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 // Unary Instructions that produce a result
34 InstrItinData<IIC_iUNAr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
35 InstrItinData<IIC_iUNAsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
36 InstrItinData<IIC_iUNAsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
38 // Zero and sign extension instructions
39 InstrItinData<IIC_iEXTr , [InstrStage<1, [V6_Pipe]>], [1, 1]>,
40 InstrItinData<IIC_iEXTAr , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
42 // Compare instructions
43 InstrItinData<IIC_iCMPi , [InstrStage<1, [V6_Pipe]>], [2]>,
44 InstrItinData<IIC_iCMPr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
45 InstrItinData<IIC_iCMPsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
46 InstrItinData<IIC_iCMPsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
48 // Move instructions, unconditional
49 InstrItinData<IIC_iMOVi , [InstrStage<1, [V6_Pipe]>], [2]>,
50 InstrItinData<IIC_iMOVr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
51 InstrItinData<IIC_iMOVsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
52 InstrItinData<IIC_iMOVsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
53 InstrItinData<IIC_iMOVix2 , [InstrStage<1, [V6_Pipe]>,
54 InstrStage<1, [V6_Pipe]>], [2]>,
56 // Move instructions, conditional
57 InstrItinData<IIC_iCMOVi , [InstrStage<1, [V6_Pipe]>], [3]>,
58 InstrItinData<IIC_iCMOVr , [InstrStage<1, [V6_Pipe]>], [3, 2]>,
59 InstrItinData<IIC_iCMOVsi , [InstrStage<1, [V6_Pipe]>], [3, 1]>,
60 InstrItinData<IIC_iCMOVsr , [InstrStage<1, [V6_Pipe]>], [4, 2, 1]>,
62 // Integer multiply pipeline
64 InstrItinData<IIC_iMUL16 , [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
65 InstrItinData<IIC_iMAC16 , [InstrStage<1, [V6_Pipe]>], [4, 1, 1, 2]>,
66 InstrItinData<IIC_iMUL32 , [InstrStage<2, [V6_Pipe]>], [5, 1, 1]>,
67 InstrItinData<IIC_iMAC32 , [InstrStage<2, [V6_Pipe]>], [5, 1, 1, 2]>,
68 InstrItinData<IIC_iMUL64 , [InstrStage<3, [V6_Pipe]>], [6, 1, 1]>,
69 InstrItinData<IIC_iMAC64 , [InstrStage<3, [V6_Pipe]>], [6, 1, 1, 2]>,
71 // Integer load pipeline
74 InstrItinData<IIC_iLoadi , [InstrStage<1, [V6_Pipe]>], [4, 1]>,
77 InstrItinData<IIC_iLoadr , [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
79 // Scaled register offset, issues over 2 cycles
80 InstrItinData<IIC_iLoadsi , [InstrStage<2, [V6_Pipe]>], [5, 2, 1]>,
82 // Immediate offset with update
83 InstrItinData<IIC_iLoadiu , [InstrStage<1, [V6_Pipe]>], [4, 2, 1]>,
85 // Register offset with update
86 InstrItinData<IIC_iLoadru , [InstrStage<1, [V6_Pipe]>], [4, 2, 1, 1]>,
88 // Scaled register offset with update, issues over 2 cycles
89 InstrItinData<IIC_iLoadsiu , [InstrStage<2, [V6_Pipe]>], [5, 2, 2, 1]>,
93 InstrItinData<IIC_iLoadm , [InstrStage<3, [V6_Pipe]>]>,
96 // Load multiple plus branch
97 InstrItinData<IIC_iLoadmBr , [InstrStage<3, [V6_Pipe]>,
98 InstrStage<1, [V6_Pipe]>]>,
101 // iLoadi + iALUr for t2LDRpci_pic.
