1 #as: -m32rx --no-warn-explicit-parallel-conflicts --hidden -O 2 #objdump: -dr 3 #name: m32rx 4 5 .*: +file format .* 6 7 Disassembly of section .text: 8 9 0+0000 <bcl>: 10 0: 78 00 f0 00 bcl 0 <bcl> \|\| nop 11 12 0+0004 <bncl>: 13 4: 79 ff f0 00 bncl 0 <bcl> \|\| nop 14 15 0+0008 <cmpz>: 16 8: 00 7d f0 00 cmpz fp \|\| nop 17 18 0+000c <cmpeq>: 19 c: 0d 6d f0 00 cmpeq fp,fp \|\| nop 20 21 0+0010 <maclh1>: 22 10: 5d cd f0 00 maclh1 fp,fp \|\| nop 23 24 0+0014 <msblo>: 25 14: 5d dd f0 00 msblo fp,fp \|\| nop 26 27 0+0018 <mulwu1>: 28 18: 5d ad f0 00 mulwu1 fp,fp \|\| nop 29 30 0+001c <macwu1>: 31 1c: 5d bd f0 00 macwu1 fp,fp \|\| nop 32 33 0+0020 <sadd>: 34 20: 50 e4 f0 00 sadd \|\| nop 35 36 0+0024 <satb>: 37 24: 8d 6d 03 00 satb fp,fp 38 39 0+0028 <mulhi>: 40 28: 3d 8d f0 00 mulhi fp,fp,a1 \|\| nop 41 42 0+002c <mullo>: 43 2c: 3d 1d f0 00 mullo fp,fp \|\| nop 44 45 0+0030 <divh>: 46 30: 9d 0d 00 10 divh fp,fp 47 48 0+0034 <machi>: 49 34: 3d cd f0 00 machi fp,fp,a1 \|\| nop 50 51 0+0038 <maclo>: 52 38: 3d 5d f0 00 maclo fp,fp \|\| nop 53 54 0+003c <mvfachi>: 55 3c: 5d f4 f0 00 mvfachi fp,a1 \|\| nop 56 57 0+0040 <mvfacmi>: 58 40: 5d f6 f0 00 mvfacmi fp,a1 \|\| nop 59 60 0+0044 <mvfaclo>: 61 44: 5d f5 f0 00 mvfaclo fp,a1 \|\| nop 62 63 0+0048 <mvtachi>: 64 48: 5d 74 f0 00 mvtachi fp,a1 \|\| nop 65 66 0+004c <mvtaclo>: 67 4c: 5d 71 f0 00 mvtaclo fp \|\| nop 68 69 0+0050 <rac>: 70 50: 54 90 f0 00 rac a1 \|\| nop 71 72 0+0054 <rac_ds>: 73 54: 54 90 f0 00 rac a1 \|\| nop 74 75 0+0058 <rac_dsi>: 76 58: 50 94 f0 00 rac a0,a1 \|\| nop 77 78 0+005c <rach>: 79 5c: 54 80 f0 00 rach a1 \|\| nop 80 81 0+0060 <rach_ds>: 82 60: 50 84 f0 00 rach a0,a1 \|\| nop 83 84 0+0064 <rach_dsi>: 85 64: 54 81 f0 00 rach a1,a0,#0x2 \|\| nop 86 87 0+0068 <bc__add>: 88 68: 7c 00 8d ad bc 68 <bc__add> \|\| add fp,fp 89 68: R_M32R_10_PCREL_RELA bcl 90 6c: 7c 00 0d ad bc 6c <bc__add\+0x4> -> add fp,fp 91 6c: R_M32R_10_PCREL_RELA bcl 92 93 0+0070 <bcl__addi>: 94 70: 78 00 cd 4d bcl 70 <bcl__addi> \|\| addi fp,#77 95 70: R_M32R_10_PCREL_RELA bcl 96 74: 78 00 cd 4d bcl 74 <bcl__addi\+0x4> \|\| addi fp,#77 97 74: R_M32R_10_PCREL_RELA bcl 98 99 0+0078 <bl__addv>: 100 78: 7e 00 8d 8d bl 78 <bl__addv> \|\| addv fp,fp 101 78: R_M32R_10_PCREL_RELA bcl 102 7c: 7e 00 8d 8d bl 7c <bl__addv\+0x4> \|\| addv fp,fp 103 7c: R_M32R_10_PCREL_RELA bcl 104 105 