projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge branch 'texformat-rework'
[mesa.git] / src / gallium / auxiliary / util / u_gen_mipmap.c
1 /**************************************************************************
2 *
3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 * Copyright 2008 VMware, Inc. All rights reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29 /**
30 * @file
31 * Mipmap generation utility
32 *
33 * @author Brian Paul
34 */
35
36
37 #include "pipe/p_context.h"
38 #include "util/u_debug.h"
39 #include "pipe/p_defines.h"
40 #include "pipe/p_inlines.h"
41 #include "pipe/p_shader_tokens.h"
42 #include "pipe/p_state.h"
43
44 #include "util/u_memory.h"
45 #include "util/u_draw_quad.h"
46 #include "util/u_gen_mipmap.h"
47 #include "util/u_simple_shaders.h"
48
49 #include "cso_cache/cso_context.h"
50
51
52 struct gen_mipmap_state
53 {
54 struct pipe_context *pipe;
55 struct cso_context *cso;
56
57 struct pipe_blend_state blend;
58 struct pipe_depth_stencil_alpha_state depthstencil;
59 struct pipe_rasterizer_state rasterizer;
60 struct pipe_sampler_state sampler;
61
62 void *vs;
63 void *fs;
64
65 struct pipe_buffer *vbuf; /**< quad vertices */
66 unsigned vbuf_slot;
67
68 float vertices[4][2][4]; /**< vertex/texcoords for quad */
69 };
70
71
72
73 enum dtype
74 {
75 DTYPE_UBYTE,
76 DTYPE_UBYTE_3_3_2,
77 DTYPE_USHORT,
78 DTYPE_USHORT_4_4_4_4,
79 DTYPE_USHORT_5_6_5,
80 DTYPE_USHORT_1_5_5_5_REV,
81 DTYPE_UINT,
82 DTYPE_FLOAT,
83 DTYPE_HALF_FLOAT
84 };
85
86
87 typedef ushort half_float;
88
89
90 static half_float
91 float_to_half(float f)
92 {
93 /* XXX fix this */
94 return 0;
95 }
96
97 static float
98 half_to_float(half_float h)
99 {
100 /* XXX fix this */
101 return 0.0f;
102 }
103
104
105
106
107 /**
108 * \name Support macros for do_row and do_row_3d
109 *
110 * The macro madness is here for two reasons. First, it compacts the code
111 * slightly. Second, it makes it much easier to adjust the specifics of the
112 * filter to tune the rounding characteristics.
113 */
114 /*@{*/
115 #define DECLARE_ROW_POINTERS(t, e) \
116 const t(*rowA)[e] = (const t(*)[e]) srcRowA; \
117 const t(*rowB)[e] = (const t(*)[e]) srcRowB; \
118 const t(*rowC)[e] = (const t(*)[e]) srcRowC; \
119 const t(*rowD)[e] = (const t(*)[e]) srcRowD; \
120 t(*dst)[e] = (t(*)[e]) dstRow
121
122 #define DECLARE_ROW_POINTERS0(t) \
123 const t *rowA = (const t *) srcRowA; \
124 const t *rowB = (const t *) srcRowB; \
125 const t *rowC = (const t *) srcRowC; \
126 const t *rowD = (const t *) srcRowD; \
127 t *dst = (t *) dstRow
128
129 #define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
130 ((unsigned) Aj + (unsigned) Ak \
131 + (unsigned) Bj + (unsigned) Bk \
132 + (unsigned) Cj + (unsigned) Ck \
133 + (unsigned) Dj + (unsigned) Dk \
134 + 4) >> 3
135
136 #define FILTER_3D(e) \
137 do { \
138 dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \
139 rowB[j][e], rowB[k][e], \
140 rowC[j][e], rowC[k][e], \
141 rowD[j][e], rowD[k][e]); \
142 } while(0)
143
144 #define FILTER_F_3D(e) \
145 do { \
146 dst[i][e] = (rowA[j][e] + rowA[k][e] \
147 + rowB[j][e] + rowB[k][e] \
148 + rowC[j][e] + rowC[k][e] \
149 + rowD[j][e] + rowD[k][e]) * 0.125F; \
150 } while(0)
151
152 #define FILTER_HF_3D(e) \
153 do { \
154 const float aj = half_to_float(rowA[j][e]); \
155 const float ak = half_to_float(rowA[k][e]); \
156 const float bj = half_to_float(rowB[j][e]); \
157 const float bk = half_to_float(rowB[k][e]); \
158 const float cj = half_to_float(rowC[j][e]); \
159 const float ck = half_to_float(rowC[k][e]); \
160 const float dj = half_to_float(rowD[j][e]); \
161 const float dk = half_to_float(rowD[k][e]); \
162 dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
163 * 0.125F); \
164 } while(0)
165 /*@}*/
166
167
168 /**
169 * Average together two rows of a source image to produce a single new
170 * row in the dest image. It's legal for the two source rows to point
171 * to the same data. The source width must be equal to either the
172 * dest width or two times the dest width.
173 * \param datatype GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
174 * \param comps number of components per pixel (1..4)
175 */
176 static void
177 do_row(enum dtype datatype, uint comps, int srcWidth,
178 const void *srcRowA, const void *srcRowB,
179 int dstWidth, void *dstRow)
180 {
181 const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
182 const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
183
184 assert(comps >= 1);
185 assert(comps <= 4);
186
187 /* This assertion is no longer valid with non-power-of-2 textures
188 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
189 */
190
191 if (datatype == DTYPE_UBYTE && comps == 4) {
192 uint i, j, k;
193 const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA;
194 const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB;
195 ubyte(*dst)[4] = (ubyte(*)[4]) dstRow;
196 for (i = j = 0, k = k0; i < (uint) dstWidth;
197 i++, j += colStride, k += colStride) {
198 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
199 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
200 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
201 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
202 }
203 }
204 else if (datatype == DTYPE_UBYTE && comps == 3) {
205 uint i, j, k;
206 const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA;
207 const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB;
208 ubyte(*dst)[3] = (ubyte(*)[3]) dstRow;
209 for (i = j = 0, k = k0; i < (uint) dstWidth;
210 i++, j += colStride, k += colStride) {
211 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
212 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
213 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
214 }
215 }
216 else if (datatype == DTYPE_UBYTE && comps == 2) {
217 uint i, j, k;
218 const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA;
219 const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB;
220 ubyte(*dst)[2] = (ubyte(*)[2]) dstRow;
221 for (i = j = 0, k = k0; i < (uint) dstWidth;
222 i++, j += colStride, k += colStride) {
223 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
224 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
225 }
226 }
227 else if (datatype == DTYPE_UBYTE && comps == 1) {
228 uint i, j, k;
229 const ubyte *rowA = (const ubyte *) srcRowA;
230 const ubyte *rowB = (const ubyte *) srcRowB;
231 ubyte *dst = (ubyte *) dstRow;
232 for (i = j = 0, k = k0; i < (uint) dstWidth;
233 i++, j += colStride, k += colStride) {
234 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
235 }
236 }
237
238 else if (datatype == DTYPE_USHORT && comps == 4) {
239 uint i, j, k;
240 const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA;
