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_Issue : FuncUnit; // issue
18 def A8_Pipe0 : FuncUnit; // pipeline 0
19 def A8_Pipe1 : FuncUnit; // pipeline 1
20 def A8_LdSt0 : FuncUnit; // pipeline 0 load/store
21 def A8_LdSt1 : FuncUnit; // pipeline 1 load/store
22 def A8_NPipe : FuncUnit; // NEON ALU/MUL pipe
23 def A8_NLSPipe : FuncUnit; // NEON LS pipe
25 // Dual issue pipeline represented by A8_Pipe0 | A8_Pipe1
27 def CortexA8Itineraries : ProcessorItineraries<
28 [A8_Issue, A8_Pipe0, A8_Pipe1, A8_LdSt0, A8_LdSt1, A8_NPipe, A8_NLSPipe],
30 // Two fully-pipelined integer ALU pipelines
33 InstrItinData<IIC_iALUx , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
35 // Binary Instructions that produce a result
36 InstrItinData<IIC_iALUi ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
37 InstrItinData<IIC_iALUr ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 2]>,
38 InstrItinData<IIC_iALUsi,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
39 InstrItinData<IIC_iALUsir,[InstrStage<1,[A8_Pipe0, A8_Pipe1]>], [2, 1, 2]>,
40 InstrItinData<IIC_iALUsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1, 1]>,
42 // Bitwise Instructions that produce a result
43 InstrItinData<IIC_iBITi ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
44 InstrItinData<IIC_iBITr ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 2]>,
45 InstrItinData<IIC_iBITsi,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
46 InstrItinData<IIC_iBITsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1, 1]>,
48 // Unary Instructions that produce a result
49 InstrItinData<IIC_iUNAr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
50 InstrItinData<IIC_iUNAsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
52 // Zero and sign extension instructions
53 InstrItinData<IIC_iEXTr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
54 InstrItinData<IIC_iEXTAr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
55 InstrItinData<IIC_iEXTAsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>],[2, 2, 1, 1]>,
57 // Compare instructions
58 InstrItinData<IIC_iCMPi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
59 InstrItinData<IIC_iCMPr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
60 InstrItinData<IIC_iCMPsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
61 InstrItinData<IIC_iCMPsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
64 InstrItinData<IIC_iTSTi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
65 InstrItinData<IIC_iTSTr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
66 InstrItinData<IIC_iTSTsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
67 InstrItinData<IIC_iTSTsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
69 // Move instructions, unconditional
70 InstrItinData<IIC_iMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1]>,
71 InstrItinData<IIC_iMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
72 InstrItinData<IIC_iMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
73 InstrItinData<IIC_iMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1, 1]>,
74 InstrItinData<IIC_iMOVix2,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
75 InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
77 // Move instructions, conditional
78 InstrItinData<IIC_iCMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
79 InstrItinData<IIC_iCMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
80 InstrItinData<IIC_iCMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
81 InstrItinData<IIC_iCMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
84 InstrItinData<IIC_iMVNi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1]>,
85 InstrItinData<IIC_iMVNr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
86 InstrItinData<IIC_iMVNsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
87 InstrItinData<IIC_iMVNsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1, 1]>,
89 // Integer multiply pipeline
90 // Result written in E5, but that is relative to the last cycle of multicycle,
91 // so we use 6 for those cases
93 InstrItinData<IIC_iMUL16 , [InstrStage<1, [A8_Pipe0]>], [5, 1, 1]>,
94 InstrItinData<IIC_iMAC16 , [InstrStage<1, [A8_Pipe1], 0>,
95 InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
96 InstrItinData<IIC_iMUL32 , [InstrStage<1, [A8_Pipe1], 0>,
97 InstrStage<2, [A8_Pipe0]>], [6, 1, 1]>,
98 InstrItinData<IIC_iMAC32 , [InstrStage<1, [A8_Pipe1], 0>,
99 InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
100 InstrItinData<IIC_iMUL64 , [InstrStage<2, [A8_Pipe1], 0>,
