1 //=- ARMScheduleA8.td - ARM Cortex-A8 Scheduling Definitions -*- tablegen -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the itinerary class data for the ARM Cortex A8 processors.
12 //===----------------------------------------------------------------------===//
15 // Scheduling information derived from "Cortex-A8 Technical Reference Manual".
17 def A8_Pipe0 : FuncUnit; // pipeline 0
18 def A8_Pipe1 : FuncUnit; // pipeline 1
19 def A8_LSPipe : FuncUnit; // Load / store pipeline
20 def A8_NPipe : FuncUnit; // NEON ALU/MUL pipe
21 def A8_NLSPipe : FuncUnit; // NEON LS pipe
23 // Dual issue pipeline represented by A8_Pipe0 | A8_Pipe1
25 def CortexA8Itineraries : ProcessorItineraries<
26 [A8_Pipe0, A8_Pipe1, A8_LSPipe, A8_NPipe, A8_NLSPipe],
28 // Two fully-pipelined integer ALU pipelines
31 InstrItinData<IIC_iALUx , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
33 // Binary Instructions that produce a result
34 InstrItinData<IIC_iALUi ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
35 InstrItinData<IIC_iALUr ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 2]>,
36 InstrItinData<IIC_iALUsi,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
37 InstrItinData<IIC_iALUsir,[InstrStage<1,[A8_Pipe0, A8_Pipe1]>], [2, 1, 2]>,
38 InstrItinData<IIC_iALUsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1, 1]>,
40 // Bitwise Instructions that produce a result
41 InstrItinData<IIC_iBITi ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
42 InstrItinData<IIC_iBITr ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 2]>,
43 InstrItinData<IIC_iBITsi,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
44 InstrItinData<IIC_iBITsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1, 1]>,
46 // Unary Instructions that produce a result
47 InstrItinData<IIC_iUNAr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
48 InstrItinData<IIC_iUNAsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
50 // Zero and sign extension instructions
51 InstrItinData<IIC_iEXTr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
52 InstrItinData<IIC_iEXTAr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
53 InstrItinData<IIC_iEXTAsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>],[2, 2, 1, 1]>,
55 // Compare instructions
56 InstrItinData<IIC_iCMPi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
57 InstrItinData<IIC_iCMPr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
58 InstrItinData<IIC_iCMPsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
59 InstrItinData<IIC_iCMPsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
62 InstrItinData<IIC_iTSTi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
63 InstrItinData<IIC_iTSTr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
64 InstrItinData<IIC_iTSTsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
65 InstrItinData<IIC_iTSTsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
67 // Move instructions, unconditional
68 InstrItinData<IIC_iMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1]>,
69 InstrItinData<IIC_iMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
70 InstrItinData<IIC_iMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
71 InstrItinData<IIC_iMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1, 1]>,
72 InstrItinData<IIC_iMOVix2,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
73 InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
75 // Move instructions, conditional
76 InstrItinData<IIC_iCMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
77 InstrItinData<IIC_iCMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
78 InstrItinData<IIC_iCMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
79 InstrItinData<IIC_iCMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
82 InstrItinData<IIC_iMVNi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1]>,
83 InstrItinData<IIC_iMVNr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
84 InstrItinData<IIC_iMVNsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
85 InstrItinData<IIC_iMVNsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1, 1]>,
87 // Integer multiply pipeline
88 // Result written in E5, but that is relative to the last cycle of multicycle,
89 // so we use 6 for those cases
91 InstrItinData<IIC_iMUL16 , [InstrStage<1, [A8_Pipe0]>], [5, 1, 1]>,
92 InstrItinData<IIC_iMAC16 , [InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
93 InstrItinData<IIC_iMUL32 , [InstrStage<2, [A8_Pipe0]>], [6, 1, 1]>,
94 InstrItinData<IIC_iMAC32 , [InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
95 InstrItinData<IIC_iMUL64 , [InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
96 InstrItinData<IIC_iMAC64 , [InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
98 // Integer load pipeline
101 InstrItinData<IIC_iLoad_i , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
102 InstrStage<1, [A8_LSPipe]>], [3, 1]>,
103 InstrItinData<IIC_iLoad_bh_i, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
104 InstrStage<1, [A8_LSPipe]>], [3, 1]>,
105 InstrItinData<IIC_iLoad_d_i, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
106 InstrStage<1, [A8_LSPipe]>], [3, 1]>,
109 InstrItinData<IIC_iLoad_r , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
110 InstrStage<1, [A8_LSPipe]>], [3, 1, 1]>,
111 InstrItinData<IIC_iLoad_bh_r, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
112 InstrStage<1, [A8_LSPipe]>], [3, 1, 1]>,
113 InstrItinData<IIC_iLoad_d_r , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
114 InstrStage<1, [A8_LSPipe]>], [3, 1, 1]>,
116 // Scaled register offset, issues over 2 cycles
117 // FIXME: lsl by 2 takes 1 cycle.
