u_gen_mipmap.c revision d2c2e9316d043ab584794a3524f22776deb4c777
1/************************************************************************** 2 * 3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. 4 * All Rights Reserved. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the 8 * "Software"), to deal in the Software without restriction, including 9 * without limitation the rights to use, copy, modify, merge, publish, 10 * distribute, sub license, and/or sell copies of the Software, and to 11 * permit persons to whom the Software is furnished to do so, subject to 12 * the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the 15 * next paragraph) shall be included in all copies or substantial portions 16 * of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 * 26 **************************************************************************/ 27 28/** 29 * @file 30 * Mipmap generation utility 31 * 32 * @author Brian Paul 33 */ 34 35 36#include "pipe/p_context.h" 37#include "pipe/p_debug.h" 38#include "pipe/p_defines.h" 39#include "pipe/p_inlines.h" 40#include "pipe/p_winsys.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 96#if 0 97extern half_float 98float_to_half(float f); 99 100extern float 101half_to_float(half_float h); 102#endif 103 104 105/** 106 * Average together two rows of a source image to produce a single new 107 * row in the dest image. It's legal for the two source rows to point 108 * to the same data. The source width must be equal to either the 109 * dest width or two times the dest width. 110 * \param datatype GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc. 111 * \param comps number of components per pixel (1..4) 112 */ 113static void 114do_row(enum dtype datatype, uint comps, int srcWidth, 115 const void *srcRowA, const void *srcRowB, 116 int dstWidth, void *dstRow) 117{ 118 const uint k0 = (srcWidth == dstWidth) ? 0 : 1; 119 const uint colStride = (srcWidth == dstWidth) ? 1 : 2; 120 121 assert(comps >= 1); 122 assert(comps <= 4); 123 124 /* This assertion is no longer valid with non-power-of-2 textures 125 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth); 126 */ 127 128 if (datatype == UBYTE && comps == 4) { 129 uint i, j, k; 130 const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA; 131 const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB; 132 ubyte(*dst)[4] = (ubyte(*)[4]) dstRow; 133 for (i = j = 0, k = k0; i < (uint) dstWidth; 134 i++, j += colStride, k += colStride) { 135 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 136 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 137 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 138 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 139 } 140 } 141 else if (datatype == UBYTE && comps == 3) { 142 uint i, j, k; 143 const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA; 144 const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB; 145 ubyte(*dst)[3] = (ubyte(*)[3]) dstRow; 146 for (i = j = 0, k = k0; i < (uint) dstWidth; 147 i++, j += colStride, k += colStride) { 148 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 149 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 150 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 151 } 152 } 153 else if (datatype == UBYTE && comps == 2) { 154 uint i, j, k; 155 const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA; 156 const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB; 157 ubyte(*dst)[2] = (ubyte(*)[2]) dstRow; 158 for (i = j = 0, k = k0; i < (uint) dstWidth; 159 i++, j += colStride, k += colStride) { 160 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2; 161 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2; 162 } 163 } 164 else if (datatype == UBYTE && comps == 1) { 165 uint i, j, k; 166 const ubyte *rowA = (const ubyte *) srcRowA; 167 const ubyte *rowB = (const ubyte *) srcRowB; 168 ubyte *dst = (ubyte *) dstRow; 169 for (i = j = 0, k = k0; i < (uint) dstWidth; 170 i++, j += colStride, k += colStride) { 171 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2; 172 } 173 } 174 175 else if (datatype == USHORT && comps == 4) { 176 uint i, j, k; 177 const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA; 178 const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB; 179 ushort(*dst)[4] = (ushort(*)[4]) dstRow; 180 for (i = j = 0, k = k0; i < (uint) dstWidth; 181 i++, j += colStride, k += colStride) { 182 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 183 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 184 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 185 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4; 186 } 187 } 188 else if (datatype == USHORT && comps == 3) { 189 uint i, j, k; 190 const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA; 191 const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB; 192 ushort(*dst)[3] = (ushort(*)[3]) dstRow; 193 for (i = j = 0, k = k0; i < (uint) dstWidth; 194 i++, j += colStride, k += colStride) { 195 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 196 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 197 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4; 198 } 199 } 200 else if (datatype == USHORT && comps == 2) { 201 uint i, j, k; 202 const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA; 203 const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB; 204 ushort(*dst)[2] = (ushort(*)[2]) dstRow; 205 for (i = j = 0, k = k0; i < (uint) dstWidth; 206 i++, j += colStride, k += colStride) { 207 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4; 208 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4; 209 } 210 } 211 else if (datatype == USHORT && comps == 1) { 212 uint i, j, k; 213 const ushort *rowA = (const ushort *) srcRowA; 214 const ushort *rowB = (const ushort *) srcRowB; 215 ushort *dst = (ushort *) dstRow; 216 for (i = j = 0, k = k0; i < (uint) dstWidth; 217 i++, j += colStride, k += colStride) { 218 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4; 219 } 220 } 221 222 else if (datatype == FLOAT && comps == 4) { 223 uint i, j, k; 224 const float(*rowA)[4] = (const float(*)[4]) srcRowA; 225 const float(*rowB)[4] = (const float(*)[4]) srcRowB; 226 float(*dst)[4] = (float(*)[4]) 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] + 230 rowB[j][0] + rowB[k][0]) * 0.25F; 231 dst[i][1] = (rowA[j][1] + rowA[k][1] + 232 rowB[j][1] + rowB[k][1]) * 0.25F; 233 dst[i][2] = (rowA[j][2] + rowA[k][2] + 234 rowB[j][2] + rowB[k][2]) * 0.25F; 235 dst[i][3] = (rowA[j][3] + rowA[k][3] + 236 rowB[j][3] + rowB[k][3]) * 0.25F; 237 } 238 } 239 else if (datatype == FLOAT && comps == 3) { 240 uint i, j, k; 241 const float(*rowA)[3] = (const float(*)[3]) srcRowA; 242 const float(*rowB)[3] = (const float(*)[3]) srcRowB; 243 float(*dst)[3] = (float(*)[3]) dstRow; 244 for (i = j = 0, k = k0; i < (uint) dstWidth; 245 i++, j += colStride, k += colStride) { 246 dst[i][0] = (rowA[j][0] + rowA[k][0] + 247 rowB[j][0] + rowB[k][0]) * 0.25F; 248 dst[i][1] = (rowA[j][1] + rowA[k][1] + 249 rowB[j][1] + rowB[k][1]) * 0.25F; 250 dst[i][2] = (rowA[j][2] + rowA[k][2] + 251 rowB[j][2] + rowB[k][2]) * 0.25F; 252 } 253 } 254 else if (datatype == FLOAT && comps == 2) { 255 uint i, j, k; 256 const float(*rowA)[2] = (const float(*)[2]) srcRowA; 257 const float(*rowB)[2] = (const float(*)[2]) srcRowB; 258 float(*dst)[2] = (float(*)[2]) dstRow; 259 for (i = j = 0, k = k0; i < (uint) dstWidth; 260 i++, j += colStride, k += colStride) { 261 dst[i][0] = (rowA[j][0] + rowA[k][0] + 262 rowB[j][0] + rowB[k][0]) * 0.25F; 263 dst[i][1] = (rowA[j][1] + rowA[k][1] + 264 rowB[j][1] + rowB[k][1]) * 