101 InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
102 InstrItinData<IIC_iMAC64 , [InstrStage<2, [A8_Pipe1], 0>,
103 InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
105 // Integer load pipeline
107 // loads have an extra cycle of latency, but are fully pipelined
108 // use A8_Issue to enforce the 1 load/store per cycle limit
111 InstrItinData<IIC_iLoadi , [InstrStage<1, [A8_Issue], 0>,
112 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
113 InstrStage<1, [A8_LdSt0]>], [3, 1]>,
116 InstrItinData<IIC_iLoadr , [InstrStage<1, [A8_Issue], 0>,
117 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
118 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
120 // Scaled register offset, issues over 2 cycles
121 InstrItinData<IIC_iLoadsi , [InstrStage<2, [A8_Issue], 0>,
122 InstrStage<1, [A8_Pipe0], 0>,
123 InstrStage<1, [A8_Pipe1]>,
124 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
125 InstrStage<1, [A8_LdSt0]>], [4, 1, 1]>,
127 // Immediate offset with update
128 InstrItinData<IIC_iLoadiu , [InstrStage<1, [A8_Issue], 0>,
129 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
130 InstrStage<1, [A8_LdSt0]>], [3, 2, 1]>,
132 // Register offset with update
133 InstrItinData<IIC_iLoadru , [InstrStage<1, [A8_Issue], 0>,
134 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
135 InstrStage<1, [A8_LdSt0]>], [3, 2, 1, 1]>,
137 // Scaled register offset with update, issues over 2 cycles
138 InstrItinData<IIC_iLoadsiu , [InstrStage<2, [A8_Issue], 0>,
139 InstrStage<1, [A8_Pipe0], 0>,
140 InstrStage<1, [A8_Pipe1]>,
141 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
142 InstrStage<1, [A8_LdSt0]>], [4, 3, 1, 1]>,
145 InstrItinData<IIC_iLoadm , [InstrStage<2, [A8_Issue], 0>,
146 InstrStage<2, [A8_Pipe0], 0>,
147 InstrStage<2, [A8_Pipe1]>,
148 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
149 InstrStage<1, [A8_LdSt0]>]>,
152 // Load multiple plus branch
153 InstrItinData<IIC_iLoadmBr , [InstrStage<2, [A8_Issue], 0>,
154 InstrStage<2, [A8_Pipe0], 0>,
155 InstrStage<2, [A8_Pipe1]>,
156 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
157 InstrStage<1, [A8_LdSt0]>,
158 InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
161 // iLoadi + iALUr for t2LDRpci_pic.
162 InstrItinData<IIC_iLoadiALU, [InstrStage<1, [A8_Issue], 0>,
163 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
164 InstrStage<1, [A8_LdSt0]>,
165 InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [4, 1]>,
168 // Integer store pipeline
170 // use A8_Issue to enforce the 1 load/store per cycle limit
173 InstrItinData<IIC_iStorei , [InstrStage<1, [A8_Issue], 0>,
174 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
175 InstrStage<1, [A8_LdSt0]>], [3, 1]>,
178 InstrItinData<IIC_iStorer , [InstrStage<1, [A8_Issue], 0>,
179 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
180 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
182 // Scaled register offset, issues over 2 cycles
183 InstrItinData<IIC_iStoresi , [InstrStage<2, [A8_Issue], 0>,
184 InstrStage<1, [A8_Pipe0], 0>,
185 InstrStage<1, [A8_Pipe1]>,
186 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
187 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
189 // Immediate offset with update
190 InstrItinData<IIC_iStoreiu , [InstrStage<1, [A8_Issue], 0>,
191 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
192 InstrStage<1, [A8_LdSt0]>], [2, 3, 1]>,
194 // Register offset with update
195 InstrItinData<IIC_iStoreru , [InstrStage<1, [A8_Issue], 0>,
196 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
197 InstrStage<1, [A8_LdSt0]>], [2, 3, 1, 1]>,
199 // Scaled register offset with update, issues over 2 cycles
200 InstrItinData<IIC_iStoresiu, [InstrStage<2, [A8_Issue], 0>,
201 InstrStage<1, [A8_Pipe0], 0>,
202 InstrStage<1, [A8_Pipe1]>,
203 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
204 InstrStage<1, [A8_LdSt0]>], [3, 3, 1, 1]>,
207 InstrItinData<IIC_iStorem , [InstrStage<2, [A8_Issue], 0>,
208 InstrStage<2, [A8_Pipe0], 0>,
209 InstrStage<2, [A8_Pipe1]>,
210 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
211 InstrStage<1, [A8_LdSt0]>]>,
215 // no delay slots, so the latency of a branch is unimportant
216 InstrItinData<IIC_Br , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
219 // Issue through integer pipeline, and execute in NEON unit. We assume
220 // RunFast mode so that NFP pipeline is used for single-precision when
223 // FP Special Register to Integer Register File Move