118 InstrItinData<IIC_iLoad_si , [InstrStage<2, [A8_Pipe0, A8_Pipe1]>,
119 InstrStage<1, [A8_LSPipe]>], [4, 1, 1]>,
120 InstrItinData<IIC_iLoad_bh_si,[InstrStage<2, [A8_Pipe0, A8_Pipe1]>,
121 InstrStage<1, [A8_LSPipe]>], [4, 1, 1]>,
123 // Immediate offset with update
124 InstrItinData<IIC_iLoad_iu , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
125 InstrStage<1, [A8_LSPipe]>], [3, 2, 1]>,
126 InstrItinData<IIC_iLoad_bh_iu,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
127 InstrStage<1, [A8_LSPipe]>], [3, 2, 1]>,
129 // Register offset with update
130 InstrItinData<IIC_iLoad_ru , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
131 InstrStage<1, [A8_LSPipe]>], [3, 2, 1, 1]>,
132 InstrItinData<IIC_iLoad_bh_ru,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
133 InstrStage<1, [A8_LSPipe]>], [3, 2, 1, 1]>,
134 InstrItinData<IIC_iLoad_d_ru, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
135 InstrStage<1, [A8_LSPipe]>], [3, 2, 1, 1]>,
137 // Scaled register offset with update, issues over 2 cycles
138 InstrItinData<IIC_iLoad_siu , [InstrStage<2, [A8_Pipe0, A8_Pipe1]>,
139 InstrStage<1, [A8_LSPipe]>], [4, 3, 1, 1]>,
140 InstrItinData<IIC_iLoad_bh_siu,[InstrStage<2, [A8_Pipe0, A8_Pipe1]>,
141 InstrStage<1, [A8_LSPipe]>], [4, 3, 1, 1]>,
143 // Load multiple, def is the 5th operand. Pipeline 0 only.
144 // FIXME: A8_LSPipe cycle time is dynamic, this assumes 3 to 4 registers.
145 InstrItinData<IIC_iLoad_m , [InstrStage<1, [A8_Pipe0]>,
146 InstrStage<2, [A8_LSPipe]>], [1, 1, 1, 1, 3]>,
148 // Load multiple + update, defs are the 1st and 5th operands.
149 InstrItinData<IIC_iLoad_mu , [InstrStage<1, [A8_Pipe0]>,
150 InstrStage<3, [A8_LSPipe]>], [2, 1, 1, 1, 3]>,
152 // Load multiple plus branch
153 InstrItinData<IIC_iLoad_mBr, [InstrStage<1, [A8_Pipe0]>,
154 InstrStage<3, [A8_LSPipe]>,
155 InstrStage<1, [A8_Pipe0, A8_Pipe1]>],
158 // Pop, def is the 3rd operand.
159 InstrItinData<IIC_iPop , [InstrStage<1, [A8_Pipe0]>,
160 InstrStage<3, [A8_LSPipe]>], [1, 1, 3]>,
162 // Push, def is the 3th operand.
163 InstrItinData<IIC_iPop_Br, [InstrStage<1, [A8_Pipe0]>,
164 InstrStage<3, [A8_LSPipe]>,
165 InstrStage<1, [A8_Pipe0, A8_Pipe1]>],
169 // iLoadi + iALUr for t2LDRpci_pic.
