u_gen_mipmap.c revision 4617981ec72f7985941bee4b03c534d97ff96bc6
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 "pipe/p_debug.h" 39#include "pipe/p_defines.h" 40#include "pipe/p_inlines.h" 41#include "pipe/p_shader_tokens.h" 42 43#include "util/u_memory.h" 44#include "util/u_draw_quad.h" 45#include "util/u_gen_mipmap.h" 46#include "util/u_simple_shaders.h" 47 48#include "tgsi/tgsi_build.h" 49#include "tgsi/tgsi_dump.h" 50#include "tgsi/tgsi_parse.h" 51 52#include "cso_cache/cso_context.h" 53 54 55struct gen_mipmap_state 56{ 57 struct pipe_context *pipe; 58 struct cso_context *cso; 59 60 struct pipe_blend_state blend; 61 struct pipe_depth_stencil_alpha_state depthstencil; 62 struct pipe_rasterizer_state rasterizer; 63 struct pipe_sampler_state sampler; 64 struct pipe_viewport_state viewport; 65 66 struct pipe_shader_state vert_shader; 67 struct pipe_shader_state frag_shader; 68 void *vs; 69 void *fs; 70 71 struct pipe_buffer *vbuf; /**< quad vertices */ 72 unsigned vbuf_slot; 73 74 float vertices[4][2][4]; /**< vertex/texcoords for quad */ 75}; 76 77 78 79enum dtype 80{ 81 UBYTE, 82 UBYTE_3_3_2, 83 USHORT, 84 USHORT_4_4_4_4, 85 USHORT_5_6_5, 86 USHORT_1_5_5_5_REV, 87 UINT, 88 FLOAT, 89 HALF_FLOAT 90}; 91 92 93typedef ushort half_float; 94 95 96static half_float 97float_to_half(float f) 98{ 99 /* XXX fix this */ 100 return 0; 101} 102 103static float 104half_to_float(half_float h) 105{ 106 /* XXX fix this */ 107 return 0.0f; 108} 109 110 111 112 113/** 114 * \name Support macros for do_row and do_row_3d 115 * 116 * The macro madness is here for two reasons. First, it compacts the code 117 * slightly. Second, it makes it much easier to adjust the specifics of the 118 * filter to tune the rounding characteristics. 119 */ 120/*@{*/ 121#define DECLARE_ROW_POINTERS(t, e) \ 122 const t(*rowA)[e] = (const t(*)[e]) srcRowA; \ 123 const t(*rowB)[e] = (const t(*)[e]) srcRowB; \ 124 const t(*rowC)[e] = (const t(*)[e]) srcRowC; \ 125 const t(*rowD)[e] = (const t(*)[e]) srcRowD; \ 126 t(*dst)[e] = (t(*)[e]) dstRow 127 128#define DECLARE_ROW_POINTERS0(t) \ 129 const t *rowA = (const t *) srcRowA; \ 130 const t *rowB = (const t *) srcRowB; \ 131 const t *rowC = (const t *) srcRowC; \ 132 const t *rowD = (const t *) srcRowD; \ 133 t *dst = (t *) dstRow 134 135#define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \ 136 ((unsigned) Aj + (unsigned) Ak \ 137 + (unsigned) Bj + (unsigned) Bk \ 138 + (unsigned) Cj + (unsigned) Ck \ 139 + (unsigned) Dj + (unsigned) Dk \ 140 + 4) >> 3 141 142#define FILTER_3D(e) \ 143 do { \ 144 dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \ 145 rowB[j][e], rowB[k][e], \ 146 rowC[j][e], rowC[k][e], \ 147 rowD[j][e], rowD[k][e]); \ 148 } while(0) 149 150#define FILTER_F_3D(e) \ 151 do { \ 152 dst[i][e] = (rowA[j][e] + rowA[k][e] \ 153 + rowB[j][e] + rowB[k][e] \ 154 + rowC[j][e] + rowC[k][e] \ 155 + rowD[j][e] + rowD[k][e]) * 0.125F; \ 156 } while(0) 157 158#define FILTER_HF_3D(e) \ 159 do { \ 160 const float aj = half_to_float(rowA[j][e]); \ 161 const float ak = half_to_float(rowA[k][e]); \ 162 const float bj = half_to_float(rowB[j][e]); \ 163 const float bk = half_to_float(rowB[k][e]); \ 164 const float cj = half_to_float(rowC[j][e]); \ 165 const float ck = half_to_float(rowC[k][e]); \ 166 const float dj = half_to_float(rowD[j][e]); \ 167 const float dk = half_to_float(rowD[k][e]); \ 168 dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \ 169 * 0.125F); \ 170 } while(0) 171/*@}*/ 172 173 174/** 175 * Average together two rows of a source image to produce a single new 176 * row in the dest image. It's legal for the two source rows to point 177 * to the same data. The source width must be equal to either the 178 * dest width or two times the dest width. 179 * \param datatype GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc. 180 * \param comps number of components per pixel (1..4) 181 */ 182static void 183do_row(enum dtype datatype, uint comps, int srcWidth, 184 const void *srcRowA, const void *srcRowB, 185 int dstWidth, void *dstRow) 186{ 187 const uint k0 = (srcWidth == dstWidth) ? 0 : 1; 188 const uint colStride = (srcWidth == dstWidth) ? 1 : 2; 189 190 assert(comps >= 1); 191 assert(comps <= 4); 192 193 /* This assertion is no longer valid with non-power-of-2 textures 194 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth); 195 */ 196 197 if (datatype == UBYTE && comps == 4) { 198 uint i, j, k; 199 const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA; 200 const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB; 201 ubyte(*dst)[4] = (ubyte(*)[4]) dstRow; 202 for (i = j = 0, k = k0; i < (uint) dstWidth; 203 i++, j += colStride, k += colStride) { 204 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 205 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 206 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 207 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 208 } 209 } 210 else if (datatype == UBYTE && comps == 3) { 211 uint i, j, k; 212 const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA; 213 const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB; 214 ubyte(*dst)[3] = (ubyte(*)[3]) dstRow; 215 for (i = j = 0, k = k0; i < (uint) dstWidth; 216 i++, j += colStride, k += colStride) { 217 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 218 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 219 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 220 } 221 } 222 else if (datatype == UBYTE && comps == 2) { 223 uint i, j, k; 224 const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA; 225 const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB; 226 ubyte(*dst)[2] = (ubyte(*)[2]) dstRow; 227 for (i = j = 0, k = k0; i < (uint) dstWidth; 228 i++, j += colStride, k += colStride) { 229 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2; 230 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2; 231 } 232 } 233 else if (datatype == UBYTE && comps == 1) { 234 uint i, j, k; 235 const ubyte *rowA = (const ubyte *) srcRowA; 236 const ubyte *rowB = (const ubyte *) srcRowB; 237 ubyte *dst = (ubyte *) dstRow; 238 for (i = j = 0, k = k0; i < (uint) dstWidth; 239 i++, j += colStride, k += colStride) { 240 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2; 241 } 242 } 243 244 else if (datatype == USHORT && comps == 4) { 245 uint i, j, k; 246 const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA; 247 const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB; 248 ushort(*dst)[4] = (ushort(*)[4]) dstRow; 249 for (i = j = 0, k = k0; i < (uint) dstWidth; 250 i++, j += colStride, k += colStride) { 251 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 252 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 253 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 254 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 255 } 256 } 257 else if (datatype == USHORT && comps == 3) { 258 uint i, j, k; 259 const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA; 260 const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB; 261 ushort(*dst)[3] = (ushort(*)[3]) dstRow; 262 for (i = j = 0, k = k0; i < (uint) dstWidth; 263 i++, j += colStride, k += colStride) { 264 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 265 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 266 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 267 } 268 } 269 else if (datatype == USHORT && comps == 2) { 270 uint i, j, k; 271 const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA; 272 const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB; 273 ushort(*dst)[2] = (ushort(*)[2]) dstRow; 274 for (i = j = 0, k = k0; i < (uint) dstWidth; 275 i++, j += colStride, k += colStride) { 276 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 277 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 278 } 279 } 280 else if (datatype == USHORT && comps == 1) { 281 uint i, j, k; 282 const ushort *rowA = (const ushort *) srcRowA; 283 const ushort *rowB = (const ushort *) srcRowB; 284 ushort *dst = (ushort *) dstRow; 285 for (i = j = 0, k = k0; i < (uint) dstWidth; 286 i++, j += colStride, k += colStride) { 287 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4; 288 } 289 } 290 291 else if (datatype == FLOAT && comps == 4) { 292 uint i, j, k; 293 const float(*rowA)[4] = (const float(*)[4]) srcRowA; 294 const float(*rowB)[4] = (const float(*)[4]) srcRowB; 295 float(*dst)[4] = (float(*)[4]) dstRow; 296 for (i = j = 0, k = k0; i < (uint) dstWidth; 297 i++, j += colStride, k += colStride) { 298 dst[i][0] = (rowA[j][0] + rowA[k][0] + 299 rowB[j][0] + rowB[k][0]) * 0.25F; 300 dst[i][1] = (rowA[j][1] + rowA[k][1] + 301 rowB[j][1] + rowB[k][1]) * 0.25F; 302 dst[i][2] = (rowA[j][2] + rowA[k][2] + 303 rowB[j][2] + rowB[k][2]) * 0.25F; 304 dst[i][3] = (rowA[j][3] + rowA[k][3] + 305 rowB[j][3] + rowB[k][3]) * 0.25F; 306 } 307 } 308 else if (datatype == FLOAT && comps == 3) { 309 uint i, j, k; 310 const float(*rowA)[3] = (const float(*)[3]) srcRowA; 311 const float(*rowB)[3] = (const float(*)[3]) srcRowB; 312 float(*dst)[3] = (float(*)[3]) dstRow; 313 for (i = j = 0, k = k0; i < (uint) dstWidth; 314 i++, j += colStride, k += colStride) { 315 dst[i][0] = (rowA[j][0] + rowA[k][0] + 316 rowB[j][0] + rowB[k][0]) * 0.25F; 317 dst[i][1] = (rowA[j][1] + rowA[k][1] + 318 rowB[j][1] + rowB[k][1]) * 0.25F; 319 dst[i][2] = (rowA[j][2] + rowA[k][2] + 320 rowB[j][2] + rowB[k][2]) * 0.25F; 321 } 322 } 323 else if (datatype == FLOAT && comps == 2) { 324 uint i, j, k; 325 const float(*rowA)[2] = (const float(*)[2]) srcRowA; 326 const float(*rowB)[2] = (const float(*)[2]) srcRowB; 327 float(*dst)[2] = (float(*)[2]) dstRow; 328 for (i = j = 0, k = k0; i < (uint) dstWidth; 329 i++, j += colStride, k += colStride) { 330 dst[i][0] = (rowA[j][0] + rowA[k][0] + 331 rowB[j][0] + rowB[k][0]) * 0.25F; 332 dst[i][1] = (rowA[j][1] + rowA[k][1] + 333 rowB[j][1] + rowB[k][1]) * 0.25F; 334 } 335 } 336 else if (datatype == FLOAT && comps == 1) { 337 uint i, j, k; 338 const float *rowA = (const float *) srcRowA; 339 const float *rowB = (const float *) srcRowB; 340 float *dst = (float *) dstRow; 341 for (i = j = 0, k = k0; i < (uint) dstWidth; 342 i++, j += colStride, k += colStride) { 343 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F; 344 } 345 } 346 347#if 0 348 else if (datatype == HALF_FLOAT && comps == 4) { 349 uint i, j, k, comp; 350 const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA; 351 const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB; 352 half_float(*dst)[4] = (half_float(*)[4]) dstRow; 353 for (i = j = 0, k = k0; i < (uint) dstWidth; 354 i++, j += colStride, k += colStride) { 355 for (comp = 0; comp < 4; comp++) { 356 float aj, ak, bj, bk; 357 aj = half_to_float(rowA[j][comp]); 358 ak = half_to_float(rowA[k][comp]); 359 bj = half_to_float(rowB[j][comp]); 360 bk = half_to_float(rowB[k][comp]); 361 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 362 } 363 } 364 } 365 else if (datatype == HALF_FLOAT && comps == 3) { 366 uint i, j, k, comp; 367 const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA; 368 const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB; 369 half_float(*dst)[3] = (half_float(*)[3]) dstRow; 370 for (i = j = 0, k = k0; i < (uint) dstWidth; 371 i++, j += colStride, k += colStride) { 372 for (comp = 0; comp < 3; comp++) { 373 float aj, ak, bj, bk; 374 aj = half_to_float(rowA[j][comp]); 375 ak = half_to_float(rowA[k][comp]); 376 bj = half_to_float(rowB[j][comp]); 377 bk = half_to_float(rowB[k][comp]); 378 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 379 } 380 } 381 } 382 else if (datatype == HALF_FLOAT && comps == 2) { 383 uint i, j, k, comp; 384 const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA; 385 const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB; 386 half_float(*dst)[2] = (half_float(*)[2]) dstRow; 387 for (i = j = 0, k = k0; i < (uint) dstWidth; 388 i++, j += colStride, k += colStride) { 389 for (comp = 0; comp < 2; comp++) { 390 float aj, ak, bj, bk; 391 aj = half_to_float(rowA[j][comp]); 392 ak = half_to_float(rowA[k][comp]); 393 bj = half_to_float(rowB[j][comp]); 394 bk = half_to_float(rowB[k][comp]); 395 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 396 } 397 } 398 } 399 else if (datatype == HALF_FLOAT && comps == 1) { 400 uint i, j, k; 401 const half_float *rowA = (const half_float *) srcRowA; 402 const half_float *rowB = (const half_float *) srcRowB; 403 half_float *dst = (half_float *) dstRow; 404 for (i = j = 0, k = k0; i < (uint) dstWidth; 405 i++, j += colStride, k += colStride) { 406 float aj, ak, bj, bk; 407 aj = half_to_float(rowA[j]); 408 ak = half_to_float(rowA[k]); 409 bj = half_to_float(rowB[j]); 410 bk = half_to_float(rowB[k]); 411 dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F); 412 } 413 } 414#endif 415 416 else if (datatype == UINT && comps == 1) { 417 uint i, j, k; 418 const uint *rowA = (const uint *) srcRowA; 419 const uint *rowB = (const uint *) srcRowB; 420 uint *dst = (uint *) dstRow; 421 for (i = j = 0, k = k0; i < (uint) dstWidth; 422 i++, j += colStride, k += colStride) { 423 dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4; 424 } 425 } 426 427 else if (datatype == USHORT_5_6_5 && comps == 3) { 428 uint i, j, k; 429 const ushort *rowA = (const ushort *) srcRowA; 430 const ushort *rowB = (const ushort *) srcRowB; 431 ushort *dst = (ushort *) dstRow; 432 for (i = j = 0, k = k0; i < (uint) dstWidth; 433 i++, j += colStride, k += colStride) { 434 const int rowAr0 = rowA[j] & 0x1f; 435 const int rowAr1 = rowA[k] & 0x1f; 436 const int rowBr0 = rowB[j] & 0x1f; 437 const int rowBr1 = rowB[k] & 0x1f; 438 const int rowAg0 = (rowA[j] >> 5) & 0x3f; 439 const int rowAg1 = (rowA[k] >> 5) & 0x3f; 440 const int rowBg0 = (rowB[j] >> 5) & 0x3f; 441 const int rowBg1 = (rowB[k] >> 5) & 0x3f; 442 const int rowAb0 = (rowA[j] >> 11) & 0x1f; 443 const int rowAb1 = (rowA[k] >> 11) & 0x1f; 444 const int rowBb0 = (rowB[j] >> 11) & 0x1f; 445 const int rowBb1 = (rowB[k] >> 11) & 0x1f; 446 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 447 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 448 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 449 dst[i] = (blue << 11) | (green << 5) | red; 450 } 451 } 452 else if (datatype == USHORT_4_4_4_4 && comps == 4) { 453 uint i, j, k; 454 const ushort *rowA = (const ushort *) srcRowA; 455 const ushort *rowB = (const ushort *) srcRowB; 456 ushort *dst = (ushort *) dstRow; 457 for (i = j = 0, k = k0; i < (uint) dstWidth; 458 i++, j += colStride, k += colStride) { 459 const int rowAr0 = rowA[j] & 0xf; 460 const int rowAr1 = rowA[k] & 0xf; 461 const int rowBr0 = rowB[j] & 0xf; 462 const int rowBr1 = rowB[k] & 0xf; 463 const int rowAg0 = (rowA[j] >> 4) & 0xf; 464 const int rowAg1 = (rowA[k] >> 4) & 0xf; 465 const int rowBg0 = (rowB[j] >> 4) & 0xf; 466 const int rowBg1 = (rowB[k] >> 4) & 0xf; 467 const int rowAb0 = (rowA[j] >> 8) & 0xf; 468 const int rowAb1 = (rowA[k] >> 8) & 0xf; 469 const int rowBb0 = (rowB[j] >> 8) & 0xf; 470 const int rowBb1 = (rowB[k] >> 8) & 0xf; 471 const int rowAa0 = (rowA[j] >> 12) & 0xf; 472 const int rowAa1 = (rowA[k] >> 12) & 0xf; 473 const int rowBa0 = (rowB[j] >> 12) & 0xf; 474 const int rowBa1 = (rowB[k] >> 12) & 0xf; 475 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 476 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 477 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 478 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 479 dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red; 480 } 481 } 482 else if (datatype == USHORT_1_5_5_5_REV && comps == 4) { 483 uint i, j, k; 484 const ushort *rowA = (const ushort *) srcRowA; 485 const ushort *rowB = (const ushort *) srcRowB; 486 ushort *dst = (ushort *) dstRow; 487 for (i = j = 0, k = k0; i < (uint) dstWidth; 488 i++, j += colStride, k += colStride) { 489 const int rowAr0 = rowA[j] & 0x1f; 490 const int rowAr1 = rowA[k] & 0x1f; 491 const int rowBr0 = rowB[j] & 0x1f; 492 const int rowBr1 = rowB[k] & 0x1f; 493 const int rowAg0 = (rowA[j] >> 5) & 0x1f; 494 const int rowAg1 = (rowA[k] >> 5) & 0x1f; 495 const int rowBg0 = (rowB[j] >> 5) & 0x1f; 496 const int rowBg1 = (rowB[k] >> 5) & 0x1f; 497 const int rowAb0 = (rowA[j] >> 10) & 0x1f; 498 const int rowAb1 = (rowA[k] >> 10) & 0x1f; 499 const int rowBb0 = (rowB[j] >> 10) & 0x1f; 500 const int rowBb1 = (rowB[k] >> 10) & 0x1f; 501 const int rowAa0 = (rowA[j] >> 15) & 0x1; 502 const int rowAa1 = (rowA[k] >> 15) & 0x1; 503 const int rowBa0 = (rowB[j] >> 15) & 0x1; 504 const int rowBa1 = (rowB[k] >> 15) & 0x1; 505 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 506 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 507 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 508 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 509 dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red; 510 } 511 } 512 else if (datatype == UBYTE_3_3_2 && comps == 3) { 513 uint i, j, k; 514 const ubyte *rowA = (const ubyte *) srcRowA; 515 const ubyte *rowB = (const ubyte *) srcRowB; 516 ubyte *dst = (ubyte *) dstRow; 517 for (i = j = 0, k = k0; i < (uint) dstWidth; 518 i++, j += colStride, k += colStride) { 519 const int rowAr0 = rowA[j] & 0x3; 520 const int rowAr1 = rowA[k] & 0x3; 521 const int rowBr0 = rowB[j] & 0x3; 522 const int rowBr1 = rowB[k] & 0x3; 523 const int rowAg0 = (rowA[j] >> 2) & 0x7; 524 const int rowAg1 = (rowA[k] >> 2) & 0x7; 525 const int rowBg0 = (rowB[j] >> 2) & 0x7; 526 const int rowBg1 = (rowB[k] >> 2) & 0x7; 527 const int rowAb0 = (rowA[j] >> 5) & 0x7; 528 const int rowAb1 = (rowA[k] >> 5) & 0x7; 529 const int rowBb0 = (rowB[j] >> 5) & 0x7; 530 const int rowBb1 = (rowB[k] >> 5) & 0x7; 531 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 532 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 533 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 534 dst[i] = (blue << 5) | (green << 2) | red; 535 } 536 } 537 else { 538 debug_printf("bad format in do_row()"); 539 } 540} 541 542 543/** 544 * Average together four rows of a source image to produce a single new 545 * row in the dest image. It's legal for the two source rows to point 546 * to the same data. The source width must be equal to either the 547 * dest width or two times the dest width. 548 * 549 * \param datatype GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT, 550 * \c GL_FLOAT, etc. 551 * \param comps number of components per pixel (1..4) 552 * \param srcWidth Width of a row in the source data 553 * \param srcRowA Pointer to one of the rows of source data 554 * \param srcRowB Pointer to one of the rows of source data 555 * \param srcRowC Pointer to one of the rows of source data 556 * \param srcRowD Pointer to one of the rows of source data 557 * \param dstWidth Width of a row in the destination data 558 * \param srcRowA Pointer to the row of destination data 559 */ 560static void 561do_row_3D(enum dtype datatype, uint comps, int srcWidth, 562 const void *srcRowA, const void *srcRowB, 563 const void *srcRowC, const void *srcRowD, 564 int dstWidth, void *dstRow) 565{ 566 const uint k0 = (srcWidth == dstWidth) ? 