projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
gallium/u_gen_mipmap: don't release vertex buffer at end of frame / in glFlush
[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 "util/u_inlines.h"
41 #include "pipe/p_shader_tokens.h"
42 #include "pipe/p_state.h"
43
44 #include "util/u_format.h"
45 #include "util/u_memory.h"
46 #include "util/u_draw_quad.h"
47 #include "util/u_gen_mipmap.h"
48 #include "util/u_simple_shaders.h"
49 #include "util/u_math.h"
50 #include "util/u_texture.h"
51 #include "util/u_half.h"
52 #include "util/u_surface.h"
53
54 #include "cso_cache/cso_context.h"
55
56
57 struct gen_mipmap_state
58 {
59 struct pipe_context *pipe;
60 struct cso_context *cso;
61
62 struct pipe_blend_state blend;
63 struct pipe_depth_stencil_alpha_state depthstencil;
64 struct pipe_rasterizer_state rasterizer;
65 struct pipe_sampler_state sampler;
66 struct pipe_vertex_element velem[2];
67
68 void *vs;
69 void *fs[TGSI_TEXTURE_COUNT]; /**< Not all are used, but simplifies code */
70
71 struct pipe_resource *vbuf; /**< quad vertices */
72 unsigned vbuf_slot;
73
74 float vertices[4][2][4]; /**< vertex/texcoords for quad */
75 };
76
77
78
79 enum dtype
80 {
81 DTYPE_UBYTE,
82 DTYPE_UBYTE_3_3_2,
83 DTYPE_USHORT,
84 DTYPE_USHORT_4_4_4_4,
85 DTYPE_USHORT_5_6_5,
86 DTYPE_USHORT_1_5_5_5_REV,
87 DTYPE_UINT,
88 DTYPE_FLOAT,
89 DTYPE_HALF_FLOAT
90 };
91
92
93 typedef uint16_t half_float;
94
95
96 /**
97 * \name Support macros for do_row and do_row_3d
98 *
99 * The macro madness is here for two reasons. First, it compacts the code
100 * slightly. Second, it makes it much easier to adjust the specifics of the
101 * filter to tune the rounding characteristics.
102 */
103 /*@{*/
104 #define DECLARE_ROW_POINTERS(t, e) \
105 const t(*rowA)[e] = (const t(*)[e]) srcRowA; \
106 const t(*rowB)[e] = (const t(*)[e]) srcRowB; \
107 const t(*rowC)[e] = (const t(*)[e]) srcRowC; \
108 const t(*rowD)[e] = (const t(*)[e]) srcRowD; \
109 t(*dst)[e] = (t(*)[e]) dstRow
110
111 #define DECLARE_ROW_POINTERS0(t) \
112 const t *rowA = (const t *) srcRowA; \
113 const t *rowB = (const t *) srcRowB; \
114 const t *rowC = (const t *) srcRowC; \
115 const t *rowD = (const t *) srcRowD; \
116 t *dst = (t *) dstRow
117
118 #define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
119 ((unsigned) Aj + (unsigned) Ak \
120 + (unsigned) Bj + (unsigned) Bk \
121 + (unsigned) Cj + (unsigned) Ck \
122 + (unsigned) Dj + (unsigned) Dk \
123 + 4) >> 3
124
125 #define FILTER_3D(e) \
126 do { \
127 dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \
128 rowB[j][e], rowB[k][e], \
129 rowC[j][e], rowC[k][e], \
130 rowD[j][e], rowD[k][e]); \
131 } while(0)
132
133 #define FILTER_F_3D(e) \
134 do { \
135 dst[i][e] = (rowA[j][e] + rowA[k][e] \
136 + rowB[j][e] + rowB[k][e] \
137 + rowC[j][e] + rowC[k][e] \
138 + rowD[j][e] + rowD[k][e]) * 0.125F; \
139 } while(0)
140
141 #define FILTER_HF_3D(e) \
142 do { \
143 const float aj = util_half_to_float(rowA[j][e]); \
144 const float ak = util_half_to_float(rowA[k][e]); \
145 const float bj = util_half_to_float(rowB[j][e]); \
146 const float bk = util_half_to_float(rowB[k][e]); \
147 const float cj = util_half_to_float(rowC[j][e]); \
148 const float ck = util_half_to_float(rowC[k][e]); \
149 const float dj = util_half_to_float(rowD[j][e]); \
150 const float dk = util_half_to_float(rowD[k][e]); \
151 dst[i][e] = util_float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
152 * 0.125F); \
153 } while(0)
154 /*@}*/
155
156
157 /**
158 * Average together two rows of a source image to produce a single new
159 * row in the dest image. It's legal for the two source rows to point
160 * to the same data. The source width must be equal to either the
161 * dest width or two times the dest width.
162 * \param datatype GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
163 * \param comps number of components per pixel (1..4)
164 */
165 static void
166 do_row(enum dtype datatype, uint comps, int srcWidth,
167 const void *srcRowA, const void *srcRowB,
168 int dstWidth, void *dstRow)
169 {
170 const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
171 const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
172
173 assert(comps >= 1);
174 assert(comps <= 4);
175
176 /* This assertion is no longer valid with non-power-of-2 textures
177 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
178 */
179
180 if (datatype == DTYPE_UBYTE && comps == 4) {
181 uint i, j, k;
182 const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA;
183 const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB;
184 ubyte(*dst)[4] = (ubyte(*)[4]) dstRow;
185 for (i = j = 0, k = k0; i < (uint) dstWidth;
186 i++, j += colStride, k += colStride) {
187 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
188 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
189 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
190 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
191 }
192 }
193 else if (datatype == DTYPE_UBYTE && comps == 3) {
194 uint i, j, k;
195 const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA;
196 const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB;
197 ubyte(*dst)[3] = (ubyte(*)[3]) dstRow;
198 for (i = j = 0, k = k0; i < (uint) dstWidth;
199 i++, j += colStride, k += colStride) {
200 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
201 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
202 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
203 }
204 }
205 else if (datatype == DTYPE_UBYTE && comps == 2) {
206 uint i, j, k;
207 const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA;
208 const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB;
209 ubyte(*dst)[2] = (ubyte(*)[2]) dstRow;
210 for (i = j = 0, k = k0; i < (uint) dstWidth;
211 i++, j += colStride, k += colStride) {
212 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
213 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
214 }
215 }
216 else if (datatype == DTYPE_UBYTE && comps == 1) {
217 uint i, j, k;
218 const ubyte *rowA = (const ubyte *) srcRowA;
219 const ubyte *rowB = (const ubyte *) srcRowB;
220 ubyte *dst = (ubyte *) dstRow;
221 for (i = j = 0, k = k0; i < (uint) dstWidth;
222 i++, j += colStride, k += colStride) {
223 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
224 }
225 }
226
227 else if (datatype == DTYPE_USHORT && comps == 4) {
228 uint i, j, k;
229 const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA;
230 const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB;
231 ushort(*dst)[4] = (ushort(*)[4]) dstRow;
232 for (i = j = 0, k = k0; i < (uint) dstWidth;
233 i++, j += colStride, k += colStride) {
234 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
235 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
236 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
237 dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
238 }
239 }
240 else if (datatype == DTYPE_USHORT && comps == 3) {
241 uint i, j, k;
242 const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA;
243 const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB;
