1 //=- AArch64SchedExynosM4.td - Samsung Exynos M4 Sched Defs --*- tablegen -*-=//
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
7 //===----------------------------------------------------------------------===//
9 // This file defines the machine model for the Samsung Exynos M4 to support
10 // instruction scheduling and other instruction cost heuristics.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // The Exynos-M4 is an advanced superscalar microprocessor with a 6-wide
16 // in-order stage for decode and dispatch and a wider issue stage.
17 // The execution units and loads and stores are out-of-order.
19 def ExynosM4Model : SchedMachineModel {
20 let IssueWidth = 6; // Up to 6 uops per cycle.
21 let MicroOpBufferSize = 228; // ROB size.
22 let LoopMicroOpBufferSize = 48; // Based on the instruction queue size.
23 let LoadLatency = 4; // Optimistic load cases.
24 let MispredictPenalty = 16; // Minimum branch misprediction penalty.
25 let CompleteModel = 1; // Use the default model otherwise.
27 list<Predicate> UnsupportedFeatures = SVEUnsupported.F;
30 //===----------------------------------------------------------------------===//
31 // Define each kind of processor resource and number available on the Exynos-M4.
33 let SchedModel = ExynosM4Model in {
35 def M4UnitA : ProcResource<2>; // Simple integer
36 def M4UnitC : ProcResource<2>; // Simple and complex integer
37 let Super = M4UnitC, BufferSize = 1 in
38 def M4UnitD : ProcResource<1>; // Integer division (inside C0, serialized)
39 let Super = M4UnitC in
40 def M4UnitE : ProcResource<1>; // CRC (inside C0)
41 def M4UnitB : ProcResource<2>; // Branch
42 def M4UnitL0 : ProcResource<1>; // Load
43 def M4UnitS0 : ProcResource<1>; // Store
44 def M4PipeLS : ProcResource<1>; // Load/Store
45 let Super = M4PipeLS in {
46 def M4UnitL1 : ProcResource<1>;
47 def M4UnitS1 : ProcResource<1>;
49 def M4PipeF0 : ProcResource<1>; // FP #0
50 let Super = M4PipeF0 in {
51 def M4UnitFMAC0 : ProcResource<1>; // FP multiplication
52 def M4UnitFADD0 : ProcResource<1>; // Simple FP
53 def M4UnitFCVT0 : ProcResource<1>; // FP conversion
54 def M4UnitNALU0 : ProcResource<1>; // Simple vector
55 def M4UnitNHAD : ProcResource<1>; // Horizontal vector
56 def M4UnitNMSC : ProcResource<1>; // FP and vector miscellanea
57 def M4UnitNMUL0 : ProcResource<1>; // Vector multiplication
58 def M4UnitNSHT0 : ProcResource<1>; // Vector shifting
59 def M4UnitNSHF0 : ProcResource<1>; // Vector shuffling
60 def M4UnitNCRY0 : ProcResource<1>; // Cryptographic
62 def M4PipeF1 : ProcResource<1>; // FP #1
63 let Super = M4PipeF1 in {
64 def M4UnitFMAC1 : ProcResource<1>; // FP multiplication
65 def M4UnitFADD1 : ProcResource<1>; // Simple FP
66 def M4UnitFDIV0 : ProcResource<2>; // FP division (serialized)
67 def M4UnitFSQR0 : ProcResource<2>; // FP square root (serialized)
68 def M4UnitFST0 : ProcResource<1>; // FP store
69 def M4UnitNALU1 : ProcResource<1>; // Simple vector
70 def M4UnitNSHT1 : ProcResource<1>; // Vector shifting
71 def M4UnitNSHF1 : ProcResource<1>; // Vector shuffling
73 def M4PipeF2 : ProcResource<1>; // FP #2
74 let Super = M4PipeF2 in {
75 def M4UnitFMAC2 : ProcResource<1>; // FP multiplication
76 def M4UnitFADD2 : ProcResource<1>; // Simple FP
77 def M4UnitFCVT1 : ProcResource<1>; // FP conversion
78 def M4UnitFDIV1 : ProcResource<2>; // FP division (serialized)
79 def M4UnitFSQR1 : ProcResource<2>; // FP square root (serialized)
80 def M4UnitFST1 : ProcResource<1>; // FP store
81 def M4UnitNALU2 : ProcResource<1>; // Simple vector
82 def M4UnitNMUL1 : ProcResource<1>; // Vector multiplication
83 def M4UnitNSHT2 : ProcResource<1>; // Vector shifting
84 def M4UnitNCRY1 : ProcResource<1>; // Cryptographic
87 def M4UnitALU : ProcResGroup<[M4UnitA,
89 def M4UnitL : ProcResGroup<[M4UnitL0,
91 def M4UnitS : ProcResGroup<[M4UnitS0,
93 def M4UnitFMAC : ProcResGroup<[M4UnitFMAC0,
96 def M4UnitFMACH : ProcResGroup<[M4UnitFMAC0,
98 def M4UnitFADD : ProcResGroup<[M4UnitFADD0,
101 def M4UnitFADDH : ProcResGroup<[M4UnitFADD0,
103 def M4UnitFCVT : ProcResGroup<[M4UnitFCVT0,
105 def M4UnitFCVTH : ProcResGroup<[M4UnitFCVT0]>;
106 def M4UnitFDIV : ProcResGroup<[M4UnitFDIV0,
108 def M4UnitFDIVH : ProcResGroup<[M4UnitFDIV0]>;
109 def M4UnitFSQR : ProcResGroup<[M4UnitFSQR0,
111 def M4UnitFSQRH : ProcResGroup<[M4UnitFSQR0]>;
112 def M4UnitFST : ProcResGroup<[M4UnitFST0,
114 def M4UnitNALU : ProcResGroup<[M4UnitNALU0,
117 def M4UnitNALUH : ProcResGroup<[M4UnitNALU0,
119 def M4UnitNMUL : ProcResGroup<[M4UnitNMUL0,
121 def M4UnitNSHT : ProcResGroup<[M4UnitNSHT0,
124 def M4UnitNSHF : ProcResGroup<[M4UnitNSHF0,
126 def M4UnitNSHFH : ProcResGroup<[M4UnitNSHF0]>;
127 def M4UnitNCRY : ProcResGroup<[M4UnitNCRY0,
130 //===----------------------------------------------------------------------===//
131 // Resources details.