0 : 1; 567 const uint colStride = (srcWidth == dstWidth) ? 1 : 2; 568 uint i, j, k; 569 570 assert(comps >= 1); 571 assert(comps <= 4); 572 573 if ((datatype == UBYTE) && (comps == 4)) { 574 DECLARE_ROW_POINTERS(ubyte, 4); 575 576 for (i = j = 0, k = k0; i < (uint) dstWidth; 577 i++, j += colStride, k += colStride) { 578 FILTER_3D(0); 579 FILTER_3D(1); 580 FILTER_3D(2); 581 FILTER_3D(3); 582 } 583 } 584 else if ((datatype == UBYTE) && (comps == 3)) { 585 DECLARE_ROW_POINTERS(ubyte, 3); 586 587 for (i = j = 0, k = k0; i < (uint) dstWidth; 588 i++, j += colStride, k += colStride) { 589 FILTER_3D(0); 590 FILTER_3D(1); 591 FILTER_3D(2); 592 } 593 } 594 else if ((datatype == UBYTE) && (comps == 2)) { 595 DECLARE_ROW_POINTERS(ubyte, 2); 596 597 for (i = j = 0, k = k0; i < (uint) dstWidth; 598 i++, j += colStride, k += colStride) { 599 FILTER_3D(0); 600 FILTER_3D(1); 601 } 602 } 603 else if ((datatype == UBYTE) && (comps == 1)) { 604 DECLARE_ROW_POINTERS(ubyte, 1); 605 606 for (i = j = 0, k = k0; i < (uint) dstWidth; 607 i++, j += colStride, k += colStride) { 608 FILTER_3D(0); 609 } 610 } 611 else if ((datatype == USHORT) && (comps == 4)) { 612 DECLARE_ROW_POINTERS(ushort, 4); 613 614 for (i = j = 0, k = k0; i < (uint) dstWidth; 615 i++, j += colStride, k += colStride) { 616 FILTER_3D(0); 617 FILTER_3D(1); 618 FILTER_3D(2); 619 FILTER_3D(3); 620 } 621 } 622 else if ((datatype == USHORT) && (comps == 3)) { 623 DECLARE_ROW_POINTERS(ushort, 3); 624 625 for (i = j = 0, k = k0; i < (uint) dstWidth; 626 i++, j += colStride, k += colStride) { 627 FILTER_3D(0); 628 FILTER_3D(1); 629 FILTER_3D(2); 630 } 631 } 632 else if ((datatype == USHORT) && (comps == 2)) { 633 DECLARE_ROW_POINTERS(ushort, 2); 634 635 for (i = j = 0, k = k0; i < (uint) dstWidth; 636 i++, j += colStride, k += colStride) { 637 FILTER_3D(0); 638 FILTER_3D(1); 639 } 640 } 641 else if ((datatype == USHORT) && (comps == 1)) { 642 DECLARE_ROW_POINTERS(ushort, 1); 643 644 for (i = j = 0, k = k0; i < (uint) dstWidth; 645 i++, j += colStride, k += colStride) { 646 FILTER_3D(0); 647 } 648 } 649 else if ((datatype == FLOAT) && (comps == 4)) { 650 DECLARE_ROW_POINTERS(float, 4); 651 652 for (i = j = 0, k = k0; i < (uint) dstWidth; 653 i++, j += colStride, k += colStride) { 654 FILTER_F_3D(0); 655 FILTER_F_3D(1); 656 FILTER_F_3D(2); 657 FILTER_F_3D(3); 658 } 659 } 660 else if ((datatype == FLOAT) && (comps == 3)) { 661 DECLARE_ROW_POINTERS(float, 3); 662 663 for (i = j = 0, k = k0; i < (uint) dstWidth; 664 i++, j += colStride, k += colStride) { 665 FILTER_F_3D(0); 666 FILTER_F_3D(1); 667 FILTER_F_3D(2); 668 } 669 } 670 else if ((datatype == FLOAT) && (comps == 2)) { 671 DECLARE_ROW_POINTERS(float, 2); 672 673 for (i = j = 0, k = k0; i < (uint) dstWidth; 674 i++, j += colStride, k += colStride) { 675 FILTER_F_3D(0); 676 FILTER_F_3D(1); 677 } 678 } 679 else if ((datatype == FLOAT) && (comps == 1)) { 680 DECLARE_ROW_POINTERS(float, 1); 681 682 for (i = j = 0, k = k0; i < (uint) dstWidth; 683 i++, j += colStride, k += colStride) { 684 FILTER_F_3D(0); 685 } 686 } 687 else if ((datatype == HALF_FLOAT) && (comps == 4)) { 688 DECLARE_ROW_POINTERS(half_float, 4); 689 690 for (i = j = 0, k = k0; i < (uint) dstWidth; 691 i++, j += colStride, k += colStride) { 692 FILTER_HF_3D(0); 693 FILTER_HF_3D(1); 694 FILTER_HF_3D(2); 695 FILTER_HF_3D(3); 696 } 697 } 698 else if ((datatype == HALF_FLOAT) && (comps == 3)) { 699 DECLARE_ROW_POINTERS(half_float, 4); 700 701 for (i = j = 0, k = k0; i < (uint) dstWidth; 702 i++, j += colStride, k += colStride) { 703 FILTER_HF_3D(0); 704 FILTER_HF_3D(1); 705 FILTER_HF_3D(2); 706 } 707 } 708 else if ((datatype == HALF_FLOAT) && (comps == 2)) { 709 DECLARE_ROW_POINTERS(half_float, 4); 710 711 for (i = j = 0, k = k0; i < (uint) dstWidth; 712 i++, j += colStride, k += colStride) { 713 FILTER_HF_3D(0); 714 FILTER_HF_3D(1); 715 } 716 } 717 else if ((datatype == HALF_FLOAT) && (comps == 1)) { 718 DECLARE_ROW_POINTERS(half_float, 4); 719 720 for (i = j = 0, k = k0; i < (uint) dstWidth; 721 i++, j += colStride, k += colStride) { 722 FILTER_HF_3D(0); 723 } 724 } 725 else if ((datatype == UINT) && (comps == 1)) { 726 const uint *rowA = (const uint *) srcRowA; 727 const uint *rowB = (const uint *) srcRowB; 728 const uint *rowC = (const uint *) srcRowC; 729 const uint *rowD = (const uint *) srcRowD; 730 float *dst = (float *) dstRow; 731 732 for (i = j = 0, k = k0; i < (uint) dstWidth; 733 i++, j += colStride, k += colStride) { 734 const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k]) 735 + ((uint64_t) rowB[j] + (uint64_t) rowB[k]) 736 + ((uint64_t) rowC[j] + (uint64_t) rowC[k]) 737 + ((uint64_t) rowD[j] + (uint64_t) rowD[k])); 738 dst[i] = (float)((double) tmp * 0.125); 739 } 740 } 741 else if ((datatype == USHORT_5_6_5) && (comps == 3)) { 742 DECLARE_ROW_POINTERS0(ushort); 743 744 for (i = j = 0, k = k0; i < (uint) dstWidth; 745 i++, j += colStride, k += colStride) { 746 const int rowAr0 = rowA[j] & 0x1f; 747 const int rowAr1 = rowA[k] & 0x1f; 748 const int rowBr0 = rowB[j] & 0x1f; 749 const int rowBr1 = rowB[k] & 0x1f; 750 const int rowCr0 = rowC[j] & 0x1f; 751 const int rowCr1 = rowC[k] & 0x1f; 752 const int rowDr0 = rowD[j] & 0x1f; 753 const int rowDr1 = rowD[k] & 0x1f; 754 const int rowAg0 = (rowA[j] >> 5) & 0x3f; 755 const int rowAg1 = (rowA[k] >> 5) & 0x3f; 756 const int rowBg0 = (rowB[j] >> 5) & 0x3f; 757 const int rowBg1 = (rowB[k] >> 5) & 0x3f; 758 const int rowCg0 = (rowC[j] >> 5) & 0x3f; 759 const int rowCg1 = (rowC[k] >> 5) & 0x3f; 760 const int rowDg0 = (rowD[j] >> 5) & 0x3f; 761 const int rowDg1 = (rowD[k] >> 5) & 0x3f; 762 const int rowAb0 = (rowA[j] >> 11) & 0x1f; 763 const int rowAb1 = (rowA[k] >> 11) & 0x1f; 764 const int rowBb0 = (rowB[j] >> 11) & 0x1f; 765 const int rowBb1 = (rowB[k] >> 11) & 0x1f; 766 const int rowCb0 = (rowC[j] >> 11) & 0x1f; 767 const int rowCb1 = (rowC[k] >> 11) & 0x1f; 768 const int rowDb0 = (rowD[j] >> 11) & 0x1f; 769 const int rowDb1 = (rowD[k] >> 11) & 0x1f; 770 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 771 rowCr0, rowCr1, rowDr0, rowDr1); 772 