1 /**************************************************************************
3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2008 VMware, Inc. All rights reserved.
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:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
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.
27 **************************************************************************/
31 * Mipmap generation utility
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"
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"
52 #include "cso_cache/cso_context.h"
55 struct gen_mipmap_state
57 struct pipe_context *pipe;
58 struct cso_context *cso;
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_clip_state clip;
65 struct pipe_vertex_element velem[2];
70 struct pipe_resource *vbuf; /**< quad vertices */
73 float vertices[4][2][4]; /**< vertex/texcoords for quad */
85 DTYPE_USHORT_1_5_5_5_REV,
92 typedef ushort half_float;
96 float_to_half(float f)
103 half_to_float(half_float h)
113 * \name Support macros for do_row and do_row_3d
115 * The macro madness is here for two reasons. First, it compacts the code
116 * slightly. Second, it makes it much easier to adjust the specifics of the
117 * filter to tune the rounding characteristics.
120 #define DECLARE_ROW_POINTERS(t, e) \
121 const t(*rowA)[e] = (const t(*)[e]) srcRowA; \
122 const t(*rowB)[e] = (const t(*)[e]) srcRowB; \
123 const t(*rowC)[e] = (const t(*)[e]) srcRowC; \
124 const t(*rowD)[e] = (const t(*)[e]) srcRowD; \
125 t(*dst)[e] = (t(*)[e]) dstRow
127 #define DECLARE_ROW_POINTERS0(t) \
128 const t *rowA = (const t *) srcRowA; \
129 const t *rowB = (const t *) srcRowB; \
130 const t *rowC = (const t *) srcRowC; \
131 const t *rowD = (const t *) srcRowD; \
132 t *dst = (t *) dstRow
134 #define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
135 ((unsigned) Aj + (unsigned) Ak \
136 + (unsigned) Bj + (unsigned) Bk \
137 + (unsigned) Cj + (unsigned) Ck \
138 + (unsigned) Dj + (unsigned) Dk \
141 #define FILTER_3D(e) \
143 dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \
144 rowB[j][e], rowB[k][e], \
145 rowC[j][e], rowC[k][e], \
146 rowD[j][e], rowD[k][e]); \
149 #define FILTER_F_3D(e) \
151 dst[i][e] = (rowA[j][e] + rowA[k][e] \
152 + rowB[j][e] + rowB[k][e] \
153 + rowC[j][e] + rowC[k][e] \
154 + rowD[j][e] + rowD[k][e]) * 0.125F; \
157 #define FILTER_HF_3D(e) \
159 const float aj = half_to_float(rowA[j][e]); \
160 const float ak = half_to_float(rowA[k][e]); \
161 const float bj = half_to_float(rowB[j][e]); \
162 const float bk = half_to_float(rowB[k][e]); \
163 const float cj = half_to_float(rowC[j][e]); \
164 const float ck = half_to_float(rowC[k][e]); \
165 const float dj = half_to_float(rowD[j][e]); \
166 const float dk = half_to_float(rowD[k][e]); \
167 dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
174 * Average together two rows of a source image to produce a single new
175 * row in the dest image. It's legal for the two source rows to point
176 * to the same data. The source width must be equal to either the
177 * dest width or two times the dest width.
178 * \param datatype GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
179 * \param comps number of components per pixel (1..4)
182 do_row(enum dtype datatype, uint comps, int srcWidth,
183 const void *srcRowA, const void *srcRowB,
184 int dstWidth, void *dstRow)
186 const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
187 const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
192 /* This assertion is no longer valid with non-power-of-2 textures
193 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
196 if (datatype == DTYPE_UBYTE && comps == 4) {
198 const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA;
199 const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB;
200 ubyte(*dst)[4] = (ubyte(*)[4]) dstRow;
201 for (i = j = 0, k = k0; i < (uint) dstWidth;
202 i++, j += colStride, k += colStride) {
203 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
204 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
205 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
206 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
209 else if (datatype == DTYPE_UBYTE && comps == 3) {
211 const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA;
212 const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB;
213 ubyte(*dst)[3] = (ubyte(*)[3]) dstRow;
214 for (i = j = 0, k = k0; i < (uint) dstWidth;
215 i++, j += colStride, k += colStride) {
216 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
217 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
218 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
221 else if (datatype == DTYPE_UBYTE && comps == 2) {
223 const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA;
224 const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB;
225 ubyte(*dst)[2] = (ubyte(*)[2]) dstRow;
226 for (i = j = 0, k = k0; i < (uint) dstWidth;
227 i++, j += colStride, k += colStride) {
228 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
229 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
232 else if (datatype == DTYPE_UBYTE && comps == 1) {
234 const ubyte *rowA = (const ubyte *) srcRowA;
235 const ubyte *rowB = (const ubyte *) srcRowB;
236 ubyte *dst = (ubyte *) dstRow;
237 for (i = j = 0, k = k0; i < (uint) dstWidth;
238 i++, j += colStride, k += colStride) {
239 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
243 else if (datatype == DTYPE_USHORT && comps == 4) {
245 const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA;
246 const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB;
247 ushort(*dst)[4] = (ushort(*)[4]) dstRow;
248 for (i = j = 0, k = k0; i < (uint) dstWidth;
249 i++, j += colStride, k += colStride) {
250 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
251 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
252 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
253 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
