1 /**************************************************************************
3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2008 VMware, Inc. All rights reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
31 * Mipmap generation utility
37 #include "pipe/p_context.h"
38 #include "util/u_debug.h"
39 #include "pipe/p_defines.h"
40 #include "pipe/p_inlines.h"
41 #include "pipe/p_shader_tokens.h"
42 #include "pipe/p_state.h"
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"
49 #include "cso_cache/cso_context.h"
52 struct gen_mipmap_state
54 struct pipe_context
*pipe
;
55 struct cso_context
*cso
;
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
;
65 struct pipe_buffer
*vbuf
; /**< quad vertices */
68 float vertices
[4][2][4]; /**< vertex/texcoords for quad */
80 DTYPE_USHORT_1_5_5_5_REV
,
87 typedef ushort half_float
;
91 float_to_half(float f
)
98 half_to_float(half_float h
)
108 * \name Support macros for do_row and do_row_3d
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.
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
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
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 \
136 #define FILTER_3D(e) \
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]); \
144 #define FILTER_F_3D(e) \
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; \
152 #define FILTER_HF_3D(e) \
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) \
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)
177 do_row(enum dtype datatype
, uint comps
, int srcWidth
,
178 const void *srcRowA
, const void *srcRowB
,
179 int dstWidth
, void *dstRow
)
181 const uint k0
= (srcWidth
== dstWidth
) ? 0 : 1;
182 const uint colStride
= (srcWidth
== dstWidth
) ? 1 : 2;
187 /* This assertion is no longer valid with non-power-of-2 textures
188 assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
191 if (datatype
== DTYPE_UBYTE
&& comps
== 4) {
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;
204 else if (datatype
== DTYPE_UBYTE
&& comps
== 3) {
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;
216 else if (datatype
== DTYPE_UBYTE
&& comps
== 2) {
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;
227 else if (datatype
== DTYPE_UBYTE
&& comps
== 1) {
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;
238 else if (datatype
== DTYPE_USHORT
&& comps
== 4) {
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;
251 else if (datatype
== DTYPE_USHORT
&& comps
== 3) {
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;
263 else if (datatype
== DTYPE_USHORT
&& comps
== 2) {
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;
274 else if (datatype
== DTYPE_USHORT
&& comps
== 1) {
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;
285 else if (datatype
== DTYPE_FLOAT
&& comps
== 4) {
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
;
302 else if (datatype
== DTYPE_FLOAT
&& comps
== 3) {
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
;
317 else if (datatype
== DTYPE_FLOAT
&& comps
== 2) {
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
;
330 else if (datatype
== DTYPE_FLOAT
&& comps
== 1) {
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
;
342 else if (datatype
== HALF_DTYPE_FLOAT
&& comps
== 4) {
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
);
359 else if (datatype
== DTYPE_HALF_FLOAT
&& comps
== 3) {
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
);
376 else if (datatype
== DTYPE_HALF_FLOAT
&& comps
== 2) {
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
);
393 else if (datatype
== DTYPE_HALF_FLOAT
&& comps
== 1) {
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
);
410 else if (datatype
== DTYPE_UINT
&& comps
== 1) {
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;
421 else if (datatype
== DTYPE_USHORT_5_6_5
&& comps
== 3) {
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
;
446 else if (datatype
== DTYPE_USHORT_4_4_4_4
&& comps
== 4) {
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
;
476 else if (datatype
== DTYPE_USHORT_1_5_5_5_REV
&& comps
== 4) {
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
;
506 else if (datatype
== DTYPE_UBYTE_3_3_2
&& comps
== 3) {
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
;
532 debug_printf("bad format in do_row()");
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.
