u_gen_mipmap.c revision ca8a91ff7eb7e3fb4595763ea71f427b97e426c6
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 "util/u_inlines.h" 41#include "pipe/p_shader_tokens.h" 42#include "pipe/p_state.h" 43 44#include "util/u_format.h" 45#include "util/u_memory.h" 46#include "util/u_draw_quad.h" 47#include "util/u_gen_mipmap.h" 48#include "util/u_simple_shaders.h" 49#include "util/u_math.h" 50#include "util/u_texture.h" 51#include "util/u_half.h" 52#include "util/u_surface.h" 53 54#include "cso_cache/cso_context.h" 55 56 57struct gen_mipmap_state 58{ 59 struct pipe_context *pipe; 60 struct cso_context *cso; 61 62 struct pipe_blend_state blend; 63 struct pipe_depth_stencil_alpha_state depthstencil; 64 struct pipe_rasterizer_state rasterizer; 65 struct pipe_sampler_state sampler; 66 struct pipe_clip_state clip; 67 struct pipe_vertex_element velem[2]; 68 69 void *vs; 70 void *fs1d, *fs2d, *fs3d, *fsCube, *fs1da, *fs2da; 71 72 struct pipe_resource *vbuf; /**< quad vertices */ 73 unsigned vbuf_slot; 74 75 float vertices[4][2][4]; /**< vertex/texcoords for quad */ 76}; 77 78 79 80enum dtype 81{ 82 DTYPE_UBYTE, 83 DTYPE_UBYTE_3_3_2, 84 DTYPE_USHORT, 85 DTYPE_USHORT_4_4_4_4, 86 DTYPE_USHORT_5_6_5, 87 DTYPE_USHORT_1_5_5_5_REV, 88 DTYPE_UINT, 89 DTYPE_FLOAT, 90 DTYPE_HALF_FLOAT 91}; 92 93 94typedef uint16_t half_float; 95 96 97/** 98 * \name Support macros for do_row and do_row_3d 99 * 100 * The macro madness is here for two reasons. First, it compacts the code 101 * slightly. Second, it makes it much easier to adjust the specifics of the 102 * filter to tune the rounding characteristics. 103 */ 104/*@{*/ 105#define DECLARE_ROW_POINTERS(t, e) \ 106 const t(*rowA)[e] = (const t(*)[e]) srcRowA; \ 107 const t(*rowB)[e] = (const t(*)[e]) srcRowB; \ 108 const t(*rowC)[e] = (const t(*)[e]) srcRowC; \ 109 const t(*rowD)[e] = (const t(*)[e]) srcRowD; \ 110 t(*dst)[e] = (t(*)[e]) dstRow 111 112#define DECLARE_ROW_POINTERS0(t) \ 113 const t *rowA = (const t *) srcRowA; \ 114 const t *rowB = (const t *) srcRowB; \ 115 const t *rowC = (const t *) srcRowC; \ 116 const t *rowD = (const t *) srcRowD; \ 117 t *dst = (t *) dstRow 118 119#define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \ 120 ((unsigned) Aj + (unsigned) Ak \ 121 + (unsigned) Bj + (unsigned) Bk \ 122 + (unsigned) Cj + (unsigned) Ck \ 123 + (unsigned) Dj + (unsigned) Dk \ 124 + 4) >> 3 125 126#define FILTER_3D(e) \ 127 do { \ 128 dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \ 129 rowB[j][e], rowB[k][e], \ 130 rowC[j][e], rowC[k][e], \ 131 rowD[j][e], rowD[k][e]); \ 132 } while(0) 133 134#define FILTER_F_3D(e) \ 135 do { \ 136 dst[i][e] = (rowA[j][e] + rowA[k][e] \ 137 + rowB[j][e] + rowB[k][e] \ 138 + rowC[j][e] + rowC[k][e] \ 139 + rowD[j][e] + rowD[k][e]) * 0.125F; \ 140 } while(0) 141 142#define FILTER_HF_3D(e) \ 143 do { \ 144 const float aj = util_half_to_float(rowA[j][e]); \ 145 const float ak = util_half_to_float(rowA[k][e]); \ 146 const float bj = util_half_to_float(rowB[j][e]); \ 147 const float bk = util_half_to_float(rowB[k][e]); \ 148 const float cj = util_half_to_float(rowC[j][e]); \ 149 const float ck = util_half_to_float(rowC[k][e]); \ 150 const float dj = util_half_to_float(rowD[j][e]); \ 151 const float dk = util_half_to_float(rowD[k][e]); \ 152 dst[i][e] = util_float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \ 153 * 0.125F); \ 154 } while(0) 155/*@}*/ 156 157 158/** 159 * Average together two rows of a source image to produce a single new 160 * row in the dest image. It's legal for the two source rows to point 161 * to the same data. The source width must be equal to either the 162 * dest width or two times the dest width. 163 * \param datatype GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc. 164 * \param comps number of components per pixel (1..4) 165 */ 166static void 167do_row(enum dtype datatype, uint comps, int srcWidth, 168 const void *srcRowA, const void *srcRowB, 169 int dstWidth, void *dstRow) 170{ 171 const uint k0 = (srcWidth == dstWidth) ? 0 : 1; 172 const uint colStride = (srcWidth == dstWidth) ? 1 : 2; 173 174 assert(comps >= 1); 175 assert(comps <= 4); 176 177 /* This assertion is no longer valid with non-power-of-2 textures 178 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth); 179 */ 180 181 if (datatype == DTYPE_UBYTE && comps == 4) { 182 uint i, j, k; 183 const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA; 184 const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB; 185 ubyte(*dst)[4] = (ubyte(*)[4]) dstRow; 186 for (i = j = 0, k = k0; i < (uint) dstWidth; 187 i++, j += colStride, k += colStride) { 188 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 189 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 190 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 191 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 192 } 193 } 194 else if (datatype == DTYPE_UBYTE && comps == 3) { 195 uint i, j, k; 196 const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA; 197 const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB; 198 ubyte(*dst)[3] = (ubyte(*)[3]) dstRow; 199 for (i = j = 0, k = k0; i < (uint) dstWidth; 200 i++, j += colStride, k += colStride) { 201 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 202 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 203 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 204 } 205 } 206 else if (datatype == DTYPE_UBYTE && comps == 2) { 207 uint i, j, k; 208 const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA; 209 const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB; 210 ubyte(*dst)[2] = (ubyte(*)[2]) dstRow; 211 for (i = j = 0, k = k0; i < (uint) dstWidth; 212 i++, j += colStride, k += colStride) { 213 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2; 214 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2; 215 } 216 } 217 else if (datatype == DTYPE_UBYTE && comps == 1) { 218 uint i, j, k; 219 const ubyte *rowA = (const ubyte *) srcRowA; 220 const ubyte *rowB = (const ubyte *) srcRowB; 221 ubyte *dst = (ubyte *) dstRow; 222 for (i = j = 0, k = k0; i < (uint) dstWidth; 223 i++, j += colStride, k += colStride) { 224 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2; 225 } 226 } 227 228 else if (datatype == DTYPE_USHORT && comps == 4) { 229 uint i, j, k; 230 const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA; 231 const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB; 232 ushort(*dst)[4] = (ushort(*)[4]) dstRow; 233 for (i = j = 0, k = k0; i < (uint) dstWidth; 234 i++, j += colStride, k += colStride) { 235 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 236 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 237 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 238 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 239 } 240 } 241 else if (datatype == DTYPE_USHORT && comps == 3) { 242 uint i, j, k; 243 const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA; 244 const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB; 245 ushort(*dst)[3] = (ushort(*)[3]) dstRow; 246 for (i = j = 0, k = k0; i < (uint) dstWidth; 247 i++, j += colStride, k += colStride) { 248 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 249 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 250 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 251 } 252 } 253 else if (datatype == DTYPE_USHORT && comps == 2) { 254 uint i, j, k; 255 const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA; 256 const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB; 257 ushort(*dst)[2] = (ushort(*)[2]) dstRow; 258 for (i = j = 0, k = k0; i < (uint) dstWidth; 259 i++, j += colStride, k += colStride) { 260 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 261 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 262 } 263 } 264 else if (datatype == DTYPE_USHORT && comps == 1) { 265 uint i, j, k; 266 const ushort *rowA = (const ushort *) srcRowA; 267 const ushort *rowB = (const ushort *) srcRowB; 268 ushort *dst = (ushort *) dstRow; 269 for (i = j = 0, k = k0; i < (uint) dstWidth; 270 i++, j += colStride, k += colStride) { 271 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4; 272 } 273 } 274 275 else if (datatype == DTYPE_FLOAT && comps == 4) { 276 uint i, j, k; 277 const float(*rowA)[4] = (const float(*)[4]) srcRowA; 278 const float(*rowB)[4] = (const float(*)[4]) srcRowB; 279 float(*dst)[4] = (float(*)[4]) dstRow; 280 for (i = j = 0, k = k0; i < (uint) dstWidth; 281 i++, j += colStride, k += colStride) { 282 dst[i][0] = (rowA[j][0] + rowA[k][0] + 283 rowB[j][0] + rowB[k][0]) * 0.25F; 284 dst[i][1] = (rowA[j][1] + rowA[k][1] + 285 rowB[j][1] + rowB[k][1]) * 0.25F; 286 dst[i][2] = (rowA[j][2] + rowA[k][2] + 287 rowB[j][2] + rowB[k][2]) * 0.25F; 288 dst[i][3] = (rowA[j][3] + rowA[k][3] + 289 rowB[j][3] + rowB[k][3]) * 0.25F; 290 } 291 } 292 else if (datatype == DTYPE_FLOAT && comps == 3) { 293 uint i, j, k; 294 const float(*rowA)[3] = (const float(*)[3]) srcRowA; 295 const float(*rowB)[3] = (const float(*)[3]) srcRowB; 296 float(*dst)[3] = (float(*)[3]) dstRow; 297 for (i = j = 0, k = k0; i < (uint) dstWidth; 298 i++, j += colStride, k += colStride) { 299 dst[i][0] = (rowA[j][0] + rowA[k][0] + 300 rowB[j][0] + rowB[k][0]) * 0.25F; 301 dst[i][1] = (rowA[j][1] + rowA[k][1] + 302 rowB[j][1] + rowB[k][1]) * 0.25F; 303 dst[i][2] = (rowA[j][2] + rowA[k][2] + 304 rowB[j][2] + rowB[k][2]) * 0.25F; 305 } 306 } 307 else if (datatype == DTYPE_FLOAT && comps == 2) { 308 uint i, j, k; 309 const float(*rowA)[2] = (const float(*)[2]) srcRowA; 310 const float(*rowB)[2] = (const float(*)[2]) srcRowB; 311 float(*dst)[2] = (float(*)[2]) dstRow; 312 for (i = j = 0, k = k0; i < (uint) dstWidth; 313 i++, j += colStride, k += colStride) { 314 dst[i][0] = (rowA[j][0] + rowA[k][0] + 315 rowB[j][0] + rowB[k][0]) * 0.25F; 316 dst[i][1] = (rowA[j][1] + rowA[k][1] + 317 rowB[j][1] + rowB[k][1]) * 0.25F; 318 } 319 } 320 else if (datatype == DTYPE_FLOAT && comps == 1) { 321 uint i, j, k; 322 const float *rowA = (const float *) srcRowA; 323 const float *rowB = (const float *) srcRowB; 324 float *dst = (float *) dstRow; 325 for (i = j = 0, k = k0; i < (uint) dstWidth; 326 i++, j += colStride, k += colStride) { 327 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F; 328 } 329 } 330 331 else if (datatype == DTYPE_HALF_FLOAT && comps == 4) { 332 uint i, j, k, comp; 333 const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA; 334 const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB; 335 half_float(*dst)[4] = (half_float(*)[4]) dstRow; 336 for (i = j = 0, k = k0; i < (uint) dstWidth; 337 i++, j += colStride, k += colStride) { 338 for (comp = 0; comp < 4; comp++) { 339 float aj, ak, bj, bk; 340 aj = util_half_to_float(rowA[j][comp]); 341 ak = util_half_to_float(rowA[k][comp]); 342 bj = util_half_to_float(rowB[j][comp]); 343 bk = util_half_to_float(rowB[k][comp]); 344 dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F); 345 } 346 } 347 } 348 else if (datatype == DTYPE_HALF_FLOAT && comps == 3) { 349 uint i, j, k, comp; 350 const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA; 351 const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB; 352 half_float(*dst)[3] = (half_float(*)[3]) dstRow; 353 for (i = j = 0, k = k0; i < (uint) dstWidth; 354 i++, j += colStride, k += colStride) { 355 for (comp = 0; comp < 3; comp++) { 356 float aj, ak, bj, bk; 357 aj = util_half_to_float(rowA[j][comp]); 358 ak = util_half_to_float(rowA[k][comp]); 359 bj = util_half_to_float(rowB[j][comp]); 360 bk = util_half_to_float(rowB[k][comp]); 361 dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F); 362 } 363 } 364 } 365 else if (datatype == DTYPE_HALF_FLOAT && comps == 2) { 366 uint i, j, k, comp; 367 const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA; 368 const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB; 369 half_float(*dst)[2] = (half_float(*)[2]) dstRow; 370 for (i = j = 0, k = k0; i < (uint) dstWidth; 371 i++, j += colStride, k += colStride) { 372 for (comp = 0; comp < 2; comp++) { 373 float aj, ak, bj, bk; 374 aj = util_half_to_float(rowA[j][comp]); 375 ak = util_half_to_float(rowA[k][comp]); 376 bj = util_half_to_float(rowB[j][comp]); 377 bk = util_half_to_float(rowB[k][comp]); 378 dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F); 379 } 380 } 381 } 382 else if (datatype == DTYPE_HALF_FLOAT && comps == 1) { 383 uint i, j, k; 384 const half_float *rowA = (const half_float *) srcRowA; 385 const half_float *rowB = (const half_float *) srcRowB; 386 half_float *dst = (half_float *) dstRow; 387 for (i = j = 0, k = k0; i < (uint) dstWidth; 388 i++, j += colStride, k += colStride) { 389 float aj, ak, bj, bk; 390 aj = util_half_to_float(rowA[j]); 391 ak = util_half_to_float(rowA[k]); 392 bj = util_half_to_float(rowB[j]); 393 bk = util_half_to_float(rowB[k]); 394 dst[i] = util_float_to_half((aj + ak + bj + bk) * 0.25F); 395 } 396 } 397 398 else if (datatype == DTYPE_UINT && comps == 1) { 399 uint i, j, k; 400 const uint *rowA = (const uint *) srcRowA; 401 const uint *rowB = (const uint *) srcRowB; 402 uint *dst = (uint *) dstRow; 403 for (i = j = 0, k = k0; i < (uint) dstWidth; 404 i++, j += colStride, k += colStride) { 405 dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4; 406 } 407 } 408 409 else if (datatype == DTYPE_USHORT_5_6_5 && comps == 3) { 410 uint i, j, k; 411 const ushort *rowA = (const ushort *) srcRowA; 412 const ushort *rowB = (const ushort *) srcRowB; 413 ushort *dst = (ushort *) dstRow; 414 for (i = j = 0, k = k0; i < (uint) dstWidth; 415 i++, j += colStride, k += colStride) { 416 const int rowAr0 = rowA[j] & 0x1f; 417 const int rowAr1 = rowA[k] & 0x1f; 418 const int rowBr0 = rowB[j] & 0x1f; 419 const int rowBr1 = rowB[k] & 0x1f; 420 const int rowAg0 = (rowA[j] >> 5) & 0x3f; 421 const int rowAg1 = (rowA[k] >> 5) & 0x3f; 422 const int rowBg0 = (rowB[j] >> 5) & 0x3f; 423 const int rowBg1 = (rowB[k] >> 5) & 0x3f; 424 const int rowAb0 = (rowA[j] >> 11) & 0x1f; 425 const int rowAb1 = (rowA[k] >> 11) & 0x1f; 426 const int rowBb0 = (rowB[j] >> 11) & 0x1f; 427 const int rowBb1 = (rowB[k] >> 11) & 0x1f; 428 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 429 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 430 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 431 dst[i] = (blue << 11) | (green << 5) | red; 432 } 433 } 434 else if (datatype == DTYPE_USHORT_4_4_4_4 && comps == 4) { 435 uint i, j, k; 436 const ushort *rowA = (const ushort *) srcRowA; 437 const ushort *rowB = (const ushort *) srcRowB; 438 ushort *dst = (ushort *) dstRow; 439 for (i = j = 0, k = k0; i < (uint) dstWidth; 440 i++, j += colStride, k += colStride) { 441 const int rowAr0 = rowA[j] & 0xf; 442 const int rowAr1 = rowA[k] & 0xf; 443 const int rowBr0 = rowB[j] & 0xf; 444 const int rowBr1 = rowB[k] & 0xf; 445 const int rowAg0 = (rowA[j] >> 4) & 0xf; 446 const int rowAg1 = (rowA[k] >> 4) & 0xf; 447 const int rowBg0 = (rowB[j] >> 4) & 0xf; 448 const int rowBg1 = (rowB[k] >> 4) & 0xf; 449 const int rowAb0 = (rowA[j] >> 8) & 0xf; 450 const int rowAb1 = (rowA[k] >> 8) & 0xf; 451 const int rowBb0 = (rowB[j] >> 8) & 0xf; 452 const int rowBb1 = (rowB[k] >> 8) & 0xf; 453 const int rowAa0 = (rowA[j] >> 12) & 0xf; 454 const int rowAa1 = (rowA[k] >> 12) & 0xf; 455 const int rowBa0 = (rowB[j] >> 12) & 0xf; 456 const int rowBa1 = (rowB[k] >> 12) & 0xf; 457 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 458 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 459 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 460 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 461 dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red; 462 } 463 } 464 else if (datatype == DTYPE_USHORT_1_5_5_5_REV && comps == 4) { 465 uint i, j, k; 466 const ushort *rowA = (const ushort *) srcRowA; 467 const ushort *rowB = (const ushort *) srcRowB; 468 ushort *dst = (ushort *) dstRow; 469 for (i = j = 0, k = k0; i < (uint) dstWidth; 470 i++, j += colStride, k += colStride) { 471 const int rowAr0 = rowA[j] & 0x1f; 472 const int rowAr1 = rowA[k] & 0x1f; 473 const int rowBr0 = rowB[j] & 0x1f; 474 const int rowBr1 = rowB[k] & 0x1f; 475 const int rowAg0 = (rowA[j] >> 5) & 0x1f; 476 const int rowAg1 = (rowA[k] >> 5) & 0x1f; 477 const int rowBg0 = (rowB[j] >> 5) & 0x1f; 478 const int rowBg1 = (rowB[k] >> 5) & 0x1f; 479 const int rowAb0 = (rowA[j] >> 10) & 0x1f; 480 const int rowAb1 = (rowA[k] >> 10) & 0x1f; 481 const int rowBb0 = (rowB[j] >> 10) & 0x1f; 482 const int rowBb1 = (rowB[k] >> 10) & 0x1f; 483 const int rowAa0 = (rowA[j] >> 15) & 0x1; 484 const int rowAa1 = (rowA[k] >> 15) & 0x1; 485 const int rowBa0 = (rowB[j] >> 15) & 0x1; 486 const int rowBa1 = (rowB[k] >> 15) & 0x1; 487 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 488 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 489 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 490 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 491 dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red; 492 } 493 } 494 else if (datatype == DTYPE_UBYTE_3_3_2 && comps == 3) { 495 uint i, j, k; 496 const ubyte *rowA = (const ubyte *) srcRowA; 497 const ubyte *rowB = (const ubyte *) srcRowB; 498 ubyte *dst = (ubyte *) dstRow; 499 for (i = j = 0, k = k0; i < (uint) dstWidth; 500 i++, j += colStride, k += colStride) { 501 const int rowAr0 = rowA[j] & 0x3; 502 const int rowAr1 = rowA[k] & 0x3; 503 const int rowBr0 = rowB[j] & 0x3; 504 const int rowBr1 = rowB[k] & 0x3; 505 const int rowAg0 = (rowA[j] >> 2) & 0x7; 506 const int rowAg1 = (rowA[k] >> 2) & 0x7; 507 const int rowBg0 = (rowB[j] >> 2) & 0x7; 508 const int rowBg1 = (rowB[k] >> 2) & 0x7; 509 const int rowAb0 = (rowA[j] >> 5) & 0x7; 510 const int rowAb1 = (rowA[k] >> 5) & 0x7; 511 const int rowBb0 = (rowB[j] >> 5) & 0x7; 512 const int rowBb1 = (rowB[k] >> 5) & 0x7; 513 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 514 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 515 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 516 dst[i] = (blue << 5) | (green << 2) | red; 517 } 518 } 519 else { 520 debug_printf("bad format in do_row()"); 521 } 522} 523 524 525/** 526 * Average together four rows of a source image to produce a single new 527 * row in the dest image. It's legal for the two source rows to point 528 * to the same data. The source width must be equal to either the 529 * dest width or two times the dest width. 530 * 531 * \param datatype GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT, 532 * \c GL_FLOAT, etc. 533 * \param comps number of components per pixel (1..4) 534 * \param srcWidth Width of a row in the source data 535 * \param srcRowA Pointer to one of the rows of source data 536 * \param srcRowB Pointer to one of the rows of source data 537 * \param srcRowC Pointer to one of the rows of source data 538 * \param srcRowD Pointer to one of the rows of source data 539 * \param dstWidth Width of a row in the destination data 540 * \param srcRowA Pointer to the row of destination data 541 */ 542static void 543do_row_3D(enum dtype datatype, uint comps, int srcWidth, 544 const void *srcRowA, const void *srcRowB, 545 const void *srcRowC, const void *srcRowD, 546 int dstWidth, void *dstRow) 547{ 548 const uint k0 = (srcWidth == dstWidth) ? 0 : 1; 549 const uint colStride = (srcWidth == dstWidth) ? 1 : 2; 550 uint i, j, k; 551 552 assert(comps >= 1); 553 assert(comps <= 4); 554 555 if ((datatype == DTYPE_UBYTE) && (comps == 4)) { 556 DECLARE_ROW_POINTERS(ubyte, 4); 557 558 for (i = j = 0, k = k0; i < (uint) dstWidth; 559 i++, j += colStride, k += colStride) { 560 FILTER_3D(0); 561 FILTER_3D(1); 562 FILTER_3D(2); 563 FILTER_3D(3); 564 } 565 } 566 else if ((datatype == DTYPE_UBYTE) && (comps == 3)) { 567 DECLARE_ROW_POINTERS(ubyte, 3); 568 569 for (i = j = 0, k = k0; i < (uint) dstWidth; 570 i++, j += colStride, k += colStride) { 571 FILTER_3D(0); 572 FILTER_3D(1); 573 FILTER_3D(2); 574 } 575 } 576 else if ((datatype == DTYPE_UBYTE) && (comps == 2)) { 577 DECLARE_ROW_POINTERS(ubyte, 2); 578 579 for (i = j = 0, k = k0; i < (uint) dstWidth; 580 i++, j += colStride, k += colStride) { 581 FILTER_3D(0); 582 FILTER_3D(1); 583 } 584 } 585 else if ((datatype == DTYPE_UBYTE) && (comps == 1)) { 586 DECLARE_ROW_POINTERS(ubyte, 1); 587 588 for (i = j = 0, k = k0; i < (uint) dstWidth; 589 i++, j += colStride, k += colStride) { 590 FILTER_3D(0); 591 } 592 } 593 else if ((datatype == DTYPE_USHORT) && (comps == 4)) { 594 DECLARE_ROW_POINTERS(ushort, 4); 595 596 for (i = j = 0, k = k0; i < (uint) dstWidth; 597 i++, j += colStride, k += colStride) { 598 FILTER_3D(0); 599 FILTER_3D(1); 600 FILTER_3D(2); 601 FILTER_3D(3); 602 } 603 } 604 else if ((datatype == DTYPE_USHORT) && (comps == 3)) { 605 DECLARE_ROW_POINTERS(ushort, 3); 606 607 for (i = j = 0, k = k0; i < (uint) dstWidth; 608 i++, j += colStride, k += colStride) { 609 FILTER_3D(0); 610 FILTER_3D(1); 611 FILTER_3D(2); 612 } 613 } 614 else if ((datatype == DTYPE_USHORT) && (comps == 2)) { 615 DECLARE_ROW_POINTERS(ushort, 2); 616 617 for (i = j = 0, k = k0; i < (uint) dstWidth; 618 i++, j += colStride, k += colStride) { 619 FILTER_3D(0); 620 FILTER_3D(1); 621 } 622 } 623 else if ((datatype == DTYPE_USHORT) && (comps == 1)) { 624 DECLARE_ROW_POINTERS(ushort, 1); 625 626 for (i = j = 0, k = k0; i < (uint) dstWidth; 627 i++, j += colStride, k += colStride) { 628 FILTER_3D(0); 629 } 630 } 631 else if ((datatype == DTYPE_FLOAT) && (comps == 4)) { 632 DECLARE_ROW_POINTERS(float, 4); 633 634 for (i = j = 0, k = k0; i < (uint) dstWidth; 635 i++, j += colStride, k += colStride) { 636 FILTER_F_3D(0); 637 FILTER_F_3D(1); 638 FILTER_F_3D(2); 639 FILTER_F_3D(3); 640 } 641 } 642 else if ((datatype == DTYPE_FLOAT) && (comps == 3)) { 643 DECLARE_ROW_POINTERS(float, 3); 644 645 for (i = j = 0, k = k0; i < (uint) dstWidth; 646 i++, j += colStride, k += colStride) { 647 FILTER_F_3D(0); 648 FILTER_F_3D(1); 649 FILTER_F_3D(2); 650 } 651 } 652 else if ((datatype == DTYPE_FLOAT) && (comps == 2)) { 653 DECLARE_ROW_POINTERS(float, 2); 654 655 for (i = j = 0, k = k0; i < (uint) dstWidth; 656 i++, j += colStride, k += colStride) { 657 FILTER_F_3D(0); 658 FILTER_F_3D(1); 659 } 660 } 661 else if ((datatype == DTYPE_FLOAT) && (comps == 1)) { 662 DECLARE_ROW_POINTERS(float, 1); 663 664 for (i = j = 0, k = k0; i < (uint) dstWidth; 665 i++, j += colStride, k += colStride) { 666 FILTER_F_3D(0); 667 } 668 } 669 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 4)) { 670 DECLARE_ROW_POINTERS(half_float, 4); 671 672 for (i = j = 0, k = k0; i < (uint) dstWidth; 673 i++, j += colStride, k += colStride) { 674 FILTER_HF_3D(0); 675 FILTER_HF_3D(1); 676 FILTER_HF_3D(2); 677 FILTER_HF_3D(3); 678 } 679 } 680 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 3)) { 681 DECLARE_ROW_POINTERS(half_float, 4); 682 683 for (i = j = 0, k = k0; i < (uint) dstWidth; 684 i++, j += colStride, k += colStride) { 685 FILTER_HF_3D(0); 686 FILTER_HF_3D(1); 687 FILTER_HF_3D(2); 688 } 689 } 690 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 2)) { 691 DECLARE_ROW_POINTERS(half_float, 4); 692 693 for (i = j = 0, k = k0; i < (uint) dstWidth; 694 i++, j += colStride, k += colStride) { 695 FILTER_HF_3D(0); 696 FILTER_HF_3D(1); 697 } 698 } 699 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 1)) { 700 DECLARE_ROW_POINTERS(half_float, 4); 701 702 for (i = j = 0, k = k0; i < (uint) dstWidth; 703 i++, j += colStride, k += colStride) { 704 FILTER_HF_3D(0); 705 } 706 } 707 else if ((datatype == DTYPE_UINT) && (comps == 1)) { 708 const uint *rowA = (const uint *) srcRowA; 709 const uint *rowB = (const uint *) srcRowB; 710 const uint *rowC = (const uint *) srcRowC; 711 const uint *rowD = (const uint *) srcRowD; 712 float *dst = (float *) dstRow; 713 714 for (i = j = 0, k = k0; i < (uint) dstWidth; 715 i++, j += colStride, k += colStride) { 716 const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k]) 717 + ((uint64_t) rowB[j] + (uint64_t) rowB[k]) 718 + ((uint64_t) rowC[j] + (uint64_t) rowC[k]) 719 + ((uint64_t) rowD[j] + (uint64_t) rowD[k])); 720 dst[i] = (float)((double) tmp * 0.125); 721 } 722 } 723 else if ((datatype == DTYPE_USHORT_5_6_5) && (comps == 3)) { 724 DECLARE_ROW_POINTERS0(ushort); 725 726 for (i = j = 0, k = k0; i < (uint) dstWidth; 727 i++, j += colStride, k += colStride) { 728 const int rowAr0 = rowA[j] & 0x1f; 729 const int rowAr1 = rowA[k] & 0x1f; 730 const int rowBr0 = rowB[j] & 0x1f; 731 const int rowBr1 = rowB[k] & 0x1f; 732 const int rowCr0 = rowC[j] & 0x1f; 733 const int rowCr1 = rowC[k] & 0x1f; 734 const int rowDr0 = rowD[j] & 0x1f; 735 const int rowDr1 = rowD[k] & 0x1f; 736 const int rowAg0 = (rowA[j] >> 5) & 0x3f; 737 const int rowAg1 = (rowA[k] >> 5) & 0x3f; 738 const int rowBg0 = (rowB[j] >> 5) & 0x3f; 739 const int rowBg1 = (rowB[k] >> 5) & 0x3f; 740 const int rowCg0 = (rowC[j] >> 5) & 0x3f; 741 const int rowCg1 = (rowC[k] >> 5) & 0x3f; 742 const int rowDg0 = (rowD[j] >> 5) & 0x3f; 743 const int rowDg1 = (rowD[k] >> 5) & 0x3f; 744 const int rowAb0 = (rowA[j] >> 11) & 0x1f; 745 const int rowAb1 = (rowA[k] >> 11) & 0x1f; 746 const int rowBb0 = (rowB[j] >> 11) & 0x1f; 747 const int rowBb1 = (rowB[k] >> 11) & 0x1f; 748 const int rowCb0 = (rowC[j] >> 11) & 0x1f; 749 const int rowCb1 = (rowC[k] >> 11) & 0x1f; 750 const int rowDb0 = (rowD[j] >> 11) & 0x1f; 751 const int rowDb1 = (rowD[k] >> 11) & 0x1f; 752 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 753 rowCr0, rowCr1, rowDr0, rowDr1); 754 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 755 rowCg0, rowCg1, rowDg0, rowDg1); 756 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 757 rowCb0, rowCb1, rowDb0, rowDb1); 758 dst[i] = (b << 11) | (g << 5) | r; 759 } 760 } 761 else if ((datatype == DTYPE_USHORT_4_4_4_4) && (comps == 4)) { 762 DECLARE_ROW_POINTERS0(ushort); 763 764 for (i = j = 0, k = k0; i < (uint) dstWidth; 765 i++, j += colStride, k += colStride) { 766 const int rowAr0 = rowA[j] & 0xf; 767 const int rowAr1 = rowA[k] & 0xf; 768 const int rowBr0 = rowB[j] & 0xf; 769 const int rowBr1 = rowB[k] & 0xf; 770 const int rowCr0 = rowC[j] & 0xf; 771 const int rowCr1 = rowC[k] & 0xf; 772 const int rowDr0 = rowD[j] & 0xf; 773 const int rowDr1 = rowD[k] & 0xf; 774 const int rowAg0 = (rowA[j] >> 4) & 0xf; 775 const int rowAg1 = (rowA[k] >> 4) & 0xf; 776 const int rowBg0 = (rowB[j] >> 4) & 0xf; 777 const int rowBg1 = (rowB[k] >> 4) & 0xf; 778 const int rowCg0 = (rowC[j] >> 4) & 0xf; 779 const int rowCg1 = (rowC[k] >> 4) & 0xf; 780 const int rowDg0 = (rowD[j] >> 4) & 0xf; 781 const int rowDg1 = (rowD[k] >> 4) & 0xf; 782 const int rowAb0 = (rowA[j] >> 8) & 0xf; 783 const int rowAb1 = (rowA[k] >> 8) & 0xf; 784 const int rowBb0 = (rowB[j] >> 8) & 0xf; 785 const int rowBb1 = (rowB[k] >> 8) & 0xf; 786 const int rowCb0 = (rowC[j] >> 8) & 0xf; 787 const int rowCb1 = (rowC[k] >> 8) & 0xf; 788 const int rowDb0 = (rowD[j] >> 8) & 0xf; 789 const int rowDb1 = (rowD[k] >> 8) & 0xf; 790 const int rowAa0 = (rowA[j] >> 12) & 0xf; 791 const int rowAa1 = (rowA[k] >> 12) & 0xf; 792 const int rowBa0 = (rowB[j] >> 12) & 0xf; 793 const int rowBa1 = (rowB[k] >> 12) & 0xf; 794 const int rowCa0 = (rowC[j] >> 12) & 0xf; 795 const int rowCa1 = (rowC[k] >> 12) & 0xf; 796 const int rowDa0 = (rowD[j] >> 12) & 0xf; 797 const int rowDa1 = (rowD[k] >> 12) & 0xf; 798 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 799 rowCr0, rowCr1, rowDr0, rowDr1); 800 