u_gen_mipmap.c revision 683e7091a953204c9aee1410ac44be3b69bae9fc
1/************************************************************************** 2 * 3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. 4 * All Rights Reserved. 5 * Copyright 2008 VMware, Inc. All rights reserved. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the 9 * "Software"), to deal in the Software without restriction, including 10 * without limitation the rights to use, copy, modify, merge, publish, 11 * distribute, sub license, and/or sell copies of the Software, and to 12 * permit persons to whom the Software is furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice and this permission notice (including the 16 * next paragraph) shall be included in all copies or substantial portions 17 * of the Software. 18 * 19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 * 27 **************************************************************************/ 28 29/** 30 * @file 31 * Mipmap generation utility 32 * 33 * @author Brian Paul 34 */ 35 36 37#include "pipe/p_context.h" 38#include "util/u_debug.h" 39#include "pipe/p_defines.h" 40#include "pipe/p_inlines.h" 41#include "pipe/p_shader_tokens.h" 42#include "pipe/p_state.h" 43 44#include "util/u_memory.h" 45#include "util/u_draw_quad.h" 46#include "util/u_gen_mipmap.h" 47#include "util/u_simple_shaders.h" 48 49#include "tgsi/tgsi_build.h" 50#include "tgsi/tgsi_dump.h" 51#include "tgsi/tgsi_parse.h" 52 53#include "cso_cache/cso_context.h" 54 55 56struct gen_mipmap_state 57{ 58 struct pipe_context *pipe; 59 struct cso_context *cso; 60 61 struct pipe_blend_state blend; 62 struct pipe_depth_stencil_alpha_state depthstencil; 63 struct pipe_rasterizer_state rasterizer; 64 struct pipe_sampler_state sampler; 65 66 void *vs; 67 void *fs; 68 69 struct pipe_buffer *vbuf; /**< quad vertices */ 70 unsigned vbuf_slot; 71 72 float vertices[4][2][4]; /**< vertex/texcoords for quad */ 73}; 74 75 76 77enum dtype 78{ 79 UBYTE, 80 UBYTE_3_3_2, 81 USHORT, 82 USHORT_4_4_4_4, 83 USHORT_5_6_5, 84 USHORT_1_5_5_5_REV, 85 UINT, 86 FLOAT, 87 HALF_FLOAT 88}; 89 90 91typedef ushort half_float; 92 93 94static half_float 95float_to_half(float f) 96{ 97 /* XXX fix this */ 98 return 0; 99} 100 101static float 102half_to_float(half_float h) 103{ 104 /* XXX fix this */ 105 return 0.0f; 106} 107 108 109 110 111/** 112 * \name Support macros for do_row and do_row_3d 113 * 114 * The macro madness is here for two reasons. First, it compacts the code 115 * slightly. Second, it makes it much easier to adjust the specifics of the 116 * filter to tune the rounding characteristics. 117 */ 118/*@{*/ 119#define DECLARE_ROW_POINTERS(t, e) \ 120 const t(*rowA)[e] = (const t(*)[e]) srcRowA; \ 121 const t(*rowB)[e] = (const t(*)[e]) srcRowB; \ 122 const t(*rowC)[e] = (const t(*)[e]) srcRowC; \ 123 const t(*rowD)[e] = (const t(*)[e]) srcRowD; \ 124 t(*dst)[e] = (t(*)[e]) dstRow 125 126#define DECLARE_ROW_POINTERS0(t) \ 127 const t *rowA = (const t *) srcRowA; \ 128 const t *rowB = (const t *) srcRowB; \ 129 const t *rowC = (const t *) srcRowC; \ 130 const t *rowD = (const t *) srcRowD; \ 131 t *dst = (t *) dstRow 132 133#define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \ 134 ((unsigned) Aj + (unsigned) Ak \ 135 + (unsigned) Bj + (unsigned) Bk \ 136 + (unsigned) Cj + (unsigned) Ck \ 137 + (unsigned) Dj + (unsigned) Dk \ 138 + 4) >> 3 139 140#define FILTER_3D(e) \ 141 do { \ 142 dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \ 143 rowB[j][e], rowB[k][e], \ 144 rowC[j][e], rowC[k][e], \ 145 rowD[j][e], rowD[k][e]); \ 146 } while(0) 147 148#define FILTER_F_3D(e) \ 149 do { \ 150 dst[i][e] = (rowA[j][e] + rowA[k][e] \ 151 + rowB[j][e] + rowB[k][e] \ 152 + rowC[j][e] + rowC[k][e] \ 153 + rowD[j][e] + rowD[k][e]) * 0.125F; \ 154 } while(0) 155 156#define FILTER_HF_3D(e) \ 157 do { \ 158 const float aj = half_to_float(rowA[j][e]); \ 159 const float ak = half_to_float(rowA[k][e]); \ 160 const float bj = half_to_float(rowB[j][e]); \ 161 const float bk = half_to_float(rowB[k][e]); \ 162 const float cj = half_to_float(rowC[j][e]); \ 163 const float ck = half_to_float(rowC[k][e]); \ 164 const float dj = half_to_float(rowD[j][e]); \ 165 const float dk = half_to_float(rowD[k][e]); \ 166 dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \ 167 * 0.125F); \ 168 } while(0) 169/*@}*/ 170 171 172/** 173 * Average together two rows of a source image to produce a single new 174 * row in the dest image. It's legal for the two source rows to point 175 * to the same data. The source width must be equal to either the 176 * dest width or two times the dest width. 177 * \param datatype GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc. 178 * \param comps number of components per pixel (1..4) 179 */ 180static void 181do_row(enum dtype datatype, uint comps, int srcWidth, 182 const void *srcRowA, const void *srcRowB, 183 int dstWidth, void *dstRow) 184{ 185 const uint k0 = (srcWidth == dstWidth) ? 0 : 1; 186 const uint colStride = (srcWidth == dstWidth) ? 1 : 2; 187 188 assert(comps >= 1); 189 assert(comps <= 4); 190 191 /* This assertion is no longer valid with non-power-of-2 textures 192 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth); 193 */ 194 195 if (datatype == UBYTE && comps == 4) { 196 uint i, j, k; 197 const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA; 198 const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB; 199 ubyte(*dst)[4] = (ubyte(*)[4]) dstRow; 200 for (i = j = 0, k = k0; i < (uint) dstWidth; 201 i++, j += colStride, k += colStride) { 202 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 203 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 204 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 205 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 206 } 207 } 208 else if (datatype == UBYTE && comps == 3) { 209 uint i, j, k; 210 const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA; 211 const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB; 212 ubyte(*dst)[3] = (ubyte(*)[3]) dstRow; 213 for (i = j = 0, k = k0; i < (uint) dstWidth; 214 i++, j += colStride, k += colStride) { 215 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 216 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 217 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 218 } 219 } 220 