543 * \param datatype GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
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
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
)
560 const uint k0
= (srcWidth
== dstWidth
) ? 0 : 1;
561 const uint colStride
= (srcWidth
== dstWidth
) ? 1 : 2;
567 if ((datatype
== DTYPE_UBYTE
) && (comps
== 4)) {
568 DECLARE_ROW_POINTERS(ubyte
, 4);
570 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
571 i
++, j
+= colStride
, k
+= colStride
) {
578 else if ((datatype
== DTYPE_UBYTE
) && (comps
== 3)) {
579 DECLARE_ROW_POINTERS(ubyte
, 3);
581 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
582 i
++, j
+= colStride
, k
+= colStride
) {
588 else if ((datatype
== DTYPE_UBYTE
) && (comps
== 2)) {
589 DECLARE_ROW_POINTERS(ubyte
, 2);
591 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
592 i
++, j
+= colStride
, k
+= colStride
) {
597 else if ((datatype
== DTYPE_UBYTE
) && (comps
== 1)) {
598 DECLARE_ROW_POINTERS(ubyte
, 1);
600 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
601 i
++, j
+= colStride
, k
+= colStride
) {
605 else if ((datatype
== DTYPE_USHORT
) && (comps
== 4)) {
606 DECLARE_ROW_POINTERS(ushort
, 4);
608 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
609 i
++, j
+= colStride
, k
+= colStride
) {
616 else if ((datatype
== DTYPE_USHORT
) && (comps
== 3)) {
617 DECLARE_ROW_POINTERS(ushort
, 3);
619 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
620 i
++, j
+= colStride
, k
+= colStride
) {
626 else if ((datatype
== DTYPE_USHORT
) && (comps
== 2)) {
627 DECLARE_ROW_POINTERS(ushort
, 2);
629 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
630 i
++, j
+= colStride
, k
+= colStride
) {
635 else if ((datatype
== DTYPE_USHORT
) && (comps
== 1)) {
636 DECLARE_ROW_POINTERS(ushort
, 1);
638 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
639 i
++, j
+= colStride
, k
+= colStride
) {
643 else if ((datatype
== DTYPE_FLOAT
) && (comps
== 4)) {
644 DECLARE_ROW_POINTERS(float, 4);
646 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
647 i
++, j
+= colStride
, k
+= colStride
) {
654 else if ((datatype
== DTYPE_FLOAT
) && (comps
== 3)) {
655 DECLARE_ROW_POINTERS(float, 3);
657 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
658 i
++, j
+= colStride
, k
+= colStride
) {
664 else if ((datatype
== DTYPE_FLOAT
) && (comps
== 2)) {
665 DECLARE_ROW_POINTERS(float, 2);
667 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
668 i
++, j
+= colStride
, k
+= colStride
) {
673 else if ((datatype
== DTYPE_FLOAT
) && (comps
== 1)) {
674 DECLARE_ROW_POINTERS(float, 1);
676 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
677 i
++, j
+= colStride
, k
+= colStride
) {
681 else if ((datatype
== DTYPE_HALF_FLOAT
) && (comps
== 4)) {
682 DECLARE_ROW_POINTERS(half_float
, 4);
684 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
685 i
++, j
+= colStride
, k
+= colStride
) {
692 else if ((datatype
== DTYPE_HALF_FLOAT
) && (comps
== 3)) {
693 DECLARE_ROW_POINTERS(half_float
, 4);
695 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
696 i
++, j
+= colStride
, k
+= colStride
) {
702 else if ((datatype
== DTYPE_HALF_FLOAT
) && (comps
== 2)) {
703 DECLARE_ROW_POINTERS(half_float
, 4);
705 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
706 i
++, j
+= colStride
, k
+= colStride
) {
711 else if ((datatype
== DTYPE_HALF_FLOAT
) && (comps
== 1)) {
712 DECLARE_ROW_POINTERS(half_float
, 4);
714 for (i
= j
= 0, k
= k0
; i
< (uint
) dstWidth
;
715 i
++, j
+= colStride
, k
+= colStride
) {
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
;
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);
