xref: /freebsd/contrib/llvm-project/llvm/lib/Target/AArch64/AArch64SchedA510.td (revision d37eb51047221dc3322b34db1038ff3aa533883f)
1//==- AArch64SchedCortexA510.td - ARM Cortex-A510 Scheduling Definitions -*- 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// This file defines the machine model for the ARM Cortex-A510 processor.
10//
11//===----------------------------------------------------------------------===//
12
13// ===---------------------------------------------------------------------===//
14// The following definitions describe the per-operand machine model.
15// This works with MachineScheduler. See MCSchedModel.h for details.
16
17// Cortex-A510 machine model for scheduling and other instruction cost heuristics.
18def CortexA510Model : SchedMachineModel {
19  let MicroOpBufferSize = 0;  // The Cortex-A510 is an in-order processor
20  let IssueWidth = 3;         // It dual-issues under most circumstances
21  let LoadLatency = 3;        // Cycles for loads to access the cache.
22                              // Most loads have a latency of 2, but some have higher latencies.
23                              // 3 seems to be a good tradeoff
24  let PostRAScheduler = 1;    // Enable PostRA scheduler pass.
25  let CompleteModel = 0;      // Covers instructions applicable to Cortex-A510.
26
27  // FIXME: Remove when all errors have been fixed.
28  let FullInstRWOverlapCheck = 0;
29}
30
31
32//===----------------------------------------------------------------------===//
33// Subtarget-specific SchedWrite types
34
35let SchedModel = CortexA510Model in {
36
37//===----------------------------------------------------------------------===//
38// Define each kind of processor resource and number available.
39
40// Modeling each pipeline as a ProcResource using the BufferSize = 0 since the
41// Cortex-A510 is in-order.
42let BufferSize = 0 in {
43  def CortexA510UnitALU0   : ProcResource<1>;    // Int ALU0
44  def CortexA510UnitALU12  : ProcResource<2>;    // Int ALU1 & ALU2
45  def CortexA510UnitMAC    : ProcResource<1>;    // Int MAC, 64-bi wide
46  def CortexA510UnitDiv    : ProcResource<1>;    // Int Division, not pipelined
47  // There are 2 LS pipes, 1 for Load/Store; 1 for Store only
48  def CortexA510UnitLdSt   : ProcResource<1>;    // Load/Store shared pipe
49  def CortexA510UnitLd1    : ProcResource<1>;    // Load pipe
50  def CortexA510UnitB      : ProcResource<1>;    // Branch
51  def CortexA510UnitPAC    : ProcResource<1>;    // Pointer Authentication (PAC) pipe
52
53  // The FP DIV/SQRT instructions execute totally differently from the FP ALU
54  // instructions, which can mostly be dual-issued; that's why for now we model
55  // them with 2 resources.
56  def CortexA510UnitVALU0  : ProcResource<1>;    // SIMD/FP/SVE ALU0
57  def CortexA510UnitVALU1  : ProcResource<1>;    // SIMD/FP/SVE ALU0
58  def CortexA510UnitVMAC   : ProcResource<2>;    // SIMD/FP/SVE MAC
59  def CortexA510UnitVMC    : ProcResource<1>;    // SIMD/FP/SVE multicycle instrs  (e.g Div, SQRT, cryptography)
60}
61
62def CortexA510UnitLd     : ProcResGroup<[CortexA510UnitLdSt, CortexA510UnitLd1]>;
63def CortexA510UnitVALU   : ProcResGroup<[CortexA510UnitVALU0, CortexA510UnitVALU1]>;
64def CortexA510UnitALU    : ProcResGroup<[CortexA510UnitALU0, CortexA510UnitALU12]>;
65// These latencies are modeled without taking into account forwarding paths
66// (the software optimisation guide lists latencies taking into account
67// typical forwarding paths).
68def : WriteRes<WriteImm, [CortexA510UnitALU]> { let Latency = 1; }    // MOVN, MOVZ
69def : WriteRes<WriteI, [CortexA510UnitALU]> { let Latency = 1; }      // ALU
70def : WriteRes<WriteISReg, [CortexA510UnitALU]> { let Latency = 2; }  // ALU of Shifted-Reg
71def : WriteRes<WriteIEReg, [CortexA510UnitALU]> { let Latency = 2; }  // ALU of Extended-Reg
72def : WriteRes<WriteExtr, [CortexA510UnitALU]> { let Latency = 2; }   // EXTR from a reg pair
73def : WriteRes<WriteIS, [CortexA510UnitALU]> { let Latency = 2; }     // Shift/Scale
74
75// MAC
76def : WriteRes<WriteIM32, [CortexA510UnitMAC]> { let Latency = 3; }   // 32-bit Multiply
77def : WriteRes<WriteIM64, [CortexA510UnitMAC]> { let Latency = 5; let ReleaseAtCycles = [2];}   // 64-bit Multiply
78
79// Div
80def : WriteRes<WriteID32, [CortexA510UnitDiv]> {
81  let Latency = 8; let ReleaseAtCycles = [8];
82}
83def : WriteRes<WriteID64, [CortexA510UnitDiv]> {
84  let Latency = 16; let ReleaseAtCycles = [16];
85}
86
87//===----------------------------------------------------------------------===//
88// Define customized scheduler read/write types specific to the Cortex A510
89
90//===----------------------------------------------------------------------===//
91class CortexA510Write<int n, ProcResourceKind res> : SchedWriteRes<[res]> {
92  let Latency = n;
93}
94
95class CortexA510MCWrite<int n, int m, ProcResourceKind res> : SchedWriteRes<[res]> {
96  let Latency = n;
97  let ReleaseAtCycles = [m];
98  let BeginGroup = 1;
99}
100
101class CortexA510MC_RC0Write<int n, ProcResourceKind res> : SchedWriteRes<[res]> {
102  let Latency = n;
103  let BeginGroup = 1;
104}
105
106//===----------------------------------------------------------------------===//
107// Define generic 2 micro-op types
108def A510Write_10cyc_1VMAC_1VALU : SchedWriteRes<[CortexA510UnitVALU, CortexA510UnitVMAC]> {
109  let Latency     = 10;
110  let NumMicroOps = 2;
111}
112
113def A510Write_15cyc_1VMAC_1VALU : SchedWriteRes<[CortexA510UnitVALU, CortexA510UnitVMAC]> {
114  let Latency     = 15;
115  let NumMicroOps = 2;
116}
117
118class A510Write_PAC_B <int lat> : SchedWriteRes<[CortexA510UnitPAC, CortexA510UnitB]> {
119  let Latency = lat;
120  let NumMicroOps = 2;
121}
122// Load
123def : WriteRes<WriteLD, [CortexA510UnitLd]> { let Latency = 2; }
124def : WriteRes<WriteLDIdx, [CortexA510UnitLd]> { let Latency = 2; }
125def : WriteRes<WriteLDHi, [CortexA510UnitLd]> { let Latency = 2; }
126
127def CortexA510WriteVLD1 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 3; }
128def CortexA510WriteVLD1SI : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 3; let SingleIssue = 1; }
129def CortexA510WriteVLD2 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 4;
130                                                  let ReleaseAtCycles = [2]; }
131def CortexA510WriteVLD3 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 5;
132                                                  let ReleaseAtCycles = [3]; }
133def CortexA510WriteVLD4 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 6;
134                                                  let ReleaseAtCycles = [4]; }
135def CortexA510WriteVLD6 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 5;
136                                                  let ReleaseAtCycles = [3]; }
137def CortexA510WriteVLD8 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 6;
138                                                  let ReleaseAtCycles = [4]; }
139
140def CortexA510WriteLDP1 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 3; }
141def CortexA510WriteLDP2 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 3; }
142def CortexA510WriteLDP4 : SchedWriteRes<[CortexA510UnitLd]> { let Latency = 3; }
143
144// Pre/Post Indexing - Performed as part of address generation
145def : WriteRes<WriteAdr, []> { let Latency = 0; }
146
147// Store
148let RetireOOO = 1 in {
149def : WriteRes<WriteST, [CortexA510UnitLdSt]> { let Latency = 1; }
150def : WriteRes<WriteSTP, [CortexA510UnitLdSt]> { let Latency = 1; }
151def : WriteRes<WriteSTIdx, [CortexA510UnitLdSt]> { let Latency = 1; }
152}
153def : WriteRes<WriteSTX, [CortexA510UnitLdSt]> { let Latency = 3; }
154
155// Vector Store - Similar to vector loads, can take 1-3 cycles to issue.
156def : WriteRes<WriteVST, [CortexA510UnitLdSt]> { let Latency = 5;
157                                          let ReleaseAtCycles = [2];}
158def CortexA510WriteVST1 : SchedWriteRes<[CortexA510UnitLdSt]> { let Latency = 4; }
159def CortexA510WriteVST2 : SchedWriteRes<[CortexA510UnitLdSt]> { let Latency = 5;
160                                                  let ReleaseAtCycles = [2]; }
161def CortexA510WriteVST3 : SchedWriteRes<[CortexA510UnitLdSt]> { let Latency = 5;
162                                                  let ReleaseAtCycles = [3]; }
163def CortexA510WriteVST4 : SchedWriteRes<[CortexA510UnitLdSt]> { let Latency = 5;
164                                                  let ReleaseAtCycles = [4]; }
165
166def : WriteRes<WriteAtomic, []> { let Unsupported = 1; }
167
168// Branch
169def : WriteRes<WriteBr, [CortexA510UnitB]>;
170def : WriteRes<WriteBrReg, [CortexA510UnitB]>;
171def : WriteRes<WriteSys, [CortexA510UnitB]>;
172def : WriteRes<WriteBarrier, [CortexA510UnitB]>;
173def : WriteRes<WriteHint, [CortexA510UnitB]>;
174
175// FP ALU
176//   As WriteF result is produced in F5 and it can be mostly forwarded
177//   to consumer at F1, the effectively Latency is set as 4.