241 const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB;
242 ushort(*dst)[4] = (ushort(*)[4]) dstRow;
243 for (i = j = 0, k = k0; i < (uint) dstWidth;
244 i++, j += colStride, k += colStride) {
245 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
246 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
247 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
248 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
249 }
250 }
251 else if (datatype == DTYPE_USHORT && comps == 3) {
252 uint i, j, k;
253 const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA;
254 const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB;
255 ushort(*dst)[3] = (ushort(*)[3]) dstRow;
256 for (i = j = 0, k = k0; i < (uint) dstWidth;
257 i++, j += colStride, k += colStride) {
258 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
259 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
260 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
261 }
262 }
263 else if (datatype == DTYPE_USHORT && comps == 2) {
264 uint i, j, k;
265 const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA;
266 const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB;
267 ushort(*dst)[2] = (ushort(*)[2]) dstRow;
268 for (i = j = 0, k = k0; i < (uint) dstWidth;
269 i++, j += colStride, k += colStride) {
270 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
271 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
272 }
273 }
274 else if (datatype == DTYPE_USHORT && comps == 1) {
275 uint i, j, k;
276 const ushort *rowA = (const ushort *) srcRowA;
277 const ushort *rowB = (const ushort *) srcRowB;
278 ushort *dst = (ushort *) dstRow;
279 for (i = j = 0, k = k0; i < (uint) dstWidth;
280 i++, j += colStride, k += colStride) {
281 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
282 }
283 }
284
285 else if (datatype == DTYPE_FLOAT && comps == 4) {
286 uint i, j, k;
287 const float(*rowA)[4] = (const float(*)[4]) srcRowA;
288 const float(*rowB)[4] = (const float(*)[4]) srcRowB;
289 float(*dst)[4] = (float(*)[4]) dstRow;
290 for (i = j = 0, k = k0; i < (uint) dstWidth;
291 i++, j += colStride, k += colStride) {
292 dst[i][0] = (rowA[j][0] + rowA[k][0] +
293 rowB[j][0] + rowB[k][0]) * 0.25F;
294 dst[i][1] = (rowA[j][1] + rowA[k][1] +
295 rowB[j][1] + rowB[k][1]) * 0.25F;
296 dst[i][2] = (rowA[j][2] + rowA[k][2] +
297 rowB[j][2] + rowB[k][2]) * 0.25F;
298 dst[i][3] = (rowA[j][3] + rowA[k][3] +
299 rowB[j][3] + rowB[k][3]) * 0.25F;
300 }
301 }
302 else if (datatype == DTYPE_FLOAT && comps == 3) {
303 uint i, j, k;
304 const float(*rowA)[3] = (const float(*)[3]) srcRowA;
305 const float(*rowB)[3] = (const float(*)[3]) srcRowB;
306 float(*dst)[3] = (float(*)[3]) dstRow;
307 for (i = j = 0, k = k0; i < (uint) dstWidth;
308 i++, j += colStride, k += colStride) {
309 dst[i][0] = (rowA[j][0] + rowA[k][0] +
310 rowB[j][0] + rowB[k][0]) * 0.25F;
311 dst[i][1] = (rowA[j][1] + rowA[k][1] +
312 rowB[j][1] + rowB[k][1]) * 0.25F;
313 dst[i][2] = (rowA[j][2] + rowA[k][2] +
314 rowB[j][2] + rowB[k][2]) * 0.25F;
315 }
316 }
317 else if (datatype == DTYPE_FLOAT && comps == 2) {
318 uint i, j, k;
319 const float(*rowA)[2] = (const float(*)[2]) srcRowA;
320 const float(*rowB)[2] = (const float(*)[2]) srcRowB;
321 float(*dst)[2] = (float(*)[2]) dstRow;
322 for (i = j = 0, k = k0; i < (uint) dstWidth;
323 i++, j += colStride, k += colStride) {
324 dst[i][0] = (rowA[j][0] + rowA[k][0] +
325 rowB[j][0] + rowB[k][0]) * 0.25F;
326 dst[i][1] = (rowA[j][1] + rowA[k][1] +
327 rowB[j][1] + rowB[k][1]) * 0.25F;
328 }
329 }
330 else if (datatype == DTYPE_FLOAT && comps == 1) {
331 uint i, j, k;
332 const float *rowA = (const float *) srcRowA;
333 const float *rowB = (const float *) srcRowB;
334 float *dst = (float *) dstRow;
335 for (i = j = 0, k = k0; i < (uint) dstWidth;
336 i++, j += colStride, k += colStride) {
337 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
338 }
339 }
340
341 #if 0
342 else if (datatype == HALF_DTYPE_FLOAT && comps == 4) {
343 uint i, j, k, comp;
344 const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA;
345 const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB;
346 half_float(*dst)[4] = (half_float(*)[4]) dstRow;
347 for (i = j = 0, k = k0; i < (uint) dstWidth;
348 i++, j += colStride, k += colStride) {
349 for (comp = 0; comp < 4; comp++) {
350 float aj, ak, bj, bk;
351 aj = half_to_float(rowA[j][comp]);
352 ak = half_to_float(rowA[k][comp]);
353 bj = half_to_float(rowB[j][comp]);
354 bk = half_to_float(rowB[k][comp]);
355 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
356 }
357 }
358 }
359 else if (datatype == DTYPE_HALF_FLOAT && comps == 3) {
360 uint i, j, k, comp;
361 const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA;
362 const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB;
363 half_float(*dst)[3] = (half_float(*)[3]) dstRow;
364 for (i = j = 0, k = k0; i < (uint) dstWidth;
365 i++, j += colStride, k += colStride) {
366 for (comp = 0; comp < 3; comp++) {
367 float aj, ak, bj, bk;
368 aj = half_to_float(rowA[j][comp]);
369 ak = half_to_float(rowA[k][comp]);
370 bj = half_to_float(rowB[j][comp]);
371 bk = half_to_float(rowB[k][comp]);
372 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
373 }
374 }
375 }
376 else if (datatype == DTYPE_HALF_FLOAT && comps == 2) {
377 uint i, j, k, comp;
378 const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA;
379 const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB;
380 half_float(*dst)[2] = (half_float(*)[2]) dstRow;
381 for (i = j = 0, k = k0; i < (uint) dstWidth;
382 i++, j += colStride, k += colStride) {
383 for (comp = 0; comp < 2; comp++) {
384 float aj, ak, bj, bk;
385 aj = half_to_float(rowA[j][comp]);
386 ak = half_to_float(rowA[k][comp]);
387 bj = half_to_float(rowB[j][comp]);
388 bk = half_to_float(rowB[k][comp]);
389 dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
390 }
391 }
392 }
393 else if (datatype == DTYPE_HALF_FLOAT && comps == 1) {
394 uint i, j, k;
395 const half_float *rowA = (const half_float *) srcRowA;
396 const half_float *rowB = (const half_float *) srcRowB;
397 half_float *dst = (half_float *) dstRow;
398 for (i = j = 0, k = k0; i < (uint) dstWidth;
399 i++, j += colStride, k += colStride) {
400 float aj, ak, bj, bk;
401 aj = half_to_float(rowA[j]);
402 ak = half_to_float(rowA[k]);
403 bj = half_to_float(rowB[j]);
404 bk = half_to_float(rowB[k]);
405 dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F);
406 }
407 }
408 #endif
409
410 else if (datatype == DTYPE_UINT && comps == 1) {
411 uint i, j, k;
412 const uint *rowA = (const uint *) srcRowA;
413 const uint *rowB = (const uint *) srcRowB;
414 uint *dst = (uint *) dstRow;
415 for (i = j = 0, k = k0; i < (uint) dstWidth;