else if (datatype == UBYTE && comps == 2) { 221 uint i, j, k; 222 const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA; 223 const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB; 224 ubyte(*dst)[2] = (ubyte(*)[2]) dstRow; 225 for (i = j = 0, k = k0; i < (uint) dstWidth; 226 i++, j += colStride, k += colStride) { 227 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2; 228 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2; 229 } 230 } 231 else if (datatype == UBYTE && comps == 1) { 232 uint i, j, k; 233 const ubyte *rowA = (const ubyte *) srcRowA; 234 const ubyte *rowB = (const ubyte *) srcRowB; 235 ubyte *dst = (ubyte *) dstRow; 236 for (i = j = 0, k = k0; i < (uint) dstWidth; 237 i++, j += colStride, k += colStride) { 238 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2; 239 } 240 } 241 242 else if (datatype == USHORT && comps == 4) { 243 uint i, j, k; 244 const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA; 245 const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB; 246 ushort(*dst)[4] = (ushort(*)[4]) dstRow; 247 for (i = j = 0, k = k0; i < (uint) dstWidth; 248 i++, j += colStride, k += colStride) { 249 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 250 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 251 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 252 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 253 } 254 } 255 else if (datatype == USHORT && comps == 3) { 256 uint i, j, k; 257 const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA; 258 const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB; 259 ushort(*dst)[3] = (ushort(*)[3]) dstRow; 260 for (i = j = 0, k = k0; i < (uint) dstWidth; 261 i++, j += colStride, k += colStride) { 262 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 263 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 264 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 265 } 266 } 267 else if (datatype == USHORT && comps == 2) { 268 uint i, j, k; 269 const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA; 270 const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB; 271 ushort(*dst)[2] = (ushort(*)[2]) dstRow; 272 for (i = j = 0, k = k0; i < (uint) dstWidth; 273 i++, j += colStride, k += colStride) { 274 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 275 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 276 } 277 } 278 else if (datatype == USHORT && comps == 1) { 279 uint i, j, k; 280 const ushort *rowA = (const ushort *) srcRowA; 281 const ushort *rowB = (const ushort *) srcRowB; 282 ushort *dst = (ushort *) dstRow; 283 for (i = j = 0, k = k0; i < (uint) dstWidth; 284 i++, j += colStride, k += colStride) { 285 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4; 286 } 287 } 288 289 else if (datatype == FLOAT && comps == 4) { 290 uint i, j, k; 291 const float(*rowA)[4] = (const float(*)[4]) srcRowA; 292 const float(*rowB)[4] = (const float(*)[4]) srcRowB; 293 float(*dst)[4] = (float(*)[4]) dstRow; 294 for (i = j = 0, k = k0; i < (uint) dstWidth; 295 i++, j += colStride, k += colStride) { 296 dst[i][0] = (rowA[j][0] + rowA[k][0] + 297 rowB[j][0] + rowB[k][0]) * 0.25F; 298 dst[i][1] = (rowA[j][1] + rowA[k][1] + 299 rowB[j][1] + rowB[k][1]) * 0.25F; 300 dst[i][2] = (rowA[j][2] + rowA[k][2] + 301 rowB[j][2] + rowB[k][2]) * 0.25F; 302 dst[i][3] = (rowA[j][3] + rowA[k][3] + 303 rowB[j][3] + rowB[k][3]) * 0.25F; 304 } 305 } 306 else if (datatype == FLOAT && comps == 3) { 307 uint i, j, k; 308 const float(*rowA)[3] = (const float(*)[3]) srcRowA; 309 const float(*rowB)[3] = (const float(*)[3]) srcRowB; 310 float(*dst)[3] = (float(*)[3]) dstRow; 311 for (i = j = 0, k = k0; i < (uint) dstWidth; 312 i++, j += colStride, k += colStride) { 313 dst[i][0] = (rowA[j][0] + rowA[k][0] + 314 rowB[j][0] + rowB[k][0]) * 0.25F; 315 dst[i][1] = (rowA[j][1] + rowA[k][1] + 316 rowB[j][1] + rowB[k][1]) * 0.25F; 317 dst[i][2] = (rowA[j][2] + rowA[k][2] + 318 rowB[j][2] + rowB[k][2]) * 0.25F; 319 } 320 } 321 else if (datatype == FLOAT && comps == 2) { 322 uint i, j, k; 323 const float(*rowA)[2] = (const float(*)[2]) srcRowA; 324 const float(*rowB)[2] = (const float(*)[2]) srcRowB; 325 float(*dst)[2] = (float(*)[2]) dstRow; 326 for (i = j = 0, k = k0; i < (uint) dstWidth; 327 i++, j += colStride, k += colStride) { 328 dst[i][0] = (rowA[j][0] + rowA[k][0] + 329 rowB[j][0] + rowB[k][0]) * 0.25F; 330 dst[i][1] = (rowA[j][1] + rowA[k][1] + 331 rowB[j][1] + rowB[k][1]) * 0.25F; 332 } 333 } 334 else if (datatype == FLOAT && comps == 1) { 335 uint i, j, k; 336 const float *rowA = (const float *) srcRowA; 337 const float *rowB = (const float *) srcRowB; 338 float *dst = (float *) dstRow; 339 for (i = j = 0, k = k0; i < (uint) dstWidth; 340 i++, j += colStride, k += colStride) { 341 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F; 342 } 343 } 344 345#if 0 346 else if (datatype == HALF_FLOAT && comps == 4) { 347 uint i, j, k, comp; 348 const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA; 349 const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB; 350 half_float(*dst)[4] = (half_float(*)[4]) dstRow; 351 for (i = j = 0, k = k0; i < (uint) dstWidth; 352 i++, j += colStride, k += colStride) { 353 for (comp = 0; comp < 4; comp++) { 354 float aj, ak, bj, bk; 355 aj = half_to_float(rowA[j][comp]); 356 ak = half_to_float(rowA[k][comp]); 357 bj = half_to_float(rowB[j][comp]); 358 bk = half_to_float(rowB[k][comp]); 359 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 360 } 361 } 362 } 363 else if (datatype == HALF_FLOAT && comps == 3) { 364 uint i, j, k, comp; 365 const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA; 366 const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB; 367 half_float(*dst)[3] = (half_float(*)[3]) dstRow; 368 for (i = j = 0, k = k0; i < (uint) dstWidth; 369 i++, j += colStride, k += colStride) { 370 for (comp = 0; comp < 3; comp++) { 371 float aj, ak, bj, bk; 372 aj = half_to_float(rowA[j][comp]); 373 ak = half_to_float(rowA[k][comp]); 374 bj = half_to_float(rowB[j][comp]); 375 bk = half_to_float(rowB[k][comp]); 376 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 377 } 378 } 379 } 380 else if (datatype == HALF_FLOAT && comps == 2) { 381 uint i, j, k, comp; 382 const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA; 383 