1 /************************************************************************** 2 * 3 * Copyright 2009 VMware, Inc. 4 * All Rights Reserved. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the 8 * "Software"), to deal in the Software without restriction, including 9 * without limitation the rights to use, copy, modify, merge, publish, 10 * distribute, sub license, and/or sell copies of the Software, and to 11 * permit persons to whom the Software is furnished to do so, subject to 12 * the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the 15 * next paragraph) shall be included in all copies or substantial portions 16 * of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR 22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 * 26 **************************************************************************/ 27 28 /** 29 * @file 30 * Helper arithmetic functions. 31 * 32 * @author Jose Fonseca <jfonseca@vmware.com> 33 */ 34 35 36 #ifndef LP_BLD_ARIT_H 37 #define LP_BLD_ARIT_H 38 39 40 #include "gallivm/lp_bld.h" 41 #include "pipe/p_compiler.h" 42 43 44 struct lp_type; 45 struct lp_build_context; 46 struct gallivm_state; 47 48 49 /** 50 * Complement, i.e., 1 - a. 51 */ 52 LLVMValueRef 53 lp_build_comp(struct lp_build_context *bld, 54 LLVMValueRef a); 55 56 LLVMValueRef 57 lp_build_add(struct lp_build_context *bld, 58 LLVMValueRef a, 59 LLVMValueRef b); 60 61 LLVMValueRef 62 lp_build_horizontal_add(struct lp_build_context *bld, 63 LLVMValueRef a); 64 65 LLVMValueRef 66 lp_build_hadd_partial4(struct lp_build_context *bld, 67 LLVMValueRef vectors[], 68 unsigned num_vecs); 69 70 LLVMValueRef 71 lp_build_sub(struct lp_build_context *bld, 72 LLVMValueRef a, 73 LLVMValueRef b); 74 75 76 LLVMValueRef 77 lp_build_mul_norm(struct gallivm_state *gallivm, 78 struct lp_type wide_type, 79 LLVMValueRef a, 80 LLVMValueRef b); 81 82 LLVMValueRef 83 lp_build_mul(struct lp_build_context *bld, 84 LLVMValueRef a, 85 LLVMValueRef b); 86 87 LLVMValueRef 88 lp_build_mul_32_lohi_cpu(struct lp_build_context *bld, 89 LLVMValueRef a, 90 LLVMValueRef b, 91 LLVMValueRef *res_hi); 92 93 LLVMValueRef 94 lp_build_mul_32_lohi(struct lp_build_context *bld, 95 LLVMValueRef a, 96 LLVMValueRef b, 97 LLVMValueRef *res_hi); 98 99 LLVMValueRef 100 lp_build_mul_imm(struct lp_build_context *bld, 101 LLVMValueRef a, 102 int b); 103 104 LLVMValueRef 105 lp_build_div(struct lp_build_context *bld, 106 LLVMValueRef a, 107 LLVMValueRef b); 108 109 110 /* llvm.fmuladd.* intrinsic */ 111 LLVMValueRef 112 lp_build_fmuladd(LLVMBuilderRef builder, 113 LLVMValueRef a, 114 LLVMValueRef b, 115 LLVMValueRef c); 116 117 /* a * b + c */ 118 LLVMValueRef 119 lp_build_mad(struct lp_build_context *bld, 120 LLVMValueRef a, 121 LLVMValueRef b, 122 LLVMValueRef c); 123 124 125 /** 126 * Set when the weights for normalized are prescaled, that is, in range 127 * 0..2**n, as opposed to range 0..2**(n-1). 128 */ 129 #define LP_BLD_LERP_PRESCALED_WEIGHTS (1 << 0) 130 131 /** 132 * Used internally when using wide intermediates for normalized lerps. 