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_iALUsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1, 1]>,
41 // Bitwise Instructions that produce a result
42 InstrItinData<IIC_iBITi ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
43 InstrItinData<IIC_iBITr ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 2]>,
44 InstrItinData<IIC_iBITsi,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
45 InstrItinData<IIC_iBITsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1, 1]>,
47 // Unary Instructions that produce a result
48 InstrItinData<IIC_iUNAr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
49 InstrItinData<IIC_iUNAsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
51 // Zero and sign extension instructions
52 InstrItinData<IIC_iEXTr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
53 InstrItinData<IIC_iEXTAr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
54 InstrItinData<IIC_iEXTAsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>],[2, 2, 1, 1]>,
56 // Compare instructions
57 InstrItinData<IIC_iCMPi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
58 InstrItinData<IIC_iCMPr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
59 InstrItinData<IIC_iCMPsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
60 InstrItinData<IIC_iCMPsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
62 // Move instructions, unconditional
63 InstrItinData<IIC_iMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1]>,
64 InstrItinData<IIC_iMOVix2,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
65 InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
66 InstrItinData<IIC_iMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
67 InstrItinData<IIC_iMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
68 InstrItinData<IIC_iMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1, 1]>,
70 // Move instructions, conditional
71 InstrItinData<IIC_iCMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
72 InstrItinData<IIC_iCMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
73 InstrItinData<IIC_iCMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
74 InstrItinData<IIC_iCMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
76 // Integer multiply pipeline
77 // Result written in E5, but that is relative to the last cycle of multicycle,
78 // so we use 6 for those cases
80 InstrItinData<IIC_iMUL16 , [InstrStage<1, [A8_Pipe0]>], [5, 1, 1]>,
81 InstrItinData<IIC_iMAC16 , [InstrStage<1, [A8_Pipe1], 0>,
82 InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
83 InstrItinData<IIC_iMUL32 , [InstrStage<1, [A8_Pipe1], 0>,
84 InstrStage<2, [A8_Pipe0]>], [6, 1, 1]>,
85 InstrItinData<IIC_iMAC32 , [InstrStage<1, [A8_Pipe1], 0>,
86 InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
87 InstrItinData<IIC_iMUL64 , [InstrStage<2, [A8_Pipe1], 0>,
88 InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
89 InstrItinData<IIC_iMAC64 , [InstrStage<2, [A8_Pipe1], 0>,
90 InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
92 // Integer load pipeline
94 // loads have an extra cycle of latency, but are fully pipelined
95 // use A8_Issue to enforce the 1 load/store per cycle limit
98 InstrItinData<IIC_iLoadi , [InstrStage<1, [A8_Issue], 0>,
99 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
100 InstrStage<1, [A8_LdSt0]>], [3, 1]>,
103 InstrItinData<IIC_iLoadr , [InstrStage<1, [A8_Issue], 0>,
104 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
105 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
107 // Scaled register offset, issues over 2 cycles
108 InstrItinData<IIC_iLoadsi , [InstrStage<2, [A8_Issue], 0>,
109 InstrStage<1, [A8_Pipe0], 0>,
110 InstrStage<1, [A8_Pipe1]>,
111 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
112 InstrStage<1, [A8_LdSt0]>], [4, 1, 1]>,
114 // Immediate offset with update
115 InstrItinData<IIC_iLoadiu , [InstrStage<1, [A8_Issue], 0>,
116 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
117 InstrStage<1, [A8_LdSt0]>], [3, 2, 1]>,
119 // Register offset with update
120 InstrItinData<IIC_iLoadru , [InstrStage<1, [A8_Issue], 0>,
121 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
122 InstrStage<1, [A8_LdSt0]>], [3, 2, 1, 1]>,
124 // Scaled register offset with update, issues over 2 cycles
125 InstrItinData<IIC_iLoadsiu , [InstrStage<2, [A8_Issue], 0>,
126 InstrStage<1, [A8_Pipe0], 0>,
127 InstrStage<1, [A8_Pipe1]>,
128 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
129 InstrStage<1, [A8_LdSt0]>], [4, 3, 1, 1]>,
132 InstrItinData<IIC_iLoadm , [InstrStage<2, [A8_Issue], 0>,
133 InstrStage<2, [A8_Pipe0], 0>,
134 InstrStage<2, [A8_Pipe1]>,
135 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
136 InstrStage<1, [A8_LdSt0]>]>,
139 // Load multiple plus branch
140 InstrItinData<IIC_iLoadmBr , [InstrStage<2, [A8_Issue], 0>,
141 InstrStage<2, [A8_Pipe0], 0>,
142 InstrStage<2, [A8_Pipe1]>,
143 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
144 InstrStage<1, [A8_LdSt0]>,
145 InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
148 // iLoadi + iALUr for t2LDRpci_pic.