224 InstrItinData<IIC_fpSTAT , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
225 InstrStage<1, [A8_NLSPipe]>]>,
227 // Single-precision FP Unary
228 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
229 InstrStage<1, [A8_NPipe]>], [7, 1]>,
231 // Double-precision FP Unary
232 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
233 InstrStage<4, [A8_NPipe], 0>,
234 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
236 // Single-precision FP Compare
237 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
238 InstrStage<1, [A8_NPipe]>], [1, 1]>,
240 // Double-precision FP Compare
241 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
242 InstrStage<4, [A8_NPipe], 0>,
243 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
245 // Single to Double FP Convert
246 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
247 InstrStage<7, [A8_NPipe], 0>,
248 InstrStage<7, [A8_NLSPipe]>], [7, 1]>,
250 // Double to Single FP Convert
251 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
252 InstrStage<5, [A8_NPipe], 0>,
253 InstrStage<5, [A8_NLSPipe]>], [5, 1]>,
255 // Single-Precision FP to Integer Convert
256 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
257 InstrStage<1, [A8_NPipe]>], [7, 1]>,
259 // Double-Precision FP to Integer Convert
260 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
261 InstrStage<8, [A8_NPipe], 0>,
262 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
264 // Integer to Single-Precision FP Convert
265 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
266 InstrStage<1, [A8_NPipe]>], [7, 1]>,
268 // Integer to Double-Precision FP Convert
269 InstrItinData<IIC_fpCVTID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
270 InstrStage<8, [A8_NPipe], 0>,
271 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
273 // Single-precision FP ALU
274 InstrItinData<IIC_fpALU32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
275 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
277 // Double-precision FP ALU
278 InstrItinData<IIC_fpALU64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
279 InstrStage<9, [A8_NPipe], 0>,
280 InstrStage<9, [A8_NLSPipe]>], [9, 1, 1]>,
282 // Single-precision FP Multiply
283 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
284 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
286 // Double-precision FP Multiply
287 InstrItinData<IIC_fpMUL64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
288 InstrStage<11, [A8_NPipe], 0>,
289 InstrStage<11, [A8_NLSPipe]>], [11, 1, 1]>,
291 // Single-precision FP MAC
292 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
293 InstrStage<1, [A8_NPipe]>], [7, 2, 1, 1]>,
295 // Double-precision FP MAC
296 InstrItinData<IIC_fpMAC64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
297 InstrStage<19, [A8_NPipe], 0>,
298 InstrStage<19, [A8_NLSPipe]>], [19, 2, 1, 1]>,
300 // Single-precision FP DIV
301 InstrItinData<IIC_fpDIV32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
302 InstrStage<20, [A8_NPipe], 0>,
303 InstrStage<20, [A8_NLSPipe]>], [20, 1, 1]>,
305 // Double-precision FP DIV
306 InstrItinData<IIC_fpDIV64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
307 InstrStage<29, [A8_NPipe], 0>,
308 InstrStage<29, [A8_NLSPipe]>], [29, 1, 1]>,
310 // Single-precision FP SQRT
311 InstrItinData<IIC_fpSQRT32, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
312 InstrStage<19, [A8_NPipe], 0>,
313 InstrStage<19, [A8_NLSPipe]>], [19, 1]>,
315 // Double-precision FP SQRT
316 InstrItinData<IIC_fpSQRT64, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
317 InstrStage<29, [A8_NPipe], 0>,
318 InstrStage<29, [A8_NLSPipe]>], [29, 1]>,
320 // Single-precision FP Load
321 // use A8_Issue to enforce the 1 load/store per cycle limit
322 InstrItinData<IIC_fpLoad32, [InstrStage<1, [A8_Issue], 0>,
323 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
324 InstrStage<1, [A8_LdSt0], 0>,
325 InstrStage<1, [A8_NLSPipe]>]>,
327 // Double-precision FP Load
328 // use A8_Issue to enforce the 1 load/store per cycle limit
329 InstrItinData<IIC_fpLoad64, [InstrStage<2, [A8_Issue], 0>,
330 InstrStage<1, [A8_Pipe0], 0>,
331 InstrStage<1, [A8_Pipe1]>,
332 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
333 InstrStage<1, [A8_LdSt0], 0>,
334 InstrStage<1, [A8_NLSPipe]>]>,
337 // use A8_Issue to enforce the 1 load/store per cycle limit
338 InstrItinData<IIC_fpLoadm, [InstrStage<3, [A8_Issue], 0>,
339 InstrStage<2, [A8_Pipe0], 0>,
340 InstrStage<2, [A8_Pipe1]>,