256 else if (datatype == DTYPE_USHORT && comps == 3) {
258 const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA;
259 const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB;
260 ushort(*dst)[3] = (ushort(*)[3]) dstRow;
261 for (i = j = 0, k = k0; i < (uint) dstWidth;
262 i++, j += colStride, k += colStride) {
263 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
264 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
265 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
268 else if (datatype == DTYPE_USHORT && comps == 2) {
270 const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA;
271 const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB;
272 ushort(*dst)[2] = (ushort(*)[2]) dstRow;
273 for (i = j = 0, k = k0; i < (uint) dstWidth;
274 i++, j += colStride, k += colStride) {
275 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
276 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
279 else if (datatype == DTYPE_USHORT && comps == 1) {
281 const ushort *rowA = (const ushort *) srcRowA;
282 const ushort *rowB = (const ushort *) srcRowB;
283 ushort *dst = (ushort *) dstRow;
284 for (i = j = 0, k = k0; i < (uint) dstWidth;
285 i++, j += colStride, k += colStride) {
286 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
290 else if (datatype == DTYPE_FLOAT && comps == 4) {
292 const float(*rowA)[4] = (const float(*)[4]) srcRowA;
293 const float(*rowB)[4] = (const float(*)[4]) srcRowB;
294 float(*dst)[4] = (float(*)[4]) dstRow;
295 for (i = j = 0, k = k0; i < (uint) dstWidth;
296 i++, j += colStride, k += colStride) {
297 dst[i][0] = (rowA[j][0] + rowA[k][0] +
298 rowB[j][0] + rowB[k][0]) * 0.25F;
299 dst[i][1] = (rowA[j][1] + rowA[k][1] +
300 rowB[j][1] + rowB[k][1]) * 0.25F;
301 dst[i][2] = (rowA[j][2] + rowA[k][2] +
302 rowB[j][2] + rowB[k][2]) * 0.25F;
303 dst[i][3] = (rowA[j][3] + rowA[k][3] +
304 rowB[j][3] + rowB[k][3]) * 0.25F;
307 else if (datatype == DTYPE_FLOAT && comps == 3) {
309 const float(*rowA)[3] = (const float(*)[3]) srcRowA;
310 const float(*rowB)[3] = (const float(*)[3]) srcRowB;
311 float(*dst)[3] = (float(*)[3]) 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 dst[i][2] = (rowA[j][2] + rowA[k][2] +
319 rowB[j][2] + rowB[k][2]) * 0.25F;
322 else if (datatype == DTYPE_FLOAT && comps == 2) {
324 const float(*rowA)[2] = (const float(*)[2]) srcRowA;
325 const float(*rowB)[2] = (const float(*)[2]) srcRowB;
326 float(*dst)[2] = (float(*)[2]) dstRow;
327 for (i = j = 0, k = k0; i < (uint) dstWidth;
328 i++, j += colStride, k += colStride) {
329 dst[i][0] = (rowA[j][0] + rowA[k][0] +
330 rowB[j][0] + rowB[k][0]) * 0.25F;
331 dst[i][1] = (rowA[j][1] + rowA[k][1] +
332 rowB[j][1] + rowB[k][1]) * 0.25F;
335 else if (datatype == DTYPE_FLOAT && comps == 1) {
337 const float *rowA = (const float *) srcRowA;
338 const float *rowB = (const float *) srcRowB;
339 float *dst = (float *) dstRow;
340 for (i = j = 0, k = k0; i < (uint) dstWidth;
341 i++, j += colStride, k += colStride) {
342 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
347 else if (datatype == HALF_DTYPE_FLOAT && comps == 4) {
349 const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA;
350 const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB;
351 half_float(*dst)[4] = (half_float(*)[4]) dstRow;
352 for (i = j = 0, k = k0; i < (uint) dstWidth;
353 i++, j += colStride, k += colStride) {
354 for (comp = 0; comp < 4; comp++) {
355 float aj, ak, bj, bk;
356 aj = half_to_float(rowA[j][comp]);
357 ak = half_to_float(rowA[k][comp]);
358 bj = half_to_float(rowB[j][comp]);
359 bk = half_to_float(rowB[k][comp]);
360 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
364 else if (datatype == DTYPE_HALF_FLOAT && comps == 3) {
366 const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA;
367 const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB;
368 half_float(*dst)[3] = (half_float(*)[3]) dstRow;
369 for (i = j = 0, k = k0; i < (uint) dstWidth;
370 i++, j += colStride, k += colStride) {
371 for (comp = 0; comp < 3; comp++) {
372 float aj, ak, bj, bk;
373 aj = half_to_float(rowA[j][comp]);
374 ak = half_to_float(rowA[k][comp]);
375 bj = half_to_float(rowB[j][comp]);
376 bk = half_to_float(rowB[k][comp]);
377 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
381 else if (datatype == DTYPE_HALF_FLOAT && comps == 2) {
383 const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA;
384 const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB;
385 half_float(*dst)[2] = (half_float(*)[2]) dstRow;
386 for (i = j = 0, k = k0; i < (uint) dstWidth;
387 i++, j += colStride, k += colStride) {
388 for (comp = 0; comp < 2; comp++) {
389 float aj, ak, bj, bk;
390 aj = half_to_float(rowA[j][comp]);
391 ak = half_to_float(rowA[k][comp]);
392 bj = half_to_float(rowB[j][comp]);
393 bk = half_to_float(rowB[k][comp]);
394 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
398 else if (datatype == DTYPE_HALF_FLOAT && comps == 1) {
400 const half_float *rowA = (const half_float *) srcRowA;
401 const half_float *rowB = (const half_float *) srcRowB;
402 half_float *dst = (half_float *) dstRow;
403 for (i = j = 0, k = k0; i < (uint) dstWidth;
404 i++, j += colStride, k += colStride) {
405 float aj, ak, bj, bk;
406 aj = half_to_float(rowA[j]);
407 ak = half_to_float(rowA[k]);
408 bj = half_to_float(rowB[j]);
409 bk = half_to_float(rowB[k]);
410 dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F);
415 else if (datatype == DTYPE_UINT && comps == 1) {
417 const uint *rowA = (const uint *) srcRowA;
418 const uint *rowB = (const uint *) srcRowB;
419 uint *dst = (uint *) dstRow;
420 for (i = j = 0, k = k0; i < (uint) dstWidth;
421 i++, j += colStride, k += colStride) {
422 dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
426 else if (datatype == DTYPE_USHORT_5_6_5 && comps == 3) {
428 const ushort *rowA = (const ushort *) srcRowA;
429 const ushort *rowB = (const ushort *) srcRowB;
430 ushort *dst = (ushort *) dstRow;
431 for (i = j = 0, k = k0; i < (uint) dstWidth;
