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], [
29 // Two fully-pipelined integer ALU pipelines
32 InstrItinData<IIC_iALUx , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
34 // Binary Instructions that produce a result
35 InstrItinData<IIC_iALUi ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
36 InstrItinData<IIC_iALUr ,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 2]>,
37 InstrItinData<IIC_iALUsi,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1]>,
38 InstrItinData<IIC_iALUsr,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2, 1, 1]>,
40 // Unary Instructions that produce a result
41 InstrItinData<IIC_iUNAr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
42 InstrItinData<IIC_iUNAsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
43 InstrItinData<IIC_iUNAsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
45 // Compare instructions
46 InstrItinData<IIC_iCMPi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
47 InstrItinData<IIC_iCMPr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
48 InstrItinData<IIC_iCMPsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
49 InstrItinData<IIC_iCMPsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
51 // Move instructions, unconditional
52 InstrItinData<IIC_iMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1]>,
53 InstrItinData<IIC_iMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
54 InstrItinData<IIC_iMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1]>,
55 InstrItinData<IIC_iMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1, 1]>,
57 // Move instructions, conditional
58 InstrItinData<IIC_iCMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
59 InstrItinData<IIC_iCMOVr , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
60 InstrItinData<IIC_iCMOVsi, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1]>,
61 InstrItinData<IIC_iCMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 1, 1]>,
63 // Integer multiply pipeline
64 // Result written in E5, but that is relative to the last cycle of multicycle,
65 // so we use 6 for those cases
67 InstrItinData<IIC_iMUL16 , [InstrStage<1, [A8_Pipe0]>], [5, 1, 1]>,
68 InstrItinData<IIC_iMAC16 , [InstrStage<1, [A8_Pipe1], 0>,
69 InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
70 InstrItinData<IIC_iMUL32 , [InstrStage<1, [A8_Pipe1], 0>,
71 InstrStage<2, [A8_Pipe0]>], [6, 1, 1]>,
72 InstrItinData<IIC_iMAC32 , [InstrStage<1, [A8_Pipe1], 0>,
73 InstrStage<2, [A8_Pipe0]>], [6, 1, 1, 4]>,
74 InstrItinData<IIC_iMUL64 , [InstrStage<2, [A8_Pipe1], 0>,
75 InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
76 InstrItinData<IIC_iMAC64 , [InstrStage<2, [A8_Pipe1], 0>,
77 InstrStage<3, [A8_Pipe0]>], [6, 6, 1, 1]>,
79 // Integer load pipeline
81 // loads have an extra cycle of latency, but are fully pipelined
82 // use A8_Issue to enforce the 1 load/store per cycle limit
85 InstrItinData<IIC_iLoadi , [InstrStage<1, [A8_Issue], 0>,
86 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
87 InstrStage<1, [A8_LdSt0]>], [3, 1]>,
90 InstrItinData<IIC_iLoadr , [InstrStage<1, [A8_Issue], 0>,
91 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
92 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
94 // Scaled register offset, issues over 2 cycles
95 InstrItinData<IIC_iLoadsi , [InstrStage<2, [A8_Issue], 0>,
96 InstrStage<1, [A8_Pipe0], 0>,
97 InstrStage<1, [A8_Pipe1]>,
98 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
99 InstrStage<1, [A8_LdSt0]>], [4, 1, 1]>,
101 // Immediate offset with update
102 InstrItinData<IIC_iLoadiu , [InstrStage<1, [A8_Issue], 0>,
103 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
104 InstrStage<1, [A8_LdSt0]>], [3, 2, 1]>,
106 // Register offset with update
107 InstrItinData<IIC_iLoadru , [InstrStage<1, [A8_Issue], 0>,
108 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
109 InstrStage<1, [A8_LdSt0]>], [3, 2, 1, 1]>,
111 // Scaled register offset with update, issues over 2 cycles
112 InstrItinData<IIC_iLoadsiu , [InstrStage<2, [A8_Issue], 0>,
113 InstrStage<1, [A8_Pipe0], 0>,
114 InstrStage<1, [A8_Pipe1]>,