735 else if ((datatype
== DTYPE_USHORT_5_6_5
) && (comps
== 3)) {
736 DECLARE_ROW_POINTERS0(ushort
);
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
;
773 else if ((datatype
== DTYPE_USHORT_4_4_4_4
) && (comps
== 4)) {
774 DECLARE_ROW_POINTERS0(ushort
);
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
);
819 dst
[i
] = (a
<< 12) | (b
<< 8) | (g
<< 4) | r
;
822 else if ((datatype
== DTYPE_USHORT_1_5_5_5_REV
) && (comps
== 4)) {
823 DECLARE_ROW_POINTERS0(ushort
);
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
);
868 dst
[i
] = (a
<< 15) | (b
<< 10) | (g
<< 5) | r
;
871 else if ((datatype
== DTYPE_UBYTE_3_3_2
) && (comps
== 3)) {
872 DECLARE_ROW_POINTERS0(ushort
);
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
;
910 debug_printf("bad format in do_row_3D()");
917 format_to_type_comps(enum pipe_format pformat
,
918 enum dtype
*datatype
, uint
*comps
)
921 case PIPE_FORMAT_A8R8G8B8_UNORM
:
922 case PIPE_FORMAT_X8R8G8B8_UNORM
:
923 case PIPE_FORMAT_B8G8R8A8_UNORM
:
924 case PIPE_FORMAT_B8G8R8X8_UNORM
:
925 *datatype
= DTYPE_UBYTE
;
928 case PIPE_FORMAT_A1R5G5B5_UNORM
:
929 *datatype
= DTYPE_USHORT_1_5_5_5_REV
;
932 case PIPE_FORMAT_A4R4G4B4_UNORM
:
933 *datatype
= DTYPE_USHORT_4_4_4_4
;
936 case PIPE_FORMAT_R5G6B5_UNORM
:
937 *datatype
= DTYPE_USHORT_5_6_5
;
940 case PIPE_FORMAT_L8_UNORM
:
941 case PIPE_FORMAT_A8_UNORM
:
942 case PIPE_FORMAT_I8_UNORM
:
943 *datatype
= DTYPE_UBYTE
;
946 case PIPE_FORMAT_A8L8_UNORM
:
947 *datatype
= DTYPE_UBYTE
;
952 *datatype
= DTYPE_UBYTE
;
960 reduce_1d(enum pipe_format pformat
,
961 int srcWidth
, const ubyte
*srcPtr
,
962 int dstWidth
, ubyte
*dstPtr
)
967 format_to_type_comps(pformat
, &datatype
, &comps
);
969 /* we just duplicate the input row, kind of hack, saves code */
970 do_row(datatype
, comps
,
971 srcWidth
, srcPtr
, srcPtr
,
977 * Strides are in bytes. If zero, it'll be computed as width * bpp.
980 reduce_2d(enum pipe_format pformat
,
981 int srcWidth
, int srcHeight
,
982 int srcRowStride
, const ubyte
*srcPtr
,
983 int dstWidth
, int dstHeight
,
984 int dstRowStride
, ubyte
*dstPtr
)
988 const int bpt
= pf_get_size(pformat
);
989 const ubyte
*srcA
, *srcB
;
993 format_to_type_comps(pformat
, &datatype
, &comps
);
996 srcRowStride
= bpt
* srcWidth
;
999 dstRowStride
= bpt
* dstWidth
;
1001 /* Compute src and dst pointers */
1004 srcB
= srcA
+ srcRowStride
;
1009 for (row
= 0; row
< dstHeight
; row
++) {
1010 do_row(datatype
, comps
,
1011 srcWidth
, srcA
, srcB
,
1013 srcA
+= 2 * srcRowStride
;
1014 srcB
+= 2 * srcRowStride
;
1015 dst
+= dstRowStride
;
1021 reduce_3d(enum pipe_format pformat
,
1022 int srcWidth
, int srcHeight
, int srcDepth
,
1023 int srcRowStride
, const ubyte
*srcPtr
,
1024 int dstWidth
, int dstHeight
, int dstDepth
,
1025 int dstRowStride
, ubyte
*dstPtr
)
1027 const int bpt
= pf_get_size(pformat
);
1028 const int border
= 0;
1030 int bytesPerSrcImage
, bytesPerDstImage
;
1031 int bytesPerSrcRow
, bytesPerDstRow
;
1032 int srcImageOffset
, srcRowOffset
;
1033 enum dtype datatype
;
1036 format_to_type_comps(pformat
, &datatype
, &comps
);
1038 bytesPerSrcImage
= srcWidth
* srcHeight
* bpt
;
1039 bytesPerDstImage
= dstWidth
* dstHeight
* bpt
;
1041 bytesPerSrcRow
= srcWidth
* bpt
;
1042 bytesPerDstRow
= dstWidth
* bpt
;
1044 /* Offset between adjacent src images to be averaged together */
1045 srcImageOffset
= (srcDepth
== dstDepth
) ? 0 : bytesPerSrcImage
;
1047 /* Offset between adjacent src rows to be averaged together */
1048 srcRowOffset
= (srcHeight
== dstHeight
) ? 0 : srcWidth
* bpt
;
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.
