1 //===- ARMScheduleV6.td - ARM v6 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 v6 processors.
12 //===----------------------------------------------------------------------===//
14 // Model based on ARM1176
17 def V6_Pipe : FuncUnit; // pipeline
19 // Scheduling information derived from "ARM1176JZF-S Technical Reference Manual"
21 def ARMV6Itineraries : ProcessorItineraries<
25 InstrItinData<IIC_iALUx , [InstrStage<1, [V6_Pipe]>]>,
27 // Binary Instructions that produce a result
28 InstrItinData<IIC_iALUi , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
29 InstrItinData<IIC_iALUr , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
30 InstrItinData<IIC_iALUsi , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
31 InstrItinData<IIC_iALUsr , [InstrStage<2, [V6_Pipe]>], [3, 3, 2, 1]>,
33 // Bitwise Instructions that produce a result
34 InstrItinData<IIC_iBITi , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
35 InstrItinData<IIC_iBITr , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
36 InstrItinData<IIC_iBITsi , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
37 InstrItinData<IIC_iBITsr , [InstrStage<2, [V6_Pipe]>], [3, 3, 2, 1]>,
39 // Unary Instructions that produce a result
40 InstrItinData<IIC_iUNAr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
41 InstrItinData<IIC_iUNAsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
43 // Zero and sign extension instructions
44 InstrItinData<IIC_iEXTr , [InstrStage<1, [V6_Pipe]>], [1, 1]>,
45 InstrItinData<IIC_iEXTAr , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
46 InstrItinData<IIC_iEXTAsr , [InstrStage<2, [V6_Pipe]>], [3, 3, 2, 1]>,
48 // Compare instructions
49 InstrItinData<IIC_iCMPi , [InstrStage<1, [V6_Pipe]>], [2]>,
50 InstrItinData<IIC_iCMPr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
51 InstrItinData<IIC_iCMPsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
52 InstrItinData<IIC_iCMPsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
55 InstrItinData<IIC_iTSTi , [InstrStage<1, [V6_Pipe]>], [2]>,
56 InstrItinData<IIC_iTSTr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
57 InstrItinData<IIC_iTSTsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
58 InstrItinData<IIC_iTSTsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
60 // Move instructions, unconditional
61 InstrItinData<IIC_iMOVi , [InstrStage<1, [V6_Pipe]>], [2]>,
62 InstrItinData<IIC_iMOVr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
63 InstrItinData<IIC_iMOVsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
64 InstrItinData<IIC_iMOVsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
65 InstrItinData<IIC_iMOVix2 , [InstrStage<1, [V6_Pipe]>,
66 InstrStage<1, [V6_Pipe]>], [2]>,
68 // Move instructions, conditional
69 InstrItinData<IIC_iCMOVi , [InstrStage<1, [V6_Pipe]>], [3]>,
70 InstrItinData<IIC_iCMOVr , [InstrStage<1, [V6_Pipe]>], [3, 2]>,
71 InstrItinData<IIC_iCMOVsi , [InstrStage<1, [V6_Pipe]>], [3, 1]>,
72 InstrItinData<IIC_iCMOVsr , [InstrStage<1, [V6_Pipe]>], [4, 2, 1]>,
73 InstrItinData<IIC_iCMOVix2 , [InstrStage<1, [V6_Pipe]>,
74 InstrStage<1, [V6_Pipe]>], [4]>,
77 InstrItinData<IIC_iMVNi , [InstrStage<1, [V6_Pipe]>], [2]>,
78 InstrItinData<IIC_iMVNr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
79 InstrItinData<IIC_iMVNsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
80 InstrItinData<IIC_iMVNsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
82 // Integer multiply pipeline
84 InstrItinData<IIC_iMUL16 , [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
85 InstrItinData<IIC_iMAC16 , [InstrStage<1, [V6_Pipe]>], [4, 1, 1, 2]>,
86 InstrItinData<IIC_iMUL32 , [InstrStage<2, [V6_Pipe]>], [5, 1, 1]>,
87 InstrItinData<IIC_iMAC32 , [InstrStage<2, [V6_Pipe]>], [5, 1, 1, 2]>,