const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB; 384 half_float(*dst)[2] = (half_float(*)[2]) dstRow; 385 for (i = j = 0, k = k0; i < (uint) dstWidth; 386 i++, j += colStride, k += colStride) { 387 for (comp = 0; comp < 2; comp++) { 388 float aj, ak, bj, bk; 389 aj = half_to_float(rowA[j][comp]); 390 ak = half_to_float(rowA[k][comp]); 391 bj = half_to_float(rowB[j][comp]); 392 bk = half_to_float(rowB[k][comp]); 393 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 394 } 395 } 396 } 397 else if (datatype == HALF_FLOAT && comps == 1) { 398 uint i, j, k; 399 const half_float *rowA = (const half_float *) srcRowA; 400 const half_float *rowB = (const half_float *) srcRowB; 401 half_float *dst = (half_float *) dstRow; 402 for (i = j = 0, k = k0; i < (uint) dstWidth; 403 i++, j += colStride, k += colStride) { 404 float aj, ak, bj, bk; 405 aj = half_to_float(rowA[j]); 406 ak = half_to_float(rowA[k]); 407 bj = half_to_float(rowB[j]); 408 bk = half_to_float(rowB[k]); 409 dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F); 410 } 411 } 412#endif 413 414 else if (datatype == UINT && comps == 1) { 415 uint i, j, k; 416 const uint *rowA = (const uint *) srcRowA; 417 const uint *rowB = (const uint *) srcRowB; 418 uint *dst = (uint *) dstRow; 419 for (i = j = 0, k = k0; i < (uint) dstWidth; 420 i++, j += colStride, k += colStride) { 421 dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4; 422 } 423 } 424 425 else if (datatype == USHORT_5_6_5 && comps == 3) { 426 uint i, j, k; 427 const ushort *rowA = (const ushort *) srcRowA; 428 const ushort *rowB = (const ushort *) srcRowB; 429 ushort *dst = (ushort *) dstRow; 430 for (i = j = 0, k = k0; i < (uint) dstWidth; 431 i++, j += colStride, k += colStride) { 432 const int rowAr0 = rowA[j] & 0x1f; 433 const int rowAr1 = rowA[k] & 0x1f; 434 const int rowBr0 = rowB[j] & 0x1f; 435 const int rowBr1 = rowB[k] & 0x1f; 436 const int rowAg0 = (rowA[j] >> 5) & 0x3f; 437 const int rowAg1 = (rowA[k] >> 5) & 0x3f; 438 const int rowBg0 = (rowB[j] >> 5) & 0x3f; 439 const int rowBg1 = (rowB[k] >> 5) & 0x3f; 440 const int rowAb0 = (rowA[j] >> 11) & 0x1f; 441 const int rowAb1 = (rowA[k] >> 11) & 0x1f; 442 const int rowBb0 = (rowB[j] >> 11) & 0x1f; 443 const int rowBb1 = (rowB[k] >> 11) & 0x1f; 444 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 445 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 446 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 447 dst[i] = (blue << 11) | (green << 5) | red; 448 } 449 } 450 else if (datatype == USHORT_4_4_4_4 && comps == 4) { 451 uint i, j, k; 452 const ushort *rowA = (const ushort *) srcRowA; 453 const ushort *rowB = (const ushort *) srcRowB; 454 ushort *dst = (ushort *) dstRow; 455 for (i = j = 0, k = k0; i < (uint) dstWidth; 456 i++, j += colStride, k += colStride) { 457 const int rowAr0 = rowA[j] & 0xf; 458 const int rowAr1 = rowA[k] & 0xf; 459 const int rowBr0 = rowB[j] & 0xf; 460 const int rowBr1 = rowB[k] & 0xf; 461 const int rowAg0 = (rowA[j] >> 4) & 0xf; 462 const int rowAg1 = (rowA[k] >> 4) & 0xf; 463 const int rowBg0 = (rowB[j] >> 4) & 0xf; 464 const int rowBg1 = (rowB[k] >> 4) & 0xf; 465 const int rowAb0 = (rowA[j] >> 8) & 0xf; 466 const int rowAb1 = (rowA[k] >> 8) & 0xf; 467 const int rowBb0 = (rowB[j] >> 8) & 0xf; 468 const int rowBb1 = (rowB[k] >> 8) & 0xf; 469 const int rowAa0 = (rowA[j] >> 12) & 0xf; 470 const int rowAa1 = (rowA[k] >> 12) & 0xf; 471 const int rowBa0 = (rowB[j] >> 12) & 0xf; 472 const int rowBa1 = (rowB[k] >> 12) & 0xf; 473 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 474 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 475 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 476 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 477 dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red; 478 } 479 } 480 else if (datatype == USHORT_1_5_5_5_REV && comps == 4) { 481 uint i, j, k; 482 const ushort *rowA = (const ushort *) srcRowA; 483 const ushort *rowB = (const ushort *) srcRowB; 484 ushort *dst = (ushort *) dstRow; 485 for (i = j = 0, k = k0; i < (uint) dstWidth; 486 i++, j += colStride, k += colStride) { 487 const int rowAr0 = rowA[j] & 0x1f; 488 const int rowAr1 = rowA[k] & 0x1f; 489 const int rowBr0 = rowB[j] & 0x1f; 490 const int rowBr1 = rowB[k] & 0x1f; 491 const int rowAg0 = (rowA[j] >> 5) & 0x1f; 492 const int rowAg1 = (rowA[k] >> 5) & 0x1f; 493 const int rowBg0 = (rowB[j] >> 5) & 0x1f; 494 const int rowBg1 = (rowB[k] >> 5) & 0x1f; 495 const int rowAb0 = (rowA[j] >> 10) & 0x1f; 496 const int rowAb1 = (rowA[k] >> 10) & 0x1f; 497 const int rowBb0 = (rowB[j] >> 10) & 0x1f; 498 const int rowBb1 = (rowB[k] >> 10) & 0x1f; 499 const int rowAa0 = (rowA[j] >> 15) & 0x1; 500 const int rowAa1 = (rowA[k] >> 15) & 0x1; 501 const int rowBa0 = (rowB[j] >> 15) & 0x1; 502 const int rowBa1 = (rowB[k] >> 15) & 0x1; 503 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 504 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 505 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 506 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 507 dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red; 508 } 509 } 510 else if (datatype == UBYTE_3_3_2 && comps == 3) { 511 uint i, j, k; 512 const ubyte *rowA = (const ubyte *) srcRowA; 513 const ubyte *rowB = (const ubyte *) srcRowB; 514 ubyte *dst = (ubyte *) dstRow; 515 for (i = j = 0, k = k0; i < (uint) dstWidth; 516 i++, j += colStride, k += colStride) { 517 const int rowAr0 = rowA[j] & 0x3; 518 const int rowAr1 = rowA[k] & 0x3; 519 const int rowBr0 = rowB[j] & 0x3; 520 const int rowBr1 = rowB[k] & 0x3; 521 const int rowAg0 = (rowA[j] >> 2) & 0x7; 522 const int rowAg1 = (rowA[k] >> 2) & 0x7; 523 const int rowBg0 = (rowB[j] >> 2) & 0x7; 524 const int rowBg1 = (rowB[k] >> 2) & 0x7; 525 const int rowAb0 = (rowA[j] >> 5) & 0x7; 526 const int rowAb1 = (rowA[k] >> 5) & 0x7; 527 const int rowBb0 = (rowB[j] >> 5) & 0x7; 528 const int rowBb1 = (rowB[k] >> 5) & 0x7; 529 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 530 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 531 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 532 dst[i] = (blue << 5) | (green << 2) | red; 533 } 534 } 535 else { 536 debug_printf("bad format in do_row()"); 537 } 538} 539 540 541/** 542 * Average together four rows of a source image to produce a single new 543 * row in the dest image. It's legal for the two source rows to point 544 * to the same data. The source width must be equal to either the 545 * dest width or two times the dest width. 