0+0080 <bnc__addx>: 106 80: 7d 00 8d 9d bnc 80 <bnc__addx> \|\| addx fp,fp 107 80: R_M32R_10_PCREL_RELA bcl 108 84: 7d 00 0d 9d bnc 84 <bnc__addx\+0x4> -> addx fp,fp 109 84: R_M32R_10_PCREL_RELA bcl 110 111 0+0088 <bncl__and>: 112 88: 79 00 8d cd bncl 88 <bncl__and> \|\| and fp,fp 113 88: R_M32R_10_PCREL_RELA bcl 114 8c: 79 00 8d cd bncl 8c <bncl__and\+0x4> \|\| and fp,fp 115 8c: R_M32R_10_PCREL_RELA bcl 116 117 0+0090 <bra__cmp>: 118 90: 7f 00 8d 4d bra 90 <bra__cmp> \|\| cmp fp,fp 119 90: R_M32R_10_PCREL_RELA bcl 120 94: 7f 00 8d 4d bra 94 <bra__cmp\+0x4> \|\| cmp fp,fp 121 94: R_M32R_10_PCREL_RELA bcl 122 123 0+0098 <jl__cmpeq>: 124 98: 1e cd 8d 6d jl fp \|\| cmpeq fp,fp 125 9c: 1e cd 8d 6d jl fp \|\| cmpeq fp,fp 126 127 0+00a0 <jmp__cmpu>: 128 a0: 1f cd 8d 5d jmp fp \|\| cmpu fp,fp 129 a4: 1f cd 8d 5d jmp fp \|\| cmpu fp,fp 130 131 0+00a8 <ld__cmpz>: 132 a8: 2d cd 80 71 ld fp,@fp \|\| cmpz r1 133 ac: 2d cd 80 71 ld fp,@fp \|\| cmpz r1 134 135 0+00b0 <ld__ldi>: 136 b0: 2d e1 e2 4d ld fp,@r1\+ \|\| ldi r2,#77 137 b4: 2d e1 e2 4d ld fp,@r1\+ \|\| ldi r2,#77 138 139 0+00b8 <ldb__mv>: 140 b8: 2d 8d 92 8d ldb fp,@fp \|\| mv r2,fp 141 bc: 2d 8d 12 8d ldb fp,@fp -> mv r2,fp 142 143 0+00c0 <ldh__neg>: 144 c0: 2d ad 82 3d ldh fp,@fp \|\| neg r2,fp 145 c4: 2d ad 02 3d ldh fp,@fp -> neg r2,fp 146 147 0+00c8 <ldub__nop>: 148 c8: 2d 9d f0 00 ldub fp,@fp \|\| nop 149 cc: 2d 9d f0 00 ldub fp,@fp \|\| nop 150 151 0+00d0 <lduh__not>: 152 d0: 2d bd 82 bd lduh fp,@fp \|\| not r2,fp 153 d4: 2d bd 02 bd lduh fp,@fp -> not r2,fp 154 155 0+00d8 <lock__or>: 156 d8: 2d dd 82 ed lock fp,@fp \|\| or r2,fp 157 dc: 2d dd 02 ed lock fp,@fp -> or r2,fp 158 159 0+00e0 <mvfc__sub>: 160 e0: 1d 91 82 2d mvfc fp,cbr \|\| sub r2,fp 161 e4: 1d 91 02 2d mvfc fp,cbr -> sub r2,fp 162 163 0+00e8 <mvtc__subv>: 164 e8: 12 ad 82 0d mvtc fp,spi \|\| subv r2,fp 165 ec: 12 ad 82 0d mvtc fp,spi \|\| subv r2,fp 166 167 0+00f0 <rte__subx>: 168 f0: 10 d6 82 2d rte \|\| sub r2,fp 169 f4: 10 d6 02 1d rte -> subx r2,fp 170 171 0+00f8 <sll__xor>: 172 f8: 1d 41 82 dd sll fp,r1 \|\| xor r2,fp 173 fc: 1d 41 02 dd sll fp,r1 -> xor r2,fp 174 175 0+0100 <slli__machi>: 176 100: 5d 56 b2 4d slli fp,#0x16 \|\| machi r2,fp 177 104: 5d 56 32 4d slli fp,#0x16 -> machi r2,fp 178 179 0+0108 <sra__maclh1>: 180 108: 1d 2d d2 cd sra fp,fp \|\| maclh1 r2,fp 181 10c: 1d 2d 52 