88 InstrItinData<IIC_iMUL64 , [InstrStage<3, [V6_Pipe]>], [6, 1, 1]>,
89 InstrItinData<IIC_iMAC64 , [InstrStage<3, [V6_Pipe]>], [6, 1, 1, 2]>,
91 // Integer load pipeline
94 InstrItinData<IIC_iLoad_i , [InstrStage<1, [V6_Pipe]>], [4, 1]>,
95 InstrItinData<IIC_iLoad_bh_i, [InstrStage<1, [V6_Pipe]>], [4, 1]>,
96 InstrItinData<IIC_iLoad_d_i , [InstrStage<1, [V6_Pipe]>], [4, 1]>,
99 InstrItinData<IIC_iLoad_r , [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
100 InstrItinData<IIC_iLoad_bh_r, [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
101 InstrItinData<IIC_iLoad_d_r , [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
103 // Scaled register offset, issues over 2 cycles
104 InstrItinData<IIC_iLoad_si , [InstrStage<2, [V6_Pipe]>], [5, 2, 1]>,
105 InstrItinData<IIC_iLoad_bh_si, [InstrStage<2, [V6_Pipe]>], [5, 2, 1]>,
107 // Immediate offset with update
108 InstrItinData<IIC_iLoad_iu , [InstrStage<1, [V6_Pipe]>], [4, 2, 1]>,
109 InstrItinData<IIC_iLoad_bh_iu, [InstrStage<1, [V6_Pipe]>], [4, 2, 1]>,
111 // Register offset with update
112 InstrItinData<IIC_iLoad_ru , [InstrStage<1, [V6_Pipe]>], [4, 2, 1, 1]>,
113 InstrItinData<IIC_iLoad_bh_ru, [InstrStage<1, [V6_Pipe]>], [4, 2, 1, 1]>,
114 InstrItinData<IIC_iLoad_d_ru , [InstrStage<1, [V6_Pipe]>], [4, 2, 1, 1]>,
116 // Scaled register offset with update, issues over 2 cycles
117 InstrItinData<IIC_iLoad_siu, [InstrStage<2, [V6_Pipe]>], [5, 2, 2, 1]>,
118 InstrItinData<IIC_iLoad_bh_siu,[InstrStage<2, [V6_Pipe]>], [5, 2, 2, 1]>,
121 // Load multiple, def is the 5th operand.
122 InstrItinData<IIC_iLoad_m , [InstrStage<3, [V6_Pipe]>], [1, 1, 1, 1, 4]>,
124 // Load multiple + update, defs are the 1st and 5th operands.
125 InstrItinData<IIC_iLoad_mu , [InstrStage<3, [V6_Pipe]>], [2, 1, 1, 1, 4]>,
127 // Load multiple plus branch
128 InstrItinData<IIC_iLoad_mBr, [InstrStage<3, [V6_Pipe]>,
129 InstrStage<1, [V6_Pipe]>], [1, 2, 1, 1, 4]>,
132 // iLoadi + iALUr for t2LDRpci_pic.
133 InstrItinData<IIC_iLoadiALU, [InstrStage<1, [V6_Pipe]>,
134 InstrStage<1, [V6_Pipe]>], [3, 1]>,
137 // Pop, def is the 3rd operand.
138 InstrItinData<IIC_iPop , [InstrStage<3, [V6_Pipe]>], [1, 1, 4]>,
140 // Pop + branch, def is the 3rd operand.
141 InstrItinData<IIC_iPop_Br, [InstrStage<3, [V6_Pipe]>,
142 InstrStage<1, [V6_Pipe]>], [1, 2, 4]>,
144 // Integer store pipeline
147 InstrItinData<IIC_iStore_i , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
148 InstrItinData<IIC_iStore_bh_i, [InstrStage<1, [V6_Pipe]>], [2, 1]>,
149 InstrItinData<IIC_iStore_d_i , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
152 InstrItinData<IIC_iStore_r , [InstrStage<1, [V6_Pipe]>], [2, 1, 1]>,
153 InstrItinData<IIC_iStore_bh_r, [InstrStage<1, [V6_Pipe]>], [2, 1, 1]>,
154 InstrItinData<IIC_iStore_d_r , [InstrStage<1, [V6_Pipe]>], [2, 1, 1]>,
156 // Scaled register offset, issues over 2 cycles
157 InstrItinData<IIC_iStore_si , [InstrStage<2, [V6_Pipe]>], [2, 2, 1]>,
158 InstrItinData<IIC_iStore_bh_si, [InstrStage<2, [V6_Pipe]>], [2, 2, 1]>,
160 // Immediate offset with update
161 InstrItinData<IIC_iStore_iu , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
162 InstrItinData<IIC_iStore_bh_iu, [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
164 // Register offset with update
165 InstrItinData<IIC_iStore_ru, [InstrStage<1, [V6_Pipe]>], [2, 2, 1, 1]>,
166 InstrItinData<IIC_iStore_bh_ru,[InstrStage<1, [V6_Pipe]>], [2, 2, 1, 1]>,
167 InstrItinData<IIC_iStore_d_ru, [InstrStage<1, [V6_Pipe]>], [2, 2, 1, 1]>,
169 // Scaled register offset with update, issues over 2 cycles
170 InstrItinData<IIC_iStore_siu, [InstrStage<2, [V6_Pipe]>], [2, 2, 2, 1]>,
171 InstrItinData<IIC_iStore_bh_siu,[InstrStage<2, [V6_Pipe]>], [2, 2, 2, 1]>,