341 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
342 InstrStage<1, [A8_LdSt0], 0>,
343 InstrStage<1, [A8_NLSPipe]>]>,
345 // Single-precision FP Store
346 // use A8_Issue to enforce the 1 load/store per cycle limit
347 InstrItinData<IIC_fpStore32,[InstrStage<1, [A8_Issue], 0>,
348 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
349 InstrStage<1, [A8_LdSt0], 0>,
350 InstrStage<1, [A8_NLSPipe]>]>,
352 // Double-precision FP Store
353 // use A8_Issue to enforce the 1 load/store per cycle limit
354 InstrItinData<IIC_fpStore64,[InstrStage<2, [A8_Issue], 0>,
355 InstrStage<1, [A8_Pipe0], 0>,
356 InstrStage<1, [A8_Pipe1]>,
357 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
358 InstrStage<1, [A8_LdSt0], 0>,
359 InstrStage<1, [A8_NLSPipe]>]>,
362 // use A8_Issue to enforce the 1 load/store per cycle limit
363 InstrItinData<IIC_fpStorem, [InstrStage<3, [A8_Issue], 0>,
364 InstrStage<2, [A8_Pipe0], 0>,
365 InstrStage<2, [A8_Pipe1]>,
366 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
367 InstrStage<1, [A8_LdSt0], 0>,
368 InstrStage<1, [A8_NLSPipe]>]>,
371 // Issue through integer pipeline, and execute in NEON unit.
374 // FIXME: We don't model this instruction properly
375 InstrItinData<IIC_VLD1, [InstrStage<1, [A8_Issue], 0>,
376 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
377 InstrStage<1, [A8_LdSt0], 0>,
378 InstrStage<1, [A8_NLSPipe]>]>,
381 // FIXME: We don't model this instruction properly
382 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Issue], 0>,
383 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
384 InstrStage<1, [A8_LdSt0], 0>,
385 InstrStage<1, [A8_NLSPipe]>], [2, 2, 1]>,
388 // FIXME: We don't model this instruction properly
389 InstrItinData<IIC_VLD3, [InstrStage<1, [A8_Issue], 0>,
390 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
391 InstrStage<1, [A8_LdSt0], 0>,
392 InstrStage<1, [A8_NLSPipe]>], [2, 2, 2, 1]>,
395 // FIXME: We don't model this instruction properly
396 InstrItinData<IIC_VLD4, [InstrStage<1, [A8_Issue], 0>,
397 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
398 InstrStage<1, [A8_LdSt0], 0>,
399 InstrStage<1, [A8_NLSPipe]>], [2, 2, 2, 2, 1]>,
402 // FIXME: We don't model this instruction properly
403 InstrItinData<IIC_VST, [InstrStage<1, [A8_Issue], 0>,
404 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
405 InstrStage<1, [A8_LdSt0], 0>,
406 InstrStage<1, [A8_NLSPipe]>]>,
408 // Double-register FP Unary
409 InstrItinData<IIC_VUNAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
410 InstrStage<1, [A8_NPipe]>], [5, 2]>,
412 // Quad-register FP Unary
413 // Result written in N5, but that is relative to the last cycle of multicycle,
414 // so we use 6 for those cases
415 InstrItinData<IIC_VUNAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
416 InstrStage<2, [A8_NPipe]>], [6, 2]>,
418 // Double-register FP Binary
419 InstrItinData<IIC_VBIND, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
420 InstrStage<1, [A8_NPipe]>], [5, 2, 2]>,
422 // Quad-register FP Binary
423 // Result written in N5, but that is relative to the last cycle of multicycle,
424 // so we use 6 for those cases
425 InstrItinData<IIC_VBINQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
426 InstrStage<2, [A8_NPipe]>], [6, 2, 2]>,
429 InstrItinData<IIC_VMOVImm, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
430 InstrStage<1, [A8_NPipe]>], [3]>,
432 // Double-register Permute Move
433 InstrItinData<IIC_VMOVD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
434 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
436 // Quad-register Permute Move
437 // Result written in N2, but that is relative to the last cycle of multicycle,
438 // so we use 3 for those cases
439 InstrItinData<IIC_VMOVQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
440 InstrStage<2, [A8_NLSPipe]>], [3, 1]>,
442 // Integer to Single-precision Move
443 InstrItinData<IIC_VMOVIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
444 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
446 // Integer to Double-precision Move
447 InstrItinData<IIC_VMOVID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
448 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
450 // Single-precision to Integer Move
451 InstrItinData<IIC_VMOVSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
452 InstrStage<1, [A8_NLSPipe]>], [20, 1]>,
454 // Double-precision to Integer Move
455 InstrItinData<IIC_VMOVDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
456 InstrStage<1, [A8_NLSPipe]>], [20, 20, 1]>,