178def : WriteRes<WriteF, [CortexA510UnitVALU]> { let Latency = 4; }
179def : WriteRes<WriteFCmp, [CortexA510UnitVALU]> { let Latency = 3; }
180def : WriteRes<WriteFCvt, [CortexA510UnitVALU]> { let Latency = 4; }
181def : WriteRes<WriteFCopy, [CortexA510UnitVALU]> { let Latency = 3; }
182def : WriteRes<WriteFImm, [CortexA510UnitVALU]> { let Latency = 3; }
183
184class CortexA510VSt<int n> : SchedWriteRes<[CortexA510UnitLdSt]> {
185  let RetireOOO = 1;
186  let ReleaseAtCycles = [n];
187}
188
189def CortexA510VSt0      : SchedWriteRes<[CortexA510UnitLdSt]> {
190  let RetireOOO = 1;
191}
192
193def : SchedAlias<WriteVd, CortexA510Write<4, CortexA510UnitVALU>>;
194def : SchedAlias<WriteVq, CortexA510Write<4, CortexA510UnitVALU>>;
195
196// FP ALU specific new schedwrite definitions
197def CortexA510WriteFPALU_F3 : SchedWriteRes<[CortexA510UnitVALU]> { let Latency = 3;}
198def CortexA510WriteFPALU_F4 : SchedWriteRes<[CortexA510UnitVALU]> { let Latency = 4;}
199
200// FP Mul, Div, Sqrt. Div/Sqrt are not pipelined
201def : WriteRes<WriteFMul, [CortexA510UnitVMAC]> { let Latency = 4; }
202
203let RetireOOO = 1 in {
204def : WriteRes<WriteFDiv, [CortexA510UnitVMC]> { let Latency = 22;
205                                            let ReleaseAtCycles = [29]; }
206def CortexA510WriteVMAC : SchedWriteRes<[CortexA510UnitVMAC]> { let Latency = 4; }
207def CortexA510WriteFDivHP : SchedWriteRes<[CortexA510UnitVMC]> { let Latency = 8;
208                                                     let ReleaseAtCycles = [5]; }
209def CortexA510WriteFDivSP : SchedWriteRes<[CortexA510UnitVMC]> { let Latency = 13;
210                                                     let ReleaseAtCycles = [10]; }
211def CortexA510WriteFDivDP : SchedWriteRes<[CortexA510UnitVMC]> { let Latency = 22;
212                                                     let ReleaseAtCycles = [19]; }
213def CortexA510WriteFSqrtHP : SchedWriteRes<[CortexA510UnitVMC]> { let Latency = 8;
214                                                      let ReleaseAtCycles = [5]; }
215def CortexA510WriteFSqrtSP : SchedWriteRes<[CortexA510UnitVMC]> { let Latency = 12;
216                                                      let ReleaseAtCycles = [9]; }
217def CortexA510WriteFSqrtDP : SchedWriteRes<[CortexA510UnitVMC]> { let Latency = 22;
218                                                      let ReleaseAtCycles = [19]; }
219}
220//===----------------------------------------------------------------------===//
221// Subtarget-specific SchedRead types.
222
223def : ReadAdvance<ReadVLD, 0>;
224def : ReadAdvance<ReadExtrHi, 0>;
225def : ReadAdvance<ReadAdrBase, 0>;
226def : ReadAdvance<ReadST, 1>;
227
228def : ReadAdvance<ReadI, 0>;
229def : ReadAdvance<ReadISReg, 0>;
230def : ReadAdvance<ReadIEReg, 0>;
231
232
233// MUL
234def : ReadAdvance<ReadIM, 0>;
235def : ReadAdvance<ReadIMA, 2>;
236
237// Div
238def : ReadAdvance<ReadID, 0>;
239
240//===----------------------------------------------------------------------===//
241// Subtarget-specific InstRWs.
242
243def A510WriteISReg : SchedWriteVariant<[
244       SchedVar<RegShiftedPred, [WriteISReg]>,
245       SchedVar<NoSchedPred, [WriteI]>]>;
246def : InstRW<[A510WriteISReg], (instregex ".*rs$")>;
247def : InstRW<[WriteIS], (instrs RBITWr, RBITXr)>;
248
249// Pointer Authentication Instructions (v8.3 PAC)
250// -----------------------------------------------------------------------------
251
252// Authenticate data address
253// Authenticate instruction address
254// Compute pointer authentication code for data address
255// Compute pointer authentication code, using generic key
256// Compute pointer authentication code for instruction address
257def : InstRW<[CortexA510Write<3, CortexA510UnitPAC>], (instregex "^AUT", "^PAC")>;
258
259// Branch and link, register, with pointer authentication
260// Branch, register, with pointer authentication
261// Branch, return, with pointer authentication
262def : InstRW<[A510Write_PAC_B<1>], (instrs BLRAA, BLRAAZ, BLRAB, BLRABZ, BRAA,
263                                            BRAAZ, BRAB, BRABZ, RETAA, RETAB,
264                                            ERETAA, ERETAB)>;
265
266// Load register, with pointer authentication
267def : InstRW<[CortexA510Write<2, CortexA510UnitPAC>], (instregex "^LDRA[AB](indexed|writeback)")>;
268
269// Strip pointer authentication code
270def : InstRW<[CortexA510Write<5, CortexA510UnitPAC>], (instrs XPACD, XPACI, XPACLRI)>;
271//---
272// Miscellaneous
273//---
274def : InstRW<[CortexA510WriteVLD1SI,CortexA510WriteLDP1], (instregex "LDPS?Wi")>;
275def : InstRW<[CortexA510WriteVLD1,CortexA510WriteLDP1], (instregex "LDPSi")>;
276def : InstRW<[CortexA510WriteVLD1,CortexA510WriteLDP2], (instregex "LDP(X|D)i")>;
277def : InstRW<[CortexA510WriteVLD1,CortexA510WriteLDP4], (instregex "LDPQi")>;
278def : InstRW<[WriteAdr, CortexA510WriteVLD1SI,CortexA510WriteLDP1], (instregex "LDPS?W(pre|post)")>;
279def : InstRW<[WriteAdr, CortexA510WriteVLD1,CortexA510WriteLDP1], (instregex "LDPS(pre|post)")>;
280def : InstRW<[WriteAdr, CortexA510WriteVLD1,CortexA510WriteLDP2], (instregex "LDP(X|D)(pre|post)")>;
281def : InstRW<[WriteAdr, CortexA510WriteVLD1,CortexA510WriteLDP4], (instregex "LDPQ(pre|post)")>;
282def : InstRW<[WriteI], (instrs COPY)>;
283//---
284// Vector Loads - 128-bit per cycle
285//---
286//   1-element structures
287def : InstRW<[CortexA510WriteVLD1], (instregex "LD1i(8|16|32|64)$")>;                // single element
288def : InstRW<[CortexA510WriteVLD1], (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; // replicate
289def : InstRW<[CortexA510WriteVLD1], (instregex "LD1Onev(8b|4h|2s|1d)$")>;
290def : InstRW<[CortexA510WriteVLD1], (instregex "LD1Onev(16b|8h|4s|2d)$")>;
291def : InstRW<[CortexA510WriteVLD1], (instregex "LD1Twov(8b|4h|2s|1d)$")>; // multiple structures
292def : InstRW<[CortexA510WriteVLD1], (instregex "LD1Twov(16b|8h|4s|2d)$")>;
293def : InstRW<[CortexA510WriteVLD2], (instregex "LD1Threev(8b|4h|2s|1d)$")>;
294def : InstRW<[CortexA510WriteVLD2], (instregex "LD1Threev(16b|8h|4s|2d)$")>;
295def : InstRW<[CortexA510WriteVLD2], (instregex "LD1Fourv(8b|4h|2s|1d)$")>;
296def : InstRW<[CortexA510WriteVLD2], (instregex "LD1Fourv(16b|8h|4s|2d)$")>;
297
298def : InstRW<[WriteAdr, CortexA510WriteVLD1], (instregex "LD1i(8|16|32|64)_POST$")>;
299def : InstRW<[WriteAdr, CortexA510WriteVLD1], (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
300def : InstRW<[WriteAdr, CortexA510WriteVLD1], (instregex "LD1Onev(8b|4h|2s|1d)_POST$")>;
301def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD1Onev(16b|8h|4s|2d)_POST$")>;
302def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD1Twov(8b|4h|2s|1d)_POST$")>;
303def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD1Twov(16b|8h|4s|2d)_POST$")>;
304def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD1Threev(8b|4h|2s|1d)_POST$")>;
305def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD1Threev(16b|8h|4s|2d)_POST$")>;
306def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD1Fourv(8b|4h|2s|1d)_POST$")>;
307def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD1Fourv(16b|8h|4s|2d)_POST$")>;
308
309//    2-element structures
310def : InstRW<[CortexA510WriteVLD2], (instregex "LD2i(8|16|32|64)$")>;
311def : InstRW<[CortexA510WriteVLD2], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
312def : InstRW<[CortexA510WriteVLD2], (instregex "LD2Twov(8b|4h|2s)$")>;
313def : InstRW<[CortexA510WriteVLD4], (instregex "LD2Twov(16b|8h|4s|2d)$")>;
314
315def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD2i(8|16|32|64)(_POST)?$")>;
316def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)(_POST)?$")>;
317def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD2Twov(8b|4h|2s)(_POST)?$")>;
318def : InstRW<[WriteAdr, CortexA510WriteVLD4], (instregex "LD2Twov(16b|8h|4s|2d)(_POST)?$")>;
319
320//    3-element structures
321def : InstRW<[CortexA510WriteVLD2], (instregex "LD3i(8|16|32|64)$")>;
322def : InstRW<[CortexA510WriteVLD2], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
323def : InstRW<[CortexA510WriteVLD3], (instregex "LD3Threev(8b|4h|2s|1d)$")>;
324def : InstRW<[CortexA510WriteVLD6], (instregex "LD3Threev(16b|8h|4s|2d)$")>;
325
326def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD3i(8|16|32|64)_POST$")>;
327def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
328def : InstRW<[WriteAdr, CortexA510WriteVLD3], (instregex "LD3Threev(8b|4h|2s|1d)_POST$")>;
329def : InstRW<[WriteAdr, CortexA510WriteVLD6], (instregex "LD3Threev(16b|8h|4s|2d)_POST$")>;
330
331//    4-element structures
332def : InstRW<[CortexA510WriteVLD2], (instregex "LD4i(8|16|32|64)$")>;                // load single 4-el structure to one lane of 4 regs.