546 * 547 * \param datatype GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT, 548 * \c GL_FLOAT, etc. 549 * \param comps number of components per pixel (1..4) 550 * \param srcWidth Width of a row in the source data 551 * \param srcRowA Pointer to one of the rows of source data 552 * \param srcRowB Pointer to one of the rows of source data 553 * \param srcRowC Pointer to one of the rows of source data 554 * \param srcRowD Pointer to one of the rows of source data 555 * \param dstWidth Width of a row in the destination data 556 * \param srcRowA Pointer to the row of destination data 557 */ 558static void 559do_row_3D(enum dtype datatype, uint comps, int srcWidth, 560 const void *srcRowA, const void *srcRowB, 561 const void *srcRowC, const void *srcRowD, 562 int dstWidth, void *dstRow) 563{ 564 const uint k0 = (srcWidth == dstWidth) ? 0 : 1; 565 const uint colStride = (srcWidth == dstWidth) ? 1 : 2; 566 uint i, j, k; 567 568 assert(comps >= 1); 569 assert(comps <= 4); 570 571 if ((datatype == UBYTE) && (comps == 4)) { 572 DECLARE_ROW_POINTERS(ubyte, 4); 573 574 for (i = j = 0, k = k0; i < (uint) dstWidth; 575 i++, j += colStride, k += colStride) { 576 FILTER_3D(0); 577 FILTER_3D(1); 578 FILTER_3D(2); 579 FILTER_3D(3); 580 } 581 } 582 else if ((datatype == UBYTE) && (comps == 3)) { 583 DECLARE_ROW_POINTERS(ubyte, 3); 584 585 for (i = j = 0, k = k0; i < (uint) dstWidth; 586 i++, j += colStride, k += colStride) { 587 FILTER_3D(0); 588 FILTER_3D(1); 589 FILTER_3D(2); 590 } 591 } 592 else if ((datatype == UBYTE) && (comps == 2)) { 593 DECLARE_ROW_POINTERS(ubyte, 2); 594 595 for (i = j = 0, k = k0; i < (uint) dstWidth; 596 i++, j += colStride, k += colStride) { 597 FILTER_3D(0); 598 FILTER_3D(1); 599 } 600 } 601 else if ((datatype == UBYTE) && (comps == 1)) { 602 DECLARE_ROW_POINTERS(ubyte, 1); 603 604 for (i = j = 0, k = k0; i < (uint) dstWidth; 605 i++, j += colStride, k += colStride) { 606 FILTER_3D(0); 607 } 608 } 609 else if ((datatype == USHORT) && (comps == 4)) { 610 DECLARE_ROW_POINTERS(ushort, 4); 611 612 for (i = j = 0, k = k0; i < (uint) dstWidth; 613 i++, j += colStride, k += colStride) { 614 FILTER_3D(0); 615 FILTER_3D(1); 616 FILTER_3D(2); 617 FILTER_3D(3); 618 } 619 } 620 else if ((datatype == USHORT) && (comps == 3)) { 621 DECLARE_ROW_POINTERS(ushort, 3); 622 623 for (i = j = 0, k = k0; i < (uint) dstWidth; 624 i++, j += colStride, k += colStride) { 625 FILTER_3D(0); 626 FILTER_3D(1); 627 FILTER_3D(2); 628 } 629 } 630 else if ((datatype == USHORT) && (comps == 2)) { 631 DECLARE_ROW_POINTERS(ushort, 2); 632 633 for (i = j = 0, k = k0; i < (uint) dstWidth; 634 i++, j += colStride, k += colStride) { 635 FILTER_3D(0); 636 FILTER_3D(1); 637 } 638 } 639 else if ((datatype == USHORT) && (comps == 1)) { 640 DECLARE_ROW_POINTERS(ushort, 1); 641 642 for (i = j = 0, k = k0; i < (uint) dstWidth; 643 i++, j += colStride, k += colStride) { 644 FILTER_3D(0); 645 } 646 } 647 else if ((datatype == FLOAT) && (comps == 4)) { 648 DECLARE_ROW_POINTERS(float, 4); 649 650 for (i = j = 0, k = k0; i < (uint) dstWidth; 651 i++, j += colStride, k += colStride) { 652 FILTER_F_3D(0); 653 FILTER_F_3D(1); 654 FILTER_F_3D(2); 655 FILTER_F_3D(3); 656 } 657 } 658 else if ((datatype == FLOAT) && (comps == 3)) { 659 DECLARE_ROW_POINTERS(float, 3); 660 661 for (i = j = 0, k = k0; i < (uint) dstWidth; 662 i++, j += colStride, k += colStride) { 663 FILTER_F_3D(0); 664 FILTER_F_3D(1); 665 FILTER_F_3D(2); 666 } 667 } 668 else if ((datatype == FLOAT) && (comps == 2)) { 669 DECLARE_ROW_POINTERS(float, 2); 670 671 for (i = j = 0, k = k0; i < (uint) dstWidth; 672 i++, j += colStride, k += colStride) { 673 FILTER_F_3D(0); 674 FILTER_F_3D(1); 675 } 676 } 677 else if ((datatype == FLOAT) && (comps == 1)) { 678 DECLARE_ROW_POINTERS(float, 1); 679 680 for (i = j = 0, k = k0; i < (uint) dstWidth; 681 i++, j += colStride, k += colStride) { 682 FILTER_F_3D(0); 683 } 684 } 685 else if ((datatype == HALF_FLOAT) && (comps == 4)) { 686 DECLARE_ROW_POINTERS(half_float, 4); 687 688 for (i = j = 0, k = k0; i < (uint) dstWidth; 689 i++, j += colStride, k += colStride) { 690 FILTER_HF_3D(0); 691 FILTER_HF_3D(1); 692 FILTER_HF_3D(2); 693 FILTER_HF_3D(3); 694 } 695 } 696 else if ((datatype == HALF_FLOAT) && (comps == 3)) { 697 DECLARE_ROW_POINTERS(half_float, 4); 698 699 for (i = j = 0, k = k0; i < (uint) dstWidth; 700 i++, j += colStride, k += colStride) { 701 FILTER_HF_3D(0); 702 FILTER_HF_3D(1); 703 FILTER_HF_3D(2); 704 } 705 } 706 else if ((datatype == HALF_FLOAT) && (comps == 2)) { 707 DECLARE_ROW_POINTERS(half_float, 4); 708 709 for (i = j = 0, k = k0; i < (uint) dstWidth; 710 i++, j += colStride, k += colStride) { 711 FILTER_HF_3D(0); 712 FILTER_HF_3D(1); 713 } 714 } 715 else if ((datatype == HALF_FLOAT) && (comps == 1)) { 716 DECLARE_ROW_POINTERS(half_float, 4); 717 718 for (i = j = 0, k = k0; i < (uint) dstWidth; 719 i++, j += colStride, k += colStride) { 720 FILTER_HF_3D(0); 721 } 722 } 723 else if ((datatype == UINT) && (comps == 1)) { 724 const uint *rowA = (const uint *) srcRowA; 725 const uint *rowB = (const uint *) srcRowB; 726 const uint *rowC = (const uint *) srcRowC; 727 const uint *rowD = (const uint *) srcRowD; 728 float *dst = (float *) dstRow; 729 730 for (i = j = 0, k = k0; i < (uint) dstWidth; 731 i++, j += colStride, k += colStride) { 732 const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k]) 733 + ((uint64_t) rowB[j] + (uint64_t) rowB[k]) 734 + ((uint64_t) rowC[j] + (uint64_t) rowC[k]) 735 + ((uint64_t) rowD[j] + (uint64_t) rowD[k])); 736 dst[i] = (float)((double) tmp * 0.125); 737 } 738 } 739 else if ((datatype == USHORT_5_6_5) && (comps == 3)) { 740 DECLARE_ROW_POINTERS0(ushort); 741 742 for (i = j = 0, k = k0; i < (uint) dstWidth; 743 i++, j += colStride, k += colStride) { 744 const int rowAr0 = rowA[j] & 0x1f; 745 const int rowAr1 = rowA[k] & 0x1f; 746 const int rowBr0 = rowB[j] & 0x1f; 747 const int rowBr1 = rowB[k] & 0x1f; 748 const int rowCr0 = rowC[j] & 0x1f; 749 const int rowCr1 = rowC[k] & 0x1f; 750 const int rowDr0 = rowD[j] & 0x1f; 751 const int rowDr1 = rowD[k] & 0x1f; 752 const int rowAg0 = (rowA[j] >> 5) & 0x3f; 