| 1 | /* |
| 2 | * This file is part of the MicroPython project, http://micropython.org/ |
| 3 | * |
| 4 | * The MIT License (MIT) |
| 5 | * |
| 6 | * Copyright (c) 2014 Fabian Vogt |
| 7 | * Copyright (c) 2013, 2014 Damien P. George |
| 8 | * |
| 9 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 10 | * of this software and associated documentation files (the "Software"), to deal |
| 11 | * in the Software without restriction, including without limitation the rights |
| 12 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 13 | * copies of the Software, and to permit persons to whom the Software is |
| 14 | * furnished to do so, subject to the following conditions: |
| 15 | * |
| 16 | * The above copyright notice and this permission notice shall be included in |
| 17 | * all copies or substantial portions of the Software. |
| 18 | * |
| 19 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 20 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 21 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 22 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 23 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 24 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 25 | * THE SOFTWARE. |
| 26 | */ |
| 27 | |
| 28 | #include <stdio.h> |
| 29 | #include <assert.h> |
| 30 | #include <string.h> |
| 31 | |
| 32 | #include "py/mpconfig.h" |
| 33 | |
| 34 | // wrapper around everything in this file |
| 35 | #if MICROPY_EMIT_ARM |
| 36 | |
| 37 | #include "py/asmarm.h" |
| 38 | |
| 39 | #define SIGNED_FIT24(x) (((x) & 0xff800000) == 0) || (((x) & 0xff000000) == 0xff000000) |
| 40 | |
| 41 | void asm_arm_end_pass(asm_arm_t *as) { |
| 42 | if (as->base.pass == MP_ASM_PASS_EMIT) { |
| 43 | #if defined(__linux__) && defined(__GNUC__) |
| 44 | char *start = mp_asm_base_get_code(&as->base); |
| 45 | char *end = start + mp_asm_base_get_code_size(&as->base); |
| 46 | __builtin___clear_cache(start, end); |
| 47 | #elif defined(__arm__) |
| 48 | // flush I- and D-cache |
| 49 | asm volatile ( |
| 50 | "0:" |
| 51 | "mrc p15, 0, r15, c7, c10, 3\n" |
| 52 | "bne 0b\n" |
| 53 | "mov r0, #0\n" |
| 54 | "mcr p15, 0, r0, c7, c7, 0\n" |
| 55 | : : : "r0" , "cc" ); |
| 56 | #endif |
| 57 | } |
| 58 | } |
| 59 | |
| 60 | // Insert word into instruction flow |
| 61 | STATIC void emit(asm_arm_t *as, uint op) { |
| 62 | uint8_t *c = mp_asm_base_get_cur_to_write_bytes(&as->base, 4); |
| 63 | if (c != NULL) { |
| 64 | *(uint32_t *)c = op; |
| 65 | } |
| 66 | } |
| 67 | |
| 68 | // Insert word into instruction flow, add "ALWAYS" condition code |
| 69 | STATIC void emit_al(asm_arm_t *as, uint op) { |
| 70 | emit(as, op | ASM_ARM_CC_AL); |
| 71 | } |
| 72 | |
| 73 | // Basic instructions without condition code |
| 74 | STATIC uint asm_arm_op_push(uint reglist) { |
| 75 | // stmfd sp!, {reglist} |
| 76 | return 0x92d0000 | (reglist & 0xFFFF); |
| 77 | } |
| 78 | |
| 79 | STATIC uint asm_arm_op_pop(uint reglist) { |
| 80 | // ldmfd sp!, {reglist} |