170 InstrItinData<IIC_iLoadiALU, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
171 InstrStage<1, [A8_LSPipe]>,
172 InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [4, 1]>,
175 // Integer store pipeline
178 InstrItinData<IIC_iStore_i , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
179 InstrStage<1, [A8_LSPipe]>], [3, 1]>,
180 InstrItinData<IIC_iStore_bh_i,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
181 InstrStage<1, [A8_LSPipe]>], [3, 1]>,
182 InstrItinData<IIC_iStore_d_i, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
183 InstrStage<1, [A8_LSPipe]>], [3, 1]>,
186 InstrItinData<IIC_iStore_r , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
187 InstrStage<1, [A8_LSPipe]>], [3, 1, 1]>,
188 InstrItinData<IIC_iStore_bh_r,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
189 InstrStage<1, [A8_LSPipe]>], [3, 1, 1]>,
190 InstrItinData<IIC_iStore_d_r, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
191 InstrStage<1, [A8_LSPipe]>], [3, 1, 1]>,
193 // Scaled register offset, issues over 2 cycles
194 InstrItinData<IIC_iStore_si , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
195 InstrStage<2, [A8_LSPipe]>], [3, 1, 1]>,
196 InstrItinData<IIC_iStore_bh_si,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
197 InstrStage<2, [A8_LSPipe]>], [3, 1, 1]>,
199 // Immediate offset with update
200 InstrItinData<IIC_iStore_iu , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
201 InstrStage<1, [A8_LSPipe]>], [2, 3, 1]>,
202 InstrItinData<IIC_iStore_bh_iu,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
203 InstrStage<1, [A8_LSPipe]>], [2, 3, 1]>,
205 // Register offset with update
206 InstrItinData<IIC_iStore_ru , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
207 InstrStage<1, [A8_LSPipe]>], [2, 3, 1, 1]>,
208 InstrItinData<IIC_iStore_bh_ru,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
209 InstrStage<1, [A8_LSPipe]>], [2, 3, 1, 1]>,
210 InstrItinData<IIC_iStore_d_ru, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
211 InstrStage<1, [A8_LSPipe]>], [2, 3, 1, 1]>,
213 // Scaled register offset with update, issues over 2 cycles
214 InstrItinData<IIC_iStore_siu, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
215 InstrStage<2, [A8_LSPipe]>], [3, 3, 1, 1]>,
216 InstrItinData<IIC_iStore_bh_siu,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
217 InstrStage<2, [A8_LSPipe]>], [3, 3, 1, 1]>,
219 // Store multiple. Pipeline 0 only.
220 // FIXME: A8_LSPipe cycle time is dynamic, this assumes 3 to 4 registers.
221 InstrItinData<IIC_iStore_m , [InstrStage<1, [A8_Pipe0]>,
222 InstrStage<2, [A8_LSPipe]>]>,
224 // Store multiple + update
225 InstrItinData<IIC_iStore_mu, [InstrStage<1, [A8_Pipe0]>,
226 InstrStage<2, [A8_LSPipe]>], [2]>,
230 // no delay slots, so the latency of a branch is unimportant
231 InstrItinData<IIC_Br , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
234 // Issue through integer pipeline, and execute in NEON unit. We assume
235 // RunFast mode so that NFP pipeline is used for single-precision when
238 // FP Special Register to Integer Register File Move
239 InstrItinData<IIC_fpSTAT , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
240 InstrStage<1, [A8_NLSPipe]>]>,
242 // Single-precision FP Unary
243 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
244 InstrStage<1, [A8_NPipe]>], [7, 1]>,
246 // Double-precision FP Unary
247 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
248 InstrStage<4, [A8_NPipe], 0>,
249 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
251 // Single-precision FP Compare
252 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
253 InstrStage<1, [A8_NPipe]>], [1, 1]>,
255 // Double-precision FP Compare
256 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
257 InstrStage<4, [A8_NPipe], 0>,
258 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
260 // Single to Double FP Convert
261 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
262 InstrStage<7, [A8_NPipe], 0>,
263 InstrStage<7, [A8_NLSPipe]>], [7, 1]>,
265 // Double to Single FP Convert
266 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
267 InstrStage<5, [A8_NPipe], 0>,
268 InstrStage<5, [A8_NLSPipe]>], [5, 1]>,
270 // Single-Precision FP to Integer Convert
271 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