416 i++, j += colStride, k += colStride) {
417 dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
418 }
419 }
420
421 else if (datatype == DTYPE_USHORT_5_6_5 && comps == 3) {
422 uint i, j, k;
423 const ushort *rowA = (const ushort *) srcRowA;
424 const ushort *rowB = (const ushort *) srcRowB;
425 ushort *dst = (ushort *) dstRow;
426 for (i = j = 0, k = k0; i < (uint) dstWidth;
427 i++, j += colStride, k += colStride) {
428 const int rowAr0 = rowA[j] & 0x1f;
429 const int rowAr1 = rowA[k] & 0x1f;
430 const int rowBr0 = rowB[j] & 0x1f;
431 const int rowBr1 = rowB[k] & 0x1f;
432 const int rowAg0 = (rowA[j] >> 5) & 0x3f;
433 const int rowAg1 = (rowA[k] >> 5) & 0x3f;
434 const int rowBg0 = (rowB[j] >> 5) & 0x3f;
435 const int rowBg1 = (rowB[k] >> 5) & 0x3f;
436 const int rowAb0 = (rowA[j] >> 11) & 0x1f;
437 const int rowAb1 = (rowA[k] >> 11) & 0x1f;
438 const int rowBb0 = (rowB[j] >> 11) & 0x1f;
439 const int rowBb1 = (rowB[k] >> 11) & 0x1f;
440 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
441 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
442 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
443 dst[i] = (blue << 11) | (green << 5) | red;
444 }
445 }
446 else if (datatype == DTYPE_USHORT_4_4_4_4 && comps == 4) {
447 uint i, j, k;
448 const ushort *rowA = (const ushort *) srcRowA;
449 const ushort *rowB = (const ushort *) srcRowB;
450 ushort *dst = (ushort *) dstRow;
451 for (i = j = 0, k = k0; i < (uint) dstWidth;
452 i++, j += colStride, k += colStride) {
453 const int rowAr0 = rowA[j] & 0xf;
454 const int rowAr1 = rowA[k] & 0xf;
455 const int rowBr0 = rowB[j] & 0xf;
456 const int rowBr1 = rowB[k] & 0xf;
457 const int rowAg0 = (rowA[j] >> 4) & 0xf;
458 const int rowAg1 = (rowA[k] >> 4) & 0xf;
459 const int rowBg0 = (rowB[j] >> 4) & 0xf;
460 const int rowBg1 = (rowB[k] >> 4) & 0xf;
461 const int rowAb0 = (rowA[j] >> 8) & 0xf;
462 const int rowAb1 = (rowA[k] >> 8) & 0xf;
463 const int rowBb0 = (rowB[j] >> 8) & 0xf;
464 const int rowBb1 = (rowB[k] >> 8) & 0xf;
465 const int rowAa0 = (rowA[j] >> 12) & 0xf;
466 const int rowAa1 = (rowA[k] >> 12) & 0xf;
467 const int rowBa0 = (rowB[j] >> 12) & 0xf;
468 const int rowBa1 = (rowB[k] >> 12) & 0xf;
469 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
470 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
471 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
472 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
473 dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
474 }
475 }
476 else if (datatype == DTYPE_USHORT_1_5_5_5_REV && comps == 4) {
477 uint i, j, k;
478 const ushort *rowA = (const ushort *) srcRowA;
479 const ushort *rowB = (const ushort *) srcRowB;
480 ushort *dst = (ushort *) dstRow;
481 for (i = j = 0, k = k0; i < (uint) dstWidth;
482 i++, j += colStride, k += colStride) {
483 const int rowAr0 = rowA[j] & 0x1f;
484 const int rowAr1 = rowA[k] & 0x1f;
485 const int rowBr0 = rowB[j] & 0x1f;
486 const int rowBr1 = rowB[k] & 0x1f;
487 const int rowAg0 = (rowA[j] >> 5) & 0x1f;
488 const int rowAg1 = (rowA[k] >> 5) & 0x1f;
489 const int rowBg0 = (rowB[j] >> 5) & 0x1f;
490 const int rowBg1 = (rowB[k] >> 5) & 0x1f;
491 const int rowAb0 = (rowA[j] >> 10) & 0x1f;
492 const int rowAb1 = (rowA[k] >> 10) & 0x1f;
493 const int rowBb0 = (rowB[j] >> 10) & 0x1f;
494 const int rowBb1 = (rowB[k] >> 10) & 0x1f;
495 const int rowAa0 = (rowA[j] >> 15) & 0x1;
496 const int rowAa1 = (rowA[k] >> 15) & 0x1;
497 const int rowBa0 = (rowB[j] >> 15) & 0x1;
498 const int rowBa1 = (rowB[k] >> 15) & 0x1;
499 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
500 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
501 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
502 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
503 dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
504 }
505 }
506 else if (datatype == DTYPE_UBYTE_3_3_2 && comps == 3) {
507 uint i, j, k;
508 const ubyte *rowA = (const ubyte *) srcRowA;
509 const ubyte *rowB = (const ubyte *) srcRowB;
510 ubyte *dst = (ubyte *) dstRow;
511 for (i = j = 0, k = k0; i < (uint) dstWidth;
512 i++, j += colStride, k += colStride) {
513 const int rowAr0 = rowA[j] & 0x3;
514 const int rowAr1 = rowA[k] & 0x3;
515 const int rowBr0 = rowB[j] & 0x3;
516 const int rowBr1 = rowB[k] & 0x3;
517 const int rowAg0 = (rowA[j] >> 2) & 0x7;
518 const int rowAg1 = (rowA[k] >> 2) & 0x7;
519 const int rowBg0 = (rowB[j] >> 2) & 0x7;
520 const int rowBg1 = (rowB[k] >> 2) & 0x7;
521 const int rowAb0 = (rowA[j] >> 5) & 0x7;
522 const int rowAb1 = (rowA[k] >> 5) & 0x7;
523 const int rowBb0 = (rowB[j] >> 5) & 0x7;
524 const int rowBb1 = (rowB[k] >> 5) & 0x7;
525 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
526 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
527 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
528 dst[i] = (blue << 5) | (green << 2) | red;
529 }
530 }
531 else {
532 debug_printf("bad format in do_row()");
533 }
534 }
535
536
537 /**
538 * Average together four rows of a source image to produce a single new
539 * row in the dest image. It's legal for the two source rows to point
540 * to the same data. The source width must be equal to either the
541 * dest width or two times the dest width.
542 *
543 * \param datatype GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
544 * \c GL_FLOAT, etc.
545 * \param comps number of components per pixel (1..4)
546 * \param srcWidth Width of a row in the source data
547 * \param srcRowA Pointer to one of the rows of source data
548 * \param srcRowB Pointer to one of the rows of source data
549 * \param srcRowC Pointer to one of the rows of source data
550 * \param srcRowD Pointer to one of the rows of source data
551 * \param dstWidth Width of a row in the destination data
552 * \param srcRowA Pointer to the row of destination data
553 */
554 static void
555 do_row_3D(enum dtype datatype, uint comps, int srcWidth,
556 const void *srcRowA, const void *srcRowB,
557 const void *srcRowC, const void *srcRowD,
558 int dstWidth, void *dstRow)
559 {
560 const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
561 const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
562 uint i, j, k;
563
564 assert(comps >= 1);
565 assert(comps <= 4);
566
567 if ((datatype == DTYPE_UBYTE) && (comps == 4)) {
568 DECLARE_ROW_POINTERS(ubyte, 4);
569
570 for (i = j = 0, k = k0; i < (uint) dstWidth;
571 i++, j += colStride, k += colStride) {
572 FILTER_3D(0);
573 FILTER_3D(1);
574 FILTER_3D(2);
575 FILTER_3D(3);
576 }
577 }
578 else if ((datatype == DTYPE_UBYTE) && (comps == 3)) {
579 DECLARE_ROW_POINTERS(ubyte, 3);
580
581 for (i = j = 0, k = k0; i < (uint) dstWidth;
582 i++, j += colStride, k += colStride) {