133 def M4WriteZ0 : SchedWriteRes<[]> { let Latency = 0; }
134 def M4WriteZ1 : SchedWriteRes<[]> { let Latency = 1;
135 let NumMicroOps = 0; }
136 def M4WriteZ4 : SchedWriteRes<[]> { let Latency = 4;
137 let NumMicroOps = 0; }
139 def M4WriteA1 : SchedWriteRes<[M4UnitALU]> { let Latency = 1; }
140 def M4WriteA2 : SchedWriteRes<[M4UnitALU]> { let Latency = 2; }
141 def M4WriteAA : SchedWriteRes<[M4UnitALU]> { let Latency = 2;
142 let ResourceCycles = [2]; }
143 def M4WriteAB : SchedWriteRes<[M4UnitALU,
144 M4UnitC]> { let Latency = 2;
145 let NumMicroOps = 2; }
146 def M4WriteAC : SchedWriteRes<[M4UnitALU,
148 M4UnitC]> { let Latency = 3;
149 let NumMicroOps = 3; }
150 def M4WriteAD : SchedWriteRes<[M4UnitALU,
151 M4UnitC]> { let Latency = 2;
152 let NumMicroOps = 2; }
153 def M4WriteAF : SchedWriteRes<[M4UnitALU]> { let Latency = 2;
154 let NumMicroOps = 2; }
155 def M4WriteAU : SchedWriteVariant<[SchedVar<IsCopyIdiomPred, [M4WriteZ0]>,
156 SchedVar<ExynosArithPred, [M4WriteA1]>,
157 SchedVar<ExynosLogicExPred, [M4WriteA1]>,
158 SchedVar<NoSchedPred, [M4WriteAA]>]>;
159 def M4WriteAV : SchedWriteVariant<[SchedVar<ExynosResetPred, [M4WriteZ0]>,
160 SchedVar<ExynosArithPred, [M4WriteA1]>,
161 SchedVar<ExynosLogicExPred, [M4WriteA1]>,
162 SchedVar<NoSchedPred, [M4WriteAA]>]>;
163 def M4WriteAX : SchedWriteVariant<[SchedVar<ExynosArithPred, [M4WriteA1]>,
164 SchedVar<ExynosLogicExPred, [M4WriteA1]>,
165 SchedVar<NoSchedPred, [M4WriteAA]>]>;
166 def M4WriteAY : SchedWriteVariant<[SchedVar<ExynosRotateRightImmPred, [M4WriteA1]>,
167 SchedVar<NoSchedPred, [M4WriteAF]>]>;
169 def M4WriteB1 : SchedWriteRes<[M4UnitB]> { let Latency = 1; }
170 def M4WriteBX : SchedWriteVariant<[SchedVar<ExynosBranchLinkLRPred, [M4WriteAC]>,
171 SchedVar<NoSchedPred, [M4WriteAB]>]>;
173 def M4WriteC1 : SchedWriteRes<[M4UnitC]> { let Latency = 1; }
174 def M4WriteC3 : SchedWriteRes<[M4UnitC]> { let Latency = 3; }
175 def M4WriteCA : SchedWriteRes<[M4UnitC]> { let Latency = 4;
176 let ResourceCycles = [2]; }
178 def M4WriteD12 : SchedWriteRes<[M4UnitD]> { let Latency = 12;
179 let ResourceCycles = [12]; }
180 def M4WriteD21 : SchedWriteRes<[M4UnitD]> { let Latency = 21;
181 let ResourceCycles = [21]; }
183 def M4WriteE2 : SchedWriteRes<[M4UnitE]> { let Latency = 2; }
185 def M4WriteL4 : SchedWriteRes<[M4UnitL]> { let Latency = 4; }
186 def M4WriteL5 : SchedWriteRes<[M4UnitL]> { let Latency = 5; }
187 def M4WriteLA : SchedWriteRes<[M4UnitL,
188 M4UnitL]> { let Latency = 5;
189 let NumMicroOps = 1; }
190 def M4WriteLB : SchedWriteRes<[M4UnitA,
191 M4UnitL]> { let Latency = 5;
192 let NumMicroOps = 2; }
193 def M4WriteLC : SchedWriteRes<[M4UnitA,
195 M4UnitL]> { let Latency = 5;
196 let NumMicroOps = 2; }
197 def M4WriteLD : SchedWriteRes<[M4UnitA,
198 M4UnitL]> { let Latency = 4;
199 let NumMicroOps = 2; }
200 def M4WriteLE : SchedWriteRes<[M4UnitA,
201 M4UnitL]> { let Latency = 6;
202 let NumMicroOps = 2; }
203 def M4WriteLH : SchedWriteRes<[]> { let Latency = 5;
204 let NumMicroOps = 0; }
205 def M4WriteLX : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M4WriteL5]>,
206 SchedVar<NoSchedPred, [M4WriteL4]>]>;
207 def M4WriteLY : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M4WriteLE]>,
208 SchedVar<NoSchedPred, [M4WriteL5]>]>;
210 def M4WriteS1 : SchedWriteRes<[M4UnitS]> { let Latency = 1; }
211 def M4WriteSA : SchedWriteRes<[M4UnitS0]> { let Latency = 3; }
212 def M4WriteSB : SchedWriteRes<[M4UnitA,
213 M4UnitS]> { let Latency = 2;
214 let NumMicroOps = 1; }
215 def M4WriteSX : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M4WriteSB]>,
216 SchedVar<NoSchedPred, [M4WriteS1]>]>;
218 def M4ReadAdrBase : SchedReadVariant<[SchedVar<
222 ExynosScaledIdxFn]>>, [ReadDefault]>,
223 SchedVar<NoSchedPred, [ReadDefault]>]>;
225 def M4WriteNEONA : SchedWriteRes<[M4UnitNSHF,
226 M4UnitFADD]> { let Latency = 3;
227 let NumMicroOps = 2; }
228 def M4WriteNEONB : SchedWriteRes<[M4UnitNALU,
229 M4UnitS0]> { let Latency = 5;
230 let NumMicroOps = 2; }
231 def M4WriteNEOND : SchedWriteRes<[M4UnitNSHF,
232 M4UnitFST]> { let Latency = 6;
233 let NumMicroOps = 2; }
234 def M4WriteNEONH : SchedWriteRes<[M4UnitNALU,
235 M4UnitS0]> { let Latency = 5;
236 let NumMicroOps = 2; }
237 def M4WriteNEONI : SchedWriteRes<[M4UnitNSHF,
238 M4UnitS0]> { let Latency = 2;
239 let NumMicroOps = 2; }
240 def M4WriteNEONJ : SchedWriteRes<[M4UnitNMSC,
241 M4UnitS0]> { let Latency = 4; }
242 def M4WriteNEONK : SchedWriteRes<[M4UnitNSHF,
244 M4UnitS0]> { let Latency = 5;
245 let NumMicroOps = 2; }
246 def M4WriteNEONL : SchedWriteRes<[M4UnitNMUL]> { let Latency = 3; }
247 def M4WriteNEONN : SchedWriteRes<[M4UnitNMSC,
248 M4UnitNMSC]> { let Latency = 5;
249 let NumMicroOps = 2; }
250 def M4WriteNEONO : SchedWriteRes<[M4UnitNMSC,
252 M4UnitNMSC]> { let Latency = 8;
253 let NumMicroOps = 3; }
254 def M4WriteNEONP : SchedWriteRes<[M4UnitNSHF,
255 M4UnitNMSC]> { let Latency = 4;
256 let NumMicroOps = 2; }
257 def M4WriteNEONQ : SchedWriteRes<[M4UnitNMSC,
258 M4UnitC]> { let Latency = 3;
259 let NumMicroOps = 1; }
260 def M4WriteNEONR : SchedWriteRes<[M4UnitFCVT0,
261 M4UnitS0]> { let Latency = 4;
262 let NumMicroOps = 1; }
263 def M4WriteNEONV : SchedWriteRes<[M4UnitFDIV,
264 M4UnitFDIV]> { let Latency = 7;
265 let ResourceCycles = [6, 6]; }
266 def M4WriteNEONVH : SchedWriteRes<[M4UnitFDIVH,
267 M4UnitFDIVH]> { let Latency = 7;
268 let ResourceCycles = [6, 6]; }
269 def M4WriteNEONW : SchedWriteRes<[M4UnitFDIV,
270 M4UnitFDIV]> { let Latency = 12;
271 let ResourceCycles = [9, 9]; }
272 def M4WriteNEONX : SchedWriteRes<[M4UnitFSQR,
273 M4UnitFSQR]> { let Latency = 8;
