| 1 | /// @ref core |
| 2 | /// @file glm/detail/func_common_simd.inl |
| 3 | |
| 4 | #if GLM_ARCH & GLM_ARCH_SSE2_BIT |
| 5 | |
| 6 | #include "../simd/common.h" |
| 7 | |
| 8 | #include <immintrin.h> |
| 9 | |
| 10 | namespace glm{ |
| 11 | namespace detail |
| 12 | { |
| 13 | template<qualifier Q> |
| 14 | struct compute_abs_vector<4, float, Q, true> |
| 15 | { |
| 16 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& v) |
| 17 | { |
| 18 | vec<4, float, Q> result; |
| 19 | result.data = glm_vec4_abs(v.data); |
| 20 | return result; |
| 21 | } |
| 22 | }; |
| 23 | |
| 24 | template<qualifier Q> |
| 25 | struct compute_abs_vector<4, int, Q, true> |
| 26 | { |
| 27 | GLM_FUNC_QUALIFIER static vec<4, int, Q> call(vec<4, int, Q> const& v) |
| 28 | { |
| 29 | vec<4, int, Q> result; |
| 30 | result.data = glm_ivec4_abs(v.data); |
| 31 | return result; |
| 32 | } |
| 33 | }; |
| 34 | |
| 35 | template<qualifier Q> |
| 36 | struct compute_floor<4, float, Q, true> |
| 37 | { |
| 38 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& v) |
| 39 | { |
| 40 | vec<4, float, Q> result; |
| 41 | result.data = glm_vec4_floor(v.data); |
| 42 | return result; |
| 43 | } |
| 44 | }; |
| 45 | |
| 46 | template<qualifier Q> |
| 47 | struct compute_ceil<4, float, Q, true> |
| 48 | { |
| 49 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& v) |
| 50 | { |
| 51 | vec<4, float, Q> result; |
| 52 | result.data = glm_vec4_ceil(v.data); |
| 53 | return result; |
| 54 | } |
| 55 | }; |
| 56 | |
| 57 | template<qualifier Q> |
| 58 | struct compute_fract<4, float, Q, true> |
| 59 | { |
| 60 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& v) |
| 61 | { |
| 62 | vec<4, float, Q> result; |
| 63 | result.data = glm_vec4_fract(v.data); |
| 64 | return result; |
| 65 | } |
| 66 | }; |
| 67 | |
| 68 | template<qualifier Q> |
| 69 | struct compute_round<4, float, Q, true> |
| 70 | { |
| 71 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& v) |
| 72 | { |
| 73 | vec<4, float, Q> result; |
| 74 | result.data = glm_vec4_round(v.data); |
| 75 | return result; |
| 76 | } |
| 77 | }; |
| 78 | |
| 79 | template<qualifier Q> |
| 80 | struct compute_mod<4, float, Q, true> |
| 81 | { |
| 82 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& x, vec<4, float, Q> const& y) |
| 83 | { |
| 84 | vec<4, float, Q> result; |
| 85 | result.data = glm_vec4_mod(x.data, y.data); |
| 86 | return result; |
| 87 | } |
| 88 | }; |
| 89 | |
| 90 | template<qualifier Q> |
| 91 | struct compute_min_vector<4, float, Q, true> |
| 92 | { |
| 93 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& v1, vec<4, float, Q> const& v2) |
| 94 | { |
| 95 | vec<4, float, Q> result; |
| 96 | result.data = _mm_min_ps(v1.data, v2.data); |
| 97 | return result; |
| 98 | } |
| 99 | }; |
| 100 | |
| 101 | template<qualifier Q> |
| 102 | struct compute_min_vector<4, int, Q, true> |
| 103 | { |
| 104 | GLM_FUNC_QUALIFIER static vec<4, int, Q> call(vec<4, int, Q> const& v1, vec<4, int, Q> const& v2) |
| 105 | { |
