xref: /third_party/mesa3d/src/mesa/main/mipmap.c (revision bf215546)
1/*
2 * Mesa 3-D graphics library
3 *
4 * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25
26/**
27 * \file mipmap.c  mipmap generation and teximage resizing functions.
28 */
29
30#include "errors.h"
31
32#include "formats.h"
33#include "glformats.h"
34#include "mipmap.h"
35#include "mtypes.h"
36#include "teximage.h"
37#include "texobj.h"
38#include "texstore.h"
39#include "image.h"
40#include "macros.h"
41#include "util/half_float.h"
42#include "util/format_rgb9e5.h"
43#include "util/format_r11g11b10f.h"
44
45#include "state_tracker/st_cb_texture.h"
46
47/**
48 * Compute the expected number of mipmap levels in the texture given
49 * the width/height/depth of the base image and the GL_TEXTURE_BASE_LEVEL/
50 * GL_TEXTURE_MAX_LEVEL settings.  This will tell us how many mipmap
51 * levels should be generated.
52 */
53unsigned
54_mesa_compute_num_levels(struct gl_context *ctx,
55                         struct gl_texture_object *texObj,
56                         GLenum target)
57{
58   const struct gl_texture_image *baseImage;
59   GLuint numLevels;
60
61   baseImage = _mesa_get_tex_image(ctx, texObj, target, texObj->Attrib.BaseLevel);
62
63   numLevels = texObj->Attrib.BaseLevel + baseImage->MaxNumLevels;
64   numLevels = MIN2(numLevels, (GLuint) texObj->Attrib.MaxLevel + 1);
65   if (texObj->Immutable)
66      numLevels = MIN2(numLevels, texObj->Attrib.NumLevels);
67   assert(numLevels >= 1);
68
69   return numLevels;
70}
71
72static GLint
73bytes_per_pixel(GLenum datatype, GLuint comps)
74{
75   GLint b;
76
77   if (datatype == GL_UNSIGNED_INT_8_24_REV_MESA ||
78       datatype == GL_UNSIGNED_INT_24_8_MESA)
79      return 4;
80
81   b = _mesa_sizeof_packed_type(datatype);
82   assert(b >= 0);
83
84   if (_mesa_type_is_packed(datatype))
85      return b;
86   else
87      return b * comps;
88}
89
90
91/**
92 * \name Support macros for do_row and do_row_3d
93 *
94 * The macro madness is here for two reasons.  First, it compacts the code
95 * slightly.  Second, it makes it much easier to adjust the specifics of the
96 * filter to tune the rounding characteristics.
97 */
98/*@{*/
99#define DECLARE_ROW_POINTERS(t, e) \
100      const t(*rowA)[e] = (const t(*)[e]) srcRowA; \
101      const t(*rowB)[e] = (const t(*)[e]) srcRowB; \
102      const t(*rowC)[e] = (const t(*)[e]) srcRowC; \
103      const t(*rowD)[e] = (const t(*)[e]) srcRowD; \
104      t(*dst)[e] = (t(*)[e]) dstRow
105
106#define DECLARE_ROW_POINTERS0(t) \
107      const t *rowA = (const t *) srcRowA; \
108      const t *rowB = (const t *) srcRowB; \
109      const t *rowC = (const t *) srcRowC; \
110      const t *rowD = (const t *) srcRowD; \
111      t *dst = (t *) dstRow
112
113#define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
114   ((unsigned) Aj + (unsigned) Ak \
115    + (unsigned) Bj + (unsigned) Bk \
116    + (unsigned) Cj + (unsigned) Ck \
117    + (unsigned) Dj + (unsigned) Dk \
118    + 4) >> 3
119
120#define FILTER_3D(e) \
121   do { \
122      dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \
123                                rowB[j][e], rowB[k][e], \
124                                rowC[j][e], rowC[k][e], \
125                                rowD[j][e], rowD[k][e]); \
126   } while(0)
127
128#define FILTER_SUM_3D_SIGNED(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
129   (Aj + Ak \
130    + Bj + Bk \
131    + Cj + Ck \
132    + Dj + Dk \
133    + 4) / 8
134
135#define FILTER_3D_SIGNED(e) \
136   do { \
137      dst[i][e] = FILTER_SUM_3D_SIGNED(rowA[j][e], rowA[k][e], \
138                                       rowB[j][e], rowB[k][e], \
139                                       rowC[j][e], rowC[k][e], \
140                                       rowD[j][e], rowD[k][e]); \
141   } while(0)
142
143#define FILTER_F_3D(e) \
144   do { \
145      dst[i][e] = (rowA[j][e] + rowA[k][e] \
146                   + rowB[j][e] + rowB[k][e] \
147                   + rowC[j][e] + rowC[k][e] \
148                   + rowD[j][e] + rowD[k][e]) * 0.125F; \
149   } while(0)
150
151#define FILTER_HF_3D(e) \
152   do { \
153      const GLfloat aj = _mesa_half_to_float(rowA[j][e]); \
154      const GLfloat ak = _mesa_half_to_float(rowA[k][e]); \
155      const GLfloat bj = _mesa_half_to_float(rowB[j][e]); \
156      const GLfloat bk = _mesa_half_to_float(rowB[k][e]); \
157      const GLfloat cj = _mesa_half_to_float(rowC[j][e]); \
158      const GLfloat ck = _mesa_half_to_float(rowC[k][e]); \
159      const GLfloat dj = _mesa_half_to_float(rowD[j][e]); \
160      const GLfloat dk = _mesa_half_to_float(rowD[k][e]); \
161      dst[i][e] = _mesa_float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
162                                      * 0.125F); \
163   } while(0)
164/*@}*/
165
166
167/**
168 * Average together two rows of a source image to produce a single new
169 * row in the dest image.  It's legal for the two source rows to point
170 * to the same data.  The source width must be equal to either the
171 * dest width or two times the dest width.
172 * \param datatype  GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
173 * \param comps  number of components per pixel (1..4)
174 */
175static void
176do_row(GLenum datatype, GLuint comps, GLint srcWidth,
177       const GLvoid *srcRowA, const GLvoid *srcRowB,
178       GLint dstWidth, GLvoid *dstRow)
179{
180   const GLuint k0 = (srcWidth == dstWidth) ? 0 : 1;
181   const GLuint colStride = (srcWidth == dstWidth) ? 1 : 2;
182
183   assert(comps >= 1);
184   assert(comps <= 4);
185
186   /* This assertion is no longer valid with non-power-of-2 textures
187   assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
188   */
189
190   if (datatype == GL_UNSIGNED_BYTE && comps == 4) {
191      GLuint i, j, k;
192      const GLubyte(*rowA)[4] = (const GLubyte(*)[4]) srcRowA;
193      const GLubyte(*rowB)[4] = (const GLubyte(*)[4]) srcRowB;
194      GLubyte(*dst)[4] = (GLubyte(*)[4]) dstRow;
195      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
196           i++, j += colStride, k += colStride) {
197         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
198         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
199         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
200         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
201      }
202   }
203   else if (datatype == GL_UNSIGNED_BYTE && comps == 3) {
204      GLuint i, j, k;
205      const GLubyte(*rowA)[3] = (const GLubyte(*)[3]) srcRowA;
206      const GLubyte(*rowB)[3] = (const GLubyte(*)[3]) srcRowB;
207      GLubyte(*dst)[3] = (GLubyte(*)[3]) dstRow;
208      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
209           i++, j += colStride, k += colStride) {
210         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
211         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
212         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
213      }
214   }
215   else if (datatype == GL_UNSIGNED_BYTE && comps == 2) {
216      GLuint i, j, k;
217      const GLubyte(*rowA)[2] = (const GLubyte(*)[2]) srcRowA;
218      const GLubyte(*rowB)[2] = (const GLubyte(*)[2]) srcRowB;
219      GLubyte(*dst)[2] = (GLubyte(*)[2]) dstRow;
220      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
221           i++, j += colStride, k += colStride) {
222         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
223         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
224      }
225   }
226   else if (datatype == GL_UNSIGNED_BYTE && comps == 1) {
227      GLuint i, j, k;
228      const GLubyte *rowA = (const GLubyte *) srcRowA;
229      const GLubyte *rowB = (const GLubyte *) srcRowB;
230      GLubyte *dst = (GLubyte *) dstRow;
231      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
232           i++, j += colStride, k += colStride) {
233         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
234      }
235   }
236
237   else if (datatype == GL_BYTE && comps == 4) {
238      GLuint i, j, k;
239      const GLbyte(*rowA)[4] = (const GLbyte(*)[4]) srcRowA;
240      const GLbyte(*rowB)[4] = (const GLbyte(*)[4]) srcRowB;
241      GLbyte(*dst)[4] = (GLbyte(*)[4]) dstRow;
242      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
243           i++, j += colStride, k += colStride) {
244         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
245         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
246         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
247         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
248      }
249   }
250   else if (datatype == GL_BYTE && comps == 3) {
251      GLuint i, j, k;
252      const GLbyte(*rowA)[3] = (const GLbyte(*)[3]) srcRowA;
253      const GLbyte(*rowB)[3] = (const GLbyte(*)[3]) srcRowB;
254      GLbyte(*dst)[3] = (GLbyte(*)[3]) dstRow;
255      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
256           i++, j += colStride, k += colStride) {
257         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
258         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
259         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
260      }
261   }
262   else if (datatype == GL_BYTE && comps == 2) {
263      GLuint i, j, k;
264      const GLbyte(*rowA)[2] = (const GLbyte(*)[2]) srcRowA;
265      const GLbyte(*rowB)[2] = (const GLbyte(*)[2]) srcRowB;
266      GLbyte(*dst)[2] = (GLbyte(*)[2]) dstRow;
267      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
268           i++, j += colStride, k += colStride) {
269         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
270         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
271      }
272   }
273   else if (datatype == GL_BYTE && comps == 1) {
274      GLuint i, j, k;
275      const GLbyte *rowA = (const GLbyte *) srcRowA;
276      const GLbyte *rowB = (const GLbyte *) srcRowB;
277      GLbyte *dst = (GLbyte *) dstRow;
278      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
279           i++, j += colStride, k += colStride) {
280         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
281      }
282   }
283
284   else if (datatype == GL_UNSIGNED_SHORT && comps == 4) {
285      GLuint i, j, k;
286      const GLushort(*rowA)[4] = (const GLushort(*)[4]) srcRowA;
287      const GLushort(*rowB)[4] = (const GLushort(*)[4]) srcRowB;
288      GLushort(*dst)[4] = (GLushort(*)[4]) dstRow;
289      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
290           i++, j += colStride, k += colStride) {
291         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
292         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
293         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
294         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
295      }
296   }
297   else if (datatype == GL_UNSIGNED_SHORT && comps == 3) {
298      GLuint i, j, k;
299      const GLushort(*rowA)[3] = (const GLushort(*)[3]) srcRowA;
300      const GLushort(*rowB)[3] = (const GLushort(*)[3]) srcRowB;
301      GLushort(*dst)[3] = (GLushort(*)[3]) dstRow;
302      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
303           i++, j += colStride, k += colStride) {
304         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
305         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
306         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
307      }
308   }
309   else if (datatype == GL_UNSIGNED_SHORT && comps == 2) {
310      GLuint i, j, k;
311      const GLushort(*rowA)[2] = (const GLushort(*)[2]) srcRowA;
312      const GLushort(*rowB)[2] = (const GLushort(*)[2]) srcRowB;
