1 /* 2 * Copyright (C) 2005 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 #ifndef ANDROID_GGL_FIXED_H 18 #define ANDROID_GGL_FIXED_H 19 20 #include <math.h> 21 #include <pixelflinger/pixelflinger.h> 22 23 // ---------------------------------------------------------------------------- 24 25 #define CONST __attribute__((const)) 26 #define ALWAYS_INLINE __attribute__((always_inline)) 27 28 const GGLfixed FIXED_BITS = 16; 29 const GGLfixed FIXED_EPSILON = 1; 30 const GGLfixed FIXED_ONE = 1L<<FIXED_BITS; 31 const GGLfixed FIXED_HALF = 1L<<(FIXED_BITS-1); 32 const GGLfixed FIXED_MIN = 0x80000000L; 33 const GGLfixed FIXED_MAX = 0x7FFFFFFFL; 34 35 inline GGLfixed gglIntToFixed(GGLfixed i) ALWAYS_INLINE ; 36 inline GGLfixed gglFixedToIntRound(GGLfixed f) ALWAYS_INLINE ; 37 inline GGLfixed gglFixedToIntFloor(GGLfixed f) ALWAYS_INLINE ; 38 inline GGLfixed gglFixedToIntCeil(GGLfixed f) ALWAYS_INLINE ; 39 inline GGLfixed gglFracx(GGLfixed v) ALWAYS_INLINE ; 40 inline GGLfixed gglFloorx(GGLfixed v) ALWAYS_INLINE ; 41 inline GGLfixed gglCeilx(GGLfixed v) ALWAYS_INLINE ; 42 inline GGLfixed gglCenterx(GGLfixed v) ALWAYS_INLINE ; 43 inline GGLfixed gglRoundx(GGLfixed v) ALWAYS_INLINE ; 44 45 GGLfixed gglIntToFixed(GGLfixed i) { 46 return i<<FIXED_BITS; 47 } 48 GGLfixed gglFixedToIntRound(GGLfixed f) { 49 return (f + FIXED_HALF)>>FIXED_BITS; 50 } 51 GGLfixed gglFixedToIntFloor(GGLfixed f) { 52 return f>>FIXED_BITS; 53 } 54 GGLfixed gglFixedToIntCeil(GGLfixed f) { 55 return (f + ((1<<FIXED_BITS) - 1))>>FIXED_BITS; 56 } 57 58 GGLfixed gglFracx(GGLfixed v) { 59 return v & ((1<<FIXED_BITS)-1); 60 } 61 GGLfixed gglFloorx(GGLfixed v) { 62 return gglFixedToIntFloor(v)<<FIXED_BITS; 63 } 64 GGLfixed gglCeilx(GGLfixed v) { 65 return gglFixedToIntCeil(v)<<FIXED_BITS; 66 } 67 GGLfixed gglCenterx(GGLfixed v) { 68 return gglFloorx(v + FIXED_HALF) | FIXED_HALF; 69 } 70 GGLfixed gglRoundx(GGLfixed v) { 71 return gglFixedToIntRound(v)<<FIXED_BITS; 72 } 73 74 // conversion from (unsigned) int, short, byte to fixed... 75 #define GGL_B_TO_X(_x) GGLfixed( ((int32_t(_x)+1)>>1)<<10 ) 76 #define GGL_S_TO_X(_x) GGLfixed( ((int32_t(_x)+1)>>1)<<2 ) 77 #define GGL_I_TO_X(_x) GGLfixed( ((int32_t(_x)>>1)+1)>>14 ) 78 #define GGL_UB_TO_X(_x) GGLfixed( uint32_t(_x) + \ 79 (uint32_t(_x)<<8) + \ 80 (uint32_t(_x)>>7) ) 81 #define GGL_US_TO_X(_x) GGLfixed( (_x) + ((_x)>>15) ) 82 #define GGL_UI_TO_X(_x) GGLfixed( (((_x)>>1)+1)>>15 ) 83 84 // ---------------------------------------------------------------------------- 85 86 GGLfixed gglPowx(GGLfixed x, GGLfixed y) CONST; 87 GGLfixed gglSqrtx(GGLfixed a) CONST; 88 GGLfixed gglSqrtRecipx(GGLfixed x) CONST; 89 GGLfixed gglFastDivx(GGLfixed n, GGLfixed d) CONST; 90 int32_t gglMulDivi(int32_t a, int32_t b, int32_t c); 91 92 int32_t gglRecipQNormalized(int32_t x, int* exponent); 93 int32_t gglRecipQ(GGLfixed x, int q) CONST; 94 95 inline GGLfixed gglRecip(GGLfixed x) CONST; 96 inline GGLfixed gglRecip(GGLfixed x) { 97 return gglRecipQ(x, 16); 98 } 99 100 inline GGLfixed gglRecip28(GGLfixed x) CONST; 101 int32_t gglRecip28(GGLfixed x) { 102 return gglRecipQ(x, 28); 103 } 104 105 // ---------------------------------------------------------------------------- 106 107 #if defined(__arm__) && !defined(__thumb__) 108 109 // inline ARM implementations 110 inline GGLfixed gglMulx(GGLfixed x, GGLfixed y, int shift) CONST; 111 inline GGLfixed gglMulx(GGLfixed x, GGLfixed y, int shift) { 112 GGLfixed result, t; 113 if (__builtin_constant_p(shift)) { 114 asm("smull %[lo], %[hi], %[x], %[y] \n" 115 "movs %[lo], %[lo], lsr %[rshift] \n" 116 "adc %[lo], %[lo], %[hi], lsl %[lshift] \n" 117 : [lo]"=r"(result), [hi]"=r"(t), [x]"=r"(x) 118 : "%[x]"(x), [y]"r"(y), [lshift] "I"(32-shift), [rshift] "I"(shift) 119 : "cc" 120 ); 121 } else { 122 asm("smull %[lo], %[hi], %[x], %[y] \n" 123 "movs %[lo], %[lo], lsr %[rshift] \n" 124 "adc %[lo], %[lo], %[hi], lsl %[lshift] \n" 125 : [lo]"=&r"(result), [hi]"=&r"(t), [x]"=&r"(x) 126 : "%[x]"(x), [y]"r"(y), [lshift] "r"(32-shift), [rshift] "r"(shift) 127 : "cc" 128 ); 129 } 130 return result; 131 } 132 133 inline GGLfixed gglMulAddx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) CONST; 134 inline GGLfixed gglMulAddx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) { 135 GGLfixed result, t; 136 if (__builtin_constant_p(shift)) { 137 asm("smull %[lo], %[hi], %[x], %[y] \n" 138 "add %[lo], %[a], %[lo], lsr %[rshift] \n" 139 "add %[lo], %[lo], %[hi], lsl %[lshift] \n" 140 : [lo]"=&r"(result), [hi]"=&r"(t), [x]"=&r"(x) 141 : "%[x]"(x), [y]"r"(y), [a]"r"(a), [lshift] "I"(32-shift), [rshift] "I"(shift) 142 ); 143 } else { 144 asm("smull %[lo], %[hi], %[x], %[y] \n" 145 "add %[lo], %[a], %[lo], lsr %[rshift] \n" 146 "add %[lo], %[lo], %[hi], lsl %[lshift] \n" 147 : [lo]"=&r"(result), [hi]"=&r"(t), [x]"=&r"(x) 148 : "%[x]"(x), [y]"r"(y), [a]"r"(a), [lshift] "r"(32-shift), [rshift] "r"(shift) 149 ); 150 } 151 return result; 152 } 153 154 inline GGLfixed gglMulSubx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) CONST; 155 inline GGLfixed gglMulSubx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) { 156 GGLfixed result, t; 157 if (__builtin_constant_p(shift)) { 158 asm("smull %[lo], %[hi], %[x], %[y] \n" 159 "rsb %[lo], %[a], %[lo], lsr %[rshift] \n" 160 "add %[lo], %[lo], %[hi], lsl %[lshift] \n" 161 : [lo]"=&r"(result), [hi]"=&r"(t), [x]"=&r"(x) 162 : "%[x]"(x), [y]"r"(y), [a]"r"(a), [lshift] "I"(32-shift), [rshift] "I"(shift) 163 ); 164 } else { 165 asm("smull %[lo], %[hi], %[x], %[y] \n" 166 "rsb %[lo], %[a], %[lo], lsr %[rshift] \n" 167 "add %[lo], %[lo], %[hi], lsl %[lshift] \n" 168 : [lo]"=&r"(result), [hi]"=&r"(t), [x]"=&r"(x) 169 : "%[x]"(x), [y]"r"(y), [a]"r"(a), [lshift] "r"(32-shift), [rshift] "r"(shift) 170 ); 171 } 172 return result; 173 } 174 175 inline int64_t gglMulii(int32_t x, int32_t y) CONST; 176 inline int64_t gglMulii(int32_t x, int32_t y) 177 { 178 // 64-bits result: r0=low, r1=high 179 union { 180 struct { 181 int32_t lo; 182 int32_t hi; 183 } s; 184 int64_t res; 185 }; 186 asm("smull %0, %1, %2, %3 \n" 187 : "=r"(s.lo), "=&r"(s.hi) 188 : "%r"(x), "r"(y) 189 : 190 ); 191 return res; 192 } 193 #elif defined(__mips__) && __mips_isa_rev < 6 194 195 /*inline MIPS implementations*/ 196 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b, int shift) CONST; 197 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b, int shift) { 198 GGLfixed result,tmp,tmp1,tmp2; 199 200 if (__builtin_constant_p(shift)) { 201 if (shift == 0) { 202 asm ("mult %[a], %[b] \t\n" 203 "mflo %[res] \t\n" 204 : [res]"=&r"(result),[tmp]"=&r"(tmp) 205 : [a]"r"(a),[b]"r"(b) 206 : "%hi","%lo" 207 ); 208 } else if (shift == 32) 209 { 210 asm ("mult %[a], %[b] \t\n" 211 "li %[tmp],1\t\n" 212 "sll %[tmp],%[tmp],0x1f\t\n" 213 "mflo %[res] \t\n" 214 "addu %[tmp1],%[tmp],%[res] \t\n" 215 "sltu %[tmp1],%[tmp1],%[tmp]\t\n" /*obit*/ 216 "sra %[tmp],%[tmp],0x1f \t\n" 217 "mfhi %[res] \t\n" 218 "addu %[res],%[res],%[tmp]\t\n" 219 "addu %[res],%[res],%[tmp1]\t\n" 220 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1) 221 : [a]"r"(a),[b]"r"(b),[shift]"I"(shift) 222 : "%hi","%lo" 223 ); 224 } else if ((shift >0) && (shift < 32)) 225 { 226 asm ("mult %[a], %[b] \t\n" 227 "li %[tmp],1 \t\n" 228 "sll %[tmp],%[tmp],%[shiftm1] \t\n" 229 "mflo %[res] \t\n" 230 "addu %[tmp1],%[tmp],%[res] \t\n" 231 "sltu %[tmp1],%[tmp1],%[tmp] \t\n" /*obit?