| 106 | vec<4, int, Q> result; |
| 107 | result.data = _mm_min_epi32(v1.data, v2.data); |
| 108 | return result; |
| 109 | } |
| 110 | }; |
| 111 | |
| 112 | template<qualifier Q> |
| 113 | struct compute_min_vector<4, uint, Q, true> |
| 114 | { |
| 115 | GLM_FUNC_QUALIFIER static vec<4, uint, Q> call(vec<4, uint, Q> const& v1, vec<4, uint, Q> const& v2) |
| 116 | { |
| 117 | vec<4, uint, Q> result; |
| 118 | result.data = _mm_min_epu32(v1.data, v2.data); |
| 119 | return result; |
| 120 | } |
| 121 | }; |
| 122 | |
| 123 | template<qualifier Q> |
| 124 | struct compute_max_vector<4, float, Q, true> |
| 125 | { |
| 126 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& v1, vec<4, float, Q> const& v2) |
| 127 | { |
| 128 | vec<4, float, Q> result; |
| 129 | result.data = _mm_max_ps(v1.data, v2.data); |
| 130 | return result; |
| 131 | } |
| 132 | }; |
| 133 | |
| 134 | template<qualifier Q> |
| 135 | struct compute_max_vector<4, int, Q, true> |
| 136 | { |
| 137 | GLM_FUNC_QUALIFIER static vec<4, int, Q> call(vec<4, int, Q> const& v1, vec<4, int, Q> const& v2) |
| 138 | { |
| 139 | vec<4, int, Q> result; |
| 140 | result.data = _mm_max_epi32(v1.data, v2.data); |
| 141 | return result; |
| 142 | } |
| 143 | }; |
| 144 | |
| 145 | template<qualifier Q> |
| 146 | struct compute_max_vector<4, uint, Q, true> |
| 147 | { |
| 148 | GLM_FUNC_QUALIFIER static vec<4, uint, Q> call(vec<4, uint, Q> const& v1, vec<4, uint, Q> const& v2) |
| 149 | { |
| 150 | vec<4, uint, Q> result; |
| 151 | result.data = _mm_max_epu32(v1.data, v2.data); |
| 152 | return result; |
| 153 | } |
| 154 | }; |
| 155 | |
| 156 | template<qualifier Q> |
| 157 | struct compute_clamp_vector<4, float, Q, true> |
| 158 | { |
| 159 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& x, vec<4, float, Q> const& minVal, vec<4, float, Q> const& maxVal) |
| 160 | { |
| 161 | vec<4, float, Q> result; |
| 162 | result.data = _mm_min_ps(_mm_max_ps(x.data, minVal.data), maxVal.data); |
| 163 | return result; |
| 164 | } |
| 165 | }; |
| 166 | |
| 167 | template<qualifier Q> |
| 168 | struct compute_clamp_vector<4, int, Q, true> |
| 169 | { |
| 170 | GLM_FUNC_QUALIFIER static vec<4, int, Q> call(vec<4, int, Q> const& x, vec<4, int, Q> const& minVal, vec<4, int, Q> const& maxVal) |
| 171 | { |
| 172 | vec<4, int, Q> result; |
| 173 | result.data = _mm_min_epi32(_mm_max_epi32(x.data, minVal.data), maxVal.data); |
| 174 | return result; |
| 175 | } |
| 176 | }; |
| 177 | |
| 178 | template<qualifier Q> |
| 179 | struct compute_clamp_vector<4, uint, Q, true> |
| 180 | { |
| 181 | GLM_FUNC_QUALIFIER static vec<4, uint, Q> call(vec<4, uint, Q> const& x, vec<4, uint, Q> const& minVal, vec<4, uint, Q> const& maxVal) |
| 182 | { |
| 183 | vec<4, uint, Q> result; |
| 184 | result.data = _mm_min_epu32(_mm_max_epu32(x.data, minVal.data), maxVal.data); |
| 185 | return result; |
| 186 | } |
| 187 | }; |
| 188 | |
| 189 | template<qualifier Q> |
| 190 | struct compute_mix_vector<4, float, bool, Q, true> |
| 191 | { |