149 InstrItinData<IIC_iLoadiALU, [InstrStage<1, [A8_Issue], 0>,
150 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
151 InstrStage<1, [A8_LdSt0]>,
152 InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [4, 1]>,
155 // Integer store pipeline
157 // use A8_Issue to enforce the 1 load/store per cycle limit
160 InstrItinData<IIC_iStorei , [InstrStage<1, [A8_Issue], 0>,
161 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
162 InstrStage<1, [A8_LdSt0]>], [3, 1]>,
165 InstrItinData<IIC_iStorer , [InstrStage<1, [A8_Issue], 0>,
166 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
167 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
169 // Scaled register offset, issues over 2 cycles
170 InstrItinData<IIC_iStoresi , [InstrStage<2, [A8_Issue], 0>,
171 InstrStage<1, [A8_Pipe0], 0>,
172 InstrStage<1, [A8_Pipe1]>,
173 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
174 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
176 // Immediate offset with update
177 InstrItinData<IIC_iStoreiu , [InstrStage<1, [A8_Issue], 0>,
178 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
179 InstrStage<1, [A8_LdSt0]>], [2, 3, 1]>,
181 // Register offset with update
182 InstrItinData<IIC_iStoreru , [InstrStage<1, [A8_Issue], 0>,
183 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
184 InstrStage<1, [A8_LdSt0]>], [2, 3, 1, 1]>,
186 // Scaled register offset with update, issues over 2 cycles
187 InstrItinData<IIC_iStoresiu, [InstrStage<2, [A8_Issue], 0>,
188 InstrStage<1, [A8_Pipe0], 0>,
189 InstrStage<1, [A8_Pipe1]>,
190 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
191 InstrStage<1, [A8_LdSt0]>], [3, 3, 1, 1]>,
194 InstrItinData<IIC_iStorem , [InstrStage<2, [A8_Issue], 0>,
195 InstrStage<2, [A8_Pipe0], 0>,
196 InstrStage<2, [A8_Pipe1]>,
197 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
198 InstrStage<1, [A8_LdSt0]>]>,
202 // no delay slots, so the latency of a branch is unimportant
203 InstrItinData<IIC_Br , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
206 // Issue through integer pipeline, and execute in NEON unit. We assume
207 // RunFast mode so that NFP pipeline is used for single-precision when
210 // FP Special Register to Integer Register File Move
211 InstrItinData<IIC_fpSTAT , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
212 InstrStage<1, [A8_NLSPipe]>]>,
214 // Single-precision FP Unary
215 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
216 InstrStage<1, [A8_NPipe]>], [7, 1]>,
218 // Double-precision FP Unary
219 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
220 InstrStage<4, [A8_NPipe], 0>,
221 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
223 // Single-precision FP Compare
224 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
225 InstrStage<1, [A8_NPipe]>], [1, 1]>,
227 // Double-precision FP Compare
228 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
229 InstrStage<4, [A8_NPipe], 0>,
230 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
232 // Single to Double FP Convert
233 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
234 InstrStage<7, [A8_NPipe], 0>,
235 InstrStage<7, [A8_NLSPipe]>], [7, 1]>,
237 // Double to Single FP Convert
238 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
239 InstrStage<5, [A8_NPipe], 0>,
240 InstrStage<5, [A8_NLSPipe]>], [5, 1]>,
242 // Single-Precision FP to Integer Convert
243 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
244 InstrStage<1, [A8_NPipe]>], [7, 1]>,
246 // Double-Precision FP to Integer Convert
247 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
248 InstrStage<8, [A8_NPipe], 0>,
249 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
251 // Integer to Single-Precision FP Convert
252 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
253 InstrStage<1, [A8_NPipe]>], [7, 1]>,
255 // Integer to Double-Precision FP Convert
256 InstrItinData<IIC_fpCVTID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