*/ 232 "addu %[res],%[res],%[tmp] \t\n" 233 "mfhi %[tmp] \t\n" 234 "addu %[tmp],%[tmp],%[tmp1] \t\n" 235 "sll %[tmp],%[tmp],%[lshift] \t\n" 236 "srl %[res],%[res],%[rshift] \t\n" 237 "or %[res],%[res],%[tmp] \t\n" 238 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 239 : [a]"r"(a),[b]"r"(b),[lshift]"I"(32-shift),[rshift]"I"(shift),[shiftm1]"I"(shift-1) 240 : "%hi","%lo" 241 ); 242 } else { 243 asm ("mult %[a], %[b] \t\n" 244 "li %[tmp],1 \t\n" 245 "sll %[tmp],%[tmp],%[shiftm1] \t\n" 246 "mflo %[res] \t\n" 247 "addu %[tmp1],%[tmp],%[res] \t\n" 248 "sltu %[tmp1],%[tmp1],%[tmp] \t\n" /*obit?*/ 249 "sra %[tmp2],%[tmp],0x1f \t\n" 250 "addu %[res],%[res],%[tmp] \t\n" 251 "mfhi %[tmp] \t\n" 252 "addu %[tmp],%[tmp],%[tmp2] \t\n" 253 "addu %[tmp],%[tmp],%[tmp1] \t\n" /*tmp=hi*/ 254 "srl %[tmp2],%[res],%[rshift] \t\n" 255 "srav %[res], %[tmp],%[rshift]\t\n" 256 "sll %[tmp],%[tmp],1 \t\n" 257 "sll %[tmp],%[tmp],%[norbits] \t\n" 258 "or %[tmp],%[tmp],%[tmp2] \t\n" 259 "movz %[res],%[tmp],%[bit5] \t\n" 260 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 261 : [a]"r"(a),[b]"r"(b),[norbits]"I"(~(shift)),[rshift]"I"(shift),[shiftm1] "I"(shift-1),[bit5]"I"(shift & 0x20) 262 : "%hi","%lo" 263 ); 264 } 265 } else { 266 asm ("mult %[a], %[b] \t\n" 267 "li %[tmp],1 \t\n" 268 "sll %[tmp],%[tmp],%[shiftm1] \t\n" 269 "mflo %[res] \t\n" 270 "addu %[tmp1],%[tmp],%[res] \t\n" 271 "sltu %[tmp1],%[tmp1],%[tmp] \t\n" /*obit?*/ 272 "sra %[tmp2],%[tmp],0x1f \t\n" 273 "addu %[res],%[res],%[tmp] \t\n" 274 "mfhi %[tmp] \t\n" 275 "addu %[tmp],%[tmp],%[tmp2] \t\n" 276 "addu %[tmp],%[tmp],%[tmp1] \t\n" /*tmp=hi*/ 277 "srl %[tmp2],%[res],%[rshift] \t\n" 278 "srav %[res], %[tmp],%[rshift]\t\n" 279 "sll %[tmp],%[tmp],1 \t\n" 280 "sll %[tmp],%[tmp],%[norbits] \t\n" 281 "or %[tmp],%[tmp],%[tmp2] \t\n" 282 "movz %[res],%[tmp],%[bit5] \t\n" 283 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 284 : [a]"r"(a),[b]"r"(b),[norbits]"r"(~(shift)),[rshift] "r"(shift),[shiftm1]"r"(shift-1),[bit5] "r"(shift & 0x20) 285 : "%hi","%lo" 286 ); 287 } 288 289 return result; 290 } 291 292 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) CONST; 293 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) { 294 GGLfixed result,t,tmp1,tmp2; 295 296 if (__builtin_constant_p(shift)) { 297 if (shift == 0) { 298 asm ("mult %[a], %[b] \t\n" 299 "mflo %[lo] \t\n" 300 "addu %[lo],%[lo],%[c] \t\n" 301 : [lo]"=&r"(result) 302 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 303 : "%hi","%lo" 304 ); 305 } else if (shift == 32) { 306 asm ("mult %[a], %[b] \t\n" 307 "mfhi %[lo] \t\n" 308 "addu %[lo],%[lo],%[c] \t\n" 309 : [lo]"=&r"(result) 310 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 311 : "%hi","%lo" 312 ); 313 } else if ((shift>0) && (shift<32)) { 314 asm ("mult %[a], %[b] \t\n" 315 "mflo %[res] \t\n" 316 "mfhi %[t] \t\n" 317 "srl %[res],%[res],%[rshift] \t\n" 318 "sll %[t],%[t],%[lshift] \t\n" 319 "or %[res],%[res],%[t] \t\n" 320 "addu %[res],%[res],%[c] \t\n" 321 : [res]"=&r"(result),[t]"=&r"(t) 322 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[lshift]"I"(32-shift),[rshift]"I"(shift) 323 : "%hi","%lo" 324 ); 325 } else { 326 asm ("mult %[a], %[b] \t\n" 327 "nor %[tmp1],$zero,%[shift]\t\n" 328 "mflo %[res] \t\n" 329 "mfhi %[t] \t\n" 330 "srl %[res],%[res],%[shift] \t\n" 331 "sll %[tmp2],%[t],1 \t\n" 332 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 333 "or %[tmp1],%[tmp2],%[res] \t\n" 334 "srav %[res],%[t],%[shift] \t\n" 335 "andi %[tmp2],%[shift],0x20\t\n" 336 "movz %[res],%[tmp1],%[tmp2]\t\n" 337 "addu %[res],%[res],%[c] \t\n" 338 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 339 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"I"(shift) 340 : "%hi","%lo" 341 ); 342 } 343 } else { 344 asm ("mult %[a], %[b] \t\n" 345 "nor %[tmp1],$zero,%[shift]\t\n" 346 "mflo %[res] \t\n" 347 "mfhi %[t] \t\n" 348 "srl %[res],%[res],%[shift] \t\n" 349 "sll %[tmp2],%[t],1 \t\n" 350 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 351 "or %[tmp1],%[tmp2],%[res] \t\n" 352 "srav %[res],%[t],%[shift] \t\n" 353 "andi %[tmp2],%[shift],0x20\t\n" 354 "movz %[res],%[tmp1],%[tmp2]\t\n" 355 "addu %[res],%[res],%[c] \t\n" 356 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 357 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"r"(shift) 358 : "%hi","%lo" 359 ); 360 } 361 return result; 362 } 363 364 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) CONST; 365 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) { 366 GGLfixed result,t,tmp1,tmp2; 367 368 if (__builtin_constant_p(shift)) { 369 if (shift == 0) { 370 asm ("mult %[a], %[b] \t\n" 371 "mflo %[lo] \t\n" 372 "subu %[lo],%[lo],%[c] \t\n" 373 : [lo]"=&r"(result) 374 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 375 : "%hi","%lo" 376 ); 377 } else if (shift == 32) { 378 asm ("mult %[a], %[b] \t\n" 379 "mfhi %[lo] \t\n" 380 "subu %[lo],%[lo],%[c] \t\n" 381 : [lo]"=&r"(result) 382 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 383 : "%hi","%lo" 384 ); 385 } else if ((shift>0) && (shift<32)) { 386 asm ("mult %[a], %[b] \t\n" 387 "mflo %[res] \t\n" 388 "mfhi %[t] \t\n" 389 "srl %[res],%[res],%[rshift] \t\n" 390 "sll %[t],%[t],%[lshift] \t\n" 391 "or %[res],%[res],%[t] \t\n" 392 "subu %[res],%[res],%[c] \t\n" 393 : [res]"=&r"(result),[t]"=&r"(t) 394 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[lshift]"I"(32-shift),[rshift]"I"(shift) 395 : "%hi","%lo" 396 ); 397 } else { 398 asm ("mult %[a], %[b] \t\n" 399 "nor %[tmp1],$zero,%[shift]\t\n" 400 "mflo %[res] \t\n" 401 "mfhi %[t] \t\n" 402 "srl %[res],%[res],%[shift] \t\n" 403 "sll %[tmp2],%[t],1 \t\n" 404 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 405 "or %[tmp1],%[tmp2],%[res] \t\n" 406 "srav %[res],%[t],%[shift] \t\n" 407 "andi %[tmp2],%[shift],0x20\t\n" 408 "movz %[res],%[tmp1],%[tmp2]\t\n" 409 "subu %[res],%[res],%[c] \t\n" 410 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 411 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"I"(shift) 412 : "%hi","%lo" 413 ); 414 } 415 } else { 416 asm ("mult %[a], %[b] \t\n" 417 "nor %[tmp1],$zero,%[shift]\t\n" 418 "mflo %[res] \t\n" 419 "mfhi %[t] \t\n" 420 "srl %[res],%[res],%[shift] \t\n" 421 "sll %[tmp2],%[t],1 \t\n" 422 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 423 "or %[tmp1],%[tmp2],%[res] \t\n" 424 "srav %[res],%[t],%[shift] \t\n" 425 "andi %[tmp2],%[shift],0x20\t\n" 426 "movz %[res],%[tmp1],%[tmp2]\t\n" 427 "subu %[res],%[res],%[c] \t\n" 428 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 429 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"r"(shift) 430 : "%hi","%lo" 431 ); 432 } 433 return result; 434 } 435 436 inline int64_t gglMulii(int32_t x, int32_t y) CONST; 437 inline int64_t gglMulii(int32_t x, int32_t y) { 438 union { 439 struct { 440 #if defined(__MIPSEL__) 441 int32_t lo; 442 int32_t hi; 443 #elif defined(__MIPSEB__) 444 int32_t hi; 445 int32_t lo; 446 #endif 447 } s; 448 int64_t res; 449 }u; 450 asm("mult %2, %3 \t\n" 451 "mfhi %1 \t\n" 452 "mflo %0 \t\n" 453 : "=r"(u.s.lo), "=&r"(u.s.hi) 454 : "%r"(x), "r"(y) 455 : "%hi","%lo" 456 ); 457 return u.res; 458 } 459 460 #elif defined(__aarch64__) 461 462 // inline AArch64 implementations 463 464 inline GGLfixed gglMulx(GGLfixed x, GGLfixed y, int shift) CONST; 465 inline GGLfixed gglMulx(GGLfixed x, GGLfixed y, int shift) 466 { 467 GGLfixed result; 468 GGLfixed round; 469 470 asm("mov %x[round], #1 \n" 471 "lsl %x[round], %x[round], %x[shift] \n" 472 "lsr %x[round], %x[round], #1 \n" 473 "smaddl %x[result], %w[x], %w[y],%x[round] \n" 474 "lsr %x[result], %x[result], %x[shift] \n" 475 : [round]"=&r"(round), [result]"=&r"(result) \ 476 : [x]"r"(x), [y]"r"(y), [shift] "r"(shift) \ 477 : 478 ); 479 return result; 480 } 481 inline GGLfixed gglMulAddx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) CONST; 482 inline GGLfixed gglMulAddx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) 483 { 484 GGLfixed result; 485 asm("smull %x[result], %w[x], %w[y] \n" 486 "lsr %x[result], %x[result], %x[shift] \n" 487 "add %w[result], %w[result], %w[a] \n" 488 : [result]"=&r"(result) \ 489 : [x]"r"(x), [y]"r"(y), [a]"r"(a), [shift] "r"(shift) \ 490 : 491 ); 492 return result; 493 } 494 495 inline GGLfixed gglMulSubx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) CONST; 496 inline GGLfixed gglMulSubx(GGLfixed x, GGLfixed y, GGLfixed a, int shift) 497 { 498 499 GGLfixed result; 500 501 asm("smull %x[result], %w[x], %w[y] \n" 502 "lsr %x[result], %x[result], %x[shift] \n" 503 "sub %w[result], %w[result], %w[a] \n" 504 : [result]"=&r"(result) \ 505 : [x]"r"(x), [y]"r"(y), [a]"r"(a), [shift] "r"(shift) \ 506 : 507 ); 508 return result; 509 } 510 inline int64_t gglMulii(int32_t x, int32_t y) CONST; 511 inline int64_t gglMulii(int32_t x, int32_t y) 512 { 513 int64_t res; 514 asm("smull %x0, %w1, %w2 \n" 515 : "=r"(res) 516 : "%r"(x), "r"(y) 517 : 518 ); 519 return res; 520 } 521 522 #elif defined(__mips__) && __mips_isa_rev == 6 523 524 /*inline MIPS implementations*/ 525 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b, int shift) CONST; 526 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b, int shift) { 527 GGLfixed result,tmp,tmp1,tmp2; 528 529 if (__builtin_constant_p(shift)) { 530 if (shift == 0) { 531 asm ("mul %[res], %[a], %[b] \t\n" 532 : [res]"=&r"(result) 533 : [a]"r"(a),[b]"r"(b) 534 ); 535 } else if (shift == 32) 536 { 537 asm ("mul %[res], %[a], %[b] \t\n" 538 "li %[tmp],1\t\n" 539 "sll %[tmp],%[tmp],0x1f\t\n" 540 "addu %[tmp1],%[tmp],%[res] \t\n" 541 "muh %[res], %[a], %[b] \t\n" 542 "sltu %[tmp1],%[tmp1],%[tmp]\t\n" /*obit*/ 543 "sra %[tmp],%[tmp],0x1f \t\n" 544 "addu %[res],%[res],%[tmp]\t\n" 545 "addu %[res],%[res],%[tmp1]\t\n" 546 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1) 547 : [a]"r"(a),[b]"r"(b),[shift]"I"(shift) 548 ); 549 } else if ((shift >0) && (shift < 32)) 550 { 551 asm ("mul %[res], %[a], %[b] \t\n" 552 "li %[tmp],1 \t\n" 553 "sll %[tmp],%[tmp],%[shiftm1] \t\n" 554 "addu %[tmp1],%[tmp],%[res] \t\n" 555 "sltu %[tmp1],%[tmp1],%[tmp] \t\n" /*obit?*/ 556 "addu %[res],%[res],%[tmp] \t\n" 557 "muh %[tmp], %[a], %[b] \t\n" 558 "addu %[tmp],%[tmp],%[tmp1] \t\n" 559 "sll %[tmp],%[tmp],%[lshift] \t\n" 560 "srl %[res],%[res],%[rshift] \t\n" 561 "or %[res],%[res],%[tmp] \t\n" 562 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 563 : [a]"r"(a),[b]"r"(b),[lshift]"I"(32-shift),[rshift]"I"(shift),[shiftm1]"I"(shift-1) 564 ); 565 } else { 566 asm ("mul %[res], %[a], %[b] \t\n" 567 "li %[tmp],1 \t\n" 568 "sll %[tmp],%[tmp],%[shiftm1] \t\n" 569 "addu %[tmp1],%[tmp],%[res] \t\n" 570 "sltu %[tmp1],%[tmp1],%[tmp] \t\n" /*obit?