115 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
116 InstrStage<1, [A8_LdSt0]>], [4, 3, 1, 1]>,
119 InstrItinData<IIC_iLoadm , [InstrStage<2, [A8_Issue], 0>,
120 InstrStage<2, [A8_Pipe0], 0>,
121 InstrStage<2, [A8_Pipe1]>,
122 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
123 InstrStage<1, [A8_LdSt0]>]>,
126 // Load multiple plus branch
127 InstrItinData<IIC_iLoadmBr , [InstrStage<2, [A8_Issue], 0>,
128 InstrStage<2, [A8_Pipe0], 0>,
129 InstrStage<2, [A8_Pipe1]>,
130 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
131 InstrStage<1, [A8_LdSt0]>,
132 InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
134 // Integer store pipeline
136 // use A8_Issue to enforce the 1 load/store per cycle limit
139 InstrItinData<IIC_iStorei , [InstrStage<1, [A8_Issue], 0>,
140 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
141 InstrStage<1, [A8_LdSt0]>], [3, 1]>,
144 InstrItinData<IIC_iStorer , [InstrStage<1, [A8_Issue], 0>,
145 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
146 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
148 // Scaled register offset, issues over 2 cycles
149 InstrItinData<IIC_iStoresi , [InstrStage<2, [A8_Issue], 0>,
150 InstrStage<1, [A8_Pipe0], 0>,
151 InstrStage<1, [A8_Pipe1]>,
152 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
153 InstrStage<1, [A8_LdSt0]>], [3, 1, 1]>,
155 // Immediate offset with update
156 InstrItinData<IIC_iStoreiu , [InstrStage<1, [A8_Issue], 0>,
157 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
158 InstrStage<1, [A8_LdSt0]>], [2, 3, 1]>,
160 // Register offset with update
161 InstrItinData<IIC_iStoreru , [InstrStage<1, [A8_Issue], 0>,
162 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
163 InstrStage<1, [A8_LdSt0]>], [2, 3, 1, 1]>,
165 // Scaled register offset with update, issues over 2 cycles
166 InstrItinData<IIC_iStoresiu, [InstrStage<2, [A8_Issue], 0>,
167 InstrStage<1, [A8_Pipe0], 0>,
168 InstrStage<1, [A8_Pipe1]>,
169 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
170 InstrStage<1, [A8_LdSt0]>], [3, 3, 1, 1]>,
173 InstrItinData<IIC_iStorem , [InstrStage<2, [A8_Issue], 0>,
174 InstrStage<2, [A8_Pipe0], 0>,
175 InstrStage<2, [A8_Pipe1]>,
176 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
177 InstrStage<1, [A8_LdSt0]>]>,
181 // no delay slots, so the latency of a branch is unimportant
182 InstrItinData<IIC_Br , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>]>,
185 // Issue through integer pipeline, and execute in NEON unit. We assume
186 // RunFast mode so that NFP pipeline is used for single-precision when
189 // FP Special Register to Integer Register File Move
190 InstrItinData<IIC_fpSTAT , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
191 InstrStage<1, [A8_NLSPipe]>]>,
193 // Single-precision FP Unary
194 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
195 InstrStage<1, [A8_NPipe]>], [7, 1]>,
197 // Double-precision FP Unary
198 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
199 InstrStage<4, [A8_NPipe], 0>,
200 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
202 // Single-precision FP Compare
203 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
204 InstrStage<1, [A8_NPipe]>], [1, 1]>,
206 // Double-precision FP Compare
207 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
208 InstrStage<4, [A8_NPipe], 0>,
209 InstrStage<4, [A8_NLSPipe]>], [4, 1]>,
211 // Single to Double FP Convert
212 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
213 InstrStage<7, [A8_NPipe], 0>,
214 InstrStage<7, [A8_NLSPipe]>], [7, 1]>,
216 // Double to Single FP Convert
217 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
218 InstrStage<5, [A8_NPipe], 0>,
219 InstrStage<5, [A8_NLSPipe]>], [5, 1]>,
221 // Single-Precision FP to Integer Convert
222 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
223 InstrStage<1, [A8_NPipe]>], [7, 1]>,
225 // Double-Precision FP to Integer Convert
226 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
227 InstrStage<8, [A8_NPipe], 0>,
228 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
230 // Integer to Single-Precision FP Convert