274 let ResourceCycles = [7, 7]; }
275 def M4WriteNEONXH : SchedWriteRes<[M4UnitFSQRH,
276 M4UnitFSQRH]> { let Latency = 7;
277 let ResourceCycles = [6, 6]; }
278 def M4WriteNEONY : SchedWriteRes<[M4UnitFSQR,
279 M4UnitFSQR]> { let Latency = 12;
280 let ResourceCycles = [9, 9]; }
281 def M4WriteNEONZ : SchedWriteVariant<[SchedVar<ExynosQFormPred, [M4WriteNEONO]>,
282 SchedVar<NoSchedPred, [M4WriteNEONN]>]>;
284 def M4WriteFADD2 : SchedWriteRes<[M4UnitFADD]> { let Latency = 2; }
285 def M4WriteFADD2H : SchedWriteRes<[M4UnitFADDH]> { let Latency = 2; }
287 def M4WriteFCVT2 : SchedWriteRes<[M4UnitFCVT]> { let Latency = 2; }
288 def M4WriteFCVT2A : SchedWriteRes<[M4UnitFCVT0]> { let Latency = 2; }
289 def M4WriteFCVT2H : SchedWriteRes<[M4UnitFCVTH]> { let Latency = 2; }
290 def M4WriteFCVT3 : SchedWriteRes<[M4UnitFCVT]> { let Latency = 3; }
291 def M4WriteFCVT3A : SchedWriteRes<[M4UnitFCVT0]> { let Latency = 3; }
292 def M4WriteFCVT3H : SchedWriteRes<[M4UnitFCVTH]> { let Latency = 3; }
293 def M4WriteFCVT4 : SchedWriteRes<[M4UnitFCVT]> { let Latency = 4; }
294 def M4WriteFCVT4A : SchedWriteRes<[M4UnitFCVT0]> { let Latency = 4; }
295 def M4WriteFCVT6A : SchedWriteRes<[M4UnitFCVT0]> { let Latency = 6; }
297 def M4WriteFDIV7 : SchedWriteRes<[M4UnitFDIV]> { let Latency = 7;
298 let ResourceCycles = [6]; }
299 def M4WriteFDIV7H : SchedWriteRes<[M4UnitFDIVH]> { let Latency = 7;
300 let ResourceCycles = [6]; }
301 def M4WriteFDIV12 : SchedWriteRes<[M4UnitFDIV]> { let Latency = 12;
302 let ResourceCycles = [9]; }
304 def M4WriteFMAC2H : SchedWriteRes<[M4UnitFMACH]> { let Latency = 2; }
305 def M4WriteFMAC3H : SchedWriteRes<[M4UnitFMACH]> { let Latency = 3; }
306 def M4WriteFMAC3 : SchedWriteRes<[M4UnitFMAC]> { let Latency = 3; }
307 def M4WriteFMAC4 : SchedWriteRes<[M4UnitFMAC]> { let Latency = 4; }
308 def M4WriteFMAC4H : SchedWriteRes<[M4UnitFMACH]> { let Latency = 4; }
309 def M4WriteFMAC5 : SchedWriteRes<[M4UnitFMAC]> { let Latency = 5; }
311 def M4WriteFSQR7H : SchedWriteRes<[M4UnitFSQRH]> { let Latency = 7;
312 let ResourceCycles = [6]; }
313 def M4WriteFSQR8 : SchedWriteRes<[M4UnitFSQR]> { let Latency = 8;
314 let ResourceCycles = [7]; }
315 def M4WriteFSQR12 : SchedWriteRes<[M4UnitFSQR]> { let Latency = 12;
316 let ResourceCycles = [9]; }
318 def M4WriteNALU1 : SchedWriteRes<[M4UnitNALU]> { let Latency = 1; }
319 def M4WriteNALU1H : SchedWriteRes<[M4UnitNALUH]> { let Latency = 1; }
321 def M4WriteNCRY1 : SchedWriteRes<[M4UnitNCRY]> { let Latency = 1; }
322 def M4WriteNCRY1A : SchedWriteRes<[M4UnitNCRY0]> { let Latency = 1; }
323 def M4WriteNCRY3A : SchedWriteRes<[M4UnitNCRY0]> { let Latency = 3; }
324 def M4WriteNCRY5A : SchedWriteRes<[M4UnitNCRY]> { let Latency = 5; }
326 def M4WriteNHAD1 : SchedWriteRes<[M4UnitNHAD]> { let Latency = 1; }
327 def M4WriteNHAD3 : SchedWriteRes<[M4UnitNHAD]> { let Latency = 3; }
329 def M4WriteNMSC1 : SchedWriteRes<[M4UnitNMSC]> { let Latency = 1; }
330 def M4WriteNMSC2 : SchedWriteRes<[M4UnitNMSC]> { let Latency = 2; }
331 def M4WriteNMSC3 : SchedWriteRes<[M4UnitNMSC]> { let Latency = 3; }
333 def M4WriteNMUL3 : SchedWriteRes<[M4UnitNMUL]> { let Latency = 3; }
335 def M4WriteNSHF1 : SchedWriteRes<[M4UnitNSHF]> { let Latency = 1; }
336 def M4WriteNSHF1H : SchedWriteRes<[M4UnitNSHFH]> { let Latency = 1; }
337 def M4WriteNSHF3 : SchedWriteRes<[M4UnitNSHF]> { let Latency = 3; }
338 def M4WriteNSHFA : SchedWriteRes<[M4UnitNSHF]> { let Latency = 1;
339 let ResourceCycles = [2]; }
340 def M4WriteNSHFB : SchedWriteRes<[M4UnitNSHF]> { let Latency = 2;
342 let ResourceCycles = [2]; }
343 def M4WriteNSHFC : SchedWriteRes<[M4UnitNSHF]> { let Latency = 3;
345 let ResourceCycles = [4]; }
346 def M4WriteNSHFD : SchedWriteRes<[M4UnitNSHF]> { let Latency = 4;
348 let ResourceCycles = [4]; }
350 def M4WriteNSHT1 : SchedWriteRes<[M4UnitNSHT]> { let Latency = 1; }
351 def M4WriteNSHT2 : SchedWriteRes<[M4UnitNSHT]> { let Latency = 2; }
352 def M4WriteNSHT3 : SchedWriteRes<[M4UnitNSHT]> { let Latency = 3; }
353 def M4WriteNSHT4A : SchedWriteRes<[M4UnitNSHT1]> { let Latency = 4; }
355 def M4WriteVLDA : SchedWriteRes<[M4UnitL,
356 M4UnitL]> { let Latency = 5;
357 let NumMicroOps = 2; }
358 def M4WriteVLDB : SchedWriteRes<[M4UnitL,
360 M4UnitL]> { let Latency = 6;
361 let NumMicroOps = 3; }
362 def M4WriteVLDC : SchedWriteRes<[M4UnitL,
365 M4UnitL]> { let Latency = 6;
366 let NumMicroOps = 4; }
367 def M4WriteVLDD : SchedWriteRes<[M4UnitL,
368 M4UnitNSHF]> { let Latency = 6;
370 let ResourceCycles = [2, 1]; }
371 def M4WriteVLDF : SchedWriteRes<[M4UnitL,
372 M4UnitL]> { let Latency = 10;
374 let ResourceCycles = [3, 3]; }
375 def M4WriteVLDG : SchedWriteRes<[M4UnitL,
377 M4UnitNSHF]> { let Latency = 6;
379 let ResourceCycles = [2, 1, 1]; }
380 def M4WriteVLDI : SchedWriteRes<[M4UnitL,
382 M4UnitL]> { let Latency = 12;
384 let ResourceCycles = [3, 3, 3]; }
385 def M4WriteVLDJ : SchedWriteRes<[M4UnitL,
388 M4UnitNSHF]> { let Latency = 7;
390 let ResourceCycles = [3, 1, 1, 1]; }
391 def M4WriteVLDK : SchedWriteRes<[M4UnitL,
395 M4UnitNSHF]> { let Latency = 7;
397 let ResourceCycles = [3, 1, 1, 1, 1]; }
398 def M4WriteVLDL : SchedWriteRes<[M4UnitL,
402 M4UnitNSHF]> { let Latency = 7;
404 let ResourceCycles = [3, 1, 1, 6, 1]; }
405 def M4WriteVLDM : SchedWriteRes<[M4UnitL,
410 M4UnitNSHF]> { let Latency = 7;
412 let ResourceCycles = [3, 1, 1, 3, 1, 1]; }
413 def M4WriteVLDN : SchedWriteRes<[M4UnitL,
416 M4UnitL]> { let Latency = 14;
418 let ResourceCycles = [3, 3, 3, 3]; }
420 def M4WriteVST1 : SchedWriteRes<[M4UnitS,
421 M4UnitFST]> { let Latency = 1;
422 let NumMicroOps = 1; }