272 InstrStage<1, [A8_NPipe]>], [7, 1]>,
274 // Double-Precision FP to Integer Convert
275 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
276 InstrStage<8, [A8_NPipe], 0>,
277 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
279 // Integer to Single-Precision FP Convert
280 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
281 InstrStage<1, [A8_NPipe]>], [7, 1]>,
283 // Integer to Double-Precision FP Convert
284 InstrItinData<IIC_fpCVTID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
285 InstrStage<8, [A8_NPipe], 0>,
286 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
288 // Single-precision FP ALU
289 InstrItinData<IIC_fpALU32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
290 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
292 // Double-precision FP ALU
293 InstrItinData<IIC_fpALU64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
294 InstrStage<9, [A8_NPipe], 0>,
295 InstrStage<9, [A8_NLSPipe]>], [9, 1, 1]>,
297 // Single-precision FP Multiply
298 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
299 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
301 // Double-precision FP Multiply
302 InstrItinData<IIC_fpMUL64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
303 InstrStage<11, [A8_NPipe], 0>,
304 InstrStage<11, [A8_NLSPipe]>], [11, 1, 1]>,
306 // Single-precision FP MAC
307 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
308 InstrStage<1, [A8_NPipe]>], [7, 2, 1, 1]>,
310 // Double-precision FP MAC
311 InstrItinData<IIC_fpMAC64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
312 InstrStage<19, [A8_NPipe], 0>,
313 InstrStage<19, [A8_NLSPipe]>], [19, 2, 1, 1]>,
315 // Single-precision FP DIV
316 InstrItinData<IIC_fpDIV32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
317 InstrStage<20, [A8_NPipe], 0>,
318 InstrStage<20, [A8_NLSPipe]>], [20, 1, 1]>,
320 // Double-precision FP DIV
321 InstrItinData<IIC_fpDIV64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
322 InstrStage<29, [A8_NPipe], 0>,
323 InstrStage<29, [A8_NLSPipe]>], [29, 1, 1]>,
325 // Single-precision FP SQRT
326 InstrItinData<IIC_fpSQRT32, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
327 InstrStage<19, [A8_NPipe], 0>,
328 InstrStage<19, [A8_NLSPipe]>], [19, 1]>,
330 // Double-precision FP SQRT
331 InstrItinData<IIC_fpSQRT64, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
332 InstrStage<29, [A8_NPipe], 0>,
333 InstrStage<29, [A8_NLSPipe]>], [29, 1]>,
335 // Single-precision FP Load
336 InstrItinData<IIC_fpLoad32, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
337 InstrStage<1, [A8_NLSPipe]>,
338 InstrStage<1, [A8_LSPipe]>],
341 // Double-precision FP Load
342 InstrItinData<IIC_fpLoad64, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
343 InstrStage<1, [A8_NLSPipe]>,
344 InstrStage<1, [A8_LSPipe]>],
348 // FIXME: A8_LSPipe cycle time is dynamic, this assumes 3 to 4 registers.
349 InstrItinData<IIC_fpLoad_m, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
350 InstrStage<1, [A8_NLSPipe]>,
351 InstrStage<1, [A8_LSPipe]>,
352 InstrStage<1, [A8_NLSPipe]>,
353 InstrStage<1, [A8_LSPipe]>], [1, 1, 1, 2]>,
355 // FP Load Multiple + update
356 InstrItinData<IIC_fpLoad_mu,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
357 InstrStage<1, [A8_NLSPipe]>,
358 InstrStage<1, [A8_LSPipe]>,
359 InstrStage<1, [A8_NLSPipe]>,
360 InstrStage<1, [A8_LSPipe]>], [2, 1, 1, 1, 2]>,
362 // Single-precision FP Store
363 InstrItinData<IIC_fpStore32,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
364 InstrStage<1, [A8_NLSPipe]>,
365 InstrStage<1, [A8_LSPipe]>],
368 // Double-precision FP Store
369 InstrItinData<IIC_fpStore64,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
370 InstrStage<1, [A8_NLSPipe]>,
371 InstrStage<1, [A8_LSPipe]>],
375 InstrItinData<IIC_fpStore_m,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
376 InstrStage<1, [A8_NLSPipe]>,
377 InstrStage<1, [A8_LSPipe]>,
378 InstrStage<1, [A8_NLSPipe]>,
379 InstrStage<1, [A8_LSPipe]>], [1, 1, 1, 1]>,
381 // FP Store Multiple + update
382 InstrItinData<IIC_fpStore_mu,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
383 InstrStage<1, [A8_NLSPipe]>,
384 InstrStage<1, [A8_LSPipe]>,
385 InstrStage<1, [A8_NLSPipe]>,
386 InstrStage<1, [A8_LSPipe]>], [2, 1, 1, 1, 1]>,
389 // Issue through integer pipeline, and execute in NEON unit.