432 i++, j += colStride, k += colStride) {
433 const int rowAr0 = rowA[j] & 0x1f;
434 const int rowAr1 = rowA[k] & 0x1f;
435 const int rowBr0 = rowB[j] & 0x1f;
436 const int rowBr1 = rowB[k] & 0x1f;
437 const int rowAg0 = (rowA[j] >> 5) & 0x3f;
438 const int rowAg1 = (rowA[k] >> 5) & 0x3f;
439 const int rowBg0 = (rowB[j] >> 5) & 0x3f;
440 const int rowBg1 = (rowB[k] >> 5) & 0x3f;
441 const int rowAb0 = (rowA[j] >> 11) & 0x1f;
442 const int rowAb1 = (rowA[k] >> 11) & 0x1f;
443 const int rowBb0 = (rowB[j] >> 11) & 0x1f;
444 const int rowBb1 = (rowB[k] >> 11) & 0x1f;
445 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
446 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
447 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
448 dst[i] = (blue << 11) | (green << 5) | red;
451 else if (datatype == DTYPE_USHORT_4_4_4_4 && comps == 4) {
453 const ushort *rowA = (const ushort *) srcRowA;
454 const ushort *rowB = (const ushort *) srcRowB;
455 ushort *dst = (ushort *) dstRow;
456 for (i = j = 0, k = k0; i < (uint) dstWidth;
457 i++, j += colStride, k += colStride) {
458 const int rowAr0 = rowA[j] & 0xf;
459 const int rowAr1 = rowA[k] & 0xf;
460 const int rowBr0 = rowB[j] & 0xf;
461 const int rowBr1 = rowB[k] & 0xf;
462 const int rowAg0 = (rowA[j] >> 4) & 0xf;
463 const int rowAg1 = (rowA[k] >> 4) & 0xf;
464 const int rowBg0 = (rowB[j] >> 4) & 0xf;
465 const int rowBg1 = (rowB[k] >> 4) & 0xf;
466 const int rowAb0 = (rowA[j] >> 8) & 0xf;
467 const int rowAb1 = (rowA[k] >> 8) & 0xf;
468 const int rowBb0 = (rowB[j] >> 8) & 0xf;
469 const int rowBb1 = (rowB[k] >> 8) & 0xf;
470 const int rowAa0 = (rowA[j] >> 12) & 0xf;
471 const int rowAa1 = (rowA[k] >> 12) & 0xf;
472 const int rowBa0 = (rowB[j] >> 12) & 0xf;
473 const int rowBa1 = (rowB[k] >> 12) & 0xf;
474 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
475 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
476 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
477 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
478 dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
481 else if (datatype == DTYPE_USHORT_1_5_5_5_REV && comps == 4) {
483 const ushort *rowA = (const ushort *) srcRowA;
484 const ushort *rowB = (const ushort *) srcRowB;
485 ushort *dst = (ushort *) dstRow;
486 for (i = j = 0, k = k0; i < (uint) dstWidth;
487 i++, j += colStride, k += colStride) {
488 const int rowAr0 = rowA[j] & 0x1f;
489 const int rowAr1 = rowA[k] & 0x1f;
490 const int rowBr0 = rowB[j] & 0x1f;
491 const int rowBr1 = rowB[k] & 0x1f;
492 const int rowAg0 = (rowA[j] >> 5) & 0x1f;
493 const int rowAg1 = (rowA[k] >> 5) & 0x1f;
494 const int rowBg0 = (rowB[j] >> 5) & 0x1f;
495 const int rowBg1 = (rowB[k] >> 5) & 0x1f;
496 const int rowAb0 = (rowA[j] >> 10) & 0x1f;
497 const int rowAb1 = (rowA[k] >> 10) & 0x1f;
498 const int rowBb0 = (rowB[j] >> 10) & 0x1f;
499 const int rowBb1 = (rowB[k] >> 10) & 0x1f;
500 const int rowAa0 = (rowA[j] >> 15) & 0x1;
501 const int rowAa1 = (rowA[k] >> 15) & 0x1;
502 const int rowBa0 = (rowB[j] >> 15) & 0x1;
503 const int rowBa1 = (rowB[k] >> 15) & 0x1;
504 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
505 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
506 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
507 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
508 dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
511 else if (datatype == DTYPE_UBYTE_3_3_2 && comps == 3) {
513 const ubyte *rowA = (const ubyte *) srcRowA;
514 const ubyte *rowB = (const ubyte *) srcRowB;
515 ubyte *dst = (ubyte *) dstRow;
516 for (i = j = 0, k = k0; i < (uint) dstWidth;
517 i++, j += colStride, k += colStride) {
518 const int rowAr0 = rowA[j] & 0x3;
519 const int rowAr1 = rowA[k] & 0x3;
520 const int rowBr0 = rowB[j] & 0x3;
521 const int rowBr1 = rowB[k] & 0x3;
522 const int rowAg0 = (rowA[j] >> 2) & 0x7;
523 const int rowAg1 = (rowA[k] >> 2) & 0x7;
524 const int rowBg0 = (rowB[j] >> 2) & 0x7;
525 const int rowBg1 = (rowB[k] >> 2) & 0x7;
526 const int rowAb0 = (rowA[j] >> 5) & 0x7;
527 const int rowAb1 = (rowA[k] >> 5) & 0x7;
528 const int rowBb0 = (rowB[j] >> 5) & 0x7;
529 const int rowBb1 = (rowB[k] >> 5) & 0x7;
530 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
531 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
532 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
533 dst[i] = (blue << 5) | (green << 2) | red;
537 debug_printf("bad format in do_row()");
543 * Average together four rows of a source image to produce a single new
544 * row in the dest image. It's legal for the two source rows to point
545 * to the same data. The source width must be equal to either the
546 * dest width or two times the dest width.
548 * \param datatype GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
550 * \param comps number of components per pixel (1..4)
551 * \param srcWidth Width of a row in the source data
552 * \param srcRowA Pointer to one of the rows of source data
553 * \param srcRowB Pointer to one of the rows of source data
554 * \param srcRowC Pointer to one of the rows of source data
555 * \param srcRowD Pointer to one of the rows of source data
556 * \param dstWidth Width of a row in the destination data
557 * \param srcRowA Pointer to the row of destination data
560 do_row_3D(enum dtype datatype, uint comps, int srcWidth,
561 const void *srcRowA, const void *srcRowB,
562 const void *srcRowC, const void *srcRowD,
563 int dstWidth, void *dstRow)
565 const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
566 const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
572 if ((datatype == DTYPE_UBYTE) && (comps == 4)) {
573 DECLARE_ROW_POINTERS(ubyte, 4);
575 for (i = j = 0, k = k0; i < (uint) dstWidth;
576 i++, j += colStride, k += colStride) {
583 else if ((datatype == DTYPE_UBYTE) && (comps == 3)) {
584 DECLARE_ROW_POINTERS(ubyte, 3);
586 for (i = j = 0, k = k0; i < (uint) dstWidth;
587 i++, j += colStride, k += colStride) {