| 192 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& x, vec<4, float, Q> const& y, vec<4, bool, Q> const& a) |
| 193 | { |
| 194 | __m128i const Load = _mm_set_epi32(-static_cast<int>(a.w), -static_cast<int>(a.z), -static_cast<int>(a.y), -static_cast<int>(a.x)); |
| 195 | __m128 const Mask = _mm_castsi128_ps(Load); |
| 196 | |
| 197 | vec<4, float, Q> Result; |
| 198 | # if 0 && GLM_ARCH & GLM_ARCH_AVX |
| 199 | Result.data = _mm_blendv_ps(x.data, y.data, Mask); |
| 200 | # else |
| 201 | Result.data = _mm_or_ps(_mm_and_ps(Mask, y.data), _mm_andnot_ps(Mask, x.data)); |
| 202 | # endif |
| 203 | return Result; |
| 204 | } |
| 205 | }; |
| 206 | /* FIXME |
| 207 | template<qualifier Q> |
| 208 | struct compute_step_vector<float, Q, tvec4> |
| 209 | { |
| 210 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& edge, vec<4, float, Q> const& x) |
| 211 | { |
| 212 | vec<4, float, Q> Result; |
| 213 | result.data = glm_vec4_step(edge.data, x.data); |
| 214 | return result; |
| 215 | } |
| 216 | }; |
| 217 | */ |
| 218 | template<qualifier Q> |
| 219 | struct compute_smoothstep_vector<4, float, Q, true> |
| 220 | { |
| 221 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& edge0, vec<4, float, Q> const& edge1, vec<4, float, Q> const& x) |
| 222 | { |
| 223 | vec<4, float, Q> Result; |
| 224 | Result.data = glm_vec4_smoothstep(edge0.data, edge1.data, x.data); |
| 225 | return Result; |
| 226 | } |
| 227 | }; |
| 228 | |
| 229 | template<qualifier Q> |
| 230 | struct compute_fma<4, float, Q, true> |
| 231 | { |
| 232 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<4, float, Q> const& a, vec<4, float, Q> const& b, vec<4, float, Q> const& c) |
| 233 | { |
| 234 | vec<4, float, Q> Result; |
| 235 | Result.data = glm_vec4_fma(a.data, b.data, c.data); |
| 236 | return Result; |
| 237 | } |
| 238 | }; |
| 239 | |
| 240 | template<qualifier Q> |
| 241 | struct compute_fma<3, float, Q, true> |
| 242 | { |
| 243 | GLM_FUNC_QUALIFIER static vec<3, float, Q> call(vec<3, float, Q> const& a, vec<3, float, Q> const& b, vec<3, float, Q> const& c) |
| 244 | { |
| 245 | vec<3, float, Q> Result; |
| 246 | Result.data = glm_vec4_fma(a.data, b.data, c.data); |
| 247 | return Result; |
| 248 | } |
| 249 | }; |
| 250 | |
| 251 | |
| 252 | template<qualifier Q> |
| 253 | struct compute_fma<4, double, Q, true> |
| 254 | { |
| 255 | GLM_FUNC_QUALIFIER static vec<4, double, Q> call(vec<4, double, Q> const& a, vec<4, double, Q> const& b, vec<4, double, Q> const& c) |
| 256 | { |
| 257 | vec<4, double, Q> Result; |
| 258 | # if (GLM_ARCH & GLM_ARCH_AVX2_BIT) && !(GLM_COMPILER & GLM_COMPILER_CLANG) |
| 259 | Result.data = _mm256_fmadd_pd(a.data, b.data, c.data); |
| 260 | # elif (GLM_ARCH & GLM_ARCH_AVX_BIT) |
| 261 | Result.data = _mm256_add_pd(_mm256_mul_pd(a.data, b.data), c.data); |
| 262 | # else |
| 263 | Result.data.setv(0, _mm_add_pd(_mm_mul_pd(a.data.getv(0), b.data.getv(0)), c.data.getv(0))); |
| 264 | Result.data.setv(1, _mm_add_pd(_mm_mul_pd(a.data.getv(1), b.data.getv(1)), c.data.getv(1))); |