392 InstrItinData<IIC_VLD1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
393 InstrStage<1, [A8_NLSPipe]>,
394 InstrStage<1, [A8_LSPipe]>]>,
396 InstrItinData<IIC_VLD1x2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
397 InstrStage<2, [A8_NLSPipe], 1>,
398 InstrStage<2, [A8_LSPipe]>],
402 InstrItinData<IIC_VLD1x3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
403 InstrStage<3, [A8_NLSPipe], 1>,
404 InstrStage<3, [A8_LSPipe]>],
408 InstrItinData<IIC_VLD1x4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
409 InstrStage<3, [A8_NLSPipe], 1>,
410 InstrStage<3, [A8_LSPipe]>],
414 InstrItinData<IIC_VLD1u, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
415 InstrStage<1, [A8_NLSPipe], 1>,
416 InstrStage<1, [A8_LSPipe]>],
420 InstrItinData<IIC_VLD1x2u, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
421 InstrStage<2, [A8_NLSPipe], 1>,
422 InstrStage<2, [A8_LSPipe]>],
426 InstrItinData<IIC_VLD1x3u, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
427 InstrStage<3, [A8_NLSPipe], 1>,
428 InstrStage<3, [A8_LSPipe]>],
432 InstrItinData<IIC_VLD1x4u, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
433 InstrStage<3, [A8_NLSPipe], 1>,
434 InstrStage<3, [A8_LSPipe]>],
438 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
439 InstrStage<1, [A8_NLSPipe], 1>,
440 InstrStage<1, [A8_LSPipe]>],
444 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
445 InstrStage<3, [A8_NLSPipe], 1>,
446 InstrStage<3, [A8_LSPipe]>],
450 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
451 InstrStage<3, [A8_NLSPipe], 1>,
452 InstrStage<3, [A8_LSPipe]>],
456 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
457 InstrStage<1, [A8_NLSPipe], 1>,
458 InstrStage<1, [A8_LSPipe]>],
462 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
463 InstrStage<3, [A8_NLSPipe], 1>,
464 InstrStage<3, [A8_LSPipe]>],
468 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
469 InstrStage<3, [A8_NLSPipe], 1>,
470 InstrStage<3, [A8_LSPipe]>],
471 [3, 3, 2, 1, 1, 1, 1, 1]>,
474 InstrItinData<IIC_VLD3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
475 InstrStage<1, [A8_NLSPipe]>,
476 InstrStage<1, [A8_LSPipe]>], [2, 2, 2, 1]>,
479 InstrItinData<IIC_VLD4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
480 InstrStage<1, [A8_NLSPipe]>,
481 InstrStage<1, [A8_LSPipe]>], [2, 2, 2, 2, 1]>,
484 // FIXME: We don't model this instruction properly
485 InstrItinData<IIC_VST, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
486 InstrStage<1, [A8_NLSPipe]>,
487 InstrStage<1, [A8_LSPipe]>]>,
489 // Double-register FP Unary
490 InstrItinData<IIC_VUNAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
491 InstrStage<1, [A8_NPipe]>], [5, 2]>,
493 // Quad-register FP Unary
494 // Result written in N5, but that is relative to the last cycle of multicycle,
495 // so we use 6 for those cases
496 InstrItinData<IIC_VUNAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
497 InstrStage<2, [A8_NPipe]>], [6, 2]>,
499 // Double-register FP Binary
500 InstrItinData<IIC_VBIND, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
501 InstrStage<1, [A8_NPipe]>], [5, 2, 2]>,
503 // Quad-register FP Binary
504 // Result written in N5, but that is relative to the last cycle of multicycle,
505 // so we use 6 for those cases
506 InstrItinData<IIC_VBINQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
507 InstrStage<2, [A8_NPipe]>], [6, 2, 2]>,
510 InstrItinData<IIC_VMOV, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
511 InstrStage<1, [A8_NPipe]>], [1, 1]>,
514 InstrItinData<IIC_VMOVImm, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
515 InstrStage<1, [A8_NPipe]>], [3]>,
517 // Double-register Permute Move
518 InstrItinData<IIC_VMOVD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
519 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
521 // Quad-register Permute Move
522 // Result written in N2, but that is relative to the last cycle of multicycle,
523 // so we use 3 for those cases
524 InstrItinData<IIC_VMOVQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
525 InstrStage<2, [A8_NLSPipe]>], [3, 1]>,
527 // Integer to Single-precision Move
528 InstrItinData<IIC_VMOVIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
529 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
531 // Integer to Double-precision Move
532 InstrItinData<IIC_VMOVID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
533 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
535 // Single-precision to Integer Move
536 InstrItinData<IIC_VMOVSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
537 InstrStage<1, [A8_NLSPipe]>], [20, 1]>,
539 // Double-precision to Integer Move