244 ushort(*dst)[3] = (ushort(*)[3]) dstRow;
245 for (i = j = 0, k = k0; i < (uint) dstWidth;
246 i++, j += colStride, k += colStride) {
247 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
248 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
249 dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
250 }
251 }
252 else if (datatype == DTYPE_USHORT && comps == 2) {
253 uint i, j, k;
254 const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA;
255 const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB;
256 ushort(*dst)[2] = (ushort(*)[2]) dstRow;
257 for (i = j = 0, k = k0; i < (uint) dstWidth;
258 i++, j += colStride, k += colStride) {
259 dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
260 dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
261 }
262 }
263 else if (datatype == DTYPE_USHORT && comps == 1) {
264 uint i, j, k;
265 const ushort *rowA = (const ushort *) srcRowA;
266 const ushort *rowB = (const ushort *) srcRowB;
267 ushort *dst = (ushort *) dstRow;
268 for (i = j = 0, k = k0; i < (uint) dstWidth;
269 i++, j += colStride, k += colStride) {
270 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
271 }
272 }
273
274 else if (datatype == DTYPE_FLOAT && comps == 4) {
275 uint i, j, k;
276 const float(*rowA)[4] = (const float(*)[4]) srcRowA;
277 const float(*rowB)[4] = (const float(*)[4]) srcRowB;
278 float(*dst)[4] = (float(*)[4]) dstRow;
279 for (i = j = 0, k = k0; i < (uint) dstWidth;
280 i++, j += colStride, k += colStride) {
281 dst[i][0] = (rowA[j][0] + rowA[k][0] +
282 rowB[j][0] + rowB[k][0]) * 0.25F;
283 dst[i][1] = (rowA[j][1] + rowA[k][1] +
284 rowB[j][1] + rowB[k][1]) * 0.25F;
285 dst[i][2] = (rowA[j][2] + rowA[k][2] +
286 rowB[j][2] + rowB[k][2]) * 0.25F;
287 dst[i][3] = (rowA[j][3] + rowA[k][3] +
288 rowB[j][3] + rowB[k][3]) * 0.25F;
289 }
290 }
291 else if (datatype == DTYPE_FLOAT && comps == 3) {
292 uint i, j, k;
293 const float(*rowA)[3] = (const float(*)[3]) srcRowA;
294 const float(*rowB)[3] = (const float(*)[3]) srcRowB;
295 float(*dst)[3] = (float(*)[3]) dstRow;
296 for (i = j = 0, k = k0; i < (uint) dstWidth;
297 i++, j += colStride, k += colStride) {
298 dst[i][0] = (rowA[j][0] + rowA[k][0] +
299 rowB[j][0] + rowB[k][0]) * 0.25F;
300 dst[i][1] = (rowA[j][1] + rowA[k][1] +
301 rowB[j][1] + rowB[k][1]) * 0.25F;
302 dst[i][2] = (rowA[j][2] + rowA[k][2] +
303 rowB[j][2] + rowB[k][2]) * 0.25F;
304 }
305 }
306 else if (datatype == DTYPE_FLOAT && comps == 2) {
307 uint i, j, k;
308 const float(*rowA)[2] = (const float(*)[2]) srcRowA;
309 const float(*rowB)[2] = (const float(*)[2]) srcRowB;
310 float(*dst)[2] = (float(*)[2]) dstRow;
311 for (i = j = 0, k = k0; i < (uint) dstWidth;
312 i++, j += colStride, k += colStride) {
313 dst[i][0] = (rowA[j][0] + rowA[k][0] +
314 rowB[j][0] + rowB[k][0]) * 0.25F;
315 dst[i][1] = (rowA[j][1] + rowA[k][1] +
316 rowB[j][1] + rowB[k][1]) * 0.25F;
317 }
318 }
319 else if (datatype == DTYPE_FLOAT && comps == 1) {
320 uint i, j, k;
321 const float *rowA = (const float *) srcRowA;
322 const float *rowB = (const float *) srcRowB;
323 float *dst = (float *) dstRow;
324 for (i = j = 0, k = k0; i < (uint) dstWidth;
325 i++, j += colStride, k += colStride) {
326 dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
327 }
328 }
329
330 else if (datatype == DTYPE_HALF_FLOAT && comps == 4) {
331 uint i, j, k, comp;
332 const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA;
333 const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB;
334 half_float(*dst)[4] = (half_float(*)[4]) dstRow;
335 for (i = j = 0, k = k0; i < (uint) dstWidth;
336 i++, j += colStride, k += colStride) {
337 for (comp = 0; comp < 4; comp++) {
338 float aj, ak, bj, bk;
339 aj = util_half_to_float(rowA[j][comp]);
340 ak = util_half_to_float(rowA[k][comp]);
341 bj = util_half_to_float(rowB[j][comp]);
342 bk = util_half_to_float(rowB[k][comp]);
343 dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
344 }
345 }
346 }
347 else if (datatype == DTYPE_HALF_FLOAT && comps == 3) {
348 uint i, j, k, comp;
349 const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA;
350 const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB;
351 half_float(*dst)[3] = (half_float(*)[3]) dstRow;
352 for (i = j = 0, k = k0; i < (uint) dstWidth;
353 i++, j += colStride, k += colStride) {
354 for (comp = 0; comp < 3; comp++) {
355 float aj, ak, bj, bk;
356 aj = util_half_to_float(rowA[j][comp]);
357 ak = util_half_to_float(rowA[k][comp]);
358 bj = util_half_to_float(rowB[j][comp]);
359 bk = util_half_to_float(rowB[k][comp]);
360 dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
361 }
362 }
363 }
364 else if (datatype == DTYPE_HALF_FLOAT && comps == 2) {
365 uint i, j, k, comp;
366 const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA;
367 const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB;
368 half_float(*dst)[2] = (half_float(*)[2]) dstRow;
369 for (i = j = 0, k = k0; i < (uint) dstWidth;
370 i++, j += colStride, k += colStride) {
371 for (comp = 0; comp < 2; comp++) {
372 float aj, ak, bj, bk;
373 aj = util_half_to_float(rowA[j][comp]);
374 ak = util_half_to_float(rowA[k][comp]);
375 bj = util_half_to_float(rowB[j][comp]);
376 bk = util_half_to_float(rowB[k][comp]);
377 dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
378 }
379 }
380 }
381 else if (datatype == DTYPE_HALF_FLOAT && comps == 1) {
382 uint i, j, k;
383 const half_float *rowA = (const half_float *) srcRowA;
384 const half_float *rowB = (const half_float *) srcRowB;
385 half_float *dst = (half_float *) dstRow;
386 for (i = j = 0, k = k0; i < (uint) dstWidth;
387 i++, j += colStride, k += colStride) {
388 float aj, ak, bj, bk;
389 aj = util_half_to_float(rowA[j]);
390 ak = util_half_to_float(rowA[k]);
391 bj = util_half_to_float(rowB[j]);
392 bk = util_half_to_float(rowB[k]);
393 dst[i] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
394 }
395 }
396
397 else if (datatype == DTYPE_UINT && comps == 1) {
398 uint i, j, k;
399 const uint *rowA = (const uint *) srcRowA;
400 const uint *rowB = (const uint *) srcRowB;
401 uint *dst = (uint *) dstRow;
402 for (i = j = 0, k = k0; i < (uint) dstWidth;
403 i++, j += colStride, k += colStride) {
404 dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
405 }
406 }
407
408 else if (datatype == DTYPE_USHORT_5_6_5 && comps == 3) {
409 uint i, j, k;
410 const ushort *rowA = (const ushort *) srcRowA;
411 const ushort *rowB = (const ushort *) srcRowB;
412 ushort *dst = (ushort *) dstRow;
413 for (i = j = 0, k = k0; i < (uint) dstWidth;
414 i++, j += colStride, k += colStride) {
415 const int rowAr0 = rowA[j] & 0x1f;
416 const int rowAr1 = rowA[k] & 0x1f;
417 const int rowBr0 = rowB[j] & 0x1f;
418 const int rowBr1 = rowB[k] & 0x1f;
419 const int rowAg0 = (rowA[j] >> 5) & 0x3f;
420 const int rowAg1 = (rowA[k] >> 5) & 0x3f;
421 const int rowBg0 = (rowB[j] >> 5) & 0x3f;
422 const int rowBg1 = (rowB[k] >> 5) & 0x3f;
423 const int rowAb0 = (rowA[j] >> 11) & 0x1f;
424 const int rowAb1 = (rowA[k] >> 11) & 0x1f;
425 const int rowBb0 = (rowB[j] >> 11) & 0x1f;
426 const int rowBb1 = (rowB[k] >> 11) & 0x1f;
427 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