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 801 rowCg0, rowCg1, rowDg0, rowDg1); 802 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 803 rowCb0, rowCb1, rowDb0, rowDb1); 804 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1, 805 rowCa0, rowCa1, rowDa0, rowDa1); 806 807 dst[i] = (a << 12) | (b << 8) | (g << 4) | r; 808 } 809 } 810 else if ((datatype == DTYPE_USHORT_1_5_5_5_REV) && (comps == 4)) { 811 DECLARE_ROW_POINTERS0(ushort); 812 813 for (i = j = 0, k = k0; i < (uint) dstWidth; 814 i++, j += colStride, k += colStride) { 815 const int rowAr0 = rowA[j] & 0x1f; 816 const int rowAr1 = rowA[k] & 0x1f; 817 const int rowBr0 = rowB[j] & 0x1f; 818 const int rowBr1 = rowB[k] & 0x1f; 819 const int rowCr0 = rowC[j] & 0x1f; 820 const int rowCr1 = rowC[k] & 0x1f; 821 const int rowDr0 = rowD[j] & 0x1f; 822 const int rowDr1 = rowD[k] & 0x1f; 823 const int rowAg0 = (rowA[j] >> 5) & 0x1f; 824 const int rowAg1 = (rowA[k] >> 5) & 0x1f; 825 const int rowBg0 = (rowB[j] >> 5) & 0x1f; 826 const int rowBg1 = (rowB[k] >> 5) & 0x1f; 827 const int rowCg0 = (rowC[j] >> 5) & 0x1f; 828 const int rowCg1 = (rowC[k] >> 5) & 0x1f; 829 const int rowDg0 = (rowD[j] >> 5) & 0x1f; 830 const int rowDg1 = (rowD[k] >> 5) & 0x1f; 831 const int rowAb0 = (rowA[j] >> 10) & 0x1f; 832 const int rowAb1 = (rowA[k] >> 10) & 0x1f; 833 const int rowBb0 = (rowB[j] >> 10) & 0x1f; 834 const int rowBb1 = (rowB[k] >> 10) & 0x1f; 835 const int rowCb0 = (rowC[j] >> 10) & 0x1f; 836 const int rowCb1 = (rowC[k] >> 10) & 0x1f; 837 const int rowDb0 = (rowD[j] >> 10) & 0x1f; 838 const int rowDb1 = (rowD[k] >> 10) & 0x1f; 839 const int rowAa0 = (rowA[j] >> 15) & 0x1; 840 const int rowAa1 = (rowA[k] >> 15) & 0x1; 841 const int rowBa0 = (rowB[j] >> 15) & 0x1; 842 const int rowBa1 = (rowB[k] >> 15) & 0x1; 843 const int rowCa0 = (rowC[j] >> 15) & 0x1; 844 const int rowCa1 = (rowC[k] >> 15) & 0x1; 845 const int rowDa0 = (rowD[j] >> 15) & 0x1; 846 const int rowDa1 = (rowD[k] >> 15) & 0x1; 847 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 848 rowCr0, rowCr1, rowDr0, rowDr1); 849 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 850 rowCg0, rowCg1, rowDg0, rowDg1); 851 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 852 rowCb0, rowCb1, rowDb0, rowDb1); 853 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1, 854 rowCa0, rowCa1, rowDa0, rowDa1); 855 856 dst[i] = (a << 15) | (b << 10) | (g << 5) | r; 857 } 858 } 859 else if ((datatype == DTYPE_UBYTE_3_3_2) && (comps == 3)) { 860 DECLARE_ROW_POINTERS0(ushort); 861 862 for (i = j = 0, k = k0; i < (uint) dstWidth; 863 i++, j += colStride, k += colStride) { 864 const int rowAr0 = rowA[j] & 0x3; 865 const int rowAr1 = rowA[k] & 0x3; 866 const int rowBr0 = rowB[j] & 0x3; 867 const int rowBr1 = rowB[k] & 0x3; 868 const int rowCr0 = rowC[j] & 0x3; 869 const int rowCr1 = rowC[k] & 0x3; 870 const int rowDr0 = rowD[j] & 0x3; 871 const int rowDr1 = rowD[k] & 0x3; 872 const int rowAg0 = (rowA[j] >> 2) & 0x7; 873 const int rowAg1 = (rowA[k] >> 2) & 0x7; 874 const int rowBg0 = (rowB[j] >> 2) & 0x7; 875 const int rowBg1 = (rowB[k] >> 2) & 0x7; 876 const int rowCg0 = (rowC[j] >> 2) & 0x7; 877 const int rowCg1 = (rowC[k] >> 2) & 0x7; 878 const int rowDg0 = (rowD[j] >> 2) & 0x7; 879 const int rowDg1 = (rowD[k] >> 2) & 0x7; 880 const int rowAb0 = (rowA[j] >> 5) & 0x7; 881 const int rowAb1 = (rowA[k] >> 5) & 0x7; 882 const int rowBb0 = (rowB[j] >> 5) & 0x7; 883 const int rowBb1 = (rowB[k] >> 5) & 0x7; 884 const int rowCb0 = (rowC[j] >> 5) & 0x7; 885 const int rowCb1 = (rowC[k] >> 5) & 0x7; 886 const int rowDb0 = (rowD[j] >> 5) & 0x7; 887 const int rowDb1 = (rowD[k] >> 5) & 0x7; 888 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1, 889 rowCr0, rowCr1, rowDr0, rowDr1); 890 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1, 891 rowCg0, rowCg1, rowDg0, rowDg1); 892 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1, 893 rowCb0, rowCb1, rowDb0, rowDb1); 894 dst[i] = (b << 5) | (g << 2) | r; 895 } 896 } 897 else { 898 debug_printf("bad format in do_row_3D()"); 899 } 900} 901 902 903 904static void 905format_to_type_comps(enum pipe_format pformat, 906 enum dtype *datatype, uint *comps) 907{ 908 /* XXX I think this could be implemented in terms of the pf_*() functions */ 909 switch (pformat) { 910 case PIPE_FORMAT_B8G8R8A8_UNORM: 911 case PIPE_FORMAT_B8G8R8X8_UNORM: 912 case PIPE_FORMAT_A8R8G8B8_UNORM: 913 case PIPE_FORMAT_X8R8G8B8_UNORM: 914 case PIPE_FORMAT_A8B8G8R8_SRGB: 915 case PIPE_FORMAT_X8B8G8R8_SRGB: 916 case PIPE_FORMAT_B8G8R8A8_SRGB: 917 case PIPE_FORMAT_B8G8R8X8_SRGB: 918 case PIPE_FORMAT_A8R8G8B8_SRGB: 919 case PIPE_FORMAT_X8R8G8B8_SRGB: 920 case PIPE_FORMAT_R8G8B8_SRGB: 921 *datatype = DTYPE_UBYTE; 922 *comps = 4; 923 return; 924 case PIPE_FORMAT_B5G5R5X1_UNORM: 925 case PIPE_FORMAT_B5G5R5A1_UNORM: 926 *datatype = DTYPE_USHORT_1_5_5_5_REV; 927 *comps = 4; 928 return; 929 case PIPE_FORMAT_B4G4R4A4_UNORM: 930 *datatype = DTYPE_USHORT_4_4_4_4; 931 *comps = 4; 932 return; 933 case PIPE_FORMAT_B5G6R5_UNORM: 934 *datatype = DTYPE_USHORT_5_6_5; 935 *comps = 3; 936 return; 937 case PIPE_FORMAT_L8_UNORM: 938 case PIPE_FORMAT_L8_SRGB: 939 case PIPE_FORMAT_A8_UNORM: 940 case PIPE_FORMAT_I8_UNORM: 941 *datatype = DTYPE_UBYTE; 942 *comps = 1; 943 return; 944 case PIPE_FORMAT_L8A8_UNORM: 945 case PIPE_FORMAT_L8A8_SRGB: 946 *datatype = DTYPE_UBYTE; 947 *comps = 2; 948 return; 949 default: 950 assert(0); 951 *datatype = DTYPE_UBYTE; 952 *comps = 0; 953 break; 954 } 955} 956 957 958static void 959reduce_1d(enum pipe_format pformat, 960 int srcWidth, const ubyte *srcPtr, 961 int dstWidth, ubyte *dstPtr) 962{ 963 enum dtype datatype; 964 uint comps; 965 966 format_to_type_comps(pformat, &datatype, &comps); 967 968 /* we just duplicate the input row, kind of hack, saves code */ 969 do_row(datatype, comps, 970 srcWidth, srcPtr, srcPtr, 971 dstWidth, dstPtr); 972} 973 974 975/** 976 * Strides are in bytes. If zero, it'll be computed as width * bpp. 977 */ 978static void 979reduce_2d(enum pipe_format pformat, 980 int srcWidth, int srcHeight, 981 int srcRowStride, const ubyte *srcPtr, 982 int dstWidth, int dstHeight, 983 int dstRowStride, ubyte *dstPtr) 984{ 985 enum dtype datatype; 986 uint comps; 987 const int bpt = util_format_get_blocksize(pformat); 988 const ubyte *srcA, *srcB; 989 ubyte *dst; 990 int row; 991 992 format_to_type_comps(pformat, &datatype, &comps); 993 994 if (!srcRowStride) 995 srcRowStride = bpt * srcWidth; 996 997 if (!dstRowStride) 998 dstRowStride = bpt * dstWidth; 999 1000 /* Compute src and dst pointers */ 1001 srcA = srcPtr; 1002 if (srcHeight > 1) 1003 srcB = srcA + srcRowStride; 1004 else 1005 srcB = srcA; 1006 dst = dstPtr; 1007 1008 for (row = 0; row < dstHeight; row++) { 1009 do_row(datatype, comps, 1010 srcWidth, srcA, srcB, 1011 dstWidth, dst); 1012 srcA += 2 * srcRowStride; 1013 srcB += 2 * srcRowStride; 1014 dst += dstRowStride; 1015 } 1016} 1017 1018 1019static void 1020reduce_3d(enum pipe_format pformat, 1021 int srcWidth, int srcHeight, int srcDepth, 1022 int srcRowStride, int srcImageStride, const ubyte *srcPtr, 1023 int dstWidth, int dstHeight, int dstDepth, 1024 int dstRowStride, int dstImageStride, ubyte *dstPtr) 1025{ 1026 const int bpt = util_format_get_blocksize(pformat); 1027 int img, row; 1028 int srcImageOffset, srcRowOffset; 1029 enum dtype datatype; 1030 uint comps; 1031 1032 format_to_type_comps(pformat, &datatype, &comps); 1033 1034 /* XXX I think we should rather assert those strides */ 1035 if (!srcImageStride) 1036 srcImageStride = srcWidth * srcHeight * bpt; 1037 if (!dstImageStride) 1038 dstImageStride = dstWidth * dstHeight * bpt; 1039 1040 if (!srcRowStride) 1041 srcRowStride = srcWidth * bpt; 1042 if (!dstRowStride) 1043 dstRowStride = dstWidth * bpt; 1044 1045 /* Offset between adjacent src images to be averaged together */ 1046 srcImageOffset = (srcDepth == dstDepth) ? 