583 FILTER_3D(0);
584 FILTER_3D(1);
585 FILTER_3D(2);
586 }
587 }
588 else if ((datatype == DTYPE_UBYTE) && (comps == 2)) {
589 DECLARE_ROW_POINTERS(ubyte, 2);
590
591 for (i = j = 0, k = k0; i < (uint) dstWidth;
592 i++, j += colStride, k += colStride) {
593 FILTER_3D(0);
594 FILTER_3D(1);
595 }
596 }
597 else if ((datatype == DTYPE_UBYTE) && (comps == 1)) {
598 DECLARE_ROW_POINTERS(ubyte, 1);
599
600 for (i = j = 0, k = k0; i < (uint) dstWidth;
601 i++, j += colStride, k += colStride) {
602 FILTER_3D(0);
603 }
604 }
605 else if ((datatype == DTYPE_USHORT) && (comps == 4)) {
606 DECLARE_ROW_POINTERS(ushort, 4);
607
608 for (i = j = 0, k = k0; i < (uint) dstWidth;
609 i++, j += colStride, k += colStride) {
610 FILTER_3D(0);
611 FILTER_3D(1);
612 FILTER_3D(2);
613 FILTER_3D(3);
614 }
615 }
616 else if ((datatype == DTYPE_USHORT) && (comps == 3)) {
617 DECLARE_ROW_POINTERS(ushort, 3);
618
619 for (i = j = 0, k = k0; i < (uint) dstWidth;
620 i++, j += colStride, k += colStride) {
621 FILTER_3D(0);
622 FILTER_3D(1);
623 FILTER_3D(2);
624 }
625 }
626 else if ((datatype == DTYPE_USHORT) && (comps == 2)) {
627 DECLARE_ROW_POINTERS(ushort, 2);
628
629 for (i = j = 0, k = k0; i < (uint) dstWidth;
630 i++, j += colStride, k += colStride) {
631 FILTER_3D(0);
632 FILTER_3D(1);
633 }
634 }
635 else if ((datatype == DTYPE_USHORT) && (comps == 1)) {
636 DECLARE_ROW_POINTERS(ushort, 1);
637
638 for (i = j = 0, k = k0; i < (uint) dstWidth;
639 i++, j += colStride, k += colStride) {
640 FILTER_3D(0);
641 }
642 }
643 else if ((datatype == DTYPE_FLOAT) && (comps == 4)) {
644 DECLARE_ROW_POINTERS(float, 4);
645
646 for (i = j = 0, k = k0; i < (uint) dstWidth;
647 i++, j += colStride, k += colStride) {
648 FILTER_F_3D(0);
649 FILTER_F_3D(1);
650 FILTER_F_3D(2);
651 FILTER_F_3D(3);
652 }
653 }
654 else if ((datatype == DTYPE_FLOAT) && (comps == 3)) {
655 DECLARE_ROW_POINTERS(float, 3);
656
657 for (i = j = 0, k = k0; i < (uint) dstWidth;
658 i++, j += colStride, k += colStride) {
659 FILTER_F_3D(0);
660 FILTER_F_3D(1);
661 FILTER_F_3D(2);
662 }
663 }
664 else if ((datatype == DTYPE_FLOAT) && (comps == 2)) {
665 DECLARE_ROW_POINTERS(float, 2);
666
667 for (i = j = 0, k = k0; i < (uint) dstWidth;
668 i++, j += colStride, k += colStride) {
669 FILTER_F_3D(0);
670 FILTER_F_3D(1);
671 }
672 }
673 else if ((datatype == DTYPE_FLOAT) && (comps == 1)) {
674 DECLARE_ROW_POINTERS(float, 1);
675
676 for (i = j = 0, k = k0; i < (uint) dstWidth;
677 i++, j += colStride, k += colStride) {
678 FILTER_F_3D(0);
679 }
680 }
681 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 4)) {
682 DECLARE_ROW_POINTERS(half_float, 4);
683
684 for (i = j = 0, k = k0; i < (uint) dstWidth;
685 i++, j += colStride, k += colStride) {
686 FILTER_HF_3D(0);
687 FILTER_HF_3D(1);
688 FILTER_HF_3D(2);
689 FILTER_HF_3D(3);
690 }
691 }
692 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 3)) {
693 DECLARE_ROW_POINTERS(half_float, 4);
694
695 for (i = j = 0, k = k0; i < (uint) dstWidth;
696 i++, j += colStride, k += colStride) {
697 FILTER_HF_3D(0);
698 FILTER_HF_3D(1);
699 FILTER_HF_3D(2);
700 }
701 }
702 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 2)) {
703 DECLARE_ROW_POINTERS(half_float, 4);
704
705 for (i = j = 0, k = k0; i < (uint) dstWidth;
706 i++, j += colStride, k += colStride) {
707 FILTER_HF_3D(0);
708 FILTER_HF_3D(1);
709 }
710 }
711 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 1)) {
712 DECLARE_ROW_POINTERS(half_float, 4);
713
714 for (i = j = 0, k = k0; i < (uint) dstWidth;
715 i++, j += colStride, k += colStride) {
716 FILTER_HF_3D(0);
717 }
718 }
719 else if ((datatype == DTYPE_UINT) && (comps == 1)) {
720 const uint *rowA = (const uint *) srcRowA;
721 const uint *rowB = (const uint *) srcRowB;
722 const uint *rowC = (const uint *) srcRowC;
723 const uint *rowD = (const uint *) srcRowD;
724 float *dst = (float *) dstRow;
725
726 for (i = j = 0, k = k0; i < (uint) dstWidth;
727 i++, j += colStride, k += colStride) {
728 const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k])
729 + ((uint64_t) rowB[j] + (uint64_t) rowB[k])
730 + ((uint64_t) rowC[j] + (uint64_t) rowC[k])
731 + ((uint64_t) rowD[j] + (uint64_t) rowD[k]));
732 dst[i] = (float)((double) tmp * 0.125);
733 }
734 }
735 else if ((datatype == DTYPE_USHORT_5_6_5) && (comps == 3)) {
736 DECLARE_ROW_POINTERS0(ushort);
737
738 for (i = j = 0, k = k0; i < (uint) dstWidth;
739 i++, j += colStride, k += colStride) {
740 const int rowAr0 = rowA[j] & 0x1f;
741 const int rowAr1 = rowA[k] & 0x1f;
742 const int rowBr0 = rowB[j] & 0x1f;
743 const int rowBr1 = rowB[k] & 0x1f;
744 const int rowCr0 = rowC[j] & 0x1f;
745 const int rowCr1 = rowC[k] & 0x1f;
746 const int rowDr0 = rowD[j] & 0x1f;
747 const int rowDr1 = rowD[k] & 0x1f;
748 const int rowAg0 = (rowA[j] >> 5) & 0x3f;
749 const int rowAg1 = (rowA[k] >> 5) & 0x3f;
750 const int rowBg0 = (rowB[j] >> 5) & 0x3f;
751 const int rowBg1 = (rowB[k] >> 5) & 0x3f;
752 const int rowCg0 = (rowC[j] >> 5) & 0x3f;
753 const int rowCg1 = (rowC[k] >> 5) & 0x3f;
754 const int rowDg0 = (rowD[j] >> 5) & 0x3f;
755 const int rowDg1 = (rowD[k] >> 5) & 0x3f;
756 const int rowAb0 = (rowA[j] >> 11) & 0x1f;
757 const int rowAb1 = (rowA[k] >> 11) & 0x1f;
758 const int rowBb0 = (rowB[j] >> 11) & 0x1f;
759 const int rowBb1 = (rowB[k] >> 11) & 0x1f;
760 const int rowCb0 = (rowC[j] >> 11) & 0x1f;
761 const int rowCb1 = (rowC[k] >> 11) & 0x1f;
762 const int rowDb0 = (rowD[j] >> 11) & 0x1f;
763 const int rowDb1 = (rowD[k] >> 11) & 0x1f;
764 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
765 rowCr0, rowCr1, rowDr0, rowDr1);
766 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
767 rowCg0, rowCg1, rowDg0, rowDg1);
768 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
769 rowCb0, rowCb1, rowDb0, rowDb1);
770 dst[i] = (b << 11) | (g << 5) | r;
771 }
772 }
773 else if ((datatype == DTYPE_USHORT_4_4_4_4) && (comps == 4)) {
774 DECLARE_ROW_POINTERS0(ushort);
775
776 for (i = j = 0, k = k0; i < (uint) dstWidth;
777 i++, j += colStride, k += colStride) {
778 const int rowAr0 = rowA[j] & 0xf;
779 const int rowAr1 = rowA[k] & 0xf;
780 const int rowBr0 = rowB[j] & 0xf;
781 const int rowBr1 = rowB[k] & 0xf;
782 const int rowCr0 = rowC[j] & 0xf;
783 const int rowCr1 = rowC[k] & 0xf;
784 const int rowDr0 = rowD[j] & 0xf;
785 const int rowDr1 = rowD[k] & 0xf;
786 const int rowAg0 = (rowA[j] >> 4) & 0xf;
787 const int rowAg1 = (rowA[k] >> 4) & 0xf;
788 const int rowBg0 = (rowB[j] >> 4) & 0xf;
789 const int rowBg1 = (rowB[k] >> 4) & 0xf;
790 const int rowCg0 = (rowC[j] >> 4) & 0xf;
791 const int rowCg1 = (rowC[k] >> 4) & 0xf;
792 const int rowDg0 = (rowD[j] >> 4) & 0xf;
793 const int rowDg1 = (rowD[k] >> 4) & 0xf;
794 const int rowAb0 = (rowA[j] >> 8) & 0xf;