0.25F; 265 } 266 } 267 else if (datatype == FLOAT && comps == 1) { 268 uint i, j, k; 269 const float *rowA = (const float *) srcRowA; 270 const float *rowB = (const float *) srcRowB; 271 float *dst = (float *) dstRow; 272 for (i = j = 0, k = k0; i < (uint) dstWidth; 273 i++, j += colStride, k += colStride) { 274 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F; 275 } 276 } 277 278#if 0 279 else if (datatype == HALF_FLOAT && comps == 4) { 280 uint i, j, k, comp; 281 const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA; 282 const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB; 283 half_float(*dst)[4] = (half_float(*)[4]) dstRow; 284 for (i = j = 0, k = k0; i < (uint) dstWidth; 285 i++, j += colStride, k += colStride) { 286 for (comp = 0; comp < 4; comp++) { 287 float aj, ak, bj, bk; 288 aj = half_to_float(rowA[j][comp]); 289 ak = half_to_float(rowA[k][comp]); 290 bj = half_to_float(rowB[j][comp]); 291 bk = half_to_float(rowB[k][comp]); 292 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 293 } 294 } 295 } 296 else if (datatype == HALF_FLOAT && comps == 3) { 297 uint i, j, k, comp; 298 const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA; 299 const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB; 300 half_float(*dst)[3] = (half_float(*)[3]) dstRow; 301 for (i = j = 0, k = k0; i < (uint) dstWidth; 302 i++, j += colStride, k += colStride) { 303 for (comp = 0; comp < 3; comp++) { 304 float aj, ak, bj, bk; 305 aj = half_to_float(rowA[j][comp]); 306 ak = half_to_float(rowA[k][comp]); 307 bj = half_to_float(rowB[j][comp]); 308 bk = half_to_float(rowB[k][comp]); 309 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 310 } 311 } 312 } 313 else if (datatype == HALF_FLOAT && comps == 2) { 314 uint i, j, k, comp; 315 const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA; 316 const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB; 317 half_float(*dst)[2] = (half_float(*)[2]) dstRow; 318 for (i = j = 0, k = k0; i < (uint) dstWidth; 319 i++, j += colStride, k += colStride) { 320 for (comp = 0; comp < 2; comp++) { 321 float aj, ak, bj, bk; 322 aj = half_to_float(rowA[j][comp]); 323 ak = half_to_float(rowA[k][comp]); 324 bj = half_to_float(rowB[j][comp]); 325 bk = half_to_float(rowB[k][comp]); 326 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F); 327 } 328 } 329 } 330 else if (datatype == HALF_FLOAT && comps == 1) { 331 uint i, j, k; 332 const half_float *rowA = (const half_float *) srcRowA; 333 const half_float *rowB = (const half_float *) srcRowB; 334 half_float *dst = (half_float *) dstRow; 335 for (i = j = 0, k = k0; i < (uint) dstWidth; 336 i++, j += colStride, k += colStride) { 337 float aj, ak, bj, bk; 338 aj = half_to_float(rowA[j]); 339 ak = half_to_float(rowA[k]); 340 bj = half_to_float(rowB[j]); 341 bk = half_to_float(rowB[k]); 342 dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F); 343 } 344 } 345#endif 346 347 else if (datatype == UINT && comps == 1) { 348 uint i, j, k; 349 const uint *rowA = (const uint *) srcRowA; 350 const uint *rowB = (const uint *) srcRowB; 351 uint *dst = (uint *) dstRow; 352 for (i = j = 0, k = k0; i < (uint) dstWidth; 353 i++, j += colStride, k += colStride) { 354 dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4; 355 } 356 } 357 358 else if (datatype == USHORT_5_6_5 && comps == 3) { 359 uint i, j, k; 360 const ushort *rowA = (const ushort *) srcRowA; 361 const ushort *rowB = (const ushort *) srcRowB; 362 ushort *dst = (ushort *) dstRow; 363 for (i = j = 0, k = k0; i < (uint) dstWidth; 364 i++, j += colStride, k += colStride) { 365 const int rowAr0 = rowA[j] & 0x1f; 366 const int rowAr1 = rowA[k] & 0x1f; 367 const int rowBr0 = rowB[j] & 0x1f; 368 const int rowBr1 = rowB[k] & 0x1f; 369 const int rowAg0 = (rowA[j] >> 5) & 0x3f; 370 const int