| 265 | # endif |
| 266 | return Result; |
| 267 | } |
| 268 | }; |
| 269 | |
| 270 | // copy vec3 to vec4 and set w to 0 |
| 271 | template<qualifier Q> |
| 272 | struct convert_vec3_to_vec4W0<float, Q, true> |
| 273 | { |
| 274 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<3, float, Q> const& a) |
| 275 | { |
| 276 | vec<4, float, Q> v; |
| 277 | #if (GLM_ARCH & GLM_ARCH_SSE41_BIT) |
| 278 | v.data = _mm_blend_ps(a.data, _mm_setzero_ps(), 8); |
| 279 | #else |
| 280 | __m128i mask = _mm_set_epi32(0, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF); |
| 281 | __m128 v0 = _mm_castsi128_ps(_mm_and_si128(_mm_castps_si128(a.data), mask)); |
| 282 | v.data = v0; |
| 283 | #endif |
| 284 | return v; |
| 285 | } |
| 286 | }; |
| 287 | |
| 288 | // copy vec3 to vec4 and set w to 1 |
| 289 | template<qualifier Q> |
| 290 | struct convert_vec3_to_vec4W1<float, Q, true> |
| 291 | { |
| 292 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<3, float, Q> const& a) |
| 293 | { |
| 294 | vec<4, float, Q> v; |
| 295 | #if (GLM_ARCH & GLM_ARCH_SSE41_BIT) |
| 296 | v.data = _mm_blend_ps(a.data, _mm_set1_ps(1.0f), 8); |
| 297 | #else |
| 298 | __m128 t1 = _mm_shuffle_ps(a.data, a.data, _MM_SHUFFLE(0, 2, 1, 3)); //permute x, w |
| 299 | __m128 t2 = _mm_move_ss(t1, _mm_set_ss(1.0f)); // set x to 1.0f |
| 300 | v.data = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(0, 2, 1, 3)); //permute x, w |
| 301 | #endif |
| 302 | return v; |
| 303 | } |
| 304 | }; |
| 305 | |
| 306 | // copy vec3 to vec4 and set w to vec3.z |
| 307 | template<qualifier Q> |
| 308 | struct convert_vec3_to_vec4WZ<float, Q, true> |
| 309 | { |
| 310 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<3, float, Q> const& a) |
| 311 | { |
| 312 | vec<4, float, Q> v; |
| 313 | v.data = _mm_shuffle_ps(a.data, a.data, _MM_SHUFFLE(2, 2, 1, 0)); |
| 314 | return v; |
| 315 | } |
| 316 | }; |
| 317 | |
| 318 | // copy vec3 to vec4 and set w to 0 |
| 319 | template<qualifier Q> |
| 320 | struct convert_vec3_to_vec4W0<double, Q, true> |
| 321 | { |
| 322 | GLM_FUNC_QUALIFIER static vec<4, double, Q> call(vec<3, double, Q> const& a) |
| 323 | { |
| 324 | vec<4, double, Q> v; |
| 325 | #if (GLM_ARCH & GLM_ARCH_AVX_BIT) |
| 326 | v.data = _mm256_blend_pd(a.data, _mm256_setzero_pd(), 8); |
| 327 | #else |
| 328 | v.data.setv(0, a.data.getv(0)); |
| 329 | glm_dvec2 av2 = a.data.getv(1); |
| 330 | av2 = _mm_shuffle_pd(av2, _mm_setzero_pd(), 2); |
| 331 | v.data.setv(1, av2); |
| 332 | #endif |
| 333 | return v; |
| 334 | } |
| 335 | }; |
| 336 | |
| 337 | // copy vec3 to vec4 and set w to vec3.z |
| 338 | template<qualifier Q> |
| 339 | struct convert_vec3_to_vec4WZ<double, Q, true> |
| 340 | { |
| 341 | GLM_FUNC_QUALIFIER static vec<4, double, Q> call(vec<3, double, Q> const& a) |
| 342 | { |
| 343 | vec<4, double, Q> v; |
| 344 | #if (GLM_ARCH & GLM_ARCH_AVX_BIT) |
| 345 | v.data = _mm256_permute_pd(a.data, 2); |
| 346 | #else |
| 347 | v.data.setv(0, a.data.getv(0)); |
| 348 | glm_dvec2 av2 = a.data.getv(1); |