| 81 | return 0x8bd0000 | (reglist & 0xFFFF); |
| 82 | } |
| 83 | |
| 84 | STATIC uint asm_arm_op_mov_reg(uint rd, uint rn) { |
| 85 | // mov rd, rn |
| 86 | return 0x1a00000 | (rd << 12) | rn; |
| 87 | } |
| 88 | |
| 89 | STATIC uint asm_arm_op_mov_imm(uint rd, uint imm) { |
| 90 | // mov rd, #imm |
| 91 | return 0x3a00000 | (rd << 12) | imm; |
| 92 | } |
| 93 | |
| 94 | STATIC uint asm_arm_op_mvn_imm(uint rd, uint imm) { |
| 95 | // mvn rd, #imm |
| 96 | return 0x3e00000 | (rd << 12) | imm; |
| 97 | } |
| 98 | |
| 99 | STATIC uint asm_arm_op_add_imm(uint rd, uint rn, uint imm) { |
| 100 | // add rd, rn, #imm |
| 101 | return 0x2800000 | (rn << 16) | (rd << 12) | (imm & 0xFF); |
| 102 | } |
| 103 | |
| 104 | STATIC uint asm_arm_op_add_reg(uint rd, uint rn, uint rm) { |
| 105 | // add rd, rn, rm |
| 106 | return 0x0800000 | (rn << 16) | (rd << 12) | rm; |
| 107 | } |
| 108 | |
| 109 | STATIC uint asm_arm_op_sub_imm(uint rd, uint rn, uint imm) { |
| 110 | // sub rd, rn, #imm |
| 111 | return 0x2400000 | (rn << 16) | (rd << 12) | (imm & 0xFF); |
| 112 | } |
| 113 | |
| 114 | STATIC uint asm_arm_op_sub_reg(uint rd, uint rn, uint rm) { |
| 115 | // sub rd, rn, rm |
| 116 | return 0x0400000 | (rn << 16) | (rd << 12) | rm; |
| 117 | } |
| 118 | |
| 119 | STATIC uint asm_arm_op_mul_reg(uint rd, uint rm, uint rs) { |
| 120 | // mul rd, rm, rs |
| 121 | assert(rd != rm); |
| 122 | return 0x0000090 | (rd << 16) | (rs << 8) | rm; |
| 123 | } |
| 124 | |
| 125 | STATIC uint asm_arm_op_and_reg(uint rd, uint rn, uint rm) { |
| 126 | // and rd, rn, rm |
| 127 | return 0x0000000 | (rn << 16) | (rd << 12) | rm; |
| 128 | } |
| 129 | |
| 130 | STATIC uint asm_arm_op_eor_reg(uint rd, uint rn, uint rm) { |
| 131 | // eor rd, rn, rm |
| 132 | return 0x0200000 | (rn << 16) | (rd << 12) | rm; |
| 133 | } |
| 134 | |
| 135 | STATIC uint asm_arm_op_orr_reg(uint rd, uint rn, uint rm) { |
| 136 | // orr rd, rn, rm |
| 137 | return 0x1800000 | (rn << 16) | (rd << 12) | rm; |
| 138 | } |
| 139 | |
| 140 | void asm_arm_bkpt(asm_arm_t *as) { |
| 141 | // bkpt #0 |
| 142 | emit_al(as, 0x1200070); |
| 143 | } |
| 144 | |
| 145 | // locals: |
| 146 | // - stored on the stack in ascending order |
| 147 | // - numbered 0 through num_locals-1 |
| 148 | // - SP points to first local |
| 149 | // |
| 150 | // | SP |
| 151 | // v |
| 152 | // l0 l1 l2 ... l(n-1) |
| 153 | // ^ ^ |
| 154 | // | low address | high address in RAM |
| 155 | |
| 156 | void asm_arm_entry(asm_arm_t *as, int num_locals) { |
| 157 | assert(num_locals >= 0); |
| 158 | |
| 159 | as->stack_adjust = 0; |
| 160 | as->push_reglist = 1 << ASM_ARM_REG_R1 |
| 161 | | 1 << ASM_ARM_REG_R2 |
| 162 | | 1 << ASM_ARM_REG_R3 |
| 163 | | 1 << ASM_ARM_REG_R4 |
| 164 | | 1 << ASM_ARM_REG_R5 |
| 165 | | 1 << ASM_ARM_REG_R6 |
| 166 | | 1 << ASM_ARM_REG_R7 |
| 167 | | 1 << ASM_ARM_REG_R8; |
| 168 | |
| 169 | // Only adjust the stack if there are more locals than usable registers |
| 170 | if (num_locals > 3) { |