428 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
429 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
430 dst[i] = (blue << 11) | (green << 5) | red;
431 }
432 }
433 else if (datatype == DTYPE_USHORT_4_4_4_4 && comps == 4) {
434 uint i, j, k;
435 const ushort *rowA = (const ushort *) srcRowA;
436 const ushort *rowB = (const ushort *) srcRowB;
437 ushort *dst = (ushort *) dstRow;
438 for (i = j = 0, k = k0; i < (uint) dstWidth;
439 i++, j += colStride, k += colStride) {
440 const int rowAr0 = rowA[j] & 0xf;
441 const int rowAr1 = rowA[k] & 0xf;
442 const int rowBr0 = rowB[j] & 0xf;
443 const int rowBr1 = rowB[k] & 0xf;
444 const int rowAg0 = (rowA[j] >> 4) & 0xf;
445 const int rowAg1 = (rowA[k] >> 4) & 0xf;
446 const int rowBg0 = (rowB[j] >> 4) & 0xf;
447 const int rowBg1 = (rowB[k] >> 4) & 0xf;
448 const int rowAb0 = (rowA[j] >> 8) & 0xf;
449 const int rowAb1 = (rowA[k] >> 8) & 0xf;
450 const int rowBb0 = (rowB[j] >> 8) & 0xf;
451 const int rowBb1 = (rowB[k] >> 8) & 0xf;
452 const int rowAa0 = (rowA[j] >> 12) & 0xf;
453 const int rowAa1 = (rowA[k] >> 12) & 0xf;
454 const int rowBa0 = (rowB[j] >> 12) & 0xf;
455 const int rowBa1 = (rowB[k] >> 12) & 0xf;
456 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
457 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
458 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
459 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
460 dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
461 }
462 }
463 else if (datatype == DTYPE_USHORT_1_5_5_5_REV && comps == 4) {
464 uint i, j, k;
465 const ushort *rowA = (const ushort *) srcRowA;
466 const ushort *rowB = (const ushort *) srcRowB;
467 ushort *dst = (ushort *) dstRow;
468 for (i = j = 0, k = k0; i < (uint) dstWidth;
469 i++, j += colStride, k += colStride) {
470 const int rowAr0 = rowA[j] & 0x1f;
471 const int rowAr1 = rowA[k] & 0x1f;
472 const int rowBr0 = rowB[j] & 0x1f;
473 const int rowBr1 = rowB[k] & 0x1f;
474 const int rowAg0 = (rowA[j] >> 5) & 0x1f;
475 const int rowAg1 = (rowA[k] >> 5) & 0x1f;
476 const int rowBg0 = (rowB[j] >> 5) & 0x1f;
477 const int rowBg1 = (rowB[k] >> 5) & 0x1f;
478 const int rowAb0 = (rowA[j] >> 10) & 0x1f;
479 const int rowAb1 = (rowA[k] >> 10) & 0x1f;
480 const int rowBb0 = (rowB[j] >> 10) & 0x1f;
481 const int rowBb1 = (rowB[k] >> 10) & 0x1f;
482 const int rowAa0 = (rowA[j] >> 15) & 0x1;
483 const int rowAa1 = (rowA[k] >> 15) & 0x1;
484 const int rowBa0 = (rowB[j] >> 15) & 0x1;
485 const int rowBa1 = (rowB[k] >> 15) & 0x1;
486 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
487 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
488 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
489 const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
490 dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
491 }
492 }
493 else if (datatype == DTYPE_UBYTE_3_3_2 && comps == 3) {
494 uint i, j, k;
495 const ubyte *rowA = (const ubyte *) srcRowA;
496 const ubyte *rowB = (const ubyte *) srcRowB;
497 ubyte *dst = (ubyte *) dstRow;
498 for (i = j = 0, k = k0; i < (uint) dstWidth;
499 i++, j += colStride, k += colStride) {
500 const int rowAr0 = rowA[j] & 0x3;
501 const int rowAr1 = rowA[k] & 0x3;
502 const int rowBr0 = rowB[j] & 0x3;
503 const int rowBr1 = rowB[k] & 0x3;
504 const int rowAg0 = (rowA[j] >> 2) & 0x7;
505 const int rowAg1 = (rowA[k] >> 2) & 0x7;
506 const int rowBg0 = (rowB[j] >> 2) & 0x7;
507 const int rowBg1 = (rowB[k] >> 2) & 0x7;
508 const int rowAb0 = (rowA[j] >> 5) & 0x7;
509 const int rowAb1 = (rowA[k] >> 5) & 0x7;
510 const int rowBb0 = (rowB[j] >> 5) & 0x7;
511 const int rowBb1 = (rowB[k] >> 5) & 0x7;
512 const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
513 const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
514 const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
515 dst[i] = (blue << 5) | (green << 2) | red;
516 }
517 }
518 else {
519 debug_printf("bad format in do_row()");
520 }
521 }
522
523
524 /**
525 * Average together four rows of a source image to produce a single new
526 * row in the dest image. It's legal for the two source rows to point
527 * to the same data. The source width must be equal to either the
528 * dest width or two times the dest width.
529 *
530 * \param datatype GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
531 * \c GL_FLOAT, etc.
532 * \param comps number of components per pixel (1..4)
533 * \param srcWidth Width of a row in the source data
534 * \param srcRowA Pointer to one of the rows of source data
535 * \param srcRowB Pointer to one of the rows of source data
536 * \param srcRowC Pointer to one of the rows of source data
537 * \param srcRowD Pointer to one of the rows of source data
538 * \param dstWidth Width of a row in the destination data
539 * \param srcRowA Pointer to the row of destination data
540 */
541 static void
542 do_row_3D(enum dtype datatype, uint comps, int srcWidth,
543 const void *srcRowA, const void *srcRowB,
544 const void *srcRowC, const void *srcRowD,
545 int dstWidth, void *dstRow)
546 {
547 const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
548 const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
549 uint i, j, k;
550
551 assert(comps >= 1);
552 assert(comps <= 4);
553
554 if ((datatype == DTYPE_UBYTE) && (comps == 4)) {
555 DECLARE_ROW_POINTERS(ubyte, 4);
556
557 for (i = j = 0, k = k0; i < (uint) dstWidth;
558 i++, j += colStride, k += colStride) {
559 FILTER_3D(0);
560 FILTER_3D(1);
561 FILTER_3D(2);
562 FILTER_3D(3);
563 }
564 }
565 else if ((datatype == DTYPE_UBYTE) && (comps == 3)) {
566 DECLARE_ROW_POINTERS(ubyte, 3);
567
568 for (i = j = 0, k = k0; i < (uint) dstWidth;
569 i++, j += colStride, k += colStride) {
570 FILTER_3D(0);
571 FILTER_3D(1);
572 FILTER_3D(2);
573 }
574 }
575 else if ((datatype == DTYPE_UBYTE) && (comps == 2)) {
576 DECLARE_ROW_POINTERS(ubyte, 2);
577
578 for (i = j = 0, k = k0; i < (uint) dstWidth;
579 i++, j += colStride, k += colStride) {
580 FILTER_3D(0);
581 FILTER_3D(1);
582 }
583 }
584 else if ((datatype == DTYPE_UBYTE) && (comps == 1)) {
585 DECLARE_ROW_POINTERS(ubyte, 1);
586
587 for (i = j = 0, k = k0; i < (uint) dstWidth;
588 i++, j += colStride, k += colStride) {
589 FILTER_3D(0);
590 }
591 }
592 else if ((datatype == DTYPE_USHORT) && (comps == 4)) {
593 DECLARE_ROW_POINTERS(ushort, 4);
594
595 for (i = j = 0, k = k0; i < (uint) dstWidth;
596 i++, j += colStride, k += colStride) {
597 FILTER_3D(0);
598 FILTER_3D(1);
599 FILTER_3D(2);
600 FILTER_3D(3);
601 }
602 }
603 else if ((datatype == DTYPE_USHORT) && (comps == 3)) {
604 DECLARE_ROW_POINTERS(ushort, 3);
605
606 for (i = j = 0, k = k0; i < (uint) dstWidth;
607 i++, j += colStride, k += colStride) {
608 FILTER_3D(0);
609 FILTER_3D(1);
610 FILTER_3D(2);
611 }
612 }
613 else if ((datatype == DTYPE_USHORT) && (comps == 2)) {
614 DECLARE_ROW_POINTERS(ushort, 2);
615
616 for (i = j = 0, k = k0; i < (uint) dstWidth;
617 i++, j += colStride, k += colStride) {
618 FILTER_3D(0);
619 FILTER_3D(1);
620 }
621 }
622 else if ((datatype == DTYPE_USHORT) && (comps == 1)) {
623 DECLARE_ROW_POINTERS(ushort, 1);
624