795 const int rowAb1 = (rowA[k] >> 8) & 0xf;
796 const int rowBb0 = (rowB[j] >> 8) & 0xf;
797 const int rowBb1 = (rowB[k] >> 8) & 0xf;
798 const int rowCb0 = (rowC[j] >> 8) & 0xf;
799 const int rowCb1 = (rowC[k] >> 8) & 0xf;
800 const int rowDb0 = (rowD[j] >> 8) & 0xf;
801 const int rowDb1 = (rowD[k] >> 8) & 0xf;
802 const int rowAa0 = (rowA[j] >> 12) & 0xf;
803 const int rowAa1 = (rowA[k] >> 12) & 0xf;
804 const int rowBa0 = (rowB[j] >> 12) & 0xf;
805 const int rowBa1 = (rowB[k] >> 12) & 0xf;
806 const int rowCa0 = (rowC[j] >> 12) & 0xf;
807 const int rowCa1 = (rowC[k] >> 12) & 0xf;
808 const int rowDa0 = (rowD[j] >> 12) & 0xf;
809 const int rowDa1 = (rowD[k] >> 12) & 0xf;
810 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
811 rowCr0, rowCr1, rowDr0, rowDr1);
812 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
813 rowCg0, rowCg1, rowDg0, rowDg1);
814 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
815 rowCb0, rowCb1, rowDb0, rowDb1);
816 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
817 rowCa0, rowCa1, rowDa0, rowDa1);
818
819 dst[i] = (a << 12) | (b << 8) | (g << 4) | r;
820 }
821 }
822 else if ((datatype == DTYPE_USHORT_1_5_5_5_REV) && (comps == 4)) {
823 DECLARE_ROW_POINTERS0(ushort);
824
825 for (i = j = 0, k = k0; i < (uint) dstWidth;
826 i++, j += colStride, k += colStride) {
827 const int rowAr0 = rowA[j] & 0x1f;
828 const int rowAr1 = rowA[k] & 0x1f;
829 const int rowBr0 = rowB[j] & 0x1f;
830 const int rowBr1 = rowB[k] & 0x1f;
831 const int rowCr0 = rowC[j] & 0x1f;
832 const int rowCr1 = rowC[k] & 0x1f;
833 const int rowDr0 = rowD[j] & 0x1f;
834 const int rowDr1 = rowD[k] & 0x1f;
835 const int rowAg0 = (rowA[j] >> 5) & 0x1f;
836 const int rowAg1 = (rowA[k] >> 5) & 0x1f;
837 const int rowBg0 = (rowB[j] >> 5) & 0x1f;
838 const int rowBg1 = (rowB[k] >> 5) & 0x1f;
839 const int rowCg0 = (rowC[j] >> 5) & 0x1f;
840 const int rowCg1 = (rowC[k] >> 5) & 0x1f;
841 const int rowDg0 = (rowD[j] >> 5) & 0x1f;
842 const int rowDg1 = (rowD[k] >> 5) & 0x1f;
843 const int rowAb0 = (rowA[j] >> 10) & 0x1f;
844 const int rowAb1 = (rowA[k] >> 10) & 0x1f;
845 const int rowBb0 = (rowB[j] >> 10) & 0x1f;
846 const int rowBb1 = (rowB[k] >> 10) & 0x1f;
847 const int rowCb0 = (rowC[j] >> 10) & 0x1f;
848 const int rowCb1 = (rowC[k] >> 10) & 0x1f;
849 const int rowDb0 = (rowD[j] >> 10) & 0x1f;
850 const int rowDb1 = (rowD[k] >> 10) & 0x1f;
851 const int rowAa0 = (rowA[j] >> 15) & 0x1;
852 const int rowAa1 = (rowA[k] >> 15) & 0x1;
853 const int rowBa0 = (rowB[j] >> 15) & 0x1;
854 const int rowBa1 = (rowB[k] >> 15) & 0x1;
855 const int rowCa0 = (rowC[j] >> 15) & 0x1;
856 const int rowCa1 = (rowC[k] >> 15) & 0x1;
857 const int rowDa0 = (rowD[j] >> 15) & 0x1;
858 const int rowDa1 = (rowD[k] >> 15) & 0x1;
859 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
860 rowCr0, rowCr1, rowDr0, rowDr1);
861 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
862 rowCg0, rowCg1, rowDg0, rowDg1);
863 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
864 rowCb0, rowCb1, rowDb0, rowDb1);
865 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
866 rowCa0, rowCa1, rowDa0, rowDa1);
867
868 dst[i] = (a << 15) | (b << 10) | (g << 5) | r;
869 }
870 }
871 else if ((datatype == DTYPE_UBYTE_3_3_2) && (comps == 3)) {
872 DECLARE_ROW_POINTERS0(ushort);
873
874 for (i = j = 0, k = k0; i < (uint) dstWidth;
875 i++, j += colStride, k += colStride) {
876 const int rowAr0 = rowA[j] & 0x3;
877 const int rowAr1 = rowA[k] & 0x3;
878 const int rowBr0 = rowB[j] & 0x3;
879 const int rowBr1 = rowB[k] & 0x3;
880 const int rowCr0 = rowC[j] & 0x3;
881 const int rowCr1 = rowC[k] & 0x3;
882 const int rowDr0 = rowD[j] & 0x3;
883 const int rowDr1 = rowD[k] & 0x3;
884 const int rowAg0 = (rowA[j] >> 2) & 0x7;
885 const int rowAg1 = (rowA[k] >> 2) & 0x7;
886 const int rowBg0 = (rowB[j] >> 2) & 0x7;
887 const int rowBg1 = (rowB[k] >> 2) & 0x7;
888 const int rowCg0 = (rowC[j] >> 2) & 0x7;
889 const int rowCg1 = (rowC[k] >> 2) & 0x7;
890 const int rowDg0 = (rowD[j] >> 2) & 0x7;
891 const int rowDg1 = (rowD[k] >> 2) & 0x7;
892 const int rowAb0 = (rowA[j] >> 5) & 0x7;
893 const int rowAb1 = (rowA[k] >> 5) & 0x7;
894 const int rowBb0 = (rowB[j] >> 5) & 0x7;
895 const int rowBb1 = (rowB[k] >> 5) & 0x7;
896 const int rowCb0 = (rowC[j] >> 5) & 0x7;
897 const int rowCb1 = (rowC[k] >> 5) & 0x7;
898 const int rowDb0 = (rowD[j] >> 5) & 0x7;
899 const int rowDb1 = (rowD[k] >> 5) & 0x7;
900 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
901 rowCr0, rowCr1, rowDr0, rowDr1);
902 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
903 rowCg0, rowCg1, rowDg0, rowDg1);
904 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
905 rowCb0, rowCb1, rowDb0, rowDb1);
906 dst[i] = (b << 5) | (g << 2) | r;
907 }
908 }
909 else {
910 debug_printf("bad format in do_row_3D()");
911 }
912 }
913
914
915
916 static void
917 format_to_type_comps(enum pipe_format pformat,
918 enum dtype *datatype, uint *comps)
919 {
920 /* XXX I think this could be implemented in terms of the pf_*() functions */
921 switch (pformat) {
922 case PIPE_FORMAT_A8R8G8B8_UNORM:
923 case PIPE_FORMAT_X8R8G8B8_UNORM:
924 case PIPE_FORMAT_B8G8R8A8_UNORM:
925 case PIPE_FORMAT_B8G8R8X8_UNORM:
926 case PIPE_FORMAT_R8G8B8A8_SRGB:
927 case PIPE_FORMAT_R8G8B8X8_SRGB:
928 case PIPE_FORMAT_A8R8G8B8_SRGB:
929 case PIPE_FORMAT_X8R8G8B8_SRGB:
930 case PIPE_FORMAT_B8G8R8A8_SRGB:
931 case PIPE_FORMAT_B8G8R8X8_SRGB:
932 case PIPE_FORMAT_R8G8B8_SRGB:
933 *datatype = DTYPE_UBYTE;
934 *comps = 4;
935 return;
936 case PIPE_FORMAT_A1R5G5B5_UNORM:
937 *datatype = DTYPE_USHORT_1_5_5_5_REV;
938 *comps = 4;
939 return;
940 case PIPE_FORMAT_A4R4G4B4_UNORM:
941 *datatype = DTYPE_USHORT_4_4_4_4;
942 *comps = 4;
943 return;
944 case PIPE_FORMAT_R5G6B5_UNORM:
945 *datatype = DTYPE_USHORT_5_6_5;
946 *comps = 3;
947 return;
948 case PIPE_FORMAT_L8_UNORM:
949 case PIPE_FORMAT_L8_SRGB:
950 case PIPE_FORMAT_A8_UNORM:
951 case PIPE_FORMAT_I8_UNORM:
952 *datatype = DTYPE_UBYTE;
953 *comps = 1;
954 return;
955 case PIPE_FORMAT_A8L8_UNORM:
956 case PIPE_FORMAT_A8L8_SRGB:
957 *datatype = DTYPE_UBYTE;
958 *comps = 2;
959 return;
960 default:
961 assert(0);
962 *datatype = DTYPE_UBYTE;
963 *comps = 0;
964 break;
965 }
966 }
967
968
969 static void
970 reduce_1d(enum pipe_format pformat,
971 int srcWidth, const ubyte *srcPtr,
972 int dstWidth, ubyte *dstPtr)
973 {
974 enum dtype datatype;
975 uint comps;
976
977 format_to_type_comps(pformat, &datatype, &comps);
978
979 /* we just duplicate the input row, kind of hack, saves code */
980 do_row(datatype, comps,
981 srcWidth, srcPtr, srcPtr,
982 dstWidth, dstPtr);
983 }
984
985
986 /**
987 * Strides are in bytes. If zero, it'll be computed as width * bpp.