540 InstrItinData<IIC_VMOVDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
541 InstrStage<1, [A8_NLSPipe]>], [20, 20, 1]>,
543 // Integer to Lane Move
544 InstrItinData<IIC_VMOVISL , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
545 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
547 // Vector narrow move
548 InstrItinData<IIC_VMOVN , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
549 InstrStage<1, [A8_NPipe]>], [2, 1]>,
551 // Double-register Permute
552 InstrItinData<IIC_VPERMD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
553 InstrStage<1, [A8_NLSPipe]>], [2, 2, 1, 1]>,
555 // Quad-register Permute
556 // Result written in N2, but that is relative to the last cycle of multicycle,
557 // so we use 3 for those cases
558 InstrItinData<IIC_VPERMQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
559 InstrStage<2, [A8_NLSPipe]>], [3, 3, 1, 1]>,
561 // Quad-register Permute (3 cycle issue)
562 // Result written in N2, but that is relative to the last cycle of multicycle,
563 // so we use 4 for those cases
564 InstrItinData<IIC_VPERMQ3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
565 InstrStage<1, [A8_NLSPipe]>,
566 InstrStage<1, [A8_NPipe], 0>,
567 InstrStage<2, [A8_NLSPipe]>], [4, 4, 1, 1]>,
569 // Double-register FP Multiple-Accumulate
570 InstrItinData<IIC_VMACD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
571 InstrStage<1, [A8_NPipe]>], [9, 3, 2, 2]>,
573 // Quad-register FP Multiple-Accumulate
574 // Result written in N9, but that is relative to the last cycle of multicycle,
575 // so we use 10 for those cases
576 InstrItinData<IIC_VMACQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
577 InstrStage<2, [A8_NPipe]>], [10, 3, 2, 2]>,
579 // Double-register Reciprical Step
580 InstrItinData<IIC_VRECSD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
581 InstrStage<1, [A8_NPipe]>], [9, 2, 2]>,
583 // Quad-register Reciprical Step
584 InstrItinData<IIC_VRECSQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
585 InstrStage<2, [A8_NPipe]>], [10, 2, 2]>,
587 // Double-register Integer Count
588 InstrItinData<IIC_VCNTiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
589 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
591 // Quad-register Integer Count
592 // Result written in N3, but that is relative to the last cycle of multicycle,
593 // so we use 4 for those cases
594 InstrItinData<IIC_VCNTiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
595 InstrStage<2, [A8_NPipe]>], [4, 2, 2]>,
597 // Double-register Integer Unary
598 InstrItinData<IIC_VUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
599 InstrStage<1, [A8_NPipe]>], [4, 2]>,
601 // Quad-register Integer Unary
602 InstrItinData<IIC_VUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
603 InstrStage<1, [A8_NPipe]>], [4, 2]>,
605 // Double-register Integer Q-Unary
606 InstrItinData<IIC_VQUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
607 InstrStage<1, [A8_NPipe]>], [4, 1]>,
609 // Quad-register Integer CountQ-Unary
610 InstrItinData<IIC_VQUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
611 InstrStage<1, [A8_NPipe]>], [4, 1]>,
613 // Double-register Integer Binary
614 InstrItinData<IIC_VBINiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
615 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
617 // Quad-register Integer Binary
618 InstrItinData<IIC_VBINiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
619 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
621 // Double-register Integer Binary (4 cycle)
622 InstrItinData<IIC_VBINi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
623 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
625 // Quad-register Integer Binary (4 cycle)
626 InstrItinData<IIC_VBINi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
627 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
630 // Double-register Integer Subtract
631 InstrItinData<IIC_VSUBiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
632 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
634 // Quad-register Integer Subtract
635 InstrItinData<IIC_VSUBiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
636 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
638 // Double-register Integer Subtract
639 InstrItinData<IIC_VSUBi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
640 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
642 // Quad-register Integer Subtract
643 InstrItinData<IIC_VSUBi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
644 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
646 // Double-register Integer Shift