625 for (i = j = 0, k = k0; i < (uint) dstWidth;
626 i++, j += colStride, k += colStride) {
627 FILTER_3D(0);
628 }
629 }
630 else if ((datatype == DTYPE_FLOAT) && (comps == 4)) {
631 DECLARE_ROW_POINTERS(float, 4);
632
633 for (i = j = 0, k = k0; i < (uint) dstWidth;
634 i++, j += colStride, k += colStride) {
635 FILTER_F_3D(0);
636 FILTER_F_3D(1);
637 FILTER_F_3D(2);
638 FILTER_F_3D(3);
639 }
640 }
641 else if ((datatype == DTYPE_FLOAT) && (comps == 3)) {
642 DECLARE_ROW_POINTERS(float, 3);
643
644 for (i = j = 0, k = k0; i < (uint) dstWidth;
645 i++, j += colStride, k += colStride) {
646 FILTER_F_3D(0);
647 FILTER_F_3D(1);
648 FILTER_F_3D(2);
649 }
650 }
651 else if ((datatype == DTYPE_FLOAT) && (comps == 2)) {
652 DECLARE_ROW_POINTERS(float, 2);
653
654 for (i = j = 0, k = k0; i < (uint) dstWidth;
655 i++, j += colStride, k += colStride) {
656 FILTER_F_3D(0);
657 FILTER_F_3D(1);
658 }
659 }
660 else if ((datatype == DTYPE_FLOAT) && (comps == 1)) {
661 DECLARE_ROW_POINTERS(float, 1);
662
663 for (i = j = 0, k = k0; i < (uint) dstWidth;
664 i++, j += colStride, k += colStride) {
665 FILTER_F_3D(0);
666 }
667 }
668 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 4)) {
669 DECLARE_ROW_POINTERS(half_float, 4);
670
671 for (i = j = 0, k = k0; i < (uint) dstWidth;
672 i++, j += colStride, k += colStride) {
673 FILTER_HF_3D(0);
674 FILTER_HF_3D(1);
675 FILTER_HF_3D(2);
676 FILTER_HF_3D(3);
677 }
678 }
679 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 3)) {
680 DECLARE_ROW_POINTERS(half_float, 4);
681
682 for (i = j = 0, k = k0; i < (uint) dstWidth;
683 i++, j += colStride, k += colStride) {
684 FILTER_HF_3D(0);
685 FILTER_HF_3D(1);
686 FILTER_HF_3D(2);
687 }
688 }
689 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 2)) {
690 DECLARE_ROW_POINTERS(half_float, 4);
691
692 for (i = j = 0, k = k0; i < (uint) dstWidth;
693 i++, j += colStride, k += colStride) {
694 FILTER_HF_3D(0);
695 FILTER_HF_3D(1);
696 }
697 }
698 else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 1)) {
699 DECLARE_ROW_POINTERS(half_float, 4);
700
701 for (i = j = 0, k = k0; i < (uint) dstWidth;
702 i++, j += colStride, k += colStride) {
703 FILTER_HF_3D(0);
704 }
705 }
706 else if ((datatype == DTYPE_UINT) && (comps == 1)) {
707 const uint *rowA = (const uint *) srcRowA;
708 const uint *rowB = (const uint *) srcRowB;
709 const uint *rowC = (const uint *) srcRowC;
710 const uint *rowD = (const uint *) srcRowD;
711 float *dst = (float *) dstRow;
712
713 for (i = j = 0, k = k0; i < (uint) dstWidth;
714 i++, j += colStride, k += colStride) {
715 const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k])
716 + ((uint64_t) rowB[j] + (uint64_t) rowB[k])
717 + ((uint64_t) rowC[j] + (uint64_t) rowC[k])
718 + ((uint64_t) rowD[j] + (uint64_t) rowD[k]));
719 dst[i] = (float)((double) tmp * 0.125);
720 }
721 }
722 else if ((datatype == DTYPE_USHORT_5_6_5) && (comps == 3)) {
723 DECLARE_ROW_POINTERS0(ushort);
724
725 for (i = j = 0, k = k0; i < (uint) dstWidth;
726 i++, j += colStride, k += colStride) {
727 const int rowAr0 = rowA[j] & 0x1f;
728 const int rowAr1 = rowA[k] & 0x1f;
729 const int rowBr0 = rowB[j] & 0x1f;
730 const int rowBr1 = rowB[k] & 0x1f;
731 const int rowCr0 = rowC[j] & 0x1f;
732 const int rowCr1 = rowC[k] & 0x1f;
733 const int rowDr0 = rowD[j] & 0x1f;
734 const int rowDr1 = rowD[k] & 0x1f;
735 const int rowAg0 = (rowA[j] >> 5) & 0x3f;
736 const int rowAg1 = (rowA[k] >> 5) & 0x3f;
737 const int rowBg0 = (rowB[j] >> 5) & 0x3f;
738 const int rowBg1 = (rowB[k] >> 5) & 0x3f;
739 const int rowCg0 = (rowC[j] >> 5) & 0x3f;
740 const int rowCg1 = (rowC[k] >> 5) & 0x3f;
741 const int rowDg0 = (rowD[j] >> 5) & 0x3f;
742 const int rowDg1 = (rowD[k] >> 5) & 0x3f;
743 const int rowAb0 = (rowA[j] >> 11) & 0x1f;
744 const int rowAb1 = (rowA[k] >> 11) & 0x1f;
745 const int rowBb0 = (rowB[j] >> 11) & 0x1f;
746 const int rowBb1 = (rowB[k] >> 11) & 0x1f;
747 const int rowCb0 = (rowC[j] >> 11) & 0x1f;
748 const int rowCb1 = (rowC[k] >> 11) & 0x1f;
749 const int rowDb0 = (rowD[j] >> 11) & 0x1f;
750 const int rowDb1 = (rowD[k] >> 11) & 0x1f;
751 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
752 rowCr0, rowCr1, rowDr0, rowDr1);
753 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
754 rowCg0, rowCg1, rowDg0, rowDg1);
755 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
756 rowCb0, rowCb1, rowDb0, rowDb1);
757 dst[i] = (b << 11) | (g << 5) | r;
758 }
759 }
760 else if ((datatype == DTYPE_USHORT_4_4_4_4) && (comps == 4)) {
761 DECLARE_ROW_POINTERS0(ushort);
762
763 for (i = j = 0, k = k0; i < (uint) dstWidth;
764 i++, j += colStride, k += colStride) {
765 const int rowAr0 = rowA[j] & 0xf;
766 const int rowAr1 = rowA[k] & 0xf;
767 const int rowBr0 = rowB[j] & 0xf;
768 const int rowBr1 = rowB[k] & 0xf;
769 const int rowCr0 = rowC[j] & 0xf;
770 const int rowCr1 = rowC[k] & 0xf;
771 const int rowDr0 = rowD[j] & 0xf;
772 const int rowDr1 = rowD[k] & 0xf;
773 const int rowAg0 = (rowA[j] >> 4) & 0xf;
774 const int rowAg1 = (rowA[k] >> 4) & 0xf;
775 const int rowBg0 = (rowB[j] >> 4) & 0xf;
776 const int rowBg1 = (rowB[k] >> 4) & 0xf;
777 const int rowCg0 = (rowC[j] >> 4) & 0xf;
778 const int rowCg1 = (rowC[k] >> 4) & 0xf;
779 const int rowDg0 = (rowD[j] >> 4) & 0xf;
780 const int rowDg1 = (rowD[k] >> 4) & 0xf;
781 const int rowAb0 = (rowA[j] >> 8) & 0xf;
782 const int rowAb1 = (rowA[k] >> 8) & 0xf;
783 const int rowBb0 = (rowB[j] >> 8) & 0xf;
784 const int rowBb1 = (rowB[k] >> 8) & 0xf;
785 const int rowCb0 = (rowC[j] >> 8) & 0xf;
786 const int rowCb1 = (rowC[k] >> 8) & 0xf;
787 const int rowDb0 = (rowD[j] >> 8) & 0xf;
788 const int rowDb1 = (rowD[k] >> 8) & 0xf;
789 const int rowAa0 = (rowA[j] >> 12) & 0xf;
790 const int rowAa1 = (rowA[k] >> 12) & 0xf;
791 const int rowBa0 = (rowB[j] >> 12) & 0xf;
792 const int rowBa1 = (rowB[k] >> 12) & 0xf;
793 const int rowCa0 = (rowC[j] >> 12) & 0xf;
794 const int rowCa1 = (rowC[k] >> 12) & 0xf;
795 const int rowDa0 = (rowD[j] >> 12) & 0xf;
796 const int rowDa1 = (rowD[k] >> 12) & 0xf;
797 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
798 rowCr0, rowCr1, rowDr0, rowDr1);
799 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
800 rowCg0, rowCg1, rowDg0, rowDg1);
801 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
802 rowCb0, rowCb1, rowDb0, rowDb1);
803 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
804 rowCa0, rowCa1, rowDa0, rowDa1);
805
806 dst[i] = (a << 12) | (b << 8) | (g << 4) | r;
807 }
808 }
809 else if ((datatype == DTYPE_USHORT_1_5_5_5_REV) && (comps == 4)) {
810 DECLARE_ROW_POINTERS0(ushort);
811
812 for (i = j = 0, k = k0; i < (uint) dstWidth;
813 i++, j += colStride, k += colStride) {
814 const int rowAr0 = rowA[j] & 0x1f;
815 const int rowAr1 = rowA[k] & 0x1f;
816 const int rowBr0 = rowB[j] & 0x1f;
817 const int rowBr1 = rowB[k] & 0x1f;
818 const int rowCr0 = rowC[j] & 0x1f;
819 const int rowCr1 = rowC[k] & 0x1f;
820 const int rowDr0 = rowD[j] & 0x1f;
821 const int rowDr1 = rowD[k] & 0x1f;
822 const int rowAg0 = (rowA[j] >> 5) & 0x1f;
823 const int rowAg1 = (rowA[k] >> 5) & 0x1f;
824 const int rowBg0 = (rowB[j] >> 5) & 0x1f;
825 const int rowBg1 = (rowB[k] >> 5) & 0x1f;