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 773 rowCg0, rowCg1, rowDg0, rowDg1); 774 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 775 rowCb0, rowCb1, rowDb0, rowDb1); 776 dst[i] = (b << 11) | (g << 5) | r; 777 } 778 } 779 else if ((datatype == USHORT_4_4_4_4) && (comps == 4)) { 780 DECLARE_ROW_POINTERS0(ushort); 781 782 for (i = j = 0, k = k0; i < (uint) dstWidth; 783 i++, j += colStride, k += colStride) { 784 const int rowAr0 = rowA[j] & 0xf; 785 const int rowAr1 = rowA[k] & 0xf; 786 const int rowBr0 = rowB[j] & 0xf; 787 const int rowBr1 = rowB[k] & 0xf; 788 const int rowCr0 = rowC[j] & 0xf; 789 const int rowCr1 = rowC[k] & 0xf; 790 const int rowDr0 = rowD[j] & 0xf; 791 const int rowDr1 = rowD[k] & 0xf; 792 const int rowAg0 = (rowA[j] >> 4) & 0xf; 793 const int rowAg1 = (rowA[k] >> 4) & 0xf; 794 const int rowBg0 = (rowB[j] >> 4) & 0xf; 795 const int rowBg1 = (rowB[k] >> 4) & 0xf; 796 const int rowCg0 = (rowC[j] >> 4) & 0xf; 797 const int rowCg1 = (rowC[k] >> 4) & 0xf; 798 const int rowDg0 = (rowD[j] >> 4) & 0xf; 799 const int rowDg1 = (rowD[k] >> 4) & 0xf; 800 const int rowAb0 = (rowA[j] >> 8) & 0xf; 801 const int rowAb1 = (rowA[k] >> 8) & 0xf; 802 const int rowBb0 = (rowB[j] >> 8) & 0xf; 803 const int rowBb1 = (rowB[k] >> 8) & 0xf; 804 const int rowCb0 = (rowC[j] >> 8) & 0xf; 805 const int rowCb1 = (rowC[k] >> 8) & 0xf; 806 const int rowDb0 = (rowD[j] >> 8) & 0xf; 807 const int rowDb1 = (rowD[k] >> 8) & 0xf; 808 const int rowAa0 = (rowA[j] >> 12) & 0xf; 809 const int rowAa1 = (rowA[k] >> 12) & 0xf; 810 const int rowBa0 = (rowB[j] >> 12) & 0xf; 811 const int rowBa1 = (rowB[k] >> 12) & 0xf; 812 const int rowCa0 = (rowC[j] >> 12) & 0xf; 813 const int rowCa1 = (rowC[k] >> 12) & 0xf; 814 const int rowDa0 = (rowD[j] >> 12) & 0xf; 815 const int rowDa1 = (rowD[k] >> 12) & 0xf; 816 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 817 rowCr0, rowCr1, rowDr0, rowDr1); 818 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 819 rowCg0, rowCg1, rowDg0, rowDg1); 820 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 821 rowCb0, rowCb1, rowDb0, rowDb1); 822 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1, 823 rowCa0, rowCa1, rowDa0, rowDa1); 824 825 dst[i] = (a << 12) | (b << 8) | (g << 4) | r; 826 } 827 } 828 else if ((datatype == USHORT_1_5_5_5_REV) && (comps == 4)) { 829 DECLARE_ROW_POINTERS0(ushort); 830 831 for (i = j = 0, k = k0; i < (uint) dstWidth; 832 i++, j += colStride, k += colStride) { 833 const int rowAr0 = rowA[j] & 0x1f; 834 const int rowAr1 = rowA[k] & 0x1f; 835 const int rowBr0 = rowB[j] & 0x1f; 836 const int rowBr1 = rowB[k] & 0x1f; 837 const int rowCr0 = rowC[j] & 0x1f; 838 const int rowCr1 = rowC[k] & 0x1f; 839 const int rowDr0 = rowD[j] & 0x1f; 840 const int rowDr1 = rowD[k] & 0x1f; 841 const int rowAg0 = (rowA[j] >> 5) & 0x1f; 842 const int rowAg1 = (rowA[k] >> 5) & 0x1f; 843 const int rowBg0 = (rowB[j] >> 5) & 0x1f; 844 const int rowBg1 = (rowB[k] >> 5) & 0x1f; 845 const int rowCg0 = (rowC[j] >> 5) & 0x1f; 846 const int rowCg1 = (rowC[k] >> 5) & 0x1f; 847 const int rowDg0 = (rowD[j] >> 5) & 0x1f; 848 const int rowDg1 = (rowD[k] >> 5) & 0x1f; 849 const int rowAb0 = (rowA[j] >> 10) & 0x1f; 850 const int rowAb1 = (rowA[k] >> 10) & 0x1f; 851 const int rowBb0 = (rowB[j] >> 10) & 0x1f; 852 const int rowBb1 = (rowB[k] >> 10) & 0x1f; 853 const int rowCb0 = (rowC[j] >> 10) & 0x1f; 854 const int rowCb1 = (rowC[k] >> 10) & 0x1f; 855 const int rowDb0 = (rowD[j] >> 10) & 0x1f; 856 const int rowDb1 = (rowD[k] >> 10) & 0x1f; 857 const int rowAa0 = (rowA[j] >> 15) & 0x1; 858 const int rowAa1 = (rowA[k] >> 15) & 0x1; 859 const int rowBa0 = (rowB[j] >> 15) & 0x1; 860 const int rowBa1 = (rowB[k] >> 15) & 0x1; 861 const int rowCa0 = (rowC[j] >> 15) & 0x1; 862 const int rowCa1 = (rowC[k] >> 15) & 0x1; 863 const int rowDa0 = (rowD[j] >> 15) & 0x1; 864 const int rowDa1 = (rowD[k] >> 15) & 0x1; 865 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 866 rowCr0, rowCr1, rowDr0, rowDr1); 867 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 868 rowCg0, rowCg1, rowDg0, rowDg1); 869 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 870 rowCb0, rowCb1, rowDb0, rowDb1); 871 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1, 872 rowCa0, rowCa1, rowDa0, rowDa1); 873 874 dst[i] = (a << 15) | (b << 10) | (g << 5) | r; 875 } 876 } 877 else if ((datatype == UBYTE_3_3_2) && (comps == 3)) { 878 DECLARE_ROW_POINTERS0(ushort); 879 880 for (i = j = 0, k = k0; i < (uint) dstWidth; 881 i++, j += colStride, k += colStride) { 882 const int rowAr0 = rowA[j] & 0x3; 883 const int rowAr1 = rowA[k] & 0x3; 884 const int rowBr0 = rowB[j] & 0x3; 885 const int rowBr1 = rowB[k] & 0x3; 886 const int rowCr0 = rowC[j] & 0x3; 887 const int rowCr1 = rowC[k] & 0x3; 888 const int rowDr0 = rowD[j] & 0x3; 889 const int rowDr1 = rowD[k] & 0x3; 890 const int rowAg0 = (rowA[j] >> 2) & 0x7; 891 const int rowAg1 = (rowA[k] >> 2) & 0x7; 892 const int rowBg0 = (rowB[j] >> 2) & 0x7; 893 const int rowBg1 = (rowB[k] >> 2) & 0x7; 894 const int rowCg0 = (rowC[j] >> 2) & 0x7; 895 const int rowCg1 = (rowC[k] >> 2) & 0x7; 896 const int rowDg0 = (rowD[j] >> 2) & 0x7; 897 const int rowDg1 = (rowD[k] >> 2) & 0x7; 898 const int rowAb0 = (rowA[j] >> 5) & 0x7; 899 const int rowAb1 = (rowA[k] >> 5) & 0x7; 900 const int rowBb0 = (rowB[j] >> 5) & 0x7; 901 const int rowBb1 = (rowB[k] >> 5) & 0x7; 902 const int rowCb0 = (rowC[j] >> 5) & 0x7; 903 const int rowCb1 = (rowC[k] >> 5) & 0x7; 904 const int rowDb0 = (rowD[j] >> 5) & 0x7; 905 const int rowDb1 = (rowD[k] >> 5) & 0x7; 906 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 907 rowCr0, rowCr1, rowDr0, rowDr1); 908 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 909 rowCg0, rowCg1, rowDg0, rowDg1); 910 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 911 rowCb0, rowCb1, rowDb0, rowDb1); 912 dst[i] = (b << 5) | (g << 2) | r; 913 } 914 } 915 else { 916 debug_printf("bad format in do_row_3D()"); 917 } 918} 919 920 921 922static void 923format_to_type_comps(enum pipe_format pformat, 924 enum dtype *datatype, uint *comps) 925{ 926 switch (pformat) { 927 case PIPE_FORMAT_A8R8G8B8_UNORM: 928 case PIPE_FORMAT_X8R8G8B8_UNORM: 929 case PIPE_FORMAT_B8G8R8A8_UNORM: 930 case PIPE_FORMAT_B8G8R8X8_UNORM: 931 *datatype = UBYTE; 932 *comps = 4; 933 return; 934 case PIPE_FORMAT_A1R5G5B5_UNORM: 