333def : InstRW<[CortexA510WriteVLD2], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; // load single 4-el structure, replicate to all lanes of 4 regs.
334def : InstRW<[CortexA510WriteVLD4], (instregex "LD4Fourv(8b|4h|2s|1d)$")>;           // load multiple 4-el structures to 4 regs.
335def : InstRW<[CortexA510WriteVLD8], (instregex "LD4Fourv(16b|8h|4s|2d)$")>;
336
337def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD4i(8|16|32|64)_POST$")>;
338def : InstRW<[WriteAdr, CortexA510WriteVLD2], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
339def : InstRW<[WriteAdr, CortexA510WriteVLD4], (instregex "LD4Fourv(8b|4h|2s|1d)_POST$")>;
340def : InstRW<[WriteAdr, CortexA510WriteVLD8], (instregex "LD4Fourv(16b|8h|4s|2d)_POST$")>;
341
342//---
343// Vector Stores
344//---
345def : InstRW<[CortexA510WriteVST1], (instregex "ST1i(8|16|32|64)$")>;
346def : InstRW<[CortexA510WriteVST1], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
347def : InstRW<[CortexA510WriteVST1], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
348def : InstRW<[CortexA510WriteVST2], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
349def : InstRW<[CortexA510WriteVST4], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
350def : InstRW<[WriteAdr, CortexA510WriteVST1], (instregex "ST1i(8|16|32|64)_POST$")>;
351def : InstRW<[WriteAdr, CortexA510WriteVST1], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
352def : InstRW<[WriteAdr, CortexA510WriteVST1], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
353def : InstRW<[WriteAdr, CortexA510WriteVST2], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
354def : InstRW<[WriteAdr, CortexA510WriteVST4], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
355
356def : InstRW<[CortexA510WriteVST2], (instregex "ST2i(8|16|32|64)$")>;
357def : InstRW<[CortexA510WriteVST2], (instregex "ST2Twov(8b|4h|2s)$")>;
358def : InstRW<[CortexA510WriteVST4], (instregex "ST2Twov(16b|8h|4s|2d)$")>;
359def : InstRW<[WriteAdr, CortexA510WriteVST2], (instregex "ST2i(8|16|32|64)_POST$")>;
360def : InstRW<[WriteAdr, CortexA510WriteVST2], (instregex "ST2Twov(8b|4h|2s)_POST$")>;
361def : InstRW<[WriteAdr, CortexA510WriteVST4], (instregex "ST2Twov(16b|8h|4s|2d)_POST$")>;
362
363def : InstRW<[CortexA510WriteVST2], (instregex "ST3i(8|16|32|64)$")>;
364def : InstRW<[CortexA510WriteVST4], (instregex "ST3Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
365def : InstRW<[WriteAdr, CortexA510WriteVST2], (instregex "ST3i(8|16|32|64)_POST$")>;
366def : InstRW<[WriteAdr, CortexA510WriteVST4], (instregex "ST3Threev(8b|4h|2s|1d|2d|16b|8h|4s|4d)_POST$")>;
367
368def : InstRW<[CortexA510WriteVST2], (instregex "ST4i(8|16|32|64)$")>;
369def : InstRW<[CortexA510WriteVST4], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
370def : InstRW<[WriteAdr, CortexA510WriteVST2], (instregex "ST4i(8|16|32|64)_POST$")>;
371def : InstRW<[WriteAdr, CortexA510WriteVST4], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
372
373//---
374// Floating Point Conversions, MAC, DIV, SQRT
375//---
376def : InstRW<[CortexA510WriteFPALU_F3], (instregex "^DUP(v2i64|v4i32|v8i16|v16i8)")>;
377def : InstRW<[CortexA510WriteFPALU_F4], (instregex "^XTN")>;
378def : InstRW<[CortexA510WriteFPALU_F4], (instregex "^FCVT[ALMNPZ][SU](S|U)?(W|X)")>;
379def : InstRW<[CortexA510WriteFPALU_F4], (instregex "^FCVT(X)?[ALMNPXZ](S|U|N)?v")>;
380
381def : InstRW<[CortexA510WriteFPALU_F4], (instregex "^(S|U)CVTF(S|U)(W|X)(H|S|D)")>;
382def : InstRW<[CortexA510WriteFPALU_F4], (instregex "^(S|U)CVTF(h|s|d)")>;
383def : InstRW<[CortexA510WriteFPALU_F4], (instregex "^(S|U)CVTFv")>;
384
385def : InstRW<[CortexA510WriteVMAC], (instregex "^FN?M(ADD|SUB).*")>;
386def : InstRW<[CortexA510WriteVMAC], (instregex "^FML(A|S)v.*")>;
387def : InstRW<[CortexA510WriteFDivHP], (instrs FDIVHrr)>;
388def : InstRW<[CortexA510WriteFDivSP], (instrs FDIVSrr)>;
389def : InstRW<[CortexA510WriteFDivDP], (instrs FDIVDrr)>;
390def : InstRW<[CortexA510WriteFDivHP], (instregex "^FDIVv.*16$")>;
391def : InstRW<[CortexA510WriteFDivSP], (instregex "^FDIVv.*32$")>;
392def : InstRW<[CortexA510WriteFDivDP], (instregex "^FDIVv.*64$")>;
393def : InstRW<[CortexA510WriteFSqrtHP], (instregex "^.*SQRT.*16$")>;
394def : InstRW<[CortexA510WriteFSqrtSP], (instregex "^.*SQRT.*32$")>;
395def : InstRW<[CortexA510WriteFSqrtDP], (instregex "^.*SQRT.*64$")>;
396
397// 4.15. Advanced SIMD integer instructions
398// ASIMD absolute diff
399def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]ABDv(2i32|4i16|8i8)")>;
400def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]ABDv(16i8|4i32|8i16)")>;
401// ASIMD absolute diff accum
402def : InstRW<[CortexA510Write<8, CortexA510UnitVALU>], (instregex "[SU]ABAL?v")>;
403// ASIMD absolute diff long
404def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]ABDLv")>;
405// ASIMD arith #1
406def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "(ADD|SUB|NEG)v(1i64|2i32|4i16|8i8)",
407  "[SU]R?HADDv(2i32|4i16|8i8)", "[SU]HSUBv(2i32|4i16|8i8)")>;
408def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "(ADD|SUB|NEG)v(2i64|4i32|8i16|16i8)",
409  "[SU]R?HADDv(8i16|4i32|16i8)", "[SU]HSUBv(8i16|4i32|16i8)")>;
410// ASIMD arith #2
411def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "ABSv(1i64|2i32|4i16|8i8)$",
412  "[SU]ADDLPv(2i32_v1i64|4i16_v2i32|8i8_v4i16)$",
413  "([SU]QADD|[SU]QSUB|SQNEG|SUQADD|USQADD)v(1i16|1i32|1i64|1i8|2i32|4i16|8i8)$",
414  "ADDPv(2i32|4i16|8i8)$")>;
415def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "ABSv(2i64|4i32|8i16|16i8)$",
416  "[SU]ADDLPv(16i8_v8i16|4i32_v2i64|8i16_v4i32)$",
417  "([SU]QADD|[SU]QSUB|SQNEG|SUQADD|USQADD)v(16i8|2i64|4i32|8i16)$",
418  "ADDPv(16i8|2i64|4i32|8i16)$")>;
419// ASIMD arith #3
420def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex  "SADDLv", "UADDLv", "SADDWv",
421  "UADDWv", "SSUBLv", "USUBLv", "SSUBWv", "USUBWv", "ADDHNv", "SUBHNv")>;
422// ASIMD arith #5
423def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "RADDHNv", "RSUBHNv")>;
424// ASIMD arith, reduce
425def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex  "ADDVv", "SADDLVv", "UADDLVv")>;
426// ASIMD compare #1
427def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "CM(EQ|GE|GT|HI|HS|LE|LT)v(1i64|2i32|4i16|8i8)")>;
428def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "CM(EQ|GE|GT|HI|HS|LE|LT)v(2i64|4i32|8i16|16i8)")>;
429// ASIMD compare #2
430def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "CMTSTv(1i64|2i32|4i16|8i8)")>;
431def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "CMTSTv(2i64|4i32|8i16|16i8)")>;
432// ASIMD logical $1
433def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "(AND|EOR|NOT|ORN)v8i8",
434  "(ORR|BIC)v(2i32|4i16|8i8)$", "MVNIv(2i|2s|4i16)")>;
435def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "(AND|EOR|NOT|ORN)v16i8",
436  "(ORR|BIC)v(16i8|4i32|8i16)$", "MVNIv(4i32|4s|8i16)")>;
437// ASIMD max/min, basic
438def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU](MIN|MAX)P?v(2i32|4i16|8i8)")>;
439def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU](MIN|MAX)P?v(16i8|4i132|8i16)")>;
440// SIMD max/min, reduce
441def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU](MAX|MIN)Vv")>;
442// ASIMD multiply, by element
443def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "MULv(2i32|4i16|4i32|8i16)_indexed$",
444  "SQR?DMULHv(1i16|1i32|2i32|4i16|4i32|8i16)_indexed$")>;
445// ASIMD multiply
446def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instrs PMULv8i8)>;
447def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instrs PMULv16i8)>;
448// ASIMD multiply accumulate
449def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "ML[AS]v(2i32|4i16|8i8)$")>;
450def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "ML[AS]v(16i8|4i32|8i16)$")>;
451def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "ML[AS]v(2i32|4i16|4i32|8i16)_indexed$")>;
452// ASIMD multiply accumulate half
453def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "SQRDML[AS]H[vi]")>;
454// ASIMD multiply accumulate long
455def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]ML[AS]Lv")>;
456// ASIMD multiply accumulate long #2
457def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "SQDML[AS]L[iv]")>;
458// ASIMD dot product
459def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]DOTv8i8")>;
460def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]DOTv16i8")>;
461// ASIMD dot product, by scalar
462def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]DOTlanev")>;
463// ASIMD multiply long
464def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]MULLv", "SQDMULL[iv]")>;
465// ASIMD polynomial (8x8) multiply long
466def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instrs PMULLv8i8, PMULLv16i8)>;
467// ASIMD pairwise add and accumulate