rowAg1 = (rowA[k] >> 5) & 0x3f; 371 const int rowBg0 = (rowB[j] >> 5) & 0x3f; 372 const int rowBg1 = (rowB[k] >> 5) & 0x3f; 373 const int rowAb0 = (rowA[j] >> 11) & 0x1f; 374 const int rowAb1 = (rowA[k] >> 11) & 0x1f; 375 const int rowBb0 = (rowB[j] >> 11) & 0x1f; 376 const int rowBb1 = (rowB[k] >> 11) & 0x1f; 377 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 378 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 379 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 380 dst[i] = (blue << 11) | (green << 5) | red; 381 } 382 } 383 else if (datatype == USHORT_4_4_4_4 && comps == 4) { 384 uint i, j, k; 385 const ushort *rowA = (const ushort *) srcRowA; 386 const ushort *rowB = (const ushort *) srcRowB; 387 ushort *dst = (ushort *) dstRow; 388 for (i = j = 0, k = k0; i < (uint) dstWidth; 389 i++, j += colStride, k += colStride) { 390 const int rowAr0 = rowA[j] & 0xf; 391 const int rowAr1 = rowA[k] & 0xf; 392 const int rowBr0 = rowB[j] & 0xf; 393 const int rowBr1 = rowB[k] & 0xf; 394 const int rowAg0 = (rowA[j] >> 4) & 0xf; 395 const int rowAg1 = (rowA[k] >> 4) & 0xf; 396 const int rowBg0 = (rowB[j] >> 4) & 0xf; 397 const int rowBg1 = (rowB[k] >> 4) & 0xf; 398 const int rowAb0 = (rowA[j] >> 8) & 0xf; 399 const int rowAb1 = (rowA[k] >> 8) & 0xf; 400 const int rowBb0 = (rowB[j] >> 8) & 0xf; 401 const int rowBb1 = (rowB[k] >> 8) & 0xf; 402 const int rowAa0 = (rowA[j] >> 12) & 0xf; 403 const int rowAa1 = (rowA[k] >> 12) & 0xf; 404 const int rowBa0 = (rowB[j] >> 12) & 0xf; 405 const int rowBa1 = (rowB[k] >> 12) & 0xf; 406 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 407 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 408 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 409 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 410 dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red; 411 } 412 } 413 else if (datatype == USHORT_1_5_5_5_REV && comps == 4) { 414 uint i, j, k; 415 const ushort *rowA = (const ushort *) srcRowA; 416 const ushort *rowB = (const ushort *) srcRowB; 417 ushort *dst = (ushort *) dstRow; 418 for (i = j = 0, k = k0; i < (uint) dstWidth; 419 i++, j += colStride, k += colStride) { 420 const int rowAr0 = rowA[j] & 0x1f; 421 const int rowAr1 = rowA[k] & 0x1f; 422 const int rowBr0 = rowB[j] & 0x1f; 423 const int rowBr1 = rowB[k] & 0xf; 424 const int rowAg0 = (rowA[j] >> 5) & 0x1f; 425 const int rowAg1 = (rowA[k] >> 5) & 0x1f; 426 const int rowBg0 = (rowB[j] >> 5) & 0x1f; 427 const int rowBg1 = (rowB[k] >> 5) & 0x1f; 428 const int rowAb0 = (rowA[j] >> 10) & 0x1f; 429 const int rowAb1 = (rowA[k] >> 10) & 0x1f; 430 const int rowBb0 = (rowB[j] >> 10) & 0x1f; 431 const int rowBb1 = (rowB[k] >> 10) & 0x1f; 432 const int rowAa0 = (rowA[j] >> 15) & 0x1; 433 const int rowAa1 = (rowA[k] >> 15) & 0x1; 434 const int rowBa0 = (rowB[j] >> 15) & 0x1; 435 const int rowBa1 = (rowB[k] >> 15) & 0x1; 436 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 437 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 438 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 439 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2; 440 dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red; 441 } 442 } 443 else if (datatype == UBYTE_3_3_2 && comps == 3) { 444 uint i, j, k; 445 const ubyte *rowA = (const ubyte *) srcRowA; 446 const ubyte *rowB = (const ubyte *) srcRowB; 447 ubyte *dst = (ubyte *) dstRow; 448 for (i = j = 0, k = k0; i < (uint) dstWidth; 449 i++, j += colStride, k += colStride) { 450 const int rowAr0 = rowA[j] & 0x3; 451 const int rowAr1 = rowA[k] & 0x3; 452 const int rowBr0 = rowB[j] & 0x3; 453 const int rowBr1 = rowB[k] & 0x3; 454 const int rowAg0 = (rowA[j] >> 2) & 0x7; 455 const int rowAg1 = (rowA[k] >> 2) & 0x7; 456 const int rowBg0 = (rowB[j] >> 2) & 0x7; 457 const int rowBg1 = (rowB[k] >> 