
735  let Inst{12-10} = 0b000;
736  let Inst{9-5}   = Rn;
737  let Inst{4}     = 0b0;
738  let Inst{3-0}   = imm4;
739}
740
741let mayStore = 1 in
742class sme_spill_inst<string opcodestr>
743    : sme_spill_fill_base<0b1, (outs),
744                          (ins MatrixOp:$ZAt, MatrixIndexGPR32Op12_15:$Rv,
745                               sme_elm_idx0_15:$imm4, GPR64sp:$Rn,
746                               imm0_15:$offset),
747                          opcodestr>;
748let mayLoad = 1 in
749class sme_fill_inst<string opcodestr>
750    : sme_spill_fill_base<0b0, (outs MatrixOp:$ZAt),
751                          (ins MatrixIndexGPR32Op12_15:$Rv,
752                               sme_elm_idx0_15:$imm4, GPR64sp:$Rn,
753                               imm0_15:$offset),
754                          opcodestr>;
755multiclass sme_spill<string opcodestr> {
756  def NAME : sme_spill_inst<opcodestr>;
757  def : InstAlias<opcodestr # "\t$ZAt[$Rv, $imm4], [$Rn]",
758                  (!cast<Instruction>(NAME) MatrixOp:$ZAt,
759                   MatrixIndexGPR32Op12_15:$Rv, sme_elm_idx0_15:$imm4, GPR64sp:$Rn, 0), 1>;
760  // base
761  def : Pat<(int_aarch64_sme_str MatrixIndexGPR32Op12_15:$idx, GPR64sp:$base),
762            (!cast<Instruction>(NAME) ZA, $idx, 0, $base, 0)>;
763}
764
765multiclass sme_fill<string opcodestr> {
766  def NAME : sme_fill_inst<opcodestr>;
767  def : InstAlias<opcodestr # "\t$ZAt[$Rv, $imm4], [$Rn]",
768                  (!cast<Instruction>(NAME) MatrixOp:$ZAt,
769                   MatrixIndexGPR32Op12_15:$Rv, sme_elm_idx0_15:$imm4, GPR64sp:$Rn, 0), 1>;
770  def NAME # _PSEUDO
771      : Pseudo<(outs),
772               (ins MatrixIndexGPR32Op12_15:$idx, imm0_15:$imm4,
773                    GPR64sp:$base), []>,
774        Sched<[]> {
775    // Translated to actual instruction in AArch64ISelLowering.cpp
776    let usesCustomInserter = 1;
777    let mayLoad = 1;
778  }
779  // base
780  def : Pat<(int_aarch64_sme_ldr MatrixIndexGPR32Op12_15:$idx, GPR64sp:$base),
781            (!cast<Instruction>(NAME # _PSEUDO) $idx, 0, $base)>;
782}
783
784//===----------------------------------------------------------------------===//
785// Move instructions
786//===----------------------------------------------------------------------===//
787
788class sme_vector_to_tile_base<bit Q, bit V, bits<2> sz, dag outs, dag ins,
789                              string mnemonic, string argstr>
790    : I<outs, ins, mnemonic, argstr, "", []>, Sched<[]> {
791  bits<2> Rv;
792  bits<3> Pg;
793  bits<5> Zn;
794  let Inst{31-24} = 0b11000000;
795  let Inst{23-22} = sz;
796  let Inst{21-17} = 0b00000;
797  let Inst{16}    = Q;
798  let Inst{15}    = V;
799  let Inst{14-13} = Rv;
800  let Inst{12-10} = Pg;
801  let Inst{9-5}   = Zn;
802  let Inst{4}     = 0b0;
803}
804
805class sme_vector_to_tile_inst<bit Q, bits<2> sz, MatrixTileVectorOperand tile_ty,
806                              bit is_col, Operand imm_ty, ZPRRegOp zpr_ty,
807                              string mnemonic>
808    : sme_vector_to_tile_base<Q, is_col, sz, (outs tile_ty:$ZAd),
809        (ins tile_ty:$_ZAd, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm, PPR3bAny:$Pg, zpr_ty:$Zn),
810        mnemonic, "\t$ZAd[$Rv, $imm], $Pg/m, $Zn">{
811
812  let Constraints = "$ZAd = $_ZAd";
813}
814
815
816multiclass sme_vector_to_tile_aliases<Instruction inst,
817                                      MatrixTileVectorOperand tile_ty,
818                                      ZPRRegOp zpr_ty, Operand imm_ty> {
819  def : InstAlias<"mov\t$ZAd[$Rv, $imm], $Pg/m, $Zn",
820                  (inst tile_ty:$ZAd, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm, PPR3bAny:$Pg, zpr_ty:$Zn), 1>;
821}
822
823multiclass sme_vector_to_tile_patterns<Instruction inst, ValueType zpr_vt,
824                                       ValueType ppr_vt, Operand imm_ty,
825                                       Operand offset_ty,
826                                       SDPatternOperator op,
827                                       ComplexPattern tileslice> {
828  def : Pat<(op imm_ty:$tile, (i32 (tileslice MatrixIndexGPR32Op12_15:$idx,
829                                              offset_ty:$imm)),
830                (ppr_vt PPR3bAny:$pg), (zpr_vt ZPRAny:$zn)),
831            (inst imm_ty:$tile, $idx, $imm, $pg, $zn)>;
832}
833
834class sme_mova_insert_pseudo<SMEMatrixTypeEnum za_flag>
835    : Pseudo<(outs), (ins i32imm:$tile, MatrixIndexGPR32Op12_15:$idx,
836                          i32imm:$imm, PPR3bAny:$pg, ZPRAny:$zn), []>,
837      Sched<[]> {
838  // Translated to the actual instructions in AArch64ISelLowering.cpp
839  let SMEMatrixType = za_flag;
840  let usesCustomInserter = 1;
841}
842
843multiclass sme_vector_v_to_tile<string mnemonic, bit is_col> {
844  def _B : sme_vector_to_tile_inst<0b0, 0b00, !if(is_col, TileVectorOpV8,
845                                                          TileVectorOpH8),
846                                   is_col, sme_elm_idx0_15, ZPR8, mnemonic>,
847                                   SMEPseudo2Instr<NAME # _B, 1> {
848    bits<4> imm;
849    let Inst{3-0} = imm;
850  }
851  def _H : sme_vector_to_tile_inst<0b0, 0b01, !if(is_col, TileVectorOpV16,
852                                                          TileVectorOpH16),
853                                   is_col, sme_elm_idx0_7, ZPR16, mnemonic>,
854                                   SMEPseudo2Instr<NAME # _H, 1> {
855    bits<1> ZAd;
856    bits<3> imm;
857    let Inst{3}   = ZAd;
858    let Inst{2-0} = imm;
859  }
860  def _S : sme_vector_to_tile_inst<0b0, 0b10, !if(is_col, TileVectorOpV32,
861                                                          TileVectorOpH32),
862                                   is_col, sme_elm_idx0_3, ZPR32, mnemonic>,
863                                   SMEPseudo2Instr<NAME # _S, 1> {
864    bits<2> ZAd;
865    bits<2> imm;
866    let Inst{3-2} = ZAd;
867    let Inst{1-0} = imm;
868  }
869  def _D : sme_vector_to_tile_inst<0b0, 0b11, !if(is_col, TileVectorOpV64,
870                                                          TileVectorOpH64),
871                                   is_col, sme_elm_idx0_1, ZPR64, mnemonic>,
872                                   SMEPseudo2Instr<NAME # _D, 1> {
873    bits<3> ZAd;
874    bits<1> imm;
875    let Inst{3-1} = ZAd;
876    let Inst{0}   = imm;
877  }
878  def _Q : sme_vector_to_tile_inst<0b1, 0b11, !if(is_col, TileVectorOpV128,
879                                                          TileVectorOpH128),
880                                   is_col, sme_elm_idx0_0, ZPR128, mnemonic>,
881                                   SMEPseudo2Instr<NAME # _Q, 1> {
882    bits<4> ZAd;
883    bits<1> imm;
884    let Inst{3-0} = ZAd;
885  }
886
887  // Pseudo instructions for lowering intrinsics, using immediates instead of
888  // tile registers.
889  def _PSEUDO_B : sme_mova_insert_pseudo<SMEMatrixTileB>, SMEPseudo2Instr<NAME # _B, 0>;
890  def _PSEUDO_H : sme_mova_insert_pseudo<SMEMatrixTileH>, SMEPseudo2Instr<NAME # _H, 0>;
891  def _PSEUDO_S : sme_mova_insert_pseudo<SMEMatrixTileS>, SMEPseudo2Instr<NAME # _S, 0>;
892  def _PSEUDO_D : sme_mova_insert_pseudo<SMEMatrixTileD>, SMEPseudo2Instr<NAME # _D, 0>;
893  def _PSEUDO_Q : sme_mova_insert_pseudo<SMEMatrixTileQ>, SMEPseudo2Instr<NAME # _Q, 0>;
894
895  defm : sme_vector_to_tile_aliases<!cast<Instruction>(NAME # _B),
896                                    !if(is_col, TileVectorOpV8,
897                                                TileVectorOpH8),
898                                    ZPR8, sme_elm_idx0_15>;
899  defm : sme_vector_to_tile_aliases<!cast<Instruction>(NAME # _H),
900                                    !if(is_col, TileVectorOpV16,
901                                                TileVectorOpH16),
902                                    ZPR16, sme_elm_idx0_7>;
903  defm : sme_vector_to_tile_aliases<!cast<Instruction>(NAME # _S),
904                                    !if(is_col, TileVectorOpV32,
905                                                TileVectorOpH32),
906                                    ZPR32, sme_elm_idx0_3>;
907  defm : sme_vector_to_tile_aliases<!cast<Instruction>(NAME # _D),
908                                    !if(is_col, TileVectorOpV64,
909                                                TileVectorOpH64),
910                                    ZPR64, sme_elm_idx0_1>;
911  defm : sme_vector_to_tile_aliases<!cast<Instruction>(NAME # _Q),
912                                    !if(is_col, TileVectorOpV128,
913                                                TileVectorOpH128),
914                                    ZPR128, sme_elm_idx0_0>;
915
916  defvar op = !if(is_col, int_aarch64_sme_write_vert,
917                          int_aarch64_sme_write_horiz);
918
919  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_B),
920                                     nxv16i8, nxv16i1, sme_elm_idx0_0, sme_elm_idx0_15,
921                                     op, tileslice8>;
922  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_H),
923                                     nxv8i16, nxv8i1, sme_elm_idx0_1, sme_elm_idx0_7,
924                                     op, tileslice16>;
925  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_H),
926                                     nxv8f16, nxv8i1, sme_elm_idx0_1, sme_elm_idx0_7,
927                                     op, tileslice16>;
928  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_H),
929                                     nxv8bf16, nxv8i1, sme_elm_idx0_1, sme_elm_idx0_7,
930                                     op, tileslice16>;
931  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_S),
932                                     nxv4i32, nxv4i1, sme_elm_idx0_3, sme_elm_idx0_3,
933                                     op, tileslice32>;
934  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_S),
935                                     nxv4f32, nxv4i1, sme_elm_idx0_3, sme_elm_idx0_3,
936                                     op, tileslice32>;
937  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_D),
938                                     nxv2i64, nxv2i1, sme_elm_idx0_7, sme_elm_idx0_1,
939                                     op, tileslice64>;
940  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_D),
941                                     nxv2f64, nxv2i1, sme_elm_idx0_7, sme_elm_idx0_1,
942                                     op, tileslice64>;
943
944  defvar opq = !if(is_col, int_aarch64_sme_writeq_vert,
945                           int_aarch64_sme_writeq_horiz);
946
947  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
948                                     nxv16i8, nxv16i1, sme_elm_idx0_15,
949                                     sme_elm_idx0_0, opq, tileslice128>;
950  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
951                                     nxv8i16, nxv8i1, sme_elm_idx0_15,
952                                     sme_elm_idx0_0, opq, tileslice128>;
953  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
954                                     nxv8f16, nxv8i1, sme_elm_idx0_15,
955                                     sme_elm_idx0_0, opq, tileslice128>;
956  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
957                                     nxv8bf16, nxv8i1, sme_elm_idx0_15,
958                                     sme_elm_idx0_0, opq, tileslice128>;
959  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
960                                     nxv4i32, nxv4i1, sme_elm_idx0_15,
961                                     sme_elm_idx0_0, opq, tileslice128>;
962  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
963                                     nxv4f32, nxv4i1, sme_elm_idx0_15,
964                                     sme_elm_idx0_0, opq, tileslice128>;
965  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
966                                     nxv2i64, nxv2i1, sme_elm_idx0_15,
967                                     sme_elm_idx0_0, opq, tileslice128>;
968  defm : sme_vector_to_tile_patterns<!cast<Instruction>(NAME # _PSEUDO_Q),
969                                     nxv2f64, nxv2i1, sme_elm_idx0_15,
970                                     sme_elm_idx0_0, opq, tileslice128>;
971}
972
973multiclass sme_vector_to_tile<string mnemonic> {
974  defm _H : sme_vector_v_to_tile<mnemonic, /*is_col=*/0b0>;
975  defm _V : sme_vector_v_to_tile<mnemonic, /*is_col=*/0b1>;
976}
977
978class sme_tile_to_vector_base<bit Q, bit V, bits<2> sz, dag outs, dag ins,
979                              string mnemonic, string argstr>
980    : I<outs, ins, mnemonic, argstr, "", []>, Sched<[]> {
981  bits<2> Rv;
982  bits<3> Pg;
983  bits<5> Zd;
984  let Inst{31-24} = 0b11000000;
985  let Inst{23-22} = sz;
986  let Inst{21-17} = 0b00001;
987  let Inst{16}    = Q;
988  let Inst{15}    = V;
989  let Inst{14-13} = Rv;
990  let Inst{12-10} = Pg;
991  let Inst{9}     = 0b0;
992  let Inst{4-0}   = Zd;
993}
994
995class sme_tile_to_vector_inst<bit Q, bits<2> sz, ZPRRegOp zpr_ty,
996                              MatrixTileVectorOperand tile_ty,
997                              bit is_col, Operand imm_ty, string mnemonic>
998    : sme_tile_to_vector_base<Q, is_col, sz, (outs zpr_ty:$Zd),
999        (ins zpr_ty:$_Zd, PPR3bAny:$Pg, tile_ty:$ZAn, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm),
1000        mnemonic, "\t$Zd, $Pg/m, $ZAn[$Rv, $imm]"> {
1001
1002  let Constraints = "$Zd = $_Zd";
1003}
1004
1005multiclass sme_tile_to_vector_aliases<Instruction inst, ZPRRegOp zpr_ty,
1006                                      MatrixTileVectorOperand tile_ty,
1007                                      Operand imm_ty > {
1008  def : InstAlias<"mov\t$Zd, $Pg/m, $ZAn[$Rv, $imm]",
1009                  (inst zpr_ty:$Zd, PPR3bAny:$Pg, tile_ty:$ZAn, MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm), 1>;
1010}
1011
1012multiclass sme_tile_to_vector_patterns<Instruction inst, ValueType zpr_vt,
1013                                       ValueType ppr_vt, Operand offset_ty,
1014                                       ComplexPattern imm2tile,
1015                                       ComplexPattern tileslice,
1016                                       SDPatternOperator op> {
1017  def : Pat<(zpr_vt (op (zpr_vt ZPRAny:$passthru), (ppr_vt PPR3bAny:$pg),
1018                        (imm2tile untyped:$tile), MatrixIndexGPR32Op12_15:$idx)),
1019            (inst $passthru, $pg, $tile, $idx, 0)>;
1020  let AddedComplexity = 1 in {
1021    def : Pat<(zpr_vt (op (zpr_vt ZPRAny:$passthru), (ppr_vt PPR3bAny:$pg),
1022                          (imm2tile untyped:$tile),
1023                          (i32 (tileslice MatrixIndexGPR32Op12_15:$idx,
1024                                          offset_ty:$imm)))),
1025              (inst $passthru, $pg, $tile, $idx, $imm)>;
1026  }
1027}
1028
1029multiclass sme_tile_to_vector_v<string mnemonic, bit is_col> {
1030  def _B : sme_tile_to_vector_inst<0b0, 0b00, ZPR8, !if(is_col, TileVectorOpV8,
1031                                                                TileVectorOpH8),
1032                                   is_col, sme_elm_idx0_15, mnemonic> {
1033    bits<4> imm;
1034    let Inst{8-5} = imm;
1035  }
1036  def _H : sme_tile_to_vector_inst<0b0, 0b01, ZPR16, !if(is_col, TileVectorOpV16,
1037                                                                 TileVectorOpH16),
1038                                   is_col, sme_elm_idx0_7, mnemonic> {
1039    bits<1> ZAn;
1040    bits<3> imm;
1041    let Inst{8}   = ZAn;
1042    let Inst{7-5} = imm;
1043  }
1044  def _S : sme_tile_to_vector_inst<0b0, 0b10, ZPR32, !if(is_col, TileVectorOpV32,
1045                                                                 TileVectorOpH32),
1046                                   is_col, sme_elm_idx0_3, mnemonic> {
1047    bits<2> ZAn;
1048    bits<2> imm;
1049    let Inst{8-7} = ZAn;
1050    let Inst{6-5} = imm;
1051  }
1052  def _D : sme_tile_to_vector_inst<0b0, 0b11, ZPR64, !if(is_col, TileVectorOpV64,
1053                                                                 TileVectorOpH64),
1054                                   is_col, sme_elm_idx0_1, mnemonic> {
1055    bits<3> ZAn;
1056    bits<1> imm;
1057    let Inst{8-6} = ZAn;
1058    let Inst{5}   = imm;
1059  }
1060  def _Q : sme_tile_to_vector_inst<0b1, 0b11, ZPR128, !if(is_col, TileVectorOpV128,
1061                                                                  TileVectorOpH128),
1062                                   is_col, sme_elm_idx0_0, mnemonic> {
1063    bits<4> ZAn;
1064    let Inst{8-5} = ZAn;
1065  }
1066
1067  defm : sme_tile_to_vector_aliases<!cast<Instruction>(NAME # _B), ZPR8,
1068                                    !if(is_col, TileVectorOpV8,
1069                                                TileVectorOpH8), sme_elm_idx0_15>;
1070  defm : sme_tile_to_vector_aliases<!cast<Instruction>(NAME # _H), ZPR16,
1071                                    !if(is_col, TileVectorOpV16,
1072                                                TileVectorOpH16), sme_elm_idx0_7>;
1073  defm : sme_tile_to_vector_aliases<!cast<Instruction>(NAME # _S), ZPR32,
1074                                    !if(is_col, TileVectorOpV32,
1075                                                TileVectorOpH32), sme_elm_idx0_3>;
1076  defm : sme_tile_to_vector_aliases<!cast<Instruction>(NAME # _D), ZPR64,
1077                                    !if(is_col, TileVectorOpV64,
1078                                                TileVectorOpH64), sme_elm_idx0_1>;
1079  defm : sme_tile_to_vector_aliases<!cast<Instruction>(NAME # _Q), ZPR128,
1080                                    !if(is_col, TileVectorOpV128,
1081                                                TileVectorOpH128), sme_elm_idx0_0>;
1082
1083  defvar op = !if(is_col, int_aarch64_sme_read_vert,
1084                          int_aarch64_sme_read_horiz);
1085
1086  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _B),
1087                                     nxv16i8, nxv16i1, sme_elm_idx0_15,
1088                                     imm_to_tile8, tileslice8, op>;
1089  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _H),
1090                                     nxv8i16, nxv8i1, sme_elm_idx0_7,
1091                                     imm_to_tile16, tileslice16, op>;
1092  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _H),
1093                                     nxv8f16, nxv8i1, sme_elm_idx0_7,
1094                                     imm_to_tile16, tileslice16, op>;
1095  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _H),
1096                                     nxv8bf16, nxv8i1, sme_elm_idx0_7,
1097                                     imm_to_tile16, tileslice16, op>;
1098  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _S),
1099                                     nxv4i32, nxv4i1, sme_elm_idx0_3,
1100                                     imm_to_tile32, tileslice32, op>;
1101  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _S),
1102                                     nxv4f32, nxv4i1, sme_elm_idx0_3,
1103                                     imm_to_tile32, tileslice32, op>;
1104  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _D),
1105                                     nxv2i64, nxv2i1, sme_elm_idx0_1,
1106                                     imm_to_tile64, tileslice64, op>;
1107  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _D),
1108                                     nxv2f64, nxv2i1, sme_elm_idx0_1,
1109                                     imm_to_tile64, tileslice64, op>;
1110
1111  defvar opq = !if(is_col, int_aarch64_sme_readq_vert,
1112                           int_aarch64_sme_readq_horiz);
1113
1114  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1115                                     nxv16i8, nxv16i1, sme_elm_idx0_0,
1116                                     imm_to_tile128, tileslice128, opq>;
1117  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1118                                     nxv8i16, nxv8i1, sme_elm_idx0_0,
1119                                     imm_to_tile128, tileslice128, opq>;
1120  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1121                                     nxv8f16, nxv8i1, sme_elm_idx0_0,
1122                                     imm_to_tile128, tileslice128, opq>;
1123  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1124                                     nxv8bf16, nxv8i1, sme_elm_idx0_0,
1125                                     imm_to_tile128, tileslice128, opq>;
1126  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1127                                     nxv4i32, nxv4i1, sme_elm_idx0_0,
1128                                     imm_to_tile128, tileslice128, opq>;
1129  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1130                                     nxv4f32, nxv4i1, sme_elm_idx0_0,
1131                                     imm_to_tile128, tileslice128, opq>;
1132  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1133                                     nxv2i64, nxv2i1, sme_elm_idx0_0,
1134                                     imm_to_tile128, tileslice128, opq>;
1135  defm : sme_tile_to_vector_patterns<!cast<Instruction>(NAME # _Q),
1136                                     nxv2f64, nxv2i1, sme_elm_idx0_0,
1137                                     imm_to_tile128, tileslice128, opq>;
1138}
1139
1140multiclass sme_tile_to_vector<string mnemonic> {
1141  defm _H : sme_tile_to_vector_v<mnemonic, /*is_col=*/0b0>;
1142  defm _V : sme_tile_to_vector_v<mnemonic, /*is_col=*/0b1>;
1143}
1144
1145//===----------------------------------------------------------------------===//
1146// SME Zero
1147//===----------------------------------------------------------------------===//
1148
1149// NOTE: This definition isn't really correct because there are outputs, i.e.
1150// the tile registers being zeroed. We fix this up in a custom inserter that
1151// marks the appropriate registers as being implicitly defined.