468def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU>], (instregex "[SU]ADALPv")>;
469// ASIMD shift accumulate
470def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU>], (instregex "[SU]SRA(d|v2i32|v4i16|v8i8)")>;
471def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU>], (instregex "[SU]SRAv(16i8|2i64|4i32|8i16)")>;
472// ASIMD shift accumulate #2
473def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU>], (instregex "[SU]RSRA[vd]")>;
474// ASIMD shift by immed
475def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "SHLd$", "SHLv",
476  "SLId$", "SRId$", "[SU]SHR[vd]", "SHRNv")>;
477// ASIMD shift by immed
478// SXTL and UXTL are aliases for SHLL
479def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[US]?SHLLv")>;
480// ASIMD shift by immed #2
481def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]RSHR(d|v2i32|v4i16|v8i8)",
482  "[SU]RSHRv(16i8|2i64|4i32|8i16)")>;
483def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "RSHRNv(2i32|4i16|8i8)",
484  "RSHRNv(16i8|4i32|8i16)")>;
485// ASIMD shift by register
486def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]SHLv(1i64|2i32|4i16|8i8)")>;
487def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]SHLv(2i64|4i32|8i16|16i8)")>;
488// ASIMD shift by register #2
489def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]RSHLv(1i64|2i32|4i16|8i8)")>;
490def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "[SU]RSHLv(2i64|4i32|8i16|16i8)")>;
491
492def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]QSHLv(1i64|2i32|4i16|8i8)")>;
493def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]QSHLv(2i64|4i32|8i16|16i8)")>;
494
495def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]QRSHLv(1i64|2i32|4i16|8i8)")>;
496def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "[SU]QRSHLv(2i64|4i32|8i16|16i8)")>;
497
498// Cryptography extensions
499// -----------------------------------------------------------------------------
500
501// Crypto AES ops
502def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^AES[DE]rr$", "^AESI?MCrr")>;
503
504// Crypto polynomial (64x64) multiply long
505def : InstRW<[CortexA510MCWrite<8, 0, CortexA510UnitVMC>], (instrs PMULLv1i64, PMULLv2i64)>;
506
507// Crypto SHA1 hash acceleration op
508// Crypto SHA1 schedule acceleration ops
509def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^SHA1(H|SU0|SU1)")>;
510
511// Crypto SHA1 hash acceleration ops
512// Crypto SHA256 hash acceleration ops
513def : InstRW<[CortexA510MCWrite<8, 0, CortexA510UnitVMC>], (instregex "^SHA1[CMP]", "^SHA256H2?")>;
514
515// Crypto SHA256 schedule acceleration ops
516def : InstRW<[CortexA510MCWrite<8, 0, CortexA510UnitVMC>], (instregex "^SHA256SU[01]")>;
517
518// Crypto SHA512 hash acceleration ops
519def : InstRW<[CortexA510MCWrite<8, 0, CortexA510UnitVMC>], (instregex "^SHA512(H|H2|SU0|SU1)")>;
520
521// Crypto SHA3 ops
522def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instrs BCAX, EOR3, XAR)>;
523def : InstRW<[CortexA510MCWrite<8, 0, CortexA510UnitVMC>], (instrs RAX1)>;
524
525
526// Crypto SM3 ops
527def : InstRW<[CortexA510MCWrite<8, 0, CortexA510UnitVMC>], (instregex "^SM3PARTW[12]$", "^SM3SS1$",
528                                                            "^SM3TT[12][AB]$")>;
529
530// Crypto SM4 ops
531def : InstRW<[CortexA510MCWrite<8, 0, CortexA510UnitVMC>], (instrs SM4E, SM4ENCKEY)>;
532
533// CRC
534// -----------------------------------------------------------------------------
535
536def : InstRW<[CortexA510MCWrite<2, 0, CortexA510UnitMAC>], (instregex "^CRC32")>;
537
538// SVE Predicate instructions
539
540// Loop control, based on predicate
541def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs BRKA_PPmP, BRKA_PPzP,
542                                                  BRKB_PPmP, BRKB_PPzP)>;
543
544// Loop control, based on predicate and flag setting
545def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs BRKAS_PPzP, BRKBS_PPzP)>;
546
547// Loop control, propagating
548def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs BRKN_PPzP, BRKPA_PPzPP, BRKPB_PPzPP)>;
549
550// Loop control, propagating and flag setting
551def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs BRKNS_PPzP)>;
552def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs BRKPAS_PPzPP, BRKPBS_PPzPP)>;
553
554
555// Loop control, based on GPR
556def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>],
557             (instregex "^WHILE(GE|GT|HI|HS|LE|LO|LS|LT)_P(WW|XX)_[BHSD]")>;
558
559def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instregex "^WHILE(RW|WR)_PXX_[BHSD]")>;
560
561// Loop terminate
562def : InstRW<[CortexA510Write<1, CortexA510UnitALU>], (instregex "^CTERM(EQ|NE)_(WW|XX)")>;
563
564// Predicate counting scalar
565def : InstRW<[CortexA510Write<1, CortexA510UnitALU>], (instrs ADDPL_XXI, ADDVL_XXI, RDVLI_XI)>;
566
567def : InstRW<[CortexA510Write<1, CortexA510UnitALU>],
568             (instregex "^CNT[BHWD]_XPiI")>;
569
570def : InstRW<[CortexA510Write<1, CortexA510UnitALU>],
571             (instregex "^(INC|DEC)[BHWD]_XPiI")>;
572
573def : InstRW<[CortexA510Write<1, CortexA510UnitALU>],
574             (instregex "^(SQINC|SQDEC|UQINC|UQDEC)[BHWD]_[XW]Pi(Wd)?I")>;
575
576// Predicate counting scalar, active predicate
577def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>],
578             (instregex "^CNTP_XPP_[BHSD]")>;
579
580def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>],
581             (instregex "^(DEC|INC)P_XP_[BHSD]")>;
582
583def : InstRW<[CortexA510Write<8, CortexA510UnitVALU0>],
584             (instregex "^(SQDEC|SQINC|UQDEC|UQINC)P_XP_[BHSD]",
585                        "^(UQDEC|UQINC)P_WP_[BHSD]",
586                        "^(SQDEC|SQINC|UQDEC|UQINC)P_XPWd_[BHSD]")>;
587
588
589// Predicate counting vector, active predicate
590def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
591             (instregex "^(DEC|INC|SQDEC|SQINC|UQDEC|UQINC)P_ZP_[HSD]")>;
592
593// Predicate logical
594def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>],
595             (instregex "^(AND|BIC|EOR|NAND|NOR|ORN|ORR)_PPzPP")>;
596
597// Predicate logical, flag setting
598def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>],
599             (instregex "^(ANDS|BICS|EORS|NANDS|NORS|ORNS|ORRS)_PPzPP")>;
600
601// Predicate reverse
602def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instregex "^REV_PP_[BHSD]")>;
603
604// Predicate select
605def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs SEL_PPPP)>;
606
607// Predicate set
608def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instregex "^PFALSE", "^PTRUE_[BHSD]")>;
609
610// Predicate set/initialize, set flags
611def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instregex "^PTRUES_[BHSD]")>;
612
613// Predicate find first/next
614def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instregex "^PFIRST_B", "^PNEXT_[BHSD]")>;
615
616// Predicate test
617def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs PTEST_PP)>;
618
619// Predicate transpose
620def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instregex "^TRN[12]_PPP_[BHSDQ]")>;
621
622// Predicate unpack and widen
623def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instrs PUNPKHI_PP, PUNPKLO_PP)>;
624
625// Predicate zip/unzip
626def : InstRW<[CortexA510Write<6, CortexA510UnitVALU0>], (instregex "^(ZIP|UZP)[12]_PPP_[BHSDQ]")>;
627
628
629// SVE integer instructions
630// -----------------------------------------------------------------------------
631// Arithmetic, absolute diff
632def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^[SU]ABD_(ZPmZ|ZPZZ)_[BHSD]")>;
633
634// Arithmetic, absolute diff accum
635def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU>], (instregex "^[SU]ABA_ZZZ_[BHSD]")>;
636
637// Arithmetic, absolute diff accum long
638def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU>], (instregex "^[SU]ABAL[TB]_ZZZ_[HSD]")>;
639
640// Arithmetic, absolute diff long
641def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^[SU]ABDL[TB]_ZZZ_[HSD]")>;
642
643// Arithmetic, basic
644def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>],
645             (instregex "^(ABS|CNOT|NEG)_ZPmZ_[BHSD]",
646                        "^(ADD|SUB|SUBR)_ZPmZ_[BHSD]",
647                        "^(ADD|SUB|SUBR)_ZPZZ_[BHSD]",
648                        "^(ADD|SUB)_ZZZ_[BHSD]",
649                        "^(ADD|SUB|SUBR)_ZI_[BHSD]",
650                        "^ADR_[SU]XTW_ZZZ_D_[0123]",
651                        "^ADR_LSL_ZZZ_[SD]_[0123]",
652                        "^[SU](ADD|SUB)[LW][BT]_ZZZ_[HSD]",
653                        "^SADDLBT_ZZZ_[HSD]",
654                        "^[SU]H(ADD|SUB|SUBR)_ZPmZ_[BHSD]",
655                        "^SSUBL(BT|TB)_ZZZ_[HSD]")>;
656
657// Arithmetic, complex
658def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