753 const int rowAg1 = (rowA[k] >> 5) & 0x3f; 754 const int rowBg0 = (rowB[j] >> 5) & 0x3f; 755 const int rowBg1 = (rowB[k] >> 5) & 0x3f; 756 const int rowCg0 = (rowC[j] >> 5) & 0x3f; 757 const int rowCg1 = (rowC[k] >> 5) & 0x3f; 758 const int rowDg0 = (rowD[j] >> 5) & 0x3f; 759 const int rowDg1 = (rowD[k] >> 5) & 0x3f; 760 const int rowAb0 = (rowA[j] >> 11) & 0x1f; 761 const int rowAb1 = (rowA[k] >> 11) & 0x1f; 762 const int rowBb0 = (rowB[j] >> 11) & 0x1f; 763 const int rowBb1 = (rowB[k] >> 11) & 0x1f; 764 const int rowCb0 = (rowC[j] >> 11) & 0x1f; 765 const int rowCb1 = (rowC[k] >> 11) & 0x1f; 766 const int rowDb0 = (rowD[j] >> 11) & 0x1f; 767 const int rowDb1 = (rowD[k] >> 11) & 0x1f; 768 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 769 rowCr0, rowCr1, rowDr0, rowDr1); 770 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 771 rowCg0, rowCg1, rowDg0, rowDg1); 772 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 773 rowCb0, rowCb1, rowDb0, rowDb1); 774 dst[i] = (b << 11) | (g << 5) | r; 775 } 776 } 777 else if ((datatype == USHORT_4_4_4_4) && (comps == 4)) { 778 DECLARE_ROW_POINTERS0(ushort); 779 780 for (i = j = 0, k = k0; i < (uint) dstWidth; 781 i++, j += colStride, k += colStride) { 782 const int rowAr0 = rowA[j] & 0xf; 783 const int rowAr1 = rowA[k] & 0xf; 784 const int rowBr0 = rowB[j] & 0xf; 785 const int rowBr1 = rowB[k] & 0xf; 786 const int rowCr0 = rowC[j] & 0xf; 787 const int rowCr1 = rowC[k] & 0xf; 788 const int rowDr0 = rowD[j] & 0xf; 789 const int rowDr1 = rowD[k] & 0xf; 790 const int rowAg0 = (rowA[j] >> 4) & 0xf; 791 const int rowAg1 = (rowA[k] >> 4) & 0xf; 792 const int rowBg0 = (rowB[j] >> 4) & 0xf; 793 const int rowBg1 = (rowB[k] >> 4) & 0xf; 794 const int rowCg0 = (rowC[j] >> 4) & 0xf; 795 const int rowCg1 = (rowC[k] >> 4) & 0xf; 796 const int rowDg0 = (rowD[j] >> 4) & 0xf; 797 const int rowDg1 = (rowD[k] >> 4) & 0xf; 798 const int rowAb0 = (rowA[j] >> 8) & 0xf; 799 const int rowAb1 = (rowA[k] >> 8) & 0xf; 800 const int rowBb0 = (rowB[j] >> 8) & 0xf; 801 const int rowBb1 = (rowB[k] >> 8) & 0xf; 802 const int rowCb0 = (rowC[j] >> 8) & 0xf; 803 const int rowCb1 = (rowC[k] >> 8) & 0xf; 804 const int rowDb0 = (rowD[j] >> 8) & 0xf; 805 const int rowDb1 = (rowD[k] >> 8) & 0xf; 806 const int rowAa0 = (rowA[j] >> 12) & 0xf; 807 const int rowAa1 = (rowA[k] >> 12) & 0xf; 808 const int rowBa0 = (rowB[j] >> 12) & 0xf; 809 const int rowBa1 = (rowB[k] >> 12) & 0xf; 810 const int rowCa0 = (rowC[j] >> 12) & 0xf; 811 const int rowCa1 = (rowC[k] >> 12) & 0xf; 812 const int rowDa0 = (rowD[j] >> 12) & 0xf; 813 const int rowDa1 = (rowD[k] >> 12) & 0xf; 814 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 815 rowCr0, rowCr1, rowDr0, rowDr1); 816 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 817 rowCg0, rowCg1, rowDg0, rowDg1); 818 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 819 rowCb0, rowCb1, rowDb0, rowDb1); 820 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1, 821 rowCa0, rowCa1, rowDa0, rowDa1); 822 823 dst[i] = (a << 12) | (b << 8) | (g << 4) | r; 824 } 825 } 826 else if ((datatype == USHORT_1_5_5_5_REV) && (comps == 4)) { 827 DECLARE_ROW_POINTERS0(ushort); 828 829 for (i = j = 0, k = k0; i < (uint) dstWidth; 830 i++, j += colStride, k += colStride) { 831 const int rowAr0 = rowA[j] & 0x1f; 832 const int rowAr1 = rowA[k] & 0x1f; 833 const int rowBr0 = rowB[j] & 0x1f; 834 const int rowBr1 = rowB[k] & 0x1f; 835 const int rowCr0 = rowC[j] & 0x1f; 836 const int rowCr1 = rowC[k] & 0x1f; 837 const int rowDr0 = rowD[j] & 0x1f; 838 const int rowDr1 = rowD[k] & 0x1f; 839 const int rowAg0 = (rowA[j] >> 5) & 0x1f; 840 const int rowAg1 = (rowA[k] >> 5) & 0x1f; 841 const int rowBg0 = (rowB[j] >> 5) & 0x1f; 842 const int rowBg1 = (rowB[k] >> 5) & 0x1f; 843 const int rowCg0 = (rowC[j] >> 5) & 0x1f; 844 const int rowCg1 = (rowC[k] >> 5) & 0x1f; 845 const int rowDg0 = (rowD[j] >> 5) & 0x1f; 846 const int rowDg1 = (rowD[k] >> 5) & 0x1f; 847 const int rowAb0 = (rowA[j] >> 10) & 0x1f; 848 const int rowAb1 = (rowA[k] >> 10) & 0x1f; 849 const int rowBb0 = (rowB[j] >> 10) & 0x1f; 850 const int rowBb1 = (rowB[k] >> 10) & 0x1f; 851 const int rowCb0 = (rowC[j] >> 10) & 0x1f; 852 const int rowCb1 = (rowC[k] >> 10) & 0x1f; 853 const int rowDb0 = (rowD[j] >> 10) & 0x1f; 854 const int rowDb1 = (rowD[k] >> 10) & 0x1f; 855 const int rowAa0 = (rowA[j] >> 15) & 0x1; 856 const int rowAa1 = (rowA[k] >> 15) & 0x1; 857 const int rowBa0 = (rowB[j] >> 15) & 0x1; 858 const int rowBa1 = (rowB[k] >> 15) & 0x1; 859 const int rowCa0 = (rowC[j] >> 15) & 0x1; 860 const int rowCa1 = (rowC[k] >> 15) & 0x1; 861 const int rowDa0 = (rowD[j] >> 15) & 0x1; 862 const int rowDa1 = (rowD[k] >> 15) & 0x1; 863 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 864 rowCr0, rowCr1, rowDr0, rowDr1); 865 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 866 rowCg0, rowCg1, rowDg0, rowDg1); 867 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 868 rowCb0, rowCb1, rowDb0, rowDb1); 869 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1, 870 rowCa0, rowCa1, rowDa0, rowDa1); 871 872 dst[i] = (a << 15) | (b << 10) | (g << 5) | r; 873 } 874 } 875 else if ((datatype == UBYTE_3_3_2) && (comps == 3)) { 876 DECLARE_ROW_POINTERS0(ushort); 877 878 for (i = j = 0, k = k0; i < (uint) dstWidth; 879 i++, j += colStride, k += colStride) { 880 const int rowAr0 = rowA[j] & 0x3; 881 const int rowAr1 = rowA[k] & 0x3; 882 const int rowBr0 = rowB[j] & 0x3; 883 const int rowBr1 = rowB[k] & 0x3; 884 const int rowCr0 = rowC[j] & 0x3; 885 const int rowCr1 = rowC[k] & 0x3; 886 const int rowDr0 = rowD[j] & 0x3; 887 const int rowDr1 = rowD[k] & 0x3; 888 const int rowAg0 = (rowA[j] >> 2) & 0x7; 889 const int rowAg1 = (rowA[k] >> 2) & 0x7; 890 const int rowBg0 = (rowB[j] >> 2) & 0x7; 891 const int rowBg1 = (rowB[k] >> 2) & 0x7; 892 const int rowCg0 = (rowC[j] >> 2) & 0x7; 893 const int rowCg1 = (rowC[k] >> 2) & 0x7; 894 const int rowDg0 = (rowD[j] >> 2) & 0x7; 895 const int rowDg1 = (rowD[k] >> 2) & 0x7; 896 const int rowAb0 = (rowA[j] >> 5) & 0x7; 897 const int rowAb1 = (rowA[k] >> 5) & 0x7; 898 const int rowBb0 = (rowB[j] >> 5) & 0x7; 899 const int rowBb1 = (rowB[k] >> 5) & 0x7; 900 const int rowCb0 = (rowC[j] >> 5) & 0x7; 901 const int rowCb1 = (rowC[k] >> 5) & 0x7; 902 const int rowDb0 = (rowD[j] >> 5) & 0x7; 903 const