313      GLushort(*dst)[2] = (GLushort(*)[2]) dstRow;
314      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
315           i++, j += colStride, k += colStride) {
316         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
317         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
318      }
319   }
320   else if (datatype == GL_UNSIGNED_SHORT && comps == 1) {
321      GLuint i, j, k;
322      const GLushort *rowA = (const GLushort *) srcRowA;
323      const GLushort *rowB = (const GLushort *) srcRowB;
324      GLushort *dst = (GLushort *) dstRow;
325      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
326           i++, j += colStride, k += colStride) {
327         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
328      }
329   }
330
331   else if (datatype == GL_SHORT && comps == 4) {
332      GLuint i, j, k;
333      const GLshort(*rowA)[4] = (const GLshort(*)[4]) srcRowA;
334      const GLshort(*rowB)[4] = (const GLshort(*)[4]) srcRowB;
335      GLshort(*dst)[4] = (GLshort(*)[4]) dstRow;
336      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
337           i++, j += colStride, k += colStride) {
338         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
339         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
340         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
341         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
342      }
343   }
344   else if (datatype == GL_SHORT && comps == 3) {
345      GLuint i, j, k;
346      const GLshort(*rowA)[3] = (const GLshort(*)[3]) srcRowA;
347      const GLshort(*rowB)[3] = (const GLshort(*)[3]) srcRowB;
348      GLshort(*dst)[3] = (GLshort(*)[3]) dstRow;
349      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
350           i++, j += colStride, k += colStride) {
351         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
352         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
353         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
354      }
355   }
356   else if (datatype == GL_SHORT && comps == 2) {
357      GLuint i, j, k;
358      const GLshort(*rowA)[2] = (const GLshort(*)[2]) srcRowA;
359      const GLshort(*rowB)[2] = (const GLshort(*)[2]) srcRowB;
360      GLshort(*dst)[2] = (GLshort(*)[2]) dstRow;
361      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
362           i++, j += colStride, k += colStride) {
363         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
364         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
365      }
366   }
367   else if (datatype == GL_SHORT && comps == 1) {
368      GLuint i, j, k;
369      const GLshort *rowA = (const GLshort *) srcRowA;
370      const GLshort *rowB = (const GLshort *) srcRowB;
371      GLshort *dst = (GLshort *) dstRow;
372      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
373           i++, j += colStride, k += colStride) {
374         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
375      }
376   }
377
378   else if (datatype == GL_FLOAT && comps == 4) {
379      GLuint i, j, k;
380      const GLfloat(*rowA)[4] = (const GLfloat(*)[4]) srcRowA;
381      const GLfloat(*rowB)[4] = (const GLfloat(*)[4]) srcRowB;
382      GLfloat(*dst)[4] = (GLfloat(*)[4]) dstRow;
383      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
384           i++, j += colStride, k += colStride) {
385         dst[i][0] = (rowA[j][0] + rowA[k][0] +
386                      rowB[j][0] + rowB[k][0]) * 0.25F;
387         dst[i][1] = (rowA[j][1] + rowA[k][1] +
388                      rowB[j][1] + rowB[k][1]) * 0.25F;
389         dst[i][2] = (rowA[j][2] + rowA[k][2] +
390                      rowB[j][2] + rowB[k][2]) * 0.25F;
391         dst[i][3] = (rowA[j][3] + rowA[k][3] +
392                      rowB[j][3] + rowB[k][3]) * 0.25F;
393      }
394   }
395   else if (datatype == GL_FLOAT && comps == 3) {
396      GLuint i, j, k;
397      const GLfloat(*rowA)[3] = (const GLfloat(*)[3]) srcRowA;
398      const GLfloat(*rowB)[3] = (const GLfloat(*)[3]) srcRowB;
399      GLfloat(*dst)[3] = (GLfloat(*)[3]) dstRow;
400      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
401           i++, j += colStride, k += colStride) {
402         dst[i][0] = (rowA[j][0] + rowA[k][0] +
403                      rowB[j][0] + rowB[k][0]) * 0.25F;
404         dst[i][1] = (rowA[j][1] + rowA[k][1] +
405                      rowB[j][1] + rowB[k][1]) * 0.25F;
406         dst[i][2] = (rowA[j][2] + rowA[k][2] +
407                      rowB[j][2] + rowB[k][2]) * 0.25F;
408      }
409   }
410   else if (datatype == GL_FLOAT && comps == 2) {
411      GLuint i, j, k;
412      const GLfloat(*rowA)[2] = (const GLfloat(*)[2]) srcRowA;
413      const GLfloat(*rowB)[2] = (const GLfloat(*)[2]) srcRowB;
414      GLfloat(*dst)[2] = (GLfloat(*)[2]) dstRow;
415      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
416           i++, j += colStride, k += colStride) {
417         dst[i][0] = (rowA[j][0] + rowA[k][0] +
418                      rowB[j][0] + rowB[k][0]) * 0.25F;
419         dst[i][1] = (rowA[j][1] + rowA[k][1] +
420                      rowB[j][1] + rowB[k][1]) * 0.25F;
421      }
422   }
423   else if (datatype == GL_FLOAT && comps == 1) {
424      GLuint i, j, k;
425      const GLfloat *rowA = (const GLfloat *) srcRowA;
426      const GLfloat *rowB = (const GLfloat *) srcRowB;
427      GLfloat *dst = (GLfloat *) dstRow;
428      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
429           i++, j += colStride, k += colStride) {
430         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
431      }
432   }
433
434   else if (datatype == GL_HALF_FLOAT_ARB && comps == 4) {
435      GLuint i, j, k, comp;
436      const GLhalfARB(*rowA)[4] = (const GLhalfARB(*)[4]) srcRowA;
437      const GLhalfARB(*rowB)[4] = (const GLhalfARB(*)[4]) srcRowB;
438      GLhalfARB(*dst)[4] = (GLhalfARB(*)[4]) dstRow;
439      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
440           i++, j += colStride, k += colStride) {
441         for (comp = 0; comp < 4; comp++) {
442            GLfloat aj, ak, bj, bk;
443            aj = _mesa_half_to_float(rowA[j][comp]);
444            ak = _mesa_half_to_float(rowA[k][comp]);
445            bj = _mesa_half_to_float(rowB[j][comp]);
446            bk = _mesa_half_to_float(rowB[k][comp]);
447            dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
448         }
449      }
450   }
451   else if (datatype == GL_HALF_FLOAT_ARB && comps == 3) {
452      GLuint i, j, k, comp;
453      const GLhalfARB(*rowA)[3] = (const GLhalfARB(*)[3]) srcRowA;
454      const GLhalfARB(*rowB)[3] = (const GLhalfARB(*)[3]) srcRowB;
455      GLhalfARB(*dst)[3] = (GLhalfARB(*)[3]) dstRow;
456      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
457           i++, j += colStride, k += colStride) {
458         for (comp = 0; comp < 3; comp++) {
459            GLfloat aj, ak, bj, bk;
460            aj = _mesa_half_to_float(rowA[j][comp]);
461            ak = _mesa_half_to_float(rowA[k][comp]);
462            bj = _mesa_half_to_float(rowB[j][comp]);
463            bk = _mesa_half_to_float(rowB[k][comp]);
464            dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
465         }
466      }
467   }
468   else if (datatype == GL_HALF_FLOAT_ARB && comps == 2) {
469      GLuint i, j, k, comp;
470      const GLhalfARB(*rowA)[2] = (const GLhalfARB(*)[2]) srcRowA;
471      const GLhalfARB(*rowB)[2] = (const GLhalfARB(*)[2]) srcRowB;
472      GLhalfARB(*dst)[2] = (GLhalfARB(*)[2]) dstRow;
473      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
474           i++, j += colStride, k += colStride) {
475         for (comp = 0; comp < 2; comp++) {
476            GLfloat aj, ak, bj, bk;
477            aj = _mesa_half_to_float(rowA[j][comp]);
478            ak = _mesa_half_to_float(rowA[k][comp]);
479            bj = _mesa_half_to_float(rowB[j][comp]);
480            bk = _mesa_half_to_float(rowB[k][comp]);
481            dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
482         }
483      }
484   }
485   else if (datatype == GL_HALF_FLOAT_ARB && comps == 1) {
486      GLuint i, j, k;
487      const GLhalfARB *rowA = (const GLhalfARB *) srcRowA;
488      const GLhalfARB *rowB = (const GLhalfARB *) srcRowB;
489      GLhalfARB *dst = (GLhalfARB *) dstRow;
490      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
491           i++, j += colStride, k += colStride) {
492         GLfloat aj, ak, bj, bk;
493         aj = _mesa_half_to_float(rowA[j]);
494         ak = _mesa_half_to_float(rowA[k]);
495         bj = _mesa_half_to_float(rowB[j]);
496         bk = _mesa_half_to_float(rowB[k]);
497         dst[i] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
498      }
499   }
500
501   else if (datatype == GL_UNSIGNED_INT && comps == 1) {
502      GLuint i, j, k;
503      const GLuint *rowA = (const GLuint *) srcRowA;
504      const GLuint *rowB = (const GLuint *) srcRowB;
505      GLuint *dst = (GLuint *) dstRow;
506      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
507           i++, j += colStride, k += colStride) {
508         dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
509      }
510   }
511
512   else if (datatype == GL_UNSIGNED_SHORT_5_6_5 && comps == 3) {
513      GLuint i, j, k;
514      const GLushort *rowA = (const GLushort *) srcRowA;
515      const GLushort *rowB = (const GLushort *) srcRowB;
516      GLushort *dst = (GLushort *) dstRow;
517      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
518           i++, j += colStride, k += colStride) {
519         const GLint rowAr0 = rowA[j] & 0x1f;
520         const GLint rowAr1 = rowA[k] & 0x1f;
521         const GLint rowBr0 = rowB[j] & 0x1f;
522         const GLint rowBr1 = rowB[k] & 0x1f;
523         const GLint rowAg0 = (rowA[j] >> 5) & 0x3f;
524         const GLint rowAg1 = (rowA[k] >> 5) & 0x3f;
525         const GLint rowBg0 = (rowB[j] >> 5) & 0x3f;
526         const GLint rowBg1 = (rowB[k] >> 5) & 0x3f;
527         const GLint rowAb0 = (rowA[j] >> 11) & 0x1f;
528         const GLint rowAb1 = (rowA[k] >> 11) & 0x1f;
529         const GLint rowBb0 = (rowB[j] >> 11) & 0x1f;
530         const GLint rowBb1 = (rowB[k] >> 11) & 0x1f;
531         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
532         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
533         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
534         dst[i] = (blue << 11) | (green << 5) | red;
535      }
536   }
537   else if (datatype == GL_UNSIGNED_SHORT_4_4_4_4 && comps == 4) {
538      GLuint i, j, k;
539      const GLushort *rowA = (const GLushort *) srcRowA;
540      const GLushort *rowB = (const GLushort *) srcRowB;
541      GLushort *dst = (GLushort *) dstRow;
542      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
543           i++, j += colStride, k += colStride) {
544         const GLint rowAr0 = rowA[j] & 0xf;
545         const GLint rowAr1 = rowA[k] & 0xf;
546         const GLint rowBr0 = rowB[j] & 0xf;
547         const GLint rowBr1 = rowB[k] & 0xf;
548         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
549         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
550         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
551         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
552         const GLint rowAb0 = (rowA[j] >> 8) & 0xf;
553         const GLint rowAb1 = (rowA[k] >> 8) & 0xf;
554         const GLint rowBb0 = (rowB[j] >> 8) & 0xf;
555         const GLint rowBb1 = (rowB[k] >> 8) & 0xf;
556         const GLint rowAa0 = (rowA[j] >> 12) & 0xf;
557         const GLint rowAa1 = (rowA[k] >> 12) & 0xf;
558         const GLint rowBa0 = (rowB[j] >> 12) & 0xf;