2) & 0x7; 458 const int rowAb0 = (rowA[j] >> 5) & 0x7; 459 const int rowAb1 = (rowA[k] >> 5) & 0x7; 460 const int rowBb0 = (rowB[j] >> 5) & 0x7; 461 const int rowBb1 = (rowB[k] >> 5) & 0x7; 462 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2; 463 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2; 464 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2; 465 dst[i] = (blue << 5) | (green << 2) | red; 466 } 467 } 468 else { 469 debug_printf("bad format in do_row()"); 470 } 471} 472 473 474static void 475format_to_type_comps(enum pipe_format pformat, 476 enum dtype *datatype, uint *comps) 477{ 478 switch (pformat) { 479 case PIPE_FORMAT_A8R8G8B8_UNORM: 480 case PIPE_FORMAT_X8R8G8B8_UNORM: 481 case PIPE_FORMAT_B8G8R8A8_UNORM: 482 case PIPE_FORMAT_B8G8R8X8_UNORM: 483 *datatype = UBYTE; 484 *comps = 4; 485 return; 486 case PIPE_FORMAT_A1R5G5B5_UNORM: 487 *datatype = USHORT_1_5_5_5_REV; 488 *comps = 4; 489 return; 490 case PIPE_FORMAT_A4R4G4B4_UNORM: 491 *datatype = USHORT_4_4_4_4; 492 *comps = 4; 493 return; 494 case PIPE_FORMAT_R5G6B5_UNORM: 495 *datatype = USHORT_5_6_5; 496 *comps = 3; 497 return; 498 case PIPE_FORMAT_L8_UNORM: 499 case PIPE_FORMAT_A8_UNORM: 500 case PIPE_FORMAT_I8_UNORM: 501 *datatype = UBYTE; 502 *comps = 1; 503 return; 504 case PIPE_FORMAT_A8L8_UNORM: 505 *datatype = UBYTE; 506 *comps = 2; 507 return; 508 default: 509 assert(0); 510 *datatype = UBYTE; 511 *comps = 0; 512 break; 513 } 514} 515 516 517static void 518reduce_1d(enum pipe_format pformat, 519 int srcWidth, const ubyte *srcPtr, 520 int dstWidth, ubyte *dstPtr) 521{ 522 enum dtype datatype; 523 uint comps; 524 525 format_to_type_comps(pformat, &datatype, &comps); 526 527 /* we just duplicate the input row, kind of hack, saves code */ 528 do_row(datatype, comps, 529 srcWidth, srcPtr, srcPtr, 530 dstWidth, dstPtr); 531} 532 533 534/** 535 * Strides are in bytes. If zero, it'll be computed as width * bpp. 536 */ 537static void 538reduce_2d(enum pipe_format pformat, 539 int srcWidth, int srcHeight, 540 int srcRowStride, const ubyte *srcPtr, 541 int dstWidth, int dstHeight, 542 int dstRowStride, ubyte *dstPtr) 543{ 544 enum dtype datatype; 545 uint comps; 546 const int bpt = pf_get_size(pformat); 547 const ubyte *srcA, *srcB; 548 ubyte *dst; 549 int row; 550 551 format_to_type_comps(pformat, &datatype, &comps); 552 553 if (!srcRowStride) 554 srcRowStride = bpt * srcWidth; 555 556 if (!dstRowStride) 557 dstRowStride = bpt * dstWidth; 558 559 /* Compute src and dst pointers */ 560 srcA = srcPtr; 561 if (srcHeight > 1) 562 srcB = srcA + srcRowStride; 563 else 564 srcB = srcA; 565 dst = dstPtr; 566 567 for (row = 0; row < dstHeight; row++) { 568 do_row(datatype, comps, 569 srcWidth, srcA, srcB, 570 dstWidth, dst); 571 srcA += 2 * srcRowStride; 572 srcB += 2 * srcRowStride; 573 dst += dstRowStride; 574 } 575} 576 577 578static void 579make_1d_mipmap(struct gen_mipmap_state *ctx, 580 struct pipe_texture *pt, 581 uint face, uint baseLevel, uint lastLevel) 582{ 583 struct pipe_context *pipe = ctx->pipe; 584 struct pipe_screen *screen = pipe->screen; 585 const uint zslice = 0; 586 uint dstLevel; 587 588 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 589 const uint srcLevel = dstLevel - 1; 590 struct pipe_surface *srcSurf, *dstSurf; 591 void *srcMap, *dstMap; 592 593 srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice, 594 PIPE_BUFFER_USAGE_CPU_READ); 595 596 dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice, 597 PIPE_BUFFER_USAGE_CPU_WRITE); 598 599 srcMap = ((ubyte *) pipe_buffer_map(screen, srcSurf->buffer, 600 PIPE_BUFFER_USAGE_CPU_READ) 601 + srcSurf->offset); 602 dstMap = ((ubyte *) pipe_buffer_map(screen, dstSurf->buffer, 603 PIPE_BUFFER_USAGE_CPU_WRITE) 604 + dstSurf->offset); 605 606 reduce_1d(pt->format, 607 srcSurf->width, srcMap, 