174 InstrItinData<IIC_iStore_m , [InstrStage<3, [V6_Pipe]>]>,
176 // Store multiple + update
177 InstrItinData<IIC_iStore_mu , [InstrStage<3, [V6_Pipe]>], [2]>,
181 // no delay slots, so the latency of a branch is unimportant
182 InstrItinData<IIC_Br , [InstrStage<1, [V6_Pipe]>]>,
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, [V6_Pipe]>], [3]>,
192 // Single-precision FP Unary
193 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
195 // Double-precision FP Unary
196 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
198 // Single-precision FP Compare
199 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
201 // Double-precision FP Compare
202 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
204 // Single to Double FP Convert
205 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
207 // Double to Single FP Convert
208 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
210 // Single-Precision FP to Integer Convert
211 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
213 // Double-Precision FP to Integer Convert
214 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
216 // Integer to Single-Precision FP Convert
217 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
219 // Integer to Double-Precision FP Convert
220 InstrItinData<IIC_fpCVTID , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
222 // Single-precision FP ALU
223 InstrItinData<IIC_fpALU32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
225 // Double-precision FP ALU
226 InstrItinData<IIC_fpALU64 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
228 // Single-precision FP Multiply
229 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
231 // Double-precision FP Multiply
232 InstrItinData<IIC_fpMUL64 , [InstrStage<2, [V6_Pipe]>], [9, 2, 2]>,
234 // Single-precision FP MAC
235 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2, 2]>,
237 // Double-precision FP MAC
238 InstrItinData<IIC_fpMAC64 , [InstrStage<2, [V6_Pipe]>], [9, 2, 2, 2]>,
240 // Single-precision FP DIV
241 InstrItinData<IIC_fpDIV32 , [InstrStage<15, [V6_Pipe]>], [20, 2, 2]>,
243 // Double-precision FP DIV
244 InstrItinData<IIC_fpDIV64 , [InstrStage<29, [V6_Pipe]>], [34, 2, 2]>,
246 // Single-precision FP SQRT
247 InstrItinData<IIC_fpSQRT32 , [InstrStage<15, [V6_Pipe]>], [20, 2, 2]>,
249 // Double-precision FP SQRT
250 InstrItinData<IIC_fpSQRT64 , [InstrStage<29, [V6_Pipe]>], [34, 2, 2]>,
252 // Integer to Single-precision Move
253 InstrItinData<IIC_fpMOVIS, [InstrStage<1, [V6_Pipe]>], [10, 1]>,
255 // Integer to Double-precision Move
256 InstrItinData<IIC_fpMOVID, [InstrStage<1, [V6_Pipe]>], [10, 1, 1]>,
258 // Single-precision to Integer Move
259 InstrItinData<IIC_fpMOVSI, [InstrStage<1, [V6_Pipe]>], [10, 1]>,
261 // Double-precision to Integer Move
262 InstrItinData<IIC_fpMOVDI, [InstrStage<1, [V6_Pipe]>], [10, 10, 1]>,
264 // Single-precision FP Load
265 InstrItinData<IIC_fpLoad32 , [InstrStage<1, [V6_Pipe]>], [5, 2, 2]>,
267 // Double-precision FP Load
268 InstrItinData<IIC_fpLoad64 , [InstrStage<1, [V6_Pipe]>], [5, 2, 2]>,
271 InstrItinData<IIC_fpLoad_m , [InstrStage<3, [V6_Pipe]>], [2, 1, 1, 5]>,
273 // FP Load Multiple + update
274 InstrItinData<IIC_fpLoad_mu, [InstrStage<3, [V6_Pipe]>], [3, 2, 1, 1, 5]>,
276 // Single-precision FP Store
277 InstrItinData<IIC_fpStore32 , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
279 // Double-precision FP Store
280 // use FU_Issue to enforce the 1 load/store per cycle limit
281 InstrItinData<IIC_fpStore64 , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
284 InstrItinData<IIC_fpStore_m, [InstrStage<3, [V6_Pipe]>], [2, 2, 2, 2]>,
286 // FP Store Multiple + update
287 InstrItinData<IIC_fpStore_mu,[InstrStage<3, [V6_Pipe]>], [3, 2, 2, 2, 2]>