231 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
232 InstrStage<1, [A8_NPipe]>], [7, 1]>,
234 // Integer to Double-Precision FP Convert
235 InstrItinData<IIC_fpCVTID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
236 InstrStage<8, [A8_NPipe], 0>,
237 InstrStage<8, [A8_NLSPipe]>], [8, 1]>,
239 // Single-precision FP ALU
240 InstrItinData<IIC_fpALU32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
241 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
243 // Double-precision FP ALU
244 InstrItinData<IIC_fpALU64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
245 InstrStage<9, [A8_NPipe], 0>,
246 InstrStage<9, [A8_NLSPipe]>], [9, 1, 1]>,
248 // Single-precision FP Multiply
249 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
250 InstrStage<1, [A8_NPipe]>], [7, 1, 1]>,
252 // Double-precision FP Multiply
253 InstrItinData<IIC_fpMUL64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
254 InstrStage<11, [A8_NPipe], 0>,
255 InstrStage<11, [A8_NLSPipe]>], [11, 1, 1]>,
257 // Single-precision FP MAC
258 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
259 InstrStage<1, [A8_NPipe]>], [7, 2, 1, 1]>,
261 // Double-precision FP MAC
262 InstrItinData<IIC_fpMAC64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
263 InstrStage<19, [A8_NPipe], 0>,
264 InstrStage<19, [A8_NLSPipe]>], [19, 2, 1, 1]>,
266 // Single-precision FP DIV
267 InstrItinData<IIC_fpDIV32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
268 InstrStage<20, [A8_NPipe], 0>,
269 InstrStage<20, [A8_NLSPipe]>], [20, 1, 1]>,
271 // Double-precision FP DIV
272 InstrItinData<IIC_fpDIV64 , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
273 InstrStage<29, [A8_NPipe], 0>,
274 InstrStage<29, [A8_NLSPipe]>], [29, 1, 1]>,
276 // Single-precision FP SQRT
277 InstrItinData<IIC_fpSQRT32, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
278 InstrStage<19, [A8_NPipe], 0>,
279 InstrStage<19, [A8_NLSPipe]>], [19, 1]>,
281 // Double-precision FP SQRT
282 InstrItinData<IIC_fpSQRT64, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
283 InstrStage<29, [A8_NPipe], 0>,
284 InstrStage<29, [A8_NLSPipe]>], [29, 1]>,
286 // Single-precision FP Load
287 // use A8_Issue to enforce the 1 load/store per cycle limit
288 InstrItinData<IIC_fpLoad32, [InstrStage<1, [A8_Issue], 0>,
289 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
290 InstrStage<1, [A8_LdSt0], 0>,
291 InstrStage<1, [A8_NLSPipe]>]>,
293 // Double-precision FP Load
294 // use A8_Issue to enforce the 1 load/store per cycle limit
295 InstrItinData<IIC_fpLoad64, [InstrStage<2, [A8_Issue], 0>,
296 InstrStage<1, [A8_Pipe0], 0>,
297 InstrStage<1, [A8_Pipe1]>,
298 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
299 InstrStage<1, [A8_LdSt0], 0>,
300 InstrStage<1, [A8_NLSPipe]>]>,
303 // use A8_Issue to enforce the 1 load/store per cycle limit
304 InstrItinData<IIC_fpLoadm, [InstrStage<3, [A8_Issue], 0>,
305 InstrStage<2, [A8_Pipe0], 0>,
306 InstrStage<2, [A8_Pipe1]>,
307 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
308 InstrStage<1, [A8_LdSt0], 0>,
309 InstrStage<1, [A8_NLSPipe]>]>,
311 // Single-precision FP Store
312 // use A8_Issue to enforce the 1 load/store per cycle limit
313 InstrItinData<IIC_fpStore32,[InstrStage<1, [A8_Issue], 0>,
314 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
315 InstrStage<1, [A8_LdSt0], 0>,
316 InstrStage<1, [A8_NLSPipe]>]>,
318 // Double-precision FP Store
319 // use A8_Issue to enforce the 1 load/store per cycle limit
320 InstrItinData<IIC_fpStore64,[InstrStage<2, [A8_Issue], 0>,
321 InstrStage<1, [A8_Pipe0], 0>,
322 InstrStage<1, [A8_Pipe1]>,
323 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
324 InstrStage<1, [A8_LdSt0], 0>,
325 InstrStage<1, [A8_NLSPipe]>]>,
328 // use A8_Issue to enforce the 1 load/store per cycle limit
329 InstrItinData<IIC_fpStorem, [InstrStage<3, [A8_Issue], 0>,
330 InstrStage<2, [A8_Pipe0], 0>,
331 InstrStage<2, [A8_Pipe1]>,
332 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
333 InstrStage<1, [A8_LdSt0], 0>,
334 InstrStage<1, [A8_NLSPipe]>]>,
337 // Issue through integer pipeline, and execute in NEON unit.