257 InstrStage<8, [A8_NPipe], 0>,
258 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
260 // Single-precision FP ALU
261 InstrItinData<IIC_fpALU32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
262 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
264 // Double-precision FP ALU
265 InstrItinData<IIC_fpALU64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
266 InstrStage<9, [A8_NPipe], 0>,
267 InstrStage<9, [A8_NLSPipe]>], [9, 1, 1]>,
269 // Single-precision FP Multiply
270 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
271 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
273 // Double-precision FP Multiply
274 InstrItinData<IIC_fpMUL64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
275 InstrStage<11, [A8_NPipe], 0>,
276 InstrStage<11, [A8_NLSPipe]>], [11, 1, 1]>,
278 // Single-precision FP MAC
279 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
280 InstrStage<1, [A8_NPipe]>], [7, 2, 1, 1]>,
282 // Double-precision FP MAC
283 InstrItinData<IIC_fpMAC64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
284 InstrStage<19, [A8_NPipe], 0>,
285 InstrStage<19, [A8_NLSPipe]>], [19, 2, 1, 1]>,
287 // Single-precision FP DIV
288 InstrItinData<IIC_fpDIV32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
289 InstrStage<20, [A8_NPipe], 0>,
290 InstrStage<20, [A8_NLSPipe]>], [20, 1, 1]>,
292 // Double-precision FP DIV
293 InstrItinData<IIC_fpDIV64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
294 InstrStage<29, [A8_NPipe], 0>,
295 InstrStage<29, [A8_NLSPipe]>], [29, 1, 1]>,
297 // Single-precision FP SQRT
298 InstrItinData<IIC_fpSQRT32, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
299 InstrStage<19, [A8_NPipe], 0>,
300 InstrStage<19, [A8_NLSPipe]>], [19, 1]>,
302 // Double-precision FP SQRT
303 InstrItinData<IIC_fpSQRT64, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
304 InstrStage<29, [A8_NPipe], 0>,
305 InstrStage<29, [A8_NLSPipe]>], [29, 1]>,
307 // Single-precision FP Load
308 // use A8_Issue to enforce the 1 load/store per cycle limit
309 InstrItinData<IIC_fpLoad32, [InstrStage<1, [A8_Issue], 0>,
310 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
311 InstrStage<1, [A8_LdSt0], 0>,
312 InstrStage<1, [A8_NLSPipe]>]>,
314 // Double-precision FP Load
315 // use A8_Issue to enforce the 1 load/store per cycle limit
316 InstrItinData<IIC_fpLoad64, [InstrStage<2, [A8_Issue], 0>,
317 InstrStage<1, [A8_Pipe0], 0>,
318 InstrStage<1, [A8_Pipe1]>,
319 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
320 InstrStage<1, [A8_LdSt0], 0>,
321 InstrStage<1, [A8_NLSPipe]>]>,
324 // use A8_Issue to enforce the 1 load/store per cycle limit
325 InstrItinData<IIC_fpLoadm, [InstrStage<3, [A8_Issue], 0>,
326 InstrStage<2, [A8_Pipe0], 0>,
327 InstrStage<2, [A8_Pipe1]>,
328 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
329 InstrStage<1, [A8_LdSt0], 0>,
330 InstrStage<1, [A8_NLSPipe]>]>,
332 // Single-precision FP Store
333 // use A8_Issue to enforce the 1 load/store per cycle limit
334 InstrItinData<IIC_fpStore32,[InstrStage<1, [A8_Issue], 0>,
335 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
336 InstrStage<1, [A8_LdSt0], 0>,
337 InstrStage<1, [A8_NLSPipe]>]>,
339 // Double-precision FP Store
340 // use A8_Issue to enforce the 1 load/store per cycle limit
341 InstrItinData<IIC_fpStore64,[InstrStage<2, [A8_Issue], 0>,
342 InstrStage<1, [A8_Pipe0], 0>,
343 InstrStage<1, [A8_Pipe1]>,
344 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
345 InstrStage<1, [A8_LdSt0], 0>,
346 InstrStage<1, [A8_NLSPipe]>]>,
349 // use A8_Issue to enforce the 1 load/store per cycle limit
350 InstrItinData<IIC_fpStorem, [InstrStage<3, [A8_Issue], 0>,
351 InstrStage<2, [A8_Pipe0], 0>,
352 InstrStage<2, [A8_Pipe1]>,
353 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
354 InstrStage<1, [A8_LdSt0], 0>,
355 InstrStage<1, [A8_NLSPipe]>]>,
358 // Issue through integer pipeline, and execute in NEON unit.