423 def M4WriteVSTA : WriteSequence<[WriteVST], 2>;
424 def M4WriteVSTB : WriteSequence<[WriteVST], 3>;
425 def M4WriteVSTC : WriteSequence<[WriteVST], 4>;
426 def M4WriteVSTD : SchedWriteRes<[M4UnitS,
427 M4UnitFST]> { let Latency = 2; }
428 def M4WriteVSTE : SchedWriteRes<[M4UnitS,
431 M4UnitFST]> { let Latency = 2;
432 let NumMicroOps = 2; }
433 def M4WriteVSTF : SchedWriteRes<[M4UnitNSHF,
437 M4UnitFST]> { let Latency = 4;
439 let ResourceCycles = [1, 2, 1, 2, 1]; }
440 def M4WriteVSTG : SchedWriteRes<[M4UnitNSHF,
448 M4UnitFST]> { let Latency = 5;
450 let ResourceCycles = [1, 1, 1, 2, 1, 2, 1, 2, 1]; }
451 def M4WriteVSTI : SchedWriteRes<[M4UnitNSHF,
462 M4UnitFST]> { let Latency = 8;
464 let ResourceCycles = [1, 1, 1, 1, 2, 1, 2, 1, 2, 1, 2, 1]; }
465 def M4WriteVSTJ : SchedWriteRes<[M4UnitA,
469 M4UnitFST]> { let Latency = 1;
470 let NumMicroOps = 2; }
471 def M4WriteVSTK : SchedWriteRes<[M4UnitA,
473 M4UnitFST]> { let Latency = 3;
474 let NumMicroOps = 2; }
475 def M4WriteVSTL : SchedWriteRes<[M4UnitNSHF,
480 M4UnitFST]> { let Latency = 4;
482 let ResourceCycles = [1, 1, 2, 1, 2, 1]; }
483 def M4WriteVSTY : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M4WriteVSTK]>,
484 SchedVar<NoSchedPred, [WriteVST]>]>;
487 def M4WriteCOPY : SchedWriteVariant<[SchedVar<ExynosFPPred, [M4WriteNALU1]>,
488 SchedVar<NoSchedPred, [M4WriteZ0]>]>;
489 def M4WriteMOVI : SchedWriteVariant<[SchedVar<IsZeroFPIdiomPred, [M4WriteZ0]>,
490 SchedVar<NoSchedPred, [M4WriteNALU1]>]>;
493 def M4ReadAESM1 : SchedReadAdvance<+1, [M4WriteNCRY1]>;
494 def M4ReadFMACM1 : SchedReadAdvance<+1, [M4WriteFMAC4,
497 def M4ReadNMULM1 : SchedReadAdvance<+1, [M4WriteNMUL3]>;
498 def M4ReadNMULP2 : SchedReadAdvance<-2, [M4WriteNMUL3]>;
501 //===----------------------------------------------------------------------===//
502 // Coarse scheduling model.
504 // Branch instructions.
505 def : SchedAlias<WriteBr, M4WriteZ0>;
506 def : SchedAlias<WriteBrReg, M4WriteC1>;
508 // Arithmetic and logical integer instructions.
509 def : SchedAlias<WriteI, M4WriteA1>;
510 def : SchedAlias<WriteIEReg, M4WriteAA>; // FIXME: M4WriteAX crashes TableGen.
511 def : SchedAlias<WriteISReg, M4WriteAA>; // FIXME: M4WriteAX crashes TableGen.
512 def : SchedAlias<WriteIS, M4WriteA1>;
514 // Move instructions.
515 def : SchedAlias<WriteImm, M4WriteA1>;
517 // Divide and multiply instructions.
518 def : SchedAlias<WriteID32, M4WriteD12>;
519 def : SchedAlias<WriteID64, M4WriteD21>;
520 def : SchedAlias<WriteIM32, M4WriteC3>;
521 def : SchedAlias<WriteIM64, M4WriteCA>;
523 // Miscellaneous instructions.
524 def : SchedAlias<WriteExtr, M4WriteAY>;
527 def : SchedAlias<WriteAdr, M4WriteZ1>;
528 def : SchedAlias<ReadAdrBase, M4ReadAdrBase>;
530 // Load instructions.
531 def : SchedAlias<WriteLD, M4WriteL4>;
532 def : SchedAlias<WriteLDHi, M4WriteZ4>;
533 def : SchedAlias<WriteLDIdx, M4WriteLX>;
535 // Store instructions.
536 def : SchedAlias<WriteST, M4WriteS1>;
537 def : SchedAlias<WriteSTP, M4WriteS1>;
538 def : SchedAlias<WriteSTX, M4WriteS1>;
539 def : SchedAlias<WriteSTIdx, M4WriteSX>;
541 // FP data instructions.
542 def : SchedAlias<WriteF, M4WriteFADD2>;
543 def : SchedAlias<WriteFCmp, M4WriteNMSC2>;
544 def : SchedAlias<WriteFDiv, M4WriteFDIV12>;
545 def : SchedAlias<WriteFMul, M4WriteFMAC3>;
547 // FP miscellaneous instructions.
548 def : SchedAlias<WriteFCvt, M4WriteFCVT2>;
549 def : SchedAlias<WriteFImm, M4WriteNALU1>;
550 def : SchedAlias<WriteFCopy, M4WriteNALU1>;
552 // FP load instructions.
553 def : SchedAlias<WriteVLD, M4WriteL5>;
555 // FP store instructions.
556 def : SchedAlias<WriteVST, M4WriteVST1>;
558 // ASIMD FP instructions.
559 def : SchedAlias<WriteV, M4WriteNALU1>;
561 // Other miscellaneous instructions.
562 def : WriteRes<WriteAtomic, []> { let Unsupported = 1; }
563 def : WriteRes<WriteBarrier, []> { let Latency = 1; }
564 def : WriteRes<WriteHint, []> { let Latency = 1; }
565 def : WriteRes<WriteSys, []> { let Latency = 1; }
567 //===----------------------------------------------------------------------===//
568 // Generic fast forwarding.
570 // TODO: Add FP register forwarding rules.
572 def : ReadAdvance<ReadI, 0>;
573 def : ReadAdvance<ReadISReg, 0>;
574 def : ReadAdvance<ReadIEReg, 0>;
575 def : ReadAdvance<ReadIM, 0>;
576 // TODO: The forwarding for 32 bits actually saves 2 cycles.
577 def : ReadAdvance<ReadIMA, 3, [WriteIM32, WriteIM64]>;
578 def : ReadAdvance<ReadID, 0>;
579 def : ReadAdvance<ReadExtrHi, 0>;
580 def : ReadAdvance<ReadAdrBase, 0>;
581 def : ReadAdvance<ReadVLD, 0>;
583 //===----------------------------------------------------------------------===//
584 // Finer scheduling model.
586 // Branch instructions
587 def : InstRW<[M4WriteB1], (instrs Bcc)>;
588 def : InstRW<[M4WriteAF], (instrs BL)>;
589 def : InstRW<[M4WriteBX], (instrs BLR)>;
590 def : InstRW<[M4WriteC1], (instregex "^CBN?Z[WX]")>;
591 def : InstRW<[M4WriteAD], (instregex "^TBN?Z[WX]")>;
593 // Arithmetic and logical integer instructions.
594 def : InstRW<[M4WriteAX], (instregex "^(ADD|AND|BIC|EON|EOR|ORN|SUB)[WX]rs$")>;
595 def : InstRW<[M4WriteAU], (instrs ORRWrs, ORRXrs)>;
596 def : InstRW<[M4WriteAX], (instregex "^(ADD|AND|BIC|SUB)S[WX]rs$")>;
597 def : InstRW<[M4WriteAX], (instregex "^(ADD|SUB)S?[WX]rx(64)?$")>;
598 def : InstRW<[M4WriteAV], (instrs ADDWri, ADDXri, ORRWri, ORRXri)>;
600 // Move instructions.
601 def : InstRW<[M4WriteCOPY], (instrs COPY)>;
602 def : InstRW<[M4WriteZ0], (instrs ADR, ADRP)>;
603 def : InstRW<[M4WriteZ0], (instregex "^MOV[NZ][WX]i")>;
605 // Divide and multiply instructions.
607 // Miscellaneous instructions.
609 // Load instructions.