826 const int rowCg0 = (rowC[j] >> 5) & 0x1f;
827 const int rowCg1 = (rowC[k] >> 5) & 0x1f;
828 const int rowDg0 = (rowD[j] >> 5) & 0x1f;
829 const int rowDg1 = (rowD[k] >> 5) & 0x1f;
830 const int rowAb0 = (rowA[j] >> 10) & 0x1f;
831 const int rowAb1 = (rowA[k] >> 10) & 0x1f;
832 const int rowBb0 = (rowB[j] >> 10) & 0x1f;
833 const int rowBb1 = (rowB[k] >> 10) & 0x1f;
834 const int rowCb0 = (rowC[j] >> 10) & 0x1f;
835 const int rowCb1 = (rowC[k] >> 10) & 0x1f;
836 const int rowDb0 = (rowD[j] >> 10) & 0x1f;
837 const int rowDb1 = (rowD[k] >> 10) & 0x1f;
838 const int rowAa0 = (rowA[j] >> 15) & 0x1;
839 const int rowAa1 = (rowA[k] >> 15) & 0x1;
840 const int rowBa0 = (rowB[j] >> 15) & 0x1;
841 const int rowBa1 = (rowB[k] >> 15) & 0x1;
842 const int rowCa0 = (rowC[j] >> 15) & 0x1;
843 const int rowCa1 = (rowC[k] >> 15) & 0x1;
844 const int rowDa0 = (rowD[j] >> 15) & 0x1;
845 const int rowDa1 = (rowD[k] >> 15) & 0x1;
846 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
847 rowCr0, rowCr1, rowDr0, rowDr1);
848 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
849 rowCg0, rowCg1, rowDg0, rowDg1);
850 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
851 rowCb0, rowCb1, rowDb0, rowDb1);
852 const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
853 rowCa0, rowCa1, rowDa0, rowDa1);
854
855 dst[i] = (a << 15) | (b << 10) | (g << 5) | r;
856 }
857 }
858 else if ((datatype == DTYPE_UBYTE_3_3_2) && (comps == 3)) {
859 DECLARE_ROW_POINTERS0(ushort);
860
861 for (i = j = 0, k = k0; i < (uint) dstWidth;
862 i++, j += colStride, k += colStride) {
863 const int rowAr0 = rowA[j] & 0x3;
864 const int rowAr1 = rowA[k] & 0x3;
865 const int rowBr0 = rowB[j] & 0x3;
866 const int rowBr1 = rowB[k] & 0x3;
867 const int rowCr0 = rowC[j] & 0x3;
868 const int rowCr1 = rowC[k] & 0x3;
869 const int rowDr0 = rowD[j] & 0x3;
870 const int rowDr1 = rowD[k] & 0x3;
871 const int rowAg0 = (rowA[j] >> 2) & 0x7;
872 const int rowAg1 = (rowA[k] >> 2) & 0x7;
873 const int rowBg0 = (rowB[j] >> 2) & 0x7;
874 const int rowBg1 = (rowB[k] >> 2) & 0x7;
875 const int rowCg0 = (rowC[j] >> 2) & 0x7;
876 const int rowCg1 = (rowC[k] >> 2) & 0x7;
877 const int rowDg0 = (rowD[j] >> 2) & 0x7;
878 const int rowDg1 = (rowD[k] >> 2) & 0x7;
879 const int rowAb0 = (rowA[j] >> 5) & 0x7;
880 const int rowAb1 = (rowA[k] >> 5) & 0x7;
881 const int rowBb0 = (rowB[j] >> 5) & 0x7;
882 const int rowBb1 = (rowB[k] >> 5) & 0x7;
883 const int rowCb0 = (rowC[j] >> 5) & 0x7;
884 const int rowCb1 = (rowC[k] >> 5) & 0x7;
885 const int rowDb0 = (rowD[j] >> 5) & 0x7;
886 const int rowDb1 = (rowD[k] >> 5) & 0x7;
887 const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
888 rowCr0, rowCr1, rowDr0, rowDr1);
889 const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
890 rowCg0, rowCg1, rowDg0, rowDg1);
891 const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
892 rowCb0, rowCb1, rowDb0, rowDb1);
893 dst[i] = (b << 5) | (g << 2) | r;
894 }
895 }
896 else {
897 debug_printf("bad format in do_row_3D()");
898 }
899 }
900
901
902
903 static void
904 format_to_type_comps(enum pipe_format pformat,
905 enum dtype *datatype, uint *comps)
906 {
907 /* XXX I think this could be implemented in terms of the pf_*() functions */
908 switch (pformat) {
909 case PIPE_FORMAT_B8G8R8A8_UNORM:
910 case PIPE_FORMAT_B8G8R8X8_UNORM:
911 case PIPE_FORMAT_A8R8G8B8_UNORM:
912 case PIPE_FORMAT_X8R8G8B8_UNORM:
913 case PIPE_FORMAT_A8B8G8R8_SRGB:
914 case PIPE_FORMAT_X8B8G8R8_SRGB:
915 case PIPE_FORMAT_B8G8R8A8_SRGB:
916 case PIPE_FORMAT_B8G8R8X8_SRGB:
917 case PIPE_FORMAT_A8R8G8B8_SRGB:
918 case PIPE_FORMAT_X8R8G8B8_SRGB:
919 case PIPE_FORMAT_R8G8B8_SRGB:
920 *datatype = DTYPE_UBYTE;
921 *comps = 4;
922 return;
923 case PIPE_FORMAT_B5G5R5X1_UNORM:
924 case PIPE_FORMAT_B5G5R5A1_UNORM:
925 *datatype = DTYPE_USHORT_1_5_5_5_REV;
926 *comps = 4;
927 return;
928 case PIPE_FORMAT_B4G4R4A4_UNORM:
929 *datatype = DTYPE_USHORT_4_4_4_4;
930 *comps = 4;
931 return;
932 case PIPE_FORMAT_B5G6R5_UNORM:
933 *datatype = DTYPE_USHORT_5_6_5;
934 *comps = 3;
935 return;
936 case PIPE_FORMAT_L8_UNORM:
937 case PIPE_FORMAT_L8_SRGB:
938 case PIPE_FORMAT_A8_UNORM:
939 case PIPE_FORMAT_I8_UNORM:
940 *datatype = DTYPE_UBYTE;
941 *comps = 1;
942 return;
943 case PIPE_FORMAT_L8A8_UNORM:
944 case PIPE_FORMAT_L8A8_SRGB:
945 *datatype = DTYPE_UBYTE;
946 *comps = 2;
947 return;
948 default:
949 assert(0);
950 *datatype = DTYPE_UBYTE;
951 *comps = 0;
952 break;
953 }
954 }
955
956
957 static void
958 reduce_1d(enum pipe_format pformat,
959 int srcWidth, const ubyte *srcPtr,
960 int dstWidth, ubyte *dstPtr)
961 {
962 enum dtype datatype;
963 uint comps;
964
965 format_to_type_comps(pformat, &datatype, &comps);
966
967 /* we just duplicate the input row, kind of hack, saves code */
968 do_row(datatype, comps,
969 srcWidth, srcPtr, srcPtr,
970 dstWidth, dstPtr);
971 }
972
973
974 /**
975 * Strides are in bytes. If zero, it'll be computed as width * bpp.
976 */
977 static void
978 reduce_2d(enum pipe_format pformat,
979 int srcWidth, int srcHeight,
980 int srcRowStride, const ubyte *srcPtr,
981 int dstWidth, int dstHeight,
982 int dstRowStride, ubyte *dstPtr)
983 {
984 enum dtype datatype;
985 uint comps;
986 const int bpt = util_format_get_blocksize(pformat);
987 const ubyte *srcA, *srcB;
988 ubyte *dst;
989 int row;
990
991 format_to_type_comps(pformat, &datatype, &comps);
992
993 if (!srcRowStride)
994 srcRowStride = bpt * srcWidth;
995
996 if (!dstRowStride)
997 dstRowStride = bpt * dstWidth;
998
999 /* Compute src and dst pointers */
1000 srcA = srcPtr;
1001 if (srcHeight > 1)
1002 srcB = srcA + srcRowStride;
1003 else
1004 srcB = srcA;
1005 dst = dstPtr;
1006
1007 for (row = 0; row < dstHeight; row++) {
1008 do_row(datatype, comps,
1009 srcWidth, srcA, srcB,
1010 dstWidth, dst);
1011 srcA += 2 * srcRowStride;
1012 srcB += 2 * srcRowStride;
1013 dst += dstRowStride;
1014 }
1015 }
1016
1017
1018 static void
1019 reduce_3d(enum pipe_format pformat,
1020 int srcWidth, int srcHeight, int srcDepth,
1021 int srcRowStride, int srcImageStride, const ubyte *srcPtr,
1022 int dstWidth, int dstHeight, int dstDepth,
1023 int dstRowStride, int dstImageStride, ubyte *dstPtr)
1024 {
1025 const int bpt = util_format_get_blocksize(pformat);
1026 int img, row;
1027 int srcImageOffset, srcRowOffset;
1028 enum dtype datatype;
1029 uint comps;
1030
1031 format_to_type_comps(pformat, &datatype, &comps);
1032
1033 /* XXX I think we should rather assert those strides */
1034 if (!srcImageStride)
1035 srcImageStride = srcWidth * srcHeight * bpt;
1036 if (!dstImageStride)
1037 dstImageStride = dstWidth * dstHeight * bpt;
1038
1039 if (!srcRowStride)
1040 srcRowStride = srcWidth * bpt;
1041 if (!dstRowStride)
1042 dstRowStride = dstWidth * bpt;
1043
1044 /* Offset between adjacent src images to be averaged together */
1045 srcImageOffset = (srcDepth == dstDepth) ? 0 : srcImageStride;
1046
1047 /* Offset between adjacent src rows to be averaged together */
1048 srcRowOffset = (srcHeight == dstHeight) ? 0 : srcRowStride;
1049
1050 /*
1051 * Need to average together up to 8 src pixels for each dest pixel.