593 else if ((datatype == DTYPE_UBYTE) && (comps == 2)) {
594 DECLARE_ROW_POINTERS(ubyte, 2);
596 for (i = j = 0, k = k0; i < (uint) dstWidth;
597 i++, j += colStride, k += colStride) {
602 else if ((datatype == DTYPE_UBYTE) && (comps == 1)) {
603 DECLARE_ROW_POINTERS(ubyte, 1);
605 for (i = j = 0, k = k0; i < (uint) dstWidth;
606 i++, j += colStride, k += colStride) {
610 else if ((datatype == DTYPE_USHORT) && (comps == 4)) {
611 DECLARE_ROW_POINTERS(ushort, 4);
613 for (i = j = 0, k = k0; i < (uint) dstWidth;
614 i++, j += colStride, k += colStride) {
621 else if ((datatype == DTYPE_USHORT) && (comps == 3)) {
622 DECLARE_ROW_POINTERS(ushort, 3);
624 for (i = j = 0, k = k0; i < (uint) dstWidth;
625 i++, j += colStride, k += colStride) {
631 else if ((datatype == DTYPE_USHORT) && (comps == 2)) {
632 DECLARE_ROW_POINTERS(ushort, 2);
634 for (i = j = 0, k = k0; i < (uint) dstWidth;
635 i++, j += colStride, k += colStride) {
640 else if ((datatype == DTYPE_USHORT) && (comps == 1)) {
641 DECLARE_ROW_POINTERS(ushort, 1);
643 for (i = j = 0, k = k0; i < (uint) dstWidth;
644 i++, j += colStride, k += colStride) {
648 else if ((datatype == DTYPE_FLOAT) && (comps == 4)) {
649 DECLARE_ROW_POINTERS(float, 4);
651 for (i = j = 0, k = k0; i < (uint) dstWidth;
652 i++, j += colStride, k += colStride) {
659 else if ((datatype == DTYPE_FLOAT) && (comps == 3)) {
660 DECLARE_ROW_POINTERS(float, 3);
662 for (i = j = 0, k = k0; i < (uint) dstWidth;
663 i++, j += colStride, k += colStride) {
669 else if ((datatype == DTYPE_FLOAT) && (comps == 2)) {
670 DECLARE_ROW_POINTERS(float, 2);
672 for (i = j = 0, k = k0; i < (uint) dstWidth;
673 i++, j += colStride, k += colStride) {
678 else if ((datatype == DTYPE_FLOAT) && (comps == 1)) {
679 DECLARE_ROW_POINTERS(float, 1);
681 for (i = j = 0, k = k0; i < (uint) dstWidth;
682 i++, j += colStride, k += colStride) {
686 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 4)) {
687 DECLARE_ROW_POINTERS(half_float, 4);
689 for (i = j = 0, k = k0; i < (uint) dstWidth;
690 i++, j += colStride, k += colStride) {
697 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 3)) {
698 DECLARE_ROW_POINTERS(half_float, 4);
700 for (i = j = 0, k = k0; i < (uint) dstWidth;
701 i++, j += colStride, k += colStride) {
707 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 2)) {
708 DECLARE_ROW_POINTERS(half_float, 4);
710 for (i = j = 0, k = k0; i < (uint) dstWidth;
711 i++, j += colStride, k += colStride) {
716 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 1)) {
717 DECLARE_ROW_POINTERS(half_float, 4);
719 for (i = j = 0, k = k0; i < (uint) dstWidth;
720 i++, j += colStride, k += colStride) {
724 else if ((datatype == DTYPE_UINT) && (comps == 1)) {
725 const uint *rowA = (const uint *) srcRowA;
726 const uint *rowB = (const uint *) srcRowB;
727 const uint *rowC = (const uint *) srcRowC;
728 const uint *rowD = (const uint *) srcRowD;
729 float *dst = (float *) dstRow;
731 for (i = j = 0, k = k0; i < (uint) dstWidth;
732 i++, j += colStride, k += colStride) {
733 const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k])
734 + ((uint64_t) rowB[j] + (uint64_t) rowB[k])
735 + ((uint64_t) rowC[j] + (uint64_t) rowC[k])
736 + ((uint64_t) rowD[j] + (uint64_t) rowD[k]));
737 dst[i] = (float)((double) tmp * 0.125);
740 else if ((datatype == DTYPE_USHORT_5_6_5) && (comps == 3)) {
741 DECLARE_ROW_POINTERS0(ushort);
743 for (i = j = 0, k = k0; i < (uint) dstWidth;
744 i++, j += colStride, k += colStride) {
745 const int rowAr0 = rowA[j] & 0x1f;
746 const int rowAr1 = rowA[k] & 0x1f;
747 const int rowBr0 = rowB[j] & 0x1f;
748 const int rowBr1 = rowB[k] & 0x1f;
749 const int rowCr0 = rowC[j] & 0x1f;
750 const int rowCr1 = rowC[k] & 0x1f;
751 const int rowDr0 = rowD[j] & 0x1f;
752 const int rowDr1 = rowD[k] & 0x1f;
753 const int rowAg0 = (rowA[j] >> 5) & 0x3f;
754 const int rowAg1 = (rowA[k] >> 5) & 0x3f;
755 const int rowBg0 = (rowB[j] >> 5) & 0x3f;
756 const int rowBg1 = (rowB[k] >> 5) & 0x3f;
757 const int rowCg0 = (rowC[j] >> 5) & 0x3f;
758 const int rowCg1 = (rowC[k] >> 5) & 0x3f;
759 const int rowDg0 = (rowD[j] >> 5) & 0x3f;
760 const int rowDg1 = (rowD[k] >> 5) & 0x3f;
761 const int rowAb0 = (rowA[j] >> 11) & 0x1f;
762 const int rowAb1 = (rowA[k] >> 11) & 0x1f;
763 const int rowBb0 = (rowB[j] >> 11) & 0x1f;
764 const int rowBb1 = (rowB[k] >> 11) & 0x1f;
765 const int rowCb0 = (rowC[j] >> 11) & 0x1f;
766 const int rowCb1 = (rowC[k] >> 11) & 0x1f;
767 const int rowDb0 = (rowD[j] >> 11) & 0x1f;
768 const int rowDb1 = (rowD[k] >> 11) & 0x1f;
769 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
770 rowCr0, rowCr1, rowDr0, rowDr1);
771 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
772 rowCg0, rowCg1, rowDg0, rowDg1);
773 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
774 rowCb0, rowCb1, rowDb0, rowDb1);
775 dst[i] = (b << 11) | (g << 5) | r;
778 else if ((datatype == DTYPE_USHORT_4_4_4_4) && (comps == 4)) {
779 DECLARE_ROW_POINTERS0(ushort);
781 for (i = j = 0, k = k0; i < (uint) dstWidth;
782 i++, j += colStride, k += colStride) {
783 const int rowAr0 = rowA[j] & 0xf;
784 const int rowAr1 = rowA[k] & 0xf;
785 const int rowBr0 = rowB[j] & 0xf;
786 const int rowBr1 = rowB[k] & 0xf;
787 const int rowCr0 = rowC[j] & 0xf;
788 const int rowCr1 = rowC[k] & 0xf;
789 const int rowDr0 = rowD[j] & 0xf;
790 const int rowDr1 = rowD[k] & 0xf;
791 const int rowAg0 = (rowA[j] >> 4) & 0xf;
792 const int rowAg1 = (rowA[k] >> 4) & 0xf;
793 const int rowBg0 = (rowB[j] >> 4) & 0xf;
794 const int rowBg1 = (rowB[k] >> 4) & 0xf;
795 const int rowCg0 = (rowC[j] >> 4) & 0xf;
796 const int rowCg1 = (rowC[k] >> 4) & 0xf;
797 const int rowDg0 = (rowD[j] >> 4) & 0xf;
798 const int rowDg1 = (rowD[k] >> 4) & 0xf;
799 const int rowAb0 = (rowA[j] >> 8) & 0xf;
800 const int rowAb1 = (rowA[k] >> 8) & 0xf;
801 const int rowBb0 = (rowB[j] >> 8) & 0xf;
802 const int rowBb1 = (rowB[k] >> 8) & 0xf;
803 const int rowCb0 = (rowC[j] >> 8) & 0xf;
804 const int rowCb1 = (rowC[k] >> 8) & 0xf;
805 const int rowDb0 = (rowD[j] >> 8) & 0xf;