610 def : InstRW<[M4WriteLD,
612 WriteAdr], (instregex "^LDP(SW|W|X)(post|pre)")>;
613 def : InstRW<[M4WriteL5,
614 ReadAdrBase], (instregex "^LDR(BB|SBW|SBX|HH|SHW|SHX|SW|W|X)roW")>;
615 def : InstRW<[WriteLDIdx,
616 ReadAdrBase], (instregex "^LDR(BB|SBW|SBX|HH|SHW|SHX|SW|W|X)roX")>;
617 def : InstRW<[M4WriteL5,
618 ReadAdrBase], (instrs PRFMroW)>;
619 def : InstRW<[WriteLDIdx,
620 ReadAdrBase], (instrs PRFMroX)>;
622 // Store instructions.
623 def : InstRW<[M4WriteSB,
624 ReadAdrBase], (instregex "^STR(BB|HH|W|X)roW")>;
625 def : InstRW<[WriteST,
626 ReadAdrBase], (instregex "^STR(BB|HH|W|X)roX")>;
628 // FP data instructions.
629 def : InstRW<[M4WriteNSHF1H], (instrs FABSHr)>;
630 def : InstRW<[M4WriteNSHF1], (instregex "^FABS[SD]r")>;
631 def : InstRW<[M4WriteFADD2H], (instregex "^F(ADD|SUB)Hrr")>;
632 def : InstRW<[M4WriteFADD2], (instregex "^F(ADD|SUB)[SD]rr")>;
633 def : InstRW<[M4WriteFADD2H], (instregex "^FADDPv.i16")>;
634 def : InstRW<[M4WriteFADD2], (instregex "^FADDPv.i(32|64)")>;
635 def : InstRW<[M4WriteNEONQ], (instregex "^FCCMPE?[HSD]rr")>;
636 def : InstRW<[M4WriteNMSC2], (instregex "^FCMPE?[HSD]r[ir]")>;
637 def : InstRW<[M4WriteNMSC1], (instregex "^F(AC|CM)(EQ|GE|GT|LE|LT)(16|32|64|v1)")>;
638 def : InstRW<[M4WriteFDIV7H], (instrs FDIVHrr)>;
639 def : InstRW<[M4WriteFDIV7], (instrs FDIVSrr)>;
640 def : InstRW<[M4WriteFDIV12], (instrs FDIVDrr)>;
641 def : InstRW<[M4WriteNMSC1], (instregex "^F(MAX|MIN)(NM)?[HSD]rr")>;
642 def : InstRW<[M4WriteFMAC3H], (instregex "^FN?MULHrr")>;
643 def : InstRW<[M4WriteFMAC3], (instregex "^FN?MUL[SD]rr")>;
644 def : InstRW<[M4WriteFMAC3H], (instrs FMULX16)>;
645 def : InstRW<[M4WriteFMAC3], (instregex "^FMULX(32|64)")>;
646 def : InstRW<[M4WriteFMAC4H,
647 M4ReadFMACM1], (instregex "^FN?M(ADD|SUB)Hrrr")>;
648 def : InstRW<[M4WriteFMAC4,
649 M4ReadFMACM1], (instregex "^FN?M(ADD|SUB)[SD]rrr")>;
650 def : InstRW<[M4WriteNALU1H], (instrs FNEGHr)>;
651 def : InstRW<[M4WriteNALU1], (instregex "^FNEG[SD]r")>;
652 def : InstRW<[M4WriteFCVT3A], (instregex "^FRINT.+r")>;
653 def : InstRW<[M4WriteNEONH], (instregex "^FCSEL[HSD]rrr")>;
654 def : InstRW<[M4WriteFSQR7H], (instrs FSQRTHr)>;
655 def : InstRW<[M4WriteFSQR8], (instrs FSQRTSr)>;
656 def : InstRW<[M4WriteFSQR12], (instrs FSQRTDr)>;
658 // FP miscellaneous instructions.
659 def : InstRW<[M4WriteFCVT2H], (instregex "^FCVTH[SD]r")>;
660 def : InstRW<[M4WriteFCVT2H], (instregex "^FCVT[SD]Hr")>;
661 def : InstRW<[M4WriteFCVT2], (instregex "^FCVT[SD][SD]r")>;
662 def : InstRW<[M4WriteFCVT6A], (instregex "^[SU]CVTF[SU][XW][HSD]ri")>;
663 def : InstRW<[M4WriteNEONR], (instregex "^FCVT[AMNPZ][SU][SU][XW][HSD]r")>;
664 def : InstRW<[M4WriteNALU1], (instregex "^FMOV[HSD][ir]")>;
665 def : InstRW<[M4WriteSA], (instregex "^FMOV[WX][HSD]r")>;
666 def : InstRW<[M4WriteNEONJ], (instregex "^FMOV[HSD][WX]r")>;
667 def : InstRW<[M4WriteNEONI], (instregex "^FMOVXDHighr")>;
668 def : InstRW<[M4WriteNEONK], (instregex "^FMOVDXHighr")>;
669 def : InstRW<[M4WriteFCVT3H], (instregex "^F(RECP|RSQRT)Ev1f16")>;
670 def : InstRW<[M4WriteFCVT3], (instregex "^F(RECP|RSQRT)Ev1i(32|64)")>;
671 def : InstRW<[M4WriteNMSC1], (instregex "^FRECPXv1")>;
672 def : InstRW<[M4WriteFMAC4H], (instregex "^F(RECP|RSQRT)S16")>;
673 def : InstRW<[M4WriteFMAC4], (instregex "^F(RECP|RSQRT)S(32|64)")>;
675 // FP load instructions.
676 def : InstRW<[WriteVLD], (instregex "^LDR[SDQ]l")>;
677 def : InstRW<[WriteVLD], (instregex "^LDUR[BHSDQ]i")>;
678 def : InstRW<[WriteVLD,
679 WriteAdr], (instregex "^LDR[BHSDQ](post|pre)")>;
680 def : InstRW<[WriteVLD], (instregex "^LDR[BHSDQ]ui")>;
681 def : InstRW<[M4WriteLE,
682 ReadAdrBase], (instregex "^LDR[BHSDQ]roW")>;
683 def : InstRW<[WriteVLD,
684 ReadAdrBase], (instregex "^LDR[BHSD]roX")>;
685 def : InstRW<[M4WriteLY,
686 ReadAdrBase], (instrs LDRQroX)>;
687 def : InstRW<[WriteVLD,
688 M4WriteLH], (instregex "^LDN?P[SD]i")>;
689 def : InstRW<[M4WriteLA,
690 M4WriteLH], (instregex "^LDN?PQi")>;
691 def : InstRW<[M4WriteL5,
693 WriteAdr], (instregex "^LDP[SD]post")>;
694 def : InstRW<[M4WriteLB,
696 WriteAdr], (instrs LDPQpost)>;
697 def : InstRW<[M4WriteLB,
699 WriteAdr], (instregex "^LDP[SD]pre")>;
700 def : InstRW<[M4WriteLC,
702 WriteAdr], (instrs LDPQpre)>;
704 // FP store instructions.
705 def : InstRW<[WriteVST], (instregex "^STUR[BHSDQ]i")>;
706 def : InstRW<[WriteVST,
707 WriteAdr], (instregex "^STR[BHSDQ](post|pre)")>;
708 def : InstRW<[WriteVST], (instregex "^STR[BHSDQ]ui")>;
709 def : InstRW<[M4WriteVSTK,
710 ReadAdrBase], (instregex "^STR[BHSD]roW")>;
711 def : InstRW<[M4WriteVSTK,
712 ReadAdrBase], (instrs STRQroW)>;
713 def : InstRW<[WriteVST,
714 ReadAdrBase], (instregex "^STR[BHSD]roX")>;
715 def : InstRW<[M4WriteVSTY,
716 ReadAdrBase], (instrs STRQroX)>;
717 def : InstRW<[WriteVST], (instregex "^STN?P[SD]i")>;
718 def : InstRW<[M4WriteVSTJ], (instregex "^STN?PQi")>;
719 def : InstRW<[WriteVST,
720 WriteAdr], (instregex "^STP[SD](post|pre)")>;
721 def : InstRW<[M4WriteVSTJ,
722 WriteAdr], (instregex "^STPQ(post|pre)")>;
724 // ASIMD instructions.