cd sra fp,fp -> maclh1 r2,fp 182 183 0+0110 <srai__maclo>: 184 110: 5d 36 b2 5d srai fp,#0x16 \|\| maclo r2,fp 185 114: 5d 36 32 5d srai fp,#0x16 -> maclo r2,fp 186 187 0+0118 <srl__macwhi>: 188 118: 1d 0d b2 6d srl fp,fp \|\| macwhi r2,fp 189 11c: 1d 0d 32 6d srl fp,fp -> macwhi r2,fp 190 191 0+0120 <srli__macwlo>: 192 120: 5d 16 b2 7d srli fp,#0x16 \|\| macwlo r2,fp 193 124: 5d 16 32 7d srli fp,#0x16 -> macwlo r2,fp 194 195 0+0128 <st__macwu1>: 196 128: 2d 4d d2 bd st fp,@fp \|\| macwu1 r2,fp 197 12c: 2d 4d d2 bd st fp,@fp \|\| macwu1 r2,fp 198 199 0+0130 <st__msblo>: 200 130: 2d 6d d2 dd st fp,@\+fp \|\| msblo r2,fp 201 134: 2d 6d 52 dd st fp,@\+fp -> msblo r2,fp 202 203 0+0138 <st__mul>: 204 138: 2d 7d 92 6d st fp,@-fp \|\| mul r2,fp 205 13c: 2d 7d 12 6d st fp,@-fp -> mul r2,fp 206 207 0+0140 <stb__mulhi>: 208 140: 2d 0d b2 0d stb fp,@fp \|\| mulhi r2,fp 209 144: 2d 0d b2 0d stb fp,@fp \|\| mulhi r2,fp 210 211 0+0148 <sth__mullo>: 212 148: 2d 2d b2 1d sth fp,@fp \|\| mullo r2,fp 213 14c: 2d 2d b2 1d sth fp,@fp \|\| mullo r2,fp 214 215 0+0150 <trap__mulwhi>: 216 150: 10 f2 b2 2d trap #0x2 \|\| mulwhi r2,fp 217 154: 10 f2 f0 00 trap #0x2 \|\| nop 218 158: 32 2d f0 00 mulwhi r2,fp \|\| nop 219 220 0+015c <unlock__mulwlo>: 221 15c: 2d 5d b2 3d unlock fp,@fp \|\| mulwlo r2,fp 222 160: 2d 5d b2 3d unlock fp,@fp \|\| mulwlo r2,fp 223 224 0+0164 <add__mulwu1>: 225 164: 0d ad d2 ad add fp,fp \|\| mulwu1 r2,fp 226 168: 0d ad 52 ad add fp,fp -> mulwu1 r2,fp 227 228 0+016c <addi__mvfachi>: 229 16c: 4d 4d d2 f0 addi fp,#77 \|\| mvfachi r2 230 170: 4d 4d d2 f0 addi fp,#77 \|\| mvfachi r2 231 232 0+0174 <addv__mvfaclo>: 233 174: 0d 8d d2 f5 addv fp,fp \|\| mvfaclo r2,a1 234 178: 0d 8d d2 f5 addv fp,fp \|\| mvfaclo r2,a1 235 236 0+017c <addx__mvfacmi>: 237 17c: 0d 9d d2 f2 addx fp,fp \|\| mvfacmi r2 238 180: 0d 9d d2 f2 addx fp,fp \|\| mvfacmi r2 239 240 0+0184 <and__mvtachi>: 241 184: 0d cd d2 70 and fp,fp \|\| mvtachi r2 242 188: 0d cd d2 70 and fp,fp \|\| mvtachi r2 243 244 0+018c <cmp__mvtaclo>: 245 18c: 0d 4d d2 71 cmp fp,fp \|\| mvtaclo r2 246 190: 0d 4d d2 71 cmp fp,fp \|\| mvtaclo r2 247 248 0+0194 <cmpeq__rac>: 249 194: 0d 6d d4 90 cmpeq fp,fp \|\| rac a1 250 198: 0d 6d d4 90 cmpeq fp,fp \|\| rac a1 251 252 0+019c <cmpu__rach>: 253 19c: 0d 5d d0 84 cmpu fp,fp \|\| rach a0,a1 254 1a0: 0d 5d d4 84 cmpu fp,fp \|\| rach a1,a1 255 256 0+01a4 <cmpz__sadd>: 257 1a4: 00 7d d0 e4 cmpz fp \|\| sadd 258 1a8: 00 7d d0 e4 cmpz fp \|\| sadd 259 260 0+01ac <sc>: 261 1ac: 74 01 d0 e4 sc \|\| sadd 262 263 0+01b0 <snc>: 264 1b0: 75 01 d0 e4 snc \|\| sadd 265 266 0+01b4 <jc>: 267 1b4: 1c cd f0 00 jc fp \|\| nop 268 269 0+01b8 <jnc>: 270 1b8: 1d cd f0 00 jnc fp \|\| nop 271 272 0+01bc <pcmpbz>: 273 1bc: 03 7d f0 00 pcmpbz fp \|\| nop 274 275 0+01c0 <sat>: 276 1c0: 8d 6d 00 00 sat fp,fp 277 278 0+01c4 <sath>: 279 1c4: 8d 6d 02 00 sath fp,fp 280 281 0+01c8 <jc__pcmpbz>: 282 1c8: 1c cd 83 7d jc fp \|\| pcmpbz fp 283 1cc: 1c cd 03 7d jc fp -> pcmpbz fp 284 285 0+01d0 <jnc__ldi>: 286 1d0: 1d cd ed 4d jnc fp \|\| ldi fp,#77 287 1d4: 1d cd 6d 4d jnc fp -> ldi fp,#77 288 289 0+01d8 <sc__mv>: 290 1d8: 74 01 9d 82 sc \|\| mv fp,r2 291 1dc: 74 01 9d 82 sc \|\| mv fp,r2 292 293 0+01e0 <snc__neg>: 294 1e0: 75 01 8d 32 snc \|\| neg fp,r2 295 1e4: 75 01 8d 32 snc \|\| neg fp,r2 296 297 0+01e8 <nop__sadd>: 298 1e8: 70 00 d0 e4 nop \|\| sadd 299 300 0+01ec <sadd__nop>: 301 1ec: 70 00 d0 e4 nop \|\| sadd 302 303 0+01f0 <sadd__nop_reverse>: 304 1f0: 70 00 d0 e4 nop \|\| sadd 305 306 0+01f4 <add__not>: 307 1f4: 00 a1 83 b5 add r0,r1 \|\| not r3,r5 308 309 0+01f8 <add__not_dest_clash>: 310 1f8: 03 a4 03 b5 add r3,r4 -> not r3,r5 311 312 0+01fc <add__not__src_clash>: 313 1fc: 03 a4 05 b3 add r3,r4 -> not r5,r3 314 315 0+0200 <add__not__no_clash>: 316 200: 03 a4 84 b5 add r3,r4 \|\| not r4,r5 317 318 0+0204 <mul__sra>: 319 204: 13 24 91 62 sra r3,r4 \|\| mul r1,r2 320 321 0+0208 <mul__sra__reverse_src_clash>: 322 208: 13 24 91 63 sra r3,r4 \|\| mul r1,r3 323 324 0+020c <bc__add_>: 325 20c: 7c 04 01 a2 bc 21c <label> -> add r1,r2 326 327 0+0210 <add__bc>: 328 210: 7c 03 83 a4 bc 21c <label> \|\| add r3,r4 329 330 0+0214 <bc__add__forced_parallel>: 331 214: 7c 02 85 a6 bc 21c <label> \|\| add r5,r6 332 333 0+0218 <add__bc__forced_parallel>: 334 218: 7c 01 87 a8 bc 21c <label> \|\| add r7,r8 335 336 0+021c <label>: 337 21c: 70 00 f0 00 nop \|\| nop 338 339 0+0220 <mulwhi>: 340 220: 3d 2d 3d ad mulwhi fp,fp -> mulwhi fp,fp,a1 341 342 0+0224 <mulwlo>: 343 224: 3d 3d 3d bd mulwlo fp,fp -> mulwlo fp,fp,a1 344 345 0+0228 <macwhi>: 346 228: 3d 6d 3d ed macwhi fp,fp -> macwhi fp,fp,a1 347 348 0+022c <macwlo>: 349 22c: 3d 7d 3d fd macwlo fp,fp -> macwlo fp,fp,a1 350