14// TODO: Add model for an ARM11
| 14// Model based on ARM1176 15// 16// Scheduling information derived from "ARM1176JZF-S Technical Reference Manual". 17// 18def ARMV6Itineraries : ProcessorItineraries<[ 19 // 20 // No operand cycles 21 InstrItinData<IIC_iALUx , [InstrStage<1, [FU_Pipe0]>]>, 22 // 23 // Binary Instructions that produce a result 24 InstrItinData<IIC_iALUi , [InstrStage<1, [FU_Pipe0]>], [2, 2]>, 25 InstrItinData<IIC_iALUr , [InstrStage<1, [FU_Pipe0]>], [2, 2, 2]>, 26 InstrItinData<IIC_iALUsi , [InstrStage<1, [FU_Pipe0]>], [2, 2, 1]>, 27 InstrItinData<IIC_iALUsr , [InstrStage<2, [FU_Pipe0]>], [3, 3, 2, 1]>, 28 // 29 // Unary Instructions that produce a result 30 InstrItinData<IIC_iUNAr , [InstrStage<1, [FU_Pipe0]>], [2, 2]>, 31 InstrItinData<IIC_iUNAsi , [InstrStage<1, [FU_Pipe0]>], [2, 1]>, 32 InstrItinData<IIC_iUNAsr , [InstrStage<2, [FU_Pipe0]>], [3, 2, 1]>, 33 // 34 // Compare instructions 35 InstrItinData<IIC_iCMPi , [InstrStage<1, [FU_Pipe0]>], [2]>, 36 InstrItinData<IIC_iCMPr , [InstrStage<1, [FU_Pipe0]>], [2, 2]>, 37 InstrItinData<IIC_iCMPsi , [InstrStage<1, [FU_Pipe0]>], [2, 1]>, 38 InstrItinData<IIC_iCMPsr , [InstrStage<2, [FU_Pipe0]>], [3, 2, 1]>, 39 // 40 // Move instructions, unconditional 41 InstrItinData<IIC_iMOVi , [InstrStage<1, [FU_Pipe0]>], [2]>, 42 InstrItinData<IIC_iMOVr , [InstrStage<1, [FU_Pipe0]>], [2, 2]>, 43 InstrItinData<IIC_iMOVsi , [InstrStage<1, [FU_Pipe0]>], [2, 1]>, 44 InstrItinData<IIC_iMOVsr , [InstrStage<2, [FU_Pipe0]>], [3, 2, 1]>, 45 // 46 // Move instructions, conditional 47 InstrItinData<IIC_iCMOVi , [InstrStage<1, [FU_Pipe0]>], [3]>, 48 InstrItinData<IIC_iCMOVr , [InstrStage<1, [FU_Pipe0]>], [3, 2]>, 49 InstrItinData<IIC_iCMOVsi , [InstrStage<1, [FU_Pipe0]>], [3, 1]>, 50 InstrItinData<IIC_iCMOVsr , [InstrStage<1, [FU_Pipe0]>], [4, 2, 1]>, 51 52 // Integer multiply pipeline 53 // 54 InstrItinData<IIC_iMUL16 , [InstrStage<1, [FU_Pipe0]>], [4, 1, 1]>, 55 InstrItinData<IIC_iMAC16 , [InstrStage<1, [FU_Pipe0]>], [4, 1, 1, 2]>, 56 InstrItinData<IIC_iMUL32 , [InstrStage<2, [FU_Pipe0]>], [5, 1, 1]>, 57 InstrItinData<IIC_iMAC32 , [InstrStage<2, [FU_Pipe0]>], [5, 1, 1, 2]>, 58 InstrItinData<IIC_iMUL64 , [InstrStage<3, [FU_Pipe0]>], [6, 1, 1]>, 59 InstrItinData<IIC_iMAC64 , [InstrStage<3, [FU_Pipe0]>], [6, 1, 1, 2]>, 60 61 // Integer load pipeline 62 // 63 // Immediate offset 64 InstrItinData<IIC_iLoadi , [InstrStage<1, [FU_Pipe0]>], [4, 1]>, 65 // 66 // Register offset 67 InstrItinData<IIC_iLoadr , [InstrStage<1, [FU_Pipe0]>], [4, 1, 1]>, 68 // 69 // Scaled register offset, issues over 2 cycles 70 InstrItinData<IIC_iLoadsi , [InstrStage<2, [FU_Pipe0]>], [5, 2, 1]>, 71 // 72 // Immediate offset with update 73 InstrItinData<IIC_iLoadiu , [InstrStage<1, [FU_Pipe0]>], [4, 2, 1]>, 74 // 75 // Register offset with update 76 InstrItinData<IIC_iLoadru , [InstrStage<1, [FU_Pipe0]>], [4, 2, 1, 1]>, 77 // 78 // Scaled register offset with update, issues over 2 cycles 79 InstrItinData<IIC_iLoadsiu , [InstrStage<2, [FU_Pipe0]>], [5, 2, 2, 1]>, 80 81 // 82 // Load multiple 83 InstrItinData<IIC_iLoadm , [InstrStage<3, [FU_Pipe0]>]>, 84 85 // Integer store pipeline 86 // 87 // Immediate offset 88 InstrItinData<IIC_iStorei , [InstrStage<1, [FU_Pipe0]>], [2, 1]>, 89 // 90 // Register offset 91 InstrItinData<IIC_iStorer , [InstrStage<1, [FU_Pipe0]>], [2, 1, 1]>, 92 93 // 94 // Scaled register offset, issues over 2 cycles 95 InstrItinData<IIC_iStoresi , [InstrStage<2, [FU_Pipe0]>], [2, 2, 1]>, 96 // 97 // Immediate offset with update 98 InstrItinData<IIC_iStoreiu , [InstrStage<1, [FU_Pipe0]>], [2, 2, 1]>, 99 // 100 // Register offset with update 101 InstrItinData<IIC_iStoreru , [InstrStage<1, [FU_Pipe0]>], [2, 2, 1, 1]>, 102 // 103 // Scaled register offset with update, issues over 2 cycles 104 InstrItinData<IIC_iStoresiu, [InstrStage<2, [FU_Pipe0]>], [2, 2, 2, 1]>, 105 // 106 // Store multiple 107 InstrItinData<IIC_iStorem , [InstrStage<3, [FU_Pipe0]>]>, 108 109 // Branch 110 // 111 // no delay slots, so the latency of a branch is unimportant 112 InstrItinData<IIC_Br , [InstrStage<1, [FU_Pipe0]>]>, 113 114 // VFP 115 // Issue through integer pipeline, and execute in NEON unit. We assume 116 // RunFast mode so that NFP pipeline is used for single-precision when 117 // possible. 