340 // FIXME: We don't model this instruction properly
341 InstrItinData<IIC_VLD1, [InstrStage<1, [A8_Issue], 0>,
342 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
343 InstrStage<1, [A8_LdSt0], 0>,
344 InstrStage<1, [A8_NLSPipe]>]>,
347 // FIXME: We don't model this instruction properly
348 InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Issue], 0>,
349 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
350 InstrStage<1, [A8_LdSt0], 0>,
351 InstrStage<1, [A8_NLSPipe]>], [2, 2, 1]>,
354 // FIXME: We don't model this instruction properly
355 InstrItinData<IIC_VLD3, [InstrStage<1, [A8_Issue], 0>,
356 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
357 InstrStage<1, [A8_LdSt0], 0>,
358 InstrStage<1, [A8_NLSPipe]>], [2, 2, 2, 1]>,
361 // FIXME: We don't model this instruction properly
362 InstrItinData<IIC_VLD4, [InstrStage<1, [A8_Issue], 0>,
363 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
364 InstrStage<1, [A8_LdSt0], 0>,
365 InstrStage<1, [A8_NLSPipe]>], [2, 2, 2, 2, 1]>,
368 // FIXME: We don't model this instruction properly
369 InstrItinData<IIC_VST, [InstrStage<1, [A8_Issue], 0>,
370 InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
371 InstrStage<1, [A8_LdSt0], 0>,
372 InstrStage<1, [A8_NLSPipe]>]>,
374 // Double-register FP Unary
375 InstrItinData<IIC_VUNAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
376 InstrStage<1, [A8_NPipe]>], [5, 2]>,
378 // Quad-register FP Unary
379 // Result written in N5, but that is relative to the last cycle of multicycle,
380 // so we use 6 for those cases
381 InstrItinData<IIC_VUNAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
382 InstrStage<2, [A8_NPipe]>], [6, 2]>,
384 // Double-register FP Binary
385 InstrItinData<IIC_VBIND, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
386 InstrStage<1, [A8_NPipe]>], [5, 2, 2]>,
388 // Quad-register FP Binary
389 // Result written in N5, but that is relative to the last cycle of multicycle,
390 // so we use 6 for those cases
391 InstrItinData<IIC_VBINQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
392 InstrStage<2, [A8_NPipe]>], [6, 2, 2]>,
395 InstrItinData<IIC_VMOVImm, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
396 InstrStage<1, [A8_NPipe]>], [3]>,
398 // Double-register Permute Move
399 InstrItinData<IIC_VMOVD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
400 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
402 // Quad-register Permute Move
403 // Result written in N2, but that is relative to the last cycle of multicycle,
404 // so we use 3 for those cases
405 InstrItinData<IIC_VMOVQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
406 InstrStage<2, [A8_NLSPipe]>], [3, 1]>,
408 // Integer to Single-precision Move
409 InstrItinData<IIC_VMOVIS , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
410 InstrStage<1, [A8_NLSPipe]>], [2, 1]>,
412 // Integer to Double-precision Move
413 InstrItinData<IIC_VMOVID , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
414 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
416 // Single-precision to Integer Move
417 InstrItinData<IIC_VMOVSI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
418 InstrStage<1, [A8_NLSPipe]>], [20, 1]>,
420 // Double-precision to Integer Move
421 InstrItinData<IIC_VMOVDI , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
422 InstrStage<1, [A8_NLSPipe]>], [20, 20, 1]>,
424 // Integer to Lane Move
425 InstrItinData<IIC_VMOVISL , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
426 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
428 // Double-register Permute
429 InstrItinData<IIC_VPERMD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
430 InstrStage<1, [A8_NLSPipe]>], [2, 2, 1, 1]>,
432 // Quad-register Permute
433 // Result written in N2, but that is relative to the last cycle of multicycle,
434 // so we use 3 for those cases
435 InstrItinData<IIC_VPERMQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
436 InstrStage<2, [A8_NLSPipe]>], [3, 3, 1, 1]>,
438 // Quad-register Permute (3 cycle issue)
439 // Result written in N2, but that is relative to the last cycle of multicycle,