int rowDb1 = (rowD[k] >> 5) & 0x7; 904 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 905 rowCr0, rowCr1, rowDr0, rowDr1); 906 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 907 rowCg0, rowCg1, rowDg0, rowDg1); 908 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 909 rowCb0, rowCb1, rowDb0, rowDb1); 910 dst[i] = (b << 5) | (g << 2) | r; 911 } 912 } 913 else { 914 debug_printf("bad format in do_row_3D()"); 915 } 916} 917 918 919 920static void 921format_to_type_comps(enum pipe_format pformat, 922 enum dtype *datatype, uint *comps) 923{ 924 switch (pformat) { 925 case PIPE_FORMAT_A8R8G8B8_UNORM: 926 case PIPE_FORMAT_X8R8G8B8_UNORM: 927 case PIPE_FORMAT_B8G8R8A8_UNORM: 928 case PIPE_FORMAT_B8G8R8X8_UNORM: 929 *datatype = UBYTE; 930 *comps = 4; 931 return; 932 case PIPE_FORMAT_A1R5G5B5_UNORM: 933 *datatype = USHORT_1_5_5_5_REV; 934 *comps = 4; 935 return; 936 case PIPE_FORMAT_A4R4G4B4_UNORM: 937 *datatype = USHORT_4_4_4_4; 938 *comps = 4; 939 return; 940 case PIPE_FORMAT_R5G6B5_UNORM: 941 *datatype = USHORT_5_6_5; 942 *comps = 3; 943 return; 944 case PIPE_FORMAT_L8_UNORM: 945 case PIPE_FORMAT_A8_UNORM: 946 case PIPE_FORMAT_I8_UNORM: 947 *datatype = UBYTE; 948 *comps = 1; 949 return; 950 case PIPE_FORMAT_A8L8_UNORM: 951 *datatype = UBYTE; 952 *comps = 2; 953 return; 954 default: 955 assert(0); 956 *datatype = UBYTE; 957 *comps = 0; 958 break; 959 } 960} 961 962 963static void 964reduce_1d(enum pipe_format pformat, 965 int srcWidth, const ubyte *srcPtr, 966 int dstWidth, ubyte *dstPtr) 967{ 968 enum dtype datatype; 969 uint comps; 970 971 format_to_type_comps(pformat, &datatype, &comps); 972 973 /* we just duplicate the input row, kind of hack, saves code */ 974 do_row(datatype, comps, 975 srcWidth, srcPtr, srcPtr, 976 dstWidth, dstPtr); 977} 978 979 980/** 981 * Strides are in bytes. If zero, it'll be computed as width * bpp. 982 */ 983static void 984reduce_2d(enum pipe_format pformat, 985 int srcWidth, int srcHeight, 986 int srcRowStride, const ubyte *srcPtr, 987 int dstWidth, int dstHeight, 988 int dstRowStride, ubyte *dstPtr) 989{ 990 enum dtype datatype; 991 uint comps; 992 const int bpt = pf_get_size(pformat); 993 const ubyte *srcA, *srcB; 994 ubyte *dst; 995 int row; 996 997 format_to_type_comps(pformat, &datatype, &comps); 998 999 if (!srcRowStride) 1000 srcRowStride = bpt * srcWidth; 1001 1002 if (!dstRowStride) 1003 dstRowStride = bpt * dstWidth; 1004 1005 /* Compute src and dst pointers */ 1006 srcA = srcPtr; 1007 if (srcHeight > 1) 1008 srcB = srcA + srcRowStride; 1009 else 1010 srcB = srcA; 1011 dst = dstPtr; 1012 1013 for (row = 0; row < dstHeight; row++) { 1014 do_row(datatype, comps, 1015 srcWidth, srcA, srcB, 1016 dstWidth, dst); 1017 srcA += 2 * srcRowStride; 1018 srcB += 2 * srcRowStride; 1019 dst += dstRowStride; 1020 } 1021} 1022 1023 1024static void 1025reduce_3d(enum pipe_format pformat, 1026 int srcWidth, int srcHeight, int srcDepth, 1027 int srcRowStride, const ubyte *srcPtr, 1028 int dstWidth, int dstHeight, int dstDepth, 1029 int dstRowStride, ubyte *dstPtr) 1030{ 1031 const int bpt = pf_get_size(pformat); 1032 const int border = 0; 1033 int img, row; 1034 int bytesPerSrcImage, bytesPerDstImage; 1035 int bytesPerSrcRow, bytesPerDstRow; 1036 int srcImageOffset, srcRowOffset; 1037 enum dtype datatype; 1038 uint comps; 1039 1040 format_to_type_comps(pformat, &datatype, &comps); 1041 1042 bytesPerSrcImage = srcWidth * srcHeight * bpt; 1043 bytesPerDstImage = dstWidth * dstHeight * bpt; 1044 1045 bytesPerSrcRow = srcWidth * bpt; 1046 bytesPerDstRow = dstWidth * bpt; 1047 1048 /* Offset between adjacent src images to be averaged together */ 1049 srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage; 1050 1051 /* Offset between adjacent src rows to be averaged together */ 1052 srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt; 1053 1054 /* 1055 * Need to average together up to 8 src pixels for each dest pixel. 1056 * Break that down into 3 operations: 1057 * 1. take two rows from source image and average them together. 1058 * 2. take two rows from next source image and average them together. 1059 * 3. take the two averaged rows and average them for the final dst row. 1060 */ 1061 1062 /* 1063 _mesa_printf("mip3d %d x %d x %d -> %d x %d x %d\n", 1064 srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth); 1065 */ 1066 1067 for (img = 0; img < dstDepth; img++) { 1068 /* first source image pointer, skipping border */ 1069 const ubyte *imgSrcA = srcPtr 1070 + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border 1071 + img * (bytesPerSrcImage + srcImageOffset); 1072 /* second source image pointer, skipping border */ 1073 const ubyte *imgSrcB = imgSrcA + srcImageOffset; 1074 /* address of the dest image, skipping border */ 1075 ubyte *imgDst = dstPtr 1076 + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border 1077 + img * bytesPerDstImage; 1078 1079 /* setup the four source row pointers and the dest row pointer */ 1080 const ubyte *srcImgARowA = imgSrcA; 1081 const ubyte *srcImgARowB = imgSrcA + srcRowOffset; 1082 const ubyte *srcImgBRowA = imgSrcB; 1083 const ubyte *srcImgBRowB = imgSrcB + srcRowOffset; 1084 ubyte *dstImgRow = imgDst; 1085 1086 for (row = 0; row < dstHeight; row++) { 1087 do_row_3D(datatype, comps, srcWidth, 1088 srcImgARowA, srcImgARowB, 1089 srcImgBRowA, srcImgBRowB, 1090 dstWidth, dstImgRow); 1091 1092 /* advance to next rows */ 1093 srcImgARowA += bytesPerSrcRow + srcRowOffset; 1094 srcImgARowB += bytesPerSrcRow + srcRowOffset; 1095 srcImgBRowA += bytesPerSrcRow + srcRowOffset; 1096 srcImgBRowB += bytesPerSrcRow + srcRowOffset; 1097 dstImgRow += bytesPerDstRow; 1098 } 1099 } 1100} 1101 1102 1103 1104 1105static void 1106make_1d_mipmap(struct gen_mipmap_state *ctx, 1107 struct pipe_texture *pt, 1108 uint face, uint baseLevel, uint lastLevel) 1109{ 1110 struct pipe_context *pipe = ctx->pipe; 1111 struct pipe_screen *screen = pipe->screen; 1112 const uint zslice = 0; 1113 uint dstLevel; 1114 1115 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1116 const uint srcLevel = dstLevel - 1; 1117 struct pipe_transfer *srcTrans, *dstTrans; 1118 void *srcMap, *dstMap; 1119 1120 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, 1121 PIPE_TRANSFER_READ, 0, 0, 1122 pt->width[srcLevel], 1123 pt->height[srcLevel]); 1124 