1152class sme_zero_inst<string mnemonic>
1153    : I<(outs), (ins MatrixTileList:$imm),
1154        mnemonic, "\t$imm", "", []>, Sched<[]> {
1155  bits<8> imm;
1156  let Inst{31-8} = 0b110000000000100000000000;
1157  let Inst{7-0}  = imm;
1158}
1159
1160multiclass sme_zero<string mnemonic> {
1161  def NAME : sme_zero_inst<mnemonic>;
1162
1163  def : InstAlias<"zero\t\\{za\\}", (!cast<Instruction>(NAME) 0b11111111), 1>;
1164  def : InstAlias<"zero\t\\{za0.h\\}", (!cast<Instruction>(NAME) 0b01010101), 1>;
1165  def : InstAlias<"zero\t\\{za1.h\\}", (!cast<Instruction>(NAME) 0b10101010), 1>;
1166  def : InstAlias<"zero\t\\{za0.s\\}", (!cast<Instruction>(NAME) 0b00010001), 1>;
1167  def : InstAlias<"zero\t\\{za1.s\\}", (!cast<Instruction>(NAME) 0b00100010), 1>;
1168  def : InstAlias<"zero\t\\{za2.s\\}", (!cast<Instruction>(NAME) 0b01000100), 1>;
1169  def : InstAlias<"zero\t\\{za3.s\\}", (!cast<Instruction>(NAME) 0b10001000), 1>;
1170  def : InstAlias<"zero\t\\{za0.s,za1.s\\}", (!cast<Instruction>(NAME) 0b00110011), 1>;
1171  def : InstAlias<"zero\t\\{za0.s,za3.s\\}", (!cast<Instruction>(NAME) 0b10011001), 1>;
1172  def : InstAlias<"zero\t\\{za1.s,za2.s\\}", (!cast<Instruction>(NAME) 0b01100110), 1>;
1173  def : InstAlias<"zero\t\\{za2.s,za3.s\\}", (!cast<Instruction>(NAME) 0b11001100), 1>;
1174  def : InstAlias<"zero\t\\{za0.s,za1.s,za2.s\\}", (!cast<Instruction>(NAME) 0b01110111), 1>;
1175  def : InstAlias<"zero\t\\{za0.s,za1.s,za3.s\\}", (!cast<Instruction>(NAME) 0b10111011), 1>;
1176  def : InstAlias<"zero\t\\{za0.s,za2.s,za3.s\\}", (!cast<Instruction>(NAME) 0b11011101), 1>;
1177  def : InstAlias<"zero\t\\{za1.s,za2.s,za3.s\\}", (!cast<Instruction>(NAME) 0b11101110), 1>;
1178
1179  def NAME # _PSEUDO : Pseudo<(outs), (ins i32imm:$tilelist), []>,
1180      Sched<[]> {
1181    // Translated to the actual instructions in AArch64ISelLowering.cpp
1182    let usesCustomInserter = 1;
1183  }
1184
1185  def : Pat<(int_aarch64_sme_zero timm32_0_255:$imm),
1186            (!cast<Instruction>(NAME # _PSEUDO) timm32_0_255:$imm)>;
1187}
1188
1189//===----------------------------------------------------------------------===//
1190// SVE2 Instructions
1191//===----------------------------------------------------------------------===//
1192
1193class sve2_int_perm_revd<string asm>
1194    : I<(outs ZPR128:$Zd), (ins ZPR128:$_Zd, PPR3bAny:$Pg, ZPR128:$Zn),
1195        asm, "\t$Zd, $Pg/m, $Zn", "", []>,
1196      Sched<[]> {
1197  bits<5> Zd;
1198  bits<3> Pg;
1199  bits<5> Zn;
1200  let Inst{31-24} = 0b00000101;
1201  let Inst{23-22} = 0b00; // size
1202  let Inst{21-13} = 0b101110100;
1203  let Inst{12-10} = Pg;
1204  let Inst{9-5}   = Zn;
1205  let Inst{4-0}   = Zd;
1206
1207  let Constraints = "$Zd = $_Zd";
1208  let DestructiveInstType = DestructiveUnary;
1209  let ElementSize = ZPR128.ElementSize;
1210}
1211
1212multiclass sve2_int_perm_revd<string asm, SDPatternOperator op> {
1213  def NAME : sve2_int_perm_revd<asm>;
1214
1215  def : SVE_1_Op_Passthru_Pat<nxv16i8, op, nxv16i1, nxv16i8, !cast<Instruction>(NAME)>;
1216  def : SVE_1_Op_Passthru_Pat<nxv8i16, op, nxv8i1,  nxv8i16, !cast<Instruction>(NAME)>;
1217  def : SVE_1_Op_Passthru_Pat<nxv4i32, op, nxv4i1,  nxv4i32, !cast<Instruction>(NAME)>;
1218  def : SVE_1_Op_Passthru_Pat<nxv2i64, op, nxv2i1,  nxv2i64, !cast<Instruction>(NAME)>;
1219}
1220
1221class sve2_clamp<string asm, bits<2> sz, bit U, ZPRRegOp zpr_ty>
1222    : I<(outs zpr_ty:$Zd), (ins zpr_ty:$Zn, zpr_ty:$Zm, zpr_ty:$_Zd),
1223        asm, "\t$Zd, $Zn, $Zm", "", []>,
1224      Sched<[]> {
1225  bits<5> Zm;
1226  bits<5> Zn;
1227  bits<5> Zd;
1228  let Inst{31-24} = 0b01000100;
1229  let Inst{23-22} = sz;
1230  let Inst{21}    = 0b0;
1231  let Inst{20-16} = Zm;
1232  let Inst{15-11} = 0b11000;
1233  let Inst{10}    = U;
1234  let Inst{9-5}   = Zn;
1235  let Inst{4-0}   = Zd;
1236
1237  let Constraints = "$Zd = $_Zd";
1238  let DestructiveInstType = DestructiveOther;
1239  let ElementSize = zpr_ty.ElementSize;
1240}
1241
1242multiclass sve2_clamp<string asm, bit U, SDPatternOperator op> {
1243  def _B : sve2_clamp<asm, 0b00, U, ZPR8>;
1244  def _H : sve2_clamp<asm, 0b01, U, ZPR16>;
1245  def _S : sve2_clamp<asm, 0b10, U, ZPR32>;
1246  def _D : sve2_clamp<asm, 0b11, U, ZPR64>;
1247
1248  def : SVE_3_Op_Pat<nxv16i8, op, nxv16i8, nxv16i8, nxv16i8, !cast<Instruction>(NAME # _B)>;
1249  def : SVE_3_Op_Pat<nxv8i16, op, nxv8i16, nxv8i16, nxv8i16, !cast<Instruction>(NAME # _H)>;
1250  def : SVE_3_Op_Pat<nxv4i32, op, nxv4i32, nxv4i32, nxv4i32, !cast<Instruction>(NAME # _S)>;
1251  def : SVE_3_Op_Pat<nxv2i64, op, nxv2i64, nxv2i64, nxv2i64, !cast<Instruction>(NAME # _D)>;
1252}
1253
1254class sve2_int_perm_sel_p<string asm, PPRRegOp ppr_ty, Operand imm_ty>
1255    : I<(outs PPRAny:$Pd), (ins PPRAny:$Pn, ppr_ty:$Pm,
1256                            MatrixIndexGPR32Op12_15:$Rv, imm_ty:$imm),
1257        asm, "\t$Pd, $Pn, $Pm[$Rv, $imm]", "", []>,
1258      Sched<[]> {
1259  bits<2> Rv;
1260  bits<4> Pn;
1261  bits<4> Pm;
1262  bits<4> Pd;
1263  let Inst{31-24} = 0b00100101;
1264  let Inst{21}    = 0b1;
1265  let Inst{17-16} = Rv;
1266  let Inst{15-14} = 0b01;
1267  let Inst{13-10} = Pn;
1268  let Inst{9}     = 0b0;
1269  let Inst{8-5}   = Pm;
1270  let Inst{4}     = 0b0;
1271  let Inst{3-0}   = Pd;
1272}
1273
1274multiclass sve2_int_perm_sel_p<string asm, SDPatternOperator op> {
1275  def _B : sve2_int_perm_sel_p<asm, PPR8, sme_elm_idx0_15> {
1276    bits<4> imm;
1277    let Inst{23-22} = imm{3-2};
1278    let Inst{20-19} = imm{1-0};
1279    let Inst{18}    = 0b1;
1280  }
1281  def _H : sve2_int_perm_sel_p<asm, PPR16, sme_elm_idx0_7> {
1282    bits<3> imm;
1283    let Inst{23-22} = imm{2-1};
1284    let Inst{20}    = imm{0};
1285    let Inst{19-18} = 0b10;
1286  }
1287  def _S : sve2_int_perm_sel_p<asm, PPR32, sme_elm_idx0_3> {
1288    bits<2> imm;
1289    let Inst{23-22} = imm{1-0};
1290    let Inst{20-18} = 0b100;
1291  }
1292  def _D : sve2_int_perm_sel_p<asm, PPR64, sme_elm_idx0_1> {
1293    bits<1> imm;
1294    let Inst{23}    = imm;
1295    let Inst{22}    = 0b1;
1296    let Inst{20-18} = 0b000;
1297  }
1298
1299  def : InstAlias<asm # "\t$Pd, $Pn, $Pm[$Rv, $imm]",
1300                  (!cast<Instruction>(NAME # _B) PNRAny:$Pd,
1301                      PNRAny:$Pn, PPR8:$Pm, MatrixIndexGPR32Op12_15:$Rv, sme_elm_idx0_15:$imm), 0>;
1302  def : InstAlias<asm # "\t$Pd, $Pn, $Pm[$Rv, $imm]",
1303                  (!cast<Instruction>(NAME # _H) PNRAny:$Pd,
1304                      PNRAny:$Pn, PPR16:$Pm, MatrixIndexGPR32Op12_15:$Rv, sme_elm_idx0_7:$imm), 0>;
1305  def : InstAlias<asm # "\t$Pd, $Pn, $Pm[$Rv, $imm]",
1306                  (!cast<Instruction>(NAME # _S) PNRAny:$Pd,
1307                      PNRAny:$Pn, PPR32:$Pm, MatrixIndexGPR32Op12_15:$Rv, sme_elm_idx0_3:$imm), 0>;
1308  def : InstAlias<asm # "\t$Pd, $Pn, $Pm[$Rv, $imm]",
1309                  (!cast<Instruction>(NAME # _D) PNRAny:$Pd,
1310                      PNRAny:$Pn, PPR64:$Pm, MatrixIndexGPR32Op12_15:$Rv, sme_elm_idx0_1:$imm), 0>;
1311
1312  def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv16i1 PPR8:$Pm),
1313             MatrixIndexGPR32Op12_15:$idx)),
1314            (!cast<Instruction>(NAME # _B) $Pn, $Pm, $idx, 0)>;
1315  def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv8i1 PPR16:$Pm),
1316             MatrixIndexGPR32Op12_15:$idx)),
1317            (!cast<Instruction>(NAME # _H) $Pn, $Pm, $idx, 0)>;
1318  def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv4i1 PPR32:$Pm),
1319             MatrixIndexGPR32Op12_15:$idx)),
1320            (!cast<Instruction>(NAME # _S) $Pn, $Pm, $idx, 0)>;
1321  def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv2i1 PPR64:$Pm),
1322             MatrixIndexGPR32Op12_15:$idx)),
1323            (!cast<Instruction>(NAME # _D) $Pn, $Pm, $idx, 0)>;
1324
1325  let AddedComplexity = 1 in {
1326    def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv16i1 PPR8:$Pm),
1327               (i32 (tileslice8 MatrixIndexGPR32Op12_15:$idx, sme_elm_idx0_15:$imm)))),
1328              (!cast<Instruction>(NAME # _B) $Pn, $Pm, $idx, $imm)>;
1329    def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv8i1 PPR16:$Pm),
1330               (i32 (tileslice16 MatrixIndexGPR32Op12_15:$idx, sme_elm_idx0_7:$imm)))),
1331              (!cast<Instruction>(NAME # _H) $Pn, $Pm, $idx, $imm)>;
1332    def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv4i1 PPR32:$Pm),
1333               (i32 (tileslice32 MatrixIndexGPR32Op12_15:$idx, sme_elm_idx0_3:$imm)))),
1334              (!cast<Instruction>(NAME # _S) $Pn, $Pm, $idx, $imm)>;
1335    def : Pat<(nxv16i1 (op (nxv16i1 PPRAny:$Pn), (nxv2i1 PPR64:$Pm),
1336               (i32 (tileslice64 MatrixIndexGPR32Op12_15:$idx, sme_elm_idx0_1:$imm)))),
1337              (!cast<Instruction>(NAME # _D) $Pn, $Pm, $idx, $imm)>;
1338  }
1339}
1340
1341//===----------------------------------------------------------------------===//
1342// SME2 Instructions
1343//===----------------------------------------------------------------------===//
1344
1345//===----------------------------------------------------------------------===//
1346// SME2 single-multi ternary int/fp, two/four registers
1347
1348class sme2_dot_mla_add_sub_array_vg24_single<bits<7> op,
1349                                         MatrixOperand matrix_ty,
1350                                         RegisterOperand multi_vector_ty,
1351                                         ZPRRegOp zpr_ty,
1352                                         string mnemonic>
1353   : I<(outs matrix_ty:$ZAd),
1354       (ins  matrix_ty:$_ZAd, MatrixIndexGPR32Op8_11:$Rv,
1355       sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, zpr_ty:$Zm),
1356       mnemonic,"\t$ZAd[$Rv, $imm3, " # !if(op{5}, "vgx4", "vgx2") # "], $Zn, $Zm",
1357       "", []> , Sched<[]> {
1358  bits<4> Zm;
1359  bits<5> Zn;
1360  bits<2> Rv;
1361  bits<3> imm3;
1362  let Inst{31-23} = 0b110000010;
1363  let Inst{22}    = op{6}; //sz
1364  let Inst{21}    = 0b1;
1365  let Inst{20}    = op{5}; //vgx4
1366  let Inst{19-16} = Zm;
1367  let Inst{15}    = 0b0;
1368  let Inst{14-13} = Rv;
1369  let Inst{12-10} = op{4-2};
1370  let Inst{9-5}   = Zn;
1371  let Inst{4-3}   = op{1-0};
1372  let Inst{2-0}   = imm3;
1373  let Constraints = "$ZAd = $_ZAd";
1374}
1375
1376multiclass sme2_dot_mla_add_sub_array_vg24_single<string mnemonic, bits<7> op,
1377                                              MatrixOperand matrix_ty,
1378                                              RegisterOperand multi_vector_ty,
1379                                              ZPRRegOp zpr_ty>{
1380  def NAME: sme2_dot_mla_add_sub_array_vg24_single<op, matrix_ty, multi_vector_ty, zpr_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
1381
1382  def : InstAlias<mnemonic # "\t$ZAd[$Rv, $imm3], $Zn, $Zm",
1383                 (!cast<Instruction>(NAME) matrix_ty:$ZAd,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, zpr_ty:$Zm), 0>;
1384}
1385
1386multiclass sme2_dot_mla_add_sub_array_vg2_single<string mnemonic, bits<7> op,
1387                                              MatrixOperand matrix_ty,
1388                                              RegisterOperand multi_vector_ty,
1389                                              ZPRRegOp zpr_ty, ValueType vty, SDPatternOperator intrinsic>{
1390  def NAME: sme2_dot_mla_add_sub_array_vg24_single<op, matrix_ty, multi_vector_ty, zpr_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
1391
1392  def : InstAlias<mnemonic # "\t$ZAd[$Rv, $imm3], $Zn, $Zm",
1393                 (!cast<Instruction>(NAME) matrix_ty:$ZAd,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, zpr_ty:$Zm), 0>;
1394
1395  def _PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, zpr_ty, SMEMatrixArray>;
1396
1397  def : SME2_ZA_TwoOp_VG2_Multi_Single_Pat<NAME, intrinsic, sme_elm_idx0_7, zpr_ty, vty, tileslice16>;
1398}
1399
1400multiclass sme2_dot_mla_add_sub_array_vg4_single<string mnemonic, bits<7> op,
1401                                              MatrixOperand matrix_ty,
1402                                              RegisterOperand multi_vector_ty,
1403                                              ZPRRegOp zpr_ty, ValueType vty, SDPatternOperator intrinsic>{
1404  def NAME: sme2_dot_mla_add_sub_array_vg24_single<op, matrix_ty, multi_vector_ty, zpr_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
1405
1406  def : InstAlias<mnemonic # "\t$ZAd[$Rv, $imm3], $Zn, $Zm",
1407                 (!cast<Instruction>(NAME) matrix_ty:$ZAd,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, zpr_ty:$Zm), 0>;
1408
1409  def _PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, zpr_ty, SMEMatrixArray>;
1410
1411  def : SME2_ZA_TwoOp_VG4_Multi_Single_Pat<NAME, intrinsic, sme_elm_idx0_7, zpr_ty, vty, tileslice16>;
1412}
1413
1414//===----------------------------------------------------------------------===//
1415// SME2 multiple vectors ternary INT/FP  two and four registers
1416class sme2_dot_mla_add_sub_array_vg2_multi<bits<6> op,
1417                                       MatrixOperand matrix_ty,
1418                                       RegisterOperand multi_vector_ty,
1419                                       string mnemonic>
1420   : I<(outs matrix_ty:$ZAd),
1421       (ins  matrix_ty:$_ZAd, MatrixIndexGPR32Op8_11:$Rv,
1422       sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, multi_vector_ty:$Zm),
1423       mnemonic, "\t$ZAd[$Rv, $imm3, vgx2], $Zn, $Zm",
1424       "", []>, Sched<[]>{
1425  bits<4> Zm;
1426  bits<4> Zn;
1427  bits<2> Rv;
1428  bits<3> imm3;
1429  let Inst{31-23} = 0b110000011;
1430  let Inst{22}    = op{5}; //sz
1431  let Inst{21}    = 0b1;
1432  let Inst{20-17} = Zm;
1433  let Inst{16-15} = 0b00;
1434  let Inst{14-13} = Rv;
1435  let Inst{12-10} = op{4-2};
1436  let Inst{9-6}   = Zn;
1437  let Inst{5}     = 0b0;
1438  let Inst{4-3}   = op{1-0};
1439  let Inst{2-0}   = imm3;
1440  let Constraints = "$ZAd = $_ZAd";
1441}
1442
1443multiclass sme2_dot_mla_add_sub_array_vg2_multi<string mnemonic, bits<6> op,
1444                                            MatrixOperand  matrix_ty,
1445                                            RegisterOperand multi_vector_ty, ValueType zpr_ty,
1446                                            SDPatternOperator intrinsic> {
1447  def NAME : sme2_dot_mla_add_sub_array_vg2_multi<op, matrix_ty, multi_vector_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
1448
1449  def _PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, SMEMatrixArray>;
1450
1451  def : SME2_ZA_TwoOp_VG2_Multi_Multi_Pat<NAME, intrinsic, sme_elm_idx0_7, zpr_ty, tileslice16>;
1452
1453  def : InstAlias<mnemonic # "\t$ZAd[$Rv, $imm3], $Zn, $Zm",
1454                  (!cast<Instruction>(NAME) matrix_ty:$ZAd,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, multi_vector_ty:$Zm), 0>;
1455}
1456
1457class sme2_dot_mla_add_sub_array_vg4_multi<bits<6> op,
1458                                            MatrixOperand matrix_ty,
1459                                            RegisterOperand multi_vector_ty,
1460                                            string mnemonic>
1461   : I<(outs matrix_ty:$ZAd),
1462       (ins  matrix_ty:$_ZAd, MatrixIndexGPR32Op8_11:$Rv,
1463        sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, multi_vector_ty:$Zm),
1464        mnemonic, "\t$ZAd[$Rv, $imm3, vgx4], $Zn, $Zm",
1465        "", []>, Sched<[]>{
1466  bits<3> Zm;
1467  bits<3> Zn;
1468  bits<2> Rv;
1469  bits<3> imm3;
1470  let Inst{31-23} = 0b110000011;
1471  let Inst{22}    = op{5}; //sz
1472  let Inst{21}    = 0b1;
1473  let Inst{20-18} = Zm;
1474  let Inst{17-15} = 0b010;
1475  let Inst{14-13} = Rv;
1476  let Inst{12-10} = op{4-2};
1477  let Inst{9-7}   = Zn;
1478  let Inst{6-5}   = 0b00;
1479  let Inst{4-3}   = op{1-0};
1480  let Inst{2-0}   = imm3;
1481  let Constraints = "$ZAd = $_ZAd";
1482}
1483
1484multiclass sme2_dot_mla_add_sub_array_vg4_multi<string mnemonic, bits<6> op,
1485                                            MatrixOperand  matrix_ty,
1486                                            RegisterOperand multi_vector_ty,
1487                                            ValueType zpr_ty, SDPatternOperator intrinsic>{
1488  def NAME : sme2_dot_mla_add_sub_array_vg4_multi<op, matrix_ty, multi_vector_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
1489
1490  def _PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, SMEMatrixArray>;
1491
1492  def : SME2_ZA_TwoOp_VG4_Multi_Multi_Pat<NAME, intrinsic, sme_elm_idx0_7, zpr_ty, tileslice16>;
1493
1494  def : InstAlias<mnemonic # "\t$ZAd[$Rv, $imm3], $Zn, $Zm",
1495                 (!cast<Instruction>(NAME) matrix_ty:$ZAd,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, multi_vector_ty:$Zn, multi_vector_ty:$Zm), 0>;
1496}
1497
1498//===----------------------------------------------------------------------===//
1499// SME2 multiple vectors binary two or four  registers
1500
1501class sme2_multivec_accum_add_sub<string mnemonic, bit sz, bit vg4, bits<3> op,
1502                                  MatrixOperand matrix_ty,
1503                                  RegisterOperand vector_ty>
1504    : I<(outs matrix_ty:$ZAdn),
1505        (ins matrix_ty:$_ZAdn, MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, vector_ty:$Zm),
1506        mnemonic, "\t$ZAdn[$Rv, $imm3, " # !if(vg4, "vgx4", "vgx2") # "], $Zm",
1507        "", []>, Sched<[]> {
1508  bits<2> Rv;
1509  bits<3> imm3;
1510  let Inst{31-23} = 0b110000011;
1511  let Inst{22}    = sz;
1512  let Inst{21-19} = 0b100;
1513  let Inst{18}    = op{2};
1514  let Inst{17}    = 0b0;
1515  let Inst{16}    = vg4;
1516  let Inst{15}    = 0b0;
1517  let Inst{14-13} = Rv;
1518  let Inst{12-10} = 0b111;
1519  let Inst{5}     = 0b0;
1520  let Inst{4-3}   = op{1-0};
1521  let Inst{2-0}   = imm3;
1522
1523  let Constraints = "$ZAdn = $_ZAdn";
1524}
1525
1526class sme2_multivec_accum_add_sub_vg2<string mnemonic, bit sz, bits<3> op,
1527                                      MatrixOperand matrix_ty,
1528                                      RegisterOperand vector_ty>
1529    : sme2_multivec_accum_add_sub<mnemonic, sz, 0b0, op, matrix_ty, vector_ty> {
1530  bits<4> Zm;
1531  let Inst{9-6} = Zm;
1532}
1533
1534
1535multiclass sme2_multivec_accum_add_sub_vg2<string mnemonic, bits<4> op,
1536                                           MatrixOperand matrix_ty,
1537                                           RegisterOperand vector_ty,
1538                                           ValueType vty,
1539                                           SDPatternOperator intrinsic> {
1540  def NAME : sme2_multivec_accum_add_sub_vg2<mnemonic, op{3}, op{2-0}, matrix_ty, vector_ty>,
1541                                             SMEPseudo2Instr<NAME, 1>;
1542  def : InstAlias<mnemonic # "\t$ZAdn[$Rv, $imm3], $Zm",
1543  (!cast<Instruction>(NAME) matrix_ty:$ZAdn,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, vector_ty:$Zm), 0>;
1544
1545  def _PSEUDO : sme2_move_to_za_pseudo<NAME, sme_elm_idx0_7, vector_ty, SMEMatrixArray>;
1546  def : SME2_ZA_VG1x2_Multi_Pat<NAME, intrinsic, vty, sme_elm_idx0_7, tileslice16>;
1547}
1548
1549class sme2_multivec_accum_add_sub_vg4<string mnemonic, bit sz, bits<3> op,
1550                                      MatrixOperand matrix_ty,
1551                                      RegisterOperand vector_ty>
1552    : sme2_multivec_accum_add_sub<mnemonic, sz, 0b1, op, matrix_ty, vector_ty> {
1553  bits<3> Zm;
1554  let Inst{9-7} = Zm;
1555  let Inst{6}   = 0b0;
1556}
1557
1558multiclass sme2_multivec_accum_add_sub_vg4<string mnemonic, bits<4> op,
1559                                           MatrixOperand matrix_ty,
1560                                           RegisterOperand vector_ty,
1561                                           ValueType vty,
1562                                           SDPatternOperator intrinsic> {
1563  def NAME : sme2_multivec_accum_add_sub_vg4<mnemonic, op{3}, op{2-0}, matrix_ty, vector_ty>,
1564                                             SMEPseudo2Instr<NAME, 1>;
1565  def : InstAlias<mnemonic # "\t$ZAdn[$Rv, $imm3], $Zm",
1566  (!cast<Instruction>(NAME) matrix_ty:$ZAdn,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3, vector_ty:$Zm), 0>;
1567
1568  def _PSEUDO : sme2_move_to_za_pseudo<NAME, sme_elm_idx0_7, vector_ty, SMEMatrixArray>;
1569  def : SME2_ZA_VG1x4_Multi_Pat<NAME, intrinsic, vty, sme_elm_idx0_7, tileslice16>;
1570}
1571
1572//===----------------------------------------------------------------------===//
1573// SME2 Multi-vector - Multiple and Single SVE Destructive
1574// Two and Four registers
1575
1576class sme2_sve_destructive_vector_vg2_single<bits<2> sz, bits<7> op,
1577                                             RegisterOperand vector_ty,
1578                                             ZPRRegOp zpr_ty,
1579                                             string mnemonic>
1580    : I<(outs vector_ty:$Zdn), (ins vector_ty:$_Zdn, zpr_ty:$Zm),
1581        mnemonic, "\t$Zdn, $_Zdn, $Zm",
1582        "", []>, Sched<[]> {
1583  bits<4> Zm;
1584  bits<4> Zdn;
1585  let Inst{31-24} = 0b11000001;
1586  let Inst{23-22} = sz;
1587  let Inst{21-20} = 0b10;
1588  let Inst{19-16} = Zm;
1589  let Inst{15-11} = 0b10100;
1590  let Inst{10-5}  = op{6-1};
1591  let Inst{4-1}   = Zdn;
1592  let Inst{0}     = op{0};
1593
1594  let Constraints = "$Zdn = $_Zdn";
1595}
1596
1597multiclass sme2_fp_sve_destructive_vector_vg2_single<string mnemonic, bits<7> op> {
1598  def _H : sme2_sve_destructive_vector_vg2_single<0b01, op, ZZ_h_mul_r, ZPR4b16, mnemonic>;
1599  def _S : sme2_sve_destructive_vector_vg2_single<0b10, op, ZZ_s_mul_r, ZPR4b32, mnemonic>;
1600  def _D : sme2_sve_destructive_vector_vg2_single<0b11, op, ZZ_d_mul_r, ZPR4b64, mnemonic>;
1601}
1602
1603multiclass sme2_int_sve_destructive_vector_vg2_single<string mnemonic, bits<7> op> {
1604  def _B : sme2_sve_destructive_vector_vg2_single<0b00, op, ZZ_b_mul_r, ZPR4b8, mnemonic>;
1605  def _H : sme2_sve_destructive_vector_vg2_single<0b01, op, ZZ_h_mul_r, ZPR4b16, mnemonic>;
1606  def _S : sme2_sve_destructive_vector_vg2_single<0b10, op, ZZ_s_mul_r, ZPR4b32, mnemonic>;
1607  def _D : sme2_sve_destructive_vector_vg2_single<0b11, op, ZZ_d_mul_r, ZPR4b64, mnemonic>;
1608}
1609
1610// SME2.1 fmax/fmin instructions.
1611multiclass sme2p1_bf_max_min_vector_vg2_single<string mnemonic, bits<7>op> {
1612  def _H : sme2_sve_destructive_vector_vg2_single<0b00, op, ZZ_h_mul_r,
1613                                                  ZPR4b16, mnemonic>;
1614}
1615
1616class sme2_sve_destructive_vector_vg4_single<bits<2> sz, bits<7> op,
1617                                             RegisterOperand vector_ty,
1618                                             ZPRRegOp zpr_ty,
1619                                             string mnemonic>
1620    : I<(outs vector_ty:$Zdn), (ins vector_ty:$_Zdn, zpr_ty:$Zm),
1621        mnemonic, "\t$Zdn, $_Zdn, $Zm",
1622        "", []>, Sched<[]> {
1623  bits<4> Zm;
1624  bits<3> Zdn;
1625  let Inst{31-24} = 0b11000001;
1626  let Inst{23-22} = sz;
1627  let Inst{21-20} = 0b10;
1628  let Inst{19-16} = Zm;
1629  let Inst{15-11} = 0b10101;
1630  let Inst{10-5}  = op{6-1};
1631  let Inst{4-2}   = Zdn;
1632  let Inst{1}     = 0b0;
1633  let Inst{0}     = op{0};
1634
1635  let Constraints = "$Zdn = $_Zdn";
1636}
1637
1638multiclass sme2_fp_sve_destructive_vector_vg4_single<string mnemonic, bits<7> op> {
1639  def _H : sme2_sve_destructive_vector_vg4_single<0b01, op, ZZZZ_h_mul_r, ZPR4b16, mnemonic>;
1640  def _S : sme2_sve_destructive_vector_vg4_single<0b10, op, ZZZZ_s_mul_r, ZPR4b32, mnemonic>;
1641  def _D : sme2_sve_destructive_vector_vg4_single<0b11, op, ZZZZ_d_mul_r, ZPR4b64, mnemonic>;
1642}
1643
1644multiclass sme2_int_sve_destructive_vector_vg4_single<string mnemonic, bits<7> op> {
1645  def _B : sme2_sve_destructive_vector_vg4_single<0b00, op, ZZZZ_b_mul_r, ZPR4b8, mnemonic>;
1646  def _H : sme2_sve_destructive_vector_vg4_single<0b01, op, ZZZZ_h_mul_r, ZPR4b16, mnemonic>;
1647  def _S : sme2_sve_destructive_vector_vg4_single<0b10, op, ZZZZ_s_mul_r, ZPR4b32, mnemonic>;
1648  def _D : sme2_sve_destructive_vector_vg4_single<0b11, op, ZZZZ_d_mul_r, ZPR4b64, mnemonic>;
1649}
1650
1651// SME2.1 fmax/fmin instructions.