988 */
989 static void
990 reduce_2d(enum pipe_format pformat,
991 int srcWidth, int srcHeight,
992 int srcRowStride, const ubyte *srcPtr,
993 int dstWidth, int dstHeight,
994 int dstRowStride, ubyte *dstPtr)
995 {
996 enum dtype datatype;
997 uint comps;
998 const int bpt = pf_get_size(pformat);
999 const ubyte *srcA, *srcB;
1000 ubyte *dst;
1001 int row;
1002
1003 format_to_type_comps(pformat, &datatype, &comps);
1004
1005 if (!srcRowStride)
1006 srcRowStride = bpt * srcWidth;
1007
1008 if (!dstRowStride)
1009 dstRowStride = bpt * dstWidth;
1010
1011 /* Compute src and dst pointers */
1012 srcA = srcPtr;
1013 if (srcHeight > 1)
1014 srcB = srcA + srcRowStride;
1015 else
1016 srcB = srcA;
1017 dst = dstPtr;
1018
1019 for (row = 0; row < dstHeight; row++) {
1020 do_row(datatype, comps,
1021 srcWidth, srcA, srcB,
1022 dstWidth, dst);
1023 srcA += 2 * srcRowStride;
1024 srcB += 2 * srcRowStride;
1025 dst += dstRowStride;
1026 }
1027 }
1028
1029
1030 static void
1031 reduce_3d(enum pipe_format pformat,
1032 int srcWidth, int srcHeight, int srcDepth,
1033 int srcRowStride, const ubyte *srcPtr,
1034 int dstWidth, int dstHeight, int dstDepth,
1035 int dstRowStride, ubyte *dstPtr)
1036 {
1037 const int bpt = pf_get_size(pformat);
1038 const int border = 0;
1039 int img, row;
1040 int bytesPerSrcImage, bytesPerDstImage;
1041 int bytesPerSrcRow, bytesPerDstRow;
1042 int srcImageOffset, srcRowOffset;
1043 enum dtype datatype;
1044 uint comps;
1045
1046 format_to_type_comps(pformat, &datatype, &comps);
1047
1048 bytesPerSrcImage = srcWidth * srcHeight * bpt;
1049 bytesPerDstImage = dstWidth * dstHeight * bpt;
1050
1051 bytesPerSrcRow = srcWidth * bpt;
1052 bytesPerDstRow = dstWidth * bpt;
1053
1054 /* Offset between adjacent src images to be averaged together */
1055 srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage;
1056
1057 /* Offset between adjacent src rows to be averaged together */
1058 srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt;
1059
1060 /*
1061 * Need to average together up to 8 src pixels for each dest pixel.
1062 * Break that down into 3 operations:
1063 * 1. take two rows from source image and average them together.
1064 * 2. take two rows from next source image and average them together.
1065 * 3. take the two averaged rows and average them for the final dst row.
1066 */
1067
1068 /*
1069 _mesa_printf("mip3d %d x %d x %d -> %d x %d x %d\n",
1070 srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
1071 */
1072
1073 for (img = 0; img < dstDepth; img++) {
1074 /* first source image pointer, skipping border */
1075 const ubyte *imgSrcA = srcPtr
1076 + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border
1077 + img * (bytesPerSrcImage + srcImageOffset);
1078 /* second source image pointer, skipping border */
1079 const ubyte *imgSrcB = imgSrcA + srcImageOffset;
1080 /* address of the dest image, skipping border */
1081 ubyte *imgDst = dstPtr
1082 + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border
1083 + img * bytesPerDstImage;
1084
1085 /* setup the four source row pointers and the dest row pointer */
1086 const ubyte *srcImgARowA = imgSrcA;
1087 const ubyte *srcImgARowB = imgSrcA + srcRowOffset;
1088 const ubyte *srcImgBRowA = imgSrcB;
1089 const ubyte *srcImgBRowB = imgSrcB + srcRowOffset;
1090 ubyte *dstImgRow = imgDst;
1091
1092 for (row = 0; row < dstHeight; row++) {
1093 do_row_3D(datatype, comps, srcWidth,
1094 srcImgARowA, srcImgARowB,
1095 srcImgBRowA, srcImgBRowB,
1096 dstWidth, dstImgRow);
1097
1098 /* advance to next rows */
1099 srcImgARowA += bytesPerSrcRow + srcRowOffset;
1100 srcImgARowB += bytesPerSrcRow + srcRowOffset;
1101 srcImgBRowA += bytesPerSrcRow + srcRowOffset;
1102 srcImgBRowB += bytesPerSrcRow + srcRowOffset;
1103 dstImgRow += bytesPerDstRow;
1104 }
1105 }
1106 }
1107
1108
1109
1110
1111 static void
1112 make_1d_mipmap(struct gen_mipmap_state *ctx,
1113 struct pipe_texture *pt,
1114 uint face, uint baseLevel, uint lastLevel)
1115 {
1116 struct pipe_context *pipe = ctx->pipe;
1117 struct pipe_screen *screen = pipe->screen;
1118 const uint zslice = 0;
1119 uint dstLevel;
1120
1121 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1122 const uint srcLevel = dstLevel - 1;
1123 struct pipe_transfer *srcTrans, *dstTrans;
1124 void *srcMap, *dstMap;
1125
1126 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice,
1127 PIPE_TRANSFER_READ, 0, 0,
1128 pt->width[srcLevel],
1129 pt->height[srcLevel]);
1130 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice,
1131 PIPE_TRANSFER_WRITE, 0, 0,
1132 pt->width[dstLevel],
1133 pt->height[dstLevel]);
1134
1135 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans);
1136 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans);
1137
1138 reduce_1d(pt->format,
1139 srcTrans->width, srcMap,
1140 dstTrans->width, dstMap);
1141
1142 screen->transfer_unmap(screen, srcTrans);
1143 screen->transfer_unmap(screen, dstTrans);
1144
1145 screen->tex_transfer_destroy(srcTrans);
1146 screen->tex_transfer_destroy(dstTrans);
1147 }
1148 }
1149
1150
1151 static void
1152 make_2d_mipmap(struct gen_mipmap_state *ctx,
1153 struct pipe_texture *pt,
1154 uint face, uint baseLevel, uint lastLevel)
1155 {
1156 struct pipe_context *pipe = ctx->pipe;
1157 struct pipe_screen *screen = pipe->screen;
1158 const uint zslice = 0;
1159 uint dstLevel;
1160
1161 assert(pt->block.width == 1);
1162 assert(pt->block.height == 1);
1163
1164 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1165 const uint srcLevel = dstLevel - 1;
1166 struct pipe_transfer *srcTrans, *dstTrans;
1167 ubyte *srcMap, *dstMap;
1168
1169 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice,
1170 PIPE_TRANSFER_READ, 0, 0,
1171 pt->width[srcLevel],
1172 pt->height[srcLevel]);
1173 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice,
1174 PIPE_TRANSFER_WRITE, 0, 0,
1175 pt->width[dstLevel],
1176 pt->height[dstLevel]);
1177
1178 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans);
1179 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans);
1180
1181 reduce_2d(pt->format,
1182 srcTrans->width, srcTrans->height,
1183 srcTrans->stride, srcMap,
1184 dstTrans->width, dstTrans->height,