725 def : InstRW<[M4WriteNHAD1], (instregex "^[SU]ABDL?v")>;
726 def : InstRW<[M4WriteNHAD3], (instregex "^[SU]ABAL?v")>;
727 def : InstRW<[M4WriteNMSC1], (instregex "^ABSv")>;
728 def : InstRW<[M4WriteNALU1], (instregex "^(ADD|NEG|SUB)v")>;
729 def : InstRW<[M4WriteNHAD3], (instregex "^[SU]?ADDL?Pv")>;
730 def : InstRW<[M4WriteNHAD3], (instregex "^[SU]H(ADD|SUB)v")>;
731 def : InstRW<[M4WriteNHAD3], (instregex "^[SU](ADD|SUB)[LW]v")>;
732 def : InstRW<[M4WriteNHAD3], (instregex "^R?(ADD|SUB)HN2?v")>;
733 def : InstRW<[M4WriteNHAD3], (instregex "^[SU]Q(ADD|SUB)v")>;
734 def : InstRW<[M4WriteNHAD3], (instregex "^(SU|US)QADDv")>;
735 def : InstRW<[M4WriteNHAD3], (instregex "^[SU]RHADDv")>;
736 def : InstRW<[M4WriteNMSC1], (instregex "^SQ(ABS|NEG)v")>;
737 def : InstRW<[M4WriteNHAD3], (instregex "^[SU]?ADDL?Vv")>;
738 def : InstRW<[M4WriteNMSC1], (instregex "^CM(EQ|GE|GT|HI|HS|LE|LT)v")>;
739 def : InstRW<[M4WriteNALU1], (instregex "^CMTSTv")>;
740 def : InstRW<[M4WriteNALU1], (instregex "^(AND|BIC|EOR|NOT|ORN|ORR)v")>;
741 def : InstRW<[M4WriteNMSC1], (instregex "^[SU](MIN|MAX)v")>;
742 def : InstRW<[M4WriteNMSC2], (instregex "^[SU](MIN|MAX)Pv")>;
743 def : InstRW<[M4WriteNHAD3], (instregex "^[SU](MIN|MAX)Vv")>;
744 def : InstRW<[M4WriteNMUL3,
745 M4ReadNMULM1], (instregex "^ML[AS]v")>;
746 def : InstRW<[M4WriteNMUL3,
747 M4ReadNMULM1], (instregex "^(SQR?D)?MULH?v")>;
748 def : InstRW<[M4WriteNMUL3,
749 M4ReadNMULM1], (instregex "^SQRDML[AS]H")>;
750 def : InstRW<[M4WriteNMUL3,
751 M4ReadNMULM1], (instregex "^(S|U|SQD)ML[AS]L(v1(i32|i64)|v2i32|v4i16|v8i8)")>;
752 def : InstRW<[M4WriteNMUL3,
753 M4ReadNMULP2], (instregex "^(S|U|SQD)ML[AS]L(v4i32|v8i16|v16i8)")>;
754 def : InstRW<[M4WriteNMUL3,
755 M4ReadNMULM1], (instregex "^(S|U|SQD)MULL(v1(i32|i64)|v2i32|v4i16|v8i8)")>;
756 def : InstRW<[M4WriteNMUL3,
757 M4ReadNMULP2], (instregex "^(S|U|SQD)MULL(v4i32|v8i16|v16i8)")>;
758 def : InstRW<[M4WriteNMUL3], (instregex "^[SU]DOT(lane)?v")>;
759 def : InstRW<[M4WriteNHAD3], (instregex "^[SU]ADALPv")>;
760 def : InstRW<[M4WriteNSHT4A], (instregex "^[SU]R?SRA[dv]")>;
761 def : InstRW<[M4WriteNSHT1], (instregex "^SHL[dv]")>;
762 def : InstRW<[M4WriteNSHT1], (instregex "^S[LR]I[dv]")>;
763 def : InstRW<[M4WriteNSHT1], (instregex "^[SU]SH[LR][dv]")>;
764 def : InstRW<[M4WriteNSHT2], (instregex "^[SU]?SHLLv")>;
765 def : InstRW<[M4WriteNSHT4A], (instregex "^[SU]?Q?R?SHRU?N[bhsv]")>;
766 def : InstRW<[M4WriteNSHT4A], (instregex "^[SU]RSH[LR][dv]")>;
767 def : InstRW<[M4WriteNSHT4A], (instregex "^[SU]QR?SHLU?[bhsdv]")>;
769 // ASIMD FP instructions.
770 def : InstRW<[M4WriteNSHF1H], (instregex "^FABSv.f16")>;
771 def : InstRW<[M4WriteNSHF1], (instregex "^FABSv.f(32|64)")>;
772 def : InstRW<[M4WriteFADD2H], (instregex "^F(ABD|ADD|SUB)v.f16")>;
773 def : InstRW<[M4WriteFADD2], (instregex "^F(ABD|ADD|SUB)v.f(32|64)")>;
774 def : InstRW<[M4WriteFADD2H], (instregex "^FADDPv.f16")>;
775 def : InstRW<[M4WriteFADD2], (instregex "^FADDPv.f(32|64)")>;
776 def : InstRW<[M4WriteNMSC1], (instregex "^F(AC|CM)(EQ|GE|GT|LE|LT)v[^1]")>;
777 def : InstRW<[M4WriteFCVT2], (instregex "^FCVT(L|N|XN)v")>;
778 def : InstRW<[M4WriteFCVT2A], (instregex "^FCVT[AMNPZ][SU]v")>;
779 def : InstRW<[M4WriteFCVT2H], (instregex "^[SU]CVTFv.[fi]16")>;
780 def : InstRW<[M4WriteFCVT2], (instregex "^[SU]CVTFv.[fi](32|64)")>;
781 def : InstRW<[M4WriteFDIV7H], (instrs FDIVv4f16)>;
782 def : InstRW<[M4WriteNEONVH], (instrs FDIVv8f16)>;
783 def : InstRW<[M4WriteFDIV7], (instrs FDIVv2f32)>;
784 def : InstRW<[M4WriteNEONV], (instrs FDIVv4f32)>;
785 def : InstRW<[M4WriteNEONW], (instrs FDIVv2f64)>;
786 def : InstRW<[M4WriteNMSC1], (instregex "^F(MAX|MIN)(NM)?v")>;
787 def : InstRW<[M4WriteNMSC2], (instregex "^F(MAX|MIN)(NM)?Pv")>;
788 def : InstRW<[M4WriteNEONZ], (instregex "^F(MAX|MIN)(NM)?Vv")>;
789 def : InstRW<[M4WriteFMAC2H], (instregex "^FMULX?v.[fi]16")>;
790 def : InstRW<[M4WriteFMAC3], (instregex "^FMULX?v.[fi](32|64)")>;
791 def : InstRW<[M4WriteFMAC4H,
792 M4ReadFMACM1], (instregex "^FML[AS]v.[fi]16")>;
793 def : InstRW<[M4WriteFMAC4,
794 M4ReadFMACM1], (instregex "^FML[AS]v.[fi](32|64)")>;
795 def : InstRW<[M4WriteNALU1H], (instregex "^FNEGv.f16")>;
796 def : InstRW<[M4WriteNALU1], (instregex "^FNEGv.f(32|64)")>;
797 def : InstRW<[M4WriteFCVT3A], (instregex "^FRINT[AIMNPXZ]v")>;
798 def : InstRW<[M4WriteFSQR7H], (instrs FSQRTv4f16)>;
799 def : InstRW<[M4WriteNEONXH], (instrs FSQRTv8f16)>;
800 def : InstRW<[M4WriteFSQR8], (instrs FSQRTv2f32)>;
801 def : InstRW<[M4WriteNEONX], (instrs FSQRTv4f32)>;
802 def : InstRW<[M4WriteNEONY], (instrs FSQRTv2f64)>;
804 // ASIMD miscellaneous instructions.