1059 _mesa_printf("mip3d %d x %d x %d -> %d x %d x %d\n",
1060 srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
1063 for (img
= 0; img
< dstDepth
; img
++) {
1064 /* first source image pointer, skipping border */
1065 const ubyte
*imgSrcA
= srcPtr
1066 + (bytesPerSrcImage
+ bytesPerSrcRow
+ border
) * bpt
* border
1067 + img
* (bytesPerSrcImage
+ srcImageOffset
);
1068 /* second source image pointer, skipping border */
1069 const ubyte
*imgSrcB
= imgSrcA
+ srcImageOffset
;
1070 /* address of the dest image, skipping border */
1071 ubyte
*imgDst
= dstPtr
1072 + (bytesPerDstImage
+ bytesPerDstRow
+ border
) * bpt
* border
1073 + img
* bytesPerDstImage
;
1075 /* setup the four source row pointers and the dest row pointer */
1076 const ubyte
*srcImgARowA
= imgSrcA
;
1077 const ubyte
*srcImgARowB
= imgSrcA
+ srcRowOffset
;
1078 const ubyte
*srcImgBRowA
= imgSrcB
;
1079 const ubyte
*srcImgBRowB
= imgSrcB
+ srcRowOffset
;
1080 ubyte
*dstImgRow
= imgDst
;
1082 for (row
= 0; row
< dstHeight
; row
++) {
1083 do_row_3D(datatype
, comps
, srcWidth
,
1084 srcImgARowA
, srcImgARowB
,
1085 srcImgBRowA
, srcImgBRowB
,
1086 dstWidth
, dstImgRow
);
1088 /* advance to next rows */
1089 srcImgARowA
+= bytesPerSrcRow
+ srcRowOffset
;
1090 srcImgARowB
+= bytesPerSrcRow
+ srcRowOffset
;
1091 srcImgBRowA
+= bytesPerSrcRow
+ srcRowOffset
;
1092 srcImgBRowB
+= bytesPerSrcRow
+ srcRowOffset
;
1093 dstImgRow
+= bytesPerDstRow
;
1102 make_1d_mipmap(struct gen_mipmap_state
*ctx
,
1103 struct pipe_texture
*pt
,
1104 uint face
, uint baseLevel
, uint lastLevel
)
1106 struct pipe_context
*pipe
= ctx
->pipe
;
1107 struct pipe_screen
*screen
= pipe
->screen
;
1108 const uint zslice
= 0;
1111 for (dstLevel
= baseLevel
+ 1; dstLevel
<= lastLevel
; dstLevel
++) {
1112 const uint srcLevel
= dstLevel
- 1;
1113 struct pipe_transfer
*srcTrans
, *dstTrans
;
1114 void *srcMap
, *dstMap
;
1116 srcTrans
= screen
->get_tex_transfer(screen
, pt
, face
, srcLevel
, zslice
,
1117 PIPE_TRANSFER_READ
, 0, 0,
1118 pt
->width
[srcLevel
],
1119 pt
->height
[srcLevel
]);
1120 dstTrans
= screen
->get_tex_transfer(screen
, pt
, face
, dstLevel
, zslice
,
1121 PIPE_TRANSFER_WRITE
, 0, 0,
1122 pt
->width
[dstLevel
],
1123 pt
->height
[dstLevel
]);
1125 srcMap
= (ubyte
*) screen
->transfer_map(screen
, srcTrans
);
1126 dstMap
= (ubyte
*) screen
->transfer_map(screen
, dstTrans
);
1128 reduce_1d(pt
->format
,
1129 srcTrans
->width
, srcMap
,
1130 dstTrans
->width
, dstMap
);
1132 screen
->transfer_unmap(screen
, srcTrans
);
1133 screen
->transfer_unmap(screen
, dstTrans
);
1135 screen
->tex_transfer_destroy(srcTrans
);
1136 screen
->tex_transfer_destroy(dstTrans
);
1142 make_2d_mipmap(struct gen_mipmap_state
*ctx
,
1143 struct pipe_texture
*pt
,
1144 uint face
, uint baseLevel
, uint lastLevel
)
1146 struct pipe_context
*pipe
= ctx
->pipe
;
1147 struct pipe_screen
*screen
= pipe
->screen
;
1148 const uint zslice
= 0;
1151 assert(pt
->block
.width
== 1);
1152 assert(pt
->block
.height