1185 dstTrans->stride, dstMap);
1186
1187 screen->transfer_unmap(screen, srcTrans);
1188 screen->transfer_unmap(screen, dstTrans);
1189
1190 screen->tex_transfer_destroy(srcTrans);
1191 screen->tex_transfer_destroy(dstTrans);
1192 }
1193 }
1194
1195
1196 static void
1197 make_3d_mipmap(struct gen_mipmap_state *ctx,
1198 struct pipe_texture *pt,
1199 uint face, uint baseLevel, uint lastLevel)
1200 {
1201 #if 0
1202 struct pipe_context *pipe = ctx->pipe;
1203 struct pipe_screen *screen = pipe->screen;
1204 uint dstLevel, zslice = 0;
1205
1206 assert(pt->block.width == 1);
1207 assert(pt->block.height == 1);
1208
1209 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1210 const uint srcLevel = dstLevel - 1;
1211 struct pipe_transfer *srcTrans, *dstTrans;
1212 ubyte *srcMap, *dstMap;
1213
1214 srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice,
1215 PIPE_TRANSFER_READ, 0, 0,
1216 pt->width[srcLevel],
1217 pt->height[srcLevel]);
1218 dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice,
1219 PIPE_TRANSFER_WRITE, 0, 0,
1220 pt->width[dstLevel],
1221 pt->height[dstLevel]);
1222
1223 srcMap = (ubyte *) screen->transfer_map(screen, srcTrans);
1224 dstMap = (ubyte *) screen->transfer_map(screen, dstTrans);
1225
1226 reduce_3d(pt->format,
1227 srcTrans->width, srcTrans->height,
1228 srcTrans->stride, srcMap,
1229 dstTrans->width, dstTrans->height,
1230 dstTrans->stride, dstMap);
1231
1232 screen->transfer_unmap(screen, srcTrans);
1233 screen->transfer_unmap(screen, dstTrans);
1234
1235 screen->tex_transfer_destroy(srcTrans);
1236 screen->tex_transfer_destroy(dstTrans);
1237 }
1238 #else
1239 (void) reduce_3d;
1240 #endif
1241 }
1242
1243
1244 static void
1245 fallback_gen_mipmap(struct gen_mipmap_state *ctx,
1246 struct pipe_texture *pt,
1247 uint face, uint baseLevel, uint lastLevel)
1248 {
1249 switch (pt->target) {
1250 case PIPE_TEXTURE_1D:
1251 make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1252 break;
1253 case PIPE_TEXTURE_2D:
1254 case PIPE_TEXTURE_CUBE:
1255 make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1256 break;
1257 case PIPE_TEXTURE_3D:
1258 make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1259 break;
1260 default:
1261 assert(0);
1262 }
1263 }
1264
1265
1266 /**
1267 * Create a mipmap generation context.
1268 * The idea is to create one of these and re-use it each time we need to
1269 * generate a mipmap.
1270 */
1271 struct gen_mipmap_state *
1272 util_create_gen_mipmap(struct pipe_context *pipe,
1273 struct cso_context *cso)
1274 {
1275 struct gen_mipmap_state *ctx;
1276 uint i;
1277
1278 ctx = CALLOC_STRUCT(gen_mipmap_state);
1279 if (!ctx)
1280 return NULL;
1281
1282 ctx->pipe = pipe;
1283 ctx->cso = cso;
1284
1285 /* disabled blending/masking */
1286 memset(&ctx->blend, 0, sizeof(ctx->blend));
1287 ctx->blend.colormask = PIPE_MASK_RGBA;
1288
1289 /* no-op depth/stencil/alpha */
1290 memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
1291
1292 /* rasterizer */
1293 memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
1294 ctx->rasterizer.front_winding = PIPE_WINDING_CW;
1295 ctx->rasterizer.cull_mode = PIPE_WINDING_NONE;
1296 ctx->rasterizer.bypass_vs_clip_and_viewport = 1;
1297 ctx->rasterizer.gl_rasterization_rules = 1;
1298
1299 /* sampler state */
1300 memset(&ctx->sampler, 0, sizeof(ctx->sampler));
1301 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1302 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1303 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1304 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
1305 ctx->sampler.normalized_coords = 1;
1306
1307 /* vertex shader - still needed to specify mapping from fragment
1308 * shader input semantics to vertex elements
1309 */
1310 {
1311 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
1312 TGSI_SEMANTIC_GENERIC };
1313 const uint semantic_indexes[] = { 0, 0 };
1314 ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
1315 semantic_indexes);
1316 }
1317
1318 /* fragment shader */
1319 ctx->fs = util_make_fragment_tex_shader(pipe);
1320
1321 /* vertex data that doesn't change */
1322 for (i = 0; i < 4; i++) {
1323 ctx->vertices[i][0][2] = 0.0f; /* z */
1324 ctx->vertices[i][0][3] = 1.0f; /* w */
1325 ctx->vertices[i][1][3] = 1.0f; /* q */
1326 }
1327
1328 /* Note: the actual vertex buffer is allocated as needed below */
1329
1330 return ctx;
1331 }
1332
1333
1334 /**
1335 * Get next "slot" of vertex space in the vertex buffer.
1336 * We're allocating one large vertex buffer and using it piece by piece.
1337 */
1338 static unsigned
1339 get_next_slot(struct gen_mipmap_state *ctx)
1340 {
1341 const unsigned max_slots = 4096 / sizeof ctx->vertices;
1342
1343 if (ctx->vbuf_slot >= max_slots)
1344 util_gen_mipmap_flush( ctx );
1345
1346 if (!ctx->vbuf) {
1347 ctx->vbuf = pipe_buffer_create(ctx->pipe->screen,
1348 32,
1349 PIPE_BUFFER_USAGE_VERTEX,
1350 max_slots * sizeof ctx->vertices);
1351 }
1352
1353 return ctx->vbuf_slot++ * sizeof ctx->vertices;
1354 }
1355
1356
1357 static unsigned
1358 set_vertex_data(struct gen_mipmap_state *ctx,
1359 enum pipe_texture_target tex_target,
1360 uint face, float width, float height)
1361 {
1362 unsigned offset;
1363
1364 /* vert[0].position */
1365 ctx->vertices[0][0][0] = 0.0f; /*x*/
1366 ctx->vertices[0][0][1] = 0.0f; /*y*/
1367
1368 /* vert[1].position */
1369 ctx->vertices[1][0][0] = width;
1370 ctx->vertices[1][0][1] = 0.0f;
1371
1372 /* vert[2].position */
1373 ctx->vertices[2][0][0] = width;
1374 ctx->vertices[2][0][1] = height;
1375
1376 /* vert[3].position */
1377 ctx->vertices[3][0][0] = 0.0f;
1378 ctx->vertices[3][0][1] = height;
1379
1380 /* Setup vertex texcoords. This is a little tricky for cube maps. */
1381 if (tex_target == PIPE_TEXTURE_CUBE) {
1382 static const float st[4][2] = {
1383 {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
1384 };
1385 float rx, ry, rz;
1386 uint i;
1387
1388 /* loop over quad verts */
1389 for (i = 0; i < 4; i++) {
1390 /* Compute sc = +/-scale and tc = +/-scale.
1391 * Not +/-1 to avoid cube face selection ambiguity near the edges,
1392 * though that can still sometimes happen with this scale factor...