806 const int rowDb1 = (rowD[k] >> 8) & 0xf;
807 const int rowAa0 = (rowA[j] >> 12) & 0xf;
808 const int rowAa1 = (rowA[k] >> 12) & 0xf;
809 const int rowBa0 = (rowB[j] >> 12) & 0xf;
810 const int rowBa1 = (rowB[k] >> 12) & 0xf;
811 const int rowCa0 = (rowC[j] >> 12) & 0xf;
812 const int rowCa1 = (rowC[k] >> 12) & 0xf;
813 const int rowDa0 = (rowD[j] >> 12) & 0xf;
814 const int rowDa1 = (rowD[k] >> 12) & 0xf;
815 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
816 rowCr0, rowCr1, rowDr0, rowDr1);
817 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
818 rowCg0, rowCg1, rowDg0, rowDg1);
819 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
820 rowCb0, rowCb1, rowDb0, rowDb1);
821 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
822 rowCa0, rowCa1, rowDa0, rowDa1);
824 dst[i] = (a << 12) | (b << 8) | (g << 4) | r;
827 else if ((datatype == DTYPE_USHORT_1_5_5_5_REV) && (comps == 4)) {
828 DECLARE_ROW_POINTERS0(ushort);
830 for (i = j = 0, k = k0; i < (uint) dstWidth;
831 i++, j += colStride, k += colStride) {
832 const int rowAr0 = rowA[j] & 0x1f;
833 const int rowAr1 = rowA[k] & 0x1f;
834 const int rowBr0 = rowB[j] & 0x1f;
835 const int rowBr1 = rowB[k] & 0x1f;
836 const int rowCr0 = rowC[j] & 0x1f;
837 const int rowCr1 = rowC[k] & 0x1f;
838 const int rowDr0 = rowD[j] & 0x1f;
839 const int rowDr1 = rowD[k] & 0x1f;
840 const int rowAg0 = (rowA[j] >> 5) & 0x1f;
841 const int rowAg1 = (rowA[k] >> 5) & 0x1f;
842 const int rowBg0 = (rowB[j] >> 5) & 0x1f;
843 const int rowBg1 = (rowB[k] >> 5) & 0x1f;
844 const int rowCg0 = (rowC[j] >> 5) & 0x1f;
845 const int rowCg1 = (rowC[k] >> 5) & 0x1f;
846 const int rowDg0 = (rowD[j] >> 5) & 0x1f;
847 const int rowDg1 = (rowD[k] >> 5) & 0x1f;
848 const int rowAb0 = (rowA[j] >> 10) & 0x1f;
849 const int rowAb1 = (rowA[k] >> 10) & 0x1f;
850 const int rowBb0 = (rowB[j] >> 10) & 0x1f;
851 const int rowBb1 = (rowB[k] >> 10) & 0x1f;
852 const int rowCb0 = (rowC[j] >> 10) & 0x1f;
853 const int rowCb1 = (rowC[k] >> 10) & 0x1f;
854 const int rowDb0 = (rowD[j] >> 10) & 0x1f;
855 const int rowDb1 = (rowD[k] >> 10) & 0x1f;
856 const int rowAa0 = (rowA[j] >> 15) & 0x1;
857 const int rowAa1 = (rowA[k] >> 15) & 0x1;
858 const int rowBa0 = (rowB[j] >> 15) & 0x1;
859 const int rowBa1 = (rowB[k] >> 15) & 0x1;
860 const int rowCa0 = (rowC[j] >> 15) & 0x1;
861 const int rowCa1 = (rowC[k] >> 15) & 0x1;
862 const int rowDa0 = (rowD[j] >> 15) & 0x1;
863 const int rowDa1 = (rowD[k] >> 15) & 0x1;
864 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
865 rowCr0, rowCr1, rowDr0, rowDr1);
866 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
867 rowCg0, rowCg1, rowDg0, rowDg1);
868 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
869 rowCb0, rowCb1, rowDb0, rowDb1);
870 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
871 rowCa0, rowCa1, rowDa0, rowDa1);
873 dst[i] = (a << 15) | (b << 10) | (g << 5) | r;
876 else if ((datatype == DTYPE_UBYTE_3_3_2) && (comps == 3)) {
877 DECLARE_ROW_POINTERS0(ushort);
879 for (i = j = 0, k = k0; i < (uint) dstWidth;
880 i++, j += colStride, k += colStride) {
881 const int rowAr0 = rowA[j] & 0x3;
882 const int rowAr1 = rowA[k] & 0x3;
883 const int rowBr0 = rowB[j] & 0x3;
884 const int rowBr1 = rowB[k] & 0x3;
885 const int rowCr0 = rowC[j] & 0x3;
886 const int rowCr1 = rowC[k] & 0x3;
887 const int rowDr0 = rowD[j] & 0x3;
888 const int rowDr1 = rowD[k] & 0x3;
889 const int rowAg0 = (rowA[j] >> 2) & 0x7;
890 const int rowAg1 = (rowA[k] >> 2) & 0x7;
891 const int rowBg0 = (rowB[j] >> 2) & 0x7;
892 const int rowBg1 = (rowB[k] >> 2) & 0x7;
893 const int rowCg0 = (rowC[j] >> 2) & 0x7;
894 const int rowCg1 = (rowC[k] >> 2) & 0x7;
895 const int rowDg0 = (rowD[j] >> 2) & 0x7;
896 const int rowDg1 = (rowD[k] >> 2) & 0x7;
897 const int rowAb0 = (rowA[j] >> 5) & 0x7;
898 const int rowAb1 = (rowA[k] >> 5) & 0x7;
899 const int rowBb0 = (rowB[j] >> 5) & 0x7;
900 const int rowBb1 = (rowB[k] >> 5) & 0x7;
901 const int rowCb0 = (rowC[j] >> 5) & 0x7;
902 const int rowCb1 = (rowC[k] >> 5) & 0x7;
903 const int rowDb0 = (rowD[j] >> 5) & 0x7;
904 const int rowDb1 = (rowD[k] >> 5) & 0x7;
905 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
906 rowCr0, rowCr1, rowDr0, rowDr1);
907 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
908 rowCg0, rowCg1, rowDg0, rowDg1);
909 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
910 rowCb0, rowCb1, rowDb0, rowDb1);
911 dst[i] = (b << 5) | (g << 2) | r;
915 debug_printf("bad format in do_row_3D()");
922 format_to_type_comps(enum pipe_format pformat,
923 enum dtype *datatype, uint *comps)
925 /* XXX I think this could be implemented in terms of the pf_*() functions */
927 case PIPE_FORMAT_B8G8R8A8_UNORM:
928 case PIPE_FORMAT_B8G8R8X8_UNORM:
929 case PIPE_FORMAT_A8R8G8B8_UNORM:
930 case PIPE_FORMAT_X8R8G8B8_UNORM:
931 case PIPE_FORMAT_A8B8G8R8_SRGB:
932 case PIPE_FORMAT_X8B8G8R8_SRGB:
933 case PIPE_FORMAT_B8G8R8A8_SRGB:
934 case PIPE_FORMAT_B8G8R8X8_SRGB:
935 case PIPE_FORMAT_A8R8G8B8_SRGB:
936 case PIPE_FORMAT_X8R8G8B8_SRGB:
937 case PIPE_FORMAT_R8G8B8_SRGB:
938 *datatype = DTYPE_UBYTE;
941 case PIPE_FORMAT_B5G5R5X1_UNORM:
942 case PIPE_FORMAT_B5G5R5A1_UNORM:
943 *datatype = DTYPE_USHORT_1_5_5_5_REV;
946 case PIPE_FORMAT_B4G4R4A4_UNORM:
947 *datatype = DTYPE_USHORT_4_4_4_4;
950 case PIPE_FORMAT_B5G6R5_UNORM:
951 *datatype = DTYPE_USHORT_5_6_5;
954 case PIPE_FORMAT_L8_UNORM:
955 case PIPE_FORMAT_L8_SRGB:
956 case PIPE_FORMAT_A8_UNORM:
957 case PIPE_FORMAT_I8_UNORM:
958 *datatype = DTYPE_UBYTE;
961 case PIPE_FORMAT_L8A8_UNORM:
962 case PIPE_FORMAT_L8A8_SRGB:
963 *datatype = DTYPE_UBYTE;
968 *datatype = DTYPE_UBYTE;
976 reduce_1d(enum pipe_format pformat,
977 int srcWidth, const ubyte *srcPtr,
978 int dstWidth, ubyte *dstPtr)
983 format_to_type_comps(pformat, &datatype, &comps);
985 /* we just duplicate the input row, kind of hack, saves code */
986 do_row(datatype, comps,
987 srcWidth, srcPtr, srcPtr,
993 * Strides are in bytes. If zero, it'll be computed as width * bpp.