659             (instregex "^R?(ADD|SUB)HN[BT]_ZZZ_[BHS]",
660                        "^SQ(ABS|NEG)_ZPmZ_[BHSD]",
661                        "^SQ(ADD|SUB|SUBR)_ZPmZ_?[BHSD]",
662                        "^[SU]Q(ADD|SUB)_ZZZ_[BHSD]",
663                        "^[SU]Q(ADD|SUB)_ZI_[BHSD]",
664                        "^(SRH|SUQ|UQ|USQ|URH)ADD_ZPmZ_[BHSD]",
665                        "^(UQSUB|UQSUBR)_ZPmZ_[BHSD]")>;
666
667// Arithmetic, large integer
668def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^(AD|SB)CL[BT]_ZZZ_[SD]")>;
669
670// Arithmetic, pairwise add
671def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^ADDP_ZPmZ_[BHSD]")>;
672
673// Arithmetic, pairwise add and accum long
674def : InstRW<[CortexA510MCWrite<7, 2, CortexA510UnitVALU>], (instregex "^[SU]ADALP_ZPmZ_[HSD]")>;
675
676// Arithmetic, shift
677def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>],
678             (instregex "^(ASR|LSL|LSR)_WIDE_ZPmZ_[BHS]",
679                        "^(ASR|LSL|LSR)_WIDE_ZZZ_[BHS]",
680                        "^(ASR|LSL|LSR)_ZPmI_[BHSD]",
681                        "^(ASR|LSL|LSR)_ZPZI_[BHSD]",
682                        "^(ASR|LSL|LSR)_ZPmZ_[BHSD]",
683                        "^(ASR|LSL|LSR)_ZPZZ_[BHSD]",
684                        "^(ASR|LSL|LSR)_ZZI_[BHSD]",
685                        "^(ASRR|LSLR|LSRR)_ZPmZ_[BHSD]")>;
686// Arithmetic, shift right for divide
687def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
688             (instregex "^ASRD_ZPmI_[BHSD]",
689                        "^ASRD_ZPZI_[BHSD]")>;
690
691// Arithmetic, shift and accumulate
692def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
693             (instregex "^(SSRA|USRA)_ZZI_[BHSD]")>;
694
695def : InstRW<[CortexA510MCWrite<7, 2, CortexA510UnitVALU>],
696             (instregex "^(SRSRA|URSRA)_ZZI_[BHSD]")>;
697
698
699// Arithmetic, shift by immediate
700// Arithmetic, shift by immediate and insert
701def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>],
702             (instregex "^(SHRNB|SHRNT|SSHLLB|SSHLLT|USHLLB|USHLLT|SLI|SRI)_ZZI_[BHSD]")>;
703
704// Arithmetic, shift complex
705def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
706             (instregex "^(SQ)?RSHRU?N[BT]_ZZI_[BHS]",
707                        "^(SQRSHL|SQRSHLR|SQSHL|SQSHLR|UQRSHL|UQRSHLR|UQSHL|UQSHLR)_(ZPmZ|ZPZZ)_[BHSD]",
708                        "^(SQSHL|SQSHLU|UQSHL)_(ZPmI|ZPZI)_[BHSD]",
709                        "^SQSHRU?N[BT]_ZZI_[BHS]",
710                        "^UQR?SHRN[BT]_ZZI_[BHS]")>;
711
712// Arithmetic, shift rounding
713def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
714             (instregex "^(SRSHL|SRSHR|SRSHLR|URSHL|URSHLR|URSHR)_(ZPmZ|ZPZZ|ZPZI)_[BHSD]",
715                        "^[SU]RSHR_ZPmI_[BHSD]")>;
716
717// Bit manipulation
718def : InstRW<[CortexA510MCWrite<14, 13, CortexA510UnitVMC>],
719             (instregex "^(BDEP|BEXT|BGRP)_ZZZ_B")>;
720
721def : InstRW<[CortexA510MCWrite<22, 21, CortexA510UnitVMC>],
722             (instregex "^(BDEP|BEXT|BGRP)_ZZZ_H")>;
723
724def : InstRW<[CortexA510MCWrite<38, 37, CortexA510UnitVMC>],
725             (instregex "^(BDEP|BEXT|BGRP)_ZZZ_S")>;
726
727def : InstRW<[CortexA510MCWrite<70, 69, CortexA510UnitVMC>],
728             (instregex "^(BDEP|BEXT|BGRP)_ZZZ_D")>;
729
730
731// Bitwise select
732def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^(BSL|BSL1N|BSL2N|NBSL)_ZZZZ")>;
733
734// Count/reverse bits
735def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^(CLS|CLZ|RBIT)_ZPmZ_[BHSD]")>;
736def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^CNT_ZPmZ_[BH]")>;
737def : InstRW<[CortexA510Write<8, CortexA510UnitVALU>], (instregex "^CNT_ZPmZ_S")>;
738def : InstRW<[CortexA510Write<12, CortexA510UnitVALU>], (instregex "^CNT_ZPmZ_D")>;
739// Broadcast logical bitmask immediate to vector
740def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instrs DUPM_ZI)>;
741
742// Compare and set flags
743def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>],
744             (instregex "^CMP(EQ|GE|GT|HI|HS|LE|LO|LS|LT|NE)_PPzZ[IZ]_[BHSD]",
745                        "^CMP(EQ|GE|GT|HI|HS|LE|LO|LS|LT|NE)_WIDE_PPzZZ_[BHS]")>;
746
747// Complex add
748def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^CADD_ZZI_[BHSD]")>;
749
750def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^SQCADD_ZZI_[BHSD]")>;
751
752// Complex dot product 8-bit element
753def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instrs CDOT_ZZZ_S, CDOT_ZZZI_S)>;
754
755// Complex dot product 16-bit element
756def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instrs CDOT_ZZZ_D, CDOT_ZZZI_D)>;
757
758// Complex multiply-add B, H, S element size
759def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^CMLA_ZZZ_[BHS]",
760                                            "^CMLA_ZZZI_[HS]")>;
761
762// Complex multiply-add D element size
763def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instrs CMLA_ZZZ_D)>;
764
765// Conditional extract operations, scalar form
766def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU>], (instregex "^CLAST[AB]_RPZ_[BHSD]")>;
767
768// Conditional extract operations, SIMD&FP scalar and vector forms
769def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^CLAST[AB]_[VZ]PZ_[BHSD]",
770                                            "^COMPACT_ZPZ_[SD]",
771                                            "^SPLICE_ZPZZ?_[BHSD]")>;
772
773// Convert to floating point, 64b to float or convert to double
774def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^[SU]CVTF_ZPmZ_Dto[SD]")>;
775
776// Convert to floating point, 64b to half
777def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^[SU]CVTF_ZPmZ_DtoH")>;
778
779// Convert to floating point, 32b to single or half
780def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^[SU]CVTF_ZPmZ_Sto[HS]")>;
781
782// Convert to floating point, 32b to double
783def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^[SU]CVTF_ZPmZ_StoD")>;
784
785// Convert to floating point, 16b to half
786def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^[SU]CVTF_ZPmZ_HtoH")>;
787
788// Copy, scalar
789def : InstRW<[CortexA510Write<3, CortexA510UnitVALU0>],(instregex "^CPY_ZPmR_[BHSD]")>;
790
791// Copy, scalar SIMD&FP or imm
792def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^CPY_ZPm[IV]_[BHSD]",
793                                           "^CPY_ZPzI_[BHSD]")>;
794
795// Divides, 32 bit
796def : InstRW<[CortexA510MCWrite<15, 12, CortexA510UnitVMC>], (instregex "^[SU]DIVR?_(ZPmZ|ZPZZ)_S")>;
797
798// Divides, 64 bit
799def : InstRW<[CortexA510MCWrite<26, 23, CortexA510UnitVMC>], (instregex "^[SU]DIVR?_(ZPmZ|ZPZZ)_D")>;
800
801// Dot product, 8 bit
802def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^[SU]DOT_ZZZI?_S")>;
803
804// Dot product, 8 bit, using signed and unsigned integers
805def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instrs SUDOT_ZZZI, USDOT_ZZZI, USDOT_ZZZ)>;
806
807// Dot product, 16 bit
808def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^[SU]DOT_ZZZI?_D")>;
809
810// Duplicate, immediate and indexed form
811def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^DUP_ZI_[BHSD]",
812                                           "^DUP_ZZI_[BHSDQ]")>;
813
814// Duplicate, scalar form
815def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^DUP_ZR_[BHSD]")>;
816
817// Extend, sign or zero
818def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^[SU]XTB_ZPmZ_[HSD]",
819                                            "^[SU]XTH_ZPmZ_[SD]",
820                                            "^[SU]XTW_ZPmZ_[D]")>;
821
822// Extract
823def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instrs EXT_ZZI, EXT_ZZI_B)>;
824
825// Extract narrow saturating
826def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^[SU]QXTN[BT]_ZZ_[BHS]",
827                                            "^SQXTUN[BT]_ZZ_[BHS]")>;
828
829// Extract/insert operation, SIMD and FP scalar form
830def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^LAST[AB]_VPZ_[BHSD]",
831                                            "^INSR_ZV_[BHSD]")>;
832
833// Extract/insert operation, scalar
834def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU0>], (instregex "^LAST[AB]_RPZ_[BHSD]",
835                                                "^INSR_ZR_[BHSD]")>;
836
837// Histogram operations
838def : InstRW<[CortexA510MCWrite<8, 2, CortexA510UnitVALU0>], (instregex "^HISTCNT_ZPzZZ_[SD]",
839                                                  "^HISTSEG_ZZZ")>;
840
841// Horizontal operations, B, H, S form, immediate operands only
842def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^INDEX_II_[BHS]")>;
843
844// Horizontal operations, B, H, S form, scalar, immediate operands/ scalar
845// operands only / immediate, scalar operands