133 * 134 * Do not use. 135 */ 136 #define LP_BLD_LERP_WIDE_NORMALIZED (1 << 1) 137 138 LLVMValueRef 139 lp_build_lerp(struct lp_build_context *bld, 140 LLVMValueRef x, 141 LLVMValueRef v0, 142 LLVMValueRef v1, 143 unsigned flags); 144 145 LLVMValueRef 146 lp_build_lerp_2d(struct lp_build_context *bld, 147 LLVMValueRef x, 148 LLVMValueRef y, 149 LLVMValueRef v00, 150 LLVMValueRef v01, 151 LLVMValueRef v10, 152 LLVMValueRef v11, 153 unsigned flags); 154 155 LLVMValueRef 156 lp_build_lerp_3d(struct lp_build_context *bld, 157 LLVMValueRef x, 158 LLVMValueRef y, 159 LLVMValueRef z, 160 LLVMValueRef v000, 161 LLVMValueRef v001, 162 LLVMValueRef v010, 163 LLVMValueRef v011, 164 LLVMValueRef v100, 165 LLVMValueRef v101, 166 LLVMValueRef v110, 167 LLVMValueRef v111, 168 unsigned flags); 169 170 /** 171 * Specifies floating point NaN behavior. 172 */ 173 enum gallivm_nan_behavior { 174 /* Results are undefined with NaN. Results in fastest code */ 175 GALLIVM_NAN_BEHAVIOR_UNDEFINED, 176 /* If one of the inputs is NaN, the other operand is returned */ 177 GALLIVM_NAN_RETURN_OTHER, 178 /* If one of the inputs is NaN, the other operand is returned, 179 * but we guarantee the second operand is not a NaN. 180 * In min/max it will be as fast as undefined with sse opcodes, 181 * and archs having native return_other can benefit too. */ 182 GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN, 183 /* If one of the inputs is NaN, NaN is returned, 184 * but we guarantee the first operand is not a NaN. 185 * In min/max it will be as fast as undefined with sse opcodes, 186 * and archs having native return_nan can benefit too. */ 187 GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN, 188 189 }; 190 191 LLVMValueRef 192 lp_build_min(struct lp_build_context *bld, 193 LLVMValueRef a, 194 LLVMValueRef b); 195 196 LLVMValueRef 197 lp_build_min_ext(struct lp_build_context *bld, 198 LLVMValueRef a, 199 LLVMValueRef b, 200 enum gallivm_nan_behavior nan_behavior); 201 202 LLVMValueRef 203 lp_build_max(struct lp_build_context *bld, 204 LLVMValueRef a, 205 LLVMValueRef b); 206 207 LLVMValueRef 208 lp_build_max_ext(struct lp_build_context *bld, 209 LLVMValueRef a, 210 LLVMValueRef b, 211 enum gallivm_nan_behavior nan_behavior); 212 213 LLVMValueRef 214 lp_build_clamp(struct lp_build_context *bld, 215 LLVMValueRef a, 216 LLVMValueRef min, 217 LLVMValueRef max); 218 219 LLVMValueRef 220 lp_build_clamp_zero_one_nanzero(struct lp_build_context *bld, 221 LLVMValueRef a); 222 223 LLVMValueRef 224 lp_build_abs(struct lp_build_context *bld, 225 LLVMValueRef a); 226 227 LLVMValueRef 228 lp_build_negate(struct lp_build_context *bld, 229 LLVMValueRef a); 230 231 LLVMValueRef 232 lp_build_sgn(struct lp_build_context *bld, 233 LLVMValueRef a); 234 235 LLVMValueRef 236 lp_build_set_sign(struct lp_build_context *bld, 237 LLVMValueRef a, LLVMValueRef sign); 238 239 LLVMValueRef 240 lp_build_int_to_float(struct lp_build_context *bld, 241 LLVMValueRef a); 242 243 LLVMValueRef 244 lp_build_round(struct lp_build_context *bld, 245 LLVMValueRef a); 246 247 LLVMValueRef 248 lp_build_floor(struct lp_build_context *bld, 249 LLVMValueRef a); 250 251 LLVMValueRef 252 lp_build_ceil(struct lp_build_context *bld, 253 LLVMValueRef a); 254 255 LLVMValueRef 256 lp_build_trunc(struct lp_build_context *bld, 257 LLVMValueRef a); 258 259 LLVMValueRef 260 lp_build_fract(struct lp_build_context *bld, 261 LLVMValueRef a); 262 263 LLVMValueRef 264 lp_build_fract_safe(struct lp_build_context *bld, 265 LLVMValueRef a); 266 267 LLVMValueRef 268 lp_build_ifloor(struct lp_build_context *bld, 269 LLVMValueRef a); 270 