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, 1125 PIPE_TRANSFER_WRITE, 0, 0, 1126 pt->width[dstLevel], 1127 pt->height[dstLevel]); 1128 1129 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); 1130 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); 1131 1132 reduce_1d(pt->format, 1133 srcTrans->width, srcMap, 1134 dstTrans->width, dstMap); 1135 1136 screen->transfer_unmap(screen, srcTrans); 1137 screen->transfer_unmap(screen, dstTrans); 1138 1139 screen->tex_transfer_destroy(srcTrans); 1140 screen->tex_transfer_destroy(dstTrans); 1141 } 1142} 1143 1144 1145static void 1146make_2d_mipmap(struct gen_mipmap_state *ctx, 1147 struct pipe_texture *pt, 1148 uint face, uint baseLevel, uint lastLevel) 1149{ 1150 struct pipe_context *pipe = ctx->pipe; 1151 struct pipe_screen *screen = pipe->screen; 1152 const uint zslice = 0; 1153 uint dstLevel; 1154 1155 assert(pt->block.width == 1); 1156 assert(pt->block.height == 1); 1157 1158 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1159 const uint srcLevel = dstLevel - 1; 1160 struct pipe_transfer *srcTrans, *dstTrans; 1161 ubyte *srcMap, *dstMap; 1162 1163 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, 1164 PIPE_TRANSFER_READ, 0, 0, 1165 pt->width[srcLevel], 1166 pt->height[srcLevel]); 1167 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, 1168 PIPE_TRANSFER_WRITE, 0, 0, 1169 pt->width[dstLevel], 1170 pt->height[dstLevel]); 1171 1172 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); 1173 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); 1174 1175 reduce_2d(pt->format, 1176 srcTrans->width, srcTrans->height, 1177 srcTrans->stride, srcMap, 1178 dstTrans->width, dstTrans->height, 1179 dstTrans->stride, dstMap); 1180 1181 screen->transfer_unmap(screen, srcTrans); 1182 screen->transfer_unmap(screen, dstTrans); 1183 1184 screen->tex_transfer_destroy(srcTrans); 1185 screen->tex_transfer_destroy(dstTrans); 1186 } 1187} 1188 1189 1190static void 1191make_3d_mipmap(struct gen_mipmap_state *ctx, 1192 struct pipe_texture *pt, 1193 uint face, uint baseLevel, uint lastLevel) 1194{ 1195#if 0 1196 struct pipe_context *pipe = ctx->pipe; 1197 struct pipe_screen *screen = pipe->screen; 1198 uint dstLevel, zslice = 0; 1199 1200 assert(pt->block.width == 1); 1201 assert(pt->block.height == 1); 1202 1203 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1204 const uint srcLevel = dstLevel - 1; 1205 struct pipe_transfer *srcTrans, *dstTrans; 1206 ubyte *srcMap, *dstMap; 1207 1208 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, 1209 PIPE_TRANSFER_READ, 0, 0, 1210 pt->width[srcLevel], 1211 pt->height[srcLevel]); 1212 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, 1213 PIPE_TRANSFER_WRITE, 0, 0, 1214 pt->width[dstLevel], 1215 pt->height[dstLevel]); 1216 1217 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); 1218 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); 1219 1220 reduce_3d(pt->format, 1221 srcTrans->width, srcTrans->height, 1222 srcTrans->stride, srcMap, 1223 dstTrans->width, dstTrans->height, 1224 dstTrans->stride, dstMap); 1225 1226 screen->transfer_unmap(screen, srcTrans); 1227 screen->transfer_unmap(screen, dstTrans); 1228 1229 screen->tex_transfer_destroy(srcTrans); 1230 screen->tex_transfer_destroy(dstTrans); 1231 } 1232#else 1233 (void) reduce_3d; 1234#endif 1235} 1236 1237 1238static void 1239fallback_gen_mipmap(struct gen_mipmap_state *ctx, 1240 struct pipe_texture *pt, 1241 uint face, uint baseLevel, uint lastLevel) 1242{ 1243 switch (pt->target) { 1244 case PIPE_TEXTURE_1D: 1245 make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel); 1246 break; 1247 case PIPE_TEXTURE_2D: 1248 case PIPE_TEXTURE_CUBE: 1249 make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel); 1250 break; 1251 case PIPE_TEXTURE_3D: 1252 make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel); 1253 break; 1254 default: 1255 assert(0); 1256 } 1257} 1258 1259 1260/** 1261 * Create a mipmap generation context. 1262 * The idea is to create one of these and re-use it each time we need to 1263 * generate a mipmap. 1264 */ 1265struct gen_mipmap_state * 1266util_create_gen_mipmap(struct pipe_context *pipe, 1267 struct cso_context *cso) 1268{ 1269 struct gen_mipmap_state *ctx; 1270 uint i; 1271 1272 ctx = CALLOC_STRUCT(gen_mipmap_state); 1273 if (!ctx) 1274 return NULL; 1275 1276 ctx->pipe = pipe; 1277 ctx->cso = cso; 1278 1279 /* disabled blending/masking */ 1280 memset(&ctx->blend, 0, sizeof(ctx->blend)); 1281 ctx->blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; 1282 ctx->blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; 1283 ctx->blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; 1284 ctx->blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; 1285 ctx->blend.colormask = PIPE_MASK_RGBA; 1286 1287 /* no-op depth/stencil/alpha */ 1288 memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil)); 1289 1290 /* rasterizer */ 1291 memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer)); 1292 ctx->rasterizer.front_winding = PIPE_WINDING_CW; 1293 ctx->rasterizer.cull_mode = PIPE_WINDING_NONE; 1294 ctx->rasterizer.bypass_vs_clip_and_viewport = 1; 1295 ctx->rasterizer.gl_rasterization_rules = 1; 1296 1297 /* sampler state */ 1298 memset(&ctx->sampler, 0, sizeof(ctx->sampler)); 1299 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1300 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1301 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1302 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST; 1303 ctx->sampler.normalized_coords = 1; 1304 1305 /* vertex shader - still needed to specify mapping from fragment 1306 * shader input semantics to vertex elements 1307 */ 1308 { 1309 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION, 1310 TGSI_SEMANTIC_GENERIC }; 1311 const uint semantic_indexes[] = { 0, 0 }; 1312 ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names, 1313 semantic_indexes); 1314 } 1315 1316 /* fragment shader */ 1317 ctx->fs = util_make_fragment_tex_shader(pipe); 1318 1319 /* vertex data that doesn't change */ 1320 for (i = 0; i < 4; i++) { 1321 ctx->vertices[i][0][2] = 0.0f; /* z */ 1322 ctx->vertices[i][0][3] = 1.0f; /* w */ 1323 ctx->vertices[i][1][2] = 0.0f; /* r */ 1324 ctx->vertices[i][1][3] = 1.0f; /* q */ 1325 } 1326 1327 /* Note: the actual vertex buffer is allocated as needed below */ 1328 1329 return ctx; 1330} 1331 1332 1333/** 1334 * Get next "slot" of vertex space in the vertex buffer. 