== 1);
1154 for (dstLevel
= baseLevel
+ 1; dstLevel
<= lastLevel
; dstLevel
++) {
1155 const uint srcLevel
= dstLevel
- 1;
1156 struct pipe_transfer
*srcTrans
, *dstTrans
;
1157 ubyte
*srcMap
, *dstMap
;
1159 srcTrans
= screen
->get_tex_transfer(screen
, pt
, face
, srcLevel
, zslice
,
1160 PIPE_TRANSFER_READ
, 0, 0,
1161 pt
->width
[srcLevel
],
1162 pt
->height
[srcLevel
]);
1163 dstTrans
= screen
->get_tex_transfer(screen
, pt
, face
, dstLevel
, zslice
,
1164 PIPE_TRANSFER_WRITE
, 0, 0,
1165 pt
->width
[dstLevel
],
1166 pt
->height
[dstLevel
]);
1168 srcMap
= (ubyte
*) screen
->transfer_map(screen
, srcTrans
);
1169 dstMap
= (ubyte
*) screen
->transfer_map(screen
, dstTrans
);
1171 reduce_2d(pt
->format
,
1172 srcTrans
->width
, srcTrans
->height
,
1173 srcTrans
->stride
, srcMap
,
1174 dstTrans
->width
, dstTrans
->height
,
1175 dstTrans
->stride
, dstMap
);
1177 screen
->transfer_unmap(screen
, srcTrans
);
1178 screen
->transfer_unmap(screen
, dstTrans
);
1180 screen
->tex_transfer_destroy(srcTrans
);
1181 screen
->tex_transfer_destroy(dstTrans
);
1187 make_3d_mipmap(struct gen_mipmap_state
*ctx
,
1188 struct pipe_texture
*pt
,
1189 uint face
, uint baseLevel
, uint lastLevel
)
1192 struct pipe_context
*pipe
= ctx
->pipe
;
1193 struct pipe_screen
*screen
= pipe
->screen
;
1194 uint dstLevel
, zslice
= 0;
1196 assert(pt
->block
.width
== 1);
1197 assert(pt
->block
.height
== 1);
1199 for (dstLevel
= baseLevel
+ 1; dstLevel
<= lastLevel
; dstLevel
++) {
1200 const uint srcLevel
= dstLevel
- 1;
1201 struct pipe_transfer
*srcTrans
, *dstTrans
;
1202 ubyte
*srcMap
, *dstMap
;
1204 srcTrans
= screen
->get_tex_transfer(screen
, pt
, face
, srcLevel
, zslice
,
1205 PIPE_TRANSFER_READ
, 0, 0,
1206 pt
->width
[srcLevel
],
1207 pt
->height
[srcLevel
]);
1208 dstTrans
= screen
->get_tex_transfer(screen
, pt
, face
, dstLevel
, zslice
,
1209 PIPE_TRANSFER_WRITE
, 0, 0,
1210 pt
->width
[dstLevel
],
1211 pt
->height
[dstLevel
]);
1213 srcMap
= (ubyte
*) screen
->transfer_map(screen
, srcTrans
);
1214 dstMap
= (ubyte
*) screen
->transfer_map(screen
, dstTrans
);
1216 reduce_3d(pt
->format
,
1217 srcTrans
->width
, srcTrans
->height
,
1218 srcTrans
->stride
, srcMap
,
1219 dstTrans
->width
, dstTrans
->height
,
1220 dstTrans
->stride
, dstMap
);
1222 screen
->transfer_unmap(screen
, srcTrans
);
1223 screen
->transfer_unmap(screen
, dstTrans
);
1225 screen
->tex_transfer_destroy(srcTrans
);
1226 screen
->tex_transfer_destroy(dstTrans
);
1235 fallback_gen_mipmap(struct gen_mipmap_state
*ctx
,
1236 struct pipe_texture
*pt
,
1237 uint face
, uint baseLevel
, uint lastLevel
)
1239 switch (pt
->target
) {
1240 case PIPE_TEXTURE_1D
:
1241 make_1d_mipmap(ctx
, pt
, face
, baseLevel
, lastLevel
);
1243 case PIPE_TEXTURE_2D
:
1244 case PIPE_TEXTURE_CUBE
:
1245 make_2d_mipmap(ctx
, pt
, face
, baseLevel
, lastLevel
);
1247 case PIPE_TEXTURE_3D
:
1248 make_3d_mipmap(ctx
, pt
, face
, baseLevel
, lastLevel
);
1257 * Create a mipmap generation context.