1052 * Break that down into 3 operations:
1053 * 1. take two rows from source image and average them together.
1054 * 2. take two rows from next source image and average them together.
1055 * 3. take the two averaged rows and average them for the final dst row.
1056 */
1057
1058 /*
1059 printf("mip3d %d x %d x %d -> %d x %d x %d\n",
1060 srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
1061 */
1062
1063 for (img = 0; img < dstDepth; img++) {
1064 /* first source image pointer */
1065 const ubyte *imgSrcA = srcPtr
1066 + img * (srcImageStride + srcImageOffset);
1067 /* second source image pointer */
1068 const ubyte *imgSrcB = imgSrcA + srcImageOffset;
1069 /* address of the dest image */
1070 ubyte *imgDst = dstPtr + img * dstImageStride;
1071
1072 /* setup the four source row pointers and the dest row pointer */
1073 const ubyte *srcImgARowA = imgSrcA;
1074 const ubyte *srcImgARowB = imgSrcA + srcRowOffset;
1075 const ubyte *srcImgBRowA = imgSrcB;
1076 const ubyte *srcImgBRowB = imgSrcB + srcRowOffset;
1077 ubyte *dstImgRow = imgDst;
1078
1079 for (row = 0; row < dstHeight; row++) {
1080 do_row_3D(datatype, comps, srcWidth,
1081 srcImgARowA, srcImgARowB,
1082 srcImgBRowA, srcImgBRowB,
1083 dstWidth, dstImgRow);
1084
1085 /* advance to next rows */
1086 srcImgARowA += srcRowStride + srcRowOffset;
1087 srcImgARowB += srcRowStride + srcRowOffset;
1088 srcImgBRowA += srcRowStride + srcRowOffset;
1089 srcImgBRowB += srcRowStride + srcRowOffset;
1090 dstImgRow += dstImageStride;
1091 }
1092 }
1093 }
1094
1095
1096
1097
1098 static void
1099 make_1d_mipmap(struct gen_mipmap_state *ctx,
1100 struct pipe_resource *pt,
1101 uint layer, uint baseLevel, uint lastLevel)
1102 {
1103 struct pipe_context *pipe = ctx->pipe;
1104 uint dstLevel;
1105
1106 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1107 const uint srcLevel = dstLevel - 1;
1108 struct pipe_transfer *srcTrans, *dstTrans;
1109 void *srcMap, *dstMap;
1110
1111 srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer,
1112 PIPE_TRANSFER_READ, 0, 0,
1113 u_minify(pt->width0, srcLevel),
1114 u_minify(pt->height0, srcLevel));
1115 dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer,
1116 PIPE_TRANSFER_WRITE, 0, 0,
1117 u_minify(pt->width0, dstLevel),
1118 u_minify(pt->height0, dstLevel));
1119
1120 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1121 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1122
1123 reduce_1d(pt->format,
1124 srcTrans->box.width, srcMap,
1125 dstTrans->box.width, dstMap);
1126
1127 pipe->transfer_unmap(pipe, srcTrans);
1128 pipe->transfer_unmap(pipe, dstTrans);
1129
1130 pipe->transfer_destroy(pipe, srcTrans);
1131 pipe->transfer_destroy(pipe, dstTrans);
1132 }
1133 }
1134
1135
1136 static void
1137 make_2d_mipmap(struct gen_mipmap_state *ctx,
1138 struct pipe_resource *pt,
1139 uint layer, uint baseLevel, uint lastLevel)
1140 {
1141 struct pipe_context *pipe = ctx->pipe;
1142 uint dstLevel;
1143
1144 assert(util_format_get_blockwidth(pt->format) == 1);
1145 assert(util_format_get_blockheight(pt->format) == 1);
1146
1147 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1148 const uint srcLevel = dstLevel - 1;
1149 struct pipe_transfer *srcTrans, *dstTrans;
1150 ubyte *srcMap, *dstMap;
1151
1152 srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer,
1153 PIPE_TRANSFER_READ, 0, 0,
1154 u_minify(pt->width0, srcLevel),
1155 u_minify(pt->height0, srcLevel));
1156 dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer,
1157 PIPE_TRANSFER_WRITE, 0, 0,
1158 u_minify(pt->width0, dstLevel),
1159 u_minify(pt->height0, dstLevel));
1160
1161 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1162 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1163
1164 reduce_2d(pt->format,
1165 srcTrans->box.width, srcTrans->box.height,
1166 srcTrans->stride, srcMap,
1167 dstTrans->box.width, dstTrans->box.height,
1168 dstTrans->stride, dstMap);
1169
1170 pipe->transfer_unmap(pipe, srcTrans);
1171 pipe->transfer_unmap(pipe, dstTrans);
1172
1173 pipe->transfer_destroy(pipe, srcTrans);
1174 pipe->transfer_destroy(pipe, dstTrans);
1175 }
1176 }
1177
1178
1179 /* XXX looks a bit more like it could work now but need to test */
1180 static void
1181 make_3d_mipmap(struct gen_mipmap_state *ctx,
1182 struct pipe_resource *pt,
1183 uint face, uint baseLevel, uint lastLevel)
1184 {
1185 struct pipe_context *pipe = ctx->pipe;
1186 uint dstLevel;
1187 struct pipe_box src_box, dst_box;
1188
1189 assert(util_format_get_blockwidth(pt->format) == 1);
1190 assert(util_format_get_blockheight(pt->format) == 1);
1191
1192 src_box.x = src_box.y = src_box.z = 0;
1193 dst_box.x = dst_box.y = dst_box.z = 0;
1194
1195 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1196 const uint srcLevel = dstLevel - 1;
1197 struct pipe_transfer *srcTrans, *dstTrans;
1198 ubyte *srcMap, *dstMap;
1199 struct pipe_box src_box, dst_box;
1200 src_box.width = u_minify(pt->width0, srcLevel);
1201 src_box.height = u_minify(pt->height0, srcLevel);
1202 src_box.depth = u_minify(pt->depth0, srcLevel);
1203 dst_box.width = u_minify(pt->width0, dstLevel);
1204 dst_box.height = u_minify(pt->height0, dstLevel);
1205 dst_box.depth = u_minify(pt->depth0, dstLevel);
1206
1207 srcTrans = pipe->get_transfer(pipe, pt, srcLevel,
1208 PIPE_TRANSFER_READ,
1209 &src_box);
1210 dstTrans = pipe->get_transfer(pipe, pt, dstLevel,
1211 PIPE_TRANSFER_WRITE,
1212 &dst_box);
1213
1214 srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1215 dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1216
1217 reduce_3d(pt->format,
1218 srcTrans->box.width, srcTrans->box.height, srcTrans->box.depth,
1219 srcTrans->stride, srcTrans->layer_stride, srcMap,
1220 dstTrans->box.width, dstTrans->box.height, dstTrans->box.depth,
1221 dstTrans->stride, dstTrans->layer_stride, dstMap);
1222
1223 pipe->transfer_unmap(pipe, srcTrans);
1224 pipe->transfer_unmap(pipe, dstTrans);
1225
1226 pipe->transfer_destroy(pipe, srcTrans);
1227 pipe->transfer_destroy(pipe, dstTrans);
1228 }
1229 }
1230
1231
1232 static void
1233 fallback_gen_mipmap(struct gen_mipmap_state *ctx,
1234 struct pipe_resource *pt,
1235 uint layer, uint baseLevel, uint lastLevel)
1236 {
1237 switch (pt->target) {
1238 case PIPE_TEXTURE_1D:
1239 make_1d_mipmap(ctx, pt, layer, baseLevel, lastLevel);
1240 break;
1241 case PIPE_TEXTURE_2D:
1242 case PIPE_TEXTURE_RECT:
1243 case PIPE_TEXTURE_CUBE:
1244 make_2d_mipmap(ctx, pt, layer, baseLevel, lastLevel);
1245 break;
1246 case PIPE_TEXTURE_3D:
1247 make_3d_mipmap(ctx, pt, layer, baseLevel, lastLevel);
1248 break;
1249 default:
1250 assert(0);
1251 }
1252 }
1253
1254
1255 /**
1256 * Create a mipmap generation context.