559         const GLint rowBa1 = (rowB[k] >> 12) & 0xf;
560         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
561         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
562         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
563         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
564         dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
565      }
566   }
567   else if (datatype == GL_UNSIGNED_SHORT_1_5_5_5_REV && comps == 4) {
568      GLuint i, j, k;
569      const GLushort *rowA = (const GLushort *) srcRowA;
570      const GLushort *rowB = (const GLushort *) srcRowB;
571      GLushort *dst = (GLushort *) dstRow;
572      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
573           i++, j += colStride, k += colStride) {
574         const GLint rowAr0 = rowA[j] & 0x1f;
575         const GLint rowAr1 = rowA[k] & 0x1f;
576         const GLint rowBr0 = rowB[j] & 0x1f;
577         const GLint rowBr1 = rowB[k] & 0x1f;
578         const GLint rowAg0 = (rowA[j] >> 5) & 0x1f;
579         const GLint rowAg1 = (rowA[k] >> 5) & 0x1f;
580         const GLint rowBg0 = (rowB[j] >> 5) & 0x1f;
581         const GLint rowBg1 = (rowB[k] >> 5) & 0x1f;
582         const GLint rowAb0 = (rowA[j] >> 10) & 0x1f;
583         const GLint rowAb1 = (rowA[k] >> 10) & 0x1f;
584         const GLint rowBb0 = (rowB[j] >> 10) & 0x1f;
585         const GLint rowBb1 = (rowB[k] >> 10) & 0x1f;
586         const GLint rowAa0 = (rowA[j] >> 15) & 0x1;
587         const GLint rowAa1 = (rowA[k] >> 15) & 0x1;
588         const GLint rowBa0 = (rowB[j] >> 15) & 0x1;
589         const GLint rowBa1 = (rowB[k] >> 15) & 0x1;
590         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
591         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
592         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
593         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
594         dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
595      }
596   }
597   else if (datatype == GL_UNSIGNED_SHORT_5_5_5_1 && comps == 4) {
598      GLuint i, j, k;
599      const GLushort *rowA = (const GLushort *) srcRowA;
600      const GLushort *rowB = (const GLushort *) srcRowB;
601      GLushort *dst = (GLushort *) dstRow;
602      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
603           i++, j += colStride, k += colStride) {
604         const GLint rowAr0 = (rowA[j] >> 11) & 0x1f;
605         const GLint rowAr1 = (rowA[k] >> 11) & 0x1f;
606         const GLint rowBr0 = (rowB[j] >> 11) & 0x1f;
607         const GLint rowBr1 = (rowB[k] >> 11) & 0x1f;
608         const GLint rowAg0 = (rowA[j] >> 6) & 0x1f;
609         const GLint rowAg1 = (rowA[k] >> 6) & 0x1f;
610         const GLint rowBg0 = (rowB[j] >> 6) & 0x1f;
611         const GLint rowBg1 = (rowB[k] >> 6) & 0x1f;
612         const GLint rowAb0 = (rowA[j] >> 1) & 0x1f;
613         const GLint rowAb1 = (rowA[k] >> 1) & 0x1f;
614         const GLint rowBb0 = (rowB[j] >> 1) & 0x1f;
615         const GLint rowBb1 = (rowB[k] >> 1) & 0x1f;
616         const GLint rowAa0 = (rowA[j] & 0x1);
617         const GLint rowAa1 = (rowA[k] & 0x1);
618         const GLint rowBa0 = (rowB[j] & 0x1);
619         const GLint rowBa1 = (rowB[k] & 0x1);
620         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
621         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
622         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
623         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
624         dst[i] = (red << 11) | (green << 6) | (blue << 1) | alpha;
625      }
626   }
627
628   else if (datatype == GL_UNSIGNED_BYTE_3_3_2 && comps == 3) {
629      GLuint i, j, k;
630      const GLubyte *rowA = (const GLubyte *) srcRowA;
631      const GLubyte *rowB = (const GLubyte *) srcRowB;
632      GLubyte *dst = (GLubyte *) dstRow;
633      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
634           i++, j += colStride, k += colStride) {
635         const GLint rowAr0 = rowA[j] & 0x3;
636         const GLint rowAr1 = rowA[k] & 0x3;
637         const GLint rowBr0 = rowB[j] & 0x3;
638         const GLint rowBr1 = rowB[k] & 0x3;
639         const GLint rowAg0 = (rowA[j] >> 2) & 0x7;
640         const GLint rowAg1 = (rowA[k] >> 2) & 0x7;
641         const GLint rowBg0 = (rowB[j] >> 2) & 0x7;
642         const GLint rowBg1 = (rowB[k] >> 2) & 0x7;
643         const GLint rowAb0 = (rowA[j] >> 5) & 0x7;
644         const GLint rowAb1 = (rowA[k] >> 5) & 0x7;
645         const GLint rowBb0 = (rowB[j] >> 5) & 0x7;
646         const GLint rowBb1 = (rowB[k] >> 5) & 0x7;
647         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
648         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
649         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
650         dst[i] = (blue << 5) | (green << 2) | red;
651      }
652   }
653
654   else if (datatype == MESA_UNSIGNED_BYTE_4_4 && comps == 2) {
655      GLuint i, j, k;
656      const GLubyte *rowA = (const GLubyte *) srcRowA;
657      const GLubyte *rowB = (const GLubyte *) srcRowB;
658      GLubyte *dst = (GLubyte *) dstRow;
659      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
660           i++, j += colStride, k += colStride) {
661         const GLint rowAr0 = rowA[j] & 0xf;
662         const GLint rowAr1 = rowA[k] & 0xf;
663         const GLint rowBr0 = rowB[j] & 0xf;
664         const GLint rowBr1 = rowB[k] & 0xf;
665         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
666         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
667         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
668         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
669         const GLint r = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
670         const GLint g = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
671         dst[i] = (g << 4) | r;
672      }
673   }
674
675   else if (datatype == GL_UNSIGNED_INT_2_10_10_10_REV && comps == 4) {
676      GLuint i, j, k;
677      const GLuint *rowA = (const GLuint *) srcRowA;
678      const GLuint *rowB = (const GLuint *) srcRowB;
679      GLuint *dst = (GLuint *) dstRow;
680      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
681           i++, j += colStride, k += colStride) {
682         const GLint rowAr0 = rowA[j] & 0x3ff;
683         const GLint rowAr1 = rowA[k] & 0x3ff;
684         const GLint rowBr0 = rowB[j] & 0x3ff;
685         const GLint rowBr1 = rowB[k] & 0x3ff;
686         const GLint rowAg0 = (rowA[j] >> 10) & 0x3ff;
687         const GLint rowAg1 = (rowA[k] >> 10) & 0x3ff;
688         const GLint rowBg0 = (rowB[j] >> 10) & 0x3ff;
689         const GLint rowBg1 = (rowB[k] >> 10) & 0x3ff;
690         const GLint rowAb0 = (rowA[j] >> 20) & 0x3ff;
691         const GLint rowAb1 = (rowA[k] >> 20) & 0x3ff;
692         const GLint rowBb0 = (rowB[j] >> 20) & 0x3ff;
693         const GLint rowBb1 = (rowB[k] >> 20) & 0x3ff;
694         const GLint rowAa0 = (rowA[j] >> 30) & 0x3;
695         const GLint rowAa1 = (rowA[k] >> 30) & 0x3;
696         const GLint rowBa0 = (rowB[j] >> 30) & 0x3;
697         const GLint rowBa1 = (rowB[k] >> 30) & 0x3;
698         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
699         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
700         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
701         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
702         dst[i] = (alpha << 30) | (blue << 20) | (green << 10) | red;
703      }
704   }
705
706   else if (datatype == GL_UNSIGNED_INT_5_9_9_9_REV && comps == 3) {
707      GLuint i, j, k;
708      const GLuint *rowA = (const GLuint*) srcRowA;
709      const GLuint *rowB = (const GLuint*) srcRowB;
710      GLuint *dst = (GLuint*)dstRow;
711      GLfloat res[3], rowAj[3], rowBj[3], rowAk[3], rowBk[3];
712      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
713           i++, j += colStride, k += colStride) {
714         rgb9e5_to_float3(rowA[j], rowAj);
715         rgb9e5_to_float3(rowB[j], rowBj);
716         rgb9e5_to_float3(rowA[k], rowAk);
717         rgb9e5_to_float3(rowB[k], rowBk);
718         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0]) * 0.25F;
719         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1]) * 0.25F;
720         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2]) * 0.25F;
721         dst[i] = float3_to_rgb9e5(res);
722      }
723   }
724
725   else if (datatype == GL_UNSIGNED_INT_10F_11F_11F_REV && comps == 3) {
726      GLuint i, j, k;
727      const GLuint *rowA = (const GLuint*) srcRowA;
728      const GLuint *rowB = (const GLuint*) srcRowB;
729      GLuint *dst = (GLuint*)dstRow;
730      GLfloat res[3], rowAj[3], rowBj[3], rowAk[3], rowBk[3];
731      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
732           i++, j += colStride, k += colStride) {
733         r11g11b10f_to_float3(rowA[j], rowAj);
734         r11g11b10f_to_float3(rowB[j], rowBj);
735         r11g11b10f_to_float3(rowA[k], rowAk);
736         r11g11b10f_to_float3(rowB[k], rowBk);
737         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0]) * 0.25F;
738         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1]) * 0.25F;
739         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2]) * 0.25F;
740         dst[i] = float3_to_r11g11b10f(res);
741      }
742   }
743
744   else if (datatype == GL_FLOAT_32_UNSIGNED_INT_24_8_REV && comps == 1) {
745      GLuint i, j, k;
746      const GLfloat *rowA = (const GLfloat *) srcRowA;
747      const GLfloat *rowB = (const GLfloat *) srcRowB;
748      GLfloat *dst = (GLfloat *) dstRow;
749      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
750           i++, j += colStride, k += colStride) {
751         dst[i*2] = (rowA[j*2] + rowA[k*2] + rowB[j*2] + rowB[k*2]) * 0.25F;
752      }
753   }
754
755   else if (datatype == GL_UNSIGNED_INT_24_8_MESA && comps == 2) {
756      GLuint i, j, k;
757      const GLuint *rowA = (const GLuint *) srcRowA;
758      const GLuint *rowB = (const GLuint *) srcRowB;
759      GLuint *dst = (GLuint *) dstRow;
760      /* note: averaging stencil values seems weird, but what else? */
761      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
762           i++, j += colStride, k += colStride) {
763         GLuint z = (((rowA[j] >> 8) + (rowA[k] >> 8) +
764                      (rowB[j] >> 8) + (rowB[k] >> 8)) / 4) << 8;
765         GLuint s = ((rowA[j] & 0xff) + (rowA[k] & 0xff) +
766                     (rowB[j] & 0xff) + (rowB[k] & 0xff)) / 4;
767         dst[i] = z | s;
768      }
769   }
770   else if (datatype == GL_UNSIGNED_INT_8_24_REV_MESA && comps == 2) {
771      GLuint i, j, k;
772      const GLuint *rowA = (const GLuint *) srcRowA;
773      const GLuint *rowB = (const GLuint *) srcRowB;
774      GLuint *dst = (GLuint *) dstRow;
775      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
776           i++, j += colStride, k += colStride) {
777         GLuint z = ((rowA[j] & 0xffffff) + (rowA[k] & 0xffffff) +
778                     (rowB[j] & 0xffffff) + (rowB[k] & 0xffffff)) / 4;
779         GLuint s = (((rowA[j] >> 24) + (rowA[k] >> 24) +
780                      (rowB[j] >> 24) + (rowB[k] >> 24)) / 4) << 24;
781         dst[i] = z | s;
782      }
783   }
784
785   else {
786      unreachable("bad format in do_row()");
787   }
788}
789
790
791/**
792 * Average together four rows of a source image to produce a single new
793 * row in the dest image.  It's legal for the two source rows to point
794 * to the same data.  The source width must be equal to either the
795 * dest width or two times the dest width.