1258 * The idea is to create one of these and re-use it each time we need to
1259 * generate a mipmap.
1261 struct gen_mipmap_state
*
1262 util_create_gen_mipmap(struct pipe_context
*pipe
,
1263 struct cso_context
*cso
)
1265 struct gen_mipmap_state
*ctx
;
1268 ctx
= CALLOC_STRUCT(gen_mipmap_state
);
1275 /* disabled blending/masking */
1276 memset(&ctx
->blend
, 0, sizeof(ctx
->blend
));
1277 ctx
->blend
.colormask
= PIPE_MASK_RGBA
;
1279 /* no-op depth/stencil/alpha */
1280 memset(&ctx
->depthstencil
, 0, sizeof(ctx
->depthstencil
));
1283 memset(&ctx
->rasterizer
, 0, sizeof(ctx
->rasterizer
));
1284 ctx
->rasterizer
.front_winding
= PIPE_WINDING_CW
;
1285 ctx
->rasterizer
.cull_mode
= PIPE_WINDING_NONE
;
1286 ctx
->rasterizer
.bypass_vs_clip_and_viewport
= 1;
1287 ctx
->rasterizer
.gl_rasterization_rules
= 1;
1290 memset(&ctx
->sampler
, 0, sizeof(ctx
->sampler
));
1291 ctx
->sampler
.wrap_s
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1292 ctx
->sampler
.wrap_t
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1293 ctx
->sampler
.wrap_r
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1294 ctx
->sampler
.min_mip_filter
= PIPE_TEX_MIPFILTER_NEAREST
;
1295 ctx
->sampler
.normalized_coords
= 1;
1297 /* vertex shader - still needed to specify mapping from fragment
1298 * shader input semantics to vertex elements
1301 const uint semantic_names
[] = { TGSI_SEMANTIC_POSITION
,
1302 TGSI_SEMANTIC_GENERIC
};
1303 const uint semantic_indexes
[] = { 0, 0 };
1304 ctx
->vs
= util_make_vertex_passthrough_shader(pipe
, 2, semantic_names
,
1308 /* fragment shader */
1309 ctx
->fs
= util_make_fragment_tex_shader(pipe
);
1311 /* vertex data that doesn't change */
1312 for (i
= 0; i
< 4; i
++) {
1313 ctx
->vertices
[i
][0][2] = 0.0f
; /* z */
1314 ctx
->vertices
[i
][0][3] = 1.0f
; /* w */
1315 ctx
->vertices
[i
][1][3] = 1.0f
; /* q */
1318 /* Note: the actual vertex buffer is allocated as needed below */
1325 * Get next "slot" of vertex space in the vertex buffer.
1326 * We're allocating one large vertex buffer and using it piece by piece.
1329 get_next_slot(struct gen_mipmap_state
*ctx
)
1331 const unsigned max_slots
= 4096 / sizeof ctx
->vertices
;
1333 if (ctx
->vbuf_slot
>= max_slots
)
1334 util_gen_mipmap_flush( ctx
);
1337 ctx
->vbuf
= pipe_buffer_create(ctx
->pipe
->screen
,
1339 PIPE_BUFFER_USAGE_VERTEX
,
1340 max_slots
* sizeof ctx
->vertices
);
1343 return ctx
->vbuf_slot
++ * sizeof ctx
->vertices
;
1348 set_vertex_data(struct gen_mipmap_state
*ctx
,
1349 enum pipe_texture_target tex_target
,
1350 uint face
, float width
, float height
)
1354 /* vert[0].position */
1355 ctx
->vertices
[0][0][0] = 0.0f
; /*x*/
1356 ctx
->vertices
[0][0][1] = 0.0f
; /*y*/
1358 /* vert[1].position */
1359 ctx
->vertices
[1][0][0] = width
;
1360 ctx
->vertices
[1][0][1] = 0.0f
;
1362 /* vert[2].position */
1363 ctx
->vertices
[2][0][0] = width
;
1364 ctx
->vertices
[2][0][1] = height
;
1366 /* vert[3].position */
1367 ctx
->vertices
[3][0][0] = 0.0f
;
1368 ctx
->vertices
[3][0][1] = height
;
1370 /* Setup vertex texcoords. This is a little tricky for cube maps. */
1371 if (tex_target
== PIPE_TEXTURE_CUBE
) {
1372 static const float st
[4][2] = {
1373 {0.0f
, 0.0f
}, {1.0f
, 0.0f
}, {1.0f
, 1.0f
}, {0.0f
, 1.0f
}
1378 /* loop over quad verts */
1379 for (i
= 0; i
< 4; i
++) {
1380 /* Compute sc = +/-scale and tc = +/-scale.
