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]>,
37 // Zero and sign extension instructions
38 InstrItinData<IIC_iEXTr , [InstrStage<1, [V6_Pipe]>], [1, 1]>,
39 InstrItinData<IIC_iEXTAr , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
41 // Compare instructions
42 InstrItinData<IIC_iCMPi , [InstrStage<1, [V6_Pipe]>], [2]>,
43 InstrItinData<IIC_iCMPr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
44 InstrItinData<IIC_iCMPsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
45 InstrItinData<IIC_iCMPsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
47 // Move instructions, unconditional
48 InstrItinData<IIC_iMOVi , [InstrStage<1, [V6_Pipe]>], [2]>,
49 InstrItinData<IIC_iMOVr , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
50 InstrItinData<IIC_iMOVsi , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
51 InstrItinData<IIC_iMOVsr , [InstrStage<2, [V6_Pipe]>], [3, 2, 1]>,
52 InstrItinData<IIC_iMOVix2 , [InstrStage<1, [V6_Pipe]>,
53 InstrStage<1, [V6_Pipe]>], [2]>,
55 // Move instructions, conditional
56 InstrItinData<IIC_iCMOVi , [InstrStage<1, [V6_Pipe]>], [3]>,
57 InstrItinData<IIC_iCMOVr , [InstrStage<1, [V6_Pipe]>], [3, 2]>,
58 InstrItinData<IIC_iCMOVsi , [InstrStage<1, [V6_Pipe]>], [3, 1]>,
59 InstrItinData<IIC_iCMOVsr , [InstrStage<1, [V6_Pipe]>], [4, 2, 1]>,
61 // Integer multiply pipeline
63 InstrItinData<IIC_iMUL16 , [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
64 InstrItinData<IIC_iMAC16 , [InstrStage<1, [V6_Pipe]>], [4, 1, 1, 2]>,
65 InstrItinData<IIC_iMUL32 , [InstrStage<2, [V6_Pipe]>], [5, 1, 1]>,
66 InstrItinData<IIC_iMAC32 , [InstrStage<2, [V6_Pipe]>], [5, 1, 1, 2]>,
67 InstrItinData<IIC_iMUL64 , [InstrStage<3, [V6_Pipe]>], [6, 1, 1]>,
68 InstrItinData<IIC_iMAC64 , [InstrStage<3, [V6_Pipe]>], [6, 1, 1, 2]>,
70 // Integer load pipeline
73 InstrItinData<IIC_iLoadi , [InstrStage<1, [V6_Pipe]>], [4, 1]>,
76 InstrItinData<IIC_iLoadr , [InstrStage<1, [V6_Pipe]>], [4, 1, 1]>,
78 // Scaled register offset, issues over 2 cycles
79 InstrItinData<IIC_iLoadsi , [InstrStage<2, [V6_Pipe]>], [5, 2, 1]>,
81 // Immediate offset with update
82 InstrItinData<IIC_iLoadiu , [InstrStage<1, [V6_Pipe]>], [4, 2, 1]>,
84 // Register offset with update
85 InstrItinData<IIC_iLoadru , [InstrStage<1, [V6_Pipe]>], [4, 2, 1, 1]>,
87 // Scaled register offset with update, issues over 2 cycles
88 InstrItinData<IIC_iLoadsiu , [InstrStage<2, [V6_Pipe]>], [5, 2, 2, 1]>,
92 InstrItinData<IIC_iLoadm , [InstrStage<3, [V6_Pipe]>]>,
95 // Load multiple plus branch
96 InstrItinData<IIC_iLoadmBr , [InstrStage<3, [V6_Pipe]>,
97 InstrStage<1, [V6_Pipe]>]>,
100 // iLoadi + iALUr for t2LDRpci_pic.