796 *
797 * \param datatype  GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
798 *                  \c GL_FLOAT, etc.
799 * \param comps     number of components per pixel (1..4)
800 * \param srcWidth  Width of a row in the source data
801 * \param srcRowA   Pointer to one of the rows of source data
802 * \param srcRowB   Pointer to one of the rows of source data
803 * \param srcRowC   Pointer to one of the rows of source data
804 * \param srcRowD   Pointer to one of the rows of source data
805 * \param dstWidth  Width of a row in the destination data
806 * \param srcRowA   Pointer to the row of destination data
807 */
808static void
809do_row_3D(GLenum datatype, GLuint comps, GLint srcWidth,
810          const GLvoid *srcRowA, const GLvoid *srcRowB,
811          const GLvoid *srcRowC, const GLvoid *srcRowD,
812          GLint dstWidth, GLvoid *dstRow)
813{
814   const GLuint k0 = (srcWidth == dstWidth) ? 0 : 1;
815   const GLuint colStride = (srcWidth == dstWidth) ? 1 : 2;
816   GLuint i, j, k;
817
818   assert(comps >= 1);
819   assert(comps <= 4);
820
821   if ((datatype == GL_UNSIGNED_BYTE) && (comps == 4)) {
822      DECLARE_ROW_POINTERS(GLubyte, 4);
823
824      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
825           i++, j += colStride, k += colStride) {
826         FILTER_3D(0);
827         FILTER_3D(1);
828         FILTER_3D(2);
829         FILTER_3D(3);
830      }
831   }
832   else if ((datatype == GL_UNSIGNED_BYTE) && (comps == 3)) {
833      DECLARE_ROW_POINTERS(GLubyte, 3);
834
835      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
836           i++, j += colStride, k += colStride) {
837         FILTER_3D(0);
838         FILTER_3D(1);
839         FILTER_3D(2);
840      }
841   }
842   else if ((datatype == GL_UNSIGNED_BYTE) && (comps == 2)) {
843      DECLARE_ROW_POINTERS(GLubyte, 2);
844
845      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
846           i++, j += colStride, k += colStride) {
847         FILTER_3D(0);
848         FILTER_3D(1);
849      }
850   }
851   else if ((datatype == GL_UNSIGNED_BYTE) && (comps == 1)) {
852      DECLARE_ROW_POINTERS(GLubyte, 1);
853
854      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
855           i++, j += colStride, k += colStride) {
856         FILTER_3D(0);
857      }
858   }
859   else if ((datatype == GL_BYTE) && (comps == 4)) {
860      DECLARE_ROW_POINTERS(GLbyte, 4);
861
862      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
863           i++, j += colStride, k += colStride) {
864         FILTER_3D_SIGNED(0);
865         FILTER_3D_SIGNED(1);
866         FILTER_3D_SIGNED(2);
867         FILTER_3D_SIGNED(3);
868      }
869   }
870   else if ((datatype == GL_BYTE) && (comps == 3)) {
871      DECLARE_ROW_POINTERS(GLbyte, 3);
872
873      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
874           i++, j += colStride, k += colStride) {
875         FILTER_3D_SIGNED(0);
876         FILTER_3D_SIGNED(1);
877         FILTER_3D_SIGNED(2);
878      }
879   }
880   else if ((datatype == GL_BYTE) && (comps == 2)) {
881      DECLARE_ROW_POINTERS(GLbyte, 2);
882
883      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
884           i++, j += colStride, k += colStride) {
885         FILTER_3D_SIGNED(0);
886         FILTER_3D_SIGNED(1);
887       }
888   }
889   else if ((datatype == GL_BYTE) && (comps == 1)) {
890      DECLARE_ROW_POINTERS(GLbyte, 1);
891
892      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
893           i++, j += colStride, k += colStride) {
894         FILTER_3D_SIGNED(0);
895      }
896   }
897   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 4)) {
898      DECLARE_ROW_POINTERS(GLushort, 4);
899
900      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
901           i++, j += colStride, k += colStride) {
902         FILTER_3D(0);
903         FILTER_3D(1);
904         FILTER_3D(2);
905         FILTER_3D(3);
906      }
907   }
908   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 3)) {
909      DECLARE_ROW_POINTERS(GLushort, 3);
910
911      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
912           i++, j += colStride, k += colStride) {
913         FILTER_3D(0);
914         FILTER_3D(1);
915         FILTER_3D(2);
916      }
917   }
918   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 2)) {
919      DECLARE_ROW_POINTERS(GLushort, 2);
920
921      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
922           i++, j += colStride, k += colStride) {
923         FILTER_3D(0);
924         FILTER_3D(1);
925      }
926   }
927   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 1)) {
928      DECLARE_ROW_POINTERS(GLushort, 1);
929
930      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
931           i++, j += colStride, k += colStride) {
932         FILTER_3D(0);
933      }
934   }
935   else if ((datatype == GL_SHORT) && (comps == 4)) {
936      DECLARE_ROW_POINTERS(GLshort, 4);
937
938      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
939           i++, j += colStride, k += colStride) {
940         FILTER_3D(0);
941         FILTER_3D(1);
942         FILTER_3D(2);
943         FILTER_3D(3);
944      }
945   }
946   else if ((datatype == GL_SHORT) && (comps == 3)) {
947      DECLARE_ROW_POINTERS(GLshort, 3);
948
949      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
950           i++, j += colStride, k += colStride) {
951         FILTER_3D(0);
952         FILTER_3D(1);
953         FILTER_3D(2);
954      }
955   }
956   else if ((datatype == GL_SHORT) && (comps == 2)) {
957      DECLARE_ROW_POINTERS(GLshort, 2);
958
959      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
960           i++, j += colStride, k += colStride) {
961         FILTER_3D(0);
962         FILTER_3D(1);
963      }
964   }
965   else if ((datatype == GL_SHORT) && (comps == 1)) {
966      DECLARE_ROW_POINTERS(GLshort, 1);
967
968      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
969           i++, j += colStride, k += colStride) {
970         FILTER_3D(0);
971      }
972   }
973   else if ((datatype == GL_FLOAT) && (comps == 4)) {
974      DECLARE_ROW_POINTERS(GLfloat, 4);
975
976      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
977           i++, j += colStride, k += colStride) {
978         FILTER_F_3D(0);
979         FILTER_F_3D(1);
980         FILTER_F_3D(2);
981         FILTER_F_3D(3);
982      }
983   }
984   else if ((datatype == GL_FLOAT) && (comps == 3)) {
985      DECLARE_ROW_POINTERS(GLfloat, 3);
986
987      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
988           i++, j += colStride, k += colStride) {
989         FILTER_F_3D(0);
990         FILTER_F_3D(1);
991         FILTER_F_3D(2);
992      }
993   }
994   else if ((datatype == GL_FLOAT) && (comps == 2)) {
995      DECLARE_ROW_POINTERS(GLfloat, 2);
996
997      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
998           i++, j += colStride, k += colStride) {
999         FILTER_F_3D(0);
1000         FILTER_F_3D(1);
1001      }
1002   }
1003   else if ((datatype == GL_FLOAT) && (comps == 1)) {
1004      DECLARE_ROW_POINTERS(GLfloat, 1);
1005
1006      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1007           i++, j += colStride, k += colStride) {
1008         FILTER_F_3D(0);
1009      }
1010   }
1011   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 4)) {
1012      DECLARE_ROW_POINTERS(GLhalfARB, 4);
1013
1014      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1015           i++, j += colStride, k += colStride) {
1016         FILTER_HF_3D(0);
1017         FILTER_HF_3D(1);
1018         FILTER_HF_3D(2);
1019         FILTER_HF_3D(3);
1020      }
1021   }
1022   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 3)) {
1023      DECLARE_ROW_POINTERS(GLhalfARB, 3);
1024
1025      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1026           i++, j += colStride, k += colStride) {
1027         FILTER_HF_3D(0);
1028         FILTER_HF_3D(1);
1029         FILTER_HF_3D(2);
1030      }
1031   }
1032   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 2)) {
1033      DECLARE_ROW_POINTERS(GLhalfARB, 2);
1034
1035      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1036           i++, j += colStride, k += colStride) {
1037         FILTER_HF_3D(0);
1038         FILTER_HF_3D(1);
1039      }
1040   }
1041   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 1)) {
1042      DECLARE_ROW_POINTERS(GLhalfARB, 1);
1043
1044      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1045           i++, j += colStride, k += colStride) {
1046         FILTER_HF_3D(0);
1047      }
1048   }
1049   else if ((datatype == GL_UNSIGNED_INT) && (comps == 1)) {
1050      const GLuint *rowA = (const GLuint *) srcRowA;
1051      const GLuint *rowB = (const GLuint *) srcRowB;
1052      const GLuint *rowC = (const GLuint *) srcRowC;
1053      const GLuint *rowD = (const GLuint *) srcRowD;
1054      GLfloat *dst = (GLfloat *) dstRow;
1055
1056      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1057           i++, j += colStride, k += colStride) {
1058         const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k])
1059                               + ((uint64_t) rowB[j] + (uint64_t) rowB[k])
1060                               + ((uint64_t) rowC[j] + (uint64_t) rowC[k])
1061                               + ((uint64_t) rowD[j] + (uint64_t) rowD[k]));
1062         dst[i] = (GLfloat)((double) tmp * 0.125);
1063      }
1064   }
1065   else if ((datatype == GL_UNSIGNED_SHORT_5_6_5) && (comps == 3)) {
1066      DECLARE_ROW_POINTERS0(GLushort);
1067