*/ 571 "sra %[tmp2],%[tmp],0x1f \t\n" 572 "addu %[res],%[res],%[tmp] \t\n" 573 "muh %[tmp], %[a], %[b] \t\n" 574 "addu %[tmp],%[tmp],%[tmp2] \t\n" 575 "addu %[tmp],%[tmp],%[tmp1] \t\n" /*tmp=hi*/ 576 "srl %[tmp2],%[res],%[rshift] \t\n" 577 "srav %[res], %[tmp],%[rshift]\t\n" 578 "sll %[tmp],%[tmp],1 \t\n" 579 "sll %[tmp],%[tmp],%[norbits] \t\n" 580 "or %[tmp],%[tmp],%[tmp2] \t\n" 581 "seleqz %[tmp],%[tmp],%[bit5] \t\n" 582 "selnez %[res],%[res],%[bit5] \t\n" 583 "or %[res],%[res],%[tmp] \t\n" 584 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 585 : [a]"r"(a),[b]"r"(b),[norbits]"I"(~(shift)),[rshift]"I"(shift),[shiftm1] "I"(shift-1),[bit5]"I"(shift & 0x20) 586 ); 587 } 588 } else { 589 asm ("mul %[res], %[a], %[b] \t\n" 590 "li %[tmp],1 \t\n" 591 "sll %[tmp],%[tmp],%[shiftm1] \t\n" 592 "addu %[tmp1],%[tmp],%[res] \t\n" 593 "sltu %[tmp1],%[tmp1],%[tmp] \t\n" /*obit?*/ 594 "sra %[tmp2],%[tmp],0x1f \t\n" 595 "addu %[res],%[res],%[tmp] \t\n" 596 "muh %[tmp], %[a], %[b] \t\n" 597 "addu %[tmp],%[tmp],%[tmp2] \t\n" 598 "addu %[tmp],%[tmp],%[tmp1] \t\n" /*tmp=hi*/ 599 "srl %[tmp2],%[res],%[rshift] \t\n" 600 "srav %[res], %[tmp],%[rshift]\t\n" 601 "sll %[tmp],%[tmp],1 \t\n" 602 "sll %[tmp],%[tmp],%[norbits] \t\n" 603 "or %[tmp],%[tmp],%[tmp2] \t\n" 604 "seleqz %[tmp],%[tmp],%[bit5] \t\n" 605 "selnez %[res],%[res],%[bit5] \t\n" 606 "or %[res],%[res],%[tmp] \t\n" 607 : [res]"=&r"(result),[tmp]"=&r"(tmp),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 608 : [a]"r"(a),[b]"r"(b),[norbits]"r"(~(shift)),[rshift] "r"(shift),[shiftm1]"r"(shift-1),[bit5] "r"(shift & 0x20) 609 ); 610 } 611 return result; 612 } 613 614 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) CONST; 615 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) { 616 GGLfixed result,t,tmp1,tmp2; 617 618 if (__builtin_constant_p(shift)) { 619 if (shift == 0) { 620 asm ("mul %[lo], %[a], %[b] \t\n" 621 "addu %[lo],%[lo],%[c] \t\n" 622 : [lo]"=&r"(result) 623 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 624 ); 625 } else if (shift == 32) { 626 asm ("muh %[lo], %[a], %[b] \t\n" 627 "addu %[lo],%[lo],%[c] \t\n" 628 : [lo]"=&r"(result) 629 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 630 ); 631 } else if ((shift>0) && (shift<32)) { 632 asm ("mul %[res], %[a], %[b] \t\n" 633 "muh %[t], %[a], %[b] \t\n" 634 "srl %[res],%[res],%[rshift] \t\n" 635 "sll %[t],%[t],%[lshift] \t\n" 636 "or %[res],%[res],%[t] \t\n" 637 "addu %[res],%[res],%[c] \t\n" 638 : [res]"=&r"(result),[t]"=&r"(t) 639 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[lshift]"I"(32-shift),[rshift]"I"(shift) 640 ); 641 } else { 642 asm ("mul %[res], %[a], %[b] \t\n" 643 "muh %[t], %[a], %[b] \t\n" 644 "nor %[tmp1],$zero,%[shift]\t\n" 645 "srl %[res],%[res],%[shift] \t\n" 646 "sll %[tmp2],%[t],1 \t\n" 647 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 648 "or %[tmp1],%[tmp2],%[res] \t\n" 649 "srav %[res],%[t],%[shift] \t\n" 650 "andi %[tmp2],%[shift],0x20\t\n" 651 "seleqz %[tmp1],%[tmp1],%[tmp2]\t\n" 652 "selnez %[res],%[res],%[tmp2]\t\n" 653 "or %[res],%[res],%[tmp1]\t\n" 654 "addu %[res],%[res],%[c] \t\n" 655 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 656 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"I"(shift) 657 ); 658 } 659 } else { 660 asm ("mul %[res], %[a], %[b] \t\n" 661 "muh %[t], %[a], %[b] \t\n" 662 "nor %[tmp1],$zero,%[shift]\t\n" 663 "srl %[res],%[res],%[shift] \t\n" 664 "sll %[tmp2],%[t],1 \t\n" 665 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 666 "or %[tmp1],%[tmp2],%[res] \t\n" 667 "srav %[res],%[t],%[shift] \t\n" 668 "andi %[tmp2],%[shift],0x20\t\n" 669 "seleqz %[tmp1],%[tmp1],%[tmp2]\t\n" 670 "selnez %[res],%[res],%[tmp2]\t\n" 671 "or %[res],%[res],%[tmp1]\t\n" 672 "addu %[res],%[res],%[c] \t\n" 673 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 674 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"r"(shift) 675 ); 676 } 677 return result; 678 } 679 680 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) CONST; 681 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) { 682 GGLfixed result,t,tmp1,tmp2; 683 684 if (__builtin_constant_p(shift)) { 685 if (shift == 0) { 686 asm ("mul %[lo], %[a], %[b] \t\n" 687 "subu %[lo],%[lo],%[c] \t\n" 688 : [lo]"=&r"(result) 689 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 690 ); 691 } else if (shift == 32) { 692 asm ("muh %[lo], %[a], %[b] \t\n" 693 "subu %[lo],%[lo],%[c] \t\n" 694 : [lo]"=&r"(result) 695 : [a]"r"(a),[b]"r"(b),[c]"r"(c) 696 ); 697 } else if ((shift>0) && (shift<32)) { 698 asm ("mul %[res], %[a], %[b] \t\n" 699 "muh %[t], %[a], %[b] \t\n" 700 "srl %[res],%[res],%[rshift] \t\n" 701 "sll %[t],%[t],%[lshift] \t\n" 702 "or %[res],%[res],%[t] \t\n" 703 "subu %[res],%[res],%[c] \t\n" 704 : [res]"=&r"(result),[t]"=&r"(t) 705 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[lshift]"I"(32-shift),[rshift]"I"(shift) 706 ); 707 } else { 708 asm ("mul %[res], %[a], %[b] \t\n" 709 "muh %[t], %[a], %[b] \t\n" 710 "nor %[tmp1],$zero,%[shift]\t\n" 711 "srl %[res],%[res],%[shift] \t\n" 712 "sll %[tmp2],%[t],1 \t\n" 713 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 714 "or %[tmp1],%[tmp2],%[res] \t\n" 715 "srav %[res],%[t],%[shift] \t\n" 716 "andi %[tmp2],%[shift],0x20\t\n" 717 "seleqz %[tmp1],%[tmp1],%[tmp2]\t\n" 718 "selnez %[res],%[res],%[tmp2]\t\n" 719 "or %[res],%[res],%[tmp1]\t\n" 720 "subu %[res],%[res],%[c] \t\n" 721 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 722 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"I"(shift) 723 ); 724 } 725 } else { 726 asm ("mul %[res], %[a], %[b] \t\n" 727 "muh %[t], %[a], %[b] \t\n" 728 "nor %[tmp1],$zero,%[shift]\t\n" 729 "srl %[res],%[res],%[shift] \t\n" 730 "sll %[tmp2],%[t],1 \t\n" 731 "sllv %[tmp2],%[tmp2],%[tmp1] \t\n" 732 "or %[tmp1],%[tmp2],%[res] \t\n" 733 "srav %[res],%[t],%[shift] \t\n" 734 "andi %[tmp2],%[shift],0x20\t\n" 735 "seleqz %[tmp1],%[tmp1],%[tmp2]\t\n" 736 "selnez %[res],%[res],%[tmp2]\t\n" 737 "or %[res],%[res],%[tmp1]\t\n" 738 "subu %[res],%[res],%[c] \t\n" 739 : [res]"=&r"(result),[t]"=&r"(t),[tmp1]"=&r"(tmp1),[tmp2]"=&r"(tmp2) 740 : [a]"r"(a),[b]"r"(b),[c]"r"(c),[shift]"r"(shift) 741 ); 742 } 743 return result; 744 } 745 746 inline int64_t gglMulii(int32_t x, int32_t y) CONST; 747 inline int64_t gglMulii(int32_t x, int32_t y) { 748 union { 749 struct { 750 #if defined(__MIPSEL__) 751 int32_t lo; 752 int32_t hi; 753 #elif defined(__MIPSEB__) 754 int32_t hi; 755 int32_t lo; 756 #endif 757 } s; 758 int64_t res; 759 }u; 760 asm("mul %0, %2, %3 \t\n" 761 "muh %1, %2, %3 \t\n" 762 : "=r"(u.s.lo), "=&r"(u.s.hi) 763 : "%r"(x), "r"(y) 764 ); 765 return u.res; 766 } 767 768 #else // ---------------------------------------------------------------------- 769 770 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b, int shift) CONST; 771 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b, int shift) { 772 return GGLfixed((int64_t(a)*b + (1<<(shift-1)))>>shift); 773 } 774 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) CONST; 