935 *datatype = USHORT_1_5_5_5_REV; 936 *comps = 4; 937 return; 938 case PIPE_FORMAT_A4R4G4B4_UNORM: 939 *datatype = USHORT_4_4_4_4; 940 *comps = 4; 941 return; 942 case PIPE_FORMAT_R5G6B5_UNORM: 943 *datatype = USHORT_5_6_5; 944 *comps = 3; 945 return; 946 case PIPE_FORMAT_L8_UNORM: 947 case PIPE_FORMAT_A8_UNORM: 948 case PIPE_FORMAT_I8_UNORM: 949 *datatype = UBYTE; 950 *comps = 1; 951 return; 952 case PIPE_FORMAT_A8L8_UNORM: 953 *datatype = UBYTE; 954 *comps = 2; 955 return; 956 default: 957 assert(0); 958 *datatype = UBYTE; 959 *comps = 0; 960 break; 961 } 962} 963 964 965static void 966reduce_1d(enum pipe_format pformat, 967 int srcWidth, const ubyte *srcPtr, 968 int dstWidth, ubyte *dstPtr) 969{ 970 enum dtype datatype; 971 uint comps; 972 973 format_to_type_comps(pformat, &datatype, &comps); 974 975 /* we just duplicate the input row, kind of hack, saves code */ 976 do_row(datatype, comps, 977 srcWidth, srcPtr, srcPtr, 978 dstWidth, dstPtr); 979} 980 981 982/** 983 * Strides are in bytes. If zero, it'll be computed as width * bpp. 984 */ 985static void 986reduce_2d(enum pipe_format pformat, 987 int srcWidth, int srcHeight, 988 int srcRowStride, const ubyte *srcPtr, 989 int dstWidth, int dstHeight, 990 int dstRowStride, ubyte *dstPtr) 991{ 992 enum dtype datatype; 993 uint comps; 994 const int bpt = pf_get_size(pformat); 995 const ubyte *srcA, *srcB; 996 ubyte *dst; 997 int row; 998 999 format_to_type_comps(pformat, &datatype, &comps); 1000 1001 if (!srcRowStride) 1002 srcRowStride = bpt * srcWidth; 1003 1004 if (!dstRowStride) 1005 dstRowStride = bpt * dstWidth; 1006 1007 /* Compute src and dst pointers */ 1008 srcA = srcPtr; 1009 if (srcHeight > 1) 1010 srcB = srcA + srcRowStride; 1011 else 1012 srcB = srcA; 1013 dst = dstPtr; 1014 1015 for (row = 0; row < dstHeight; row++) { 1016 do_row(datatype, comps, 1017 srcWidth, srcA, srcB, 1018 dstWidth, dst); 1019 srcA += 2 * srcRowStride; 1020 srcB += 2 * srcRowStride; 1021 dst += dstRowStride; 1022 } 1023} 1024 1025 1026static void 1027reduce_3d(enum pipe_format pformat, 1028 int srcWidth, int srcHeight, int srcDepth, 1029 int srcRowStride, const ubyte *srcPtr, 1030 int dstWidth, int dstHeight, int dstDepth, 1031 int dstRowStride, ubyte *dstPtr) 1032{ 1033 const int bpt = pf_get_size(pformat); 1034 const int border = 0; 1035 int img, row; 1036 int bytesPerSrcImage, bytesPerDstImage; 1037 int bytesPerSrcRow, bytesPerDstRow; 1038 int srcImageOffset, srcRowOffset; 1039 enum dtype datatype; 1040 uint comps; 1041 1042 format_to_type_comps(pformat, &datatype, &comps); 1043 1044 bytesPerSrcImage = srcWidth * srcHeight * bpt; 1045 bytesPerDstImage = dstWidth * dstHeight * bpt; 1046 1047 bytesPerSrcRow = srcWidth * bpt; 1048 bytesPerDstRow = dstWidth * bpt; 1049 1050 /* Offset between adjacent src images to be averaged together */ 1051 srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage; 1052 1053 /* Offset between adjacent src rows to be averaged together */ 1054 srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt; 1055 1056 /* 1057 * Need to average together up to 8 src pixels for each dest pixel. 1058 * Break that down into 3 operations: 1059 * 1. take two rows from source image and average them together. 1060 * 2. take two rows from next source image and average them together. 1061 * 3. take the two averaged rows and average them for the final dst row. 1062 */ 1063 1064 /* 1065 _mesa_printf("mip3d %d x %d x %d -> %d x %d x %d\n", 1066 srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth); 1067 */ 1068 1069 for (img = 0; img < dstDepth; img++) { 1070 /* first source image pointer, skipping border */ 1071 const ubyte *imgSrcA = srcPtr 1072 + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border 1073 + img * (bytesPerSrcImage + srcImageOffset); 1074 /* second source image pointer, skipping border */ 1075 const ubyte *imgSrcB = imgSrcA + srcImageOffset; 1076 /* address of the dest image, skipping border */ 1077 ubyte *imgDst = dstPtr 1078 + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border 1079 + img * bytesPerDstImage; 1080 1081 /* setup the four source row pointers and the dest row pointer */ 1082 const ubyte *srcImgARowA = imgSrcA; 1083 const ubyte *srcImgARowB = imgSrcA + srcRowOffset; 1084 const ubyte *srcImgBRowA = imgSrcB; 1085 const ubyte *srcImgBRowB = imgSrcB + srcRowOffset; 1086 ubyte *dstImgRow = imgDst; 1087 1088 for (row = 0; row < dstHeight; row++) { 1089 do_row_3D(datatype, comps, srcWidth, 1090 srcImgARowA, srcImgARowB, 1091 srcImgBRowA, srcImgBRowB, 1092 dstWidth, dstImgRow); 1093 1094 /* advance to next rows */ 1095 srcImgARowA += bytesPerSrcRow + srcRowOffset; 1096 srcImgARowB += bytesPerSrcRow + srcRowOffset; 1097 srcImgBRowA += bytesPerSrcRow + srcRowOffset; 1098 srcImgBRowB += bytesPerSrcRow + srcRowOffset; 1099 dstImgRow += bytesPerDstRow; 1100 } 1101 } 1102} 1103 1104 1105 1106 1107static void 1108make_1d_mipmap(struct gen_mipmap_state *ctx, 1109 struct pipe_texture *pt, 1110 uint face, uint baseLevel, uint lastLevel) 1111{ 1112 struct pipe_context *pipe = ctx->pipe; 1113 struct pipe_screen *screen = pipe->screen; 1114 const uint zslice = 0; 1115 uint dstLevel; 1116 1117 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1118 const uint srcLevel = dstLevel - 1; 1119 struct pipe_transfer *srcTrans, *dstTrans; 1120 void *srcMap, *dstMap; 1121 1122 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, 1123 PIPE_TRANSFER_READ, 0, 0, 1124 pt->width[srcLevel], 1125 pt->height[srcLevel]); 1126 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, 1127 PIPE_TRANSFER_WRITE, 0, 0, 1128 pt->width[dstLevel], 1129 pt->height[dstLevel]); 1130 1131 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); 1132 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); 1133 1134 reduce_1d(pt->format, 1135 srcTrans->width, srcMap, 1136 dstTrans->width, dstMap); 1137 1138 screen->transfer_unmap(screen, srcTrans); 1139 screen->transfer_unmap(screen, dstTrans); 1140 1141 screen->tex_transfer_release(screen, &srcTrans); 1142 screen->tex_transfer_release(screen, &dstTrans); 1143 } 1144} 1145 1146 1147static void 1148make_2d_mipmap(struct gen_mipmap_state *ctx, 1149 struct pipe_texture *pt, 1150 uint face, uint baseLevel, uint lastLevel) 1151{ 1152 struct pipe_context *pipe = ctx->pipe; 1153 struct