1381 * Not +/-1 to avoid cube face selection ambiguity near the edges,
1382 * though that can still sometimes happen with this scale factor...
1384 const float scale
= 0.9999f
;
1385 const float sc
= (2.0f
* st
[i
][0] - 1.0f
) * scale
;
1386 const float tc
= (2.0f
* st
[i
][1] - 1.0f
) * scale
;
1389 case PIPE_TEX_FACE_POS_X
:
1394 case PIPE_TEX_FACE_NEG_X
:
1399 case PIPE_TEX_FACE_POS_Y
:
1404 case PIPE_TEX_FACE_NEG_Y
:
1409 case PIPE_TEX_FACE_POS_Z
:
1414 case PIPE_TEX_FACE_NEG_Z
:
1423 ctx
->vertices
[i
][1][0] = rx
; /*s*/
1424 ctx
->vertices
[i
][1][1] = ry
; /*t*/
1425 ctx
->vertices
[i
][1][2] = rz
; /*r*/
1430 ctx
->vertices
[0][1][0] = 0.0f
; /*s*/
1431 ctx
->vertices
[0][1][1] = 0.0f
; /*t*/
1432 ctx
->vertices
[0][1][2] = 0.0f
; /*r*/
1434 ctx
->vertices
[1][1][0] = 1.0f
;
1435 ctx
->vertices
[1][1][1] = 0.0f
;
1436 ctx
->vertices
[1][1][2] = 0.0f
;
1438 ctx
->vertices
[2][1][0] = 1.0f
;
1439 ctx
->vertices
[2][1][1] = 1.0f
;
1440 ctx
->vertices
[2][1][2] = 0.0f
;
1442 ctx
->vertices
[3][1][0] = 0.0f
;
1443 ctx
->vertices
[3][1][1] = 1.0f
;
1444 ctx
->vertices
[3][1][2] = 0.0f
;
1447 offset
= get_next_slot( ctx
);
1449 pipe_buffer_write(ctx
->pipe
->screen
, ctx
->vbuf
,
1450 offset
, sizeof(ctx
->vertices
), ctx
->vertices
);
1458 * Destroy a mipmap generation context
1461 util_destroy_gen_mipmap(struct gen_mipmap_state
*ctx
)
1463 struct pipe_context
*pipe
= ctx
->pipe
;
1465 pipe
->delete_vs_state(pipe
, ctx
->vs
);
1466 pipe
->delete_fs_state(pipe
, ctx
->fs
);
1468 pipe_buffer_reference(&ctx
->vbuf
, NULL
);
1475 /* Release vertex buffer at end of frame to avoid synchronous
1478 void util_gen_mipmap_flush( struct gen_mipmap_state
*ctx
)
1480 pipe_buffer_reference(&ctx
->vbuf
, NULL
);
1486 * Generate mipmap images. It's assumed all needed texture memory is
1487 * already allocated.