1068      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1069           i++, j += colStride, k += colStride) {
1070         const GLint rowAr0 = rowA[j] & 0x1f;
1071         const GLint rowAr1 = rowA[k] & 0x1f;
1072         const GLint rowBr0 = rowB[j] & 0x1f;
1073         const GLint rowBr1 = rowB[k] & 0x1f;
1074         const GLint rowCr0 = rowC[j] & 0x1f;
1075         const GLint rowCr1 = rowC[k] & 0x1f;
1076         const GLint rowDr0 = rowD[j] & 0x1f;
1077         const GLint rowDr1 = rowD[k] & 0x1f;
1078         const GLint rowAg0 = (rowA[j] >> 5) & 0x3f;
1079         const GLint rowAg1 = (rowA[k] >> 5) & 0x3f;
1080         const GLint rowBg0 = (rowB[j] >> 5) & 0x3f;
1081         const GLint rowBg1 = (rowB[k] >> 5) & 0x3f;
1082         const GLint rowCg0 = (rowC[j] >> 5) & 0x3f;
1083         const GLint rowCg1 = (rowC[k] >> 5) & 0x3f;
1084         const GLint rowDg0 = (rowD[j] >> 5) & 0x3f;
1085         const GLint rowDg1 = (rowD[k] >> 5) & 0x3f;
1086         const GLint rowAb0 = (rowA[j] >> 11) & 0x1f;
1087         const GLint rowAb1 = (rowA[k] >> 11) & 0x1f;
1088         const GLint rowBb0 = (rowB[j] >> 11) & 0x1f;
1089         const GLint rowBb1 = (rowB[k] >> 11) & 0x1f;
1090         const GLint rowCb0 = (rowC[j] >> 11) & 0x1f;
1091         const GLint rowCb1 = (rowC[k] >> 11) & 0x1f;
1092         const GLint rowDb0 = (rowD[j] >> 11) & 0x1f;
1093         const GLint rowDb1 = (rowD[k] >> 11) & 0x1f;
1094         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1095                                       rowCr0, rowCr1, rowDr0, rowDr1);
1096         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1097                                       rowCg0, rowCg1, rowDg0, rowDg1);
1098         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1099                                       rowCb0, rowCb1, rowDb0, rowDb1);
1100         dst[i] = (b << 11) | (g << 5) | r;
1101      }
1102   }
1103   else if ((datatype == GL_UNSIGNED_SHORT_4_4_4_4) && (comps == 4)) {
1104      DECLARE_ROW_POINTERS0(GLushort);
1105
1106      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1107           i++, j += colStride, k += colStride) {
1108         const GLint rowAr0 = rowA[j] & 0xf;
1109         const GLint rowAr1 = rowA[k] & 0xf;
1110         const GLint rowBr0 = rowB[j] & 0xf;
1111         const GLint rowBr1 = rowB[k] & 0xf;
1112         const GLint rowCr0 = rowC[j] & 0xf;
1113         const GLint rowCr1 = rowC[k] & 0xf;
1114         const GLint rowDr0 = rowD[j] & 0xf;
1115         const GLint rowDr1 = rowD[k] & 0xf;
1116         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
1117         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
1118         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
1119         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
1120         const GLint rowCg0 = (rowC[j] >> 4) & 0xf;
1121         const GLint rowCg1 = (rowC[k] >> 4) & 0xf;
1122         const GLint rowDg0 = (rowD[j] >> 4) & 0xf;
1123         const GLint rowDg1 = (rowD[k] >> 4) & 0xf;
1124         const GLint rowAb0 = (rowA[j] >> 8) & 0xf;
1125         const GLint rowAb1 = (rowA[k] >> 8) & 0xf;
1126         const GLint rowBb0 = (rowB[j] >> 8) & 0xf;
1127         const GLint rowBb1 = (rowB[k] >> 8) & 0xf;
1128         const GLint rowCb0 = (rowC[j] >> 8) & 0xf;
1129         const GLint rowCb1 = (rowC[k] >> 8) & 0xf;
1130         const GLint rowDb0 = (rowD[j] >> 8) & 0xf;
1131         const GLint rowDb1 = (rowD[k] >> 8) & 0xf;
1132         const GLint rowAa0 = (rowA[j] >> 12) & 0xf;
1133         const GLint rowAa1 = (rowA[k] >> 12) & 0xf;
1134         const GLint rowBa0 = (rowB[j] >> 12) & 0xf;
1135         const GLint rowBa1 = (rowB[k] >> 12) & 0xf;
1136         const GLint rowCa0 = (rowC[j] >> 12) & 0xf;
1137         const GLint rowCa1 = (rowC[k] >> 12) & 0xf;
1138         const GLint rowDa0 = (rowD[j] >> 12) & 0xf;
1139         const GLint rowDa1 = (rowD[k] >> 12) & 0xf;
1140         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1141                                       rowCr0, rowCr1, rowDr0, rowDr1);
1142         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1143                                       rowCg0, rowCg1, rowDg0, rowDg1);
1144         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1145                                       rowCb0, rowCb1, rowDb0, rowDb1);
1146         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1147                                       rowCa0, rowCa1, rowDa0, rowDa1);
1148
1149         dst[i] = (a << 12) | (b << 8) | (g << 4) | r;
1150      }
1151   }
1152   else if ((datatype == GL_UNSIGNED_SHORT_1_5_5_5_REV) && (comps == 4)) {
1153      DECLARE_ROW_POINTERS0(GLushort);
1154
1155      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1156           i++, j += colStride, k += colStride) {
1157         const GLint rowAr0 = rowA[j] & 0x1f;
1158         const GLint rowAr1 = rowA[k] & 0x1f;
1159         const GLint rowBr0 = rowB[j] & 0x1f;
1160         const GLint rowBr1 = rowB[k] & 0x1f;
1161         const GLint rowCr0 = rowC[j] & 0x1f;
1162         const GLint rowCr1 = rowC[k] & 0x1f;
1163         const GLint rowDr0 = rowD[j] & 0x1f;
1164         const GLint rowDr1 = rowD[k] & 0x1f;
1165         const GLint rowAg0 = (rowA[j] >> 5) & 0x1f;
1166         const GLint rowAg1 = (rowA[k] >> 5) & 0x1f;
1167         const GLint rowBg0 = (rowB[j] >> 5) & 0x1f;
1168         const GLint rowBg1 = (rowB[k] >> 5) & 0x1f;
1169         const GLint rowCg0 = (rowC[j] >> 5) & 0x1f;
1170         const GLint rowCg1 = (rowC[k] >> 5) & 0x1f;
1171         const GLint rowDg0 = (rowD[j] >> 5) & 0x1f;
1172         const GLint rowDg1 = (rowD[k] >> 5) & 0x1f;
1173         const GLint rowAb0 = (rowA[j] >> 10) & 0x1f;
1174         const GLint rowAb1 = (rowA[k] >> 10) & 0x1f;
1175         const GLint rowBb0 = (rowB[j] >> 10) & 0x1f;
1176         const GLint rowBb1 = (rowB[k] >> 10) & 0x1f;
1177         const GLint rowCb0 = (rowC[j] >> 10) & 0x1f;
1178         const GLint rowCb1 = (rowC[k] >> 10) & 0x1f;
1179         const GLint rowDb0 = (rowD[j] >> 10) & 0x1f;
1180         const GLint rowDb1 = (rowD[k] >> 10) & 0x1f;
1181         const GLint rowAa0 = (rowA[j] >> 15) & 0x1;
1182         const GLint rowAa1 = (rowA[k] >> 15) & 0x1;
1183         const GLint rowBa0 = (rowB[j] >> 15) & 0x1;
1184         const GLint rowBa1 = (rowB[k] >> 15) & 0x1;
1185         const GLint rowCa0 = (rowC[j] >> 15) & 0x1;
1186         const GLint rowCa1 = (rowC[k] >> 15) & 0x1;
1187         const GLint rowDa0 = (rowD[j] >> 15) & 0x1;
1188         const GLint rowDa1 = (rowD[k] >> 15) & 0x1;
1189         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1190                                       rowCr0, rowCr1, rowDr0, rowDr1);
1191         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1192                                       rowCg0, rowCg1, rowDg0, rowDg1);
1193         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1194                                       rowCb0, rowCb1, rowDb0, rowDb1);
1195         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1196                                       rowCa0, rowCa1, rowDa0, rowDa1);
1197
1198         dst[i] = (a << 15) | (b << 10) | (g << 5) | r;
1199      }
1200   }
1201   else if ((datatype == GL_UNSIGNED_SHORT_5_5_5_1) && (comps == 4)) {
1202      DECLARE_ROW_POINTERS0(GLushort);
1203
1204      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1205           i++, j += colStride, k += colStride) {
1206         const GLint rowAr0 = (rowA[j] >> 11) & 0x1f;
1207         const GLint rowAr1 = (rowA[k] >> 11) & 0x1f;
1208         const GLint rowBr0 = (rowB[j] >> 11) & 0x1f;
1209         const GLint rowBr1 = (rowB[k] >> 11) & 0x1f;
1210         const GLint rowCr0 = (rowC[j] >> 11) & 0x1f;
1211         const GLint rowCr1 = (rowC[k] >> 11) & 0x1f;
1212         const GLint rowDr0 = (rowD[j] >> 11) & 0x1f;
1213         const GLint rowDr1 = (rowD[k] >> 11) & 0x1f;
1214         const GLint rowAg0 = (rowA[j] >> 6) & 0x1f;
1215         const GLint rowAg1 = (rowA[k] >> 6) & 0x1f;
1216         const GLint rowBg0 = (rowB[j] >> 6) & 0x1f;
1217         const GLint rowBg1 = (rowB[k] >> 6) & 0x1f;
1218         const GLint rowCg0 = (rowC[j] >> 6) & 0x1f;
1219         const GLint rowCg1 = (rowC[k] >> 6) & 0x1f;
1220         const GLint rowDg0 = (rowD[j] >> 6) & 0x1f;
1221         const GLint rowDg1 = (rowD[k] >> 6) & 0x1f;
1222         const GLint rowAb0 = (rowA[j] >> 1) & 0x1f;
1223         const GLint rowAb1 = (rowA[k] >> 1) & 0x1f;
1224         const GLint rowBb0 = (rowB[j] >> 1) & 0x1f;
1225         const GLint rowBb1 = (rowB[k] >> 1) & 0x1f;
1226         const GLint rowCb0 = (rowC[j] >> 1) & 0x1f;
1227         const GLint rowCb1 = (rowC[k] >> 1) & 0x1f;
1228         const GLint rowDb0 = (rowD[j] >> 1) & 0x1f;
1229         const GLint rowDb1 = (rowD[k] >> 1) & 0x1f;
1230         const GLint rowAa0 = (rowA[j] & 0x1);
1231         const GLint rowAa1 = (rowA[k] & 0x1);
1232         const GLint rowBa0 = (rowB[j] & 0x1);
1233         const GLint rowBa1 = (rowB[k] & 0x1);
1234         const GLint rowCa0 = (rowC[j] & 0x1);
1235         const GLint rowCa1 = (rowC[k] & 0x1);
1236         const GLint rowDa0 = (rowD[j] & 0x1);
1237         const GLint rowDa1 = (rowD[k] & 0x1);
1238         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1239                                       rowCr0, rowCr1, rowDr0, rowDr1);
1240         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1241                                       rowCg0, rowCg1, rowDg0, rowDg1);
1242         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1243                                       rowCb0, rowCb1, rowDb0, rowDb1);