846def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^INDEX_(IR|RI|RR)_[BHS]")>;
847
848// Horizontal operations, D form, immediate operands only
849def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instrs INDEX_II_D)>;
850
851// Horizontal operations, D form, scalar, immediate operands)/ scalar operands
852// only / immediate, scalar operands
853def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^INDEX_(IR|RI|RR)_D")>;
854
855// Logical
856def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>],
857             (instregex "^(AND|EOR|ORR)_ZI",
858                        "^(AND|BIC|EOR|EOR|ORR)_ZZZ",
859                        "^(AND|BIC|EOR|NOT|ORR)_ZPmZ_[BHSD]",
860                        "^(AND|BIC|EOR|NOT|ORR)_ZPZZ_[BHSD]")>;
861
862def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
863             (instregex "^EOR(BT|TB)_ZZZ_[BHSD]")>;
864
865// Max/min, basic and pairwise
866def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^[SU](MAX|MIN)_ZI_[BHSD]",
867                                           "^[SU](MAX|MIN)P?_(ZPmZ|ZPZZ)_[BHSD]")>;
868
869// Matching operations
870def : InstRW<[CortexA510MCWrite<7, 2, CortexA510UnitVALU>], (instregex "^N?MATCH_PPzZZ_[BH]")>;
871
872// Matrix multiply-accumulate
873def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instrs SMMLA_ZZZ, UMMLA_ZZZ, USMMLA_ZZZ)>;
874
875// Move prefix
876def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^MOVPRFX_ZP[mz]Z_[BHSD]",
877                                           "^MOVPRFX_ZZ")>;
878
879// Multiply, B, H, S element size
880def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^MUL_(ZI|ZPmZ|ZZZI|ZZZ|ZPZZ)_[BHS]",
881                                            "^[SU]MULH_(ZPmZ|ZZZ|ZPZZ)_[BHS]")>;
882
883// Multiply, D element size
884def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^MUL_(ZI|ZPmZ|ZZZI|ZZZ|ZPZZ)_D",
885                                            "^[SU]MULH_(ZPmZ|ZZZ|ZPZZ)_D")>;
886
887// Multiply long
888def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^[SU]MULL[BT]_ZZZI_[SD]",
889                                            "^[SU]MULL[BT]_ZZZ_[HSD]")>;
890
891// Multiply accumulate, B, H, S element size
892def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^ML[AS]_(ZZZI|ZPZZZ)_[BHS]",
893                                            "^(ML[AS]|MAD|MSB)_ZPmZZ_[BHS]")>;
894
895// Multiply accumulate, D element size
896def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^ML[AS]_(ZZZI|ZPZZZ)_D",
897                                            "^(ML[AS]|MAD|MSB)_ZPmZZ_D")>;
898
899// Multiply accumulate long
900def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^[SU]ML[AS]L[BT]_ZZZ_[HSD]",
901                                            "^[SU]ML[AS]L[BT]_ZZZI_[SD]")>;
902
903// Multiply accumulate saturating doubling long regular
904def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^SQDML[AS](LB|LT|LBT)_ZZZ_[HSD]",
905                                            "^SQDML[AS](LB|LT)_ZZZI_[SD]")>;
906
907// Multiply saturating doubling high, B, H, S element size
908def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^SQDMULH_ZZZ_[BHS]",
909                                            "^SQDMULH_ZZZI_[HS]")>;
910
911// Multiply saturating doubling high, D element size
912def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instrs SQDMULH_ZZZ_D, SQDMULH_ZZZI_D)>;
913
914// Multiply saturating doubling long
915def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^SQDMULL[BT]_ZZZ_[HSD]",
916                                            "^SQDMULL[BT]_ZZZI_[SD]")>;
917
918// Multiply saturating rounding doubling regular/complex accumulate, B, H, S
919// element size
920def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^SQRDML[AS]H_ZZZ_[BHS]",
921                                            "^SQRDCMLAH_ZZZ_[BHS]",
922                                            "^SQRDML[AS]H_ZZZI_[HS]",
923                                            "^SQRDCMLAH_ZZZI_[HS]")>;
924
925// Multiply saturating rounding doubling regular/complex accumulate, D element
926// size
927def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^SQRDML[AS]H_ZZZI?_D",
928                                            "^SQRDCMLAH_ZZZ_D")>;
929
930// Multiply saturating rounding doubling regular/complex, B, H, S element size
931def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^SQRDMULH_ZZZ_[BHS]",
932                                            "^SQRDMULH_ZZZI_[HS]")>;
933
934// Multiply saturating rounding doubling regular/complex, D element size
935def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^SQRDMULH_ZZZI?_D")>;
936
937// Multiply/multiply long, (8x8) polynomial
938def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^PMUL_ZZZ_B")>;
939
940def : InstRW<[CortexA510Write<6, CortexA510UnitVMC>], (instregex "^PMULL[BT]_ZZZ_[HDQ]")>;
941
942
943// Predicate counting vector
944def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
945             (instregex "^(DEC|INC|SQDEC|SQINC|UQDEC|UQINC)[HWD]_ZPiI")>;
946
947// Reciprocal estimate
948def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^URECPE_ZPmZ_S", "^URSQRTE_ZPmZ_S")>;
949
950// Reduction, arithmetic, B form
951def : InstRW<[CortexA510Write<4, CortexA510UnitVALU0>], (instregex "^[SU](ADD|MAX|MIN)V_VPZ_B")>;
952
953// Reduction, arithmetic, H form
954def : InstRW<[CortexA510Write<4, CortexA510UnitVALU0>], (instregex "^[SU](ADD|MAX|MIN)V_VPZ_H")>;
955
956// Reduction, arithmetic, S form
957def : InstRW<[CortexA510Write<4, CortexA510UnitVALU0>], (instregex "^[SU](ADD|MAX|MIN)V_VPZ_S")>;
958
959// Reduction, arithmetic, D form
960def : InstRW<[CortexA510Write<4, CortexA510UnitVALU0>], (instregex "^[SU](ADD|MAX|MIN)V_VPZ_D")>;
961
962// Reduction, logical
963def : InstRW<[CortexA510Write<4, CortexA510UnitVALU0>], (instregex "^(ANDV|EORV|ORV)_VPZ_[BHSD]")>;
964
965// Reverse, vector
966def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^REV_ZZ_[BHSD]",
967                                           "^REVB_ZPmZ_[HSD]",
968                                           "^REVH_ZPmZ_[SD]",
969                                           "^REVW_ZPmZ_D")>;
970
971// Select, vector form
972def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^SEL_ZPZZ_[BHSD]")>;
973
974// Table lookup
975def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^TBL_ZZZZ?_[BHSD]")>;
976
977// Table lookup extension
978def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^TBX_ZZZ_[BHSD]")>;
979
980// Transpose, vector form
981def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^TRN[12]_ZZZ_[BHSDQ]")>;
982
983// Unpack and extend
984def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^[SU]UNPK(HI|LO)_ZZ_[HSD]")>;
985
986// Zip/unzip
987def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^(UZP|ZIP)[12]_ZZZ_[BHSDQ]")>;
988
989// SVE floating-point instructions
990// -----------------------------------------------------------------------------
991
992// Floating point absolute value/difference
993def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FAB[SD]_ZPmZ_[HSD]",
994                                                                  "^FAB[SD]_ZPZZ_[HSD]")>;
995
996// Floating point arithmetic
997def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^F(ADD|SUB)_(ZPm[IZ]|ZZZ|ZPZI|ZPZZ)_[HSD]",
998                                           "^FADDP_ZPmZZ_[HSD]",
999                                           "^FNEG_ZPmZ_[HSD]",
1000                                           "^FSUBR_(ZPm[IZ]|ZPZ[IZ])_[HSD]")>;
1001
1002// Floating point associative add, F16
1003def : InstRW<[CortexA510MCWrite<32, 29, CortexA510UnitVALU>], (instrs FADDA_VPZ_H)>;
1004
1005// Floating point associative add, F32
1006def : InstRW<[CortexA510MCWrite<16, 13, CortexA510UnitVALU>], (instrs FADDA_VPZ_S)>;
1007
1008// Floating point associative add, F64
1009def : InstRW<[CortexA510MCWrite<8, 5, CortexA510UnitVALU>], (instrs FADDA_VPZ_D)>;
1010
1011// Floating point compare
1012def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FACG[ET]_PPzZZ_[HSD]",
1013                                            "^FCM(EQ|GE|GT|NE)_PPzZ[0Z]_[HSD]",
1014                                            "^FCM(LE|LT)_PPzZ0_[HSD]",
1015                                            "^FCMUO_PPzZZ_[HSD]")>;
1016
1017// Floating point complex add
1018def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FCADD_ZPmZ_[HSD]")>;
1019
1020// Floating point complex multiply add
1021def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FCMLA_ZPmZZ_[HSD]",
1022                                           "^FCMLA_ZZZI_[HS]")>;
1023
1024// Floating point convert, long or narrow (F16 to F32 or F32 to F16)
1025def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FCVT_ZPmZ_(HtoS|StoH)",