1652multiclass sme2p1_bf_max_min_vector_vg4_single<string mnemonic, bits<7>op> {
1653  def _H : sme2_sve_destructive_vector_vg4_single<0b00, op, ZZZZ_h_mul_r,
1654                                                  ZPR4b16, mnemonic>;
1655}
1656
1657class sme2_sve_destructive_vector_vg2_multi<bits<2> sz, bits<7> op,
1658                                            RegisterOperand vector_ty,
1659                                            string mnemonic>
1660    : I<(outs vector_ty:$Zdn), (ins vector_ty:$_Zdn, vector_ty:$Zm),
1661        mnemonic, "\t$Zdn, $_Zdn, $Zm",
1662        "", []>, Sched<[]> {
1663  bits<4> Zm;
1664  bits<4> Zdn;
1665  let Inst{31-24} = 0b11000001;
1666  let Inst{23-22} = sz;
1667  let Inst{21}    = 0b1;
1668  let Inst{20-17} = Zm;
1669  let Inst{16-11} = 0b010110;
1670  let Inst{10-5}  = op{6-1};
1671  let Inst{4-1}   = Zdn;
1672  let Inst{0}     = op{0};
1673
1674  let Constraints = "$Zdn = $_Zdn";
1675}
1676
1677multiclass sme2_fp_sve_destructive_vector_vg2_multi<string mnemonic, bits<7> op> {
1678  def _H : sme2_sve_destructive_vector_vg2_multi<0b01, op, ZZ_h_mul_r, mnemonic>;
1679  def _S : sme2_sve_destructive_vector_vg2_multi<0b10, op, ZZ_s_mul_r, mnemonic>;
1680  def _D : sme2_sve_destructive_vector_vg2_multi<0b11, op, ZZ_d_mul_r, mnemonic>;
1681}
1682
1683multiclass sme2_int_sve_destructive_vector_vg2_multi<string mnemonic, bits<7> op> {
1684  def _B : sme2_sve_destructive_vector_vg2_multi<0b00, op, ZZ_b_mul_r, mnemonic>;
1685  def _H : sme2_sve_destructive_vector_vg2_multi<0b01, op, ZZ_h_mul_r, mnemonic>;
1686  def _S : sme2_sve_destructive_vector_vg2_multi<0b10, op, ZZ_s_mul_r, mnemonic>;
1687  def _D : sme2_sve_destructive_vector_vg2_multi<0b11, op, ZZ_d_mul_r, mnemonic>;
1688}
1689
1690// SME2.1 fmax/fmin instructions.
1691multiclass sme2p1_bf_max_min_vector_vg2_multi<string mnemonic, bits<7>op> {
1692  def _H : sme2_sve_destructive_vector_vg2_multi<0b00, op, ZZ_h_mul_r,
1693                                                 mnemonic>;
1694}
1695
1696class sme2_sve_destructive_vector_vg4_multi<bits<2> sz, bits<7> op,
1697                                            RegisterOperand vector_ty,
1698                                            string mnemonic>
1699    : I<(outs vector_ty:$Zdn), (ins vector_ty:$_Zdn, vector_ty:$Zm),
1700        mnemonic, "\t$Zdn, $_Zdn, $Zm",
1701        "", []>, Sched<[]> {
1702  bits<3> Zm;
1703  bits<3> Zdn;
1704  let Inst{31-24} = 0b11000001;
1705  let Inst{23-22} = sz;
1706  let Inst{21}    = 0b1;
1707  let Inst{20-18} = Zm;
1708  let Inst{17-11} = 0b0010111;
1709  let Inst{10-5}  = op{6-1};
1710  let Inst{4-2}   = Zdn;
1711  let Inst{1}     = 0b0;
1712  let Inst{0}     = op{0};
1713
1714  let Constraints = "$Zdn = $_Zdn";
1715}
1716
1717multiclass sme2_fp_sve_destructive_vector_vg4_multi<string mnemonic, bits<7> op> {
1718  def _H : sme2_sve_destructive_vector_vg4_multi<0b01, op, ZZZZ_h_mul_r, mnemonic>;
1719  def _S : sme2_sve_destructive_vector_vg4_multi<0b10, op, ZZZZ_s_mul_r, mnemonic>;
1720  def _D : sme2_sve_destructive_vector_vg4_multi<0b11, op, ZZZZ_d_mul_r, mnemonic>;
1721}
1722
1723multiclass sme2_int_sve_destructive_vector_vg4_multi<string mnemonic, bits<7> op> {
1724  def _B : sme2_sve_destructive_vector_vg4_multi<0b00, op, ZZZZ_b_mul_r, mnemonic>;
1725  def _H : sme2_sve_destructive_vector_vg4_multi<0b01, op, ZZZZ_h_mul_r, mnemonic>;
1726  def _S : sme2_sve_destructive_vector_vg4_multi<0b10, op, ZZZZ_s_mul_r, mnemonic>;
1727  def _D : sme2_sve_destructive_vector_vg4_multi<0b11, op, ZZZZ_d_mul_r, mnemonic>;
1728}
1729
1730// SME2.1 fmax/fmin instructions.
1731multiclass sme2p1_bf_max_min_vector_vg4_multi<string mnemonic, bits<7>op> {
1732  def _H : sme2_sve_destructive_vector_vg4_multi<0b00, op, ZZZZ_h_mul_r,
1733                                                 mnemonic>;
1734}
1735
1736//===----------------------------------------------------------------------===//
1737// SME2 Multi-vector - Index/Single/Multi Array Vectors FMA sources
1738
1739class sme2_mla_long_array_index_base<bits<2> op0, bits<2> op, Operand index_ty,
1740                                     RegisterOperand multi_vector_ty,
1741                                     string mnemonic, string vg_acronym="">
1742    : I<(outs MatrixOp32:$ZAda),
1743        (ins MatrixOp32:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, index_ty:$imm, multi_vector_ty:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i3),
1744        mnemonic, "\t$ZAda[$Rv, $imm" # !if(!eq(vg_acronym, ""), "", ", " # vg_acronym) # "], $Zn, $Zm$i3",
1745        "", []>, Sched<[]> {
1746  bits<4> Zm;
1747  bits<2> Rv;
1748  let Inst{31-24} = 0b11000001;
1749  let Inst{23-22} = op0;
1750  let Inst{21}    = 0b0;
1751  let Inst{20}    = !if(!eq(vg_acronym, ""), 0, 1);
1752  let Inst{19-16} = Zm;
1753  let Inst{14-13} = Rv;
1754  let Inst{12}    = 0b1;
1755  let Inst{4-3}   = op;
1756
1757  let Constraints = "$ZAda = $_ZAda";
1758}
1759
1760multiclass sme2_mla_long_array_index<string mnemonic, bits<2> op0, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
1761  def _S : sme2_mla_long_array_index_base<op0, op, uimm3s2range, ZPR16,
1762                                          mnemonic>, SMEPseudo2Instr<NAME # _S, 1> {
1763    bits<3> i3;
1764    bits<5> Zn;
1765    bits<3> imm;
1766    let Inst{15}    = i3{2};
1767    let Inst{11-10} = i3{1-0};
1768    let Inst{9-5}   = Zn;
1769    let Inst{2-0}   = imm;
1770  }
1771
1772  def _S_PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME # _S, uimm3s2range, ZPR16, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
1773
1774  def : SME2_ZA_TwoOp_Multi_Index_Pat<NAME # _S, intrinsic, uimm3s2range, ZPR4b16, zpr_ty, VectorIndexH32b_timm, tileslicerange3s2>;
1775}
1776
1777class sme2_mla_long_array_vg2_index<string mnemonic, bits<2> op0, bits<2> op>
1778    : sme2_mla_long_array_index_base<op0, op, uimm2s2range, ZZ_h_mul_r,
1779                                     mnemonic, "vgx2"> {
1780  bits<3> i3;
1781  bits<4> Zn;
1782  bits<2> imm;
1783  let Inst{15}    = 0b0;
1784  let Inst{11-10} = i3{2-1};
1785  let Inst{9-6}   = Zn;
1786  let Inst{5}     = 0b0;
1787  let Inst{2}     = i3{0};
1788  let Inst{1-0}   = imm;
1789}
1790
1791multiclass sme2_fp_mla_long_array_vg2_index<string mnemonic, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
1792  def _S : sme2_mla_long_array_vg2_index<mnemonic, 0b10, op>, SMEPseudo2Instr<NAME # _S, 1>;
1793
1794  def _S_PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME # _S, uimm2s2range, ZZ_h_mul_r, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
1795
1796  def : SME2_ZA_TwoOp_VG2_Multi_Index_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, zpr_ty, VectorIndexH32b_timm, tileslicerange2s2>;
1797
1798  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i3",
1799                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i3), 0>;
1800}
1801
1802multiclass sme2_int_mla_long_array_vg2_index<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
1803  def _S : sme2_mla_long_array_vg2_index<mnemonic, 0b11, op>, SMEPseudo2Instr<NAME # _S, 1>;
1804
1805  def _S_PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME # _S, uimm2s2range, ZZ_h_mul_r, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
1806
1807  def : SME2_ZA_TwoOp_VG2_Multi_Index_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, nxv8i16, VectorIndexH32b_timm, tileslicerange2s2>;
1808
1809  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i3",
1810                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i3), 0>;
1811}
1812
1813class sme2_mla_long_array_vg4_index<string mnemonic, bits<2> op0, bits<2> op>
1814    : sme2_mla_long_array_index_base<op0, op, uimm2s2range, ZZZZ_h_mul_r,
1815                                      mnemonic, "vgx4"> {
1816  bits<3> i3;
1817  bits<3> Zn;
1818  bits<2> imm;
1819  let Inst{15}    = 0b1;
1820  let Inst{11-10} = i3{2-1};
1821  let Inst{9-7}   = Zn;
1822  let Inst{6-5}   = 0b00;
1823  let Inst{2}     = i3{0};
1824  let Inst{1-0}   = imm;
1825}
1826
1827multiclass sme2_fp_mla_long_array_vg4_index<string mnemonic, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
1828  def _S : sme2_mla_long_array_vg4_index<mnemonic, 0b10, op>, SMEPseudo2Instr<NAME # _S, 1>;
1829
1830  def _S_PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME # _S, uimm2s2range, ZZZZ_h_mul_r, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
1831
1832  def : SME2_ZA_TwoOp_VG4_Multi_Index_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, zpr_ty, VectorIndexH32b_timm, tileslicerange2s2>;
1833
1834  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i3",
1835                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i3), 0>;
1836}
1837
1838multiclass sme2_int_mla_long_array_vg4_index<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
1839  def _S : sme2_mla_long_array_vg4_index<mnemonic, 0b11, op>, SMEPseudo2Instr<NAME # _S, 1>;
1840
1841  def _S_PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME # _S, uimm2s2range, ZZZZ_h_mul_r, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
1842
1843  def : SME2_ZA_TwoOp_VG4_Multi_Index_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, nxv8i16, VectorIndexH32b_timm, tileslicerange2s2>;
1844
1845  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i3",
1846                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i3), 0>;
1847}
1848
1849class sme2_mla_long_array<bits<2>op0, bits<2> op, Operand index_ty,
1850                          RegisterOperand first_vector_ty,
1851                          RegisterOperand second_vector_ty,
1852                          string mnemonic, string vg_acronym="">
1853   : I<(outs MatrixOp32:$ZAda),
1854       (ins  MatrixOp32:$_ZAda, MatrixIndexGPR32Op8_11:$Rv,
1855       index_ty:$imm, first_vector_ty:$Zn, second_vector_ty:$Zm),
1856       mnemonic,"\t$ZAda[$Rv, $imm" # !if(!eq(vg_acronym, ""), "", ", " # vg_acronym) # "], $Zn, $Zm",
1857       "", []> , Sched<[]> {
1858  bits<2> Rv;
1859  let Inst{31-24} = 0b11000001;
1860  let Inst{23-22} = op0;
1861  let Inst{21}    = 0b1;
1862  let Inst{15}    = 0b0;
1863  let Inst{14-13} = Rv;
1864  let Inst{12-11} = 0b01;
1865  let Inst{10}    = !if(!eq(vg_acronym, ""), 1, 0);
1866  let Inst{4-3}   = op;
1867
1868  let Constraints = "$ZAda = $_ZAda";
1869}
1870
1871multiclass sme2_mla_long_array_single<string mnemonic, bits<2> op0, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
1872  def _S : sme2_mla_long_array<op0, op, uimm3s2range, ZPR16, ZPR4b16,
1873                               mnemonic> , SMEPseudo2Instr<NAME # _S, 1>{
1874    bits<4> Zm;
1875    bits<5> Zn;
1876    bits<3> imm;
1877    let Inst{20}    = 0b0;
1878    let Inst{19-16} = Zm;
1879    let Inst{9-5}   = Zn;
1880    let Inst{2-0}   = imm;
1881  }
1882
1883  def _S_PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME # _S, uimm3s2range, ZPR16, ZPR4b16, SMEMatrixArray>;
1884
1885  def : SME2_ZA_TwoOp_Multi_Single_Pat<NAME # _S, intrinsic, uimm3s2range, ZPR4b16, zpr_ty, tileslicerange3s2>;
1886}
1887
1888class sme2_mla_long_array_vg24_single<bits<2> op0, bit vg4, bits<2> op,
1889                                      RegisterOperand first_vector_ty,
1890                                      string mnemonic, string vg_acronym>
1891    : sme2_mla_long_array<op0, op, uimm2s2range, first_vector_ty, ZPR4b16,
1892                          mnemonic, vg_acronym> {
1893  bits<4> Zm;
1894  bits<5> Zn;
1895  bits<2> imm;
1896  let Inst{20}    = vg4;
1897  let Inst{19-16} = Zm;
1898  let Inst{9-5}   = Zn;
1899  let Inst{2}     = 0b0;
1900  let Inst{1-0}   = imm;
1901}
1902
1903multiclass sme2_fp_mla_long_array_vg2_single<string mnemonic, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
1904  def _S : sme2_mla_long_array_vg24_single<0b00, 0b0, op, ZZ_h, mnemonic,
1905                                           "vgx2">, SMEPseudo2Instr<NAME # _S, 1>;
1906
1907  def _S_PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME # _S, uimm2s2range, ZZ_h, ZPR4b16, SMEMatrixArray>;
1908
1909  def : SME2_ZA_TwoOp_VG2_Multi_Single_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, zpr_ty, tileslicerange2s2>;
1910
1911  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
1912                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZ_h:$Zn, ZPR4b16:$Zm), 0>;
1913}
1914
1915multiclass sme2_int_mla_long_array_vg2_single<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
1916  def _S : sme2_mla_long_array_vg24_single<0b01, 0b0, op, ZZ_h, mnemonic,
1917                                           "vgx2">, SMEPseudo2Instr<NAME # _S, 1>;
1918
1919  def _S_PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME # _S, uimm2s2range, ZZ_h, ZPR4b16, SMEMatrixArray>;
1920
1921  def : SME2_ZA_TwoOp_VG2_Multi_Single_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, nxv8i16, tileslicerange2s2>;
1922
1923  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
1924                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZ_h:$Zn, ZPR4b16:$Zm), 0>;
1925}
1926
1927multiclass sme2_fp_mla_long_array_vg4_single<string mnemonic, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
1928  def _S : sme2_mla_long_array_vg24_single<0b00, 0b1, op, ZZZZ_h, mnemonic,
1929                                           "vgx4">, SMEPseudo2Instr<NAME # _S, 1>;
1930
1931  def _S_PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME # _S, uimm2s2range, ZZZZ_h, ZPR4b16, SMEMatrixArray>;
1932
1933  def : SME2_ZA_TwoOp_VG4_Multi_Single_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, zpr_ty, tileslicerange2s2>;
1934
1935  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
1936                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZZZ_h:$Zn, ZPR4b16:$Zm), 0>;
1937}
1938
1939multiclass sme2_int_mla_long_array_vg4_single<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
1940  def _S : sme2_mla_long_array_vg24_single<0b01, 0b1, op, ZZZZ_h, mnemonic,
1941                                           "vgx4">, SMEPseudo2Instr<NAME # _S, 1>;
1942
1943  def _S_PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME # _S, uimm2s2range, ZZZZ_h, ZPR4b16, SMEMatrixArray>;
1944
1945  def : SME2_ZA_TwoOp_VG4_Multi_Single_Pat<NAME # _S, intrinsic, uimm2s2range, ZPR4b16, nxv8i16, tileslicerange2s2>;
1946
1947  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
1948                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZZZ_h:$Zn, ZPR4b16:$Zm), 0>;
1949}
1950class sme2_mla_long_array_vg2_multi<string mnemonic, bits<2> op0, bits<2> op>
1951   : sme2_mla_long_array<op0, op, uimm2s2range, ZZ_h_mul_r, ZZ_h_mul_r,  mnemonic,
1952                         "vgx2"> {
1953  bits<4> Zm;
1954  bits<4> Zn;
1955  bits<2> imm;
1956  let Inst{20-17} = Zm;
1957  let Inst{16}    = 0b0;
1958  let Inst{9-6}   = Zn;
1959  let Inst{5}     = 0b0;
1960  let Inst{2}     = 0b0;
1961  let Inst{1-0}   = imm;
1962}
1963
1964multiclass sme2_fp_mla_long_array_vg2_multi<string mnemonic, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
1965  def _S : sme2_mla_long_array_vg2_multi<mnemonic, 0b10, op>, SMEPseudo2Instr<NAME # _S, 1>;
1966
1967  def _S_PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME # _S, uimm2s2range, ZZ_h_mul_r, SMEMatrixArray>;
1968
1969  def : SME2_ZA_TwoOp_VG2_Multi_Multi_Pat<NAME # _S, intrinsic, uimm2s2range, zpr_ty, tileslicerange2s2>;
1970
1971  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
1972                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZ_h_mul_r:$Zn, ZZ_h_mul_r:$Zm), 0>;
1973}
1974
1975multiclass sme2_int_mla_long_array_vg2_multi<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
1976  def _S : sme2_mla_long_array_vg2_multi<mnemonic, 0b11, op>, SMEPseudo2Instr<NAME # _S, 1>;
1977
1978  def _S_PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME # _S, uimm2s2range, ZZ_h_mul_r, SMEMatrixArray>;
1979
1980  def : SME2_ZA_TwoOp_VG2_Multi_Multi_Pat<NAME # _S, intrinsic, uimm2s2range, nxv8i16, tileslicerange2s2>;
1981
1982  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm2], $Zn, $Zm",
1983                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm2, ZZ_h_mul_r:$Zn, ZZ_h_mul_r:$Zm), 0>;
1984}
1985
1986class sme2_mla_long_array_vg4_multi<string mnemonic, bits<2> op0, bits<2> op>
1987   : sme2_mla_long_array<op0, op, uimm2s2range, ZZZZ_h_mul_r, ZZZZ_h_mul_r, mnemonic,
1988                        "vgx4"> {
1989  bits<3> Zm;
1990  bits<3> Zn;
1991  bits<2> imm;
1992  let Inst{20-18} = Zm;
1993  let Inst{17}    = 0b0;
1994  let Inst{16}    = 0b1;
1995  let Inst{9-7}   = Zn;
1996  let Inst{6-5}   = 0b00;
1997  let Inst{2}     = 0b0;
1998  let Inst{1-0}   = imm;
1999}
2000
2001multiclass sme2_fp_mla_long_array_vg4_multi<string mnemonic, bits<2> op, ValueType zpr_ty, SDPatternOperator intrinsic> {
2002  def _S : sme2_mla_long_array_vg4_multi<mnemonic, 0b10, op>, SMEPseudo2Instr<NAME # _S, 1>;
2003
2004  def _S_PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME # _S, uimm2s2range, ZZZZ_h_mul_r, SMEMatrixArray>;
2005
2006  def : SME2_ZA_TwoOp_VG4_Multi_Multi_Pat<NAME # _S, intrinsic, uimm2s2range, zpr_ty, tileslicerange2s2>;
2007
2008  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
2009                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm, ZZZZ_h_mul_r:$Zn, ZZZZ_h_mul_r:$Zm), 0>;
2010}
2011
2012multiclass sme2_int_mla_long_array_vg4_multi<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
2013  def _S : sme2_mla_long_array_vg4_multi<mnemonic, 0b11, op>, SMEPseudo2Instr<NAME # _S, 1>;
2014
2015  def _S_PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME # _S, uimm2s2range, ZZZZ_h_mul_r, SMEMatrixArray>;
2016
2017  def : SME2_ZA_TwoOp_VG4_Multi_Multi_Pat<NAME # _S, intrinsic, uimm2s2range, nxv8i16, tileslicerange2s2>;
2018
2019  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm2], $Zn, $Zm",
2020                 (!cast<Instruction>(NAME #_S) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm2s2range:$imm2, ZZZZ_h_mul_r:$Zn, ZZZZ_h_mul_r:$Zm), 0>;
2021}
2022
2023//===----------------------------------------------------------------------===//
2024class sme2_frint_cvt_vg2_multi<bits<2>sz, bits<5>op, RegisterOperand first_ty,
2025                               RegisterOperand second_ty, string mnemonic>
2026    : I<(outs first_ty:$Zd), (ins second_ty:$Zn),
2027        mnemonic, "\t$Zd, $Zn", "", []>, Sched<[]> {
2028  bits<4> Zn;
2029  bits<4> Zd;
2030  let Inst{31-24} = 0b11000001;
2031  let Inst{23-22} = sz;
2032  let Inst{21-20} = 0b10;
2033  let Inst{19-16} = op{4-1};
2034  let Inst{15-10} = 0b111000;
2035  let Inst{9-6}   = Zn;
2036  let Inst{5}     = op{0};
2037  let Inst{4-1}   = Zd;
2038  let Inst{0}     = 0b0;
2039}
2040
2041// SME2 multi-vec FP to int convert two registers
2042// SME2 multi-vec int to FP two registers
2043multiclass sme2_fp_cvt_vg2_multi<string mnemonic, bits<5> op> {
2044  def NAME : sme2_frint_cvt_vg2_multi<0b00, op, ZZ_s_mul_r, ZZ_s_mul_r, mnemonic>;
2045}
2046
2047// SME2 multi-vec FRINT two registers
2048multiclass sme2_frint_vector_vg2_multi<string mnemonic, bits<5> op> {
2049  def _S : sme2_frint_cvt_vg2_multi<0b10, op, ZZ_s_mul_r, ZZ_s_mul_r, mnemonic>;
2050}
2051
2052class sme2_frint_zip_cvt_vg4_multi<bits<2>sz, bits<7>op, RegisterOperand first_ty,
2053                                   RegisterOperand second_ty, string mnemonic>