| 349 | __m128d t1 = _mm_shuffle_pd(av2, av2, 0); |
| 350 | v.data.setv(1, t1); |
| 351 | #endif |
| 352 | return v; |
| 353 | } |
| 354 | }; |
| 355 | |
| 356 | // copy vec3 to vec4 and set w to 1 |
| 357 | template<qualifier Q> |
| 358 | struct convert_vec3_to_vec4W1<double, Q, true> |
| 359 | { |
| 360 | GLM_FUNC_QUALIFIER static vec<4, double, Q> call(vec<3, double, Q> const& a) |
| 361 | { |
| 362 | vec<4, double, Q> v; |
| 363 | #if (GLM_ARCH & GLM_ARCH_AVX_BIT) |
| 364 | v.data = _mm256_blend_pd(a.data, _mm256_set1_pd(1.0), 8); |
| 365 | #else |
| 366 | v.data.setv(0, a.data.getv(0)); |
| 367 | glm_dvec2 av2 = a.data.getv(1); |
| 368 | av2 = _mm_shuffle_pd(av2, _mm_set1_pd(1.), 2); |
| 369 | v.data.setv(1, av2); |
| 370 | #endif |
| 371 | return v; |
| 372 | } |
| 373 | }; |
| 374 | |
| 375 | template<qualifier Q> |
| 376 | struct convert_vec4_to_vec3<float, Q, true> { |
| 377 | GLM_FUNC_QUALIFIER static vec<3, float, Q> call(vec<4, float, Q> const& a) |
| 378 | { |
| 379 | vec<3, float, Q> v; |
| 380 | v.data = a.data; |
| 381 | return v; |
| 382 | } |
| 383 | }; |
| 384 | |
| 385 | template<qualifier Q> |
| 386 | struct convert_vec4_to_vec3<double, Q, true> { |
| 387 | GLM_FUNC_QUALIFIER static vec<3, double, Q> call(vec<4, double, Q> const& a) |
| 388 | { |
| 389 | vec<3, double, Q> v; |
| 390 | #if GLM_ARCH & GLM_ARCH_AVX_BIT |
| 391 | v.data = a.data; |
| 392 | #else |
| 393 | v.data.setv(0, a.data.getv(0)); |
| 394 | v.data.setv(1, a.data.getv(1)); |
| 395 | #endif |
| 396 | return v; |
| 397 | } |
| 398 | }; |
| 399 | |
| 400 | |
| 401 | // set all coordinates to same value vec[c] |
| 402 | template<length_t L, qualifier Q> |
| 403 | struct convert_splat<L, float, Q, true> { |
| 404 | template<int c> |
| 405 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, float, Q> call(vec<L, float, Q> const& a) |
| 406 | { |
| 407 | vec<L, float, Q> Result; |
| 408 | const int s = _MM_SHUFFLE(c, c, c, c); |
| 409 | glm_f32vec4 va = static_cast<glm_f32vec4>(a.data); |
| 410 | # if GLM_ARCH & GLM_ARCH_AVX_BIT |
| 411 | Result.data = _mm_permute_ps(va, s); |
| 412 | # else |
| 413 | Result.data = _mm_shuffle_ps(va, va, s); |
| 414 | # endif |
| 415 | return Result; |
| 416 | } |
| 417 | }; |
| 418 | |
| 419 | // set all coordinates to same value vec[c] |
| 420 | template<length_t L, qualifier Q> |
| 421 | struct convert_splat<L, double, Q, true> { |
| 422 | |
| 423 | template<bool, int c> |
| 424 | struct detailSSE |
| 425 | {}; |
| 426 | |
| 427 | template<int c> |
| 428 | struct detailSSE<true, c> |
| 429 | { |
| 430 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, double, Q> call(vec<L, double, Q> const& a) |
| 431 | { |
| 432 | vec<L, double, Q> Result; |
| 433 | glm_f64vec2 r0 = _mm_shuffle_pd(a.data.getv(0), a.data.getv(0), c | c << 1); |
| 434 | Result.data.setv(0, r0); |
| 435 | Result.data.setv(1, r0); |
| 436 | return Result; |
| 437 | } |
| 438 | }; |
| 439 | |
| 440 | template<int c> |
| 441 | struct detailSSE<false, c> |
| 442 | { |