1026                                            "^FCVTLT_ZPmZ_HtoS",
1027                                            "^FCVTNT_ZPmZ_StoH")>;
1028
1029// Floating point convert, long or narrow (F16 to F64, F32 to F64, F64 to F32
1030// or F64 to F16)
1031def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FCVT_ZPmZ_(HtoD|StoD|DtoS|DtoH)",
1032                                            "^FCVTLT_ZPmZ_StoD",
1033                                            "^FCVTNT_ZPmZ_DtoS")>;
1034
1035// Floating point convert, round to odd
1036def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FCVTX_ZPmZ_DtoS", "FCVTXNT_ZPmZ_DtoS")>;
1037
1038// Floating point base2 log, F16
1039def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FLOGB_(ZPmZ|ZPZZ)_H")>;
1040
1041// Floating point base2 log, F32
1042def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FLOGB_(ZPmZ|ZPZZ)_S")>;
1043
1044// Floating point base2 log, F64
1045def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FLOGB_(ZPmZ|ZPZZ)_D")>;
1046
1047// Floating point convert to integer, F16
1048def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FCVTZ[SU]_ZPmZ_HtoH")>;
1049
1050// Floating point convert to integer, F32
1051def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FCVTZ[SU]_ZPmZ_(HtoS|StoS)")>;
1052
1053// Floating point convert to integer, F64
1054def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>],
1055             (instregex "^FCVTZ[SU]_ZPmZ_(HtoD|StoD|DtoS|DtoD)")>;
1056
1057// Floating point copy
1058def : InstRW<[CortexA510Write<3, CortexA510UnitVALU0>], (instregex "^FCPY_ZPmI_[HSD]",
1059                                           "^FDUP_ZI_[HSD]")>;
1060
1061// Floating point divide, F16
1062def : InstRW<[CortexA510MCWrite<8, 5, CortexA510UnitVMC>], (instregex "^FDIVR?_(ZPmZ|ZPZZ)_H")>;
1063
1064// Floating point divide, F32
1065def : InstRW<[CortexA510MCWrite<13, 10, CortexA510UnitVMC>], (instregex "^FDIVR?_(ZPmZ|ZPZZ)_S")>;
1066
1067// Floating point divide, F64
1068def : InstRW<[CortexA510MCWrite<22, 19, CortexA510UnitVMC>], (instregex "^FDIVR?_(ZPmZ|ZPZZ)_D")>;
1069
1070// Floating point min/max pairwise
1071def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^F(MAX|MIN)(NM)?P_ZPmZZ_[HSD]")>;
1072
1073// Floating point min/max
1074def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^F(MAX|MIN)(NM)?_(ZPm[IZ]|ZPZZ|ZPZI)_[HSD]")>;
1075
1076// Floating point multiply
1077def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^(FSCALE|FMULX)_(ZPmZ|ZPZZ)_[HSD]",
1078                                           "^FMUL_(ZPm[IZ]|ZZZI?|ZPZI|ZPZZ)_[HSD]")>;
1079
1080// Floating point multiply accumulate
1081def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>],
1082             (instregex "^FML[AS]_(ZPmZZ|ZZZI|ZPZZZ)_[HSD]",
1083                        "^(FMAD|FNMAD|FNML[AS]|FN?MSB)_(ZPmZZ|ZPZZZ)_[HSD]")>;
1084
1085// Floating point multiply add/sub accumulate long
1086def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FML[AS]L[BT]_ZZZI?_SHH")>;
1087
1088// Floating point reciprocal estimate, F16
1089def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FRECPE_ZZ_H", "^FRECPX_ZPmZ_H",
1090                                         "^FRSQRTE_ZZ_H")>;
1091
1092// Floating point reciprocal estimate, F32
1093def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FRECPE_ZZ_S", "^FRECPX_ZPmZ_S",
1094                                         "^FRSQRTE_ZZ_S")>;
1095// Floating point reciprocal estimate, F64
1096def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>],(instregex "^FRECPE_ZZ_D", "^FRECPX_ZPmZ_D",
1097                                         "^FRSQRTE_ZZ_D")>;
1098
1099// Floating point reciprocal step
1100def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^F(RECPS|RSQRTS)_ZZZ_[HSD]")>;
1101
1102// Floating point reduction, F16
1103def : InstRW<[CortexA510Write<4, CortexA510UnitVALU0>],
1104             (instregex "^(FMAXNMV|FMAXV|FMINNMV|FMINV)_VPZ_[HSD]")>;
1105
1106// Floating point reduction, F32
1107def : InstRW<[CortexA510MCWrite<12, 11, CortexA510UnitVALU0>],
1108             (instregex "^FADDV_VPZ_H")>;
1109
1110def : InstRW<[CortexA510MCWrite<8, 5, CortexA510UnitVALU0>],
1111             (instregex "^FADDV_VPZ_S")>;
1112
1113def : InstRW<[CortexA510Write<4, CortexA510UnitVALU0>],
1114             (instregex "^FADDV_VPZ_D")>;
1115
1116
1117// Floating point round to integral, F16
1118def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FRINT[AIMNPXZ]_ZPmZ_H")>;
1119
1120// Floating point round to integral, F32
1121def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FRINT[AIMNPXZ]_ZPmZ_S")>;
1122
1123// Floating point round to integral, F64
1124def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FRINT[AIMNPXZ]_ZPmZ_D")>;
1125
1126// Floating point square root, F16
1127def : InstRW<[CortexA510MCWrite<8, 5, CortexA510UnitVMC>], (instregex "^FSQRT_ZPmZ_H")>;
1128
1129// Floating point square root, F32
1130def : InstRW<[CortexA510MCWrite<12, 9, CortexA510UnitVMC>], (instregex "^FSQRT_ZPmZ_S")>;
1131
1132// Floating point square root, F64
1133def : InstRW<[CortexA510MCWrite<22, 19, CortexA510UnitVMC>], (instregex "^FSQRT_ZPmZ_D")>;
1134
1135// Floating point trigonometric exponentiation
1136def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FEXPA_ZZ_[HSD]")>;
1137
1138// Floating point trigonometric multiply add
1139def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FTMAD_ZZI_[HSD]")>;
1140
1141// Floating point trigonometric, miscellaneous
1142def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^FTSMUL_ZZZ_[HSD]")>;
1143def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instregex "^FTSSEL_ZZZ_[HSD]")>;
1144
1145
1146// SVE BFloat16 (BF16) instructions
1147// -----------------------------------------------------------------------------
1148
1149// Convert, F32 to BF16
1150def : InstRW<[CortexA510Write<4, CortexA510UnitVALU>], (instrs BFCVT_ZPmZ, BFCVTNT_ZPmZ)>;
1151
1152// Dot product
1153def : InstRW<[A510Write_10cyc_1VMAC_1VALU], (instrs BFDOT_ZZI, BFDOT_ZZZ)>;
1154
1155// Matrix multiply accumulate
1156def : InstRW<[A510Write_15cyc_1VMAC_1VALU], (instrs BFMMLA_ZZZ)>;
1157
1158// Multiply accumulate long
1159def : InstRW<[CortexA510Write<4, CortexA510UnitVMAC>], (instregex "^BFMLAL[BT]_ZZZ(I)?")>;
1160
1161// SVE Load instructions
1162// -----------------------------------------------------------------------------
1163
1164// Load vector
1165def : InstRW<[CortexA510Write<3, CortexA510UnitLd>], (instrs LDR_ZXI)>;
1166
1167// Load predicate
1168def : InstRW<[CortexA510Write<3, CortexA510UnitLdSt>], (instrs LDR_PXI)>;
1169
1170// Contiguous load, scalar + imm
1171def : InstRW<[CortexA510Write<3, CortexA510UnitLd>], (instregex "^LD1[BHWD]_IMM$",
1172                                           "^LD1S?B_[HSD]_IMM$",
1173                                           "^LD1S?H_[SD]_IMM$",
1174                                           "^LD1S?W_D_IMM$" )>;
1175// Contiguous load, scalar + scalar
1176def : InstRW<[CortexA510Write<3, CortexA510UnitLd>], (instregex "^LD1[BHWD]$",
1177                                             "^LD1S?B_[HSD]$",
1178                                             "^LD1S?H_[SD]$",
1179                                             "^LD1S?W_D$" )>;
1180
1181// Contiguous load broadcast, scalar + imm
1182def : InstRW<[CortexA510Write<3, CortexA510UnitLd>], (instregex "^LD1R[BHWD]_IMM$",
1183                                           "^LD1RSW_IMM$",
1184                                           "^LD1RS?B_[HSD]_IMM$",
1185                                           "^LD1RS?H_[SD]_IMM$",
1186                                           "^LD1RS?W_D_IMM$",
1187                                           "^LD1RQ_[BHWD]_IMM$")>;
1188
1189// Contiguous load broadcast, scalar + scalar
1190def : InstRW<[CortexA510Write<3, CortexA510UnitLdSt>], (instregex "^LD1RQ_[BHWD]$")>;
1191
1192// Non temporal load, scalar + imm
1193def : InstRW<[CortexA510Write<3, CortexA510UnitLdSt>], (instregex "^LDNT1[BHWD]_ZRI$")>;
1194
1195// Non temporal load, scalar + scalar
1196def : InstRW<[CortexA510Write<3, CortexA510UnitLdSt>], (instregex "^LDNT1[BHWD]_ZRR$")>;
1197
1198// Non temporal gather load, vector + scalar 32-bit element size
1199def : InstRW<[CortexA510MCWrite<9, 9, CortexA510UnitLdSt>], (instregex "^LDNT1[BHW]_ZZR_S_REAL$",
1200                                              "^LDNT1S[BH]_ZZR_S_REAL$")>;
1201
1202// Non temporal gather load, vector + scalar 64-bit element size
1203def : InstRW<[CortexA510MCWrite<7, 7, CortexA510UnitLdSt>], (instregex "^LDNT1S?[BHW]_ZZR_D_REAL$")>;
1204def : InstRW<[CortexA510MCWrite<7, 7, CortexA510UnitLdSt>], (instrs LDNT1D_ZZR_D_REAL)>;
1205
1206// Contiguous first faulting load, scalar + scalar
1207def : InstRW<[CortexA510Write<3, CortexA510UnitLd>], (instregex "^LDFF1[BHWD]_REAL$",