361 // FIXME: We don't model this instruction properly
362 InstrItinData<IIC_VLD1, [InstrStage<1, [A8_Issue], 0>,
363 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
364 InstrStage<1, [A8_LdSt0], 0>,
365 InstrStage<1, [A8_NLSPipe]>]>,
368 // FIXME: We don't model this instruction properly
369 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Issue], 0>,
370 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
371 InstrStage<1, [A8_LdSt0], 0>,
372 InstrStage<1, [A8_NLSPipe]>], [2, 2, 1]>,
375 // FIXME: We don't model this instruction properly
376 InstrItinData<IIC_VLD3, [InstrStage<1, [A8_Issue], 0>,
377 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
378 InstrStage<1, [A8_LdSt0], 0>,
379 InstrStage<1, [A8_NLSPipe]>], [2, 2, 2, 1]>,
382 // FIXME: We don't model this instruction properly
383 InstrItinData<IIC_VLD4, [InstrStage<1, [A8_Issue], 0>,
384 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
385 InstrStage<1, [A8_LdSt0], 0>,
386 InstrStage<1, [A8_NLSPipe]>], [2, 2, 2, 2, 1]>,
389 // FIXME: We don't model this instruction properly
390 InstrItinData<IIC_VST, [InstrStage<1, [A8_Issue], 0>,
391 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
392 InstrStage<1, [A8_LdSt0], 0>,
393 InstrStage<1, [A8_NLSPipe]>]>,
395 // Double-register FP Unary
396 InstrItinData<IIC_VUNAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
397 InstrStage<1, [A8_NPipe]>], [5, 2]>,
399 // Quad-register FP Unary
400 // Result written in N5, but that is relative to the last cycle of multicycle,
401 // so we use 6 for those cases
402 InstrItinData<IIC_VUNAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
403 InstrStage<2, [A8_NPipe]>], [6, 2]>,
405 // Double-register FP Binary
406 InstrItinData<IIC_VBIND, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
407 InstrStage<1, [A8_NPipe]>], [5, 2, 2]>,
409 // Quad-register FP Binary
410 // Result written in N5, but that is relative to the last cycle of multicycle,
411 // so we use 6 for those cases
412 InstrItinData<IIC_VBINQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
413 InstrStage<2, [A8_NPipe]>], [6, 2, 2]>,
416 InstrItinData<IIC_VMOVImm, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
417 InstrStage<1, [A8_NPipe]>], [3]>,
419 // Double-register Permute Move
420 InstrItinData<IIC_VMOVD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
421 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
423 // Quad-register Permute Move
424 // Result written in N2, but that is relative to the last cycle of multicycle,
425 // so we use 3 for those cases
426 InstrItinData<IIC_VMOVQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
427 InstrStage<2, [A8_NLSPipe]>], [3, 1]>,
429 // Integer to Single-precision Move
430 InstrItinData<IIC_VMOVIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
431 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
433 // Integer to Double-precision Move
434 InstrItinData<IIC_VMOVID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
435 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
437 // Single-precision to Integer Move
438 InstrItinData<IIC_VMOVSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
439 InstrStage<1, [A8_NLSPipe]>], [20, 1]>,
441 // Double-precision to Integer Move
442 InstrItinData<IIC_VMOVDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
443 InstrStage<1, [A8_NLSPipe]>], [20, 20, 1]>,
445 // Integer to Lane Move
446 InstrItinData<IIC_VMOVISL , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
447 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
449 // Double-register Permute
450 InstrItinData<IIC_VPERMD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
451 InstrStage<1, [A8_NLSPipe]>], [2, 2, 1, 1]>,
453 // Quad-register Permute
454 // Result written in N2, but that is relative to the last cycle of multicycle,
455 // so we use 3 for those cases
456 InstrItinData<IIC_VPERMQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
457 InstrStage<2, [A8_NLSPipe]>], [3, 3, 1, 1]>,
459 // Quad-register Permute (3 cycle issue)
460 // Result written in N2, but that is relative to the last cycle of multicycle,