102 InstrItinData<IIC_iLoadiALU, [InstrStage<1, [V6_Pipe]>,
103 InstrStage<1, [V6_Pipe]>], [3, 1]>,
105 // Integer store pipeline
108 InstrItinData<IIC_iStorei , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
111 InstrItinData<IIC_iStorer , [InstrStage<1, [V6_Pipe]>], [2, 1, 1]>,
114 // Scaled register offset, issues over 2 cycles
115 InstrItinData<IIC_iStoresi , [InstrStage<2, [V6_Pipe]>], [2, 2, 1]>,
117 // Immediate offset with update
118 InstrItinData<IIC_iStoreiu , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
120 // Register offset with update
121 InstrItinData<IIC_iStoreru , [InstrStage<1, [V6_Pipe]>], [2, 2, 1, 1]>,
123 // Scaled register offset with update, issues over 2 cycles
124 InstrItinData<IIC_iStoresiu, [InstrStage<2, [V6_Pipe]>], [2, 2, 2, 1]>,
127 InstrItinData<IIC_iStorem , [InstrStage<3, [V6_Pipe]>]>,
131 // no delay slots, so the latency of a branch is unimportant
132 InstrItinData<IIC_Br , [InstrStage<1, [V6_Pipe]>]>,
135 // Issue through integer pipeline, and execute in NEON unit. We assume
136 // RunFast mode so that NFP pipeline is used for single-precision when
139 // FP Special Register to Integer Register File Move
140 InstrItinData<IIC_fpSTAT , [InstrStage<1, [V6_Pipe]>], [3]>,
142 // Single-precision FP Unary
143 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
145 // Double-precision FP Unary
146 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
148 // Single-precision FP Compare
149 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
151 // Double-precision FP Compare
152 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
154 // Single to Double FP Convert
155 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
157 // Double to Single FP Convert
158 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
160 // Single-Precision FP to Integer Convert
161 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
163 // Double-Precision FP to Integer Convert
164 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
166 // Integer to Single-Precision FP Convert
167 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
169 // Integer to Double-Precision FP Convert
170 InstrItinData<IIC_fpCVTID , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
172 // Single-precision FP ALU
173 InstrItinData<IIC_fpALU32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
175 // Double-precision FP ALU
176 InstrItinData<IIC_fpALU64 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
178 // Single-precision FP Multiply
179 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
181 // Double-precision FP Multiply
182 InstrItinData<IIC_fpMUL64 , [InstrStage<2, [V6_Pipe]>], [9, 2, 2]>,
184 // Single-precision FP MAC
185 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2, 2]>,
187 // Double-precision FP MAC
188 InstrItinData<IIC_fpMAC64 , [InstrStage<2, [V6_Pipe]>], [9, 2, 2, 2]>,
190 // Single-precision FP DIV
191 InstrItinData<IIC_fpDIV32 , [InstrStage<15, [V6_Pipe]>], [20, 2, 2]>,
193 // Double-precision FP DIV
194 InstrItinData<IIC_fpDIV64 , [InstrStage<29, [V6_Pipe]>], [34, 2, 2]>,
196 // Single-precision FP SQRT
197 InstrItinData<IIC_fpSQRT32 , [InstrStage<15, [V6_Pipe]>], [20, 2, 2]>,
199 // Double-precision FP SQRT
200 InstrItinData<IIC_fpSQRT64 , [InstrStage<29, [V6_Pipe]>], [34, 2, 2]>,
202 // Single-precision FP Load
203 InstrItinData<IIC_fpLoad32 , [InstrStage<1, [V6_Pipe]>], [5, 2, 2]>,
205 // Double-precision FP Load
206 InstrItinData<IIC_fpLoad64 , [InstrStage<1, [V6_Pipe]>], [5, 2, 2]>,
209 InstrItinData<IIC_fpLoadm , [InstrStage<3, [V6_Pipe]>]>,
211 // Single-precision FP Store
212 InstrItinData<IIC_fpStore32 , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
214 // Double-precision FP Store
215 // use FU_Issue to enforce the 1 load/store per cycle limit
216 InstrItinData<IIC_fpStore64 , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
219 InstrItinData<IIC_fpStorem , [InstrStage<3, [V6_Pipe]>]>