1208                                              "^LDFF1S?B_[HSD]_REAL$",
1209                                              "^LDFF1S?H_[SD]_REAL$",
1210                                              "^LDFF1S?W_D_REAL$")>;
1211
1212// Contiguous non faulting load, scalar + imm
1213def : InstRW<[CortexA510Write<3, CortexA510UnitLd>], (instregex "^LDNF1[BHWD]_IMM_REAL$",
1214                                           "^LDNF1S?B_[HSD]_IMM_REAL$",
1215                                           "^LDNF1S?H_[SD]_IMM_REAL$",
1216                                           "^LDNF1S?W_D_IMM_REAL$")>;
1217
1218// Contiguous Load two structures to two vectors, scalar + imm
1219def : InstRW<[CortexA510MCWrite<3, 1, CortexA510UnitLdSt>], (instregex "^LD2[BHWD]_IMM$")>;
1220
1221// Contiguous Load two structures to two vectors, scalar + scalar
1222def : InstRW<[CortexA510MCWrite<3, 2, CortexA510UnitLdSt>], (instregex "^LD2[BHWD]$")>;
1223
1224// Contiguous Load three structures to three vectors, scalar + imm
1225def : InstRW<[CortexA510MCWrite<5, 3, CortexA510UnitLdSt>], (instregex "^LD3[BHWD]_IMM$")>;
1226
1227// Contiguous Load three structures to three vectors, scalar + scalar
1228def : InstRW<[CortexA510MCWrite<5, 3, CortexA510UnitLdSt>], (instregex "^LD3[BHWD]$")>;
1229
1230// Contiguous Load four structures to four vectors, scalar + imm
1231def : InstRW<[CortexA510MCWrite<5, 3, CortexA510UnitLdSt>], (instregex "^LD4[BHWD]_IMM$")>;
1232
1233// Contiguous Load four structures to four vectors, scalar + scalar
1234def : InstRW<[CortexA510MCWrite<5, 3, CortexA510UnitLdSt>], (instregex "^LD4[BHWD]$")>;
1235
1236// Gather load, vector + imm, 32-bit element size
1237def : InstRW<[CortexA510MCWrite<9, 9, CortexA510UnitLdSt>], (instregex "^GLD(FF)?1S?[BH]_S_IMM_REAL$",
1238                                              "^GLD(FF)?1W_IMM_REAL$")>;
1239
1240// Gather load, vector + imm, 64-bit element size
1241def : InstRW<[CortexA510MCWrite<7, 7, CortexA510UnitLdSt>], (instregex "^GLD(FF)?1S?[BHW]_D_IMM_REAL$",
1242                                              "^GLD(FF)?1D_IMM_REAL$")>;
1243
1244// Gather load, 64-bit element size
1245def : InstRW<[CortexA510MCWrite<7, 7, CortexA510UnitLdSt>],
1246             (instregex "^GLD(FF)?1S?[BHW]_D_[SU]XTW_(SCALED_)?REAL$",
1247                        "^GLD(FF)?1S?[BHW]_D_(SCALED_)?REAL$",
1248                        "^GLD(FF)?1D_[SU]XTW_(SCALED_)?REAL$",
1249                        "^GLD(FF)?1D_(SCALED_)?REAL$")>;
1250
1251// Gather load, 32-bit scaled offset
1252def : InstRW<[CortexA510MCWrite<9, 9, CortexA510UnitLd>],
1253             (instregex "^GLD(FF)?1S?[HW]_S_[SU]XTW_SCALED_REAL$",
1254                        "^GLD(FF)?1W_[SU]XTW_SCALED_REAL")>;
1255
1256// Gather load, 32-bit unpacked unscaled offset
1257def : InstRW<[CortexA510MCWrite<9, 9, CortexA510UnitLd>], (instregex "^GLD(FF)?1S?[BH]_S_[SU]XTW_REAL$",
1258                                              "^GLD(FF)?1W_[SU]XTW_REAL$")>;
1259
1260def : InstRW<[CortexA510Write<0, CortexA510UnitVALU>], (instregex "^PRF(B|H|W|D).*")>;
1261// SVE Store instructions
1262// -----------------------------------------------------------------------------
1263
1264// Store from predicate reg
1265def : InstRW<[CortexA510VSt0], (instrs STR_PXI)>;
1266
1267// Store from vector reg
1268def : InstRW<[CortexA510VSt0], (instrs STR_ZXI)>;
1269
1270// Contiguous store, scalar + imm
1271def : InstRW<[CortexA510VSt0], (instregex "^ST1[BHWD]_IMM$",
1272                                                "^ST1B_[HSD]_IMM$",
1273                                                "^ST1H_[SD]_IMM$",
1274                                                "^ST1W_D_IMM$")>;
1275
1276// Contiguous store, scalar + scalar
1277def : InstRW<[CortexA510VSt0], (instregex "^ST1H(_[SD])?$")>;
1278def : InstRW<[CortexA510VSt0], (instregex "^ST1[BWD]$",
1279                                                "^ST1B_[HSD]$",
1280                                                "^ST1W_D$")>;
1281
1282// Contiguous store two structures from two vectors, scalar + imm
1283def : InstRW<[CortexA510VSt<11>], (instregex "^ST2[BHWD]_IMM$")>;
1284
1285// Contiguous store two structures from two vectors, scalar + scalar
1286def : InstRW<[CortexA510VSt<11>], (instrs ST2H)>;
1287
1288// Contiguous store two structures from two vectors, scalar + scalar
1289def : InstRW<[CortexA510VSt<11>], (instregex "^ST2[BWD]$")>;
1290
1291// Contiguous store three structures from three vectors, scalar + imm
1292def : InstRW<[CortexA510VSt<25>], (instregex "^ST3[BHW]_IMM$")>;
1293def : InstRW<[CortexA510VSt<14>], (instregex "^ST3D_IMM$")>;
1294
1295// Contiguous store three structures from three vectors, scalar + scalar
1296def : InstRW<[CortexA510VSt<25>], (instregex "^ST3[BHW]$")>;
1297def : InstRW<[CortexA510VSt<14>], (instregex "^ST3D$")>;
1298
1299// Contiguous store four structures from four vectors, scalar + imm
1300def : InstRW<[CortexA510VSt<50>], (instregex "^ST4[BHW]_IMM$")>;
1301def : InstRW<[CortexA510VSt<25>], (instregex "^ST4D_IMM$")>;
1302
1303// Contiguous store four structures from four vectors, scalar + scalar
1304def : InstRW<[CortexA510VSt<50>], (instregex "^ST4[BHW]$")>;
1305
1306// Contiguous store four structures from four vectors, scalar + scalar
1307def : InstRW<[CortexA510VSt<25>], (instregex "^ST4D$")>;
1308
1309// Non temporal store, scalar + imm
1310def : InstRW<[CortexA510VSt0], (instregex "^STNT1[BHWD]_ZRI$")>;
1311
1312// Non temporal store, scalar + scalar
1313def : InstRW<[CortexA510VSt0], (instrs STNT1H_ZRR)>;
1314def : InstRW<[CortexA510VSt0], (instregex "^STNT1[BWD]_ZRR$")>;
1315
1316// Scatter non temporal store, vector + scalar 32-bit element size
1317def : InstRW<[CortexA510VSt<9>], (instregex "^STNT1[BHW]_ZZR_S")>;
1318
1319// Scatter non temporal store, vector + scalar 64-bit element size
1320def : InstRW<[CortexA510VSt<7>], (instregex "^STNT1[BHWD]_ZZR_D")>;
1321
1322// Scatter store vector + imm 32-bit element size
1323def : InstRW<[CortexA510VSt<9>], (instregex "^SST1[BH]_S_IMM$",
1324                                                "^SST1W_IMM$")>;
1325
1326// Scatter store vector + imm 64-bit element size
1327def : InstRW<[CortexA510VSt<7>], (instregex "^SST1[BHW]_D_IMM$",
1328                                                "^SST1D_IMM$")>;
1329
1330// Scatter store, 32-bit scaled offset
1331def : InstRW<[CortexA510VSt<8>],
1332             (instregex "^SST1(H_S|W)_[SU]XTW_SCALED$")>;
1333
1334// Scatter store, 32-bit unpacked unscaled offset
1335def : InstRW<[CortexA510VSt<8>], (instregex "^SST1[BHW]_D_[SU]XTW$",
1336                                                "^SST1D_[SU]XTW$")>;
1337
1338// Scatter store, 32-bit unpacked scaled offset
1339def : InstRW<[CortexA510VSt<8>], (instregex "^SST1[HW]_D_[SU]XTW_SCALED$",
1340                                                "^SST1D_[SU]XTW_SCALED$")>;
1341
1342// Scatter store, 32-bit unscaled offset
1343def : InstRW<[CortexA510VSt<8>], (instregex "^SST1[BH]_S_[SU]XTW$",
1344                                                "^SST1W_[SU]XTW$")>;
1345
1346// Scatter store, 64-bit scaled offset
1347def : InstRW<[CortexA510VSt<8>], (instregex "^SST1[HW]_D_SCALED$",
1348                                                "^SST1D_SCALED$")>;
1349
1350// Scatter store, 64-bit unscaled offset
1351def : InstRW<[CortexA510VSt<8>], (instregex "^SST1[BHW]_D$",
1352                                                "^SST1D$")>;
1353
1354// SVE Miscellaneous instructions
1355// -----------------------------------------------------------------------------
1356
1357// Read first fault register, unpredicated
1358def : InstRW<[CortexA510Write<1, CortexA510UnitALU>], (instrs RDFFR_P_REAL)>;
1359
1360// Read first fault register, predicated
1361def : InstRW<[CortexA510Write<3, CortexA510UnitALU0>], (instrs RDFFR_PPz_REAL)>;
1362
1363// Read first fault register and set flags
1364def : InstRW<[CortexA510Write<3, CortexA510UnitALU0>], (instrs RDFFRS_PPz)>;
1365
1366// Set first fault register
1367// Write to first fault register
1368def : InstRW<[CortexA510Write<1, CortexA510UnitALU>], (instrs SETFFR, WRFFR)>;
1369
1370// SVE Cryptographic instructions
1371// -----------------------------------------------------------------------------
1372
1373// Crypto AES ops
1374def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^AES[DE]_ZZZ_B$",
1375                                           "^AESI?MC_ZZ_B$")>;
1376
1377// Crypto SHA3 ops
1378def : InstRW<[CortexA510Write<3, CortexA510UnitVALU>], (instregex "^(BCAX|EOR3)_ZZZZ$",
1379                                            "^XAR_ZZZI_[BHSD]$")>;
1380
1381def : InstRW<[CortexA510MC_RC0Write<8, CortexA510UnitVMC>], (instregex "^RAX1_ZZZ_D$")>;
1382
1383// Crypto SM4 ops
1384def : InstRW<[CortexA510MC_RC0Write<8, CortexA510UnitVMC>], (instregex "^SM4E(KEY)?_ZZZ_S$")>;
1385
1386}
1387