LLVMValueRef 271 lp_build_iceil(struct lp_build_context *bld, 272 LLVMValueRef a); 273 274 LLVMValueRef 275 lp_build_iround(struct lp_build_context *bld, 276 LLVMValueRef a); 277 278 LLVMValueRef 279 lp_build_itrunc(struct lp_build_context *bld, 280 LLVMValueRef a); 281 282 void 283 lp_build_ifloor_fract(struct lp_build_context *bld, 284 LLVMValueRef a, 285 LLVMValueRef *out_ipart, 286 LLVMValueRef *out_fpart); 287 288 void 289 lp_build_ifloor_fract_safe(struct lp_build_context *bld, 290 LLVMValueRef a, 291 LLVMValueRef *out_ipart, 292 LLVMValueRef *out_fpart); 293 294 LLVMValueRef 295 lp_build_sqrt(struct lp_build_context *bld, 296 LLVMValueRef a); 297 298 LLVMValueRef 299 lp_build_rcp(struct lp_build_context *bld, 300 LLVMValueRef a); 301 302 LLVMValueRef 303 lp_build_rsqrt(struct lp_build_context *bld, 304 LLVMValueRef a); 305 306 boolean 307 lp_build_fast_rsqrt_available(struct lp_type type); 308 309 LLVMValueRef 310 lp_build_fast_rsqrt(struct lp_build_context *bld, 311 LLVMValueRef a); 312 313 LLVMValueRef 314 lp_build_polynomial(struct lp_build_context *bld, 315 LLVMValueRef x, 316 const double *coeffs, 317 unsigned num_coeffs); 318 319 LLVMValueRef 320 lp_build_cos(struct lp_build_context *bld, 321 LLVMValueRef a); 322 323 LLVMValueRef 324 lp_build_sin(struct lp_build_context *bld, 325 LLVMValueRef a); 326 327 LLVMValueRef 328 lp_build_pow(struct lp_build_context *bld, 329 LLVMValueRef a, 330 LLVMValueRef b); 331 332 LLVMValueRef 333 lp_build_exp(struct lp_build_context *bld, 334 LLVMValueRef a); 335 336 LLVMValueRef 337 lp_build_log(struct lp_build_context *bld, 338 LLVMValueRef a); 339 340 LLVMValueRef 341 lp_build_log_safe(struct lp_build_context *bld, 342 LLVMValueRef a); 343 344 LLVMValueRef 345 lp_build_exp2(struct lp_build_context *bld, 346 LLVMValueRef a); 347 348 LLVMValueRef 349 lp_build_extract_exponent(struct lp_build_context *bld, 350 LLVMValueRef x, 351 int bias); 352 353 LLVMValueRef 354 lp_build_extract_mantissa(struct lp_build_context *bld, 355 LLVMValueRef x); 356 357 LLVMValueRef 358 lp_build_log2(struct lp_build_context *bld, 359 LLVMValueRef a); 360 361 LLVMValueRef 362 lp_build_log2_safe(struct lp_build_context *bld, 363 LLVMValueRef a); 364 365 LLVMValueRef 366 lp_build_fast_log2(struct lp_build_context *bld, 367 LLVMValueRef a); 368 369 LLVMValueRef 370 lp_build_ilog2(struct lp_build_context *bld, 371 LLVMValueRef x); 372 373 void 374 lp_build_log2_approx(struct lp_build_context *bld, 375 LLVMValueRef x, 376 LLVMValueRef *p_exp, 377 LLVMValueRef *p_floor_log2, 378 LLVMValueRef *p_log2, 379 boolean handle_nans); 380 381 LLVMValueRef 382 lp_build_mod(struct lp_build_context *bld, 383 LLVMValueRef x, 384 LLVMValueRef y); 385 386 LLVMValueRef 387 lp_build_isnan(struct lp_build_context *bld, 388 LLVMValueRef x); 389 390 LLVMValueRef 391 lp_build_isfinite(struct lp_build_context *bld, 392 LLVMValueRef x); 393 394 395 LLVMValueRef 396 lp_build_is_inf_or_nan(struct gallivm_state *gallivm, 397 const struct lp_type type, 398 LLVMValueRef x); 399 400 401 LLVMValueRef 402 lp_build_fpstate_get(struct gallivm_state *gallivm); 403 404 void 405 lp_build_fpstate_set_denorms_zero(struct gallivm_state *gallivm, 406 boolean zero); 407 void 408 lp_build_fpstate_set(struct gallivm_state *gallivm, 409 LLVMValueRef mxcsr); 410 411 #endif /* !LP_BLD_ARIT_H */ 412