1393 */
1394 const float scale = 0.9999f;
1395 const float sc = (2.0f * st[i][0] - 1.0f) * scale;
1396 const float tc = (2.0f * st[i][1] - 1.0f) * scale;
1397
1398 switch (face) {
1399 case PIPE_TEX_FACE_POS_X:
1400 rx = 1.0f;
1401 ry = -tc;
1402 rz = -sc;
1403 break;
1404 case PIPE_TEX_FACE_NEG_X:
1405 rx = -1.0f;
1406 ry = -tc;
1407 rz = sc;
1408 break;
1409 case PIPE_TEX_FACE_POS_Y:
1410 rx = sc;
1411 ry = 1.0f;
1412 rz = tc;
1413 break;
1414 case PIPE_TEX_FACE_NEG_Y:
1415 rx = sc;
1416 ry = -1.0f;
1417 rz = -tc;
1418 break;
1419 case PIPE_TEX_FACE_POS_Z:
1420 rx = sc;
1421 ry = -tc;
1422 rz = 1.0f;
1423 break;
1424 case PIPE_TEX_FACE_NEG_Z:
1425 rx = -sc;
1426 ry = -tc;
1427 rz = -1.0f;
1428 break;
1429 default:
1430 rx = ry = rz = 0.0f;
1431 assert(0);
1432 }
1433
1434 ctx->vertices[i][1][0] = rx; /*s*/
1435 ctx->vertices[i][1][1] = ry; /*t*/
1436 ctx->vertices[i][1][2] = rz; /*r*/
1437 }
1438 }
1439 else {
1440 /* 1D/2D */
1441 ctx->vertices[0][1][0] = 0.0f; /*s*/
1442 ctx->vertices[0][1][1] = 0.0f; /*t*/
1443 ctx->vertices[0][1][2] = 0.0f; /*r*/
1444
1445 ctx->vertices[1][1][0] = 1.0f;
1446 ctx->vertices[1][1][1] = 0.0f;
1447 ctx->vertices[1][1][2] = 0.0f;
1448
1449 ctx->vertices[2][1][0] = 1.0f;
1450 ctx->vertices[2][1][1] = 1.0f;
1451 ctx->vertices[2][1][2] = 0.0f;
1452
1453 ctx->vertices[3][1][0] = 0.0f;
1454 ctx->vertices[3][1][1] = 1.0f;
1455 ctx->vertices[3][1][2] = 0.0f;
1456 }
1457
1458 offset = get_next_slot( ctx );
1459
1460 pipe_buffer_write(ctx->pipe->screen, ctx->vbuf,
1461 offset, sizeof(ctx->vertices), ctx->vertices);
1462
1463 return offset;
1464 }
1465
1466
1467
1468 /**
1469 * Destroy a mipmap generation context
1470 */
1471 void
1472 util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
1473 {
1474 struct pipe_context *pipe = ctx->pipe;
1475
1476 pipe->delete_vs_state(pipe, ctx->vs);
1477 pipe->delete_fs_state(pipe, ctx->fs);
1478
1479 pipe_buffer_reference(&ctx->vbuf, NULL);
1480
1481 FREE(ctx);
1482 }
1483
1484
1485
1486 /* Release vertex buffer at end of frame to avoid synchronous
1487 * rendering.
1488 */
1489 void util_gen_mipmap_flush( struct gen_mipmap_state *ctx )
1490 {
1491 pipe_buffer_reference(&ctx->vbuf, NULL);
1492 ctx->vbuf_slot = 0;
1493 }
1494
1495
1496 /**
1497 * Generate mipmap images. It's assumed all needed texture memory is
1498 * already allocated.
1499 *
1500 * \param pt the texture to generate mipmap levels for
1501 * \param face which cube face to generate mipmaps for (0 for non-cube maps)
1502 * \param baseLevel the first mipmap level to use as a src
1503 * \param lastLevel the last mipmap level to generate
1504 * \param filter the minification filter used to generate mipmap levels with
1505 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST
1506 */
1507 void
1508 util_gen_mipmap(struct gen_mipmap_state *ctx,
1509 struct pipe_texture *pt,
1510 uint face, uint baseLevel, uint lastLevel, uint filter)
1511 {
1512 struct pipe_context *pipe = ctx->pipe;
1513 struct pipe_screen *screen = pipe->screen;
1514 struct pipe_framebuffer_state fb;
1515 uint dstLevel;
1516 uint zslice = 0;
1517 uint offset;
1518
1519 /* The texture object should have room for the levels which we're
1520 * about to generate.
1521 */
1522 assert(lastLevel <= pt->last_level);
1523
1524 /* If this fails, why are we here? */
1525 assert(lastLevel > baseLevel);
1526
1527 assert(filter == PIPE_TEX_FILTER_LINEAR ||
1528 filter == PIPE_TEX_FILTER_NEAREST);
1529
1530 /* check if we can render in the texture's format */
1531 if (!screen->is_format_supported(screen, pt->format, PIPE_TEXTURE_2D,
1532 PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) {
1533 fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
1534 return;
1535 }
1536
1537 /* save state (restored below) */
1538 cso_save_blend(ctx->cso);
1539 cso_save_depth_stencil_alpha(ctx->cso);
1540 cso_save_rasterizer(ctx->cso);
1541 cso_save_samplers(ctx->cso);
1542 cso_save_sampler_textures(ctx->cso);
1543 cso_save_framebuffer(ctx->cso);
1544 cso_save_fragment_shader(ctx->cso);
1545 cso_save_vertex_shader(ctx->cso);
1546
1547 /* bind our state */
1548 cso_set_blend(ctx->cso, &ctx->blend);
1549 cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
1550 cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
1551
1552 cso_set_fragment_shader_handle(ctx->cso, ctx->fs);
1553 cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
1554
1555 /* init framebuffer state */
1556 memset(&fb, 0, sizeof(fb));
1557 fb.nr_cbufs = 1;
1558
1559 /* set min/mag to same filter for faster sw speed */
1560 ctx->sampler.mag_img_filter = filter;
1561 ctx->sampler.min_img_filter = filter;
1562
1563 /*
1564 * XXX for small mipmap levels, it may be faster to use the software
1565 * fallback path...
1566 */
1567 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1568 const uint srcLevel = dstLevel - 1;
1569
1570 struct pipe_surface *surf =
1571 screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
1572 PIPE_BUFFER_USAGE_GPU_WRITE);
1573
1574 /*
1575 * Setup framebuffer / dest surface
1576 */
1577 fb.cbufs[0] = surf;
1578 fb.width = pt->width[dstLevel];
1579 fb.height = pt->height[dstLevel];
1580 cso_set_framebuffer(ctx->cso, &fb);
1581
1582 /*
1583 * Setup sampler state
1584 * Note: we should only have to set the min/max LOD clamps to ensure
1585 * we grab texels from the right mipmap level. But some hardware
1586 * has trouble with min clamping so we also set the lod_bias to
1587 * try to work around that.
1588 */
1589 ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel;
1590 ctx->sampler.lod_bias = (float) srcLevel;
1591 cso_single_sampler(ctx->cso, 0, &ctx->sampler);
1592 cso_single_sampler_done(ctx->cso);
1593
1594 cso_set_sampler_textures(ctx->cso, 1, &pt);
1595
1596 /* quad coords in window coords (bypassing vs, clip and viewport) */
1597 offset = set_vertex_data(ctx,
1598 pt->target,
1599 face,
1600 (float) pt->width[dstLevel],
1601 (float) pt->height[dstLevel]);
1602
1603 util_draw_vertex_buffer(ctx->pipe,
1604 ctx->vbuf,
1605 offset,
1606 PIPE_PRIM_TRIANGLE_FAN,
1607 4, /* verts */
1608 2); /* attribs/vert */
1609
1610 pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
1611
1612 /* need to signal that the texture has changed _after_ rendering to it */
1613 pipe_surface_reference( &surf, NULL );
1614 }
1615
1616 /* restore state we changed */
1617 cso_restore_blend(ctx->cso);
1618 cso_restore_depth_stencil_alpha(ctx->cso);
1619 cso_restore_rasterizer(ctx->cso);
1620 cso_restore_samplers(ctx->cso);
1621 cso_restore_sampler_textures(ctx->cso);
1622 cso_restore_framebuffer(ctx->cso);
1623 cso_restore_fragment_shader(ctx->cso);
1624 cso_restore_vertex_shader(ctx->cso);
1625 }