pipe_screen *screen = pipe->screen; 1154 const uint zslice = 0; 1155 uint dstLevel; 1156 1157 assert(pt->block.width == 1); 1158 assert(pt->block.height == 1); 1159 1160 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1161 const uint srcLevel = dstLevel - 1; 1162 struct pipe_transfer *srcTrans, *dstTrans; 1163 ubyte *srcMap, *dstMap; 1164 1165 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, 1166 PIPE_TRANSFER_READ, 0, 0, 1167 pt->width[srcLevel], 1168 pt->height[srcLevel]); 1169 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, 1170 PIPE_TRANSFER_WRITE, 0, 0, 1171 pt->width[dstLevel], 1172 pt->height[dstLevel]); 1173 1174 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); 1175 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); 1176 1177 reduce_2d(pt->format, 1178 srcTrans->width, srcTrans->height, 1179 srcTrans->stride, srcMap, 1180 dstTrans->width, dstTrans->height, 1181 dstTrans->stride, dstMap); 1182 1183 screen->transfer_unmap(screen, srcTrans); 1184 screen->transfer_unmap(screen, dstTrans); 1185 1186 screen->tex_transfer_release(screen, &srcTrans); 1187 screen->tex_transfer_release(screen, &dstTrans); 1188 } 1189} 1190 1191 1192static void 1193make_3d_mipmap(struct gen_mipmap_state *ctx, 1194 struct pipe_texture *pt, 1195 uint face, uint baseLevel, uint lastLevel) 1196{ 1197#if 0 1198 struct pipe_context *pipe = ctx->pipe; 1199 struct pipe_screen *screen = pipe->screen; 1200 uint dstLevel, zslice = 0; 1201 1202 assert(pt->block.width == 1); 1203 assert(pt->block.height == 1); 1204 1205 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1206 const uint srcLevel = dstLevel - 1; 1207 struct pipe_transfer *srcTrans, *dstTrans; 1208 ubyte *srcMap, *dstMap; 1209 1210 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, 1211 PIPE_TRANSFER_READ, 0, 0, 1212 pt->width[srcLevel], 1213 pt->height[srcLevel]); 1214 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, 1215 PIPE_TRANSFER_WRITE, 0, 0, 1216 pt->width[dstLevel], 1217 pt->height[dstLevel]); 1218 1219 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); 1220 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); 1221 1222 reduce_3d(pt->format, 1223 srcTrans->width, srcTrans->height, 1224 srcTrans->stride, srcMap, 1225 dstTrans->width, dstTrans->height, 1226 dstTrans->stride, dstMap); 1227 1228 screen->transfer_unmap(screen, srcTrans); 1229 screen->transfer_unmap(screen, dstTrans); 1230 1231 screen->tex_transfer_release(screen, &srcTrans); 1232 screen->tex_transfer_release(screen, &dstTrans); 1233 } 1234#else 1235 (void) reduce_3d; 1236#endif 1237} 1238 1239 1240static void 1241fallback_gen_mipmap(struct gen_mipmap_state *ctx, 1242 struct pipe_texture *pt, 1243 uint face, uint baseLevel, uint lastLevel) 1244{ 1245 switch (pt->target) { 1246 case PIPE_TEXTURE_1D: 1247 make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel); 1248 break; 1249 case PIPE_TEXTURE_2D: 1250 case PIPE_TEXTURE_CUBE: 1251 make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel); 1252 break; 1253 case PIPE_TEXTURE_3D: 1254 make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel); 1255 break; 1256 default: 1257 assert(0); 1258 } 1259} 1260 1261 1262/** 1263 * Create a mipmap generation context. 1264 * The idea is to create one of these and re-use it each time we need to 1265 * generate a mipmap. 1266 */ 1267struct gen_mipmap_state * 1268util_create_gen_mipmap(struct pipe_context *pipe, 1269 struct cso_context *cso) 1270{ 1271 struct gen_mipmap_state *ctx; 1272 uint i; 1273 1274 ctx = CALLOC_STRUCT(gen_mipmap_state); 1275 if (!ctx) 1276 return NULL; 1277 1278 ctx->pipe = pipe; 1279 ctx->cso = cso; 1280 1281 /* disabled blending/masking */ 1282 memset(&ctx->blend, 0, sizeof(ctx->blend)); 1283 ctx->blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; 1284 ctx->blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; 1285 ctx->blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; 1286 ctx->blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; 1287 ctx->blend.colormask = PIPE_MASK_RGBA; 1288 1289 /* no-op depth/stencil/alpha */ 1290 memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil)); 1291 1292 /* rasterizer */ 1293 memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer)); 1294 ctx->rasterizer.front_winding = PIPE_WINDING_CW; 1295 ctx->rasterizer.cull_mode = PIPE_WINDING_NONE; 1296 ctx->rasterizer.bypass_clipping = 1; 1297 /*ctx->rasterizer.bypass_vs = 1;*/ 1298 ctx->rasterizer.gl_rasterization_rules = 1; 1299 1300 /* sampler state */ 1301 memset(&ctx->sampler, 0, sizeof(ctx->sampler)); 1302 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1303 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1304 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1305 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST; 1306 ctx->sampler.normalized_coords = 1; 1307 1308 /* viewport state (identity, verts are in wincoords) */ 1309 ctx->viewport.scale[0] = 1.0; 1310 ctx->viewport.scale[1] = 1.0; 1311 ctx->viewport.scale[2] = 1.0; 1312 ctx->viewport.scale[3] = 1.0; 1313 ctx->viewport.translate[0] = 0.0; 1314 ctx->viewport.translate[1] = 0.0; 1315 ctx->viewport.translate[2] = 0.0; 1316 ctx->viewport.translate[3] = 0.0; 1317 1318 /* vertex shader */ 1319 { 1320 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION, 1321 TGSI_SEMANTIC_GENERIC }; 1322 const uint semantic_indexes[] = { 0, 0 }; 1323 ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names, 1324 semantic_indexes, 1325 &ctx->vert_shader); 1326 } 1327 1328 /* fragment shader */ 1329 ctx->fs = util_make_fragment_tex_shader(pipe, &ctx->frag_shader); 1330 1331 /* vertex data that doesn't change */ 1332 for (i = 0; i < 4; i++) { 1333 ctx->vertices[i][0][2] = 0.0f; /* z */ 1334 ctx->vertices[i][0][3] = 1.0f; /* w */ 1335 ctx->vertices[i][1][2] = 0.0f; /* r */ 1336 ctx->vertices[i][1][3] = 1.0f; /* q */ 1337 } 1338 1339 /* Note: the actual vertex buffer is allocated as needed below */ 1340 1341 return ctx; 1342} 1343 1344 1345/** 1346 * Get next "slot" of vertex space in the vertex buffer. 1347 * We're allocating one large vertex buffer and using it piece by piece. 