996 reduce_2d(enum pipe_format pformat,
997 int srcWidth, int srcHeight,
998 int srcRowStride, const ubyte *srcPtr,
999 int dstWidth, int dstHeight,
1000 int dstRowStride, ubyte *dstPtr)
1002 enum dtype datatype;
1004 const int bpt = util_format_get_blocksize(pformat);
1005 const ubyte *srcA, *srcB;
1009 format_to_type_comps(pformat, &datatype, &comps);
1012 srcRowStride = bpt * srcWidth;
1015 dstRowStride = bpt * dstWidth;
1017 /* Compute src and dst pointers */
1020 srcB = srcA + srcRowStride;
1025 for (row = 0; row < dstHeight; row++) {
1026 do_row(datatype, comps,
1027 srcWidth, srcA, srcB,
1029 srcA += 2 * srcRowStride;
1030 srcB += 2 * srcRowStride;
1031 dst += dstRowStride;
1037 reduce_3d(enum pipe_format pformat,
1038 int srcWidth, int srcHeight, int srcDepth,
1039 int srcRowStride, const ubyte *srcPtr,
1040 int dstWidth, int dstHeight, int dstDepth,
1041 int dstRowStride, ubyte *dstPtr)
1043 const int bpt = util_format_get_blocksize(pformat);
1044 const int border = 0;
1046 int bytesPerSrcImage, bytesPerDstImage;
1047 int bytesPerSrcRow, bytesPerDstRow;
1048 int srcImageOffset, srcRowOffset;
1049 enum dtype datatype;
1052 format_to_type_comps(pformat, &datatype, &comps);
1054 bytesPerSrcImage = srcWidth * srcHeight * bpt;
1055 bytesPerDstImage = dstWidth * dstHeight * bpt;
1057 bytesPerSrcRow = srcWidth * bpt;
1058 bytesPerDstRow = dstWidth * bpt;
1060 /* Offset between adjacent src images to be averaged together */
1061 srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage;
1063 /* Offset between adjacent src rows to be averaged together */
1064 srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt;
1067 * Need to average together up to 8 src pixels for each dest pixel.
1068 * Break that down into 3 operations:
1069 * 1. take two rows from source image and average them together.
1070 * 2. take two rows from next source image and average them together.
1071 * 3. take the two averaged rows and average them for the final dst row.
1075 printf("mip3d %d x %d x %d -> %d x %d x %d\n",
1076 srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
1079 for (img = 0; img < dstDepth; img++) {
1080 /* first source image pointer, skipping border */
1081 const ubyte *imgSrcA = srcPtr
1082 + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border
1083 + img * (bytesPerSrcImage + srcImageOffset);
1084 /* second source image pointer, skipping border */
1085 const ubyte *imgSrcB = imgSrcA + srcImageOffset;
1086 /* address of the dest image, skipping border */
1087 ubyte *imgDst = dstPtr
1088 + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border
1089 + img * bytesPerDstImage;
1091 /* setup the four source row pointers and the dest row pointer */
1092 const ubyte *srcImgARowA = imgSrcA;
1093 const ubyte *srcImgARowB = imgSrcA + srcRowOffset;
1094 const ubyte *srcImgBRowA = imgSrcB;
1095 const ubyte *srcImgBRowB = imgSrcB + srcRowOffset;
1096 ubyte *dstImgRow = imgDst;
1098 for (row = 0; row < dstHeight; row++) {
1099 do_row_3D(datatype, comps, srcWidth,
1100 srcImgARowA, srcImgARowB,
1101 srcImgBRowA, srcImgBRowB,
1102 dstWidth, dstImgRow);
1104 /* advance to next rows */
1105 srcImgARowA += bytesPerSrcRow + srcRowOffset;
1106 srcImgARowB += bytesPerSrcRow + srcRowOffset;
1107 srcImgBRowA += bytesPerSrcRow + srcRowOffset;
1108 srcImgBRowB += bytesPerSrcRow + srcRowOffset;
1109 dstImgRow += bytesPerDstRow;
1118 make_1d_mipmap(struct gen_mipmap_state *ctx,
1119 struct pipe_resource *pt,
1120 uint face, uint baseLevel, uint lastLevel)
1122 struct pipe_context *pipe = ctx->pipe;
1123 const uint zslice = 0;
1126 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1127 const uint srcLevel = dstLevel - 1;
1128 struct pipe_transfer *srcTrans, *dstTrans;
1129 void *srcMap, *dstMap;
1131 srcTrans = pipe_get_transfer(pipe, pt, face, srcLevel, zslice,
1132 PIPE_TRANSFER_READ, 0, 0,
1133 u_minify(pt->width0, srcLevel),
1134 u_minify(pt->height0, srcLevel));
1135 dstTrans = pipe_get_transfer(pipe, pt, face, dstLevel, zslice,
1136 PIPE_TRANSFER_WRITE, 0, 0,
1137 u_minify(pt->width0, dstLevel),
1138 u_minify(pt->height0, dstLevel));
1140 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1141 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1143 reduce_1d(pt->format,
1144 srcTrans->box.width, srcMap,
1145 dstTrans->box.width, dstMap);
1147 pipe->transfer_unmap(pipe, srcTrans);
1148 pipe->transfer_unmap(pipe, dstTrans);
1150 pipe->transfer_destroy(pipe, srcTrans);
1151 pipe->transfer_destroy(pipe, dstTrans);
1157 make_2d_mipmap(struct gen_mipmap_state *ctx,
1158 struct pipe_resource *pt,
1159 uint face, uint baseLevel, uint lastLevel)
1161 struct pipe_context *pipe = ctx->pipe;
1162 const uint zslice = 0;
1165 assert(util_format_get_blockwidth(pt->format) == 1);
1166 assert(util_format_get_blockheight(pt->format) == 1);
1168 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1169 const uint srcLevel = dstLevel - 1;
1170 struct pipe_transfer *srcTrans, *dstTrans;
1171 ubyte *srcMap, *dstMap;