1489 * \param pt the texture to generate mipmap levels for
1490 * \param face which cube face to generate mipmaps for (0 for non-cube maps)
1491 * \param baseLevel the first mipmap level to use as a src
1492 * \param lastLevel the last mipmap level to generate
1493 * \param filter the minification filter used to generate mipmap levels with
1494 * \param filter one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST
1497 util_gen_mipmap(struct gen_mipmap_state
*ctx
,
1498 struct pipe_texture
*pt
,
1499 uint face
, uint baseLevel
, uint lastLevel
, uint filter
)
1501 struct pipe_context
*pipe
= ctx
->pipe
;
1502 struct pipe_screen
*screen
= pipe
->screen
;
1503 struct pipe_framebuffer_state fb
;
1508 /* check if we can render in the texture's format */
1509 if (!screen
->is_format_supported(screen
, pt
->format
, PIPE_TEXTURE_2D
,
1510 PIPE_TEXTURE_USAGE_RENDER_TARGET
, 0)) {
1511 fallback_gen_mipmap(ctx
, pt
, face
, baseLevel
, lastLevel
);
1515 /* save state (restored below) */
1516 cso_save_blend(ctx
->cso
);
1517 cso_save_depth_stencil_alpha(ctx
->cso
);
1518 cso_save_rasterizer(ctx
->cso
);
1519 cso_save_samplers(ctx
->cso
);
1520 cso_save_sampler_textures(ctx
->cso
);
1521 cso_save_framebuffer(ctx
->cso
);
1522 cso_save_fragment_shader(ctx
->cso
);
1523 cso_save_vertex_shader(ctx
->cso
);
1525 /* bind our state */
1526 cso_set_blend(ctx
->cso
, &ctx
->blend
);
1527 cso_set_depth_stencil_alpha(ctx
->cso
, &ctx
->depthstencil
);
1528 cso_set_rasterizer(ctx
->cso
, &ctx
->rasterizer
);
1530 cso_set_fragment_shader_handle(ctx
->cso
, ctx
->fs
);
1531 cso_set_vertex_shader_handle(ctx
->cso
, ctx
->vs
);
1533 /* init framebuffer state */
1534 memset(&fb
, 0, sizeof(fb
));
1537 /* set min/mag to same filter for faster sw speed */
1538 ctx
->sampler
.mag_img_filter
= filter
;
1539 ctx
->sampler
.min_img_filter
= filter
;
1542 * XXX for small mipmap levels, it may be faster to use the software
1545 for (dstLevel
= baseLevel
+ 1; dstLevel
<= lastLevel
; dstLevel
++) {
1546 const uint srcLevel
= dstLevel
- 1;
1548 struct pipe_surface
*surf
=
1549 screen
->get_tex_surface(screen
, pt
, face
, dstLevel
, zslice
,
1550 PIPE_BUFFER_USAGE_GPU_WRITE
);
1553 * Setup framebuffer / dest surface
1556 fb
.width
= pt
->width
[dstLevel
];
1557 fb
.height
= pt
->height
[dstLevel
];
1558 cso_set_framebuffer(ctx
->cso
, &fb
);
1561 * Setup sampler state
1562 * Note: we should only have to set the min/max LOD clamps to ensure
1563 * we grab texels from the right mipmap level. But some hardware
1564 * has trouble with min clamping so we also set the lod_bias to
1565 * try to work around that.
1567 ctx
->sampler
.min_lod
= ctx
->sampler
.max_lod
= (float) srcLevel
;
1568 ctx
->sampler
.lod_bias
= (float) srcLevel
;
1569 cso_single_sampler(ctx
->cso
, 0, &ctx
->sampler
);
1570 cso_single_sampler_done(ctx
->cso
);
1572 cso_set_sampler_textures(ctx
->cso
, 1, &pt
);
1574 /* quad coords in window coords (bypassing vs, clip and viewport) */
1575 offset
= set_vertex_data(ctx
,
1578 (float) pt
->width
[dstLevel
],
1579 (float) pt
->height
[dstLevel
]);
1581 util_draw_vertex_buffer(ctx
->pipe
,
1584 PIPE_PRIM_TRIANGLE_FAN
,
1586 2); /* attribs/vert */
1588 pipe
->flush(pipe
, PIPE_FLUSH_RENDER_CACHE
, NULL
);
1590 /* need to signal that the texture has changed _after_ rendering to it */
1591 pipe_surface_reference( &surf
, NULL
);
1594 /* restore state we changed */
1595 cso_restore_blend(ctx
->cso
);
1596 cso_restore_depth_stencil_alpha(ctx
->cso
);
1597 cso_restore_rasterizer(ctx
->cso
);
1598 cso_restore_samplers(ctx
->cso
);
1599 cso_restore_sampler_textures(ctx
->cso
);
1600 cso_restore_framebuffer(ctx
->cso
);
1601 cso_restore_fragment_shader(ctx
->cso
);
1602 cso_restore_vertex_shader(ctx
->cso
);