1257 * The idea is to create one of these and re-use it each time we need to
1258 * generate a mipmap.
1259 */
1260 struct gen_mipmap_state *
1261 util_create_gen_mipmap(struct pipe_context *pipe,
1262 struct cso_context *cso)
1263 {
1264 struct gen_mipmap_state *ctx;
1265 uint i;
1266
1267 ctx = CALLOC_STRUCT(gen_mipmap_state);
1268 if (!ctx)
1269 return NULL;
1270
1271 ctx->pipe = pipe;
1272 ctx->cso = cso;
1273
1274 /* disabled blending/masking */
1275 memset(&ctx->blend, 0, sizeof(ctx->blend));
1276 ctx->blend.rt[0].colormask = PIPE_MASK_RGBA;
1277
1278 /* no-op depth/stencil/alpha */
1279 memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
1280
1281 /* rasterizer */
1282 memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
1283 ctx->rasterizer.cull_face = PIPE_FACE_NONE;
1284 ctx->rasterizer.gl_rasterization_rules = 1;
1285 ctx->rasterizer.depth_clip = 1;
1286
1287 /* sampler state */
1288 memset(&ctx->sampler, 0, sizeof(ctx->sampler));
1289 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1290 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1291 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1292 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
1293 ctx->sampler.normalized_coords = 1;
1294
1295 /* vertex elements state */
1296 memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
1297 for (i = 0; i < 2; i++) {
1298 ctx->velem[i].src_offset = i * 4 * sizeof(float);
1299 ctx->velem[i].instance_divisor = 0;
1300 ctx->velem[i].vertex_buffer_index = 0;
1301 ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
1302 }
1303
1304 /* vertex data that doesn't change */
1305 for (i = 0; i < 4; i++) {
1306 ctx->vertices[i][0][2] = 0.0f; /* z */
1307 ctx->vertices[i][0][3] = 1.0f; /* w */
1308 ctx->vertices[i][1][3] = 1.0f; /* q */
1309 }
1310
1311 /* Note: the actual vertex buffer is allocated as needed below */
1312
1313 return ctx;
1314 }
1315
1316
1317 /**
1318 * Helper function to set the fragment shaders.
1319 */
1320 static INLINE void
1321 set_fragment_shader(struct gen_mipmap_state *ctx, uint type)
1322 {
1323 if (!ctx->fs[type])
1324 ctx->fs[type] =
1325 util_make_fragment_tex_shader(ctx->pipe, type,
1326 TGSI_INTERPOLATE_LINEAR);
1327
1328 cso_set_fragment_shader_handle(ctx->cso, ctx->fs[type]);
1329 }
1330
1331
1332 /**
1333 * Helper function to set the vertex shader.
1334 */
1335 static INLINE void
1336 set_vertex_shader(struct gen_mipmap_state *ctx)
1337 {
1338 /* vertex shader - still required to provide the linkage between
1339 * fragment shader input semantics and vertex_element/buffers.
1340 */
1341 if (!ctx->vs)
1342 {
1343 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
1344 TGSI_SEMANTIC_GENERIC };
1345 const uint semantic_indexes[] = { 0, 0 };
1346 ctx->vs = util_make_vertex_passthrough_shader(ctx->pipe, 2,
1347 semantic_names,
1348 semantic_indexes);
1349 }
1350
1351 cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
1352 }
1353
1354
1355 /**
1356 * Get next "slot" of vertex space in the vertex buffer.
1357 * We're allocating one large vertex buffer and using it piece by piece.
1358 */
1359 static unsigned
1360 get_next_slot(struct gen_mipmap_state *ctx)
1361 {
1362 const unsigned max_slots = 4096 / sizeof ctx->vertices;
1363
1364 if (ctx->vbuf_slot >= max_slots) {
1365 pipe_resource_reference(&ctx->vbuf, NULL);
1366 ctx->vbuf_slot = 0;
1367 }
1368
1369 if (!ctx->vbuf) {
1370 ctx->vbuf = pipe_buffer_create(ctx->pipe->screen,
1371 PIPE_BIND_VERTEX_BUFFER,
1372 PIPE_USAGE_STREAM,
1373 max_slots * sizeof ctx->vertices);
1374 }
1375
1376 return ctx->vbuf_slot++ * sizeof ctx->vertices;
1377 }
1378
1379
1380 static unsigned
1381 set_vertex_data(struct gen_mipmap_state *ctx,
1382 enum pipe_texture_target tex_target,
1383 uint layer, float r)
1384 {
1385 unsigned offset;
1386
1387 /* vert[0].position */
1388 ctx->vertices[0][0][0] = -1.0f; /*x*/
1389 ctx->vertices[0][0][1] = -1.0f; /*y*/
1390
1391 /* vert[1].position */
1392 ctx->vertices[1][0][0] = 1.0f;
1393 ctx->vertices[1][0][1] = -1.0f;
1394
1395 /* vert[2].position */
1396 ctx->vertices[2][0][0] = 1.0f;
1397 ctx->vertices[2][0][1] = 1.0f;
1398
1399 /* vert[3].position */
1400 ctx->vertices[3][0][0] = -1.0f;
1401 ctx->vertices[3][0][1] = 1.0f;
1402
1403 /* Setup vertex texcoords. This is a little tricky for cube maps. */
1404 if (tex_target == PIPE_TEXTURE_CUBE) {
1405 static const float st[4][2] = {
1406 {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
1407 };
1408
1409 util_map_texcoords2d_onto_cubemap(layer, &st[0][0], 2,
1410 &ctx->vertices[0][1][0], 8);
1411 }
1412 else if (tex_target == PIPE_TEXTURE_1D_ARRAY) {
1413 /* 1D texture array */
1414 ctx->vertices[0][1][0] = 0.0f; /*s*/
1415 ctx->vertices[0][1][1] = r; /*t*/
1416 ctx->vertices[0][1][2] = 0.0f; /*r*/
1417
1418 ctx->vertices[1][1][0] = 1.0f;
1419 ctx->vertices[1][1][1] = r;
1420 ctx->vertices[1][1][2] = 0.0f;
1421
1422 ctx->vertices[2][1][0] = 1.0f;
1423 ctx->vertices[2][1][1] = r;
1424 ctx->vertices[2][1][2] = 0.0f;
1425
1426 ctx->vertices[3][1][0] = 0.0f;
1427 ctx->vertices[3][1][1] = r;
1428 ctx->vertices[3][1][2] = 0.0f;
1429 } else {
1430 /* 1D/2D/3D/2D array */
1431 ctx->vertices[0][1][0] = 0.0f; /*s*/
1432 ctx->vertices[0][1][1] = 0.0f; /*t*/
1433 ctx->vertices[0][1][2] = r; /*r*/
1434
1435 ctx->vertices[1][1][0] = 1.0f;
1436 ctx->vertices[1][1][1] = 0.0f;
1437 ctx->vertices[1][1][2] = r;
1438
1439 ctx->vertices[2][1][0] = 1.0f;
1440 ctx->vertices[2][1][1] = 1.0f;
1441 ctx->vertices[2][1][2] = r;
1442
1443 ctx->vertices[3][1][0] = 0.0f;
1444 ctx->vertices[3][1][1] = 1.0f;
1445 ctx->vertices[3][1][2] = r;
1446 }
1447
1448 offset = get_next_slot( ctx );
1449
1450 pipe_buffer_write_nooverlap(ctx->pipe, ctx->vbuf,
1451 offset, sizeof(ctx->vertices), ctx->vertices);
1452
1453 return offset;
1454 }
1455
1456
1457
1458 /**
1459 * Destroy a mipmap generation context
1460 */
1461 void
1462 util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
1463 {
1464 struct pipe_context *pipe = ctx->pipe;
1465 unsigned i;
1466
1467 for (i = 0; i < Elements(ctx->fs); i++)
1468 if (ctx->fs[i])
1469 pipe->delete_fs_state(pipe, ctx->fs[i]);
1470
1471 if (ctx->vs)
1472 pipe->delete_vs_state(pipe, ctx->vs);
1473
1474 pipe_resource_reference(&ctx->vbuf, NULL);
1475
1476 FREE(ctx);
1477 }
1478
1479
1480 /**
1481 * Generate mipmap images. It's assumed all needed texture memory is
1482 * already allocated.
