Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/arch/arm/lib/lib1funcs.S: Optimized ARM division routines |
| 3 | * |
| 4 | * Author: Nicolas Pitre <nico@fluxnic.net> |
| 5 | * - contributed to gcc-3.4 on Sep 30, 2003 |
| 6 | * - adapted for the Linux kernel on Oct 2, 2003 |
| 7 | */ |
| 8 | |
| 9 | /* Copyright 1995, 1996, 1998, 1999, 2000, 2003 Free Software Foundation, Inc. |
| 10 | |
| 11 | This file is free software; you can redistribute it and/or modify it |
| 12 | under the terms of the GNU General Public License as published by the |
| 13 | Free Software Foundation; either version 2, or (at your option) any |
| 14 | later version. |
| 15 | |
| 16 | In addition to the permissions in the GNU General Public License, the |
| 17 | Free Software Foundation gives you unlimited permission to link the |
| 18 | compiled version of this file into combinations with other programs, |
| 19 | and to distribute those combinations without any restriction coming |
| 20 | from the use of this file. (The General Public License restrictions |
| 21 | do apply in other respects; for example, they cover modification of |
| 22 | the file, and distribution when not linked into a combine |
| 23 | executable.) |
| 24 | |
| 25 | This file is distributed in the hope that it will be useful, but |
| 26 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 27 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 28 | General Public License for more details. |
Patrick Georgi | a73b931 | 2015-10-31 11:55:10 +0100 | [diff] [blame] | 29 | */ |
Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 30 | |
| 31 | |
| 32 | #if defined __GNUC__ |
| 33 | |
| 34 | #include <arch/asm.h> |
| 35 | |
| 36 | .macro ARM_DIV_BODY dividend, divisor, result, curbit |
| 37 | |
| 38 | #if __COREBOOT_ARM_ARCH__ >= 5 |
| 39 | |
| 40 | clz \curbit, \divisor |
| 41 | clz \result, \dividend |
| 42 | sub \result, \curbit, \result |
| 43 | mov \curbit, #1 |
| 44 | mov \divisor, \divisor, lsl \result |
| 45 | mov \curbit, \curbit, lsl \result |
| 46 | mov \result, #0 |
| 47 | |
| 48 | #else |
| 49 | |
| 50 | @ Initially shift the divisor left 3 bits if possible, |
| 51 | @ set curbit accordingly. This allows for curbit to be located |
| 52 | @ at the left end of each 4 bit nibbles in the division loop |
| 53 | @ to save one loop in most cases. |
| 54 | tst \divisor, #0xe0000000 |
| 55 | moveq \divisor, \divisor, lsl #3 |
| 56 | moveq \curbit, #8 |
| 57 | movne \curbit, #1 |
| 58 | |
| 59 | @ Unless the divisor is very big, shift it up in multiples of |
| 60 | @ four bits, since this is the amount of unwinding in the main |
| 61 | @ division loop. Continue shifting until the divisor is |
| 62 | @ larger than the dividend. |
| 63 | 1: cmp \divisor, #0x10000000 |
| 64 | cmplo \divisor, \dividend |
| 65 | movlo \divisor, \divisor, lsl #4 |