2054    : I<(outs first_ty:$Zd), (ins second_ty:$Zn),
2055        mnemonic, "\t$Zd, $Zn", "", []>, Sched<[]> {
2056  bits<3> Zn;
2057  bits<3> Zd;
2058  let Inst{31-24} = 0b11000001;
2059  let Inst{23-22} = sz;
2060  let Inst{21-20} = 0b11;
2061  let Inst{19-16} = op{6-3};
2062  let Inst{15-10} = 0b111000;
2063  let Inst{9-7}   = Zn;
2064  let Inst{6-5}   = op{2-1};
2065  let Inst{4-2}   = Zd;
2066  let Inst{1}     = op{0};
2067  let Inst{0}     = 0b0;
2068}
2069
2070// SME2 multi-vec FP to int convert four registers
2071// SME2 multi-vec int to FP four registers
2072multiclass sme2_fp_cvt_vg4_multi<string mnemonic, bits<7> op> {
2073  def NAME : sme2_frint_zip_cvt_vg4_multi<0b00, op, ZZZZ_s_mul_r, ZZZZ_s_mul_r, mnemonic>;
2074}
2075
2076// SME2 multi-vec quadwords ZIP four registers
2077multiclass sme2_zip_vector_vg4<string mnemonic, bits<7> op> {
2078  def _B : sme2_frint_zip_cvt_vg4_multi<0b00, op, ZZZZ_b_mul_r, ZZZZ_b_mul_r,
2079                                        mnemonic>;
2080  def _H : sme2_frint_zip_cvt_vg4_multi<0b01, op, ZZZZ_h_mul_r, ZZZZ_h_mul_r,
2081                                        mnemonic>;
2082  def _S : sme2_frint_zip_cvt_vg4_multi<0b10, op, ZZZZ_s_mul_r, ZZZZ_s_mul_r,
2083                                        mnemonic>;
2084  def _D : sme2_frint_zip_cvt_vg4_multi<0b11, op, ZZZZ_d_mul_r, ZZZZ_d_mul_r,
2085                                         mnemonic>;
2086}
2087
2088// SME2 multi-vec quadwords ZIP four registers
2089multiclass sme2_zip_vector_vg4_Q<string mnemonic, bits<7> op> {
2090  def NAME: sme2_frint_zip_cvt_vg4_multi<0b00, op, ZZZZ_q_mul_r, ZZZZ_q_mul_r,
2091                                         mnemonic>;
2092}
2093
2094// SME2 multi-vec FRINT four registers
2095multiclass sme2_frint_vector_vg4_multi<string mnemonic, bits<7> op> {
2096  def _S :  sme2_frint_zip_cvt_vg4_multi<0b10, op, ZZZZ_s_mul_r, ZZZZ_s_mul_r,
2097                                         mnemonic>;
2098}
2099
2100class sme2_cvt_vg2_single<string mnemonic, bits<4> op>
2101    : I<(outs ZPR16:$Zd), (ins ZZ_s_mul_r:$Zn),
2102        mnemonic, "\t$Zd, $Zn", "", []>, Sched<[]> {
2103  bits<4> Zn;
2104  bits<5> Zd;
2105  let Inst{31-23} = 0b110000010;
2106  let Inst{22}    = op{3};
2107  let Inst{21-18} = 0b1000;
2108  let Inst{17-16} = op{2-1};
2109  let Inst{15-10} = 0b111000;
2110  let Inst{9-6}   = Zn;
2111  let Inst{5}     = op{0};
2112  let Inst{4-0}   = Zd;
2113}
2114
2115// SME2 multi-vec FP down convert two registers
2116// SME2 multi-vec int down convert two registers
2117multiclass sme2_cvt_vg2_single<string mnemonic, bits<4> op, ValueType out_vt,
2118                               ValueType in_vt, SDPatternOperator intrinsic> {
2119  def NAME :  sme2_cvt_vg2_single<mnemonic, op>;
2120  def : SVE2p1_Cvt_VG2_Pat<NAME, intrinsic, out_vt, in_vt>;
2121}
2122
2123class sme2_cvt_unpk_vector_vg2<bits<2>sz, bits<3> op, bit u, RegisterOperand first_ty,
2124                           RegisterOperand second_ty, string mnemonic>
2125    : I<(outs first_ty:$Zd), (ins second_ty:$Zn),
2126        mnemonic, "\t$Zd, $Zn", "", []>, Sched<[]> {
2127  bits<5> Zn;
2128  bits<4> Zd;
2129  let Inst{31-24} = 0b11000001;
2130  let Inst{23-22} = sz;
2131  let Inst{21-19} = 0b100;
2132  let Inst{18-16} = op;
2133  let Inst{15-10} = 0b111000;
2134  let Inst{9-5}   = Zn;
2135  let Inst{4-1}   = Zd;
2136  let Inst{0}     = u;
2137}
2138
2139// SME2 multi-vec unpack two registers
2140multiclass sme2_unpk_vector_vg2<string mnemonic, bit u> {
2141  def _H : sme2_cvt_unpk_vector_vg2<0b01, 0b101, u, ZZ_h_mul_r, ZPR8, mnemonic>;
2142  def _S : sme2_cvt_unpk_vector_vg2<0b10, 0b101, u, ZZ_s_mul_r, ZPR16, mnemonic>;
2143  def _D : sme2_cvt_unpk_vector_vg2<0b11, 0b101, u, ZZ_d_mul_r, ZPR32, mnemonic>;
2144}
2145
2146// SME2.1 multi-vec convert two registers
2147multiclass sme2p1_fp_cvt_vector_vg2_single<string mnemonic, bit l> {
2148  def _S : sme2_cvt_unpk_vector_vg2<0b10, 0b000, l, ZZ_s_mul_r, ZPR16, mnemonic>;
2149}
2150
2151class sme2_cvt_vg4_single<bit sz, bits<3> op, RegisterOperand first_ty,
2152                          RegisterOperand second_ty, string mnemonic>
2153    : I<(outs first_ty:$Zd), (ins second_ty:$Zn),
2154        mnemonic, "\t$Zd, $Zn", "", []>, Sched<[]> {
2155  bits<3> Zn;
2156  bits<5> Zd;
2157  let Inst{31-24} = 0b11000001;
2158  let Inst{23}    = sz;
2159  let Inst{22}    = op{2};
2160  let Inst{21-10} = 0b110011111000;
2161  let Inst{9-7}   = Zn;
2162  let Inst{6-5}   = op{1-0};
2163  let Inst{4-0}   = Zd;
2164}
2165
2166// SME2 multi-vec int down convert four registers
2167multiclass sme2_int_cvt_vg4_single<string mnemonic, bits<3> op, SDPatternOperator intrinsic> {
2168  def _StoB : sme2_cvt_vg4_single<0, op, ZPR8, ZZZZ_s_mul_r, mnemonic>;
2169  def _DtoH : sme2_cvt_vg4_single<1, op, ZPR16, ZZZZ_d_mul_r, mnemonic>;
2170
2171  def : SME2_Cvt_VG4_Pat<NAME # _StoB, intrinsic, nxv16i8, nxv4i32>;
2172  def : SME2_Cvt_VG4_Pat<NAME # _DtoH, intrinsic, nxv8i16, nxv2i64>;
2173}
2174
2175class sme2_unpk_vector_vg4<bits<2>sz, bit u, RegisterOperand first_ty,
2176                           RegisterOperand second_ty, string mnemonic>
2177    : I<(outs first_ty:$Zd), (ins second_ty:$Zn),
2178        mnemonic, "\t$Zd, $Zn", "", []>, Sched<[]> {
2179  bits<4> Zn;
2180  bits<3> Zd;
2181  let Inst{31-24} = 0b11000001;
2182  let Inst{23-22} = sz;
2183  let Inst{21-10} = 0b110101111000;
2184  let Inst{9-6}   = Zn;
2185  let Inst{5}     = 0b0;
2186  let Inst{4-2}   = Zd;
2187  let Inst{1}     = 0b0;
2188  let Inst{0}     = u;
2189}
2190
2191// SME2 multi-vec UNPK four registers
2192multiclass sme2_unpk_vector_vg4<string mnemonic, bit u> {
2193  def _H : sme2_unpk_vector_vg4<0b01, u, ZZZZ_h_mul_r, ZZ_b_mul_r, mnemonic>;
2194  def _S : sme2_unpk_vector_vg4<0b10, u, ZZZZ_s_mul_r, ZZ_h_mul_r, mnemonic>;
2195  def _D : sme2_unpk_vector_vg4<0b11, u, ZZZZ_d_mul_r, ZZ_s_mul_r, mnemonic>;
2196}
2197
2198//===----------------------------------------------------------------------===//
2199// SME2 multi-vec CLAMP registers
2200
2201class sme2_clamp_vector_vg24_multi<bits<2> sz, bits<3> op1, bit u,
2202                                   RegisterOperand multi_vector_ty,
2203                                   ZPRRegOp vector_ty, string mnemonic>
2204    : I<(outs multi_vector_ty:$Zd),
2205        (ins  multi_vector_ty:$_Zd, vector_ty:$Zn, vector_ty:$Zm),
2206        mnemonic, "\t$Zd, $Zn, $Zm",
2207        "", []>, Sched<[]>{
2208  bits<5> Zm;
2209  bits<5> Zn;
2210  let Inst{31-24} = 0b11000001;
2211  let Inst{23-22} = sz;
2212  let Inst{21}    = 0b1;
2213  let Inst{20-16} = Zm;
2214  let Inst{15-13} = 0b110;
2215  let Inst{12-10} = op1;
2216  let Inst{9-5}   = Zn;
2217  let Inst{0}     = u;
2218
2219  let Constraints = "$Zd = $_Zd";
2220}
2221
2222class sme2_clamp_vector_vg2_multi<bits<2> sz, bits<3> op1, bit u,
2223                                  RegisterOperand multi_vector_ty,
2224                                  ZPRRegOp vector_ty, string mnemonic>
2225    : sme2_clamp_vector_vg24_multi<sz, op1, u, multi_vector_ty, vector_ty,
2226                                   mnemonic>{
2227  bits<4> Zd;
2228  let Inst{4-1} = Zd;
2229}
2230
2231multiclass sme2_fp_clamp_vector_vg2_multi<string mnemonic>{
2232  def _H : sme2_clamp_vector_vg2_multi<0b01, 0b000, 0b0, ZZ_h_mul_r, ZPR16, mnemonic>;
2233  def _S : sme2_clamp_vector_vg2_multi<0b10, 0b000, 0b0, ZZ_s_mul_r, ZPR32, mnemonic>;
2234  def _D : sme2_clamp_vector_vg2_multi<0b11, 0b000, 0b0, ZZ_d_mul_r, ZPR64, mnemonic>;
2235}
2236
2237multiclass sme2_int_clamp_vector_vg2_multi<string mnemonic, bit u>{
2238  def _B : sme2_clamp_vector_vg2_multi<0b00, 0b001, u, ZZ_b_mul_r, ZPR8, mnemonic>;
2239  def _H : sme2_clamp_vector_vg2_multi<0b01, 0b001, u, ZZ_h_mul_r, ZPR16, mnemonic>;
2240  def _S : sme2_clamp_vector_vg2_multi<0b10, 0b001, u, ZZ_s_mul_r, ZPR32, mnemonic>;
2241  def _D : sme2_clamp_vector_vg2_multi<0b11, 0b001, u, ZZ_d_mul_r, ZPR64, mnemonic>;
2242}
2243
2244// SME2.1 multi-vec FCLAMP two registers
2245multiclass sme2p1_bfclamp_vector_vg2_multi<string mnemonic> {
2246  def _H : sme2_clamp_vector_vg2_multi<0b00, 0b000, 0b0, ZZ_h_mul_r, ZPR16,
2247                                           mnemonic>;
2248}
2249
2250class sme2_clamp_vector_vg4_multi<bits<2> sz, bits<3> op1, bit u,
2251                                  RegisterOperand multi_vector_ty,
2252                                  ZPRRegOp vector_ty, string mnemonic>
2253    : sme2_clamp_vector_vg24_multi<sz, op1, u,  multi_vector_ty, vector_ty,
2254                                   mnemonic>{
2255  bits<3> Zd;
2256  let Inst{4-2} = Zd;
2257  let Inst{1}   = 0b0;
2258}
2259
2260multiclass sme2_fp_clamp_vector_vg4_multi<string mnemonic>{
2261  def _H : sme2_clamp_vector_vg4_multi<0b01, 0b010, 0b0, ZZZZ_h_mul_r, ZPR16, mnemonic>;
2262  def _S : sme2_clamp_vector_vg4_multi<0b10, 0b010, 0b0, ZZZZ_s_mul_r, ZPR32, mnemonic>;
2263  def _D : sme2_clamp_vector_vg4_multi<0b11, 0b010, 0b0, ZZZZ_d_mul_r, ZPR64, mnemonic>;
2264}
2265
2266multiclass sme2_int_clamp_vector_vg4_multi<string mnemonic, bit u>{
2267  def _B : sme2_clamp_vector_vg4_multi<0b00, 0b011, u, ZZZZ_b_mul_r, ZPR8, mnemonic>;
2268  def _H : sme2_clamp_vector_vg4_multi<0b01, 0b011, u, ZZZZ_h_mul_r, ZPR16, mnemonic>;
2269  def _S : sme2_clamp_vector_vg4_multi<0b10, 0b011, u, ZZZZ_s_mul_r, ZPR32, mnemonic>;
2270  def _D : sme2_clamp_vector_vg4_multi<0b11, 0b011, u, ZZZZ_d_mul_r, ZPR64, mnemonic>;
2271}
2272
2273// SME2.1 multi-vec FCLAMP four registers
2274multiclass sme2p1_bfclamp_vector_vg4_multi<string mnemonic> {
2275  def _H : sme2_clamp_vector_vg4_multi<0b00, 0b010, 0b0, ZZZZ_h_mul_r, ZPR16,
2276                                       mnemonic>;
2277}
2278
2279// SME2 multi-vec ZIP two registers
2280class sme2_zip_vector_vg2<bits<2> sz, bit q, bit u,
2281                         RegisterOperand multi_vector_ty,
2282                         ZPRRegOp vector_ty, string mnemonic>
2283    : I<(outs multi_vector_ty:$Zd), (ins vector_ty:$Zn, vector_ty:$Zm),
2284        mnemonic, "\t$Zd, $Zn, $Zm",
2285        "", []>, Sched<[]>{
2286  bits<4> Zd;
2287  bits<5> Zm;
2288  bits<5> Zn;
2289  let Inst{31-24} = 0b11000001;
2290  let Inst{23-22} = sz;
2291  let Inst{21}    = 0b1;
2292  let Inst{20-16} = Zm;
2293  let Inst{15-11} = 0b11010;
2294  let Inst{10}    = q;
2295  let Inst{9-5}   = Zn;
2296  let Inst{4-1}   = Zd;
2297  let Inst{0}     = u;
2298}
2299
2300multiclass sme2_zip_vector_vg2<string mnemonic, bit op> {
2301  def _B : sme2_zip_vector_vg2<0b00, 0b0, op, ZZ_b_mul_r, ZPR8, mnemonic>;
2302  def _H : sme2_zip_vector_vg2<0b01, 0b0, op, ZZ_h_mul_r, ZPR16, mnemonic>;
2303  def _S : sme2_zip_vector_vg2<0b10, 0b0, op, ZZ_s_mul_r, ZPR32, mnemonic>;
2304  def _D : sme2_zip_vector_vg2<0b11, 0b0, op, ZZ_d_mul_r, ZPR64, mnemonic>;
2305  def _Q : sme2_zip_vector_vg2<0b00, 0b1, op, ZZ_q_mul_r, ZPR128, mnemonic>;
2306}
2307
2308//===----------------------------------------------------------------------===//
2309// SME2 Dot Products and MLA
2310
2311class sme2_multi_vec_array_vg2_index<bit sz, bits<6> op, MatrixOperand matrix_ty,
2312                                     RegisterOperand multi_vector_ty,
2313                                     ZPRRegOp vector_ty, Operand index_ty,
2314                                     string mnemonic>
2315    : I<(outs matrix_ty:$ZAda),
2316        (ins matrix_ty:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2317         multi_vector_ty:$Zn, vector_ty:$Zm, index_ty:$i),
2318         mnemonic, "\t$ZAda[$Rv, $imm3, vgx2], $Zn, $Zm$i",
2319        "", []>, Sched<[]> {
2320  bits<4> Zm;
2321  bits<2> Rv;
2322  bits<4> Zn;
2323  bits<3> imm3;
2324  let Inst{31-23} = 0b110000010;
2325  let Inst{22}    = sz;
2326  let Inst{21-20} = 0b01;
2327  let Inst{19-16} = Zm;
2328  let Inst{15}    = 0b0;
2329  let Inst{14-13} = Rv;
2330  let Inst{12-10} = op{5-3};
2331  let Inst{9-6}   = Zn;
2332  let Inst{5-3}   = op{2-0};
2333  let Inst{2-0}   = imm3;
2334
2335  let Constraints = "$ZAda = $_ZAda";
2336}
2337
2338// SME2 multi-vec ternary indexed two registers 32-bit
2339multiclass sme2_multi_vec_array_vg2_index_32b<string mnemonic, bits<4> op,
2340                                              RegisterOperand multi_vector_ty,
2341                                              ZPRRegOp vector_ty, ValueType vt,
2342                                              SDPatternOperator intrinsic> {
2343  def NAME : sme2_multi_vec_array_vg2_index<0b1, {op{3},?,?,op{2-0}}, MatrixOp32, multi_vector_ty, vector_ty,
2344                                             VectorIndexS32b_timm,  mnemonic>, SMEPseudo2Instr<NAME, 1> {
2345    bits<2> i;
2346    let Inst{11-10} = i;
2347  }
2348
2349  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, vector_ty, VectorIndexS32b_timm, SMEMatrixArray>;
2350
2351  def : SME2_ZA_TwoOp_VG2_Multi_Index_Pat<NAME, intrinsic, sme_elm_idx0_7, vector_ty, vt, VectorIndexS32b_timm, tileslice16>;
2352
2353  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm3], $Zn, $Zm$i",
2354        (!cast<Instruction>(NAME) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2355        multi_vector_ty:$Zn, vector_ty:$Zm, VectorIndexS32b_timm:$i), 0>;
2356}
2357
2358// SME2.1 multi-vec ternary indexed two registers 16-bit
2359multiclass sme2p1_multi_vec_array_vg2_index_16b<string mnemonic, bits<2> op> {
2360  def NAME : sme2_multi_vec_array_vg2_index<0b0, {0b1,?,?,op,?}, MatrixOp16,
2361                                            ZZ_h_mul_r, ZPR4b16,
2362                                            VectorIndexH, mnemonic> {
2363    bits<3> i;
2364    let Inst{11-10} = i{2-1};
2365    let Inst{3}     = i{0};
2366  }
2367  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm3], $Zn, $Zm$i",
2368        (!cast<Instruction>(NAME) MatrixOp16:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2369        ZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH:$i), 0>;
2370}
2371
2372// SME2 multi-vec ternary indexed two registers 64-bit
2373
2374class sme2_multi_vec_array_vg2_index_64b<bits<2> op,
2375                                         RegisterOperand multi_vector_ty,
2376                                         ZPRRegOp vector_ty,
2377                                         string mnemonic>
2378    : I<(outs MatrixOp64:$ZAda),
2379        (ins MatrixOp64:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2380         multi_vector_ty:$Zn, vector_ty:$Zm, VectorIndexD32b_timm:$i1),
2381        mnemonic, "\t$ZAda[$Rv, $imm3, vgx2], $Zn, $Zm$i1",
2382        "", []>, Sched<[]> {
2383  bits<4> Zm;
2384  bits<2> Rv;
2385  bits<1> i1;
2386  bits<4> Zn;
2387  bits<3> imm3;
2388  let Inst{31-20} = 0b110000011101;
2389  let Inst{19-16} = Zm;
2390  let Inst{15}    = 0b0;
2391  let Inst{14-13} = Rv;
2392  let Inst{12-11} = 0b00;
2393  let Inst{10}    = i1;
2394  let Inst{9-6}   = Zn;
2395  let Inst{5}     = 0b0;
2396  let Inst{4-3}   = op;
2397  let Inst{2-0}   = imm3;
2398
2399  let Constraints = "$ZAda = $_ZAda";
2400}
2401
2402multiclass sme2_multi_vec_array_vg2_index_64b<string mnemonic, bits<2> op,
2403                                              RegisterOperand multi_vector_ty,
2404                                              ZPRRegOp vector_ty, ValueType vt,
2405                                              SDPatternOperator intrinsic> {
2406  def NAME : sme2_multi_vec_array_vg2_index_64b<op, multi_vector_ty, vector_ty,
2407                                                mnemonic>, SMEPseudo2Instr<NAME, 1>;
2408
2409  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, vector_ty, VectorIndexD32b_timm, SMEMatrixArray>;
2410
2411  def : SME2_ZA_TwoOp_VG2_Multi_Index_Pat<NAME, intrinsic, sme_elm_idx0_7, vector_ty, vt, VectorIndexD32b_timm, tileslice16>;
2412
2413  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm3], $Zn, $Zm$i1",
2414        (!cast<Instruction>(NAME) MatrixOp64:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2415        multi_vector_ty:$Zn, vector_ty:$Zm, VectorIndexD32b_timm:$i1), 0>;
2416}
2417
2418class sme2_multi_vec_array_vg4_index<bit sz, bits<6> op, MatrixOperand matrix_ty,
2419                                     RegisterOperand multi_vector_ty,
2420                                     ZPRRegOp vector_ty, Operand index_ty,
2421                                     string mnemonic>
2422    : I<(outs matrix_ty:$ZAda),
2423        (ins matrix_ty:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2424         multi_vector_ty:$Zn, vector_ty:$Zm, index_ty:$i),
2425         mnemonic, "\t$ZAda[$Rv, $imm3, vgx4], $Zn, $Zm$i",
2426        "", []>, Sched<[]> {
2427  bits<4> Zm;
2428  bits<2> Rv;
2429  bits<3> Zn;
2430  bits<3> imm3;
2431  let Inst{31-23} = 0b110000010;
2432  let Inst{22}    = sz;
2433  let Inst{21-20} = 0b01;
2434  let Inst{19-16} = Zm;
2435  let Inst{15}    = 0b1;
2436  let Inst{14-13} = Rv;
2437  let Inst{12-10} = op{5-3};
2438  let Inst{9-7}   = Zn;
2439  let Inst{6}     = 0b0;
2440  let Inst{5-3}   = op{2-0};
2441  let Inst{2-0}   = imm3;
2442
2443  let Constraints = "$ZAda = $_ZAda";
2444}
2445
2446// SME2 multi-vec ternary indexed four registers 32-bit
2447multiclass sme2_multi_vec_array_vg4_index_32b<string mnemonic, bits<4> op,
2448                                              RegisterOperand multi_vector_ty,
2449                                              ZPRRegOp vector_ty, ValueType vt,
2450                                              SDPatternOperator intrinsic> {
2451  def NAME : sme2_multi_vec_array_vg4_index<0b1, {op{3},?,?,op{2-0}}, MatrixOp32,  multi_vector_ty,
2452                                            vector_ty, VectorIndexS32b_timm, mnemonic>, SMEPseudo2Instr<NAME, 1> {
2453   bits<2> i;
2454   let Inst{11-10} = i;
2455  }
2456
2457  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, vector_ty, VectorIndexS32b_timm, SMEMatrixArray>;
2458
2459  def : SME2_ZA_TwoOp_VG4_Multi_Index_Pat<NAME, intrinsic, sme_elm_idx0_7, vector_ty, vt, VectorIndexS32b_timm, tileslice16>;
2460
2461  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm3], $Zn, $Zm$i",
2462        (!cast<Instruction>(NAME) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2463        multi_vector_ty:$Zn, vector_ty:$Zm, VectorIndexS32b_timm:$i), 0>;
2464}
2465
2466// SME2.1 multi-vec ternary indexed four registers 16-bit
2467multiclass sme2p1_multi_vec_array_vg4_index_16b<string mnemonic, bits<2> op> {
2468  def NAME : sme2_multi_vec_array_vg4_index<0b0,{0b1,?,?,op,?}, MatrixOp16,
2469                                            ZZZZ_h_mul_r, ZPR4b16,
2470                                            VectorIndexH, mnemonic>{
2471    bits<3> i;
2472    let Inst{11-10} = i{2-1};
2473    let Inst{3}     = i{0};
2474  }
2475
2476  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm3], $Zn, $Zm$i",
2477        (!cast<Instruction>(NAME) MatrixOp16:$ZAda,  MatrixIndexGPR32Op8_11:$Rv,
2478        sme_elm_idx0_7:$imm3, ZZZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH:$i), 0>;
2479}
2480
2481// SME2 multi-vec ternary indexed four registers 64-bit
2482class sme2_multi_vec_array_vg4_index_64b<bits<3> op,
2483                                         RegisterOperand multi_vector_ty,
2484                                         ZPRRegOp vector_ty,
2485                                         string mnemonic>
2486    : I<(outs MatrixOp64:$ZAda),