1483 *
1484 * \param psv the sampler view to the texture to generate mipmap levels for
1485 * \param face which cube face to generate mipmaps for (0 for non-cube maps)
1486 * \param baseLevel the first mipmap level to use as a src
1487 * \param lastLevel the last mipmap level to generate
1488 * \param filter the minification filter used to generate mipmap levels with
1489 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST
1490 */
1491 void
1492 util_gen_mipmap(struct gen_mipmap_state *ctx,
1493 struct pipe_sampler_view *psv,
1494 uint face, uint baseLevel, uint lastLevel, uint filter)
1495 {
1496 struct pipe_context *pipe = ctx->pipe;
1497 struct pipe_screen *screen = pipe->screen;
1498 struct pipe_framebuffer_state fb;
1499 struct pipe_resource *pt = psv->texture;
1500 uint dstLevel;
1501 uint offset;
1502 uint type;
1503
1504 /* The texture object should have room for the levels which we're
1505 * about to generate.
1506 */
1507 assert(lastLevel <= pt->last_level);
1508
1509 /* If this fails, why are we here? */
1510 assert(lastLevel > baseLevel);
1511
1512 assert(filter == PIPE_TEX_FILTER_LINEAR ||
1513 filter == PIPE_TEX_FILTER_NEAREST);
1514
1515 switch (pt->target) {
1516 case PIPE_TEXTURE_1D:
1517 type = TGSI_TEXTURE_1D;
1518 break;
1519 case PIPE_TEXTURE_2D:
1520 type = TGSI_TEXTURE_2D;
1521 break;
1522 case PIPE_TEXTURE_3D:
1523 type = TGSI_TEXTURE_3D;
1524 break;
1525 case PIPE_TEXTURE_CUBE:
1526 type = TGSI_TEXTURE_CUBE;
1527 break;
1528 case PIPE_TEXTURE_1D_ARRAY:
1529 type = TGSI_TEXTURE_1D_ARRAY;
1530 break;
1531 case PIPE_TEXTURE_2D_ARRAY:
1532 type = TGSI_TEXTURE_2D_ARRAY;
1533 break;
1534 default:
1535 assert(0);
1536 type = TGSI_TEXTURE_2D;
1537 }
1538
1539 /* check if we can render in the texture's format */
1540 if (!screen->is_format_supported(screen, psv->format, pt->target,
1541 pt->nr_samples, PIPE_BIND_RENDER_TARGET)) {
1542 fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
1543 return;
1544 }
1545
1546 /* save state (restored below) */
1547 cso_save_blend(ctx->cso);
1548 cso_save_depth_stencil_alpha(ctx->cso);
1549 cso_save_rasterizer(ctx->cso);
1550 cso_save_samplers(ctx->cso);
1551 cso_save_fragment_sampler_views(ctx->cso);
1552 cso_save_stream_outputs(ctx->cso);
1553 cso_save_framebuffer(ctx->cso);
1554 cso_save_fragment_shader(ctx->cso);
1555 cso_save_vertex_shader(ctx->cso);
1556 cso_save_geometry_shader(ctx->cso);
1557 cso_save_viewport(ctx->cso);
1558 cso_save_vertex_elements(ctx->cso);
1559
1560 /* bind our state */
1561 cso_set_blend(ctx->cso, &ctx->blend);
1562 cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
1563 cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
1564 cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
1565 cso_set_stream_outputs(ctx->cso, 0, NULL, 0);
1566
1567 set_fragment_shader(ctx, type);
1568 set_vertex_shader(ctx);
1569 cso_set_geometry_shader_handle(ctx->cso, NULL);
1570
1571 /* init framebuffer state */
1572 memset(&fb, 0, sizeof(fb));
1573 fb.nr_cbufs = 1;
1574
1575 /* set min/mag to same filter for faster sw speed */
1576 ctx->sampler.mag_img_filter = filter;
1577 ctx->sampler.min_img_filter = filter;
1578
1579 /*
1580 * XXX for small mipmap levels, it may be faster to use the software
1581 * fallback path...
1582 */
1583 for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1584 const uint srcLevel = dstLevel - 1;
1585 struct pipe_viewport_state vp;
1586 unsigned nr_layers, layer, i;
1587 float rcoord = 0.0f;
1588
1589 if (pt->target == PIPE_TEXTURE_3D)
1590 nr_layers = u_minify(pt->depth0, dstLevel);
1591 else if (pt->target == PIPE_TEXTURE_2D_ARRAY || pt->target == PIPE_TEXTURE_1D_ARRAY)
1592 nr_layers = pt->array_size;
1593 else
1594 nr_layers = 1;
1595
1596 for (i = 0; i < nr_layers; i++) {
1597 struct pipe_surface *surf, surf_templ;
1598 if (pt->target == PIPE_TEXTURE_3D) {
1599 /* in theory with geom shaders and driver with full layer support
1600 could do that in one go. */
1601 layer = i;
1602 /* XXX hmm really? */
1603 rcoord = (float)layer / (float)nr_layers + 1.0f / (float)(nr_layers * 2);
1604 } else if (pt->target == PIPE_TEXTURE_2D_ARRAY || pt->target == PIPE_TEXTURE_1D_ARRAY) {
1605 layer = i;
1606 rcoord = (float)layer;
1607 } else
1608 layer = face;
1609
1610 memset(&surf_templ, 0, sizeof(surf_templ));
1611 u_surface_default_template(&surf_templ, pt, PIPE_BIND_RENDER_TARGET);
1612 surf_templ.u.tex.level = dstLevel;
1613 surf_templ.u.tex.first_layer = layer;
1614 surf_templ.u.tex.last_layer = layer;
1615 surf = pipe->create_surface(pipe, pt, &surf_templ);
1616
1617 /*
1618 * Setup framebuffer / dest surface
1619 */
1620 fb.cbufs[0] = surf;
1621 fb.width = u_minify(pt->width0, dstLevel);
1622 fb.height = u_minify(pt->height0, dstLevel);
1623 cso_set_framebuffer(ctx->cso, &fb);
1624
1625 /* viewport */
1626 vp.scale[0] = 0.5f * fb.width;
1627 vp.scale[1] = 0.5f * fb.height;
1628 vp.scale[2] = 1.0f;
1629 vp.scale[3] = 1.0f;
1630 vp.translate[0] = 0.5f * fb.width;
1631 vp.translate[1] = 0.5f * fb.height;
1632 vp.translate[2] = 0.0f;
1633 vp.translate[3] = 0.0f;
1634 cso_set_viewport(ctx->cso, &vp);
1635
1636 /*
1637 * Setup sampler state
1638 * Note: we should only have to set the min/max LOD clamps to ensure
1639 * we grab texels from the right mipmap level. But some hardware
1640 * has trouble with min clamping so we also set the lod_bias to
1641 * try to work around that.
1642 */
1643 ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel;
1644 ctx->sampler.lod_bias = (float) srcLevel;
1645 cso_single_sampler(ctx->cso, 0, &ctx->sampler);
1646 cso_single_sampler_done(ctx->cso);
1647
1648 cso_set_fragment_sampler_views(ctx->cso, 1, &psv);
1649
1650 /* quad coords in clip coords */
1651 offset = set_vertex_data(ctx,
1652 pt->target,
1653 face,
1654 rcoord);
1655
1656 util_draw_vertex_buffer(ctx->pipe,
1657 ctx->cso,
1658 ctx->vbuf,
1659 offset,
1660 PIPE_PRIM_TRIANGLE_FAN,
1661 4, /* verts */
1662 2); /* attribs/vert */
1663
1664 /* need to signal that the texture has changed _after_ rendering to it */
1665 pipe_surface_reference( &surf, NULL );
1666 }
1667 }
1668
1669 /* restore state we changed */
1670 cso_restore_blend(ctx->cso);
1671 cso_restore_depth_stencil_alpha(ctx->cso);
1672 cso_restore_rasterizer(ctx->cso);
1673 cso_restore_samplers(ctx->cso);
1674 cso_restore_fragment_sampler_views(ctx->cso);
1675 cso_restore_framebuffer(ctx->cso);
1676 cso_restore_fragment_shader(ctx->cso);
1677 cso_restore_vertex_shader(ctx->cso);
1678 cso_restore_geometry_shader(ctx->cso);
1679 cso_restore_viewport(ctx->cso);
1680 cso_restore_vertex_elements(ctx->cso);
1681 cso_restore_stream_outputs(ctx->cso);
1682 }