| 443 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, double, Q> call(vec<L, double, Q> const& a) |
| 444 | { |
| 445 | vec<L, double, Q> Result; |
| 446 | const unsigned int d = static_cast<unsigned int>(c - 2); |
| 447 | glm_f64vec2 r0 = _mm_shuffle_pd(a.data.getv(1), a.data.getv(1), d | d << 1); |
| 448 | Result.data.setv(0, r0); |
| 449 | Result.data.setv(1, r0); |
| 450 | return Result; |
| 451 | } |
| 452 | }; |
| 453 | |
| 454 | #if GLM_ARCH & GLM_ARCH_AVX_BIT |
| 455 | template<bool, int c> //note: bool is useless but needed to compil on linux (gcc) |
| 456 | struct detailAVX |
| 457 | {}; |
| 458 | |
| 459 | template<bool b> |
| 460 | struct detailAVX<b, 0> |
| 461 | { |
| 462 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, double, Q> call(vec<L, double, Q> const& a) |
| 463 | { |
| 464 | vec<L, double, Q> Result; |
| 465 | __m256d t1 = _mm256_permute2f128_pd(a.data, a.data, 0x0); |
| 466 | Result.data = _mm256_permute_pd(t1, 0); |
| 467 | return Result; |
| 468 | } |
| 469 | }; |
| 470 | |
| 471 | template<bool b> |
| 472 | struct detailAVX<b, 1> |
| 473 | { |
| 474 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, double, Q> call(vec<L, double, Q> const& a) |
| 475 | { |
| 476 | vec<L, double, Q> Result; |
| 477 | __m256d t1 = _mm256_permute2f128_pd(a.data, a.data, 0x0); |
| 478 | Result.data = _mm256_permute_pd(t1, 0xf); |
| 479 | return Result; |
| 480 | } |
| 481 | }; |
| 482 | |
| 483 | template<bool b> |
| 484 | struct detailAVX<b, 2> |
| 485 | { |
| 486 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, double, Q> call(vec<L, double, Q> const& a) |
| 487 | { |
| 488 | vec<L, double, Q> Result; |
| 489 | __m256d t2 = _mm256_permute2f128_pd(a.data, a.data, 0x11); |
| 490 | Result.data = _mm256_permute_pd(t2, 0x0); |
| 491 | return Result; |
| 492 | } |
| 493 | }; |
| 494 | |
| 495 | template<bool b> |
| 496 | struct detailAVX<b, 3> |
| 497 | { |
| 498 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, double, Q> call(vec<L, double, Q> const& a) |
| 499 | { |
| 500 | vec<L, double, Q> Result; |
| 501 | __m256d t2 = _mm256_permute2f128_pd(a.data, a.data, 0x11); |
| 502 | Result.data = _mm256_permute_pd(t2, 0xf); |
| 503 | return Result; |
| 504 | } |
| 505 | }; |
| 506 | #endif //GLM_ARCH & GLM_ARCH_AVX_BIT |
| 507 | |
| 508 | template<int c> |
| 509 | GLM_FUNC_QUALIFIER GLM_CONSTEXPR static vec<L, double, Q> call(vec<L, double, Q> const& a) |
| 510 | { |
| 511 | //return compute_splat<L, double, Q, false>::call<c>(a); |
| 512 | vec<L, double, Q> Result; |
| 513 | # if GLM_ARCH & GLM_ARCH_AVX2_BIT |
| 514 | Result.data = _mm256_permute4x64_pd(a.data, _MM_SHUFFLE(c, c, c, c)); |
| 515 | # elif GLM_ARCH & GLM_ARCH_AVX_BIT |
| 516 | Result = detailAVX<true, c>::call(a); |
| 517 | # else |
| 518 | #if 1 //detail<(c <= 1), c>::call2(a) is equivalent to following code but without if constexpr usage |
| 519 | Result = detailSSE<(c <= 1), c>::call(a); |
| 520 | #else |
| 521 | if constexpr (c <= 1) |
| 522 | { |