2487        (ins MatrixOp64:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2488         multi_vector_ty:$Zn, vector_ty:$Zm, VectorIndexD32b_timm:$i1),
2489        mnemonic, "\t$ZAda[$Rv, $imm3, vgx4], $Zn, $Zm$i1",
2490        "", []>, Sched<[]> {
2491  bits<4> Zm;
2492  bits<2> Rv;
2493  bits<1> i1;
2494  bits<3> Zn;
2495  bits<3> imm3;
2496  let Inst{31-20} = 0b110000011101;
2497  let Inst{19-16} = Zm;
2498  let Inst{15}    = 0b1;
2499  let Inst{14-13} = Rv;
2500  let Inst{12}    = 0b0;
2501  let Inst{11}    = op{2};
2502  let Inst{10}    = i1;
2503  let Inst{9-7}   = Zn;
2504  let Inst{6-5}   = 0b00;
2505  let Inst{4-3}   = op{1-0};
2506  let Inst{2-0}   = imm3;
2507
2508  let Constraints = "$ZAda = $_ZAda";
2509}
2510
2511multiclass sme2_multi_vec_array_vg4_index_64b<string mnemonic, bits<3> op,
2512                                              RegisterOperand multi_vector_ty,
2513                                              ZPRRegOp vector_ty, ValueType vty,
2514                                              SDPatternOperator intrinsic> {
2515  def NAME : sme2_multi_vec_array_vg4_index_64b<op, multi_vector_ty, vector_ty,
2516                                                mnemonic>, SMEPseudo2Instr<NAME, 1>;
2517
2518  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, sme_elm_idx0_7, multi_vector_ty, vector_ty, VectorIndexD32b_timm, SMEMatrixArray>;
2519
2520  def : SME2_ZA_TwoOp_VG4_Multi_Index_Pat<NAME, intrinsic, sme_elm_idx0_7, vector_ty, vty, VectorIndexD32b_timm, tileslice16>;
2521
2522  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm3], $Zn, $Zm$i1",
2523        (!cast<Instruction>(NAME) MatrixOp64:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, sme_elm_idx0_7:$imm3,
2524        multi_vector_ty:$Zn, vector_ty:$Zm, VectorIndexD32b_timm:$i1), 0>;
2525}
2526//===----------------------------------------------------------------------===//
2527// SME2 multi-vec indexed long long MLA one source 32-bit
2528class sme2_mla_ll_array_index_32b<string mnemonic, bits<3> op>
2529    : I<(outs MatrixOp32:$ZAda),
2530        (ins MatrixOp32:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm2s4range:$imm2, ZPR8:$Zn, ZPR4b8:$Zm, VectorIndexB32b_timm:$i),
2531        mnemonic, "\t$ZAda[$Rv, $imm2], $Zn, $Zm$i",
2532        "", []>, Sched<[]> {
2533  bits<4> Zm;
2534  bits<2> Rv;
2535  bits<4> i;
2536  bits<5> Zn;
2537  bits<2> imm2;
2538  let Inst{31-20} = 0b110000010000;
2539  let Inst{19-16} = Zm;
2540  let Inst{15}    = i{3};
2541  let Inst{14-13} = Rv;
2542  let Inst{12-10} = i{2-0};
2543  let Inst{9-5}   = Zn;
2544  let Inst{4-2}   = op;
2545  let Inst{1-0}   = imm2;
2546
2547  let Constraints = "$ZAda = $_ZAda";
2548}
2549
2550multiclass sme2_mla_ll_array_index_32b<string mnemonic, bits<3> op, SDPatternOperator intrinsic> {
2551  def NAME : sme2_mla_ll_array_index_32b<mnemonic, op>, SMEPseudo2Instr<NAME, 1>;
2552
2553  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, uimm2s4range, ZPR8, ZPR4b8, VectorIndexB32b_timm, SMEMatrixArray>;
2554
2555  def : SME2_ZA_TwoOp_Multi_Index_Pat<NAME, intrinsic, uimm2s4range, ZPR4b8, nxv16i8, VectorIndexB32b_timm, tileslicerange2s4>;
2556}
2557
2558// SME2 multi-vec indexed long long MLA one source 64-bit
2559
2560class sme2_mla_ll_array_index_64b<string mnemonic, bits<2> op>
2561    : I<(outs MatrixOp64:$ZAda),
2562        (ins MatrixOp64:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm2s4range:$imm2, ZPR16:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i),
2563        mnemonic, "\t$ZAda[$Rv, $imm2], $Zn, $Zm$i",
2564        "", []>, Sched<[]> {
2565  bits<4> Zm;
2566  bits<2> Rv;
2567  bits<3> i;
2568  bits<5> Zn;
2569  bits<2> imm2;
2570  let Inst{31-20} = 0b110000011000;
2571  let Inst{19-16} = Zm;
2572  let Inst{15}    = i{2};
2573  let Inst{14-13} = Rv;
2574  let Inst{12}    = 0b0;
2575  let Inst{11-10} = i{1-0};
2576  let Inst{9-5}   = Zn;
2577  let Inst{4-3}   = op;
2578  let Inst{2}     = 0b0;
2579  let Inst{1-0}   = imm2;
2580
2581  let Constraints = "$ZAda = $_ZAda";
2582}
2583
2584multiclass sme2_mla_ll_array_index_64b<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
2585  def NAME : sme2_mla_ll_array_index_64b<mnemonic, op>, SMEPseudo2Instr<NAME, 1>;
2586
2587  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, uimm2s4range, ZPR16, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
2588
2589  def : SME2_ZA_TwoOp_Multi_Index_Pat<NAME, intrinsic, uimm2s4range, ZPR4b16, nxv8i16, VectorIndexH32b_timm, tileslicerange2s4>;
2590}
2591
2592class sme2_mla_ll_array_vg24_index_32b<bit vg4, bits<3> op,
2593                                       RegisterOperand vector_ty,
2594                                       string mnemonic>
2595    : I<(outs MatrixOp32:$ZAda),
2596        (ins MatrixOp32:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm,
2597             vector_ty:$Zn, ZPR4b8:$Zm, VectorIndexB32b_timm:$i),
2598        mnemonic, "\t$ZAda[$Rv, $imm, " # !if(vg4, "vgx4", "vgx2") # "], $Zn, $Zm$i",
2599        "", []>, Sched<[]> {
2600  bits<4> Zm;
2601  bits<2> Rv;
2602  bits<4> i;
2603  bit     imm;
2604  let Inst{31-20} = 0b110000010001;
2605  let Inst{19-16} = Zm;
2606  let Inst{15}    = vg4;
2607  let Inst{14-13} = Rv;
2608  let Inst{12}    = 0b0;
2609  let Inst{11-10} = i{3-2};
2610  let Inst{5-3}   = op;
2611  let Inst{2-1}   = i{1-0};
2612  let Inst{0}     = imm;
2613
2614  let Constraints = "$ZAda = $_ZAda";
2615}
2616
2617//SME2 multi-vec indexed long long MLA two sources 32-bit
2618
2619multiclass sme2_mla_ll_array_vg2_index_32b<string mnemonic, bits<3> op, SDPatternOperator intrinsic> {
2620  def NAME: sme2_mla_ll_array_vg24_index_32b<0b0, op, ZZ_b_mul_r, mnemonic>, SMEPseudo2Instr<NAME, 1> {
2621   bits<4> Zn;
2622   let Inst{9-6} = Zn;
2623  }
2624
2625  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, uimm1s4range, ZZ_b_mul_r, ZPR4b8, VectorIndexB32b_timm, SMEMatrixArray>;
2626
2627  def : SME2_ZA_TwoOp_VG2_Multi_Index_Pat<NAME, intrinsic, uimm1s4range, ZPR4b8, nxv16i8, VectorIndexB32b_timm, tileslicerange1s4>;
2628
2629  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i",
2630                 (!cast<Instruction>(NAME) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm, ZZ_b_mul_r:$Zn, ZPR4b8:$Zm, VectorIndexB32b_timm:$i), 0>;
2631}
2632
2633// SME2 multi-vec indexed long long MLA four sources 32-bit
2634
2635multiclass sme2_mla_ll_array_vg4_index_32b<string mnemonic, bits<3> op, SDPatternOperator intrinsic> {
2636  def NAME: sme2_mla_ll_array_vg24_index_32b<0b1, op, ZZZZ_b_mul_r, mnemonic>, SMEPseudo2Instr<NAME, 1> {
2637   bits<3> Zn;
2638   let Inst{9-7} = Zn;
2639   let Inst{6}   = 0b0;
2640  }
2641
2642  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, uimm1s4range, ZZZZ_b_mul_r, ZPR4b8, VectorIndexB32b_timm, SMEMatrixArray>;
2643
2644  def : SME2_ZA_TwoOp_VG4_Multi_Index_Pat<NAME, intrinsic, uimm1s4range, ZPR4b8, nxv16i8, VectorIndexB32b_timm, tileslicerange1s4>;
2645
2646  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i",
2647                 (!cast<Instruction>(NAME) MatrixOp32:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm, ZZZZ_b_mul_r:$Zn, ZPR4b8:$Zm, VectorIndexB32b_timm:$i), 0>;
2648}
2649class sme2_mla_ll_array_vg24_index_64b<bit vg4,  bits<2> op,
2650                                       RegisterOperand vector_ty,
2651                                       string mnemonic>
2652    : I<(outs MatrixOp64:$ZAda),
2653        (ins MatrixOp64:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm,
2654             vector_ty:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i),
2655        mnemonic, "\t$ZAda[$Rv, $imm, " # !if(vg4, "vgx4", "vgx2") # "], $Zn, $Zm$i",
2656        "", []>, Sched<[]> {
2657  bits<4> Zm;
2658  bits<2> Rv;
2659  bits<3> i;
2660  bit     imm;
2661  let Inst{31-20} = 0b110000011001;
2662  let Inst{19-16} = Zm;
2663  let Inst{15}    = vg4;
2664  let Inst{14-13} = Rv;
2665  let Inst{12-11} = 0b00;
2666  let Inst{10}    = i{2};
2667  let Inst{5}     = 0b0;
2668  let Inst{4-3}   = op;
2669  let Inst{2-1}   = i{1-0};
2670  let Inst{0}     = imm;
2671
2672  let Constraints = "$ZAda = $_ZAda";
2673}
2674
2675// SME2 multi-vec indexed long long MLA two sources 64-bit
2676
2677multiclass sme2_mla_ll_array_vg2_index_64b<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
2678  def NAME: sme2_mla_ll_array_vg24_index_64b<0b0, op, ZZ_h_mul_r, mnemonic>, SMEPseudo2Instr<NAME, 1> {
2679    bits<4> Zn;
2680    let Inst{9-6} = Zn;
2681  }
2682
2683  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, uimm1s4range, ZZ_h_mul_r, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
2684
2685  def : SME2_ZA_TwoOp_VG2_Multi_Index_Pat<NAME, intrinsic, uimm1s4range, ZPR4b16, nxv8i16, VectorIndexH32b_timm, tileslicerange1s4>;
2686
2687  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i",
2688                 (!cast<Instruction>(NAME) MatrixOp64:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm, ZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i), 0>;
2689}
2690
2691// SME2 multi-vec indexed long long MLA four sources 64-bit
2692
2693multiclass sme2_mla_ll_array_vg4_index_64b<string mnemonic, bits<2> op, SDPatternOperator intrinsic> {
2694  def NAME: sme2_mla_ll_array_vg24_index_64b<0b1, op, ZZZZ_h_mul_r,  mnemonic>, SMEPseudo2Instr<NAME, 1> {
2695    bits<3> Zn;
2696    let Inst{9-7} = Zn;
2697    let Inst{6}   = 0b0;
2698  }
2699
2700  def _PSEUDO : sme2_za_array_2op_multi_index_pseudo<NAME, uimm1s4range, ZZZZ_h_mul_r, ZPR4b16, VectorIndexH32b_timm, SMEMatrixArray>;
2701
2702  def : SME2_ZA_TwoOp_VG4_Multi_Index_Pat<NAME, intrinsic, uimm1s4range, ZPR4b16, nxv8i16, VectorIndexH32b_timm, tileslicerange1s4>;
2703
2704  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm$i",
2705                 (!cast<Instruction>(NAME) MatrixOp64:$ZAda,  MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm, ZZZZ_h_mul_r:$Zn, ZPR4b16:$Zm, VectorIndexH32b_timm:$i), 0>;
2706}
2707
2708
2709//SME2 multiple and single vector long long FMA one source
2710
2711class sme2_mla_ll_array_single<string mnemonic, bits<4> op,
2712                               MatrixOperand matrix_ty, ZPRRegOp vector_ty,
2713                               ZPRRegOp zpr_ty>
2714    : I<(outs matrix_ty:$ZAda),
2715        (ins matrix_ty:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm2s4range:$imm,
2716             vector_ty:$Zn, zpr_ty:$Zm),
2717        mnemonic, "\t$ZAda[$Rv, $imm], $Zn, $Zm",
2718        "", []>, Sched<[]> {
2719  bits<4> Zm;
2720  bits<2> Rv;
2721  bits<5> Zn;
2722  bits<2> imm;
2723  let Inst{31-23} = 0b110000010;
2724  let Inst{22}    = op{3}; //sz
2725  let Inst{21-20} = 0b10;
2726  let Inst{19-16} = Zm;
2727  let Inst{15}    = 0b0;
2728  let Inst{14-13} = Rv;
2729  let Inst{12-10} = 0b001;
2730  let Inst{9-5}   = Zn;
2731  let Inst{4-2}   = op{2-0};
2732  let Inst{1-0}   = imm;
2733
2734  let Constraints = "$ZAda = $_ZAda";
2735}
2736
2737multiclass sme2_mla_ll_array_single<string mnemonic, bits<4> op,
2738                                    MatrixOperand matrix_ty, ZPRRegOp vector_ty,
2739                                    ZPRRegOp zpr_ty, ValueType vt, SDPatternOperator intrinsic> {
2740  def NAME : sme2_mla_ll_array_single<mnemonic, op, matrix_ty, vector_ty, zpr_ty>, SMEPseudo2Instr<NAME, 1>;
2741
2742  def NAME # _PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME, uimm2s4range, vector_ty, zpr_ty, SMEMatrixArray>;
2743
2744  def : SME2_ZA_TwoOp_Multi_Single_Pat<NAME, intrinsic, uimm2s4range, zpr_ty, vt, tileslicerange2s4>;
2745}
2746
2747class sme2_mla_ll_array_vg24_single<bits<5> op, MatrixOperand matrix_ty,
2748                                    RegisterOperand vector_ty, ZPRRegOp zpr_ty,
2749                                    string mnemonic>
2750    : I<(outs matrix_ty:$ZAda),
2751        (ins matrix_ty:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm,
2752             vector_ty:$Zn, zpr_ty:$Zm),
2753        mnemonic, "\t$ZAda[$Rv, $imm,  " # !if(op{3}, "vgx4", "vgx2") # "], $Zn, $Zm",
2754        "", []>, Sched<[]> {
2755  bits<4> Zm;
2756  bits<2> Rv;
2757  bits<5> Zn;
2758  bit     imm;
2759  let Inst{31-23} = 0b110000010;
2760  let Inst{22}    = op{4}; //sz
2761  let Inst{21}    = 0b1;
2762  let Inst{20}    = op{3}; //vg4
2763  let Inst{19-16} = Zm;
2764  let Inst{15}    = 0b0;
2765  let Inst{14-13} = Rv;
2766  let Inst{12-10} = 0b000;
2767  let Inst{9-5}   = Zn;
2768  let Inst{4-2}   = op{2-0};
2769  let Inst{1}     = 0b0;
2770  let Inst{0}     = imm;
2771
2772  let Constraints = "$ZAda = $_ZAda";
2773}
2774
2775//SME2 single-multi long long MLA two and four sources
2776
2777multiclass sme2_mla_ll_array_vg24_single<string mnemonic, bits<5> op,
2778                                          MatrixOperand matrix_ty,
2779                                          RegisterOperand multi_vector_ty,
2780                                          ZPRRegOp zpr_ty> {
2781  def NAME: sme2_mla_ll_array_vg24_single<op, matrix_ty, multi_vector_ty,
2782                                          zpr_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
2783
2784  def NAME # _PSEUDO : sme2_za_array_2op_multi_single_pseudo<NAME, uimm1s4range, multi_vector_ty, zpr_ty, SMEMatrixArray>;
2785
2786  def : InstAlias<mnemonic # "\t$ZAd[$Rv, $imm], $Zn, $Zm",
2787                 (!cast<Instruction>(NAME) matrix_ty:$ZAd,  MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm, multi_vector_ty:$Zn, zpr_ty:$Zm), 0>;
2788}
2789
2790multiclass sme2_mla_ll_array_vg2_single<string mnemonic, bits<5> op,
2791                                        MatrixOperand matrix_ty,
2792                                        RegisterOperand multi_vector_ty,
2793                                        ZPRRegOp zpr_ty, ValueType vt, SDPatternOperator intrinsic> {
2794
2795  defm NAME: sme2_mla_ll_array_vg24_single<mnemonic, op, matrix_ty, multi_vector_ty, zpr_ty>;
2796
2797  def : SME2_ZA_TwoOp_VG2_Multi_Single_Pat<NAME, intrinsic, uimm1s4range, zpr_ty, vt, tileslicerange1s4>;
2798}
2799
2800multiclass sme2_mla_ll_array_vg4_single<string mnemonic, bits<5> op,
2801                                        MatrixOperand matrix_ty,
2802                                        RegisterOperand multi_vector_ty,
2803                                        ZPRRegOp zpr_ty, ValueType vt, SDPatternOperator intrinsic> {
2804  defm NAME: sme2_mla_ll_array_vg24_single<mnemonic, op, matrix_ty, multi_vector_ty, zpr_ty>;
2805
2806  def : SME2_ZA_TwoOp_VG4_Multi_Single_Pat<NAME, intrinsic, uimm1s4range, zpr_ty, vt, tileslicerange1s4>;
2807}
2808
2809// SME2 multiple vectors long long MLA two sources
2810
2811class sme2_mla_ll_array_vg2_multi<bits<4> op, MatrixOperand matrix_ty,
2812                                  RegisterOperand vector_ty,string mnemonic>
2813    : I<(outs matrix_ty:$ZAda),
2814        (ins matrix_ty:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm,
2815             vector_ty:$Zn, vector_ty:$Zm),
2816        mnemonic, "\t$ZAda[$Rv, $imm, vgx2], $Zn, $Zm",
2817        "", []>, Sched<[]> {
2818  bits<4> Zm;
2819  bits<2> Rv;
2820  bits<4> Zn;
2821  bit     imm;
2822  let Inst{31-23} = 0b110000011;
2823  let Inst{22}    = op{3};  // sz
2824  let Inst{21}    = 0b1;
2825  let Inst{20-17} = Zm;
2826  let Inst{16-15} = 0b00;
2827  let Inst{14-13} = Rv;
2828  let Inst{12-10} = 0b000;
2829  let Inst{9-6}   = Zn;
2830  let Inst{5}     = 0b0;
2831  let Inst{4-2}   = op{2-0};
2832  let Inst{1}     = 0b0;
2833  let Inst{0}     = imm;
2834
2835  let Constraints = "$ZAda = $_ZAda";
2836}
2837
2838multiclass sme2_mla_ll_array_vg2_multi<string mnemonic, bits<4> op,
2839                                       MatrixOperand matrix_ty,
2840                                       RegisterOperand vector_ty,
2841                                       ValueType vt, SDPatternOperator intrinsic> {
2842  def NAME : sme2_mla_ll_array_vg2_multi<op, matrix_ty, vector_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
2843
2844  def _PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME, uimm1s4range, vector_ty, SMEMatrixArray>;
2845
2846  def : SME2_ZA_TwoOp_VG2_Multi_Multi_Pat<NAME, intrinsic, uimm1s4range, vt, tileslicerange1s4>;
2847
2848  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
2849                 (!cast<Instruction>(NAME) matrix_ty:$ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm, vector_ty:$Zn, vector_ty:$Zm), 0>;
2850}
2851
2852// SME2 multiple vectors long long MLA four sources
2853
2854class sme2_mla_ll_array_vg4_multi<bits<4> op,MatrixOperand matrix_ty,
2855                                  RegisterOperand vector_ty,
2856                                  string mnemonic>
2857    : I<(outs matrix_ty:$ZAda),
2858        (ins matrix_ty:$_ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm,
2859             vector_ty:$Zn, vector_ty:$Zm),
2860        mnemonic, "\t$ZAda[$Rv, $imm, vgx4], $Zn, $Zm",
2861        "", []>, Sched<[]> {
2862  bits<3> Zm;
2863  bits<2> Rv;
2864  bits<3> Zn;
2865  bit     imm;
2866  let Inst{31-23} = 0b110000011;
2867  let Inst{22}    = op{3}; // sz
2868  let Inst{21}    = 0b1;
2869  let Inst{20-18} = Zm;
2870  let Inst{17-15} = 0b010;
2871  let Inst{14-13} = Rv;
2872  let Inst{12-10} = 0b000;
2873  let Inst{9-7}   = Zn;
2874  let Inst{6-5}   = 0b00;
2875  let Inst{4-2}   = op{2-0};
2876  let Inst{1}     = 0b0;
2877  let Inst{0}     = imm;
2878
2879  let Constraints = "$ZAda = $_ZAda";
2880}
2881
2882multiclass sme2_mla_ll_array_vg4_multi<string mnemonic, bits<4> op,
2883                                       MatrixOperand matrix_ty,
2884                                       RegisterOperand vector_ty,
2885                                       ValueType vt, SDPatternOperator intrinsic> {
2886  def NAME : sme2_mla_ll_array_vg4_multi<op, matrix_ty, vector_ty, mnemonic>, SMEPseudo2Instr<NAME, 1>;
2887
2888  def _PSEUDO : sme2_za_array_2op_multi_multi_pseudo<NAME, uimm1s4range, vector_ty, SMEMatrixArray>;
2889
2890  def : SME2_ZA_TwoOp_VG4_Multi_Multi_Pat<NAME, intrinsic, uimm1s4range, vt, tileslicerange1s4>;
2891
2892  def : InstAlias<mnemonic # "\t$ZAda[$Rv, $imm], $Zn, $Zm",
2893                 (!cast<Instruction>(NAME) matrix_ty:$ZAda, MatrixIndexGPR32Op8_11:$Rv, uimm1s4range:$imm, vector_ty:$Zn, vector_ty:$Zm), 0>;
2894}
2895
2896//===----------------------------------------------------------------------===//
2897// SME2 Outer Product and Accumulate
2898
2899multiclass sme2_int_mopx_tile<string mnemonic, bits<3> op, SDPatternOperator intrinsic> {
2900  def NAME : sme_int_outer_product_inst<op, 0b0, 0b1, TileOp32, ZPR16, mnemonic>, SMEPseudo2Instr<NAME, 1> {
2901    bits<2> ZAda;
2902    let Inst{1-0} = ZAda;
2903    let Inst{2}   = 0b0;
2904  }
2905
2906  def _PSEUDO : sme_outer_product_pseudo<ZPR16, SMEMatrixTileS>, SMEPseudo2Instr<NAME, 0>;
2907
2908  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, intrinsic, timm32_0_3, nxv8i1, nxv8i16>;
2909}
2910
2911multiclass  sme2_int_bmopx_tile<string mnemonic, bits<3> op, SDPatternOperator intrinsic> {
2912  def NAME : sme_outer_product_widening_inst<op, ZPR32, mnemonic>, SMEPseudo2Instr<NAME, 1>;
2913
2914  def _PSEUDO : sme_outer_product_pseudo<ZPR32, SMEMatrixTileS>, SMEPseudo2Instr<NAME, 0>;
2915
2916  def : SME_ZA_Tile_TwoPred_TwoVec_Pat<NAME, intrinsic, timm32_0_3, nxv4i1, nxv4i32>;
2917}
2918
2919//===----------------------------------------------------------------------===///
2920// SME2 Zero Lookup Table.