| 171 | as->stack_adjust = num_locals * 4; |
| 172 | // Align stack to 8 bytes |
| 173 | if (num_locals & 1) { |
| 174 | as->stack_adjust += 4; |
| 175 | } |
| 176 | } |
| 177 | |
| 178 | emit_al(as, asm_arm_op_push(as->push_reglist | 1 << ASM_ARM_REG_LR)); |
| 179 | if (as->stack_adjust > 0) { |
| 180 | emit_al(as, asm_arm_op_sub_imm(ASM_ARM_REG_SP, ASM_ARM_REG_SP, as->stack_adjust)); |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | void asm_arm_exit(asm_arm_t *as) { |
| 185 | if (as->stack_adjust > 0) { |
| 186 | emit_al(as, asm_arm_op_add_imm(ASM_ARM_REG_SP, ASM_ARM_REG_SP, as->stack_adjust)); |
| 187 | } |
| 188 | |
| 189 | emit_al(as, asm_arm_op_pop(as->push_reglist | (1 << ASM_ARM_REG_PC))); |
| 190 | } |
| 191 | |
| 192 | void asm_arm_push(asm_arm_t *as, uint reglist) { |
| 193 | emit_al(as, asm_arm_op_push(reglist)); |
| 194 | } |
| 195 | |
| 196 | void asm_arm_pop(asm_arm_t *as, uint reglist) { |
| 197 | emit_al(as, asm_arm_op_pop(reglist)); |
| 198 | } |
| 199 | |
| 200 | void asm_arm_mov_reg_reg(asm_arm_t *as, uint reg_dest, uint reg_src) { |
| 201 | emit_al(as, asm_arm_op_mov_reg(reg_dest, reg_src)); |
| 202 | } |
| 203 | |
| 204 | size_t asm_arm_mov_reg_i32(asm_arm_t *as, uint rd, int imm) { |
| 205 | // Insert immediate into code and jump over it |
| 206 | emit_al(as, 0x59f0000 | (rd << 12)); // ldr rd, [pc] |
| 207 | emit_al(as, 0xa000000); // b pc |
| 208 | size_t loc = mp_asm_base_get_code_pos(&as->base); |
| 209 | emit(as, imm); |
| 210 | return loc; |
| 211 | } |
| 212 | |
| 213 | void asm_arm_mov_reg_i32_optimised(asm_arm_t *as, uint rd, int imm) { |
| 214 | // TODO: There are more variants of immediate values |
| 215 | if ((imm & 0xFF) == imm) { |
| 216 | emit_al(as, asm_arm_op_mov_imm(rd, imm)); |
| 217 | } else if (imm < 0 && imm >= -256) { |
| 218 | // mvn is "move not", not "move negative" |
| 219 | emit_al(as, asm_arm_op_mvn_imm(rd, ~imm)); |
| 220 | } else { |
| 221 | asm_arm_mov_reg_i32(as, rd, imm); |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | void asm_arm_mov_local_reg(asm_arm_t *as, int local_num, uint rd) { |
| 226 | // str rd, [sp, #local_num*4] |
| 227 | emit_al(as, 0x58d0000 | (rd << 12) | (local_num << 2)); |
| 228 | } |
| 229 | |
| 230 | void asm_arm_mov_reg_local(asm_arm_t *as, uint rd, int local_num) { |
| 231 | // ldr rd, [sp, #local_num*4] |
| 232 | emit_al(as, 0x59d0000 | (rd << 12) | (local_num << 2)); |
| 233 | } |
| 234 | |
| 235 | void asm_arm_cmp_reg_i8(asm_arm_t *as, uint rd, int imm) { |
| 236 | // cmp rd, #imm |
| 237 | emit_al(as, 0x3500000 | (rd << 16) | (imm & 0xFF)); |
| 238 | } |
| 239 | |
| 240 | void asm_arm_cmp_reg_reg(asm_arm_t *as, uint rd, uint rn) { |
| 241 | // cmp rd, rn |
| 242 | emit_al(as, 0x1500000 | (rd << 16) | rn); |
| 243 | } |
| 244 | |
| 245 | void asm_arm_setcc_reg(asm_arm_t *as, uint rd, uint cond) { |
| 246 | emit(as, asm_arm_op_mov_imm(rd, 1) | cond); // movCOND rd, #1 |
| 247 | emit(as, asm_arm_op_mov_imm(rd, 0) | (cond ^ (1 << 28))); // mov!COND rd, #0 |