805 def : InstRW<[M4WriteNALU1], (instregex "^RBITv")>;
806 def : InstRW<[M4WriteNALU1], (instregex "^(BIF|BIT|BSL)v")>;
807 def : InstRW<[M4WriteNALU1], (instregex "^CL[STZ]v")>;
808 def : InstRW<[M4WriteNEONB], (instregex "^DUPv.+gpr")>;
809 def : InstRW<[M4WriteNSHF1], (instregex "^CPY")>;
810 def : InstRW<[M4WriteNSHF1], (instregex "^DUPv.+lane")>;
811 def : InstRW<[M4WriteNSHF1], (instregex "^EXTv")>;
812 def : InstRW<[M4WriteNSHT4A], (instregex "^XTNv")>;
813 def : InstRW<[M4WriteNSHT4A], (instregex "^[SU]?QXTU?Nv")>;
814 def : InstRW<[M4WriteNEONB], (instregex "^INSv.+gpr")>;
815 def : InstRW<[M4WriteNSHF1], (instregex "^INSv.+lane")>;
816 def : InstRW<[M4WriteMOVI], (instregex "^(MOV|MVN)I")>;
817 def : InstRW<[M4WriteNALU1H], (instregex "^FMOVv.f16")>;
818 def : InstRW<[M4WriteNALU1], (instregex "^FMOVv.f(32|64)")>;
819 def : InstRW<[M4WriteFCVT3H], (instregex "^F(RECP|RSQRT)Ev[248]f16")>;
820 def : InstRW<[M4WriteFCVT3], (instregex "^F(RECP|RSQRT)Ev[248]f(32|64)")>;
821 def : InstRW<[M4WriteFCVT3], (instregex "^U(RECP|RSQRT)Ev[24]i32")>;
822 def : InstRW<[M4WriteFMAC4H], (instregex "^F(RECP|RSQRT)Sv.f16")>;
823 def : InstRW<[M4WriteFMAC4], (instregex "^F(RECP|RSQRT)Sv.f(32|64)")>;
824 def : InstRW<[M4WriteNSHF1], (instregex "^REV(16|32|64)v")>;
825 def : InstRW<[M4WriteNSHFA], (instregex "^TB[LX]v(8|16)i8One")>;
826 def : InstRW<[M4WriteNSHFB], (instregex "^TB[LX]v(8|16)i8Two")>;
827 def : InstRW<[M4WriteNSHFC], (instregex "^TB[LX]v(8|16)i8Three")>;
828 def : InstRW<[M4WriteNSHFD], (instregex "^TB[LX]v(8|16)i8Four")>;
829 def : InstRW<[M4WriteNEONP], (instregex "^[SU]MOVv")>;
830 def : InstRW<[M4WriteNSHF1], (instregex "^(TRN|UZP|ZIP)[12]v")>;
832 // ASIMD load instructions.
833 def : InstRW<[WriteVLD], (instregex "LD1Onev(8b|4h|2s|1d)$")>;
834 def : InstRW<[WriteVLD,
835 M4WriteA1], (instregex "LD1Onev(8b|4h|2s|1d)_POST$")>;
836 def : InstRW<[WriteVLD], (instregex "LD1Onev(16b|8h|4s|2d)$")>;
837 def : InstRW<[WriteVLD,
838 M4WriteA1], (instregex "LD1Onev(16b|8h|4s|2d)_POST$")>;
840 def : InstRW<[M4WriteVLDA], (instregex "LD1Twov(8b|4h|2s|1d)$")>;
841 def : InstRW<[M4WriteVLDA,
842 M4WriteA1], (instregex "LD1Twov(8b|4h|2s|1d)_POST$")>;
843 def : InstRW<[M4WriteVLDA], (instregex "LD1Twov(16b|8h|4s|2d)$")>;
844 def : InstRW<[M4WriteVLDA,
845 M4WriteA1], (instregex "LD1Twov(16b|8h|4s|2d)_POST$")>;
847 def : InstRW<[M4WriteVLDB], (instregex "LD1Threev(8b|4h|2s|1d)$")>;
848 def : InstRW<[M4WriteVLDB,
849 M4WriteA1], (instregex "LD1Threev(8b|4h|2s|1d)_POST$")>;
850 def : InstRW<[M4WriteVLDB], (instregex "LD1Threev(16b|8h|4s|2d)$")>;
851 def : InstRW<[M4WriteVLDB,
852 M4WriteA1], (instregex "LD1Threev(16b|8h|4s|2d)_POST$")>;
854 def : InstRW<[M4WriteVLDC], (instregex "LD1Fourv(8b|4h|2s|1d)$")>;
855 def : InstRW<[M4WriteVLDC,
856 M4WriteA1], (instregex "LD1Fourv(8b|4h|2s|1d)_POST$")>;
857 def : InstRW<[M4WriteVLDC], (instregex "LD1Fourv(16b|8h|4s|2d)$")>;
858 def : InstRW<[M4WriteVLDC,
859 M4WriteA1], (instregex "LD1Fourv(16b|8h|4s|2d)_POST$")>;
861 def : InstRW<[M4WriteVLDD], (instregex "LD1i(8|16|32|64)$")>;
862 def : InstRW<[M4WriteVLDD,
863 M4WriteA1], (instregex "LD1i(8|16|32|64)_POST$")>;
865 def : InstRW<[WriteVLD], (instregex "LD1Rv(8b|4h|2s|1d)$")>;
866 def : InstRW<[WriteVLD,
867 M4WriteA1], (instregex "LD1Rv(8b|4h|2s|1d)_POST$")>;
868 def : InstRW<[WriteVLD], (instregex "LD1Rv(16b|8h|4s|2d)$")>;
869 def : InstRW<[WriteVLD,
870 M4WriteA1], (instregex "LD1Rv(16b|8h|4s|2d)_POST$")>;
872 def : InstRW<[M4WriteVLDF], (instregex "LD2Twov(8b|4h|2s)$")>;
873 def : InstRW<[M4WriteVLDF,
874 M4WriteA1], (instregex "LD2Twov(8b|4h|2s)_POST$")>;
875 def : InstRW<[M4WriteVLDF], (instregex "LD2Twov(16b|8h|4s|2d)$")>;
876 def : InstRW<[M4WriteVLDF,
877 M4WriteA1], (instregex "LD2Twov(16b|8h|4s|2d)_POST$")>;
879 def : InstRW<[M4WriteVLDG], (instregex "LD2i(8|16|32|64)$")>;
880 def : InstRW<[M4WriteVLDG,
881 M4WriteA1], (instregex "LD2i(8|16|32|64)_POST$")>;
883 def : InstRW<[M4WriteVLDA], (instregex "LD2Rv(8b|4h|2s|1d)$")>;
884 def : InstRW<[M4WriteVLDA,
885 M4WriteA1], (instregex "LD2Rv(8b|4h|2s|1d)_POST$")>;
886 def : InstRW<[M4WriteVLDA], (instregex "LD2Rv(16b|8h|4s|2d)$")>;
887 def : InstRW<[M4WriteVLDA,
888 M4WriteA1], (instregex "LD2Rv(16b|8h|4s|2d)_POST$")>;
890 def : InstRW<[M4WriteVLDI], (instregex "LD3Threev(8b|4h|2s)$")>;
891 def : InstRW<[M4WriteVLDI,
892 M4WriteA1], (instregex "LD3Threev(8b|4h|2s)_POST$")>;
893 def : InstRW<[M4WriteVLDI], (instregex "LD3Threev(16b|8h|4s|2d)$")>;
894 def : InstRW<[M4WriteVLDI,
895 M4WriteA1], (instregex "LD3Threev(16b|8h|4s|2d)_POST$")>;
897 def : InstRW<[M4WriteVLDJ], (instregex "LD3i(8|16|32)$")>;
898 def : InstRW<[M4WriteVLDJ,
899 M4WriteA1], (instregex "LD3i(8|16|32)_POST$")>;
900 def : InstRW<[M4WriteVLDL], (instregex "LD3i64$")>;
901 def : InstRW<[M4WriteVLDL,
902 M4WriteA1], (instregex "LD3i64_POST$")>;
904 def : InstRW<[M4WriteVLDB], (instregex "LD3Rv(8b|4h|2s|1d)$")>;
905 def : InstRW<[M4WriteVLDB,
906 M4WriteA1], (instregex "LD3Rv(8b|4h|2s|1d)_POST$")>;
907 def : InstRW<[M4WriteVLDB], (instregex "LD3Rv(16b|8h|4s|2d)$")>;
908 def : InstRW<[M4WriteVLDB,
909 M4WriteA1], (instregex "LD3Rv(16b|8h|4s|2d)_POST$")>;
911 def : InstRW<[M4WriteVLDN], (instregex "LD4Fourv(8b|4h|2s)$")>;
912 def : InstRW<[M4WriteVLDN,
913 M4WriteA1], (instregex "LD4Fourv(8b|4h|2s)_POST$")>;
914 def : InstRW<[M4WriteVLDN], (instregex "LD4Fourv(16b|8h|4s|2d)$")>;
915 def : InstRW<[M4WriteVLDN,
916 M4WriteA1], (instregex "LD4Fourv(16b|8h|4s|2d)_POST$")>;
918 def : InstRW<[M4WriteVLDK], (instregex "LD4i(8|16|32)$")>;
919 def : InstRW<[M4WriteVLDK,
920 M4WriteA1], (instregex "LD4i(8|16|32)_POST$")>;
921 def : InstRW<[M4WriteVLDM], (instregex "LD4i64$")>;
922 def : InstRW<[M4WriteVLDM,
923 M4WriteA1], (instregex "LD4i64_POST$")>;
925 def : InstRW<[M4WriteVLDC], (instregex "LD4Rv(8b|4h|2s|1d)$")>;
926 def : InstRW<[M4WriteVLDC,
927 M4WriteA1], (instregex "LD4Rv(8b|4h|2s|1d)_POST$")>;
928 def : InstRW<[M4WriteVLDC], (instregex "LD4Rv(16b|8h|4s|2d)$")>;
929 def : InstRW<[M4WriteVLDC,
930 M4WriteA1], (instregex "LD4Rv(16b|8h|4s|2d)_POST$")>;
932 // ASIMD store instructions.