1173 srcTrans = pipe_get_transfer(pipe, pt, face, srcLevel, zslice,
1174 PIPE_TRANSFER_READ, 0, 0,
1175 u_minify(pt->width0, srcLevel),
1176 u_minify(pt->height0, srcLevel));
1177 dstTrans = pipe_get_transfer(pipe, pt, face, dstLevel, zslice,
1178 PIPE_TRANSFER_WRITE, 0, 0,
1179 u_minify(pt->width0, dstLevel),
1180 u_minify(pt->height0, dstLevel));
1182 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1183 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1185 reduce_2d(pt->format,
1186 srcTrans->box.width, srcTrans->box.height,
1187 srcTrans->stride, srcMap,
1188 dstTrans->box.width, dstTrans->box.height,
1189 dstTrans->stride, dstMap);
1191 pipe->transfer_unmap(pipe, srcTrans);
1192 pipe->transfer_unmap(pipe, dstTrans);
1194 pipe->transfer_destroy(pipe, srcTrans);
1195 pipe->transfer_destroy(pipe, dstTrans);
1201 make_3d_mipmap(struct gen_mipmap_state *ctx,
1202 struct pipe_resource *pt,
1203 uint face, uint baseLevel, uint lastLevel)
1206 struct pipe_context *pipe = ctx->pipe;
1207 struct pipe_screen *screen = pipe->screen;
1208 uint dstLevel, zslice = 0;
1210 assert(util_format_get_blockwidth(pt->format) == 1);
1211 assert(util_format_get_blockheight(pt->format) == 1);
1213 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1214 const uint srcLevel = dstLevel - 1;
1215 struct pipe_transfer *srcTrans, *dstTrans;
1216 ubyte *srcMap, *dstMap;
1218 srcTrans = pipe->get_transfer(pipe, pt, face, srcLevel, zslice,
1219 PIPE_TRANSFER_READ, 0, 0,
1220 u_minify(pt->width0, srcLevel),
1221 u_minify(pt->height0, srcLevel));
1222 dstTrans = pipe->get_transfer(pipe, pt, face, dstLevel, zslice,
1223 PIPE_TRANSFER_WRITE, 0, 0,
1224 u_minify(pt->width0, dstLevel),
1225 u_minify(pt->height0, dstLevel));
1227 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1228 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1230 reduce_3d(pt->format,
1231 srcTrans->width, srcTrans->height,
1232 srcTrans->stride, srcMap,
1233 dstTrans->width, dstTrans->height,
1234 dstTrans->stride, dstMap);
1236 pipe->transfer_unmap(pipe, srcTrans);
1237 pipe->transfer_unmap(pipe, dstTrans);
1239 pipe->transfer_destroy(pipe, srcTrans);
1240 pipe->transfer_destroy(pipe, dstTrans);
1249 fallback_gen_mipmap(struct gen_mipmap_state *ctx,
1250 struct pipe_resource *pt,
1251 uint face, uint baseLevel, uint lastLevel)
1253 switch (pt->target) {
1254 case PIPE_TEXTURE_1D:
1255 make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1257 case PIPE_TEXTURE_2D:
1258 case PIPE_TEXTURE_CUBE:
1259 make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1261 case PIPE_TEXTURE_3D:
1262 make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1271 * Create a mipmap generation context.
1272 * The idea is to create one of these and re-use it each time we need to
1273 * generate a mipmap.
1275 struct gen_mipmap_state *
1276 util_create_gen_mipmap(struct pipe_context *pipe,
1277 struct cso_context *cso)
1279 struct gen_mipmap_state *ctx;
1282 ctx = CALLOC_STRUCT(gen_mipmap_state);
1289 /* disabled blending/masking */
1290 memset(&ctx->blend, 0, sizeof(ctx->blend));
1291 ctx->blend.rt[0].colormask = PIPE_MASK_RGBA;
1293 /* no-op depth/stencil/alpha */
1294 memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
1297 memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
1298 ctx->rasterizer.front_winding = PIPE_WINDING_CW;
1299 ctx->rasterizer.cull_mode = PIPE_WINDING_NONE;
1300 ctx->rasterizer.gl_rasterization_rules = 1;
1303 memset(&ctx->sampler, 0, sizeof(ctx->sampler));
1304 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1305 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1306 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1307 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
1308 ctx->sampler.normalized_coords = 1;
1310 /* vertex elements state */
1311 memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
1312 for (i = 0; i < 2; i++) {
1313 ctx->velem[i].src_offset = i * 4 * sizeof(float);
1314 ctx->velem[i].instance_divisor = 0;
1315 ctx->velem[i].vertex_buffer_index = 0;
1316 ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
1319 /* vertex shader - still needed to specify mapping from fragment
1320 * shader input semantics to vertex elements
1323 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
1324 TGSI_SEMANTIC_GENERIC };
1325 const uint semantic_indexes[] = { 0, 0 };
1326 ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
1330 /* fragment shader */
1331 ctx->fs2d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_2D);
1332 ctx->fsCube = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_CUBE);
1334 /* vertex data that doesn't change */
1335 for (i = 0; i < 4; i++) {
1336 ctx->vertices[i][0][2] = 0.0f; /* z */
1337 ctx->vertices[i][0][3] = 1.0f; /* w */
1338 ctx->vertices[i][1][3] = 1.0f; /* q */
1341 /* Note: the actual vertex buffer is allocated as needed below */
1348 * Get next "slot" of vertex space in the vertex buffer.
1349 * We're allocating one large vertex buffer and using it piece by piece.