461 // so we use 4 for those cases
462 InstrItinData<IIC_VPERMQ3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
463 InstrStage<1, [A8_NLSPipe]>,
464 InstrStage<1, [A8_NPipe], 0>,
465 InstrStage<2, [A8_NLSPipe]>], [4, 4, 1, 1]>,
467 // Double-register FP Multiple-Accumulate
468 InstrItinData<IIC_VMACD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
469 InstrStage<1, [A8_NPipe]>], [9, 3, 2, 2]>,
471 // Quad-register FP Multiple-Accumulate
472 // Result written in N9, but that is relative to the last cycle of multicycle,
473 // so we use 10 for those cases
474 InstrItinData<IIC_VMACQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
475 InstrStage<2, [A8_NPipe]>], [10, 3, 2, 2]>,
477 // Double-register Reciprical Step
478 InstrItinData<IIC_VRECSD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
479 InstrStage<1, [A8_NPipe]>], [9, 2, 2]>,
481 // Quad-register Reciprical Step
482 InstrItinData<IIC_VRECSQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
483 InstrStage<2, [A8_NPipe]>], [10, 2, 2]>,
485 // Double-register Integer Count
486 InstrItinData<IIC_VCNTiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
487 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
489 // Quad-register Integer Count
490 // Result written in N3, but that is relative to the last cycle of multicycle,
491 // so we use 4 for those cases
492 InstrItinData<IIC_VCNTiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
493 InstrStage<2, [A8_NPipe]>], [4, 2, 2]>,
495 // Double-register Integer Unary
496 InstrItinData<IIC_VUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
497 InstrStage<1, [A8_NPipe]>], [4, 2]>,
499 // Quad-register Integer Unary
500 InstrItinData<IIC_VUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
501 InstrStage<1, [A8_NPipe]>], [4, 2]>,
503 // Double-register Integer Q-Unary
504 InstrItinData<IIC_VQUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
505 InstrStage<1, [A8_NPipe]>], [4, 1]>,
507 // Quad-register Integer CountQ-Unary
508 InstrItinData<IIC_VQUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
509 InstrStage<1, [A8_NPipe]>], [4, 1]>,
511 // Double-register Integer Binary
512 InstrItinData<IIC_VBINiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
513 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
515 // Quad-register Integer Binary
516 InstrItinData<IIC_VBINiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
517 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
519 // Double-register Integer Binary (4 cycle)
520 InstrItinData<IIC_VBINi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
521 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
523 // Quad-register Integer Binary (4 cycle)
524 InstrItinData<IIC_VBINi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
525 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
528 // Double-register Integer Subtract
529 InstrItinData<IIC_VSUBiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
530 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
532 // Quad-register Integer Subtract
533 InstrItinData<IIC_VSUBiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
534 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
536 // Double-register Integer Subtract
537 InstrItinData<IIC_VSUBi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
538 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
540 // Quad-register Integer Subtract
541 InstrItinData<IIC_VSUBi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
542 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
544 // Double-register Integer Shift
545 InstrItinData<IIC_VSHLiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
546 InstrStage<1, [A8_NPipe]>], [3, 1, 1]>,
548 // Quad-register Integer Shift
549 InstrItinData<IIC_VSHLiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
550 InstrStage<2, [A8_NPipe]>], [4, 1, 1]>,
552 // Double-register Integer Shift (4 cycle)
553 InstrItinData<IIC_VSHLi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
554 InstrStage<1, [A8_NPipe]>], [4, 1, 1]>,
556 // Quad-register Integer Shift (4 cycle)