440 // so we use 4 for those cases
441 InstrItinData<IIC_VPERMQ3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
442 InstrStage<1, [A8_NLSPipe]>,
443 InstrStage<1, [A8_NPipe], 0>,
444 InstrStage<2, [A8_NLSPipe]>], [4, 4, 1, 1]>,
446 // Double-register FP Multiple-Accumulate
447 InstrItinData<IIC_VMACD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
448 InstrStage<1, [A8_NPipe]>], [9, 3, 2, 2]>,
450 // Quad-register FP Multiple-Accumulate
451 // Result written in N9, but that is relative to the last cycle of multicycle,
452 // so we use 10 for those cases
453 InstrItinData<IIC_VMACQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
454 InstrStage<2, [A8_NPipe]>], [10, 3, 2, 2]>,
456 // Double-register Reciprical Step
457 InstrItinData<IIC_VRECSD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
458 InstrStage<1, [A8_NPipe]>], [9, 2, 2]>,
460 // Quad-register Reciprical Step
461 InstrItinData<IIC_VRECSQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
462 InstrStage<2, [A8_NPipe]>], [10, 2, 2]>,
464 // Double-register Integer Count
465 InstrItinData<IIC_VCNTiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
466 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
468 // Quad-register Integer Count
469 // Result written in N3, but that is relative to the last cycle of multicycle,
470 // so we use 4 for those cases
471 InstrItinData<IIC_VCNTiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
472 InstrStage<2, [A8_NPipe]>], [4, 2, 2]>,
474 // Double-register Integer Unary
475 InstrItinData<IIC_VUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
476 InstrStage<1, [A8_NPipe]>], [4, 2]>,
478 // Quad-register Integer Unary
479 InstrItinData<IIC_VUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
480 InstrStage<1, [A8_NPipe]>], [4, 2]>,
482 // Double-register Integer Q-Unary
483 InstrItinData<IIC_VQUNAiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
484 InstrStage<1, [A8_NPipe]>], [4, 1]>,
486 // Quad-register Integer CountQ-Unary
487 InstrItinData<IIC_VQUNAiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
488 InstrStage<1, [A8_NPipe]>], [4, 1]>,
490 // Double-register Integer Binary
491 InstrItinData<IIC_VBINiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
492 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
494 // Quad-register Integer Binary
495 InstrItinData<IIC_VBINiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
496 InstrStage<1, [A8_NPipe]>], [3, 2, 2]>,
498 // Double-register Integer Binary (4 cycle)
499 InstrItinData<IIC_VBINi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
500 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
502 // Quad-register Integer Binary (4 cycle)
503 InstrItinData<IIC_VBINi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
504 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
507 // Double-register Integer Subtract
508 InstrItinData<IIC_VSUBiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
509 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
511 // Quad-register Integer Subtract
512 InstrItinData<IIC_VSUBiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
513 InstrStage<1, [A8_NPipe]>], [3, 2, 1]>,
515 // Double-register Integer Subtract
516 InstrItinData<IIC_VSUBi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
517 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
519 // Quad-register Integer Subtract
520 InstrItinData<IIC_VSUBi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
521 InstrStage<1, [A8_NPipe]>], [4, 2, 1]>,
523 // Double-register Integer Shift
524 InstrItinData<IIC_VSHLiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
525 InstrStage<1, [A8_NPipe]>], [3, 1, 1]>,
527 // Quad-register Integer Shift
528 InstrItinData<IIC_VSHLiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
529 InstrStage<2, [A8_NPipe]>], [4, 1, 1]>,
531 // Double-register Integer Shift (4 cycle)
532 InstrItinData<IIC_VSHLi4D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
533 InstrStage<1, [A8_NPipe]>], [4, 1, 1]>,
535 // Quad-register Integer Shift (4 cycle)