101 InstrItinData<IIC_iLoadiALU, [InstrStage<1, [V6_Pipe]>,
102 InstrStage<1, [V6_Pipe]>], [3, 1]>,
104 // Integer store pipeline
107 InstrItinData<IIC_iStorei , [InstrStage<1, [V6_Pipe]>], [2, 1]>,
110 InstrItinData<IIC_iStorer , [InstrStage<1, [V6_Pipe]>], [2, 1, 1]>,
113 // Scaled register offset, issues over 2 cycles
114 InstrItinData<IIC_iStoresi , [InstrStage<2, [V6_Pipe]>], [2, 2, 1]>,
116 // Immediate offset with update
117 InstrItinData<IIC_iStoreiu , [InstrStage<1, [V6_Pipe]>], [2, 2, 1]>,
119 // Register offset with update
120 InstrItinData<IIC_iStoreru , [InstrStage<1, [V6_Pipe]>], [2, 2, 1, 1]>,
122 // Scaled register offset with update, issues over 2 cycles
123 InstrItinData<IIC_iStoresiu, [InstrStage<2, [V6_Pipe]>], [2, 2, 2, 1]>,
126 InstrItinData<IIC_iStorem , [InstrStage<3, [V6_Pipe]>]>,
130 // no delay slots, so the latency of a branch is unimportant
131 InstrItinData<IIC_Br , [InstrStage<1, [V6_Pipe]>]>,
134 // Issue through integer pipeline, and execute in NEON unit. We assume
135 // RunFast mode so that NFP pipeline is used for single-precision when
138 // FP Special Register to Integer Register File Move
139 InstrItinData<IIC_fpSTAT , [InstrStage<1, [V6_Pipe]>], [3]>,
141 // Single-precision FP Unary
142 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
144 // Double-precision FP Unary
145 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
147 // Single-precision FP Compare
148 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
150 // Double-precision FP Compare
151 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [V6_Pipe]>], [2, 2]>,
153 // Single to Double FP Convert
154 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
156 // Double to Single FP Convert
157 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [V6_Pipe]>], [5, 2]>,
159 // Single-Precision FP to Integer Convert
160 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
162 // Double-Precision FP to Integer Convert
163 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
165 // Integer to Single-Precision FP Convert
166 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
168 // Integer to Double-Precision FP Convert
169 InstrItinData<IIC_fpCVTID , [InstrStage<1, [V6_Pipe]>], [9, 2]>,
171 // Single-precision FP ALU
172 InstrItinData<IIC_fpALU32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
174 // Double-precision FP ALU
175 InstrItinData<IIC_fpALU64 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
177 // Single-precision FP Multiply
178 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2]>,
180 // Double-precision FP Multiply
181 InstrItinData<IIC_fpMUL64 , [InstrStage<2, [V6_Pipe]>], [9, 2, 2]>,
183 // Single-precision FP MAC
184 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [V6_Pipe]>], [9, 2, 2, 2]>,
186 // Double-precision FP MAC
187 InstrItinData<IIC_fpMAC64 , [InstrStage<2, [V6_Pipe]>], [9, 2, 2, 2]>,
189 // Single-precision FP DIV
190 InstrItinData<IIC_fpDIV32 , [InstrStage<15, [V6_Pipe]>], [20, 2, 2]>,
192 // Double-precision FP DIV
193 InstrItinData<IIC_fpDIV64 , [InstrStage<29, [V6_Pipe]>], [34, 2, 2]>,
195 // Single-precision FP SQRT
196 InstrItinData<IIC_fpSQRT32 , [InstrStage<15, [V6_Pipe]>], [20, 2, 2]>,
198 // Double-precision FP SQRT
199 InstrItinData<IIC_fpSQRT64 , [InstrStage<29, [V6_Pipe]>], [34, 2, 2]>,
201 // Single-precision FP Load
202 InstrItinData<IIC_fpLoad32 , [InstrStage<1, [V6_Pipe]>], [5, 2, 2]>,
204 // Double-precision FP Load
205 InstrItinData<IIC_fpLoad64 , [InstrStage<1, [V6_Pipe]>], [5, 2, 2]>,
208 InstrItinData<IIC_fpLoadm , [InstrStage<3, [V6_Pipe]>]>,
210 // Single-precision FP Store
211 InstrItinData<IIC_fpStore32 , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
213 // Double-precision FP Store
214 // use FU_Issue to enforce the 1 load/store per cycle limit
215 InstrItinData<IIC_fpStore64 , [InstrStage<1, [V6_Pipe]>], [2, 2, 2]>,
218 InstrItinData<IIC_fpStorem , [InstrStage<3, [V6_Pipe]>]>