2921class sme2_zero_zt<string mnemonic, bits<4> opc>
2922    : I<(outs ZTR:$ZT), (ins ),
2923         mnemonic, "\t\\{ $ZT \\}",
2924         "", []>, Sched<[]> {
2925  let Inst{31-4} = 0b1100000001001000000000000000;
2926  let Inst{3-0}  = opc;
2927}
2928
2929//===----------------------------------------------------------------------===//
2930// SME2 lookup table load/store
2931class sme2_spill_fill_vector<string mnemonic, bits<8> opc>
2932    : I<!if(opc{7}, (outs ), (outs ZTR:$ZTt)),
2933        !if(opc{7}, (ins ZTR:$ZTt, GPR64sp:$Rn), (ins GPR64sp:$Rn)),
2934        mnemonic, "\t$ZTt, [$Rn]",
2935        "", []>, Sched<[]> {
2936  bits<5> Rn;
2937  let Inst{31-22} = 0b1110000100;
2938  let Inst{21-16} = opc{7-2};
2939  let Inst{15-10} = 0b100000;
2940  let Inst{9-5}   = Rn;
2941  let Inst{4-2}   = 0b000;
2942  let Inst{1-0}   = opc{1-0};
2943
2944  let mayLoad     = !not(opc{7});
2945  let mayStore    = opc{7};
2946}
2947
2948//===----------------------------------------------------------------------===///
2949// SME2 move to/from lookup table
2950class sme2_movt_zt_to_scalar<string mnemonic, bits<7> opc>
2951    : I<(outs GPR64:$Rt), (ins ZTR:$ZTt, uimm3s8:$imm3),
2952         mnemonic, "\t$Rt, $ZTt$imm3",
2953         "", []>, Sched<[]> {
2954  bits<3> imm3;
2955  bits<5> Rt;
2956  let Inst{31-15} = 0b11000000010011000;
2957  let Inst{14-12} = imm3;
2958  let Inst{11-5}  = opc;
2959  let Inst{4-0}   = Rt;
2960}
2961
2962class sme2_movt_scalar_to_zt<string mnemonic, bits<7> opc>
2963    : I<(outs ZTR:$ZTt), (ins uimm3s8:$imm3, GPR64:$Rt),
2964         mnemonic, "\t$ZTt$imm3, $Rt",
2965         "", []>, Sched<[]> {
2966  bits<3> imm3;
2967  bits<5> Rt;
2968  let Inst{31-15} = 0b11000000010011100;
2969  let Inst{14-12} = imm3;
2970  let Inst{11-5}  = opc;
2971  let Inst{4-0}   = Rt;
2972}
2973
2974//===----------------------------------------------------------------------===//
2975// SME2 lookup table expand one register
2976class sme2_luti_vector_index<bits<2> sz, bits<7> opc, RegisterOperand vector_ty,
2977                             AsmVectorIndexOpnd index_ty, string mnemonic>
2978    : I<(outs vector_ty:$Zd),
2979        (ins ZTR:$ZTt, ZPRAny:$Zn, index_ty:$i),
2980        mnemonic, "\t$Zd, $ZTt, $Zn$i",
2981        "", []>, Sched<[]> {
2982  bits<5> Zn;
2983  bits<5> Zd;
2984  let Inst{31-19} = 0b1100000011001;
2985  let Inst{18-14} = opc{6-2};
2986  let Inst{13-12} = sz;
2987  let Inst{11-10} = opc{1-0};
2988  let Inst{9-5}   = Zn;
2989  let Inst{4-0}   = Zd;
2990}
2991
2992class sme2_luti2_vector_index<bits<2> sz, RegisterOperand vector_ty,
2993                              string mnemonic>
2994    : sme2_luti_vector_index<sz, {1,?,?,?,?,0,0}, vector_ty, VectorIndexB, mnemonic> {
2995  bits<4> i;
2996  let Inst{17-14} = i;
2997}
2998
2999multiclass sme2_luti2_vector_index<string mnemonic> {
3000  def _B : sme2_luti2_vector_index<0b00, ZPR8, mnemonic>;
3001  def _H : sme2_luti2_vector_index<0b01, ZPR16, mnemonic>;
3002  def _S : sme2_luti2_vector_index<0b10, ZPR32, mnemonic>;
3003}
3004
3005class sme2_luti4_vector_index<bits<2> sz, RegisterOperand vector_ty,
3006                              string mnemonic>
3007    : sme2_luti_vector_index<sz, {0,1,?,?,?,0,0}, vector_ty, VectorIndexH, mnemonic> {
3008  bits<3> i;
3009  let Inst{16-14} = i;
3010}
3011
3012multiclass sme2_luti4_vector_index<string mnemonic> {
3013  def _B : sme2_luti4_vector_index<0b00, ZPR8, mnemonic>;
3014  def _H : sme2_luti4_vector_index<0b01, ZPR16, mnemonic>;
3015  def _S : sme2_luti4_vector_index<0b10, ZPR32, mnemonic>;
3016}
3017
3018// SME2 lookup table expand two contiguous registers
3019class sme2_luti_vector_vg2_index<bits<2> sz, bits<6> opc, RegisterOperand vector_ty,
3020                                 AsmVectorIndexOpnd index_ty, string mnemonic>
3021    : I<(outs vector_ty:$Zd),
3022        (ins ZTR:$ZTt, ZPRAny:$Zn, index_ty:$i),
3023        mnemonic, "\t$Zd, $ZTt, $Zn$i",
3024        "", []>, Sched<[]> {
3025  bits<5> Zn;
3026  bits<4> Zd;
3027  let Inst{31-19} = 0b1100000010001;
3028  let Inst{18-15} = opc{5-2};
3029  let Inst{14}    = 0b1;
3030  let Inst{13-12} = sz;
3031  let Inst{11-10} = opc{1-0};
3032  let Inst{9-5}   = Zn;
3033  let Inst{4-1}   = Zd;
3034  let Inst{0}     = 0b0;
3035}
3036
3037class sme2_luti2_vector_vg2_index<bits<2> sz, RegisterOperand vector_ty,
3038                                  string mnemonic>
3039    : sme2_luti_vector_vg2_index<sz, {1,?,?,?,0,0}, vector_ty, VectorIndexH, mnemonic> {
3040  bits<3> i;
3041  let Inst{17-15} = i;
3042}
3043
3044multiclass sme2_luti2_vector_vg2_index<string mnemonic> {
3045  def _B : sme2_luti2_vector_vg2_index<0b00, ZZ_b_mul_r, mnemonic>;
3046  def _H : sme2_luti2_vector_vg2_index<0b01, ZZ_h_mul_r, mnemonic>;
3047  def _S : sme2_luti2_vector_vg2_index<0b10, ZZ_s_mul_r, mnemonic>;
3048}
3049
3050class sme2_luti4_vector_vg2_index<bits<2> sz, RegisterOperand vector_ty,
3051                                 string mnemonic>
3052    : sme2_luti_vector_vg2_index<sz, {0,1,?,?,0,0}, vector_ty, VectorIndexS, mnemonic> {
3053  bits<2> i;
3054  let Inst{16-15} = i;
3055}
3056
3057multiclass sme2_luti4_vector_vg2_index<string mnemonic> {
3058  def _B : sme2_luti4_vector_vg2_index<0b00, ZZ_b_mul_r, mnemonic>;
3059  def _H : sme2_luti4_vector_vg2_index<0b01, ZZ_h_mul_r, mnemonic>;
3060  def _S : sme2_luti4_vector_vg2_index<0b10, ZZ_s_mul_r, mnemonic>;
3061}
3062
3063// SME2 lookup table expand four contiguous registers
3064class sme2_luti_vector_vg4_index<bits<2> sz, bits<5>opc, RegisterOperand vector_ty,
3065                                 AsmVectorIndexOpnd index_ty, string mnemonic>
3066    : I<(outs vector_ty:$Zd),
3067        (ins ZTR:$ZTt, ZPRAny:$Zn, index_ty:$i),
3068        mnemonic, "\t$Zd, $ZTt, $Zn$i",
3069        "", []>, Sched<[]> {
3070  bits<5> Zn;
3071  bits<3> Zd;
3072  let Inst{31-19} = 0b1100000010001;
3073  let Inst{18-16} = opc{4-2};
3074  let Inst{15-14} = 0b10;
3075  let Inst{13-12} = sz;
3076  let Inst{11-10} = opc{1-0};
3077  let Inst{9-5}   = Zn;
3078  let Inst{4-2}   = Zd;
3079  let Inst{1-0}   = 0b00;
3080}
3081
3082class sme2_luti2_vector_vg4_index<bits<2> sz, RegisterOperand vector_ty,
3083                                  string mnemonic>
3084    : sme2_luti_vector_vg4_index<sz, {1,?,?,0,0}, vector_ty, VectorIndexS, mnemonic> {
3085  bits<2> i;
3086  let Inst{17-16} = i;
3087}
3088
3089multiclass sme2_luti2_vector_vg4_index<string mnemonic> {
3090  def _B : sme2_luti2_vector_vg4_index<0b00, ZZZZ_b_mul_r, mnemonic>;
3091  def _H : sme2_luti2_vector_vg4_index<0b01, ZZZZ_h_mul_r, mnemonic>;
3092  def _S : sme2_luti2_vector_vg4_index<0b10, ZZZZ_s_mul_r, mnemonic>;
3093}
3094
3095class sme2_luti4_vector_vg4_index<bits<2> sz, RegisterOperand vector_ty,
3096                                  string mnemonic>
3097    : sme2_luti_vector_vg4_index<sz, {0,1,?,0,0}, vector_ty, VectorIndexD, mnemonic> {
3098  bits<1> i;
3099  let Inst{16}    = i;
3100}
3101
3102multiclass sme2_luti4_vector_vg4_index<string mnemonic> {
3103  def _H : sme2_luti4_vector_vg4_index<0b01, ZZZZ_h_mul_r, mnemonic>;
3104  def _S : sme2_luti4_vector_vg4_index<0b10, ZZZZ_s_mul_r, mnemonic>;
3105}
3106
3107//===----------------------------------------------------------------------===//
3108// SME2 MOV
3109class sme2_mova_vec_to_tile_vg2_multi_base<bits<2> sz, bit v,
3110                                           RegisterOperand tile_ty,
3111                                           Operand index_ty,
3112                                           RegisterOperand vector_ty,
3113                                           string mnemonic>
3114   : I<(outs tile_ty:$ZAd),
3115       (ins tile_ty:$_ZAd, MatrixIndexGPR32Op12_15:$Rs, index_ty:$imm, vector_ty:$Zn),
3116       mnemonic, "\t$ZAd[$Rs, $imm], $Zn",
3117       "", []>, Sched<[]> {
3118  bits<2> Rs;
3119  bits<4> Zn;
3120  let Inst{31-24} = 0b11000000;
3121  let Inst{23-22} = sz;
3122  let Inst{21-16} = 0b000100;
3123  let Inst{15}    = v;
3124  let Inst{14-13} = Rs;
3125  let Inst{12-10} = 0b000;
3126  let Inst{9-6}   = Zn;
3127  let Inst{5-3}   = 0b000;
3128
3129  let Constraints = "$ZAd = $_ZAd";
3130}
3131
3132multiclass sme2_mova_vec_to_tile_or_array_aliases<int prefer, Instruction inst,
3133                                                  RegisterOperand tile_or_array_ty,
3134                                                  RegisterOperand  rv_ty,
3135                                                  Operand index_ty,
3136                                                  RegisterOperand vector_ty,
3137                                                  string mnemonic,
3138                                                  string vg_acronym=""> {
3139  def : InstAlias<mnemonic # "\t$ZAd[$Rs, $imm" # !if(!eq(vg_acronym, ""), "", ", " # vg_acronym) # "], $Zn",
3140                  (inst tile_or_array_ty:$ZAd, rv_ty:$Rs, index_ty:$imm, vector_ty:$Zn), prefer>;
3141
3142}
3143
3144// SME2 move vector to tile, two registers
3145multiclass sme2_mova_vec_to_tile_vg2_multi_base<bit v, string mnemonic, SDPatternOperator intrinsic> {
3146
3147  def _B : sme2_mova_vec_to_tile_vg2_multi_base<0b00, v,
3148                                                !if(v, TileVectorOpV8,
3149                                                       TileVectorOpH8),
3150                                                uimm3s2range,  ZZ_b_mul_r,
3151                                                mnemonic>, SMEPseudo2Instr<NAME # _B, 1> {
3152    bits<3> imm;
3153    let Inst{2-0} = imm;
3154  }
3155
3156  def _H : sme2_mova_vec_to_tile_vg2_multi_base<0b01, v,
3157                                                !if(v, TileVectorOpV16,
3158                                                       TileVectorOpH16),
3159                                                uimm2s2range, ZZ_h_mul_r,
3160                                                mnemonic>, SMEPseudo2Instr<NAME # _H, 1> {
3161    bits<1> ZAd;
3162    bits<2> imm;
3163    let Inst{2}   = ZAd;
3164    let Inst{1-0} = imm;
3165  }
3166
3167  def _S : sme2_mova_vec_to_tile_vg2_multi_base<0b10, v,
3168                                                !if(v, TileVectorOpV32,
3169                                                       TileVectorOpH32),
3170                                                 uimm1s2range, ZZ_s_mul_r,
3171                                                 mnemonic>, SMEPseudo2Instr<NAME # _S, 1> {
3172    bits<2> ZAd;
3173    bits<1> imm;
3174    let Inst{2-1} = ZAd;
3175    let Inst{0}   = imm;
3176  }
3177
3178  def _D : sme2_mova_vec_to_tile_vg2_multi_base<0b11, v,
3179                                                !if(v, TileVectorOpV64,
3180                                                       TileVectorOpH64),
3181                                                uimm0s2range, ZZ_d_mul_r,
3182                                                mnemonic>, SMEPseudo2Instr<NAME # _D, 1> {
3183    bits<3> ZAd;
3184    let Inst{2-0} = ZAd;
3185   }
3186
3187  def NAME # _B_PSEUDO : sme2_move_to_tile_pseudo<NAME # _B, sme_elm_idx0_0, uimm3s2range, ZZ_b_mul_r, SMEMatrixTileB>;
3188  def NAME # _H_PSEUDO : sme2_move_to_tile_pseudo<NAME # _H, sme_elm_idx0_1, uimm2s2range, ZZ_h_mul_r, SMEMatrixTileH>;
3189  def NAME # _S_PSEUDO : sme2_move_to_tile_pseudo<NAME # _S, sme_elm_idx0_3, uimm1s2range, ZZ_s_mul_r, SMEMatrixTileS>;
3190  def NAME # _D_PSEUDO : sme2_move_to_tile_pseudo<NAME # _D, sme_elm_idx0_7, uimm0s2range, ZZ_d_mul_r, SMEMatrixTileD>;
3191
3192  def : SME2_Tile_VG2_Multi_Pat<NAME # _B, intrinsic, sme_elm_idx0_0, nxv16i8, uimm3s2range, tileslicerange3s2>;
3193  def : SME2_Tile_VG2_Multi_Pat<NAME # _H, intrinsic, sme_elm_idx0_1, nxv8i16, uimm2s2range, tileslicerange2s2>;
3194  def : SME2_Tile_VG2_Multi_Pat<NAME # _H, intrinsic, sme_elm_idx0_1, nxv8f16, uimm2s2range, tileslicerange2s2>;
3195  def : SME2_Tile_VG2_Multi_Pat<NAME # _H, intrinsic, sme_elm_idx0_1, nxv8bf16, uimm2s2range, tileslicerange2s2>;
3196  def : SME2_Tile_VG2_Multi_Pat<NAME # _S, intrinsic, sme_elm_idx0_3, nxv4i32, uimm1s2range, tileslicerange1s2>;
3197  def : SME2_Tile_VG2_Multi_Pat<NAME # _S, intrinsic, sme_elm_idx0_3, nxv4f32, uimm1s2range, tileslicerange1s2>;
3198  def : SME2_Tile_VG2_Multi_Pat<NAME # _D, intrinsic, sme_elm_idx0_7, nxv2i64, uimm0s2range, tileslicerange0s2>;
3199  def : SME2_Tile_VG2_Multi_Pat<NAME # _D, intrinsic, sme_elm_idx0_7, nxv2f64, uimm0s2range, tileslicerange0s2>;
3200
3201  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _B),
3202                                                !if(v, TileVectorOpV8,
3203                                                       TileVectorOpH8),
3204                                                MatrixIndexGPR32Op12_15,
3205                                                uimm3s2range,  ZZ_b_mul_r,
3206                                                "mov">;
3207  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _H),
3208                                                !if(v, TileVectorOpV16,
3209                                                       TileVectorOpH16),
3210                                                MatrixIndexGPR32Op12_15,
3211                                                uimm2s2range,  ZZ_h_mul_r,
3212                                                "mov">;
3213  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _S),
3214                                                !if(v, TileVectorOpV32,
3215                                                       TileVectorOpH32),
3216                                                MatrixIndexGPR32Op12_15,
3217                                                uimm1s2range,  ZZ_s_mul_r,
3218                                                "mov">;
3219  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _D),
3220                                                !if(v, TileVectorOpV64,
3221                                                       TileVectorOpH64),
3222                                                MatrixIndexGPR32Op12_15,
3223                                                uimm0s2range,  ZZ_d_mul_r,
3224                                                "mov">;
3225
3226  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _B),
3227                                                !if(v, TileVectorOpV8,
3228                                                       TileVectorOpH8),
3229                                                MatrixIndexGPR32Op12_15,
3230                                                uimm3s2range,  ZZ_b_mul_r,
3231                                                "mova">;
3232  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _H),
3233                                                !if(v, TileVectorOpV16,
3234                                                       TileVectorOpH16),
3235                                                MatrixIndexGPR32Op12_15,
3236                                                uimm2s2range,  ZZ_h_mul_r,
3237                                                "mova">;
3238  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _S),
3239                                                !if(v, TileVectorOpV32,
3240                                                       TileVectorOpH32),
3241                                                MatrixIndexGPR32Op12_15,
3242                                                uimm1s2range,  ZZ_s_mul_r,
3243                                                "mova">;
3244  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _D),
3245                                                !if(v, TileVectorOpV64,
3246                                                       TileVectorOpH64),
3247                                                MatrixIndexGPR32Op12_15,
3248                                                uimm0s2range,  ZZ_d_mul_r,
3249                                                "mova">;
3250
3251  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _B),
3252                                                !if(v, TileVectorOpV8,
3253                                                       TileVectorOpH8),
3254                                                MatrixIndexGPR32Op12_15,
3255                                                uimm3s2range,  ZZ_b_mul_r,
3256                                                "mova">;
3257  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _H),
3258                                                !if(v, TileVectorOpV16,
3259                                                       TileVectorOpH16),
3260                                                MatrixIndexGPR32Op12_15,
3261                                                uimm2s2range,  ZZ_h_mul_r,
3262                                                "mova">;
3263  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _S),
3264                                                !if(v, TileVectorOpV32,
3265                                                       TileVectorOpH32),
3266                                                MatrixIndexGPR32Op12_15,
3267                                                uimm1s2range,  ZZ_s_mul_r,
3268                                                "mova">;
3269  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _D),
3270                                                !if(v, TileVectorOpV64,
3271                                                       TileVectorOpH64),
3272                                                MatrixIndexGPR32Op12_15,
3273                                                uimm0s2range,  ZZ_d_mul_r,
3274                                                "mova">;
3275}
3276
3277multiclass sme2_mova_vec_to_tile_vg2_multi<string mnemonic,
3278                                           SDPatternOperator int_h, SDPatternOperator int_v>{
3279 defm _H : sme2_mova_vec_to_tile_vg2_multi_base<0b0, mnemonic, int_h>;
3280 defm _V : sme2_mova_vec_to_tile_vg2_multi_base<0b1, mnemonic, int_v>;
3281}
3282
3283class sme2_mova_vec_to_tile_vg4_multi_base<bits<2> sz, bit v, bits<3> op,
3284                                           RegisterOperand tile_ty,
3285                                           Operand index_ty,
3286                                           RegisterOperand vector_ty,
3287                                           string mnemonic>
3288   : I<(outs tile_ty:$ZAd),
3289       (ins tile_ty:$_ZAd, MatrixIndexGPR32Op12_15:$Rs, index_ty:$imm,
3290            vector_ty:$Zn),
3291       mnemonic,
3292       "\t$ZAd[$Rs, $imm], $Zn",
3293       "", []>, Sched<[]> {
3294  bits<2> Rs;
3295  bits<3> Zn;
3296  let Inst{31-24} = 0b11000000;
3297  let Inst{23-22} = sz;
3298  let Inst{21-16} = 0b000100;
3299  let Inst{15}    = v;
3300  let Inst{14-13} = Rs;
3301  let Inst{12-10} = 0b001;
3302  let Inst{9-7}   = Zn;
3303  let Inst{6-3}   = 0b0000;
3304  let Inst{2-0}   = op;
3305  let Constraints = "$ZAd = $_ZAd";
3306}
3307
3308// SME2 move vector to tile, four registers
3309multiclass sme2_mova_vec_to_tile_vg4_multi_base<bit v, string mnemonic, SDPatternOperator intrinsic> {
3310
3311  def _B : sme2_mova_vec_to_tile_vg4_multi_base<0b00, v, {0,?,?},
3312                                                !if(v, TileVectorOpV8,
3313                                                       TileVectorOpH8),
3314                                                uimm2s4range, ZZZZ_b_mul_r,
3315                                                mnemonic>, SMEPseudo2Instr<NAME # _B, 1> {
3316    bits<2> imm;
3317    let Inst{1-0} = imm;
3318  }
3319
3320  def _H : sme2_mova_vec_to_tile_vg4_multi_base<0b01, v, {0,?,?},
3321                                                !if(v, TileVectorOpV16,
3322                                                       TileVectorOpH16),
3323                                                uimm1s4range, ZZZZ_h_mul_r,
3324                                                mnemonic>, SMEPseudo2Instr<NAME # _H, 1> {
3325    bits<1> ZAd;
3326    bits<1> imm;
3327    let Inst{1}   = ZAd;
3328    let Inst{0}   = imm;
3329  }
3330
3331  def _S : sme2_mova_vec_to_tile_vg4_multi_base<0b10, v, {0,?,?},
3332                                                !if(v, TileVectorOpV32,
3333                                                       TileVectorOpH32),
3334                                                 uimm0s4range, ZZZZ_s_mul_r,
3335                                                 mnemonic>, SMEPseudo2Instr<NAME # _S, 1> {
3336    bits<2> ZAd;
3337    let Inst{1-0} = ZAd;
3338  }
3339
3340  def _D : sme2_mova_vec_to_tile_vg4_multi_base<0b11, v, {?,?,?},
3341                                                !if(v, TileVectorOpV64,
3342                                                       TileVectorOpH64),
3343                                                uimm0s4range, ZZZZ_d_mul_r,
3344                                                mnemonic>, SMEPseudo2Instr<NAME # _D, 1> {
3345    bits<3> ZAd;
3346    let Inst{2-0} = ZAd;
3347  }
3348
3349  def NAME # _B_PSEUDO : sme2_move_to_tile_pseudo<NAME # _B, sme_elm_idx0_0, uimm2s4range, ZZZZ_b_mul_r, SMEMatrixTileB>;
3350  def NAME # _H_PSEUDO : sme2_move_to_tile_pseudo<NAME # _H, sme_elm_idx0_1, uimm1s4range, ZZZZ_h_mul_r, SMEMatrixTileH>;
3351  def NAME # _S_PSEUDO : sme2_move_to_tile_pseudo<NAME # _S, sme_elm_idx0_3, uimm0s4range, ZZZZ_s_mul_r, SMEMatrixTileS>;
3352  def NAME # _D_PSEUDO : sme2_move_to_tile_pseudo<NAME # _D, sme_elm_idx0_7, uimm0s4range, ZZZZ_d_mul_r, SMEMatrixTileD>;
3353
3354  def : SME2_Tile_VG4_Multi_Pat<NAME # _B, intrinsic, sme_elm_idx0_0, nxv16i8, uimm2s4range, tileslicerange2s4>;
3355  def : SME2_Tile_VG4_Multi_Pat<NAME # _H, intrinsic, sme_elm_idx0_1, nxv8i16, uimm1s4range, tileslicerange1s4>;
3356  def : SME2_Tile_VG4_Multi_Pat<NAME # _H, intrinsic, sme_elm_idx0_1, nxv8f16, uimm1s4range, tileslicerange1s4>;
3357  def : SME2_Tile_VG4_Multi_Pat<NAME # _H, intrinsic, sme_elm_idx0_1, nxv8bf16, uimm1s4range, tileslicerange1s4>;
3358  def : SME2_Tile_VG4_Multi_Pat<NAME # _S, intrinsic, sme_elm_idx0_3, nxv4i32, uimm0s4range, tileslicerange0s4>;
3359  def : SME2_Tile_VG4_Multi_Pat<NAME # _S, intrinsic, sme_elm_idx0_3, nxv4f32, uimm0s4range, tileslicerange0s4>;
3360  def : SME2_Tile_VG4_Multi_Pat<NAME # _D, intrinsic, sme_elm_idx0_7, nxv2i64, uimm0s4range, tileslicerange0s4>;
3361  def : SME2_Tile_VG4_Multi_Pat<NAME # _D, intrinsic, sme_elm_idx0_7, nxv2f64, uimm0s4range, tileslicerange0s4>;
3362
3363  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _B),
3364                                                !if(v, TileVectorOpV8,
3365                                                       TileVectorOpH8),
3366                                                MatrixIndexGPR32Op12_15,
3367                                                uimm2s4range, ZZZZ_b_mul_r,
3368                                                "mov">;
3369  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _H),
3370                                                !if(v, TileVectorOpV16,
3371                                                       TileVectorOpH16),
3372                                                MatrixIndexGPR32Op12_15,
3373                                                uimm1s4range, ZZZZ_h_mul_r,
3374                                                "mov">;
3375  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _S),
3376                                                !if(v, TileVectorOpV32,
3377                                                       TileVectorOpH32),
3378                                                MatrixIndexGPR32Op12_15,
3379                                                uimm0s4range, ZZZZ_s_mul_r,
3380                                                "mov">;
3381  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME # _D),
3382                                                !if(v, TileVectorOpV64,
3383                                                       TileVectorOpH64),
3384                                                MatrixIndexGPR32Op12_15,
3385                                                uimm0s4range, ZZZZ_d_mul_r,
3386                                                "mov">;
3387
3388  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _B),
3389                                                !if(v, TileVectorOpV8,
3390                                                       TileVectorOpH8),
3391                                                MatrixIndexGPR32Op12_15,
3392                                                uimm2s4range, ZZZZ_b_mul_r,
3393                                                "mova">;
3394  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _H),
3395                                                !if(v, TileVectorOpV16,
3396                                                       TileVectorOpH16),
3397                                                MatrixIndexGPR32Op12_15,
3398                                                uimm1s4range, ZZZZ_h_mul_r,
3399                                                "mova">;
3400  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _S),
3401                                                !if(v, TileVectorOpV32,
3402                                                       TileVectorOpH32),
3403                                                MatrixIndexGPR32Op12_15,
3404                                                uimm0s4range, ZZZZ_s_mul_r,
3405                                                "mova">;
3406  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME # _D),
3407                                                !if(v, TileVectorOpV64,
3408                                                       TileVectorOpH64),
3409                                                MatrixIndexGPR32Op12_15,
3410                                                uimm0s4range, ZZZZ_d_mul_r,
3411                                                "mova">;
3412
3413}
3414
3415multiclass sme2_mova_vec_to_tile_vg4_multi<string mnemonic,
3416                                           SDPatternOperator int_h, SDPatternOperator int_v>{
3417 defm _H : sme2_mova_vec_to_tile_vg4_multi_base<0b0, mnemonic, int_h>;
3418 defm _V : sme2_mova_vec_to_tile_vg4_multi_base<0b1, mnemonic, int_v>;
3419}
3420
3421// SME Move into Array
3422class sme2_mova_vec_to_array_vg24_multi< bits<5> op, RegisterOperand array_ty,
3423                                        RegisterOperand vector_ty,
3424                                        string mnemonic,
3425                                        string vg_acronym="">
3426   : I<(outs array_ty:$ZAd),
3427       (ins array_ty:$_ZAd, MatrixIndexGPR32Op8_11:$Rs, sme_elm_idx0_7:$imm,
3428            vector_ty:$Zn),
3429       mnemonic, "\t$ZAd[$Rs, $imm, " # vg_acronym # "], $Zn",
3430       "", []>, Sched<[]> {
3431  bits<2> Rs;
3432  bits<3> imm;
3433  let Inst{31-15} = 0b11000000000001000;
3434  let Inst{14-13} = Rs;
3435  let Inst{12-11} = 0b01;
3436  let Inst{10-6}  = op;
3437  let Inst{5-3}   = 0b000;
3438  let Inst{2-0}   = imm;
3439
3440  let Constraints = "$ZAd = $_ZAd";
3441}
3442
3443// MOVA (vector to array, two registers)
3444multiclass sme2_mova_vec_to_array_vg2_multi<string mnemonic, SDPatternOperator intrinsic> {
3445  def NAME : sme2_mova_vec_to_array_vg24_multi<{0,?,?,?,?}, MatrixOp64,
3446                                               ZZ_d_mul_r, mnemonic, "vgx2">, SMEPseudo2Instr<NAME, 1> {
3447   bits<4> Zn;
3448   let Inst{9-6} = Zn;
3449  }
3450
3451  def NAME # _PSEUDO : sme2_move_to_za_pseudo<NAME, sme_elm_idx0_7, ZZ_d_mul_r, SMEMatrixArray>;
3452
3453  def : SME2_ZA_VG1x2_Multi_Pat<NAME, intrinsic, nxv2i64, sme_elm_idx0_7, tileslice16>;
3454  def : SME2_ZA_VG1x2_Multi_Pat<NAME, intrinsic, nxv2f64, sme_elm_idx0_7, tileslice16>;
3455
3456  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3457                                                MatrixOp8,
3458                                                MatrixIndexGPR32Op8_11,
3459                                                sme_elm_idx0_7, ZZ_b_mul_r,
3460                                                "mova">;
3461  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3462                                                MatrixOp16,
3463                                                MatrixIndexGPR32Op8_11,
3464                                                sme_elm_idx0_7, ZZ_h_mul_r,
3465                                                "mova">;
3466  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3467                                                MatrixOp32,
3468                                                MatrixIndexGPR32Op8_11,
3469                                                sme_elm_idx0_7, ZZ_s_mul_r,