1348 */ 1349static unsigned 1350get_next_slot(struct gen_mipmap_state *ctx) 1351{ 1352 const unsigned max_slots = 4096 / sizeof ctx->vertices; 1353 1354 if (ctx->vbuf_slot >= max_slots) 1355 util_gen_mipmap_flush( ctx ); 1356 1357 if (!ctx->vbuf) { 1358 ctx->vbuf = pipe_buffer_create(ctx->pipe->screen, 1359 32, 1360 PIPE_BUFFER_USAGE_VERTEX, 1361 max_slots * sizeof ctx->vertices); 1362 } 1363 1364 return ctx->vbuf_slot++ * sizeof ctx->vertices; 1365} 1366 1367 1368static unsigned 1369set_vertex_data(struct gen_mipmap_state *ctx, float width, float height) 1370{ 1371 void *buf; 1372 unsigned offset; 1373 1374 ctx->vertices[0][0][0] = 0.0f; /*x*/ 1375 ctx->vertices[0][0][1] = 0.0f; /*y*/ 1376 ctx->vertices[0][1][0] = 0.0f; /*s*/ 1377 ctx->vertices[0][1][1] = 0.0f; /*t*/ 1378 1379 ctx->vertices[1][0][0] = width; 1380 ctx->vertices[1][0][1] = 0.0f; 1381 ctx->vertices[1][1][0] = 1.0f; 1382 ctx->vertices[1][1][1] = 0.0f; 1383 1384 ctx->vertices[2][0][0] = width; 1385 ctx->vertices[2][0][1] = height; 1386 ctx->vertices[2][1][0] = 1.0f; 1387 ctx->vertices[2][1][1] = 1.0f; 1388 1389 ctx->vertices[3][0][0] = 0.0f; 1390 ctx->vertices[3][0][1] = height; 1391 ctx->vertices[3][1][0] = 0.0f; 1392 ctx->vertices[3][1][1] = 1.0f; 1393 1394 offset = get_next_slot( ctx ); 1395 1396 buf = pipe_buffer_map(ctx->pipe->screen, ctx->vbuf, 1397 PIPE_BUFFER_USAGE_CPU_WRITE); 1398 1399 memcpy((char *)buf + offset, ctx->vertices, sizeof(ctx->vertices)); 1400 1401 pipe_buffer_unmap(ctx->pipe->screen, ctx->vbuf); 1402 1403 return offset; 1404} 1405 1406 1407 1408/** 1409 * Destroy a mipmap generation context 1410 */ 1411void 1412util_destroy_gen_mipmap(struct gen_mipmap_state *ctx) 1413{ 1414 struct pipe_context *pipe = ctx->pipe; 1415 1416 pipe->delete_vs_state(pipe, ctx->vs); 1417 pipe->delete_fs_state(pipe, ctx->fs); 1418 1419 FREE((void*) ctx->vert_shader.tokens); 1420 FREE((void*) ctx->frag_shader.tokens); 1421 1422 pipe_buffer_reference(pipe->screen, &ctx->vbuf, NULL); 1423 1424 FREE(ctx); 1425} 1426 1427 1428 1429/* Release vertex buffer at end of frame to avoid synchronous 1430 * rendering. 1431 */ 1432void util_gen_mipmap_flush( struct gen_mipmap_state *ctx ) 1433{ 1434 pipe_buffer_reference(ctx->pipe->screen, &ctx->vbuf, NULL); 1435 ctx->vbuf_slot = 0; 1436} 1437 1438 1439/** 1440 * Generate mipmap images. It's assumed all needed texture memory is 1441 * already allocated. 1442 * 1443 * \param pt the texture to generate mipmap levels for 1444 * \param face which cube face to generate mipmaps for (0 for non-cube maps) 1445 * \param baseLevel the first mipmap level to use as a src 1446 * \param lastLevel the last mipmap level to generate 1447 * \param filter the minification filter used to generate mipmap levels with 1448 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST 1449 */ 1450void 1451util_gen_mipmap(struct gen_mipmap_state *ctx, 1452 struct pipe_texture *pt, 1453 uint face, uint baseLevel, uint lastLevel, uint filter) 1454{ 1455 struct pipe_context *pipe = ctx->pipe; 1456 struct pipe_screen *screen = pipe->screen; 1457 struct pipe_framebuffer_state fb; 1458 uint dstLevel; 1459 uint zslice = 0; 1460 uint offset; 1461 1462 /* check if we can render in the texture's format */ 1463 if (!screen->is_format_supported(screen, pt->format, PIPE_TEXTURE_2D, 1464 PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) { 1465 fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel); 1466 return; 1467 } 1468 1469 /* save state (restored below) */ 1470 cso_save_blend(ctx->cso); 1471 cso_save_depth_stencil_alpha(ctx->cso); 1472 cso_save_rasterizer(ctx->cso); 1473 cso_save_samplers(ctx->cso); 1474 cso_save_sampler_textures(ctx->cso); 1475 cso_save_framebuffer(ctx->cso); 1476 cso_save_fragment_shader(ctx->cso); 1477 cso_save_vertex_shader(ctx->cso); 1478 cso_save_viewport(ctx->cso); 1479 1480 /* bind our state */ 1481 cso_set_blend(ctx->cso, &ctx->blend); 1482 cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil); 1483 cso_set_rasterizer(ctx->cso, &ctx->rasterizer); 1484 cso_set_viewport(ctx->cso, &ctx->viewport); 1485 1486 cso_set_fragment_shader_handle(ctx->cso, ctx->fs); 1487 cso_set_vertex_shader_handle(ctx->cso, ctx->vs); 1488 1489 /* init framebuffer state */ 1490 memset(&fb, 0, sizeof(fb)); 1491 fb.nr_cbufs = 1; 1492 1493 /* set min/mag to same filter for faster sw speed */ 1494 ctx->sampler.mag_img_filter = filter; 1495 ctx->sampler.min_img_filter = filter; 1496 1497 /* 1498 * XXX for small mipmap levels, it may be faster to use the software 1499 * fallback path... 1500 */ 1501 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1502 const uint srcLevel = dstLevel - 1; 1503 1504 struct pipe_surface *surf = 1505 screen->get_tex_surface(screen, pt, face, dstLevel, zslice, 1506 PIPE_BUFFER_USAGE_GPU_WRITE); 1507 1508 /* 1509 * Setup framebuffer / dest surface 1510 */ 1511 fb.cbufs[0] = surf; 1512 fb.width = pt->width[dstLevel]; 1513 fb.height = pt->height[dstLevel]; 1514 cso_set_framebuffer(ctx->cso, &fb); 1515 1516 /* 1517 * Setup sampler state 1518 * Note: we should only have to set the min/max LOD clamps to ensure 1519 * we grab texels from the right mipmap level. But some hardware 1520 * has trouble with min clamping so we also set the lod_bias to 1521 * try to work around that. 1522 */ 1523 ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel; 1524 ctx->sampler.lod_bias = (float) srcLevel; 1525 cso_single_sampler(ctx->cso, 0, &ctx->sampler); 1526 cso_single_sampler_done(ctx->cso); 1527 1528 cso_set_sampler_textures(ctx->cso, 1, &pt); 1529 1530 /* quad coords in window coords (bypassing clipping, viewport mapping) */ 1531 offset = set_vertex_data(ctx, 1532 (float) pt->width[dstLevel], 1533 (float) pt->height[dstLevel]); 1534 1535 util_draw_vertex_buffer(ctx->pipe, 1536 ctx->vbuf, 1537 offset, 1538 PIPE_PRIM_TRIANGLE_FAN, 1539 4, /* verts */ 1540 2); /* attribs/vert */ 1541 1542 pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); 1543 1544 /* need to signal that the texture has changed _after_ rendering to it */ 1545 pipe_surface_reference( &surf, NULL ); 1546 } 1547 1548 /* restore state we changed */ 1549 cso_restore_blend(ctx->cso); 1550 cso_restore_depth_stencil_alpha(ctx->cso); 1551 cso_restore_rasterizer(ctx->cso); 1552 cso_restore_samplers(ctx->cso); 1553 cso_restore_sampler_textures(ctx->cso); 1554 cso_restore_framebuffer(ctx->cso); 1555 cso_restore_fragment_shader(ctx->cso); 1556 cso_restore_vertex_shader(ctx->cso); 1557 cso_restore_viewport(ctx->cso); 1558} 1559