933 def : InstRW<[WriteVST], (instregex "ST1Onev(8b|4h|2s|1d)$")>;
934 def : InstRW<[WriteVST,
935 M4WriteA1], (instregex "ST1Onev(8b|4h|2s|1d)_POST$")>;
936 def : InstRW<[WriteVST], (instregex "ST1Onev(16b|8h|4s|2d)$")>;
937 def : InstRW<[WriteVST,
938 M4WriteA1], (instregex "ST1Onev(16b|8h|4s|2d)_POST$")>;
940 def : InstRW<[M4WriteVSTA], (instregex "ST1Twov(8b|4h|2s|1d)$")>;
941 def : InstRW<[M4WriteVSTA,
942 M4WriteA1], (instregex "ST1Twov(8b|4h|2s|1d)_POST$")>;
943 def : InstRW<[M4WriteVSTA], (instregex "ST1Twov(16b|8h|4s|2d)$")>;
944 def : InstRW<[M4WriteVSTA,
945 M4WriteA1], (instregex "ST1Twov(16b|8h|4s|2d)_POST$")>;
947 def : InstRW<[M4WriteVSTB], (instregex "ST1Threev(8b|4h|2s|1d)$")>;
948 def : InstRW<[M4WriteVSTB,
949 M4WriteA1], (instregex "ST1Threev(8b|4h|2s|1d)_POST$")>;
950 def : InstRW<[M4WriteVSTB], (instregex "ST1Threev(16b|8h|4s|2d)$")>;
951 def : InstRW<[M4WriteVSTB,
952 M4WriteA1], (instregex "ST1Threev(16b|8h|4s|2d)_POST$")>;
954 def : InstRW<[M4WriteVSTC], (instregex "ST1Fourv(8b|4h|2s|1d)$")>;
955 def : InstRW<[M4WriteVSTC,
956 M4WriteA1], (instregex "ST1Fourv(8b|4h|2s|1d)_POST$")>;
957 def : InstRW<[M4WriteVSTC], (instregex "ST1Fourv(16b|8h|4s|2d)$")>;
958 def : InstRW<[M4WriteVSTC,
959 M4WriteA1], (instregex "ST1Fourv(16b|8h|4s|2d)_POST$")>;
961 def : InstRW<[WriteVST], (instregex "ST1i(8|16|32|64)$")>;
962 def : InstRW<[WriteVST,
963 M4WriteA1], (instregex "ST1i(8|16|32|64)_POST$")>;
965 def : InstRW<[M4WriteVSTD], (instregex "ST2Twov(8b|4h|2s)$")>;
966 def : InstRW<[M4WriteVSTD,
967 M4WriteA1], (instregex "ST2Twov(8b|4h|2s)_POST$")>;
968 def : InstRW<[M4WriteVSTE], (instregex "ST2Twov(16b|8h|4s|2d)$")>;
969 def : InstRW<[M4WriteVSTE,
970 M4WriteA1], (instregex "ST2Twov(16b|8h|4s|2d)_POST$")>;
972 def : InstRW<[M4WriteVSTD], (instregex "ST2i(8|16|32|64)$")>;
973 def : InstRW<[M4WriteVSTD,
974 M4WriteA1], (instregex "ST2i(8|16|32|64)_POST$")>;
976 def : InstRW<[M4WriteVSTF], (instregex "ST3Threev(8b|4h|2s)$")>;
977 def : InstRW<[M4WriteVSTF,
978 M4WriteA1], (instregex "ST3Threev(8b|4h|2s)_POST$")>;
979 def : InstRW<[M4WriteVSTG], (instregex "ST3Threev(16b|8h|4s|2d)$")>;
980 def : InstRW<[M4WriteVSTG,
981 M4WriteA1], (instregex "ST3Threev(16b|8h|4s|2d)_POST$")>;
983 def : InstRW<[M4WriteVSTE], (instregex "ST3i(8|16|32|64)$")>;
984 def : InstRW<[M4WriteVSTE,
985 M4WriteA1], (instregex "ST3i(8|16|32|64)_POST$")>;
987 def : InstRW<[M4WriteVSTL], (instregex "ST4Fourv(8b|4h|2s)$")>;
988 def : InstRW<[M4WriteVSTL,
989 M4WriteA1], (instregex "ST4Fourv(8b|4h|2s)_POST$")>;
990 def : InstRW<[M4WriteVSTI], (instregex "ST4Fourv(16b|8h|4s|2d)$")>;
991 def : InstRW<[M4WriteVSTI,
992 M4WriteA1], (instregex "ST4Fourv(16b|8h|4s|2d)_POST$")>;
994 def : InstRW<[M4WriteVSTE], (instregex "ST4i(8|16|32|64)$")>;
995 def : InstRW<[M4WriteVSTE,
996 M4WriteA1], (instregex "ST4i(8|16|32|64)_POST$")>;
998 // Cryptography instructions.
999 def : InstRW<[M4WriteNCRY1], (instregex "^AES[DE]")>;
1000 def : InstRW<[M4WriteNCRY1,
1001 M4ReadAESM1], (instregex "^AESI?MC")>;
1002 def : InstRW<[M4WriteNCRY1A], (instregex "^PMULv")>;
1003 def : InstRW<[M4WriteNCRY1A], (instregex "^PMULLv(1|8)i")>;
1004 def : InstRW<[M4WriteNCRY3A], (instregex "^PMULLv(2|16)i")>;
1005 def : InstRW<[M4WriteNCRY1A], (instregex "^SHA1([CHMP]|SU[01])")>;
1006 def : InstRW<[M4WriteNCRY1A], (instrs SHA256SU0rr)>;
1007 def : InstRW<[M4WriteNCRY5A], (instrs SHA256SU1rrr)>;
1008 def : InstRW<[M4WriteNCRY5A], (instrs SHA256H2rrr)>;
1010 // CRC instructions.
1011 def : InstRW<[M4WriteE2], (instregex "^CRC32C?[BHWX]rr$")>;
1013 } // SchedModel = ExynosM4Model