0 : srcImageStride; 1047 1048 /* Offset between adjacent src rows to be averaged together */ 1049 srcRowOffset = (srcHeight == dstHeight) ? 0 : srcRowStride; 1050 1051 /* 1052 * Need to average together up to 8 src pixels for each dest pixel. 1053 * Break that down into 3 operations: 1054 * 1. take two rows from source image and average them together. 1055 * 2. take two rows from next source image and average them together. 1056 * 3. take the two averaged rows and average them for the final dst row. 1057 */ 1058 1059 /* 1060 printf("mip3d %d x %d x %d -> %d x %d x %d\n", 1061 srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth); 1062 */ 1063 1064 for (img = 0; img < dstDepth; img++) { 1065 /* first source image pointer */ 1066 const ubyte *imgSrcA = srcPtr 1067 + img * (srcImageStride + srcImageOffset); 1068 /* second source image pointer */ 1069 const ubyte *imgSrcB = imgSrcA + srcImageOffset; 1070 /* address of the dest image */ 1071 ubyte *imgDst = dstPtr + img * dstImageStride; 1072 1073 /* setup the four source row pointers and the dest row pointer */ 1074 const ubyte *srcImgARowA = imgSrcA; 1075 const ubyte *srcImgARowB = imgSrcA + srcRowOffset; 1076 const ubyte *srcImgBRowA = imgSrcB; 1077 const ubyte *srcImgBRowB = imgSrcB + srcRowOffset; 1078 ubyte *dstImgRow = imgDst; 1079 1080 for (row = 0; row < dstHeight; row++) { 1081 do_row_3D(datatype, comps, srcWidth, 1082 srcImgARowA, srcImgARowB, 1083 srcImgBRowA, srcImgBRowB, 1084 dstWidth, dstImgRow); 1085 1086 /* advance to next rows */ 1087 srcImgARowA += srcRowStride + srcRowOffset; 1088 srcImgARowB += srcRowStride + srcRowOffset; 1089 srcImgBRowA += srcRowStride + srcRowOffset; 1090 srcImgBRowB += srcRowStride + srcRowOffset; 1091 dstImgRow += dstImageStride; 1092 } 1093 } 1094} 1095 1096 1097 1098 1099static void 1100make_1d_mipmap(struct gen_mipmap_state *ctx, 1101 struct pipe_resource *pt, 1102 uint layer, uint baseLevel, uint lastLevel) 1103{ 1104 struct pipe_context *pipe = ctx->pipe; 1105 uint dstLevel; 1106 1107 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1108 const uint srcLevel = dstLevel - 1; 1109 struct pipe_transfer *srcTrans, *dstTrans; 1110 void *srcMap, *dstMap; 1111 1112 srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer, 1113 PIPE_TRANSFER_READ, 0, 0, 1114 u_minify(pt->width0, srcLevel), 1115 u_minify(pt->height0, srcLevel)); 1116 dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer, 1117 PIPE_TRANSFER_WRITE, 0, 0, 1118 u_minify(pt->width0, dstLevel), 1119 u_minify(pt->height0, dstLevel)); 1120 1121 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); 1122 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); 1123 1124 reduce_1d(pt->format, 1125 srcTrans->box.width, srcMap, 1126 dstTrans->box.width, dstMap); 1127 1128 pipe->transfer_unmap(pipe, srcTrans); 1129 pipe->transfer_unmap(pipe, dstTrans); 1130 1131 pipe->transfer_destroy(pipe, srcTrans); 1132 pipe->transfer_destroy(pipe, dstTrans); 1133 } 1134} 1135 1136 1137static void 1138make_2d_mipmap(struct gen_mipmap_state *ctx, 1139 struct pipe_resource *pt, 1140 uint layer, uint baseLevel, uint lastLevel) 1141{ 1142 struct pipe_context *pipe = ctx->pipe; 1143 uint dstLevel; 1144 1145 assert(util_format_get_blockwidth(pt->format) == 1); 1146 assert(util_format_get_blockheight(pt->format) == 1); 1147 1148 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1149 const uint srcLevel = dstLevel - 1; 1150 struct pipe_transfer *srcTrans, *dstTrans; 1151 ubyte *srcMap, *dstMap; 1152 1153 srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer, 1154 PIPE_TRANSFER_READ, 0, 0, 1155 u_minify(pt->width0, srcLevel), 1156 u_minify(pt->height0, srcLevel)); 1157 dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer, 1158 PIPE_TRANSFER_WRITE, 0, 0, 1159 u_minify(pt->width0, dstLevel), 1160 u_minify(pt->height0, dstLevel)); 1161 1162 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); 1163 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); 1164 1165 reduce_2d(pt->format, 1166 srcTrans->box.width, srcTrans->box.height, 1167 srcTrans->stride, srcMap, 1168 dstTrans->box.width, dstTrans->box.height, 1169 dstTrans->stride, dstMap); 1170 1171 pipe->transfer_unmap(pipe, srcTrans); 1172 pipe->transfer_unmap(pipe, dstTrans); 1173 1174 pipe->transfer_destroy(pipe, srcTrans); 1175 pipe->transfer_destroy(pipe, dstTrans); 1176 } 1177} 1178 1179 1180/* XXX looks a bit more like it could work now but need to test */ 1181static void 1182make_3d_mipmap(struct gen_mipmap_state *ctx, 1183 struct pipe_resource *pt, 1184 uint face, uint baseLevel, uint lastLevel) 1185{ 1186 struct pipe_context *pipe = ctx->pipe; 1187 uint dstLevel; 1188 struct pipe_box src_box, dst_box; 1189 1190 assert(util_format_get_blockwidth(pt->format) == 1); 1191 assert(util_format_get_blockheight(pt->format) == 1); 1192 1193 src_box.x = src_box.y = src_box.z = 0; 1194 dst_box.x = dst_box.y = dst_box.z = 0; 1195 1196 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1197 const uint srcLevel = dstLevel - 1; 1198 struct pipe_transfer *srcTrans, *dstTrans; 1199 ubyte *srcMap, *dstMap; 1200 struct pipe_box src_box, dst_box; 1201 src_box.width = u_minify(pt->width0, srcLevel); 1202 src_box.height = u_minify(pt->height0, srcLevel); 1203 src_box.depth = u_minify(pt->depth0, srcLevel); 1204 dst_box.width = u_minify(pt->width0, dstLevel); 1205 dst_box.height = u_minify(pt->height0, dstLevel); 1206 dst_box.depth = u_minify(pt->depth0, dstLevel); 1207 1208 srcTrans = pipe->get_transfer(pipe, pt, srcLevel, 1209 PIPE_TRANSFER_READ, 1210 &src_box); 1211 dstTrans = pipe->get_transfer(pipe, pt, dstLevel, 1212 PIPE_TRANSFER_WRITE, 1213 &dst_box); 1214 1215 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); 1216 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); 1217 1218 reduce_3d(pt->format, 1219 