| 66 | movlo \curbit, \curbit, lsl #4 |
| 67 | blo 1b |
| 68 | |
| 69 | @ For very big divisors, we must shift it a bit at a time, or |
| 70 | @ we will be in danger of overflowing. |
| 71 | 1: cmp \divisor, #0x80000000 |
| 72 | cmplo \divisor, \dividend |
| 73 | movlo \divisor, \divisor, lsl #1 |
| 74 | movlo \curbit, \curbit, lsl #1 |
| 75 | blo 1b |
| 76 | |
| 77 | mov \result, #0 |
| 78 | |
| 79 | #endif |
| 80 | |
| 81 | @ Division loop |
| 82 | 1: cmp \dividend, \divisor |
| 83 | subhs \dividend, \dividend, \divisor |
| 84 | orrhs \result, \result, \curbit |
| 85 | cmp \dividend, \divisor, lsr #1 |
| 86 | subhs \dividend, \dividend, \divisor, lsr #1 |
| 87 | orrhs \result, \result, \curbit, lsr #1 |
| 88 | cmp \dividend, \divisor, lsr #2 |
| 89 | subhs \dividend, \dividend, \divisor, lsr #2 |
| 90 | orrhs \result, \result, \curbit, lsr #2 |
| 91 | cmp \dividend, \divisor, lsr #3 |
| 92 | subhs \dividend, \dividend, \divisor, lsr #3 |
| 93 | orrhs \result, \result, \curbit, lsr #3 |
| 94 | cmp \dividend, #0 @ Early termination? |
| 95 | movnes \curbit, \curbit, lsr #4 @ No, any more bits to do? |
| 96 | movne \divisor, \divisor, lsr #4 |
| 97 | bne 1b |
| 98 | |
| 99 | .endm |
| 100 | |
| 101 | |
| 102 | .macro ARM_DIV2_ORDER divisor, order |
| 103 | |
| 104 | #if __COREBOOT_ARM_ARCH__ >= 5 |
| 105 | |
| 106 | clz \order, \divisor |
| 107 | rsb \order, \order, #31 |
| 108 | |
| 109 | #else |
| 110 | |
| 111 | cmp \divisor, #(1 << 16) |
| 112 | movhs \divisor, \divisor, lsr #16 |
| 113 | movhs \order, #16 |
| 114 | movlo \order, #0 |
| 115 | |
| 116 | cmp \divisor, #(1 << 8) |
| 117 | movhs \divisor, \divisor, lsr #8 |
| 118 | addhs \order, \order, #8 |
| 119 | |
| 120 | cmp \divisor, #(1 << 4) |
| 121 | movhs \divisor, \divisor, lsr #4 |
| 122 | addhs \order, \order, #4 |
| 123 | |
| 124 | cmp \divisor, #(1 << 2) |
| 125 | addhi \order, \order, #3 |
| 126 | addls \order, \order, \divisor, lsr #1 |
| 127 | |
| 128 | #endif |
| 129 | |
| 130 | .endm |
| 131 | |
| 132 | |
| 133 | .macro ARM_MOD_BODY dividend, divisor, order, spare |
| 134 | |
| 135 | #if __COREBOOT_ARM_ARCH__ >= 5 |
| 136 | |
| 137 | clz \order, \divisor |
| 138 | clz \spare, \dividend |
| 139 | sub \order, \order, \spare |
| 140 | mov \divisor, \divisor, lsl \order |
| 141 | |
| 142 | #else |
| 143 | |
| 144 | mov \order, #0 |
| 145 | |
| 146 | @ Unless the divisor is very big, shift it up in multiples of |
| 147 | @ four bits, since this is the amount of unwinding in the main |
| 148 | @ division loop. Continue shifting until the divisor is |
| 149 | @ larger than the dividend. |
| 150 | 1: cmp \divisor, #0x10000000 |
| 151 | cmplo \divisor, \dividend |
| 152 | movlo \divisor, \divisor, lsl #4 |
| 153 | addlo \order, \order, #4 |
| 154 | blo 1b |
| 155 | |
| 156 | @ For very big divisors, we must shift it a bit at a time, or |
| 157 | @ we will be in danger of overflowing. |