536 InstrItinData<IIC_VSHLi4Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
537 InstrStage<2, [A8_NPipe]>], [5, 1, 1]>,
539 // Double-register Integer Pair Add Long
540 InstrItinData<IIC_VPALiD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
541 InstrStage<1, [A8_NPipe]>], [6, 3, 1]>,
543 // Quad-register Integer Pair Add Long
544 InstrItinData<IIC_VPALiQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
545 InstrStage<2, [A8_NPipe]>], [7, 3, 1]>,
547 // Double-register Absolute Difference and Accumulate
548 InstrItinData<IIC_VABAD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
549 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 1]>,
551 // Quad-register Absolute Difference and Accumulate
552 InstrItinData<IIC_VABAQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
553 InstrStage<2, [A8_NPipe]>], [6, 3, 2, 1]>,
556 // Double-register Integer Multiply (.8, .16)
557 InstrItinData<IIC_VMULi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
558 InstrStage<1, [A8_NPipe]>], [6, 2, 2]>,
560 // Double-register Integer Multiply (.32)
561 InstrItinData<IIC_VMULi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
562 InstrStage<2, [A8_NPipe]>], [7, 2, 1]>,
564 // Quad-register Integer Multiply (.8, .16)
565 InstrItinData<IIC_VMULi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
566 InstrStage<2, [A8_NPipe]>], [7, 2, 2]>,
568 // Quad-register Integer Multiply (.32)
569 InstrItinData<IIC_VMULi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
570 InstrStage<1, [A8_NPipe]>,
571 InstrStage<2, [A8_NLSPipe], 0>,
572 InstrStage<3, [A8_NPipe]>], [9, 2, 1]>,
574 // Double-register Integer Multiply-Accumulate (.8, .16)
575 InstrItinData<IIC_VMACi16D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
576 InstrStage<1, [A8_NPipe]>], [6, 3, 2, 2]>,
578 // Double-register Integer Multiply-Accumulate (.32)
579 InstrItinData<IIC_VMACi32D, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
580 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 1]>,
582 // Quad-register Integer Multiply-Accumulate (.8, .16)
583 InstrItinData<IIC_VMACi16Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
584 InstrStage<2, [A8_NPipe]>], [7, 3, 2, 2]>,
586 // Quad-register Integer Multiply-Accumulate (.32)
587 InstrItinData<IIC_VMACi32Q, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
588 InstrStage<1, [A8_NPipe]>,
589 InstrStage<2, [A8_NLSPipe], 0>,
590 InstrStage<3, [A8_NPipe]>], [9, 3, 2, 1]>,
592 // Double-register VEXT
593 InstrItinData<IIC_VEXTD, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
594 InstrStage<1, [A8_NLSPipe]>], [2, 1, 1]>,
596 // Quad-register VEXT
597 InstrItinData<IIC_VEXTQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
598 InstrStage<2, [A8_NLSPipe]>], [3, 1, 1]>,
601 InstrItinData<IIC_VTB1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
602 InstrStage<2, [A8_NLSPipe]>], [3, 2, 1]>,
603 InstrItinData<IIC_VTB2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
604 InstrStage<2, [A8_NLSPipe]>], [3, 2, 2, 1]>,
605 InstrItinData<IIC_VTB3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
606 InstrStage<1, [A8_NLSPipe]>,
607 InstrStage<1, [A8_NPipe], 0>,
608 InstrStage<2, [A8_NLSPipe]>], [4, 2, 2, 3, 1]>,
609 InstrItinData<IIC_VTB4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
610 InstrStage<1, [A8_NLSPipe]>,
611 InstrStage<1, [A8_NPipe], 0>,
612 InstrStage<2, [A8_NLSPipe]>],[4, 2, 2, 3, 3, 1]>,
615 InstrItinData<IIC_VTBX1, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
616 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 1]>,
617 InstrItinData<IIC_VTBX2, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
618 InstrStage<2, [A8_NLSPipe]>], [3, 1, 2, 2, 1]>,
619 InstrItinData<IIC_VTBX3, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
620 InstrStage<1, [A8_NLSPipe]>,
621 InstrStage<1, [A8_NPipe], 0>,
622 InstrStage<2, [A8_NLSPipe]>],[4, 1, 2, 2, 3, 1]>,
623 InstrItinData<IIC_VTBX4, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
624 InstrStage<1, [A8_NLSPipe]>,
625 InstrStage<1, [A8_NPipe], 0>,
626 InstrStage<2, [A8_NLSPipe]>], [4, 1, 2, 2, 3, 3, 1]>