557 InstrItinData<IIC_VSHLi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
558 InstrStage<2, [A8_NPipe]>], [5, 1, 1]>,
560 // Double-register Integer Pair Add Long
561 InstrItinData<IIC_VPALiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
562 InstrStage<1, [A8_NPipe]>], [6, 3, 1]>,
564 // Quad-register Integer Pair Add Long
565 InstrItinData<IIC_VPALiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
566 InstrStage<2, [A8_NPipe]>], [7, 3, 1]>,
568 // Double-register Absolute Difference and Accumulate
569 InstrItinData<IIC_VABAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
570 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 1]>,
572 // Quad-register Absolute Difference and Accumulate
573 InstrItinData<IIC_VABAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
574 InstrStage<2, [A8_NPipe]>], [6, 3, 2, 1]>,
577 // Double-register Integer Multiply (.8, .16)
578 InstrItinData<IIC_VMULi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
579 InstrStage<1, [A8_NPipe]>], [6, 2, 2]>,
581 // Double-register Integer Multiply (.32)
582 InstrItinData<IIC_VMULi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
583 InstrStage<2, [A8_NPipe]>], [7, 2, 1]>,
585 // Quad-register Integer Multiply (.8, .16)
586 InstrItinData<IIC_VMULi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
587 InstrStage<2, [A8_NPipe]>], [7, 2, 2]>,
589 // Quad-register Integer Multiply (.32)
590 InstrItinData<IIC_VMULi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
591 InstrStage<1, [A8_NPipe]>,
592 InstrStage<2, [A8_NLSPipe], 0>,
593 InstrStage<3, [A8_NPipe]>], [9, 2, 1]>,
595 // Double-register Integer Multiply-Accumulate (.8, .16)
596 InstrItinData<IIC_VMACi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
597 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 2]>,
599 // Double-register Integer Multiply-Accumulate (.32)
600 InstrItinData<IIC_VMACi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
601 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 1]>,
603 // Quad-register Integer Multiply-Accumulate (.8, .16)
604 InstrItinData<IIC_VMACi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
605 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 2]>,
607 // Quad-register Integer Multiply-Accumulate (.32)
608 InstrItinData<IIC_VMACi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
609 InstrStage<1, [A8_NPipe]>,
610 InstrStage<2, [A8_NLSPipe], 0>,
611 InstrStage<3, [A8_NPipe]>], [9, 3, 2, 1]>,
613 // Double-register VEXT
614 InstrItinData<IIC_VEXTD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
615 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
617 // Quad-register VEXT
618 InstrItinData<IIC_VEXTQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
619 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
622 InstrItinData<IIC_VTB1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
623 InstrStage<2, [A8_NLSPipe]>], [3, 2, 1]>,
624 InstrItinData<IIC_VTB2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
625 InstrStage<2, [A8_NLSPipe]>], [3, 2, 2, 1]>,
626 InstrItinData<IIC_VTB3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
627 InstrStage<1, [A8_NLSPipe]>,
628 InstrStage<1, [A8_NPipe], 0>,
629 InstrStage<2, [A8_NLSPipe]>], [4, 2, 2, 3, 1]>,
630 InstrItinData<IIC_VTB4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
631 InstrStage<1, [A8_NLSPipe]>,
632 InstrStage<1, [A8_NPipe], 0>,
633 InstrStage<2, [A8_NLSPipe]>],[4, 2, 2, 3, 3, 1]>,
636 InstrItinData<IIC_VTBX1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
637 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 1]>,
638 InstrItinData<IIC_VTBX2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
639 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 2, 1]>,
640 InstrItinData<IIC_VTBX3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
641 InstrStage<1, [A8_NLSPipe]>,
642 InstrStage<1, [A8_NPipe], 0>,
643 InstrStage<2, [A8_NLSPipe]>],[4, 1, 2, 2, 3, 1]>,
644 InstrItinData<IIC_VTBX4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
645 InstrStage<1, [A8_NLSPipe]>,
646 InstrStage<1, [A8_NPipe], 0>,
647 InstrStage<2, [A8_NLSPipe]>], [4, 1, 2, 2, 3, 3, 1]>