118 // 119 // FP Special Register to Integer Register File Move 120 InstrItinData<IIC_fpSTAT , [InstrStage<1, [FU_Pipe0]>], [3]>, 121 // 122 // Single-precision FP Unary 123 InstrItinData<IIC_fpUNA32 , [InstrStage<1, [FU_Pipe0]>], [5, 2]>, 124 // 125 // Double-precision FP Unary 126 InstrItinData<IIC_fpUNA64 , [InstrStage<1, [FU_Pipe0]>], [5, 2]>, 127 // 128 // Single-precision FP Compare 129 InstrItinData<IIC_fpCMP32 , [InstrStage<1, [FU_Pipe0]>], [2, 2]>, 130 // 131 // Double-precision FP Compare 132 InstrItinData<IIC_fpCMP64 , [InstrStage<1, [FU_Pipe0]>], [2, 2]>, 133 // 134 // Single to Double FP Convert 135 InstrItinData<IIC_fpCVTSD , [InstrStage<1, [FU_Pipe0]>], [5, 2]>, 136 // 137 // Double to Single FP Convert 138 InstrItinData<IIC_fpCVTDS , [InstrStage<1, [FU_Pipe0]>], [5, 2]>, 139 // 140 // Single-Precision FP to Integer Convert 141 InstrItinData<IIC_fpCVTSI , [InstrStage<1, [FU_Pipe0]>], [9, 2]>, 142 // 143 // Double-Precision FP to Integer Convert 144 InstrItinData<IIC_fpCVTDI , [InstrStage<1, [FU_Pipe0]>], [9, 2]>, 145 // 146 // Integer to Single-Precision FP Convert 147 InstrItinData<IIC_fpCVTIS , [InstrStage<1, [FU_Pipe0]>], [9, 2]>, 148 // 149 // Integer to Double-Precision FP Convert 150 InstrItinData<IIC_fpCVTID , [InstrStage<1, [FU_Pipe0]>], [9, 2]>, 151 // 152 // Single-precision FP ALU 153 InstrItinData<IIC_fpALU32 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2]>, 154 // 155 // Double-precision FP ALU 156 InstrItinData<IIC_fpALU64 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2]>, 157 // 158 // Single-precision FP Multiply 159 InstrItinData<IIC_fpMUL32 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2]>, 160 // 161 // Double-precision FP Multiply 162 InstrItinData<IIC_fpMUL64 , [InstrStage<2, [FU_Pipe0]>], [9, 2, 2]>, 163 // 164 // Single-precision FP MAC 165 InstrItinData<IIC_fpMAC32 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2, 2]>, 166 // 167 // Double-precision FP MAC 168 InstrItinData<IIC_fpMAC64 , [InstrStage<2, [FU_Pipe0]>], [9, 2, 2, 2]>, 169 // 170 // Single-precision FP DIV 171 InstrItinData<IIC_fpDIV32 , [InstrStage<15, [FU_Pipe0]>], [20, 2, 2]>, 172 // 173 // Double-precision FP DIV 174 InstrItinData<IIC_fpDIV64 , [InstrStage<29, [FU_Pipe0]>], [34, 2, 2]>, 175 // 176 // Single-precision FP SQRT 177 InstrItinData<IIC_fpSQRT32 , [InstrStage<15, [FU_Pipe0]>], [20, 2, 2]>, 178 // 179 // Double-precision FP SQRT 180 InstrItinData<IIC_fpSQRT64 , [InstrStage<29, [FU_Pipe0]>], [34, 2, 2]>, 181 // 182 // Single-precision FP Load 183 InstrItinData<IIC_fpLoad32 , [InstrStage<1, [FU_Pipe0]>], [5, 2, 2]>, 184 // 185 // Double-precision FP Load 186 InstrItinData<IIC_fpLoad64 , [InstrStage<1, [FU_Pipe0]>], [5, 2, 2]>, 187 // 188 // FP Load Multiple 189 InstrItinData<IIC_fpLoadm , [InstrStage<3, [FU_Pipe0]>]>, 190 // 191 // Single-precision FP Store 192 InstrItinData<IIC_fpStore32 , [InstrStage<1, [FU_Pipe0]>], [2, 2, 2]>, 193 // 194 // Double-precision FP Store 195 // use FU_Issue to enforce the 1 load/store per cycle limit 196 InstrItinData<IIC_fpStore64 , [InstrStage<1, [FU_Pipe0]>], [2, 2, 2]>, 197 // 198 // FP Store Multiple 199 InstrItinData<IIC_fpStorem , [InstrStage<3, [FU_Pipe0]>]> 200]>;
|