608 dstSurf->width, dstMap); 609 610 pipe_buffer_unmap(screen, srcSurf->buffer); 611 pipe_buffer_unmap(screen, dstSurf->buffer); 612 613 pipe_surface_reference(&srcSurf, NULL); 614 pipe_surface_reference(&dstSurf, NULL); 615 } 616} 617 618 619static void 620make_2d_mipmap(struct gen_mipmap_state *ctx, 621 struct pipe_texture *pt, 622 uint face, uint baseLevel, uint lastLevel) 623{ 624 struct pipe_context *pipe = ctx->pipe; 625 struct pipe_screen *screen = pipe->screen; 626 const uint zslice = 0; 627 uint dstLevel; 628 629 assert(pt->block.width == 1); 630 assert(pt->block.height == 1); 631 632 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 633 const uint srcLevel = dstLevel - 1; 634 struct pipe_surface *srcSurf, *dstSurf; 635 ubyte *srcMap, *dstMap; 636 637 srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice, 638 PIPE_BUFFER_USAGE_CPU_READ); 639 dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice, 640 PIPE_BUFFER_USAGE_CPU_WRITE); 641 642 srcMap = ((ubyte *) pipe_buffer_map(screen, srcSurf->buffer, 643 PIPE_BUFFER_USAGE_CPU_READ) 644 + srcSurf->offset); 645 dstMap = ((ubyte *) pipe_buffer_map(screen, dstSurf->buffer, 646 PIPE_BUFFER_USAGE_CPU_WRITE) 647 + dstSurf->offset); 648 649 reduce_2d(pt->format, 650 srcSurf->width, srcSurf->height, 651 srcSurf->stride, srcMap, 652 dstSurf->width, dstSurf->height, 653 dstSurf->stride, dstMap); 654 655 pipe_buffer_unmap(screen, srcSurf->buffer); 656 pipe_buffer_unmap(screen, dstSurf->buffer); 657 658 pipe_surface_reference(&srcSurf, NULL); 659 pipe_surface_reference(&dstSurf, NULL); 660 } 661} 662 663 664static void 665make_3d_mipmap(struct gen_mipmap_state *ctx, 666 struct pipe_texture *pt, 667 uint face, uint baseLevel, uint lastLevel) 668{ 669} 670 671 672static void 673fallback_gen_mipmap(struct gen_mipmap_state *ctx, 674 struct pipe_texture *pt, 675 uint face, uint baseLevel, uint lastLevel) 676{ 677 switch (pt->target) { 678 case PIPE_TEXTURE_1D: 679 make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel); 680 break; 681 case PIPE_TEXTURE_2D: 682 case PIPE_TEXTURE_CUBE: 683 make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel); 684 break; 685 case PIPE_TEXTURE_3D: 686 make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel); 687 break; 688 default: 689 assert(0); 690 } 691} 692 693 694/** 695 * Create a mipmap generation context. 696 * The idea is to create one of these and re-use it each time we need to 697 * generate a mipmap. 698 */ 699struct gen_mipmap_state * 700util_create_gen_mipmap(struct pipe_context *pipe, 701 struct cso_context *cso) 702{ 703 struct gen_mipmap_state *ctx; 704 uint i; 705 706 ctx = CALLOC_STRUCT(gen_mipmap_state); 707 if (!ctx) 708 return NULL; 709 710 ctx->pipe = pipe; 711 ctx->cso = cso; 712 713 /* disabled blending/masking */ 714 memset(&ctx->blend, 0, sizeof(ctx->blend)); 715 ctx->blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE; 716 ctx->blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE; 717 ctx->blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO; 718 ctx->blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO; 719 ctx->blend.colormask = PIPE_MASK_RGBA; 720 721 /* no-op depth/stencil/alpha */ 722 memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil)); 723 724 /* rasterizer */ 725 memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer)); 726 ctx->rasterizer.front_winding = PIPE_WINDING_CW; 727 ctx->rasterizer.cull_mode = PIPE_WINDING_NONE; 728 ctx->rasterizer.bypass_clipping = 1; 729 /*ctx->rasterizer.bypass_vs = 1;*/ 730 ctx->rasterizer.gl_rasterization_rules = 1; 731 732 /* sampler state */ 733 memset(&ctx->sampler, 0, sizeof(ctx->sampler)); 734 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 735 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 736 