3470                                                "mova">;
3471  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3472                                                MatrixOp64,
3473                                                MatrixIndexGPR32Op8_11,
3474                                                sme_elm_idx0_7, ZZ_d_mul_r,
3475                                                "mova">;
3476
3477  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3478                                                MatrixOp8,
3479                                                MatrixIndexGPR32Op8_11,
3480                                                sme_elm_idx0_7, ZZ_b_mul_r,
3481                                                "mova", "vgx2">;
3482  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3483                                                MatrixOp16,
3484                                                MatrixIndexGPR32Op8_11,
3485                                                sme_elm_idx0_7, ZZ_h_mul_r,
3486                                                "mova", "vgx2">;
3487  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3488                                                MatrixOp32,
3489                                                MatrixIndexGPR32Op8_11,
3490                                                sme_elm_idx0_7, ZZ_s_mul_r,
3491                                                "mova", "vgx2">;
3492
3493  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3494                                                MatrixOp8,
3495                                                MatrixIndexGPR32Op8_11,
3496                                                sme_elm_idx0_7, ZZ_b_mul_r,
3497                                                "mov">;
3498  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3499                                                MatrixOp16,
3500                                                MatrixIndexGPR32Op8_11,
3501                                                sme_elm_idx0_7, ZZ_h_mul_r,
3502                                                "mov">;
3503  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3504                                                MatrixOp32,
3505                                                MatrixIndexGPR32Op8_11,
3506                                                sme_elm_idx0_7, ZZ_s_mul_r,
3507                                                "mov">;
3508  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3509                                                MatrixOp64,
3510                                                MatrixIndexGPR32Op8_11,
3511                                                sme_elm_idx0_7, ZZ_d_mul_r,
3512                                                "mov">;
3513
3514  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3515                                                MatrixOp8,
3516                                                MatrixIndexGPR32Op8_11,
3517                                                sme_elm_idx0_7, ZZ_b_mul_r,
3518                                                "mov", "vgx2">;
3519  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3520                                                MatrixOp16,
3521                                                MatrixIndexGPR32Op8_11,
3522                                                sme_elm_idx0_7, ZZ_h_mul_r,
3523                                                "mov", "vgx2">;
3524  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3525                                                MatrixOp32,
3526                                                MatrixIndexGPR32Op8_11,
3527                                                sme_elm_idx0_7, ZZ_s_mul_r,
3528                                                "mov", "vgx2">;
3529  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME),
3530                                                MatrixOp64,
3531                                                MatrixIndexGPR32Op8_11,
3532                                                sme_elm_idx0_7, ZZ_d_mul_r,
3533                                                "mov", "vgx2">;
3534}
3535
3536// MOVA (vector to array, four registers)
3537multiclass sme2_mova_vec_to_array_vg4_multi<string mnemonic, SDPatternOperator intrinsic> {
3538  def NAME : sme2_mova_vec_to_array_vg24_multi<{1,?,?,?,0}, MatrixOp64,
3539                                               ZZZZ_d_mul_r, mnemonic, "vgx4">, SMEPseudo2Instr<NAME, 1> {
3540    bits<3> Zn;
3541    let Inst{9-7} = Zn;
3542  }
3543
3544  def NAME # _PSEUDO : sme2_move_to_za_pseudo<NAME, sme_elm_idx0_7, ZZZZ_d_mul_r, SMEMatrixArray>;
3545
3546  def : SME2_ZA_VG1x4_Multi_Pat<NAME, intrinsic, nxv2i64, sme_elm_idx0_7, tileslice16>;
3547  def : SME2_ZA_VG1x4_Multi_Pat<NAME, intrinsic, nxv2f64, sme_elm_idx0_7, tileslice16>;
3548
3549  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3550                                                MatrixOp8,
3551                                                MatrixIndexGPR32Op8_11,
3552                                                sme_elm_idx0_7, ZZZZ_b_mul_r,
3553                                                "mova">;
3554  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3555                                                MatrixOp16,
3556                                                MatrixIndexGPR32Op8_11,
3557                                                sme_elm_idx0_7, ZZZZ_h_mul_r,
3558                                                "mova">;
3559  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3560                                                MatrixOp32,
3561                                                MatrixIndexGPR32Op8_11,
3562                                                sme_elm_idx0_7, ZZZZ_s_mul_r,
3563                                                "mova">;
3564  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3565                                                MatrixOp64,
3566                                                MatrixIndexGPR32Op8_11,
3567                                                sme_elm_idx0_7, ZZZZ_d_mul_r,
3568                                                "mova">;
3569
3570  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3571                                                MatrixOp8,
3572                                                MatrixIndexGPR32Op8_11,
3573                                                sme_elm_idx0_7, ZZZZ_b_mul_r,
3574                                                "mova", "vgx4">;
3575  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3576                                                MatrixOp16,
3577                                                MatrixIndexGPR32Op8_11,
3578                                                sme_elm_idx0_7, ZZZZ_h_mul_r,
3579                                                "mova", "vgx4">;
3580  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3581                                                MatrixOp32,
3582                                                MatrixIndexGPR32Op8_11,
3583                                                sme_elm_idx0_7, ZZZZ_s_mul_r,
3584                                                "mova", "vgx4">;
3585
3586  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3587                                                MatrixOp8,
3588                                                MatrixIndexGPR32Op8_11,
3589                                                sme_elm_idx0_7, ZZZZ_b_mul_r,
3590                                                "mov">;
3591  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3592                                                MatrixOp16,
3593                                                MatrixIndexGPR32Op8_11,
3594                                                sme_elm_idx0_7, ZZZZ_h_mul_r,
3595                                                "mov">;
3596  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3597                                                MatrixOp32,
3598                                                MatrixIndexGPR32Op8_11,
3599                                                sme_elm_idx0_7, ZZZZ_s_mul_r,
3600                                                "mov">;
3601  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3602                                                MatrixOp64,
3603                                                MatrixIndexGPR32Op8_11,
3604                                                sme_elm_idx0_7, ZZZZ_d_mul_r,
3605                                                "mov">;
3606
3607  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3608                                                MatrixOp8,
3609                                                MatrixIndexGPR32Op8_11,
3610                                                sme_elm_idx0_7, ZZZZ_b_mul_r,
3611                                                "mov", "vgx4">;
3612  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3613                                                MatrixOp16,
3614                                                MatrixIndexGPR32Op8_11,
3615                                                sme_elm_idx0_7, ZZZZ_h_mul_r,
3616                                                "mov", "vgx4">;
3617  defm : sme2_mova_vec_to_tile_or_array_aliases<0, !cast<Instruction>(NAME),
3618                                                MatrixOp32,
3619                                                MatrixIndexGPR32Op8_11,
3620                                                sme_elm_idx0_7, ZZZZ_s_mul_r,
3621                                                "mov", "vgx4">;
3622  defm : sme2_mova_vec_to_tile_or_array_aliases<1, !cast<Instruction>(NAME),
3623                                                MatrixOp64,
3624                                                MatrixIndexGPR32Op8_11,
3625                                                sme_elm_idx0_7, ZZZZ_d_mul_r,
3626                                                "mov", "vgx4">;
3627
3628}
3629
3630class sme2_mova_tile_to_vec_vg2_multi_base<bits<2> sz, bit v, bits<3> op,
3631                                           RegisterOperand vector_ty,
3632                                           RegisterOperand tile_ty,
3633                                           Operand index_ty,
3634                                           string mnemonic>
3635   : I<!if(op{1}, (outs vector_ty:$Zd, tile_ty:$_ZAn), (outs vector_ty:$Zd)),
3636       (ins tile_ty:$ZAn, MatrixIndexGPR32Op12_15:$Rs, index_ty:$imm),
3637       mnemonic,
3638       "\t$Zd, $ZAn[$Rs, $imm]",
3639       "", []>, Sched<[]> {
3640  bits<4> Zd;
3641  bits<2> Rs;
3642  let Inst{31-24} = 0b11000000;
3643  let Inst{23-22} = sz;
3644  let Inst{21-16} = 0b000110;
3645  let Inst{15}    = v;
3646  let Inst{14-13} = Rs;
3647  let Inst{12-11} = 0b00;
3648  let Inst{10-8}  = op;
3649  let Inst{4-1}   = Zd;
3650  let Inst{0}     = 0b0;
3651
3652  let Constraints = !if(op{1}, "$ZAn = $_ZAn", "");
3653}
3654
3655multiclass sme2_mova_tile_or_array_to_vec_aliases<int op, Instruction inst,
3656                                                  RegisterOperand vector_ty,
3657                                                  RegisterOperand tile_or_array_ty,
3658                                                  RegisterOperand rv_ty,
3659                                                  Operand index_ty,
3660                                                  string mnemonic,
3661                                                  string vg_acronym=""> {
3662def : InstAlias<mnemonic # "\t$Zd, $ZAn[$Rs, $imm" # !if(!eq(vg_acronym, ""), "", ", " # vg_acronym) # "]",
3663                  (inst vector_ty:$Zd, tile_or_array_ty:$ZAn, rv_ty:$Rs, index_ty:$imm), op>;
3664
3665}
3666
3667multiclass sme2_mova_tile_to_vec_vg2_multi_inst<bit v, bits<3> opc, string mnemonic> {
3668
3669  def _B : sme2_mova_tile_to_vec_vg2_multi_base<0b00, v, opc, ZZ_b_mul_r,
3670                                                !if(v, TileVectorOpV8,
3671                                                       TileVectorOpH8),
3672                                                 uimm3s2range, mnemonic> {
3673    bits<3> imm;
3674    let Inst{7-5} = imm;
3675  }
3676
3677  def _H : sme2_mova_tile_to_vec_vg2_multi_base<0b01, v, opc, ZZ_h_mul_r,
3678                                                !if(v, TileVectorOpV16,
3679                                                       TileVectorOpH16),
3680                                                 uimm2s2range, mnemonic> {
3681    bits<1> ZAn;
3682    bits<2> imm;
3683    let Inst{7}   = ZAn;
3684    let Inst{6-5} = imm;
3685  }
3686
3687  def _S : sme2_mova_tile_to_vec_vg2_multi_base<0b10, v, opc, ZZ_s_mul_r,
3688                                                !if(v, TileVectorOpV32,
3689                                                       TileVectorOpH32),
3690                                                 uimm1s2range, mnemonic> {
3691    bits<2> ZAn;
3692    bits<1> imm;
3693    let Inst{7-6} = ZAn;
3694    let Inst{5}   = imm;
3695  }
3696
3697  def _D : sme2_mova_tile_to_vec_vg2_multi_base<0b11, v, opc, ZZ_d_mul_r,
3698                                                !if(v, TileVectorOpV64,
3699                                                       TileVectorOpH64),
3700                                                uimm0s2range, mnemonic> {
3701    bits<3> ZAn;
3702    let Inst{7-5} = ZAn;
3703  }
3704
3705  if !eq(mnemonic, "mova") then {
3706  defm : sme2_mova_tile_or_array_to_vec_aliases<1,!cast<Instruction>(NAME # _B),
3707                                                ZZ_b_mul_r,
3708                                               !if(v, TileVectorOpV8,
3709                                                      TileVectorOpH8),
3710                                                MatrixIndexGPR32Op12_15,
3711                                                uimm3s2range, "mov">;
3712  defm : sme2_mova_tile_or_array_to_vec_aliases<1,!cast<Instruction>(NAME # _H),
3713                                                ZZ_h_mul_r,
3714                                                !if(v, TileVectorOpV16,
3715                                                       TileVectorOpH16),
3716                                                MatrixIndexGPR32Op12_15,
3717                                                uimm2s2range, "mov">;
3718  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME # _S),
3719                                                ZZ_s_mul_r,
3720                                                !if(v, TileVectorOpV32,
3721                                                       TileVectorOpH32),
3722                                                MatrixIndexGPR32Op12_15,
3723                                                uimm1s2range, "mov">;
3724  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME # _D),
3725                                                ZZ_d_mul_r,
3726                                                !if(v, TileVectorOpV64,
3727                                                       TileVectorOpH64),
3728                                                MatrixIndexGPR32Op12_15,
3729                                                uimm0s2range, "mov">;
3730  }
3731
3732  defm : sme2_mova_tile_or_array_to_vec_aliases<0,!cast<Instruction>(NAME # _B),
3733                                                ZZ_b_mul_r,
3734                                               !if(v, TileVectorOpV8,
3735                                                      TileVectorOpH8),
3736                                                MatrixIndexGPR32Op12_15,
3737                                                uimm3s2range, mnemonic>;
3738  defm : sme2_mova_tile_or_array_to_vec_aliases<0,!cast<Instruction>(NAME # _H),
3739                                                ZZ_h_mul_r,
3740                                                !if(v, TileVectorOpV16,
3741                                                       TileVectorOpH16),
3742                                                MatrixIndexGPR32Op12_15,
3743                                                uimm2s2range, mnemonic>;
3744  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME # _S),
3745                                                ZZ_s_mul_r,
3746                                                !if(v, TileVectorOpV32,
3747                                                       TileVectorOpH32),
3748                                                MatrixIndexGPR32Op12_15,
3749                                                uimm1s2range, mnemonic>;
3750  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME # _D),
3751                                                ZZ_d_mul_r,
3752                                                !if(v, TileVectorOpV64,
3753                                                       TileVectorOpH64),
3754                                                MatrixIndexGPR32Op12_15,
3755                                                uimm0s2range, mnemonic>;
3756
3757}
3758
3759// SME2 move tile to vector, two registers
3760multiclass sme2_mova_tile_to_vec_vg2_multi<string mnemonic>{
3761 defm _H : sme2_mova_tile_to_vec_vg2_multi_inst<0b0, 0b000, mnemonic>;
3762 defm _V : sme2_mova_tile_to_vec_vg2_multi_inst<0b1, 0b000, mnemonic>;
3763}
3764
3765// SME2p1 move tile to vector and zero tile, two registers
3766multiclass sme2p1_movaz_tile_to_vec_vg2<string mnemonic>{
3767 defm _H : sme2_mova_tile_to_vec_vg2_multi_inst<0b0, 0b010, mnemonic>;
3768 defm _V : sme2_mova_tile_to_vec_vg2_multi_inst<0b1, 0b010, mnemonic>;
3769}
3770
3771class sme2_mova_tile_to_vec_vg4_multi_base<bits<2> sz, bit v, bits<6> op,
3772                                           RegisterOperand vector_ty,
3773                                           RegisterOperand tile_ty,
3774                                           Operand index_ty,
3775                                           string mnemonic>
3776   : I<!if(op{4}, (outs vector_ty:$Zd, tile_ty:$_ZAn), (outs vector_ty:$Zd)),
3777       (ins tile_ty:$ZAn, MatrixIndexGPR32Op12_15:$Rs, index_ty:$imm),
3778       mnemonic,
3779       "\t$Zd, $ZAn[$Rs, $imm]",
3780       "", []>, Sched<[]> {
3781  bits<3> Zd;
3782  bits<2> Rs;
3783  let Inst{31-24} = 0b11000000;
3784  let Inst{23-22} = sz;
3785  let Inst{21-16} = 0b000110;
3786  let Inst{15}    = v;
3787  let Inst{14-13} = Rs;
3788  let Inst{12-11} = 0b00;
3789  let Inst{10-5}  = op{5-0};
3790  let Inst{4-2}   = Zd;
3791  let Inst{1-0}   = 0b00;
3792
3793  let Constraints = !if(op{4}, "$ZAn = $_ZAn", "");
3794}
3795
3796multiclass sme2_mova_tile_to_vec_vg4_multi_base<bit v, bits<3> opc, string mnemonic> {
3797
3798  def _B : sme2_mova_tile_to_vec_vg4_multi_base<0b00, v, {opc,0,?,?},
3799                                                ZZZZ_b_mul_r,
3800                                                !if(v, TileVectorOpV8,
3801                                                       TileVectorOpH8),
3802                                                uimm2s4range, mnemonic> {
3803    bits<2> imm;
3804    let Inst{6-5} = imm;
3805  }
3806
3807  def _H : sme2_mova_tile_to_vec_vg4_multi_base<0b01, v, {opc,0,?,?},
3808                                                ZZZZ_h_mul_r,
3809                                                !if(v, TileVectorOpV16,
3810                                                       TileVectorOpH16),
3811                                                uimm1s4range, mnemonic> {
3812    bits<1> ZAn;
3813    bits<1> imm;
3814    let Inst{6}   = ZAn;
3815    let Inst{5}   = imm;
3816  }
3817
3818  def _S : sme2_mova_tile_to_vec_vg4_multi_base<0b10, v, {opc,0,?,?},
3819                                                ZZZZ_s_mul_r,
3820                                                !if(v, TileVectorOpV32,
3821                                                       TileVectorOpH32),
3822                                                 uimm0s4range, mnemonic> {
3823    bits<2> ZAn;
3824    let Inst{6-5} = ZAn;
3825  }
3826
3827  def _D : sme2_mova_tile_to_vec_vg4_multi_base<0b11, v, {opc,?,?,?},
3828                                                ZZZZ_d_mul_r,
3829                                                !if(v, TileVectorOpV64,
3830                                                       TileVectorOpH64),
3831                                                uimm0s4range, mnemonic> {
3832    bits<3> ZAn;
3833    let Inst{7-5} = ZAn;
3834  }
3835
3836  if !eq(mnemonic, "mova") then {
3837  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME # _B),
3838                                                ZZZZ_b_mul_r,
3839                                                !if(v, TileVectorOpV8,
3840                                                      TileVectorOpH8),
3841                                                MatrixIndexGPR32Op12_15,
3842                                                uimm2s4range, "mov">;
3843  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME # _H),
3844                                                ZZZZ_h_mul_r,
3845                                                !if(v, TileVectorOpV16,
3846                                                       TileVectorOpH16),
3847                                                MatrixIndexGPR32Op12_15,
3848                                                uimm1s4range, "mov">;
3849  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME # _S),
3850                                                ZZZZ_s_mul_r,
3851                                                !if(v, TileVectorOpV32,
3852                                                      TileVectorOpH32),
3853                                                MatrixIndexGPR32Op12_15,
3854                                                uimm0s4range, "mov">;
3855  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME # _D),
3856                                                ZZZZ_d_mul_r,
3857                                                !if(v, TileVectorOpV64,
3858                                                       TileVectorOpH64),
3859                                                MatrixIndexGPR32Op12_15,
3860                                                uimm0s4range, "mov">;
3861  }
3862
3863  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME # _B),
3864                                                ZZZZ_b_mul_r,
3865                                                !if(v, TileVectorOpV8,
3866                                                       TileVectorOpH8),
3867                                                MatrixIndexGPR32Op12_15,
3868                                                uimm2s4range, mnemonic>;
3869  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME # _H),
3870                                                ZZZZ_h_mul_r,
3871                                                !if(v, TileVectorOpV16,
3872                                                       TileVectorOpH16),
3873                                                MatrixIndexGPR32Op12_15,
3874                                                uimm1s4range, mnemonic>;
3875  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME # _S),
3876                                                ZZZZ_s_mul_r,
3877                                                !if(v, TileVectorOpV32,
3878                                                      TileVectorOpH32),
3879                                                MatrixIndexGPR32Op12_15,
3880                                                uimm0s4range, mnemonic>;
3881  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME # _D),
3882                                                ZZZZ_d_mul_r,
3883                                                !if(v, TileVectorOpV64,
3884                                                       TileVectorOpH64),
3885                                                MatrixIndexGPR32Op12_15,
3886                                                uimm0s4range, mnemonic>;
3887
3888}
3889
3890// SME2 move tile to vector, four registers
3891multiclass sme2_mova_tile_to_vec_vg4_multi<string mnemonic>{
3892 defm _H : sme2_mova_tile_to_vec_vg4_multi_base<0b0, 0b100, mnemonic>;
3893 defm _V : sme2_mova_tile_to_vec_vg4_multi_base<0b1, 0b100, mnemonic>;
3894}
3895
3896// SME2p1 move tile to vector and zero tile, four registers
3897multiclass sme2p1_movaz_tile_to_vec_vg4<string mnemonic>{
3898 defm _H : sme2_mova_tile_to_vec_vg4_multi_base<0b0, 0b110, mnemonic>;
3899 defm _V : sme2_mova_tile_to_vec_vg4_multi_base<0b1, 0b110, mnemonic>;
3900}
3901
3902
3903class sme2_mova_array_to_vec_vg24_multi<bits<4>op, RegisterOperand vector_ty,
3904                                        RegisterOperand array_ty,
3905                                        string mnemonic, string vg_acronym>
3906   : I<!if(op{2}, (outs vector_ty:$Zd, array_ty:$_ZAn), (outs vector_ty:$Zd)),
3907       (ins array_ty:$ZAn, MatrixIndexGPR32Op8_11:$Rs, sme_elm_idx0_7:$imm),
3908       mnemonic,
3909       "\t$Zd, $ZAn[$Rs, $imm, " # vg_acronym # "]",
3910       "", []>, Sched<[]> {
3911  bits<2> Rs;
3912  bits<3> imm;
3913  let Inst{31-15} = 0b11000000000001100;
3914  let Inst{14-13} = Rs;
3915  let Inst{12-11} = 0b01;
3916  let Inst{10-8}  = op{3-1};
3917  let Inst{7-5}   = imm;
3918  let Inst{1}     = op{0};
3919  let Inst{0}     = 0b0;
3920  let Constraints = !if(op{2}, "$ZAn = $_ZAn", "");
3921}
3922
3923// move array to vector, two registers.
3924multiclass sme2_mova_array_to_vec_vg2_multi<bits<3> opc, string mnemonic> {
3925  def NAME : sme2_mova_array_to_vec_vg24_multi<{opc,?}, ZZ_d_mul_r, MatrixOp64,
3926                                               mnemonic, "vgx2"> {
3927    bits<4> Zd;
3928    let Inst{4-1} = Zd;
3929  }
3930
3931  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3932                                                ZZ_b_mul_r, MatrixOp8,
3933                                                MatrixIndexGPR32Op8_11,
3934                                                sme_elm_idx0_7, mnemonic>;
3935  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3936                                                ZZ_h_mul_r, MatrixOp16,
3937                                                MatrixIndexGPR32Op8_11,
3938                                                sme_elm_idx0_7, mnemonic>;
3939  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3940                                                ZZ_s_mul_r, MatrixOp32,
3941                                                MatrixIndexGPR32Op8_11,
3942                                                sme_elm_idx0_7, mnemonic>;
3943  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3944                                                ZZ_d_mul_r,  MatrixOp64,
3945                                                MatrixIndexGPR32Op8_11,
3946                                                sme_elm_idx0_7, mnemonic>;
3947
3948  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3949                                                ZZ_b_mul_r, MatrixOp8,
3950                                                MatrixIndexGPR32Op8_11,
3951                                                sme_elm_idx0_7, mnemonic, "vgx2">;
3952  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3953                                                ZZ_h_mul_r, MatrixOp16,
3954                                                MatrixIndexGPR32Op8_11,
3955                                                sme_elm_idx0_7, mnemonic, "vgx2">;
3956  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3957                                                ZZ_s_mul_r, MatrixOp32,
3958                                                MatrixIndexGPR32Op8_11,
3959                                                sme_elm_idx0_7, mnemonic, "vgx2">;
3960
3961  if !eq(mnemonic, "mova") then {
3962  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3963                                                ZZ_b_mul_r, MatrixOp8,
3964                                                MatrixIndexGPR32Op8_11,
3965                                                sme_elm_idx0_7, "mov">;
3966  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3967                                                ZZ_h_mul_r, MatrixOp16,
3968                                                MatrixIndexGPR32Op8_11,
3969                                                sme_elm_idx0_7, "mov">;
3970  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3971                                                ZZ_s_mul_r, MatrixOp32,