| 523 | glm_f64vec2 r0 = _mm_shuffle_pd(a.data.getv(0), a.data.getv(0), c | c << 1); |
| 524 | Result.data.setv(0, r0); |
| 525 | Result.data.setv(1, r0); |
| 526 | } |
| 527 | else |
| 528 | { |
| 529 | const unsigned int d = (unsigned int)(c - 2); |
| 530 | glm_f64vec2 r0 = _mm_shuffle_pd(a.data.getv(1), a.data.getv(1), d | d << 1); |
| 531 | Result.data.setv(0, r0); |
| 532 | Result.data.setv(1, r0); |
| 533 | } |
| 534 | #endif |
| 535 | # endif |
| 536 | return Result; |
| 537 | } |
| 538 | }; |
| 539 | |
| 540 | |
| 541 | }//namespace detail |
| 542 | }//namespace glm |
| 543 | |
| 544 | #endif//GLM_ARCH & GLM_ARCH_SSE2_BIT |
| 545 | |
| 546 | #if GLM_ARCH & GLM_ARCH_NEON_BIT |
| 547 | namespace glm { |
| 548 | namespace detail { |
| 549 | /* FIXME: Breaking Windows ARM Github C.I. |
| 550 | template<qualifier Q> |
| 551 | struct convert_vec3_to_vec4W0<float, Q, true> |
| 552 | { |
| 553 | GLM_FUNC_QUALIFIER static vec<4, float, Q> call(vec<3, float, Q> const& a) |
| 554 | { |
| 555 | vec<4, float, Q> v; |
| 556 | static const uint32x4_t mask = { 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0 }; |
| 557 | v.data = vbslq_f32(mask, a.data, vdupq_n_f32(0)); |
| 558 | return v; |
| 559 | } |
| 560 | }; |
| 561 | */ |
| 562 | template<qualifier Q> |
| 563 | struct convert_vec4_to_vec3<float, Q, true> { |
| 564 | GLM_FUNC_QUALIFIER static vec<3, float, Q> call(vec<4, float, Q> const& a) |
| 565 | { |
| 566 | vec<3, float, Q> v; |
| 567 | v.data = a.data; |
| 568 | return v; |
| 569 | } |
| 570 | }; |
| 571 | |
| 572 | /* compute_splat is never called? |
| 573 | template<length_t L, qualifier Q> |
| 574 | struct compute_splat<L, float, Q, true> { |
| 575 | template<int c> |
| 576 | GLM_FUNC_QUALIFIER static vec<L, float, Q> call(vec<L, float, Q> const& a) |
| 577 | { |
| 578 | (void)a; |
| 579 | } |
| 580 | |
| 581 | template<> |
| 582 | GLM_FUNC_QUALIFIER static vec<L, float, Q> call<0>(vec<L, float, Q> const& a) |
| 583 | { |
| 584 | vec<L, float, Q> Result; |
| 585 | Result.data = vdupq_lane_f32(vget_low_f32(a.data), 0); |
| 586 | return Result; |
| 587 | } |
| 588 | |
| 589 | template<> |
| 590 | GLM_FUNC_QUALIFIER static vec<L, float, Q> call<1>(vec<L, float, Q> const& a) |
| 591 | { |
| 592 | vec<L, float, Q> Result; |
| 593 | Result.data = vdupq_lane_f32(vget_low_f32(a.data), 1); |
| 594 | return Result; |
| 595 | } |
| 596 | |
| 597 | template<> |
| 598 | GLM_FUNC_QUALIFIER static vec<L, float, Q> call<2>(vec<L, float, Q> const& a) |
| 599 | { |
| 600 | vec<L, float, Q> Result; |
| 601 | Result.data = vdupq_lane_f32(vget_high_f32(a.data), 0); |
| 602 | return Result; |
| 603 | } |
| 604 | |
| 605 | template<> |
| 606 | GLM_FUNC_QUALIFIER static vec<L, float, Q> call<3>(vec<L, float, Q> const& a) |
| 607 | { |
| 608 | vec<L, float, Q> Result; |
| 609 | Result.data = vdupq_lane_f32(vget_high_f32(a.data), 1); |
| 610 | return Result; |
| 611 | } |
| 612 | }; |
| 613 | */ |
| 614 | }//namespace detail |
| 615 | }//namespace glm |
| 616 | #endif //GLM_ARCH & GLM_ARCH_NEON_BIT |