| 248 | } |
| 249 | |
| 250 | void asm_arm_add_reg_reg_reg(asm_arm_t *as, uint rd, uint rn, uint rm) { |
| 251 | // add rd, rn, rm |
| 252 | emit_al(as, asm_arm_op_add_reg(rd, rn, rm)); |
| 253 | } |
| 254 | |
| 255 | void asm_arm_sub_reg_reg_reg(asm_arm_t *as, uint rd, uint rn, uint rm) { |
| 256 | // sub rd, rn, rm |
| 257 | emit_al(as, asm_arm_op_sub_reg(rd, rn, rm)); |
| 258 | } |
| 259 | |
| 260 | void asm_arm_mul_reg_reg_reg(asm_arm_t *as, uint rd, uint rs, uint rm) { |
| 261 | // rs and rm are swapped because of restriction rd!=rm |
| 262 | // mul rd, rm, rs |
| 263 | emit_al(as, asm_arm_op_mul_reg(rd, rm, rs)); |
| 264 | } |
| 265 | |
| 266 | void asm_arm_and_reg_reg_reg(asm_arm_t *as, uint rd, uint rn, uint rm) { |
| 267 | // and rd, rn, rm |
| 268 | emit_al(as, asm_arm_op_and_reg(rd, rn, rm)); |
| 269 | } |
| 270 | |
| 271 | void asm_arm_eor_reg_reg_reg(asm_arm_t *as, uint rd, uint rn, uint rm) { |
| 272 | // eor rd, rn, rm |
| 273 | emit_al(as, asm_arm_op_eor_reg(rd, rn, rm)); |
| 274 | } |
| 275 | |
| 276 | void asm_arm_orr_reg_reg_reg(asm_arm_t *as, uint rd, uint rn, uint rm) { |
| 277 | // orr rd, rn, rm |
| 278 | emit_al(as, asm_arm_op_orr_reg(rd, rn, rm)); |
| 279 | } |
| 280 | |
| 281 | void asm_arm_mov_reg_local_addr(asm_arm_t *as, uint rd, int local_num) { |
| 282 | // add rd, sp, #local_num*4 |
| 283 | emit_al(as, asm_arm_op_add_imm(rd, ASM_ARM_REG_SP, local_num << 2)); |
| 284 | } |
| 285 | |
| 286 | void asm_arm_mov_reg_pcrel(asm_arm_t *as, uint reg_dest, uint label) { |
| 287 | assert(label < as->base.max_num_labels); |
| 288 | mp_uint_t dest = as->base.label_offsets[label]; |
| 289 | mp_int_t rel = dest - as->base.code_offset; |
| 290 | rel -= 12 + 8; // adjust for load of rel, and then PC+8 prefetch of add_reg_reg_reg |
| 291 | |
| 292 | // To load rel int reg_dest, insert immediate into code and jump over it |
| 293 | emit_al(as, 0x59f0000 | (reg_dest << 12)); // ldr rd, [pc] |
| 294 | emit_al(as, 0xa000000); // b pc |
| 295 | emit(as, rel); |
| 296 | |
| 297 | // Do reg_dest += PC |
| 298 | asm_arm_add_reg_reg_reg(as, reg_dest, reg_dest, ASM_ARM_REG_PC); |
| 299 | } |
| 300 | |
| 301 | void asm_arm_lsl_reg_reg(asm_arm_t *as, uint rd, uint rs) { |
| 302 | // mov rd, rd, lsl rs |
| 303 | emit_al(as, 0x1a00010 | (rd << 12) | (rs << 8) | rd); |
| 304 | } |
| 305 | |
| 306 | void asm_arm_lsr_reg_reg(asm_arm_t *as, uint rd, uint rs) { |
| 307 | // mov rd, rd, lsr rs |
| 308 | emit_al(as, 0x1a00030 | (rd << 12) | (rs << 8) | rd); |
| 309 | } |
| 310 | |
| 311 | void asm_arm_asr_reg_reg(asm_arm_t *as, uint rd, uint rs) { |
| 312 | // mov rd, rd, asr rs |
| 313 | emit_al(as, 0x1a00050 | (rd << 12) | (rs << 8) | rd); |
| 314 | } |
| 315 | |
| 316 | void asm_arm_ldr_reg_reg(asm_arm_t *as, uint rd, uint rn, uint byte_offset) { |
| 317 | // ldr rd, [rn, #off] |
| 318 | emit_al(as, 0x5900000 | (rn << 16) | (rd << 12) | byte_offset); |
| 319 | } |
| 320 | |
| 321 | void asm_arm_ldrh_reg_reg(asm_arm_t *as, uint rd, uint rn) { |