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 737 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST; 738 ctx->sampler.normalized_coords = 1; 739 740 /* viewport state (identity, verts are in wincoords) */ 741 ctx->viewport.scale[0] = 1.0; 742 ctx->viewport.scale[1] = 1.0; 743 ctx->viewport.scale[2] = 1.0; 744 ctx->viewport.scale[3] = 1.0; 745 ctx->viewport.translate[0] = 0.0; 746 ctx->viewport.translate[1] = 0.0; 747 ctx->viewport.translate[2] = 0.0; 748 ctx->viewport.translate[3] = 0.0; 749 750 /* vertex shader */ 751 { 752 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION, 753 TGSI_SEMANTIC_GENERIC }; 754 const uint semantic_indexes[] = { 0, 0 }; 755 ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names, 756 semantic_indexes, 757 &ctx->vert_shader); 758 } 759 760 /* fragment shader */ 761 ctx->fs = util_make_fragment_tex_shader(pipe, &ctx->frag_shader); 762 763 ctx->vbuf = pipe_buffer_create(pipe->screen, 764 32, 765 PIPE_BUFFER_USAGE_VERTEX, 766 sizeof(ctx->vertices)); 767 if (!ctx->vbuf) { 768 FREE(ctx); 769 return NULL; 770 } 771 772 /* vertex data that doesn't change */ 773 for (i = 0; i < 4; i++) { 774 ctx->vertices[i][0][2] = 0.0f; /* z */ 775 ctx->vertices[i][0][3] = 1.0f; /* w */ 776 ctx->vertices[i][1][2] = 0.0f; /* r */ 777 ctx->vertices[i][1][3] = 1.0f; /* q */ 778 } 779 780 return ctx; 781} 782 783 784static unsigned get_next_slot( struct gen_mipmap_state *ctx ) 785{ 786 const unsigned max_slots = 4096 / sizeof ctx->vertices; 787 788 if (ctx->vbuf_slot >= max_slots) 789 util_gen_mipmap_flush( ctx ); 790 791 if (!ctx->vbuf) { 792 ctx->vbuf = pipe_buffer_create(ctx->pipe->screen, 793 32, 794 PIPE_BUFFER_USAGE_VERTEX, 795 max_slots * sizeof ctx->vertices); 796 } 797 798 return ctx->vbuf_slot++ * sizeof ctx->vertices; 799} 800 801static unsigned 802set_vertex_data(struct gen_mipmap_state *ctx, float width, float height) 803{ 804 void *buf; 805 unsigned offset; 806 807 ctx->vertices[0][0][0] = 0.0f; /*x*/ 808 ctx->vertices[0][0][1] = 0.0f; /*y*/ 809 ctx->vertices[0][1][0] = 0.0f; /*s*/ 810 ctx->vertices[0][1][1] = 0.0f; /*t*/ 811 812 ctx->vertices[1][0][0] = width; 813 ctx->vertices[1][0][1] = 0.0f; 814 ctx->vertices[1][1][0] = 1.0f; 815 ctx->vertices[1][1][1] = 0.0f; 816 817 ctx->vertices[2][0][0] = width; 818 ctx->vertices[2][0][1] = height; 819 ctx->vertices[2][1][0] = 1.0f; 820 ctx->vertices[2][1][1] = 1.0f; 821 822 ctx->vertices[3][0][0] = 0.0f; 823 ctx->vertices[3][0][1] = height; 824 ctx->vertices[3][1][0] = 0.0f; 825 ctx->vertices[3][1][1] = 1.0f; 826 827 offset = get_next_slot( ctx ); 828 829 buf = pipe_buffer_map(ctx->pipe->screen, ctx->vbuf, 830 PIPE_BUFFER_USAGE_CPU_WRITE); 831 832 memcpy((char *)buf + offset, ctx->vertices, sizeof(ctx->vertices)); 833 834 pipe_buffer_unmap(ctx->pipe->screen, ctx->vbuf); 835 836 return offset; 837} 838 839 840 841/** 842 * Destroy a mipmap generation context 843 */ 844void 845util_destroy_gen_mipmap(struct gen_mipmap_state *ctx) 846{ 847 struct pipe_context *pipe = ctx->pipe; 848 849 pipe->delete_vs_state(pipe, ctx->vs); 850 pipe->delete_fs_state(pipe, ctx->fs); 851 852 FREE((void*) ctx->vert_shader.tokens); 853 FREE((void*) ctx->frag_shader.tokens); 854 855 pipe_buffer_reference(pipe->screen, &ctx->vbuf, NULL); 856 857 FREE(ctx); 858} 859 860 861 862/* Release vertex buffer at end of frame to avoid synchronous 863 * rendering. 864 */ 865void util_gen_mipmap_flush( struct gen_mipmap_state *ctx ) 866{ 867 pipe_buffer_reference(ctx->pipe->screen, &ctx->vbuf, NULL); 868 ctx->vbuf_slot = 0; 869} 870 871 872/** 873 * Generate mipmap images. It's assumed all needed texture memory is 874 * already allocated. 