775 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) { 776 return GGLfixed((int64_t(a)*b)>>shift) + c; 777 } 778 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) CONST; 779 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c, int shift) { 780 return GGLfixed((int64_t(a)*b)>>shift) - c; 781 } 782 inline int64_t gglMulii(int32_t a, int32_t b) CONST; 783 inline int64_t gglMulii(int32_t a, int32_t b) { 784 return int64_t(a)*b; 785 } 786 787 #endif 788 789 // ------------------------------------------------------------------------ 790 791 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b) CONST; 792 inline GGLfixed gglMulx(GGLfixed a, GGLfixed b) { 793 return gglMulx(a, b, 16); 794 } 795 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c) CONST; 796 inline GGLfixed gglMulAddx(GGLfixed a, GGLfixed b, GGLfixed c) { 797 return gglMulAddx(a, b, c, 16); 798 } 799 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c) CONST; 800 inline GGLfixed gglMulSubx(GGLfixed a, GGLfixed b, GGLfixed c) { 801 return gglMulSubx(a, b, c, 16); 802 } 803 804 // ------------------------------------------------------------------------ 805 806 inline int32_t gglClz(int32_t x) CONST; 807 inline int32_t gglClz(int32_t x) 808 { 809 #if (defined(__arm__) && !defined(__thumb__)) || defined(__mips__) || defined(__aarch64__) 810 return __builtin_clz(x); 811 #else 812 if (!x) return 32; 813 int32_t exp = 31; 814 if (x & 0xFFFF0000) { exp -=16; x >>= 16; } 815 if (x & 0x0000ff00) { exp -= 8; x >>= 8; } 816 if (x & 0x000000f0) { exp -= 4; x >>= 4; } 817 if (x & 0x0000000c) { exp -= 2; x >>= 2; } 818 if (x & 0x00000002) { exp -= 1; } 819 return exp; 820 #endif 821 } 822 823 // ------------------------------------------------------------------------ 824 825 int32_t gglDivQ(GGLfixed n, GGLfixed d, int32_t i) CONST; 826 827 inline int32_t gglDivQ16(GGLfixed n, GGLfixed d) CONST; 828 inline int32_t gglDivQ16(GGLfixed n, GGLfixed d) { 829 return gglDivQ(n, d, 16); 830 } 831 832 inline int32_t gglDivx(GGLfixed n, GGLfixed d) CONST; 833 inline int32_t gglDivx(GGLfixed n, GGLfixed d) { 834 return gglDivQ(n, d, 16); 835 } 836 837 // ------------------------------------------------------------------------ 838 839 inline GGLfixed gglRecipFast(GGLfixed x) CONST; 840 inline GGLfixed gglRecipFast(GGLfixed x) 841 { 842 // This is a really bad approximation of 1/x, but it's also 843 // very fast. x must be strictly positive. 844 // if x between [0.5, 1[ , then 1/x = 3-2*x 845 // (we use 2.30 fixed-point) 846 const int32_t lz = gglClz(x); 847 return (0xC0000000 - (x << (lz - 1))) >> (30-lz); 848 } 849 850 // ------------------------------------------------------------------------ 851 852 inline GGLfixed gglClampx(GGLfixed c) CONST; 853 inline GGLfixed gglClampx(GGLfixed c) 854 { 855 #if defined(__thumb__) 856 // clamp without branches 857 c &= ~(c>>31); c = FIXED_ONE - c; 858 c &= ~(c>>31); c = FIXED_ONE - c; 859 #else 860 #if defined(__arm__) 861 // I don't know why gcc thinks its smarter than me! The code below 862 // clamps to zero in one instruction, but gcc won't generate it and 863 // replace it by a cmp + movlt (it's quite amazing actually). 864 asm("bic %0, %1, %1, asr #31\n" : "=r"(c) : "r"(c)); 865 #elif defined(__aarch64__) 866 asm("bic %w0, %w1, %w1, asr #31\n" : "=r"(c) : "r"(c)); 867 #else 868 c &= ~(c>>31); 869 #endif 870 if (c>FIXED_ONE) 871 c = FIXED_ONE; 872 #endif 873 return c; 874 } 875 876 // ------------------------------------------------------------------------ 877 878 #endif // ANDROID_GGL_FIXED_H 879