458 // Integer to Lane Move
459 InstrItinData<IIC_VMOVISL , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
460 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
462 // Double-register Permute
463 InstrItinData<IIC_VPERMD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
464 InstrStage<1, [A8_NLSPipe]>], [2, 2, 1, 1]>,
466 // Quad-register Permute
467 // Result written in N2, but that is relative to the last cycle of multicycle,
468 // so we use 3 for those cases
469 InstrItinData<IIC_VPERMQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
470 InstrStage<2, [A8_NLSPipe]>], [3, 3, 1, 1]>,
472 // Quad-register Permute (3 cycle issue)
473 // Result written in N2, but that is relative to the last cycle of multicycle,
474 // so we use 4 for those cases
475 InstrItinData<IIC_VPERMQ3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
476 InstrStage<1, [A8_NLSPipe]>,
477 InstrStage<1, [A8_NPipe], 0>,
478 InstrStage<2, [A8_NLSPipe]>], [4, 4, 1, 1]>,
480 // Double-register FP Multiple-Accumulate
481 InstrItinData<IIC_VMACD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
482 InstrStage<1, [A8_NPipe]>], [9, 3, 2, 2]>,
484 // Quad-register FP Multiple-Accumulate
485 // Result written in N9, but that is relative to the last cycle of multicycle,
486 // so we use 10 for those cases
487 InstrItinData<IIC_VMACQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
488 InstrStage<2, [A8_NPipe]>], [10, 3, 2, 2]>,
490 // Double-register Reciprical Step
491 InstrItinData<IIC_VRECSD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
492 InstrStage<1, [A8_NPipe]>], [9, 2, 2]>,
494 // Quad-register Reciprical Step
495 InstrItinData<IIC_VRECSQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
496 InstrStage<2, [A8_NPipe]>], [10, 2, 2]>,
498 // Double-register Integer Count
499 InstrItinData<IIC_VCNTiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
500 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
502 // Quad-register Integer Count
503 // Result written in N3, but that is relative to the last cycle of multicycle,
504 // so we use 4 for those cases
505 InstrItinData<IIC_VCNTiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
506 InstrStage<2, [A8_NPipe]>], [4, 2, 2]>,
508 // Double-register Integer Unary
509 InstrItinData<IIC_VUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
510 InstrStage<1, [A8_NPipe]>], [4, 2]>,
512 // Quad-register Integer Unary
513 InstrItinData<IIC_VUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
514 InstrStage<1, [A8_NPipe]>], [4, 2]>,
516 // Double-register Integer Q-Unary
517 InstrItinData<IIC_VQUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
518 InstrStage<1, [A8_NPipe]>], [4, 1]>,
520 // Quad-register Integer CountQ-Unary
521 InstrItinData<IIC_VQUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
522 InstrStage<1, [A8_NPipe]>], [4, 1]>,
524 // Double-register Integer Binary
525 InstrItinData<IIC_VBINiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
526 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
528 // Quad-register Integer Binary
529 InstrItinData<IIC_VBINiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
530 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
532 // Double-register Integer Binary (4 cycle)
533 InstrItinData<IIC_VBINi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
534 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
536 // Quad-register Integer Binary (4 cycle)
537 InstrItinData<IIC_VBINi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
538 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
541 // Double-register Integer Subtract
542 InstrItinData<IIC_VSUBiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
543 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
545 // Quad-register Integer Subtract
546 InstrItinData<IIC_VSUBiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
547 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
549 // Double-register Integer Subtract
550 InstrItinData<IIC_VSUBi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
551 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
553 // Quad-register Integer Subtract
554 InstrItinData<IIC_VSUBi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
555 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
557 // Double-register Integer Shift
558 InstrItinData<IIC_VSHLiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
559 InstrStage<1, [A8_NPipe]>], [3, 1, 1]>,
561 // Quad-register Integer Shift