1244         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1245                                       rowCa0, rowCa1, rowDa0, rowDa1);
1246
1247         dst[i] = (r << 11) | (g << 6) | (b << 1) | a;
1248      }
1249   }
1250   else if ((datatype == GL_UNSIGNED_BYTE_3_3_2) && (comps == 3)) {
1251      DECLARE_ROW_POINTERS0(GLubyte);
1252
1253      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1254           i++, j += colStride, k += colStride) {
1255         const GLint rowAr0 = rowA[j] & 0x3;
1256         const GLint rowAr1 = rowA[k] & 0x3;
1257         const GLint rowBr0 = rowB[j] & 0x3;
1258         const GLint rowBr1 = rowB[k] & 0x3;
1259         const GLint rowCr0 = rowC[j] & 0x3;
1260         const GLint rowCr1 = rowC[k] & 0x3;
1261         const GLint rowDr0 = rowD[j] & 0x3;
1262         const GLint rowDr1 = rowD[k] & 0x3;
1263         const GLint rowAg0 = (rowA[j] >> 2) & 0x7;
1264         const GLint rowAg1 = (rowA[k] >> 2) & 0x7;
1265         const GLint rowBg0 = (rowB[j] >> 2) & 0x7;
1266         const GLint rowBg1 = (rowB[k] >> 2) & 0x7;
1267         const GLint rowCg0 = (rowC[j] >> 2) & 0x7;
1268         const GLint rowCg1 = (rowC[k] >> 2) & 0x7;
1269         const GLint rowDg0 = (rowD[j] >> 2) & 0x7;
1270         const GLint rowDg1 = (rowD[k] >> 2) & 0x7;
1271         const GLint rowAb0 = (rowA[j] >> 5) & 0x7;
1272         const GLint rowAb1 = (rowA[k] >> 5) & 0x7;
1273         const GLint rowBb0 = (rowB[j] >> 5) & 0x7;
1274         const GLint rowBb1 = (rowB[k] >> 5) & 0x7;
1275         const GLint rowCb0 = (rowC[j] >> 5) & 0x7;
1276         const GLint rowCb1 = (rowC[k] >> 5) & 0x7;
1277         const GLint rowDb0 = (rowD[j] >> 5) & 0x7;
1278         const GLint rowDb1 = (rowD[k] >> 5) & 0x7;
1279         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1280                                       rowCr0, rowCr1, rowDr0, rowDr1);
1281         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1282                                       rowCg0, rowCg1, rowDg0, rowDg1);
1283         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1284                                       rowCb0, rowCb1, rowDb0, rowDb1);
1285         dst[i] = (b << 5) | (g << 2) | r;
1286      }
1287   }
1288   else if (datatype == MESA_UNSIGNED_BYTE_4_4 && comps == 2) {
1289      DECLARE_ROW_POINTERS0(GLubyte);
1290
1291      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1292           i++, j += colStride, k += colStride) {
1293         const GLint rowAr0 = rowA[j] & 0xf;
1294         const GLint rowAr1 = rowA[k] & 0xf;
1295         const GLint rowBr0 = rowB[j] & 0xf;
1296         const GLint rowBr1 = rowB[k] & 0xf;
1297         const GLint rowCr0 = rowC[j] & 0xf;
1298         const GLint rowCr1 = rowC[k] & 0xf;
1299         const GLint rowDr0 = rowD[j] & 0xf;
1300         const GLint rowDr1 = rowD[k] & 0xf;
1301         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
1302         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
1303         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
1304         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
1305         const GLint rowCg0 = (rowC[j] >> 4) & 0xf;
1306         const GLint rowCg1 = (rowC[k] >> 4) & 0xf;
1307         const GLint rowDg0 = (rowD[j] >> 4) & 0xf;
1308         const GLint rowDg1 = (rowD[k] >> 4) & 0xf;
1309         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1310                                       rowCr0, rowCr1, rowDr0, rowDr1);
1311         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1312                                       rowCg0, rowCg1, rowDg0, rowDg1);
1313         dst[i] = (g << 4) | r;
1314      }
1315   }
1316   else if ((datatype == GL_UNSIGNED_INT_2_10_10_10_REV) && (comps == 4)) {
1317      DECLARE_ROW_POINTERS0(GLuint);
1318
1319      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1320           i++, j += colStride, k += colStride) {
1321         const GLint rowAr0 = rowA[j] & 0x3ff;
1322         const GLint rowAr1 = rowA[k] & 0x3ff;
1323         const GLint rowBr0 = rowB[j] & 0x3ff;
1324         const GLint rowBr1 = rowB[k] & 0x3ff;
1325         const GLint rowCr0 = rowC[j] & 0x3ff;
1326         const GLint rowCr1 = rowC[k] & 0x3ff;
1327         const GLint rowDr0 = rowD[j] & 0x3ff;
1328         const GLint rowDr1 = rowD[k] & 0x3ff;
1329         const GLint rowAg0 = (rowA[j] >> 10) & 0x3ff;
1330         const GLint rowAg1 = (rowA[k] >> 10) & 0x3ff;
1331         const GLint rowBg0 = (rowB[j] >> 10) & 0x3ff;
1332         const GLint rowBg1 = (rowB[k] >> 10) & 0x3ff;
1333         const GLint rowCg0 = (rowC[j] >> 10) & 0x3ff;
1334         const GLint rowCg1 = (rowC[k] >> 10) & 0x3ff;
1335         const GLint rowDg0 = (rowD[j] >> 10) & 0x3ff;
1336         const GLint rowDg1 = (rowD[k] >> 10) & 0x3ff;
1337         const GLint rowAb0 = (rowA[j] >> 20) & 0x3ff;
1338         const GLint rowAb1 = (rowA[k] >> 20) & 0x3ff;
1339         const GLint rowBb0 = (rowB[j] >> 20) & 0x3ff;
1340         const GLint rowBb1 = (rowB[k] >> 20) & 0x3ff;
1341         const GLint rowCb0 = (rowC[j] >> 20) & 0x3ff;
1342         const GLint rowCb1 = (rowC[k] >> 20) & 0x3ff;
1343         const GLint rowDb0 = (rowD[j] >> 20) & 0x3ff;
1344         const GLint rowDb1 = (rowD[k] >> 20) & 0x3ff;
1345         const GLint rowAa0 = (rowA[j] >> 30) & 0x3;
1346         const GLint rowAa1 = (rowA[k] >> 30) & 0x3;
1347         const GLint rowBa0 = (rowB[j] >> 30) & 0x3;
1348         const GLint rowBa1 = (rowB[k] >> 30) & 0x3;
1349         const GLint rowCa0 = (rowC[j] >> 30) & 0x3;
1350         const GLint rowCa1 = (rowC[k] >> 30) & 0x3;
1351         const GLint rowDa0 = (rowD[j] >> 30) & 0x3;
1352         const GLint rowDa1 = (rowD[k] >> 30) & 0x3;
1353         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1354                                       rowCr0, rowCr1, rowDr0, rowDr1);
1355         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1356                                       rowCg0, rowCg1, rowDg0, rowDg1);
1357         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1358                                       rowCb0, rowCb1, rowDb0, rowDb1);
1359         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1360                                       rowCa0, rowCa1, rowDa0, rowDa1);
1361
1362         dst[i] = (a << 30) | (b << 20) | (g << 10) | r;
1363      }
1364   }
1365
1366   else if (datatype == GL_UNSIGNED_INT_5_9_9_9_REV && comps == 3) {
1367      DECLARE_ROW_POINTERS0(GLuint);
1368
1369      GLfloat res[3];
1370      GLfloat rowAj[3], rowBj[3], rowCj[3], rowDj[3];
1371      GLfloat rowAk[3], rowBk[3], rowCk[3], rowDk[3];
1372
1373      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1374           i++, j += colStride, k += colStride) {
1375         rgb9e5_to_float3(rowA[j], rowAj);
1376         rgb9e5_to_float3(rowB[j], rowBj);
1377         rgb9e5_to_float3(rowC[j], rowCj);
1378         rgb9e5_to_float3(rowD[j], rowDj);
1379         rgb9e5_to_float3(rowA[k], rowAk);
1380         rgb9e5_to_float3(rowB[k], rowBk);
1381         rgb9e5_to_float3(rowC[k], rowCk);
1382         rgb9e5_to_float3(rowD[k], rowDk);
1383         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0] +
1384                   rowCj[0] + rowCk[0] + rowDj[0] + rowDk[0]) * 0.125F;
1385         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1] +
1386                   rowCj[1] + rowCk[1] + rowDj[1] + rowDk[1]) * 0.125F;
1387         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2] +
1388                   rowCj[2] + rowCk[2] + rowDj[2] + rowDk[2]) * 0.125F;
1389         dst[i] = float3_to_rgb9e5(res);
1390      }
1391   }
1392
1393   else if (datatype == GL_UNSIGNED_INT_10F_11F_11F_REV && comps == 3) {
1394      DECLARE_ROW_POINTERS0(GLuint);
1395
1396      GLfloat res[3];
1397      GLfloat rowAj[3], rowBj[3], rowCj[3], rowDj[3];
1398      GLfloat rowAk[3], rowBk[3], rowCk[3], rowDk[3];
1399
1400      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1401           i++, j += colStride, k += colStride) {
1402         r11g11b10f_to_float3(rowA[j], rowAj);
1403         r11g11b10f_to_float3(rowB[j], rowBj);
1404         r11g11b10f_to_float3(rowC[j], rowCj);
1405         r11g11b10f_to_float3(rowD[j], rowDj);
1406         r11g11b10f_to_float3(rowA[k], rowAk);
1407         r11g11b10f_to_float3(rowB[k], rowBk);
1408         r11g11b10f_to_float3(rowC[k], rowCk);
1409         r11g11b10f_to_float3(rowD[k], rowDk);
1410         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0] +
1411                   rowCj[0] + rowCk[0] + rowDj[0] + rowDk[0]) * 0.125F;
1412         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1] +
1413                   rowCj[1] + rowCk[1] + rowDj[1] + rowDk[1]) * 0.125F;
1414         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2] +
1415                   rowCj[2] + rowCk[2] + rowDj[2] + rowDk[2]) * 0.125F;
1416         dst[i] = float3_to_r11g11b10f(res);
1417      }
1418   }
1419
1420   else if (datatype == GL_FLOAT_32_UNSIGNED_INT_24_8_REV && comps == 1) {
1421      DECLARE_ROW_POINTERS(GLfloat, 2);
1422
1423      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1424           i++, j += colStride, k += colStride) {
1425         FILTER_F_3D(0);
1426      }
1427   }
1428
1429   else {
1430      unreachable("bad format in do_row()");
1431   }
1432}
1433
1434
1435/*
1436 * These functions generate a 1/2-size mipmap image from a source image.
1437 * Texture borders are handled by copying or averaging the source image's
1438 * border texels, depending on the scale-down factor.