srcTrans->box.width, srcTrans->box.height, srcTrans->box.depth, 1220 srcTrans->stride, srcTrans->layer_stride, srcMap, 1221 dstTrans->box.width, dstTrans->box.height, dstTrans->box.depth, 1222 dstTrans->stride, dstTrans->layer_stride, dstMap); 1223 1224 pipe->transfer_unmap(pipe, srcTrans); 1225 pipe->transfer_unmap(pipe, dstTrans); 1226 1227 pipe->transfer_destroy(pipe, srcTrans); 1228 pipe->transfer_destroy(pipe, dstTrans); 1229 } 1230} 1231 1232 1233static void 1234fallback_gen_mipmap(struct gen_mipmap_state *ctx, 1235 struct pipe_resource *pt, 1236 uint layer, uint baseLevel, uint lastLevel) 1237{ 1238 switch (pt->target) { 1239 case PIPE_TEXTURE_1D: 1240 make_1d_mipmap(ctx, pt, layer, baseLevel, lastLevel); 1241 break; 1242 case PIPE_TEXTURE_2D: 1243 case PIPE_TEXTURE_RECT: 1244 case PIPE_TEXTURE_CUBE: 1245 make_2d_mipmap(ctx, pt, layer, baseLevel, lastLevel); 1246 break; 1247 case PIPE_TEXTURE_3D: 1248 make_3d_mipmap(ctx, pt, layer, baseLevel, lastLevel); 1249 break; 1250 default: 1251 assert(0); 1252 } 1253} 1254 1255 1256/** 1257 * Create a mipmap generation context. 1258 * The idea is to create one of these and re-use it each time we need to 1259 * generate a mipmap. 1260 */ 1261struct gen_mipmap_state * 1262util_create_gen_mipmap(struct pipe_context *pipe, 1263 struct cso_context *cso) 1264{ 1265 struct gen_mipmap_state *ctx; 1266 uint i; 1267 1268 ctx = CALLOC_STRUCT(gen_mipmap_state); 1269 if (!ctx) 1270 return NULL; 1271 1272 ctx->pipe = pipe; 1273 ctx->cso = cso; 1274 1275 /* disabled blending/masking */ 1276 memset(&ctx->blend, 0, sizeof(ctx->blend)); 1277 ctx->blend.rt[0].colormask = PIPE_MASK_RGBA; 1278 1279 /* no-op depth/stencil/alpha */ 1280 memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil)); 1281 1282 /* rasterizer */ 1283 memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer)); 1284 ctx->rasterizer.cull_face = PIPE_FACE_NONE; 1285 ctx->rasterizer.gl_rasterization_rules = 1; 1286 1287 /* sampler state */ 1288 memset(&ctx->sampler, 0, sizeof(ctx->sampler)); 1289 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1290 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1291 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1292 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST; 1293 ctx->sampler.normalized_coords = 1; 1294 1295 /* vertex elements state */ 1296 memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2); 1297 for (i = 0; i < 2; i++) { 1298 ctx->velem[i].src_offset = i * 4 * sizeof(float); 1299 ctx->velem[i].instance_divisor = 0; 1300 ctx->velem[i].vertex_buffer_index = 0; 1301 ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT; 1302 } 1303 1304 /* vertex shader - still needed to specify mapping from fragment 1305 * shader input semantics to vertex elements 1306 */ 1307 { 1308 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION, 1309 TGSI_SEMANTIC_GENERIC }; 1310 const uint semantic_indexes[] = { 0, 0 }; 1311 ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names, 1312 semantic_indexes); 1313 } 1314 1315 /* fragment shader */ 1316 ctx->fs1d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_1D, 1317 TGSI_INTERPOLATE_LINEAR); 1318 ctx->fs2d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_2D, 1319 TGSI_INTERPOLATE_LINEAR); 1320 ctx->fs3d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_3D, 1321 TGSI_INTERPOLATE_LINEAR); 1322 ctx->fsCube = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_CUBE, 1323 TGSI_INTERPOLATE_LINEAR); 1324 if (pipe->screen->get_param(pipe->screen, PIPE_CAP_ARRAY_TEXTURES)) { 1325 ctx->fs1da = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_1D_ARRAY, 1326 TGSI_INTERPOLATE_LINEAR); 1327 ctx->fs2da = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_2D_ARRAY, 1328 TGSI_INTERPOLATE_LINEAR); 1329 } 1330 1331 1332 /* vertex data that doesn't change */ 1333 for (i = 0; i < 4; i++) { 1334 ctx->vertices[i][0][2] = 0.0f; /* z */ 1335 ctx->vertices[i][0][3] = 1.0f; /* w */ 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 PIPE_BIND_VERTEX_BUFFER, 1360 PIPE_USAGE_STREAM, 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, 1370 enum pipe_texture_target tex_target, 1371 uint layer, float r) 1372{ 1373 unsigned offset; 1374 1375 /* vert[0].position */ 1376 ctx->vertices[0][0][0] = -1.0f; /*x*/ 1377 ctx->vertices[0][0][1] = -1.0f; /*y*/ 1378 1379 /* vert[1].position */ 1380 ctx->vertices[1][0][0] = 1.0f; 1381 ctx->vertices[1][0][1] = -1.0f; 1382 1383 /* vert[2].position */ 1384 ctx->vertices[2][0][0] = 1.0f; 1385 ctx->vertices[2][0][1] = 1.0f; 1386 1387 /* vert[3].position */ 1388 ctx->vertices[3][0][0] = -1.0f; 1389 ctx->vertices[3][0][1] = 1.0f; 1390 1391 /* Setup vertex texcoords. This is a little tricky for cube maps. */ 1392 if (tex_target == PIPE_TEXTURE_CUBE) { 1393 static const float st[4][2] = { 1394 {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f} 1395 }; 1396 1397 util_map_texcoords2d_onto_cubemap(layer, &st[0][0], 2, 1398 &ctx->vertices[0][1][0], 8); 1399 } 1400 else if (tex_target == PIPE_TEXTURE_1D_ARRAY) { 1401 /* 1D texture array */ 1402 ctx->vertices[0][1][0] = 0.0f; /*s*/ 1403 ctx->vertices[0][1][1] = r; /*t*/ 1404 ctx->vertices[0][1][2] = 0.0f; /*r*/ 1405 1406 ctx->vertices[1][1][0] = 1.0f; 1407 ctx->vertices[1][1][1] = r; 1408 ctx->vertices[1][1][2] = 0.0f; 1409 1410 ctx->vertices[2][1][0] = 1.0f; 1411 ctx->vertices[2][1][1] = r; 1412 ctx->vertices[2][1][2] = 0.0f; 1413 1414 ctx->vertices[3][1][0] = 0.0f; 1415 ctx->vertices[3][1][1] = r; 1416 ctx->vertices[3][1][2] = 0.0f; 1417 } else { 1418 /* 1D/2D/3D/2D array */ 1419 ctx->vertices[0][1][0] = 0.0f; /*s*/ 1420 ctx->vertices[0][1][1] = 0.0f; /*t*/ 1421 ctx->vertices[0][1][2] = r; /*r*/ 1422 1423 ctx->vertices[1][1][0] = 1.0f; 1424 ctx->vertices[1][1][1] = 0.0f; 1425 ctx->vertices[1][1][2] = r; 1426 1427 ctx->vertices[2][1][0] = 1.0f; 1428 ctx->vertices[2][1][1] = 1.0f; 1429 ctx->vertices[2][1][2] = r; 1430 1431 ctx->vertices[3][1][0] = 0.0f; 1432 ctx->vertices[3][1][1] = 1.0f; 1433 ctx->vertices[3][1][2] = r; 1434 } 1435 1436 offset = get_next_slot( ctx ); 1437 1438 pipe_buffer_write_nooverlap(ctx->pipe, ctx->vbuf, 1439 offset, sizeof(ctx->vertices), ctx->vertices); 1440 1441 return offset; 1442} 1443 1444 1445 1446/** 1447 * Destroy a mipmap generation context 1448 */ 1449void 1450util_destroy_gen_mipmap(struct gen_mipmap_state *ctx) 1451{ 1452 struct pipe_context *pipe = ctx->pipe; 1453 1454 if (ctx->fs2da) 1455 pipe->delete_fs_state(pipe, ctx->fs2da); 1456 if (ctx->fs1da) 1457 pipe->delete_fs_state(pipe, ctx->fs1da); 1458 pipe->delete_fs_state(pipe, ctx->fsCube); 1459 pipe->delete_fs_state(pipe, ctx->fs3d); 1460 pipe->delete_fs_state(pipe, ctx->fs2d); 1461 pipe->delete_fs_state(pipe, ctx->fs1d); 1462 pipe->delete_vs_state(pipe, ctx->vs); 1463 1464 pipe_resource_reference(&ctx->vbuf, NULL); 1465 1466 FREE(ctx); 1467} 1468 1469 1470 1471/* Release vertex buffer at end of frame to avoid synchronous 1472 * rendering. 