3972                                                MatrixIndexGPR32Op8_11,
3973                                                sme_elm_idx0_7, "mov">;
3974  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3975                                                ZZ_d_mul_r,  MatrixOp64,
3976                                                MatrixIndexGPR32Op8_11,
3977                                                sme_elm_idx0_7, "mov">;
3978
3979  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3980                                                ZZ_b_mul_r, MatrixOp8,
3981                                                MatrixIndexGPR32Op8_11,
3982                                                sme_elm_idx0_7, "mov", "vgx2">;
3983  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3984                                                ZZ_h_mul_r, MatrixOp16,
3985                                                MatrixIndexGPR32Op8_11,
3986                                                sme_elm_idx0_7, "mov", "vgx2">;
3987  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
3988                                                ZZ_s_mul_r, MatrixOp32,
3989                                                MatrixIndexGPR32Op8_11,
3990                                                sme_elm_idx0_7, "mov", "vgx2">;
3991  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME),
3992                                                ZZ_d_mul_r,  MatrixOp64,
3993                                                MatrixIndexGPR32Op8_11,
3994                                                sme_elm_idx0_7, "mov", "vgx2">;
3995  }
3996}
3997
3998// move array to vector, four registers
3999multiclass sme2_mova_array_to_vec_vg4_multi<bits<4> opc, string mnemonic> {
4000  def NAME : sme2_mova_array_to_vec_vg24_multi<opc, ZZZZ_d_mul_r, MatrixOp64,
4001                                               mnemonic, "vgx4"> {
4002    bits<3> Zd;
4003    let Inst{4-2} = Zd;
4004  }
4005
4006  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4007                                                ZZZZ_b_mul_r, MatrixOp8,
4008                                                MatrixIndexGPR32Op8_11,
4009                                                sme_elm_idx0_7, mnemonic>;
4010  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4011                                                ZZZZ_h_mul_r, MatrixOp16,
4012                                                MatrixIndexGPR32Op8_11,
4013                                                sme_elm_idx0_7, mnemonic>;
4014  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4015                                                ZZZZ_s_mul_r, MatrixOp32,
4016                                                MatrixIndexGPR32Op8_11,
4017                                                sme_elm_idx0_7, mnemonic>;
4018  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4019                                                ZZZZ_d_mul_r, MatrixOp64,
4020                                                MatrixIndexGPR32Op8_11,
4021                                                sme_elm_idx0_7, mnemonic>;
4022
4023  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4024                                                ZZZZ_b_mul_r, MatrixOp8,
4025                                                MatrixIndexGPR32Op8_11,
4026                                                sme_elm_idx0_7, mnemonic, "vgx4">;
4027  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4028                                                ZZZZ_h_mul_r, MatrixOp16,
4029                                                MatrixIndexGPR32Op8_11,
4030                                                sme_elm_idx0_7, mnemonic, "vgx4">;
4031  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4032                                                ZZZZ_s_mul_r, MatrixOp32,
4033                                                MatrixIndexGPR32Op8_11,
4034                                                sme_elm_idx0_7, mnemonic, "vgx4">;
4035
4036  if !eq(mnemonic, "mova") then {
4037  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4038                                                ZZZZ_b_mul_r, MatrixOp8,
4039                                                MatrixIndexGPR32Op8_11,
4040                                                sme_elm_idx0_7, "mov">;
4041  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4042                                                ZZZZ_h_mul_r, MatrixOp16,
4043                                                MatrixIndexGPR32Op8_11,
4044                                                sme_elm_idx0_7, "mov">;
4045  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4046                                                ZZZZ_s_mul_r, MatrixOp32,
4047                                                MatrixIndexGPR32Op8_11,
4048                                                sme_elm_idx0_7, "mov">;
4049  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4050                                                ZZZZ_d_mul_r, MatrixOp64,
4051                                                MatrixIndexGPR32Op8_11,
4052                                                sme_elm_idx0_7, "mov">;
4053
4054  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4055                                                ZZZZ_b_mul_r, MatrixOp8,
4056                                                MatrixIndexGPR32Op8_11,
4057                                                sme_elm_idx0_7, "mov", "vgx4">;
4058  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4059                                                ZZZZ_h_mul_r, MatrixOp16,
4060                                                MatrixIndexGPR32Op8_11,
4061                                                sme_elm_idx0_7, "mov", "vgx4">;
4062  defm : sme2_mova_tile_or_array_to_vec_aliases<0, !cast<Instruction>(NAME),
4063                                                ZZZZ_s_mul_r, MatrixOp32,
4064                                                MatrixIndexGPR32Op8_11,
4065                                                sme_elm_idx0_7, "mov", "vgx4">;
4066  defm : sme2_mova_tile_or_array_to_vec_aliases<1, !cast<Instruction>(NAME),
4067                                                ZZZZ_d_mul_r, MatrixOp64,
4068                                                MatrixIndexGPR32Op8_11,
4069                                                sme_elm_idx0_7, "mov", "vgx4">;
4070  }
4071}
4072
4073//===----------------------------------------------------------------------===//
4074// SME2 multi-vec saturating shift right narrow
4075class sme2_sat_shift_vector_vg2<string mnemonic, bit op, bit u>
4076    : I<(outs ZPR16:$Zd), (ins ZZ_s_mul_r:$Zn, tvecshiftR16:$imm4),
4077        mnemonic, "\t$Zd, $Zn, $imm4",
4078        "", []>, Sched<[]> {
4079  bits<4> imm4;
4080  bits<4> Zn;
4081  bits<5> Zd;
4082  let Inst{31-21} = 0b11000001111;
4083  let Inst{20}    = op;
4084  let Inst{19-16} = imm4;
4085  let Inst{15-10} = 0b110101;
4086  let Inst{9-6}   = Zn;
4087  let Inst{5}     = u;
4088  let Inst{4-0}   = Zd;
4089}
4090
4091multiclass sme2_sat_shift_vector_vg2<string mnemonic, bit op, bit u, SDPatternOperator intrinsic> {
4092  def _H : sme2_sat_shift_vector_vg2<mnemonic, op, u>;
4093
4094  def : SME2_Sat_Shift_VG2_Pat<NAME # _H, intrinsic, nxv8i16, nxv4i32, tvecshiftR16>;
4095}
4096
4097class sme2_sat_shift_vector_vg4<bits<2> sz, bits<3> op, ZPRRegOp zpr_ty,
4098                                RegisterOperand vector_ty, Operand imm_ty,
4099                                string mnemonic>
4100    : I<(outs zpr_ty:$Zd), (ins vector_ty:$Zn, imm_ty:$imm),
4101        mnemonic, "\t$Zd, $Zn, $imm",
4102        "", []>, Sched<[]> {
4103  bits<3> Zn;
4104  bits<5> Zd;
4105  let Inst{31-24} = 0b11000001;
4106  let Inst{23-22} = sz;
4107  let Inst{21}    = 0b1;
4108  //  Inst{20-16} = imm5;
4109  let Inst{15-11} = 0b11011;
4110  let Inst{10}    = op{2};
4111  let Inst{9-7}   = Zn;
4112  let Inst{6-5}   = op{1-0};
4113  let Inst{4-0}   = Zd;
4114}
4115
4116multiclass sme2_sat_shift_vector_vg4<string mnemonic, bits<3> op, SDPatternOperator intrinsic> {
4117  def _B : sme2_sat_shift_vector_vg4<{0,1}, op, ZPR8, ZZZZ_s_mul_r, tvecshiftR32,
4118                                     mnemonic>{
4119    bits<5> imm;
4120    let Inst{20-16} = imm;
4121  }
4122  def _H : sme2_sat_shift_vector_vg4<{1,?}, op, ZPR16, ZZZZ_d_mul_r, tvecshiftR64,
4123                                      mnemonic> {
4124    bits<6> imm;
4125    let Inst{22}    = imm{5};
4126    let Inst{20-16} = imm{4-0};
4127  }
4128
4129  def : SME2_Sat_Shift_VG4_Pat<NAME # _B, intrinsic, nxv16i8, nxv4i32, tvecshiftR32>;
4130  def : SME2_Sat_Shift_VG4_Pat<NAME # _H, intrinsic, nxv8i16, nxv2i64, tvecshiftR64>;
4131}
4132
4133//===----------------------------------------------------------------------===//
4134// SME2 Multi-vector - SVE Select
4135class sme2_sel_vector_vg24<bits<2> sz, bits<4> op, RegisterOperand vector_ty,
4136                           string mnemonic>
4137    : I<(outs vector_ty:$Zd),
4138        (ins PNRAny_p8to15:$PNg, vector_ty:$Zn, vector_ty:$Zm),
4139        mnemonic, "\t$Zd, $PNg, $Zn, $Zm",
4140        "", []>, Sched<[]> {
4141  bits<3> PNg;
4142  let Inst{31-24} = 0b11000001;
4143  let Inst{23-22} = sz;
4144  let Inst{21}    = 0b1;
4145  let Inst{17-16} = op{3-2};
4146  let Inst{15-13} = 0b100;
4147  let Inst{12-10} = PNg;
4148  let Inst{6}     = op{1};
4149  let Inst{5}     = 0b0;
4150  let Inst{1}     = op{0};
4151  let Inst{0}     = 0b0;
4152}
4153
4154class sme2_sel_vector_vg2<bits<2> sz, RegisterOperand vector_ty,
4155                          string mnemonic>
4156     : sme2_sel_vector_vg24<sz, {?,0,?,?}, vector_ty, mnemonic> {
4157  bits<4> Zm;
4158  bits<4> Zn;
4159  bits<4> Zd;
4160  let Inst{20-17} = Zm;
4161  let Inst{9-6}   = Zn;
4162  let Inst{4-1}   = Zd;
4163}
4164
4165multiclass sme2_sel_vector_vg2<string mnemonic>{
4166  def _B : sme2_sel_vector_vg2<0b00, ZZ_b_mul_r, mnemonic>;
4167  def _H : sme2_sel_vector_vg2<0b01, ZZ_h_mul_r, mnemonic>;
4168  def _S : sme2_sel_vector_vg2<0b10, ZZ_s_mul_r, mnemonic>;
4169  def _D : sme2_sel_vector_vg2<0b11, ZZ_d_mul_r, mnemonic>;
4170}
4171class sme2_sel_vector_vg4<bits<2> sz, RegisterOperand vector_ty,
4172                          string mnemonic>
4173     : sme2_sel_vector_vg24<sz, 0b0100, vector_ty, mnemonic> {
4174  bits<3> Zm;
4175  bits<3> Zn;
4176  bits<3> Zd;
4177  let Inst{20-18} = Zm;
4178  let Inst{9-7}   = Zn;
4179  let Inst{4-2}   = Zd;
4180}
4181multiclass sme2_sel_vector_vg4<string mnemonic> {
4182  def _B : sme2_sel_vector_vg4<0b00, ZZZZ_b_mul_r, mnemonic>;
4183  def _H : sme2_sel_vector_vg4<0b01, ZZZZ_h_mul_r, mnemonic>;
4184  def _S : sme2_sel_vector_vg4<0b10, ZZZZ_s_mul_r, mnemonic>;
4185  def _D : sme2_sel_vector_vg4<0b11, ZZZZ_d_mul_r, mnemonic>;
4186}
4187
4188//===----------------------------------------------------------------------===//
4189// Non contiguous Load and Store
4190
4191class sme2_ld_vector_vg2_multi_scalar_scalar<bits<2> msz, bit n,
4192                                             RegisterOperand multi_vector_ty,
4193                                             RegisterOperand gpr_ty,
4194                                             string mnemonic>
4195   : I<(outs multi_vector_ty:$Zt),
4196       (ins PNRAny_p8to15:$PNg, GPR64sp:$Rn, gpr_ty:$Rm),
4197       mnemonic, "\t$Zt, $PNg/z, [$Rn, $Rm]",
4198       "", []>, Sched<[]> {
4199   bits<5> Rm;
4200   bits<3> PNg;
4201   bits<5> Rn;
4202   bits<4> Zt;
4203   let Inst{31-21} = 0b10100001000;
4204   let Inst{20-16} = Rm;
4205   let Inst{15}    = 0b0;
4206   let Inst{14-13} = msz;
4207   let Inst{12-10} = PNg;
4208   let Inst{9-5}   = Rn;
4209   let Inst{4}     = Zt{3};
4210   let Inst{3}     = n;
4211   let Inst{2-0}   = Zt{2-0};
4212
4213   let mayLoad = 1;
4214}
4215
4216class sme2_ld_vector_vg4_multi_scalar_scalar<bits<2> msz, bit n,
4217                                             RegisterOperand multi_vector_ty,
4218                                             RegisterOperand gpr_ty,
4219                                             string mnemonic>
4220   : I<(outs multi_vector_ty:$Zt),
4221       (ins PNRAny_p8to15:$PNg, GPR64sp:$Rn, gpr_ty:$Rm),
4222       mnemonic, "\t$Zt, $PNg/z, [$Rn, $Rm]",
4223       "", []>, Sched<[]> {
4224   bits<5> Rm;
4225   bits<3> PNg;
4226   bits<5> Rn;
4227   bits<3> Zt;
4228   let Inst{31-21} = 0b10100001000;
4229   let Inst{20-16} = Rm;
4230   let Inst{15}    = 0b1;
4231   let Inst{14-13} = msz;
4232   let Inst{12-10} = PNg;
4233   let Inst{9-5}   = Rn;
4234   let Inst{4}     = Zt{2};
4235   let Inst{3}     = n;
4236   let Inst{2}     = 0b0;
4237   let Inst{1-0}   = Zt{1-0};
4238
4239   let mayLoad = 1;
4240}
4241
4242class sme2_ld_vector_vg24_multi_scalar_immediate<bits<2> msz, bit n, bits<2> op,
4243                                                 RegisterOperand multi_vector_ty,
4244                                                 Operand index_ty,
4245                                                 string mnemonic>
4246    : I<(outs multi_vector_ty:$Zt),
4247        (ins PNRAny_p8to15:$PNg, GPR64sp:$Rn, index_ty:$imm4),
4248        mnemonic,  "\t$Zt, $PNg/z, [$Rn, $imm4, mul vl]",
4249        "", []>, Sched<[]> {
4250   bits<4> imm4;
4251   bits<3> PNg;
4252   bits<5> Rn;
4253   let Inst{31-20} = 0b101000010100;
4254   let Inst{19-16} = imm4;
4255   let Inst{15}    = op{1};
4256   let Inst{14-13} = msz;
4257   let Inst{12-10} = PNg;
4258   let Inst{9-5}   = Rn;
4259   let Inst{3}     = n;
4260   let Inst{2}     = op{0};
4261
4262   let mayLoad = 1;
4263}
4264
4265multiclass sme2_ld_vector_vg2_multi_scalar_immediate<bits<2> msz, bit n,
4266                                                     RegisterOperand multi_vector_ty,
4267                                                     Operand index_ty,
4268                                                     string mnemonic>{
4269  def NAME : sme2_ld_vector_vg24_multi_scalar_immediate<msz, n, {0,?},
4270                                                        multi_vector_ty,
4271                                                        index_ty, mnemonic> {
4272    bits<4> Zt;
4273    let Inst{4} = Zt{3};
4274    let Inst{2-0} = Zt{2-0};
4275  }
4276
4277   def : InstAlias<mnemonic # "\t$Zt, $PNg/z, [$Rn]",
4278                  (!cast<Instruction>(NAME) multi_vector_ty:$Zt, PNRAny_p8to15:$PNg, GPR64sp:$Rn, 0), 1>;
4279}
4280
4281multiclass sme2_ld_vector_vg4_multi_scalar_immediate<bits<2> msz, bit n,
4282                                                     RegisterOperand multi_vector_ty,
4283                                                     Operand index_ty,
4284                                                     string mnemonic> {
4285  def NAME : sme2_ld_vector_vg24_multi_scalar_immediate<msz, n, 0b10,
4286                                                        multi_vector_ty,
4287                                                        index_ty, mnemonic> {
4288    bits<3> Zt;
4289    let Inst{4} = Zt{2};
4290    let Inst{1-0} = Zt{1-0};
4291  }
4292
4293   def : InstAlias<mnemonic # "\t$Zt, $PNg/z, [$Rn]",
4294                   (!cast<Instruction>(NAME) multi_vector_ty:$Zt, PNRAny_p8to15:$PNg, GPR64sp:$Rn, 0), 1>;
4295}
4296
4297//===----------------------------------------------------------------------===//
4298// SME2 Non-Contiguous Store
4299class sme2_st_vector_vg2_multi_scalar_scalar<bits<2> msz, bit n,
4300                                             RegisterOperand multi_vector_ty,
4301                                             RegisterOperand gpr_ty,
4302                                             string mnemonic>
4303   : I<(outs ),
4304       (ins multi_vector_ty:$Zt, PNRAny_p8to15:$PNg, GPR64sp:$Rn, gpr_ty:$Rm),
4305       mnemonic, "\t$Zt, $PNg, [$Rn, $Rm]",
4306       "", []>, Sched<[]> {
4307   bits<5> Rm;
4308   bits<3> PNg;
4309   bits<5> Rn;
4310   bits<4> Zt;
4311   let Inst{31-21} = 0b10100001001;
4312   let Inst{20-16} = Rm;
4313   let Inst{15}    = 0b0;
4314   let Inst{14-13} = msz;
4315   let Inst{12-10} = PNg;
4316   let Inst{9-5}   = Rn;
4317   let Inst{4}     = Zt{3};
4318   let Inst{3}     = n;
4319   let Inst{2-0}   = Zt{2-0};
4320
4321   let mayStore    = 1;
4322}
4323
4324class sme2_st_vector_vg4_multi_scalar_scalar<bits<2> msz, bit n,
4325                                             RegisterOperand multi_vector_ty,
4326                                             RegisterOperand gpr_ty,
4327                                             string mnemonic>
4328   : I<(outs ),
4329       (ins multi_vector_ty:$Zt, PNRAny_p8to15:$PNg, GPR64sp:$Rn, gpr_ty:$Rm),
4330       mnemonic, "\t$Zt, $PNg, [$Rn, $Rm]",
4331       "", []>, Sched<[]> {
4332   bits<5> Rm;
4333   bits<3> PNg;
4334   bits<5> Rn;
4335   bits<3> Zt;
4336   let Inst{31-21} = 0b10100001001;
4337   let Inst{20-16} = Rm;
4338   let Inst{15}     = 0b1;
4339   let Inst{14-13} = msz;
4340   let Inst{12-10} = PNg;
4341   let Inst{9-5}   = Rn;
4342   let Inst{4}     = Zt{2};
4343   let Inst{3}     = n;
4344   let Inst{2}     = 0b0;
4345   let Inst{1-0}   = Zt{1-0};
4346
4347   let mayStore    = 1;
4348}
4349
4350class sme2_st_vector_vg24_multi_scalar_immediate<bits<2> msz, bit n, bits<2> op,
4351                                                 RegisterOperand multi_vector_ty,
4352                                                 Operand index_ty,
4353                                                 string mnemonic>
4354    : I<(outs ),
4355        (ins multi_vector_ty:$Zt, PNRAny_p8to15:$PNg, GPR64sp:$Rn, index_ty:$imm4),
4356        mnemonic,  "\t$Zt, $PNg, [$Rn, $imm4, mul vl]",
4357        "", []>, Sched<[]> {
4358   bits<4> imm4;
4359   bits<3> PNg;
4360   bits<5> Rn;
4361   let Inst{31-20} = 0b101000010110;
4362   let Inst{19-16} = imm4;
4363   let Inst{15}    = op{1};
4364   let Inst{14-13} = msz;
4365   let Inst{12-10} = PNg;
4366   let Inst{9-5}   = Rn;
4367   let Inst{3}     = n;
4368   let Inst{2}     = op{0};
4369
4370   let mayStore    = 1;
4371}
4372
4373
4374multiclass sme2_st_vector_vg2_multi_scalar_immediate<bits<2> msz, bit n,
4375                                                     RegisterOperand multi_vector_ty,
4376                                                     Operand index_ty,
4377                                                     string mnemonic> {
4378  def NAME: sme2_st_vector_vg24_multi_scalar_immediate<msz, n, {0,?},
4379                                                       multi_vector_ty,
4380                                                       index_ty, mnemonic> {
4381    bits<4> Zt;
4382    let Inst{4}   = Zt{3};
4383    let Inst{2-0} = Zt{2-0};
4384  }
4385
4386    def : InstAlias<mnemonic # "\t$Zt, $PNg, [$Rn]",
4387                   (!cast<Instruction>(NAME) multi_vector_ty:$Zt, PNRAny_p8to15:$PNg, GPR64sp:$Rn,0), 1>;
4388}
4389
4390multiclass sme2_st_vector_vg4_multi_scalar_immediate<bits<2> msz, bit n,
4391                                                     RegisterOperand multi_vector_ty,
4392                                                     Operand index_ty,
4393                                                     string mnemonic> {
4394  def NAME : sme2_st_vector_vg24_multi_scalar_immediate<msz, n, 0b10,
4395                                                        multi_vector_ty,
4396                                                        index_ty, mnemonic> {
4397    bits<3> Zt;
4398    let Inst{4}   = Zt{2};
4399    let Inst{1-0} = Zt{1-0};
4400  }
4401
4402    def : InstAlias<mnemonic # "\t$Zt, $PNg, [$Rn]",
4403                   (!cast<Instruction>(NAME) multi_vector_ty:$Zt, PNRAny_p8to15:$PNg, GPR64sp:$Rn,0), 1>;
4404}
4405
4406//===----------------------------------------------------------------------===//
4407// SME2.1
4408//===----------------------------------------------------------------------===//
4409// SME zeroing move array to vector
4410class sme2p1_movaz_tile_to_vec_base<bits<2> sz, bit q, bit v, ZPRRegOp vector_ty,
4411                                    RegisterOperand tile_ty, Operand index_ty,
4412                                    string mnemonic>
4413    : I<(outs vector_ty:$Zd, tile_ty:$ZAn),
4414        (ins tile_ty:$_ZAn, MatrixIndexGPR32Op12_15:$Rs, index_ty:$imm),
4415        mnemonic, "\t$Zd, $ZAn[$Rs, $imm]",
4416        "", []>, Sched<[]> {
4417  bits<2> Rs;
4418  bits<5> Zd;
4419  let Inst{31-24} = 0b11000000;
4420  let Inst{23-22} = sz;
4421  let Inst{21-17} = 0b00001;
4422  let Inst{16}    = q;
4423  let Inst{15}    = v;
4424  let Inst{14-13} = Rs;
4425  let Inst{12-9}  = 0b0001;
4426  let Inst{4-0}   = Zd;
4427  let Constraints = "$ZAn = $_ZAn";
4428}
4429
4430multiclass sme2p1_movaz_tile_to_vec_base<bit v, string mnemonic> {
4431  def _B : sme2p1_movaz_tile_to_vec_base<0b00, 0b0, v, ZPR8,
4432                                    !if(v, TileVectorOpV8, TileVectorOpH8),
4433                                    sme_elm_idx0_15, mnemonic> {
4434    bits<4> imm;
4435    let Inst{8-5} = imm;
4436  }
4437
4438  def _H : sme2p1_movaz_tile_to_vec_base<0b01, 0b0, v, ZPR16,
4439                                    !if(v, TileVectorOpV16, TileVectorOpH16),
4440                                    sme_elm_idx0_7, mnemonic> {
4441    bits<1> ZAn;
4442    bits<3> imm;
4443    let Inst{8}   = ZAn;
4444    let Inst{7-5} = imm;
4445  }
4446
4447  def _S : sme2p1_movaz_tile_to_vec_base<0b10, 0b0, v, ZPR32,
4448                                    !if(v, TileVectorOpV32, TileVectorOpH32),
4449                                    sme_elm_idx0_3, mnemonic> {
4450    bits<2> ZAn;
4451    bits<2> imm;
4452    let Inst{8-7} = ZAn;
4453    let Inst{6-5} = imm;
4454  }
4455
4456  def _D : sme2p1_movaz_tile_to_vec_base<0b11, 0b0, v, ZPR64,
4457                                    !if(v, TileVectorOpV64, TileVectorOpH64),
4458                                    sme_elm_idx0_1, mnemonic> {
4459    bits<3> ZAn;
4460    bits<1> imm;
4461    let Inst{8-6} = ZAn;
4462    let Inst{5}   = imm;
4463  }
4464
4465  def _Q : sme2p1_movaz_tile_to_vec_base<0b11, 0b1, v, ZPR128,
4466                                    !if(v, TileVectorOpV128, TileVectorOpH128),
4467                                    sme_elm_idx0_0, mnemonic> {
4468    bits<4> ZAn;
4469    let Inst{8-5} = ZAn;
4470  }
4471}
4472
4473multiclass sme2p1_movaz_tile_to_vec<string mnemonic>{
4474 defm _H : sme2p1_movaz_tile_to_vec_base<0b0, mnemonic>;
4475 defm _V : sme2p1_movaz_tile_to_vec_base<0b1, mnemonic>;
4476}
4477
4478//===----------------------------------------------------------------------===//
4479// SME2.1 multiple vectors zero array
4480
4481class sme2p1_zero_matrix<bits<6> opc, Operand index_ty, string mnemonic,
4482                         string vg_acronym="">
4483    : I<(outs MatrixOp64:$ZAd),
4484        (ins MatrixOp64:$_ZAd, MatrixIndexGPR32Op8_11:$Rv, index_ty:$imm),
4485        mnemonic, "\t$ZAd[$Rv, $imm" # !if(!eq(vg_acronym, ""), "", ", " # vg_acronym) # "]",
4486        "", []>, Sched<[]> {
4487  bits <2> Rv;
4488  let Inst{31-18} = 0b11000000000011;
4489  let Inst{17-15} = opc{5-3};
4490  let Inst{14-13} = Rv;
4491  let Inst{12-3} = 0b0000000000;
4492  let Inst{2-0}  = opc{2-0};
4493  let Constraints = "$ZAd = $_ZAd";
4494}
4495
4496multiclass sme2p1_zero_matrix<string mnemonic> {
4497  def _VG2_Z : sme2p1_zero_matrix<{0b000,?,?,?}, sme_elm_idx0_7, mnemonic, "vgx2"> {
4498    bits<3> imm;
4499    let Inst{2-0} = imm;
4500  }
4501  def _2Z : sme2p1_zero_matrix<{0b001,?,?,?}, uimm3s2range, mnemonic> {
4502    bits<3> imm;
4503    let Inst{2-0} = imm;
4504  }
4505  def _VG2_2Z : sme2p1_zero_matrix<{0b0100,?,?}, uimm2s2range, mnemonic, "vgx2"> {
4506    bits<2> imm;
4507    let Inst{1-0} = imm;
4508  }
4509  def _VG4_2Z : sme2p1_zero_matrix<{0b0110,?,?}, uimm2s2range, mnemonic, "vgx4"> {
4510    bits<2> imm;
4511    let Inst{1-0} = imm;
4512  }
4513  def _VG4_Z : sme2p1_zero_matrix<{0b100,?,?,?}, sme_elm_idx0_7, mnemonic, "vgx4"> {
4514    bits<3> imm;
4515    let Inst{2-0} = imm;
4516  }
4517  def _4Z : sme2p1_zero_matrix<{0b1010,?,?}, uimm2s4range, mnemonic> {
4518    bits<2> imm;
4519    let Inst{1-0} = imm;
4520  }
4521  def _VG2_4Z :sme2p1_zero_matrix<{0b11000,?}, uimm1s4range, mnemonic, "vgx2"> {
4522    bits<1> imm;
4523    let Inst{0}   = imm;
4524  }
4525  def _VG4_4Z :sme2p1_zero_matrix<{0b11100,?}, uimm1s4range, mnemonic, "vgx4"> {
4526    bits<1> imm;
4527    let Inst{0}   = imm;
4528  }
4529}
4530
4531//===----------------------------------------------------------------------===//
4532// SME2.1 lookup table expand two non-contiguous registers
4533
4534class sme2p1_luti_vector_vg2_index<bits<4> op, bits<2> sz, RegisterOperand vector_ty,
4535                                   AsmVectorIndexOpnd index_ty,
4536                                   string mnemonic>
4537    :  I<(outs vector_ty:$Zd), (ins ZTR:$ZTt, ZPRAny:$Zn, index_ty:$i),
4538          mnemonic, "\t$Zd, $ZTt, $Zn$i",
4539          "", []>, Sched<[]> {
4540  bits<5> Zn;
4541  bits<4> Zd;
4542  let Inst{31-19} = 0b1100000010011;
4543  let Inst{18-15} = op;
4544  let Inst{14}    = 0b1;
4545  let Inst{13-12} = sz;
4546  let Inst{11-10} = 0b00;
4547  let Inst{9-5}   = Zn;
4548  let Inst{4}     = Zd{3};
4549  let Inst{3}     = 0b0;
4550  let Inst{2-0}   = Zd{2-0};
4551}
4552
4553class sme2p1_luti2_vector_vg2_index<bits<2> sz, RegisterOperand vector_ty,
4554                                    AsmVectorIndexOpnd index_ty,
4555                                    string mnemonic>
4556  : sme2p1_luti_vector_vg2_index<{1,?,?,?}, sz, vector_ty, index_ty, mnemonic> {
4557  bits<3> i;
4558  let Inst{17-15} = i;
4559}
4560
4561multiclass sme2p1_luti2_vector_vg2_index<string mnemonic> {
4562  def _B : sme2p1_luti2_vector_vg2_index<0b00, ZZ_b_strided, VectorIndexH,
4563                                         mnemonic>;
4564  def _H : sme2p1_luti2_vector_vg2_index<0b01, ZZ_h_strided, VectorIndexH,
4565                                         mnemonic>;
4566}
4567
4568class sme2p1_luti4_vector_vg2_index<bits<2> sz, RegisterOperand vector_ty,
4569                                    AsmVectorIndexOpnd index_ty,
4570                                    string mnemonic>
4571  : sme2p1_luti_vector_vg2_index<{0b01,?,?}, sz, vector_ty, index_ty, mnemonic> {
4572  bits<2> i;
4573  let Inst{16-15} = i;
4574}
4575multiclass sme2p1_luti4_vector_vg2_index<string mnemonic> {
4576  def _B : sme2p1_luti4_vector_vg2_index<0b00, ZZ_b_strided, VectorIndexS,
4577                                         mnemonic>;
4578  def _H : sme2p1_luti4_vector_vg2_index<0b01, ZZ_h_strided, VectorIndexS,
4579                                         mnemonic>;
4580}
4581
4582// SME2.1 lookup table expand four non-contiguous registers
4583class sme2p1_luti_vector_vg4_index<bits<3> op, bits<2> sz, RegisterOperand vector_ty,
4584                                   AsmVectorIndexOpnd index_ty,
4585                                   string mnemonic>
4586    :  I<(outs vector_ty:$Zd), (ins ZTR:$ZTt, ZPRAny:$Zn, index_ty:$i),
4587          mnemonic, "\t$Zd, $ZTt, $Zn$i",
4588          "", []>, Sched<[]> {
4589  bits<5> Zn;
4590  bits<3> Zd;
4591  let Inst{31-19} = 0b1100000010011;
4592  let Inst{18-16} = op;
4593  let Inst{15-14} = 0b10;
4594  let Inst{13-12} = sz;
4595  let Inst{11-10} = 0b00;
4596  let Inst{9-5}   = Zn;
4597  let Inst{4}     = Zd{2};
4598  let Inst{3-2}   = 0b00;
4599  let Inst{1-0}   = Zd{1-0};
4600}
4601
4602class sme2p1_luti2_vector_vg4_index<bits<2> sz, RegisterOperand vector_ty,
4603                                    AsmVectorIndexOpnd index_ty,
4604                                    string mnemonic>
4605  : sme2p1_luti_vector_vg4_index<{1,?,?}, sz, vector_ty, index_ty, mnemonic> {
4606  bits<2> i;
4607  let Inst{17-16} = i;
4608}
4609
4610multiclass sme2p1_luti2_vector_vg4_index<string mnemonic> {
4611  def _B : sme2p1_luti2_vector_vg4_index<0b00, ZZZZ_b_strided, VectorIndexS,
4612                                         mnemonic>;
4613  def _H : sme2p1_luti2_vector_vg4_index<0b01, ZZZZ_h_strided, VectorIndexS,
4614                                         mnemonic>;
4615}
4616
4617class sme2p1_luti4_vector_vg4_index<bits<2> sz, RegisterOperand vector_ty,
4618                                    AsmVectorIndexOpnd index_ty,
4619                                    string mnemonic>
4620  : sme2p1_luti_vector_vg4_index<{0b01,?}, sz, vector_ty, index_ty, mnemonic> {
4621  bit i;
4622  let Inst{16}    = i;
4623}
4624
4625multiclass sme2p1_luti4_vector_vg4_index<string mnemonic> {
4626  def _H: sme2p1_luti4_vector_vg4_index<0b01, ZZZZ_h_strided, VectorIndexD, mnemonic>;
4627}
4628