1439 */
1440
1441static void
1442make_1d_mipmap(GLenum datatype, GLuint comps, GLint border,
1443               GLint srcWidth, const GLubyte *srcPtr,
1444               GLint dstWidth, GLubyte *dstPtr)
1445{
1446   const GLint bpt = bytes_per_pixel(datatype, comps);
1447   const GLubyte *src;
1448   GLubyte *dst;
1449
1450   /* skip the border pixel, if any */
1451   src = srcPtr + border * bpt;
1452   dst = dstPtr + border * bpt;
1453
1454   /* we just duplicate the input row, kind of hack, saves code */
1455   do_row(datatype, comps, srcWidth - 2 * border, src, src,
1456          dstWidth - 2 * border, dst);
1457
1458   if (border) {
1459      /* copy left-most pixel from source */
1460      assert(dstPtr);
1461      assert(srcPtr);
1462      memcpy(dstPtr, srcPtr, bpt);
1463      /* copy right-most pixel from source */
1464      memcpy(dstPtr + (dstWidth - 1) * bpt,
1465             srcPtr + (srcWidth - 1) * bpt,
1466             bpt);
1467   }
1468}
1469
1470
1471static void
1472make_2d_mipmap(GLenum datatype, GLuint comps, GLint border,
1473               GLint srcWidth, GLint srcHeight,
1474               const GLubyte *srcPtr, GLint srcRowStride,
1475               GLint dstWidth, GLint dstHeight,
1476               GLubyte *dstPtr, GLint dstRowStride)
1477{
1478   const GLint bpt = bytes_per_pixel(datatype, comps);
1479   const GLint srcWidthNB = srcWidth - 2 * border;  /* sizes w/out border */
1480   const GLint dstWidthNB = dstWidth - 2 * border;
1481   const GLint dstHeightNB = dstHeight - 2 * border;
1482   const GLubyte *srcA, *srcB;
1483   GLubyte *dst;
1484   GLint row, srcRowStep;
1485
1486   /* Compute src and dst pointers, skipping any border */
1487   srcA = srcPtr + border * ((srcWidth + 1) * bpt);
1488   if (srcHeight > 1 && srcHeight > dstHeight) {
1489      /* sample from two source rows */
1490      srcB = srcA + srcRowStride;
1491      srcRowStep = 2;
1492   }
1493   else {
1494      /* sample from one source row */
1495      srcB = srcA;
1496      srcRowStep = 1;
1497   }
1498
1499   dst = dstPtr + border * ((dstWidth + 1) * bpt);
1500
1501   for (row = 0; row < dstHeightNB; row++) {
1502      do_row(datatype, comps, srcWidthNB, srcA, srcB,
1503             dstWidthNB, dst);
1504      srcA += srcRowStep * srcRowStride;
1505      srcB += srcRowStep * srcRowStride;
1506      dst += dstRowStride;
1507   }
1508
1509   /* This is ugly but probably won't be used much */
1510   if (border > 0) {
1511      /* fill in dest border */
1512      /* lower-left border pixel */
1513      assert(dstPtr);
1514      assert(srcPtr);
1515      memcpy(dstPtr, srcPtr, bpt);
1516      /* lower-right border pixel */
1517      memcpy(dstPtr + (dstWidth - 1) * bpt,
1518             srcPtr + (srcWidth - 1) * bpt, bpt);
1519      /* upper-left border pixel */
1520      memcpy(dstPtr + dstWidth * (dstHeight - 1) * bpt,
1521             srcPtr + srcWidth * (srcHeight - 1) * bpt, bpt);
1522      /* upper-right border pixel */
1523      memcpy(dstPtr + (dstWidth * dstHeight - 1) * bpt,
1524             srcPtr + (srcWidth * srcHeight - 1) * bpt, bpt);
1525      /* lower border */
1526      do_row(datatype, comps, srcWidthNB,
1527             srcPtr + bpt,
1528             srcPtr + bpt,
1529             dstWidthNB, dstPtr + bpt);
1530      /* upper border */
1531      do_row(datatype, comps, srcWidthNB,
1532             srcPtr + (srcWidth * (srcHeight - 1) + 1) * bpt,
1533             srcPtr + (srcWidth * (srcHeight - 1) + 1) * bpt,
1534             dstWidthNB,
1535             dstPtr + (dstWidth * (dstHeight - 1) + 1) * bpt);
1536      /* left and right borders */
1537      if (srcHeight == dstHeight) {
1538         /* copy border pixel from src to dst */
1539         for (row = 1; row < srcHeight; row++) {
1540            memcpy(dstPtr + dstWidth * row * bpt,
1541                   srcPtr + srcWidth * row * bpt, bpt);
1542            memcpy(dstPtr + (dstWidth * row + dstWidth - 1) * bpt,
1543                   srcPtr + (srcWidth * row + srcWidth - 1) * bpt, bpt);
1544         }
1545      }
1546      else {
1547         /* average two src pixels each dest pixel */
1548         for (row = 0; row < dstHeightNB; row += 2) {
1549            do_row(datatype, comps, 1,
1550                   srcPtr + (srcWidth * (row * 2 + 1)) * bpt,
1551                   srcPtr + (srcWidth * (row * 2 + 2)) * bpt,
1552                   1, dstPtr + (dstWidth * row + 1) * bpt);
1553            do_row(datatype, comps, 1,
1554                   srcPtr + (srcWidth * (row * 2 + 1) + srcWidth - 1) * bpt,
1555                   srcPtr + (srcWidth * (row * 2 + 2) + srcWidth - 1) * bpt,
1556                   1, dstPtr + (dstWidth * row + 1 + dstWidth - 1) * bpt);
1557         }
1558      }
1559   }
1560}
1561
1562
1563static void
1564make_3d_mipmap(GLenum datatype, GLuint comps, GLint border,
1565               GLint srcWidth, GLint srcHeight, GLint srcDepth,
1566               const GLubyte **srcPtr, GLint srcRowStride,
1567               GLint dstWidth, GLint dstHeight, GLint dstDepth,
1568               GLubyte **dstPtr, GLint dstRowStride)
1569{
1570   const GLint bpt = bytes_per_pixel(datatype, comps);
1571   const GLint srcWidthNB = srcWidth - 2 * border;  /* sizes w/out border */
1572   const GLint srcDepthNB = srcDepth - 2 * border;
1573   const GLint dstWidthNB = dstWidth - 2 * border;
1574   const GLint dstHeightNB = dstHeight - 2 * border;
1575   const GLint dstDepthNB = dstDepth - 2 * border;
1576   GLint img, row;
1577   GLint bytesPerSrcImage, bytesPerDstImage;
1578   GLint srcImageOffset, srcRowOffset;
1579
1580   (void) srcDepthNB; /* silence warnings */
1581
1582   bytesPerSrcImage = srcRowStride * srcHeight * bpt;
1583   bytesPerDstImage = dstRowStride * dstHeight * bpt;
1584
1585   /* Offset between adjacent src images to be averaged together */
1586   srcImageOffset = (srcDepth == dstDepth) ? 0 : 1;
1587
1588   /* Offset between adjacent src rows to be averaged together */
1589   srcRowOffset = (srcHeight == dstHeight) ? 0 : srcRowStride;
1590
1591   /*
1592    * Need to average together up to 8 src pixels for each dest pixel.
1593    * Break that down into 3 operations:
1594    *   1. take two rows from source image and average them together.
1595    *   2. take two rows from next source image and average them together.
1596    *   3. take the two averaged rows and average them for the final dst row.
1597    */
1598
1599   /*
1600   printf("mip3d %d x %d x %d  ->  %d x %d x %d\n",
1601          srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
1602   */
1603
1604   for (img = 0; img < dstDepthNB; img++) {
1605      /* first source image pointer, skipping border */
1606      const GLubyte *imgSrcA = srcPtr[img * 2 + border]
1607         + srcRowStride * border + bpt * border;
1608      /* second source image pointer, skipping border */
1609      const GLubyte *imgSrcB = srcPtr[img * 2 + srcImageOffset + border]
1610         + srcRowStride * border + bpt * border;
1611
1612      /* address of the dest image, skipping border */
1613      GLubyte *imgDst = dstPtr[img + border]
1614         + dstRowStride * border + bpt * border;
1615
1616      /* setup the four source row pointers and the dest row pointer */
1617      const GLubyte *srcImgARowA = imgSrcA;
1618      const GLubyte *srcImgARowB = imgSrcA + srcRowOffset;
1619      const GLubyte *srcImgBRowA = imgSrcB;
1620      const GLubyte *srcImgBRowB = imgSrcB + srcRowOffset;
1621      GLubyte *dstImgRow = imgDst;
1622
1623      for (row = 0; row < dstHeightNB; row++) {
1624         do_row_3D(datatype, comps, srcWidthNB,
1625                   srcImgARowA, srcImgARowB,
1626                   srcImgBRowA, srcImgBRowB,
1627                   dstWidthNB, dstImgRow);
1628
1629         /* advance to next rows */
1630         srcImgARowA += srcRowStride + srcRowOffset;
1631         srcImgARowB += srcRowStride + srcRowOffset;
1632         srcImgBRowA += srcRowStride + srcRowOffset;
1633         srcImgBRowB += srcRowStride + srcRowOffset;
1634         dstImgRow += dstRowStride;
1635      }
1636   }
1637
1638
1639   /* Luckily we can leverage the make_2d_mipmap() function here! */
1640   if (border > 0) {
1641      /* do front border image */
1642      make_2d_mipmap(datatype, comps, 1,
1643                     srcWidth, srcHeight, srcPtr[0], srcRowStride,
1644                     dstWidth, dstHeight, dstPtr[0], dstRowStride);
1645      /* do back border image */
1646      make_2d_mipmap(datatype, comps, 1,
1647                     srcWidth, srcHeight, srcPtr[srcDepth - 1], srcRowStride,
1648                     dstWidth, dstHeight, dstPtr[dstDepth - 1], dstRowStride);
1649
1650      /* do four remaining border edges that span the image slices */
1651      if (srcDepth == dstDepth) {
1652         /* just copy border pixels from src to dst */
1653         for (img = 0; img < dstDepthNB; img++) {
1654            const GLubyte *src;
1655            GLubyte *dst;
1656
1657            /* do border along [img][row=0][col=0] */
1658            src = srcPtr[img * 2];
1659            dst = dstPtr[img];
1660            memcpy(dst, src, bpt);
1661
1662            /* do border along [img][row=dstHeight-1][col=0] */
1663            src = srcPtr[img * 2] + (srcHeight - 1) * srcRowStride;
1664            dst = dstPtr[img] + (dstHeight - 1) * dstRowStride;
1665            memcpy(dst, src, bpt);
1666
1667            /* do border along [img][row=0][col=dstWidth-1] */
1668            src = srcPtr[img * 2] + (srcWidth - 1) * bpt;
1669            dst = dstPtr[img] + (dstWidth - 1) * bpt;
1670            memcpy(dst, src, bpt);
1671
1672            /* do border along [img][row=dstHeight-1][col=dstWidth-1] */
1673            src = srcPtr[img * 2] + (bytesPerSrcImage - bpt);
1674            dst = dstPtr[img] + (bytesPerDstImage - bpt);
1675            memcpy(dst, src, bpt);
1676         }
1677      }
1678      else {
1679         /* average border pixels from adjacent src image pairs */
1680         assert(srcDepthNB == 2 * dstDepthNB);
1681         for (img = 0; img < dstDepthNB; img++) {
1682            const GLubyte *srcA, *srcB;
1683            GLubyte *dst;
1684
1685            /* do border along [img][row=0][col=0] */
1686            srcA = srcPtr[img * 2 + 0];
1687            srcB = srcPtr[img * 2 + srcImageOffset];
1688            dst = dstPtr[img];
1689            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1690
1691            /* do border along [img][row=dstHeight-1][col=0] */
1692            srcA = srcPtr[img * 2 + 0]
1693               + (srcHeight - 1) * srcRowStride;
1694            srcB = srcPtr[img * 2 + srcImageOffset]
1695               + (srcHeight - 1) * srcRowStride;
1696            dst = dstPtr[img] + (dstHeight - 1) * dstRowStride;
1697            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1698
1699            /* do border along [img][row=0][col=dstWidth-1] */
1700            srcA = srcPtr[img * 2 + 0] + (srcWidth - 1) * bpt;
1701            srcB = srcPtr[img * 2 + srcImageOffset] + (srcWidth - 1) * bpt;
1702            dst = dstPtr[img] + (dstWidth - 1) * bpt;
1703            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1704
1705            /* do border along [img][row=dstHeight-1][col=dstWidth-1] */
1706            srcA = srcPtr[img * 2 + 0] + (bytesPerSrcImage - bpt);
1707            srcB = srcPtr[img * 2 + srcImageOffset] + (bytesPerSrcImage - bpt);
1708            dst = dstPtr[img] + (bytesPerDstImage - bpt);
1709            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1710         }
1711      }
1712   }
1713}
1714
1715
1716/**
1717 * Down-sample a texture image to produce the next lower mipmap level.