647 InstrItinData<IIC_VSHLiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
648 InstrStage<1, [A8_NPipe]>], [3, 1, 1]>,
650 // Quad-register Integer Shift
651 InstrItinData<IIC_VSHLiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
652 InstrStage<2, [A8_NPipe]>], [4, 1, 1]>,
654 // Double-register Integer Shift (4 cycle)
655 InstrItinData<IIC_VSHLi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
656 InstrStage<1, [A8_NPipe]>], [4, 1, 1]>,
658 // Quad-register Integer Shift (4 cycle)
659 InstrItinData<IIC_VSHLi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
660 InstrStage<2, [A8_NPipe]>], [5, 1, 1]>,
662 // Double-register Integer Pair Add Long
663 InstrItinData<IIC_VPALiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
664 InstrStage<1, [A8_NPipe]>], [6, 3, 1]>,
666 // Quad-register Integer Pair Add Long
667 InstrItinData<IIC_VPALiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
668 InstrStage<2, [A8_NPipe]>], [7, 3, 1]>,
670 // Double-register Absolute Difference and Accumulate
671 InstrItinData<IIC_VABAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
672 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 1]>,
674 // Quad-register Absolute Difference and Accumulate
675 InstrItinData<IIC_VABAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
676 InstrStage<2, [A8_NPipe]>], [6, 3, 2, 1]>,
679 // Double-register Integer Multiply (.8, .16)
680 InstrItinData<IIC_VMULi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
681 InstrStage<1, [A8_NPipe]>], [6, 2, 2]>,
683 // Double-register Integer Multiply (.32)
684 InstrItinData<IIC_VMULi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
685 InstrStage<2, [A8_NPipe]>], [7, 2, 1]>,
687 // Quad-register Integer Multiply (.8, .16)
688 InstrItinData<IIC_VMULi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
689 InstrStage<2, [A8_NPipe]>], [7, 2, 2]>,
691 // Quad-register Integer Multiply (.32)
692 InstrItinData<IIC_VMULi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
693 InstrStage<1, [A8_NPipe]>,
694 InstrStage<2, [A8_NLSPipe], 0>,
695 InstrStage<3, [A8_NPipe]>], [9, 2, 1]>,
697 // Double-register Integer Multiply-Accumulate (.8, .16)
698 InstrItinData<IIC_VMACi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
699 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 2]>,
701 // Double-register Integer Multiply-Accumulate (.32)
702 InstrItinData<IIC_VMACi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
703 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 1]>,
705 // Quad-register Integer Multiply-Accumulate (.8, .16)
706 InstrItinData<IIC_VMACi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
707 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 2]>,
709 // Quad-register Integer Multiply-Accumulate (.32)
710 InstrItinData<IIC_VMACi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
711 InstrStage<1, [A8_NPipe]>,
712 InstrStage<2, [A8_NLSPipe], 0>,
713 InstrStage<3, [A8_NPipe]>], [9, 3, 2, 1]>,
715 // Double-register VEXT
716 InstrItinData<IIC_VEXTD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
717 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
719 // Quad-register VEXT
720 InstrItinData<IIC_VEXTQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
721 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
724 InstrItinData<IIC_VTB1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
725 InstrStage<2, [A8_NLSPipe]>], [3, 2, 1]>,
726 InstrItinData<IIC_VTB2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
727 InstrStage<2, [A8_NLSPipe]>], [3, 2, 2, 1]>,
728 InstrItinData<IIC_VTB3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
729 InstrStage<1, [A8_NLSPipe]>,
730 InstrStage<1, [A8_NPipe], 0>,
731 InstrStage<2, [A8_NLSPipe]>], [4, 2, 2, 3, 1]>,
732 InstrItinData<IIC_VTB4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
733 InstrStage<1, [A8_NLSPipe]>,
734 InstrStage<1, [A8_NPipe], 0>,
735 InstrStage<2, [A8_NLSPipe]>],[4, 2, 2, 3, 3, 1]>,
738 InstrItinData<IIC_VTBX1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
739 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 1]>,
740 InstrItinData<IIC_VTBX2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
741 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 2, 1]>,
742 InstrItinData<IIC_VTBX3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
743 InstrStage<1, [A8_NLSPipe]>,
744 InstrStage<1, [A8_NPipe], 0>,
745 InstrStage<2, [A8_NLSPipe]>],[4, 1, 2, 2, 3, 1]>,
746 InstrItinData<IIC_VTBX4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
747 InstrStage<1, [A8_NLSPipe]>,
748 InstrStage<1, [A8_NPipe], 0>,
749 InstrStage<2, [A8_NLSPipe]>], [4, 1, 2, 2, 3, 3, 1]>