| 158 | 1: cmp \divisor, #0x80000000 |
| 159 | cmplo \divisor, \dividend |
| 160 | movlo \divisor, \divisor, lsl #1 |
| 161 | addlo \order, \order, #1 |
| 162 | blo 1b |
| 163 | |
| 164 | #endif |
| 165 | |
| 166 | @ Perform all needed substractions to keep only the reminder. |
| 167 | @ Do comparisons in batch of 4 first. |
| 168 | subs \order, \order, #3 @ yes, 3 is intended here |
| 169 | blt 2f |
| 170 | |
| 171 | 1: cmp \dividend, \divisor |
| 172 | subhs \dividend, \dividend, \divisor |
| 173 | cmp \dividend, \divisor, lsr #1 |
| 174 | subhs \dividend, \dividend, \divisor, lsr #1 |
| 175 | cmp \dividend, \divisor, lsr #2 |
| 176 | subhs \dividend, \dividend, \divisor, lsr #2 |
| 177 | cmp \dividend, \divisor, lsr #3 |
| 178 | subhs \dividend, \dividend, \divisor, lsr #3 |
| 179 | cmp \dividend, #1 |
| 180 | mov \divisor, \divisor, lsr #4 |
| 181 | subges \order, \order, #4 |
| 182 | bge 1b |
| 183 | |
| 184 | tst \order, #3 |
| 185 | teqne \dividend, #0 |
| 186 | beq 5f |
| 187 | |
| 188 | @ Either 1, 2 or 3 comparison/substractions are left. |
| 189 | 2: cmn \order, #2 |
| 190 | blt 4f |
| 191 | beq 3f |
| 192 | cmp \dividend, \divisor |
| 193 | subhs \dividend, \dividend, \divisor |
| 194 | mov \divisor, \divisor, lsr #1 |
| 195 | 3: cmp \dividend, \divisor |
| 196 | subhs \dividend, \dividend, \divisor |
| 197 | mov \divisor, \divisor, lsr #1 |
| 198 | 4: cmp \dividend, \divisor |
| 199 | subhs \dividend, \dividend, \divisor |
| 200 | 5: |
| 201 | .endm |
| 202 | |
| 203 | |
| 204 | ENTRY(__udivsi3) |
Julius Werner | 7c6e489 | 2014-01-24 16:23:08 -0800 | [diff] [blame] | 205 | .global __aeabi_uidiv |
| 206 | __aeabi_uidiv: |
Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 207 | |
| 208 | subs r2, r1, #1 |
| 209 | moveq pc, lr |
| 210 | bcc Ldiv0 |
| 211 | cmp r0, r1 |
| 212 | bls 11f |
| 213 | tst r1, r2 |
| 214 | beq 12f |
| 215 | |
| 216 | ARM_DIV_BODY r0, r1, r2, r3 |
| 217 | |
| 218 | mov r0, r2 |
| 219 | mov pc, lr |
| 220 | |
| 221 | 11: moveq r0, #1 |
| 222 | movne r0, #0 |
| 223 | mov pc, lr |
| 224 | |
| 225 | 12: ARM_DIV2_ORDER r1, r2 |
| 226 | |
| 227 | mov r0, r0, lsr r2 |
| 228 | mov pc, lr |
| 229 | |
Julius Werner | 7c6e489 | 2014-01-24 16:23:08 -0800 | [diff] [blame] | 230 | .type __aeabi_uidiv, %function |
| 231 | .size __aeabi_uidiv, .-__aeabi_uidiv |
Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 232 | ENDPROC(__udivsi3) |
Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 233 | |
| 234 | ENTRY(__umodsi3) |
| 235 | |
| 236 | subs r2, r1, #1 @ compare divisor with 1 |
| 237 | bcc Ldiv0 |
| 238 | cmpne r0, r1 @ compare dividend with divisor |
| 239 | moveq r0, #0 |
| 240 | tsthi r1, r2 @ see if divisor is power of 2 |
| 241 | andeq r0, r0, r2 |
| 242 | movls pc, lr |
| 243 | |
| 244 | ARM_MOD_BODY r0, r1, r2, r3 |
| 245 | |
| 246 | mov pc, lr |