875 * 876 * \param pt the texture to generate mipmap levels for 877 * \param face which cube face to generate mipmaps for (0 for non-cube maps) 878 * \param baseLevel the first mipmap level to use as a src 879 * \param lastLevel the last mipmap level to generate 880 * \param filter the minification filter used to generate mipmap levels with 881 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST 882 */ 883void 884util_gen_mipmap(struct gen_mipmap_state *ctx, 885 struct pipe_texture *pt, 886 uint face, uint baseLevel, uint lastLevel, uint filter) 887{ 888 struct pipe_context *pipe = ctx->pipe; 889 struct pipe_screen *screen = pipe->screen; 890 struct pipe_framebuffer_state fb; 891 uint dstLevel; 892 uint zslice = 0; 893 uint offset; 894 895 /* check if we can render in the texture's format */ 896 if (!screen->is_format_supported(screen, pt->format, PIPE_TEXTURE_2D, 897 PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) { 898 fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel); 899 return; 900 } 901 902 /* save state (restored below) */ 903 cso_save_blend(ctx->cso); 904 cso_save_depth_stencil_alpha(ctx->cso); 905 cso_save_rasterizer(ctx->cso); 906 cso_save_samplers(ctx->cso); 907 cso_save_sampler_textures(ctx->cso); 908 cso_save_framebuffer(ctx->cso); 909 cso_save_fragment_shader(ctx->cso); 910 cso_save_vertex_shader(ctx->cso); 911 cso_save_viewport(ctx->cso); 912 913 /* bind our state */ 914 cso_set_blend(ctx->cso, &ctx->blend); 915 cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil); 916 cso_set_rasterizer(ctx->cso, &ctx->rasterizer); 917 cso_set_viewport(ctx->cso, &ctx->viewport); 918 919 cso_set_fragment_shader_handle(ctx->cso, ctx->fs); 920 cso_set_vertex_shader_handle(ctx->cso, ctx->vs); 921 922 /* init framebuffer state */ 923 memset(&fb, 0, sizeof(fb)); 924 fb.num_cbufs = 1; 925 926 /* set min/mag to same filter for faster sw speed */ 927 ctx->sampler.mag_img_filter = filter; 928 ctx->sampler.min_img_filter = filter; 929 930 /* 931 * XXX for small mipmap levels, it may be faster to use the software 932 * fallback path... 933 */ 934 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) { 935 const uint srcLevel = dstLevel - 1; 936 937 struct pipe_surface *surf = 938 screen->get_tex_surface(screen, pt, face, dstLevel, zslice, 939 PIPE_BUFFER_USAGE_GPU_WRITE); 940 941 /* 942 * Setup framebuffer / dest surface 943 */ 944 fb.cbufs[0] = surf; 945 fb.width = pt->width[dstLevel]; 946 fb.height = pt->height[dstLevel]; 947 cso_set_framebuffer(ctx->cso, &fb); 948 949 /* 950 * Setup sampler state 951 * Note: we should only have to set the min/max LOD clamps to ensure 952 * we grab texels from the right mipmap level. But some hardware 953 * has trouble with min clamping so we also set the lod_bias to 954 * try to work around that. 955 */ 956 ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel; 957 ctx->sampler.lod_bias = (float) srcLevel; 958 cso_single_sampler(ctx->cso, 0, &ctx->sampler); 959 cso_single_sampler_done(ctx->cso); 960 961 cso_set_sampler_textures(ctx->cso, 1, &pt); 962 963 /* quad coords in window coords (bypassing clipping, viewport mapping) */ 964 offset = set_vertex_data(ctx, 965 (float) pt->width[dstLevel], 966 (float) pt->height[dstLevel]); 967 968 util_draw_vertex_buffer(ctx->pipe, 969 ctx->vbuf, 970 offset, 971 PIPE_PRIM_TRIANGLE_FAN, 972 4, /* verts */ 973 2); /* attribs/vert */ 974 975 pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL); 976 977 /* need to signal that the texture has changed _after_ rendering to it */ 978 pipe_surface_reference( &surf, NULL ); 979 } 980 981 /* restore state we changed */ 982 cso_restore_blend(ctx->cso); 983 cso_restore_depth_stencil_alpha(ctx->cso); 984 cso_restore_rasterizer(ctx->cso); 985 cso_restore_samplers(ctx->cso); 986 cso_restore_sampler_textures(ctx->cso); 987 cso_restore_framebuffer(ctx->cso); 988 cso_restore_fragment_shader(ctx->cso); 989 cso_restore_vertex_shader(ctx->cso); 990 cso_restore_viewport(ctx->cso); 991} 992