1335 * We're allocating one large vertex buffer and using it piece by piece. 1336 */ 1337static unsigned 1338get_next_slot(struct gen_mipmap_state *ctx) 1339{ 1340 const unsigned max_slots = 4096 / sizeof ctx->vertices; 1341 1342 if (ctx->vbuf_slot >= max_slots) 1343 util_gen_mipmap_flush( ctx ); 1344 1345 if (!ctx->vbuf) { 1346 ctx->vbuf = pipe_buffer_create(ctx->pipe->screen, 1347 32, 1348 PIPE_BUFFER_USAGE_VERTEX, 1349 max_slots * sizeof ctx->vertices); 1350 } 1351 1352 return ctx->vbuf_slot++ * sizeof ctx->vertices; 1353} 1354 1355 1356static unsigned 1357set_vertex_data(struct gen_mipmap_state *ctx, float width, float height) 1358{ 1359 unsigned offset; 1360 1361 ctx->vertices[0][0][0] = 0.0f; /*x*/ 1362 ctx->vertices[0][0][1] = 0.0f; /*y*/ 1363 ctx->vertices[0][1][0] = 0.0f; /*s*/ 1364 ctx->vertices[0][1][1] = 0.0f; /*t*/ 1365 1366 ctx->vertices[1][0][0] = width; 1367 ctx->vertices[1][0][1] = 0.0f; 1368 ctx->vertices[1][1][0] = 1.0f; 1369 ctx->vertices[1][1][1] = 0.0f; 1370 1371 ctx->vertices[2][0][0] = width; 1372 ctx->vertices[2][0][1] = height; 1373 ctx->vertices[2][1][0] = 1.0f; 1374 ctx->vertices[2][1][1] = 1.0f; 1375 1376 ctx->vertices[3][0][0] = 0.0f; 1377 ctx->vertices[3][0][1] = height; 1378 ctx->vertices[3][1][0] = 0.0f; 1379 ctx->vertices[3][1][1] = 1.0f; 1380 1381 offset = get_next_slot( ctx ); 1382 1383 pipe_buffer_write(ctx->pipe->screen, ctx->vbuf, 1384 offset, sizeof(ctx->vertices), ctx->vertices); 1385 1386 return offset; 1387} 1388 1389 1390 1391/** 1392 * Destroy a mipmap generation context 1393 */ 1394void 1395util_destroy_gen_mipmap(struct gen_mipmap_state *ctx) 1396{ 1397 struct pipe_context *pipe = ctx->pipe; 1398 1399 pipe->delete_vs_state(pipe, ctx->vs); 1400 pipe->delete_fs_state(pipe, ctx->fs); 1401 1402 pipe_buffer_reference(&ctx->vbuf, NULL); 1403 1404 FREE(ctx); 1405} 1406 1407 1408 1409/* Release vertex buffer at end of frame to avoid synchronous 1410 * rendering. 1411 */ 1412void util_gen_mipmap_flush( struct gen_mipmap_state *ctx ) 1413{ 1414 pipe_buffer_reference(&ctx->vbuf, NULL); 1415 ctx->vbuf_slot = 0; 1416} 1417 1418 1419/** 1420 * Generate mipmap images. It's assumed all needed texture memory is 1421 * already allocated. 1422 * 1423 * \param pt the texture to generate mipmap levels for 1424 * \param face which cube face to generate mipmaps for (0 for non-cube maps) 1425 * \param baseLevel the first mipmap level to use as a src 1426 * \param lastLevel the last mipmap level to generate 1427 * \param filter the minification filter used to generate mipmap levels with 1428 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST 1429 */ 1430void 1431util_gen_mipmap(struct gen_mipmap_state *ctx, 1432 struct pipe_texture *pt, 1433 uint face, uint baseLevel, uint lastLevel, uint filter) 1434{ 1435 struct pipe_context *pipe = ctx->pipe; 1436 struct pipe_screen *screen = pipe->screen; 1437 struct pipe_framebuffer_state fb; 1438 uint dstLevel; 1439 uint zslice = 0; 1440 uint offset; 1441 1442 /* check if we can render in the texture's format */ 1443 if (!screen->is_format_supported(screen, pt->format, PIPE_TEXTURE_2D, 1444 PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) { 1445 fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel); 1446 return; 1447 } 1448 1449 /* save state (restored below) */ 1450 cso_save_blend(ctx->cso); 1451 cso_save_depth_stencil_alpha(ctx->cso); 1452 cso_save_rasterizer(ctx->cso); 1453 cso_save_samplers(ctx->cso); 1454 cso_save_sampler_textures(ctx->cso); 1455 cso_save_framebuffer(ctx->cso); 1456 cso_save_fragment_shader(ctx->cso); 1457 cso_save_vertex_shader(ctx->cso); 1458 1459 /* bind our state */ 1460 cso_set_blend(ctx->cso, &ctx->blend); 1461 cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil); 1462 cso_set_rasterizer(ctx->cso, &ctx->rasterizer); 1463 1464 cso_set_fragment_shader_handle(ctx->cso, ctx->fs); 1465 cso_set_vertex_shader_handle(ctx->cso, ctx->vs); 1466 1467 /* init framebuffer state */ 1468 memset(&fb, 0, sizeof(fb)); 1469 fb.nr_cbufs = 1; 1470 1471 /* set min/mag to same filter for faster sw speed */ 1472 ctx->sampler.mag_img_filter = filter; 1473 ctx->sampler.min_img_filter = filter; 1474 1475 /* 1476 * XXX for small mipmap levels, it may be faster to use the software 1477 * fallback path... 1478 */ 1479 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1480 const uint srcLevel = dstLevel - 1; 1481 1482 struct pipe_surface *surf = 1483 screen->get_tex_surface(screen, pt, face, dstLevel, zslice, 1484 PIPE_BUFFER_USAGE_GPU_WRITE); 1485 1486 /* 1487 * Setup framebuffer / dest surface 1488 */ 1489 fb.cbufs[0] = surf; 1490 fb.width = pt->width[dstLevel]; 1491 fb.height = pt->height[dstLevel]; 1492 cso_set_framebuffer(ctx->cso, &fb); 1493 1494 /* 1495 * Setup sampler state 1496 * Note: we should only have to set the min/max LOD clamps to ensure 1497 * we grab texels from the right mipmap level. But some hardware 1498 * has trouble with min clamping so we also set the lod_bias to 1499 * try to work around that. 1500 */ 1501 ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel; 1502 ctx->sampler.lod_bias = (float) srcLevel; 1503 cso_single_sampler(ctx->cso, 0, &ctx->sampler); 1504 cso_single_sampler_done(ctx->cso); 1505 1506 cso_set_sampler_textures(ctx->cso, 1, &pt); 1507 1508 /* quad coords in window coords (bypassing vs, clip and viewport) */ 1509 offset = set_vertex_data(ctx, 1510 (float) pt->width[dstLevel], 1511 (float) pt->height[dstLevel]); 1512 1513 util_draw_vertex_buffer(ctx->pipe, 1514 ctx->vbuf, 1515 offset, 1516 PIPE_PRIM_TRIANGLE_FAN, 1517 4, /* verts */ 1518 2); /* attribs/vert */ 1519 1520 pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); 1521 1522 /* need to signal that the texture has changed _after_ rendering to it */ 1523 pipe_surface_reference( &surf, NULL ); 1524 } 1525 1526 /* restore state we changed */ 1527 cso_restore_blend(ctx->cso); 1528 cso_restore_depth_stencil_alpha(ctx->cso); 1529 cso_restore_rasterizer(ctx->cso); 1530 cso_restore_samplers(ctx->cso); 1531 cso_restore_sampler_textures(ctx->cso); 1532 cso_restore_framebuffer(ctx->cso); 1533 cso_restore_fragment_shader(ctx->cso); 1534 cso_restore_vertex_shader(ctx->cso); 1535} 1536