562 InstrItinData<IIC_VSHLiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
563 InstrStage<2, [A8_NPipe]>], [4, 1, 1]>,
565 // Double-register Integer Shift (4 cycle)
566 InstrItinData<IIC_VSHLi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
567 InstrStage<1, [A8_NPipe]>], [4, 1, 1]>,
569 // Quad-register Integer Shift (4 cycle)
570 InstrItinData<IIC_VSHLi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
571 InstrStage<2, [A8_NPipe]>], [5, 1, 1]>,
573 // Double-register Integer Pair Add Long
574 InstrItinData<IIC_VPALiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
575 InstrStage<1, [A8_NPipe]>], [6, 3, 1]>,
577 // Quad-register Integer Pair Add Long
578 InstrItinData<IIC_VPALiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
579 InstrStage<2, [A8_NPipe]>], [7, 3, 1]>,
581 // Double-register Absolute Difference and Accumulate
582 InstrItinData<IIC_VABAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
583 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 1]>,
585 // Quad-register Absolute Difference and Accumulate
586 InstrItinData<IIC_VABAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
587 InstrStage<2, [A8_NPipe]>], [6, 3, 2, 1]>,
590 // Double-register Integer Multiply (.8, .16)
591 InstrItinData<IIC_VMULi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
592 InstrStage<1, [A8_NPipe]>], [6, 2, 2]>,
594 // Double-register Integer Multiply (.32)
595 InstrItinData<IIC_VMULi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
596 InstrStage<2, [A8_NPipe]>], [7, 2, 1]>,
598 // Quad-register Integer Multiply (.8, .16)
599 InstrItinData<IIC_VMULi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
600 InstrStage<2, [A8_NPipe]>], [7, 2, 2]>,
602 // Quad-register Integer Multiply (.32)
603 InstrItinData<IIC_VMULi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
604 InstrStage<1, [A8_NPipe]>,
605 InstrStage<2, [A8_NLSPipe], 0>,
606 InstrStage<3, [A8_NPipe]>], [9, 2, 1]>,
608 // Double-register Integer Multiply-Accumulate (.8, .16)
609 InstrItinData<IIC_VMACi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
610 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 2]>,
612 // Double-register Integer Multiply-Accumulate (.32)
613 InstrItinData<IIC_VMACi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
614 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 1]>,
616 // Quad-register Integer Multiply-Accumulate (.8, .16)
617 InstrItinData<IIC_VMACi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
618 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 2]>,
620 // Quad-register Integer Multiply-Accumulate (.32)
621 InstrItinData<IIC_VMACi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
622 InstrStage<1, [A8_NPipe]>,
623 InstrStage<2, [A8_NLSPipe], 0>,
624 InstrStage<3, [A8_NPipe]>], [9, 3, 2, 1]>,
626 // Double-register VEXT
627 InstrItinData<IIC_VEXTD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
628 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
630 // Quad-register VEXT
631 InstrItinData<IIC_VEXTQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
632 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
635 InstrItinData<IIC_VTB1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
636 InstrStage<2, [A8_NLSPipe]>], [3, 2, 1]>,
637 InstrItinData<IIC_VTB2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
638 InstrStage<2, [A8_NLSPipe]>], [3, 2, 2, 1]>,
639 InstrItinData<IIC_VTB3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
640 InstrStage<1, [A8_NLSPipe]>,
641 InstrStage<1, [A8_NPipe], 0>,
642 InstrStage<2, [A8_NLSPipe]>], [4, 2, 2, 3, 1]>,
643 InstrItinData<IIC_VTB4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
644 InstrStage<1, [A8_NLSPipe]>,
645 InstrStage<1, [A8_NPipe], 0>,
646 InstrStage<2, [A8_NLSPipe]>],[4, 2, 2, 3, 3, 1]>,
649 InstrItinData<IIC_VTBX1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
650 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 1]>,
651 InstrItinData<IIC_VTBX2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
652 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 2, 1]>,
653 InstrItinData<IIC_VTBX3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
654 InstrStage<1, [A8_NLSPipe]>,
655 InstrStage<1, [A8_NPipe], 0>,
656 InstrStage<2, [A8_NLSPipe]>],[4, 1, 2, 2, 3, 1]>,
657 InstrItinData<IIC_VTBX4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
658 InstrStage<1, [A8_NLSPipe]>,
659 InstrStage<1, [A8_NPipe], 0>,
660 InstrStage<2, [A8_NLSPipe]>], [4, 1, 2, 2, 3, 3, 1]>