1352 get_next_slot(struct gen_mipmap_state *ctx)
1354 const unsigned max_slots = 4096 / sizeof ctx->vertices;
1356 if (ctx->vbuf_slot >= max_slots)
1357 util_gen_mipmap_flush( ctx );
1360 ctx->vbuf = pipe_buffer_create(ctx->pipe->screen,
1361 PIPE_BIND_VERTEX_BUFFER,
1362 max_slots * sizeof ctx->vertices);
1365 return ctx->vbuf_slot++ * sizeof ctx->vertices;
1370 set_vertex_data(struct gen_mipmap_state *ctx,
1371 enum pipe_texture_target tex_target,
1376 /* vert[0].position */
1377 ctx->vertices[0][0][0] = -1.0f; /*x*/
1378 ctx->vertices[0][0][1] = -1.0f; /*y*/
1380 /* vert[1].position */
1381 ctx->vertices[1][0][0] = 1.0f;
1382 ctx->vertices[1][0][1] = -1.0f;
1384 /* vert[2].position */
1385 ctx->vertices[2][0][0] = 1.0f;
1386 ctx->vertices[2][0][1] = 1.0f;
1388 /* vert[3].position */
1389 ctx->vertices[3][0][0] = -1.0f;
1390 ctx->vertices[3][0][1] = 1.0f;
1392 /* Setup vertex texcoords. This is a little tricky for cube maps. */
1393 if (tex_target == PIPE_TEXTURE_CUBE) {
1394 static const float st[4][2] = {
1395 {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
1398 util_map_texcoords2d_onto_cubemap(face, &st[0][0], 2,
1399 &ctx->vertices[0][1][0], 8);
1403 ctx->vertices[0][1][0] = 0.0f; /*s*/
1404 ctx->vertices[0][1][1] = 0.0f; /*t*/
1405 ctx->vertices[0][1][2] = 0.0f; /*r*/
1407 ctx->vertices[1][1][0] = 1.0f;
1408 ctx->vertices[1][1][1] = 0.0f;
1409 ctx->vertices[1][1][2] = 0.0f;
1411 ctx->vertices[2][1][0] = 1.0f;
1412 ctx->vertices[2][1][1] = 1.0f;
1413 ctx->vertices[2][1][2] = 0.0f;
1415 ctx->vertices[3][1][0] = 0.0f;
1416 ctx->vertices[3][1][1] = 1.0f;
1417 ctx->vertices[3][1][2] = 0.0f;
1420 offset = get_next_slot( ctx );
1422 pipe_buffer_write_nooverlap(ctx->pipe, ctx->vbuf,
1423 offset, sizeof(ctx->vertices), ctx->vertices);
1431 * Destroy a mipmap generation context
1434 util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
1436 struct pipe_context *pipe = ctx->pipe;
1438 pipe->delete_vs_state(pipe, ctx->vs);
1439 pipe->delete_fs_state(pipe, ctx->fs2d);
1440 pipe->delete_fs_state(pipe, ctx->fsCube);
1442 pipe_resource_reference(&ctx->vbuf, NULL);
1449 /* Release vertex buffer at end of frame to avoid synchronous
1452 void util_gen_mipmap_flush( struct gen_mipmap_state *ctx )
1454 pipe_resource_reference(&ctx->vbuf, NULL);
1460 * Generate mipmap images. It's assumed all needed texture memory is
1461 * already allocated.
1463 * \param psv the sampler view to the texture to generate mipmap levels for
1464 * \param face which cube face to generate mipmaps for (0 for non-cube maps)
1465 * \param baseLevel the first mipmap level to use as a src
1466 * \param lastLevel the last mipmap level to generate
1467 * \param filter the minification filter used to generate mipmap levels with
1468 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST
1471 util_gen_mipmap(struct gen_mipmap_state *ctx,
1472 struct pipe_sampler_view *psv,
1473 uint face, uint baseLevel, uint lastLevel, uint filter)
1475 struct pipe_context *pipe = ctx->pipe;
1476 struct pipe_screen *screen = pipe->screen;
1477 struct pipe_framebuffer_state fb;
1478 struct pipe_resource *pt = psv->texture;
1479 void *fs = (pt->target == PIPE_TEXTURE_CUBE) ? ctx->fsCube : ctx->fs2d;
1484 /* The texture object should have room for the levels which we're
1485 * about to generate.
1487 assert(lastLevel <= pt->last_level);
1489 /* If this fails, why are we here? */
1490 assert(lastLevel > baseLevel);
1492 assert(filter == PIPE_TEX_FILTER_LINEAR ||
1493 filter == PIPE_TEX_FILTER_NEAREST);
1495 /* check if we can render in the texture's format */
1496 if (!screen->is_format_supported(screen, psv->format, PIPE_TEXTURE_2D,
1497 PIPE_BIND_RENDER_TARGET, 0)) {
1498 fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
1502 /* save state (restored below) */
1503 cso_save_blend(ctx->cso);
1504 cso_save_depth_stencil_alpha(ctx->cso);
1505 cso_save_rasterizer(ctx->cso);
1506 cso_save_samplers(ctx->cso);
1507 cso_save_fragment_sampler_views(ctx->cso);
1508 cso_save_framebuffer(ctx->cso);
1509 cso_save_fragment_shader(ctx->cso);
1510 cso_save_vertex_shader(ctx->cso);
1511 cso_save_viewport(ctx->cso);
1512 cso_save_clip(ctx->cso);
1513 cso_save_vertex_elements(ctx->cso);
1515 /* bind our state */
1516 cso_set_blend(ctx->cso, &ctx->blend);
1517 cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
1518 cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
1519 cso_set_clip(ctx->cso, &ctx->clip);
1520 cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
1522 cso_set_fragment_shader_handle(ctx->cso, fs);
1523 cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
1525 /* init framebuffer state */
1526 memset(&fb, 0, sizeof(fb));
1529 /* set min/mag to same filter for faster sw speed */
1530 ctx->sampler.mag_img_filter = filter;
1531 ctx->sampler.min_img_filter = filter;
1534 * XXX for small mipmap levels, it may be faster to use the software
1537 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1538 const uint srcLevel = dstLevel - 1;
1539 struct pipe_viewport_state vp;
1541 struct pipe_surface *surf =
1542 screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
1543 PIPE_BIND_RENDER_TARGET);
1546 * Setup framebuffer / dest surface
1549 fb.width = u_minify(pt->width0, dstLevel);
1550 fb.height = u_minify(pt->height0, dstLevel);
1551 cso_set_framebuffer(ctx->cso, &fb);
1554 vp.scale[0] = 0.5f * fb.width;
1555 vp.scale[1] = 0.5f * fb.height;
1558 vp.translate[0] = 0.5f * fb.width;
1559 vp.translate[1] = 0.5f * fb.height;
1560 vp.translate[2] = 0.0f;
1561 vp.translate[3] = 0.0f;
1562 cso_set_viewport(ctx->cso, &vp);
1565 * Setup sampler state
1566 * Note: we should only have to set the min/max LOD clamps to ensure
1567 * we grab texels from the right mipmap level. But some hardware
1568 * has trouble with min clamping so we also set the lod_bias to
1569 * try to work around that.
1571 ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel;
1572 ctx->sampler.lod_bias = (float) srcLevel;
1573 cso_single_sampler(ctx->cso, 0, &ctx->sampler);
1574 cso_single_sampler_done(ctx->cso);
1576 cso_set_fragment_sampler_views(ctx->cso, 1, &psv);
1578 /* quad coords in clip coords */
1579 offset = set_vertex_data(ctx,
1583 util_draw_vertex_buffer(ctx->pipe,
1586 PIPE_PRIM_TRIANGLE_FAN,
1588 2); /* attribs/vert */
1590 pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
1592 /* need to signal that the texture has changed _after_ rendering to it */
1593 pipe_surface_reference( &surf, NULL );
1596 /* restore state we changed */
1597 cso_restore_blend(ctx->cso);
1598 cso_restore_depth_stencil_alpha(ctx->cso);
1599 cso_restore_rasterizer(ctx->cso);
1600 cso_restore_samplers(ctx->cso);
1601 cso_restore_fragment_sampler_views(ctx->cso);
1602 cso_restore_framebuffer(ctx->cso);
1603 cso_restore_fragment_shader(ctx->cso);
1604 cso_restore_vertex_shader(ctx->cso);
1605 cso_restore_viewport(ctx->cso);
1606 cso_restore_clip(ctx->cso);
1607 cso_restore_vertex_elements(ctx->cso);