1473 */ 1474void util_gen_mipmap_flush( struct gen_mipmap_state *ctx ) 1475{ 1476 pipe_resource_reference(&ctx->vbuf, NULL); 1477 ctx->vbuf_slot = 0; 1478} 1479 1480 1481/** 1482 * Generate mipmap images. It's assumed all needed texture memory is 1483 * already allocated. 1484 * 1485 * \param psv the sampler view to the texture to generate mipmap levels for 1486 * \param face which cube face to generate mipmaps for (0 for non-cube maps) 1487 * \param baseLevel the first mipmap level to use as a src 1488 * \param lastLevel the last mipmap level to generate 1489 * \param filter the minification filter used to generate mipmap levels with 1490 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST 1491 */ 1492void 1493util_gen_mipmap(struct gen_mipmap_state *ctx, 1494 struct pipe_sampler_view *psv, 1495 uint face, uint baseLevel, uint lastLevel, uint filter) 1496{ 1497 struct pipe_context *pipe = ctx->pipe; 1498 struct pipe_screen *screen = pipe->screen; 1499 struct pipe_framebuffer_state fb; 1500 struct pipe_resource *pt = psv->texture; 1501 void *fs; 1502 uint dstLevel; 1503 uint offset; 1504 1505 /* The texture object should have room for the levels which we're 1506 * about to generate. 1507 */ 1508 assert(lastLevel <= pt->last_level); 1509 1510 /* If this fails, why are we here? */ 1511 assert(lastLevel > baseLevel); 1512 1513 assert(filter == PIPE_TEX_FILTER_LINEAR || 1514 filter == PIPE_TEX_FILTER_NEAREST); 1515 1516 switch (pt->target) { 1517 case PIPE_TEXTURE_1D: 1518 fs = ctx->fs1d; 1519 break; 1520 case PIPE_TEXTURE_2D: 1521 fs = ctx->fs2d; 1522 break; 1523 case PIPE_TEXTURE_3D: 1524 fs = ctx->fs3d; 1525 break; 1526 case PIPE_TEXTURE_CUBE: 1527 fs = ctx->fsCube; 1528 break; 1529 case PIPE_TEXTURE_1D_ARRAY: 1530 fs = ctx->fs1da; 1531 break; 1532 case PIPE_TEXTURE_2D_ARRAY: 1533 fs = ctx->fs2da; 1534 break; 1535 default: 1536 assert(0); 1537 fs = ctx->fs2d; 1538 } 1539 1540 /* check if we can render in the texture's format */ 1541 if (!screen->is_format_supported(screen, psv->format, pt->target, 1542 pt->nr_samples, PIPE_BIND_RENDER_TARGET, 0)) { 1543 fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel); 1544 return; 1545 } 1546 1547 /* save state (restored below) */ 1548 cso_save_blend(ctx->cso); 1549 cso_save_depth_stencil_alpha(ctx->cso); 1550 cso_save_rasterizer(ctx->cso); 1551 cso_save_samplers(ctx->cso); 1552 cso_save_fragment_sampler_views(ctx->cso); 1553 cso_save_framebuffer(ctx->cso); 1554 cso_save_fragment_shader(ctx->cso); 1555 cso_save_vertex_shader(ctx->cso); 1556 cso_save_viewport(ctx->cso); 1557 cso_save_clip(ctx->cso); 1558 cso_save_vertex_elements(ctx->cso); 1559 1560 /* bind our state */ 1561 cso_set_blend(ctx->cso, &ctx->blend); 1562 cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil); 1563 cso_set_rasterizer(ctx->cso, &ctx->rasterizer); 1564 cso_set_clip(ctx->cso, &ctx->clip); 1565 cso_set_vertex_elements(ctx->cso, 2, ctx->velem); 1566 1567 cso_set_fragment_shader_handle(ctx->cso, fs); 1568 cso_set_vertex_shader_handle(ctx->cso, ctx->vs); 1569 1570 /* init framebuffer state */ 1571 memset(&fb, 0, sizeof(fb)); 1572 fb.nr_cbufs = 1; 1573 1574 /* set min/mag to same filter for faster sw speed */ 1575 ctx->sampler.mag_img_filter = filter; 1576 ctx->sampler.min_img_filter = filter; 1577 1578 /* 1579 * XXX for small mipmap levels, it may be faster to use the software 1580 * fallback path... 1581 */ 1582 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 1583 const uint srcLevel = dstLevel - 1; 1584 struct pipe_viewport_state vp; 1585 unsigned nr_layers, layer, i; 1586 float rcoord = 0.0f; 1587 1588 if (pt->target == PIPE_TEXTURE_3D) 1589 nr_layers = u_minify(pt->depth0, dstLevel); 1590 else if (pt->target == PIPE_TEXTURE_2D_ARRAY || pt->target == PIPE_TEXTURE_1D_ARRAY) 1591 nr_layers = pt->array_size; 1592 else 1593 nr_layers = 1; 1594 1595 for (i = 0; i < nr_layers; i++) { 1596 struct pipe_surface *surf, surf_templ; 1597 if (pt->target == PIPE_TEXTURE_3D) { 1598 /* in theory with geom shaders and driver with full layer support 1599 could do that in one go. */ 1600 layer = i; 1601 /* XXX hmm really? */ 1602 rcoord = (float)layer / (float)nr_layers + 1.0f / (float)(nr_layers * 2); 1603 } else if (pt->target == PIPE_TEXTURE_2D_ARRAY || pt->target == PIPE_TEXTURE_1D_ARRAY) { 1604 layer = i; 1605 rcoord = (float)layer; 1606 } else 1607 layer = face; 1608 1609 memset(&surf_templ, 0, sizeof(surf_templ)); 1610 u_surface_default_template(&surf_templ, pt, PIPE_BIND_RENDER_TARGET); 1611 surf_templ.u.tex.level = dstLevel; 1612 surf_templ.u.tex.first_layer = layer; 1613 surf_templ.u.tex.last_layer = layer; 1614 surf = pipe->create_surface(pipe, pt, &surf_templ); 1615 1616 /* 1617 * Setup framebuffer / dest surface 1618 */ 1619 fb.cbufs[0] = surf; 1620 fb.width = u_minify(pt->width0, dstLevel); 1621 fb.height = u_minify(pt->height0, dstLevel); 1622 cso_set_framebuffer(ctx->cso, &fb); 1623 1624 /* viewport */ 1625 vp.scale[0] = 0.5f * fb.width; 1626 vp.scale[1] = 0.5f * fb.height; 1627 vp.scale[2] = 1.0f; 1628 vp.scale[3] = 1.0f; 1629 vp.translate[0] = 0.5f * fb.width; 1630 vp.translate[1] = 0.5f * fb.height; 1631 vp.translate[2] = 0.0f; 1632 vp.translate[3] = 0.0f; 1633 cso_set_viewport(ctx->cso, &vp); 1634 1635 /* 1636 * Setup sampler state 1637 * Note: we should only have to set the min/max LOD clamps to ensure 1638 * we grab texels from the right mipmap level. But some hardware 1639 * has trouble with min clamping so we also set the lod_bias to 1640 * try to work around that. 1641 */ 1642 ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel; 1643 ctx->sampler.lod_bias = (float) srcLevel; 1644 cso_single_sampler(ctx->cso, 0, &ctx->sampler); 1645 cso_single_sampler_done(ctx->cso); 1646 1647 cso_set_fragment_sampler_views(ctx->cso, 1, &psv); 1648 1649 /* quad coords in clip coords */ 1650 offset = set_vertex_data(ctx, 1651 pt->target, 1652 face, 1653 rcoord); 1654 1655 util_draw_vertex_buffer(ctx->pipe, 1656 ctx->cso, 1657 ctx->vbuf, 1658 offset, 1659 PIPE_PRIM_TRIANGLE_FAN, 1660 4, /* verts */ 1661 2); /* attribs/vert */ 1662 1663 pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); 1664 1665 /* need to signal that the texture has changed _after_ rendering to it */ 1666 pipe_surface_reference( &surf, NULL ); 1667 } 1668 } 1669 1670 /* restore state we changed */ 1671 cso_restore_blend(ctx->cso); 1672 cso_restore_depth_stencil_alpha(ctx->cso); 1673 cso_restore_rasterizer(ctx->cso); 1674 cso_restore_samplers(ctx->cso); 1675 cso_restore_fragment_sampler_views(ctx->cso); 1676 cso_restore_framebuffer(ctx->cso); 1677 cso_restore_fragment_shader(ctx->cso); 1678 cso_restore_vertex_shader(ctx->cso); 1679 cso_restore_viewport(ctx->cso); 1680 cso_restore_clip(ctx->cso); 1681 cso_restore_vertex_elements(ctx->cso); 1682} 1683