| 247 | |
| 248 | ENDPROC(__umodsi3) |
| 249 | |
| 250 | ENTRY(__divsi3) |
Julius Werner | 7c6e489 | 2014-01-24 16:23:08 -0800 | [diff] [blame] | 251 | .global __aeabi_idiv |
| 252 | __aeabi_idiv: |
Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 253 | |
| 254 | cmp r1, #0 |
| 255 | eor ip, r0, r1 @ save the sign of the result. |
| 256 | beq Ldiv0 |
| 257 | rsbmi r1, r1, #0 @ loops below use unsigned. |
| 258 | subs r2, r1, #1 @ division by 1 or -1 ? |
| 259 | beq 10f |
| 260 | movs r3, r0 |
| 261 | rsbmi r3, r0, #0 @ positive dividend value |
| 262 | cmp r3, r1 |
| 263 | bls 11f |
| 264 | tst r1, r2 @ divisor is power of 2 ? |
| 265 | beq 12f |
| 266 | |
| 267 | ARM_DIV_BODY r3, r1, r0, r2 |
| 268 | |
| 269 | cmp ip, #0 |
| 270 | rsbmi r0, r0, #0 |
| 271 | mov pc, lr |
| 272 | |
| 273 | 10: teq ip, r0 @ same sign ? |
| 274 | rsbmi r0, r0, #0 |
| 275 | mov pc, lr |
| 276 | |
| 277 | 11: movlo r0, #0 |
| 278 | moveq r0, ip, asr #31 |
| 279 | orreq r0, r0, #1 |
| 280 | mov pc, lr |
| 281 | |
| 282 | 12: ARM_DIV2_ORDER r1, r2 |
| 283 | |
| 284 | cmp ip, #0 |
| 285 | mov r0, r3, lsr r2 |
| 286 | rsbmi r0, r0, #0 |
| 287 | mov pc, lr |
| 288 | |
Julius Werner | 7c6e489 | 2014-01-24 16:23:08 -0800 | [diff] [blame] | 289 | .type __aeabi_idiv, %function |
| 290 | .size __aeabi_idiv, .-__aeabi_idiv |
Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 291 | ENDPROC(__divsi3) |
Gabe Black | 169c0df | 2013-10-08 18:24:10 -0700 | [diff] [blame] | 292 | |
| 293 | ENTRY(__modsi3) |
| 294 | |
| 295 | cmp r1, #0 |
| 296 | beq Ldiv0 |
| 297 | rsbmi r1, r1, #0 @ loops below use unsigned. |
| 298 | movs ip, r0 @ preserve sign of dividend |
| 299 | rsbmi r0, r0, #0 @ if negative make positive |
| 300 | subs r2, r1, #1 @ compare divisor with 1 |
| 301 | cmpne r0, r1 @ compare dividend with divisor |
| 302 | moveq r0, #0 |
| 303 | tsthi r1, r2 @ see if divisor is power of 2 |
| 304 | andeq r0, r0, r2 |
| 305 | bls 10f |
| 306 | |
| 307 | ARM_MOD_BODY r0, r1, r2, r3 |
| 308 | |
| 309 | 10: cmp ip, #0 |
| 310 | rsbmi r0, r0, #0 |
| 311 | mov pc, lr |
| 312 | |
| 313 | ENDPROC(__modsi3) |
| 314 | |
| 315 | ENTRY(__aeabi_uidivmod) |
| 316 | |
| 317 | stmfd sp!, {r0, r1, ip, lr} |
| 318 | bl __aeabi_uidiv |
| 319 | ldmfd sp!, {r1, r2, ip, lr} |
| 320 | mul r3, r0, r2 |
| 321 | sub r1, r1, r3 |
| 322 | mov pc, lr |
| 323 | |
| 324 | ENDPROC(__aeabi_uidivmod) |
| 325 | |
| 326 | ENTRY(__aeabi_idivmod) |
| 327 | stmfd sp!, {r0, r1, ip, lr} |
| 328 | bl __aeabi_idiv |
| 329 | ldmfd sp!, {r1, r2, ip, lr} |
| 330 | mul r3, r0, r2 |
| 331 | sub r1, r1, r3 |
| 332 | mov pc, lr |
| 333 | |
| 334 | ENDPROC(__aeabi_idivmod) |
| 335 | |
| 336 | |
| 337 | Ldiv0: |
| 338 | str lr, [sp, #-8]! |
| 339 | bl __div0 |
| 340 | mov r0, #0 @ About as wrong as it could be. |
| 341 | ldr pc, [sp], #8 |
| 342 | ENDPROC(Ldiv0) |
| 343 | |
| 344 | #endif |