| 322 | // ldrh rd, [rn] |
| 323 | emit_al(as, 0x1d000b0 | (rn << 16) | (rd << 12)); |
| 324 | } |
| 325 | |
| 326 | void asm_arm_ldrb_reg_reg(asm_arm_t *as, uint rd, uint rn) { |
| 327 | // ldrb rd, [rn] |
| 328 | emit_al(as, 0x5d00000 | (rn << 16) | (rd << 12)); |
| 329 | } |
| 330 | |
| 331 | void asm_arm_str_reg_reg(asm_arm_t *as, uint rd, uint rm, uint byte_offset) { |
| 332 | // str rd, [rm, #off] |
| 333 | emit_al(as, 0x5800000 | (rm << 16) | (rd << 12) | byte_offset); |
| 334 | } |
| 335 | |
| 336 | void asm_arm_strh_reg_reg(asm_arm_t *as, uint rd, uint rm) { |
| 337 | // strh rd, [rm] |
| 338 | emit_al(as, 0x1c000b0 | (rm << 16) | (rd << 12)); |
| 339 | } |
| 340 | |
| 341 | void asm_arm_strb_reg_reg(asm_arm_t *as, uint rd, uint rm) { |
| 342 | // strb rd, [rm] |
| 343 | emit_al(as, 0x5c00000 | (rm << 16) | (rd << 12)); |
| 344 | } |
| 345 | |
| 346 | void asm_arm_str_reg_reg_reg(asm_arm_t *as, uint rd, uint rm, uint rn) { |
| 347 | // str rd, [rm, rn, lsl #2] |
| 348 | emit_al(as, 0x7800100 | (rm << 16) | (rd << 12) | rn); |
| 349 | } |
| 350 | |
| 351 | void asm_arm_strh_reg_reg_reg(asm_arm_t *as, uint rd, uint rm, uint rn) { |
| 352 | // strh doesn't support scaled register index |
| 353 | emit_al(as, 0x1a00080 | (ASM_ARM_REG_R8 << 12) | rn); // mov r8, rn, lsl #1 |
| 354 | emit_al(as, 0x18000b0 | (rm << 16) | (rd << 12) | ASM_ARM_REG_R8); // strh rd, [rm, r8] |
| 355 | } |
| 356 | |
| 357 | void asm_arm_strb_reg_reg_reg(asm_arm_t *as, uint rd, uint rm, uint rn) { |
| 358 | // strb rd, [rm, rn] |
| 359 | emit_al(as, 0x7c00000 | (rm << 16) | (rd << 12) | rn); |
| 360 | } |
| 361 | |
| 362 | void asm_arm_bcc_label(asm_arm_t *as, int cond, uint label) { |
| 363 | assert(label < as->base.max_num_labels); |
| 364 | mp_uint_t dest = as->base.label_offsets[label]; |
| 365 | mp_int_t rel = dest - as->base.code_offset; |
| 366 | rel -= 8; // account for instruction prefetch, PC is 8 bytes ahead of this instruction |
| 367 | rel >>= 2; // in ARM mode the branch target is 32-bit aligned, so the 2 LSB are omitted |
| 368 | |
| 369 | if (SIGNED_FIT24(rel)) { |
| 370 | emit(as, cond | 0xa000000 | (rel & 0xffffff)); |
| 371 | } else { |
| 372 | printf("asm_arm_bcc: branch does not fit in 24 bits\n" ); |
| 373 | } |
| 374 | } |
| 375 | |
| 376 | void asm_arm_b_label(asm_arm_t *as, uint label) { |
| 377 | asm_arm_bcc_label(as, ASM_ARM_CC_AL, label); |
| 378 | } |
| 379 | |
| 380 | void asm_arm_bl_ind(asm_arm_t *as, uint fun_id, uint reg_temp) { |
| 381 | // The table offset should fit into the ldr instruction |
| 382 | assert(fun_id < (0x1000 / 4)); |
| 383 | emit_al(as, asm_arm_op_mov_reg(ASM_ARM_REG_LR, ASM_ARM_REG_PC)); // mov lr, pc |
| 384 | emit_al(as, 0x597f000 | (fun_id << 2)); // ldr pc, [r7, #fun_id*4] |
| 385 | } |
| 386 | |
| 387 | void asm_arm_bx_reg(asm_arm_t *as, uint reg_src) { |
| 388 | emit_al(as, 0x012fff10 | reg_src); |
| 389 | } |
| 390 | |
| 391 | #endif // MICROPY_EMIT_ARM |
| 392 | |