1718 * \param comps  components per texel (1, 2, 3 or 4)
1719 * \param srcData  array[slice] of pointers to source image slices
1720 * \param dstData  array[slice] of pointers to dest image slices
1721 * \param srcRowStride  stride between source rows, in bytes
1722 * \param dstRowStride  stride between destination rows, in bytes
1723 */
1724static void
1725_mesa_generate_mipmap_level(GLenum target,
1726                            GLenum datatype, GLuint comps,
1727                            GLint border,
1728                            GLint srcWidth, GLint srcHeight, GLint srcDepth,
1729                            const GLubyte **srcData,
1730                            GLint srcRowStride,
1731                            GLint dstWidth, GLint dstHeight, GLint dstDepth,
1732                            GLubyte **dstData,
1733                            GLint dstRowStride)
1734{
1735   int i;
1736
1737   switch (target) {
1738   case GL_TEXTURE_1D:
1739      make_1d_mipmap(datatype, comps, border,
1740                     srcWidth, srcData[0],
1741                     dstWidth, dstData[0]);
1742      break;
1743   case GL_TEXTURE_2D:
1744   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
1745   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
1746   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
1747   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
1748   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
1749   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
1750      make_2d_mipmap(datatype, comps, border,
1751                     srcWidth, srcHeight, srcData[0], srcRowStride,
1752                     dstWidth, dstHeight, dstData[0], dstRowStride);
1753      break;
1754   case GL_TEXTURE_3D:
1755      make_3d_mipmap(datatype, comps, border,
1756                     srcWidth, srcHeight, srcDepth,
1757                     srcData, srcRowStride,
1758                     dstWidth, dstHeight, dstDepth,
1759                     dstData, dstRowStride);
1760      break;
1761   case GL_TEXTURE_1D_ARRAY_EXT:
1762      assert(srcHeight == 1);
1763      assert(dstHeight == 1);
1764      for (i = 0; i < dstDepth; i++) {
1765         make_1d_mipmap(datatype, comps, border,
1766                        srcWidth, srcData[i],
1767                        dstWidth, dstData[i]);
1768      }
1769      break;
1770   case GL_TEXTURE_2D_ARRAY_EXT:
1771   case GL_TEXTURE_CUBE_MAP_ARRAY:
1772      for (i = 0; i < dstDepth; i++) {
1773         make_2d_mipmap(datatype, comps, border,
1774                        srcWidth, srcHeight, srcData[i], srcRowStride,
1775                        dstWidth, dstHeight, dstData[i], dstRowStride);
1776      }
1777      break;
1778   case GL_TEXTURE_RECTANGLE_NV:
1779   case GL_TEXTURE_EXTERNAL_OES:
1780      /* no mipmaps, do nothing */
1781      break;
1782   default:
1783      unreachable("bad tex target in _mesa_generate_mipmaps");
1784   }
1785}
1786
1787
1788/**
1789 * compute next (level+1) image size
1790 * \return GL_FALSE if no smaller size can be generated (eg. src is 1x1x1 size)
1791 */
1792GLboolean
1793_mesa_next_mipmap_level_size(GLenum target, GLint border,
1794                       GLint srcWidth, GLint srcHeight, GLint srcDepth,
1795                       GLint *dstWidth, GLint *dstHeight, GLint *dstDepth)
1796{
1797   if (srcWidth - 2 * border > 1) {
1798      *dstWidth = (srcWidth - 2 * border) / 2 + 2 * border;
1799   }
1800   else {
1801      *dstWidth = srcWidth; /* can't go smaller */
1802   }
1803
1804   if ((srcHeight - 2 * border > 1) &&
1805       target != GL_TEXTURE_1D_ARRAY_EXT &&
1806       target != GL_PROXY_TEXTURE_1D_ARRAY_EXT) {
1807      *dstHeight = (srcHeight - 2 * border) / 2 + 2 * border;
1808   }
1809   else {
1810      *dstHeight = srcHeight; /* can't go smaller */
1811   }
1812
1813   if ((srcDepth - 2 * border > 1) &&
1814       target != GL_TEXTURE_2D_ARRAY_EXT &&
1815       target != GL_PROXY_TEXTURE_2D_ARRAY_EXT &&
1816       target != GL_TEXTURE_CUBE_MAP_ARRAY &&
1817       target != GL_PROXY_TEXTURE_CUBE_MAP_ARRAY) {
1818      *dstDepth = (srcDepth - 2 * border) / 2 + 2 * border;
1819   }
1820   else {
1821      *dstDepth = srcDepth; /* can't go smaller */
1822   }
1823
1824   if (*dstWidth == srcWidth &&
1825       *dstHeight == srcHeight &&
1826       *dstDepth == srcDepth) {
1827      return GL_FALSE;
1828   }
1829   else {
1830      return GL_TRUE;
1831   }
1832}
1833
1834
1835/**
1836 * Helper function for mipmap generation.
1837 * Make sure the specified destination mipmap level is the right size/format
1838 * for mipmap generation.  If not, (re) allocate it.
1839 * \return GL_TRUE if successful, GL_FALSE if mipmap generation should stop
1840 */
1841static GLboolean
1842prepare_mipmap_level(struct gl_context *ctx,
1843                     struct gl_texture_object *texObj, GLuint level,
1844                     GLsizei width, GLsizei height, GLsizei depth,
1845                     GLsizei border, GLenum intFormat, mesa_format format)
1846{
1847   const GLuint numFaces = _mesa_num_tex_faces(texObj->Target);
1848   GLuint face;
1849
1850   if (texObj->Immutable) {
1851      /* The texture was created with glTexStorage() so the number/size of
1852       * mipmap levels is fixed and the storage for all images is already
1853       * allocated.
1854       */
1855      if (!texObj->Image[0][level]) {
1856         /* No more levels to create - we're done */
1857         return GL_FALSE;
1858      }
1859      else {
1860         /* Nothing to do - the texture memory must have already been
1861          * allocated to the right size so we're all set.
1862          */
1863         return GL_TRUE;
1864      }
1865   }
1866
1867   for (face = 0; face < numFaces; face++) {
1868      struct gl_texture_image *dstImage;
1869      const GLenum target = _mesa_cube_face_target(texObj->Target, face);
1870
1871      dstImage = _mesa_get_tex_image(ctx, texObj, target, level);
1872      if (!dstImage) {
1873         /* out of memory */
1874         return GL_FALSE;
1875      }
1876
1877      if (dstImage->Width != width ||
1878          dstImage->Height != height ||
1879          dstImage->Depth != depth ||
1880          dstImage->Border != border ||
1881          dstImage->InternalFormat != intFormat ||
1882          dstImage->TexFormat != format) {
1883         /* need to (re)allocate image */
1884         st_FreeTextureImageBuffer(ctx, dstImage);
1885
1886         _mesa_init_teximage_fields(ctx, dstImage,
1887                                    width, height, depth,
1888                                    border, intFormat, format);
1889
1890         st_AllocTextureImageBuffer(ctx, dstImage);
1891
1892         /* in case the mipmap level is part of an FBO: */
1893         _mesa_update_fbo_texture(ctx, texObj, face, level);
1894
1895         ctx->NewState |= _NEW_TEXTURE_OBJECT;
1896         ctx->PopAttribState |= GL_TEXTURE_BIT;
1897      }
1898   }
1899
1900   return GL_TRUE;
1901}
1902
1903
1904/**
1905 * Prepare all mipmap levels beyond 'baseLevel' for mipmap generation.
1906 * When finished, all the gl_texture_image structures for the smaller
1907 * mipmap levels will be consistent with the base level (in terms of
1908 * dimensions, format, etc).
1909 */
1910void
1911_mesa_prepare_mipmap_levels(struct gl_context *ctx,
1912                            struct gl_texture_object *texObj,
1913                            unsigned baseLevel, unsigned maxLevel)
1914{
1915   const struct gl_texture_image *baseImage =
1916      _mesa_select_tex_image(texObj, texObj->Target, baseLevel);
1917
1918   if (baseImage == NULL)
1919      return;
1920
1921   const GLint border = 0;
1922   GLint width = baseImage->Width;
1923   GLint height = baseImage->Height;
1924   GLint depth = baseImage->Depth;
1925   const GLenum intFormat = baseImage->InternalFormat;
1926   const mesa_format texFormat = baseImage->TexFormat;
1927   GLint newWidth, newHeight, newDepth;
1928
1929   /* Prepare baseLevel + 1, baseLevel + 2, ... */
1930   for (unsigned level = baseLevel + 1; level <= maxLevel; level++) {
1931      if (!_mesa_next_mipmap_level_size(texObj->Target, border,
1932                                        width, height, depth,
1933                                        &newWidth, &newHeight, &newDepth)) {
1934         /* all done */
1935         break;
1936      }
1937
1938      if (!prepare_mipmap_level(ctx, texObj, level,
1939                                newWidth, newHeight, newDepth,
1940                                border, intFormat, texFormat)) {
1941         break;
1942      }
1943
1944      width = newWidth;
1945      height = newHeight;
1946      depth = newDepth;
1947   }
1948}
1949
1950
1951static void
1952generate_mipmap_uncompressed(struct gl_context *ctx, GLenum target,
1953                             struct gl_texture_object *texObj,
1954                             const struct gl_texture_image *srcImage,
1955                             GLuint maxLevel)
1956{
1957   GLuint level;
1958   GLenum datatype;
1959   GLuint comps;
1960
1961   _mesa_uncompressed_format_to_type_and_comps(srcImage->TexFormat, &datatype, &comps);
1962
1963   for (level = texObj->Attrib.BaseLevel; level < maxLevel; level++) {
1964      /* generate image[level+1] from image[level] */
1965      struct gl_texture_image *srcImage, *dstImage;
1966      GLint srcRowStride, dstRowStride;
1967      GLint srcWidth, srcHeight, srcDepth;
1968      GLint dstWidth, dstHeight, dstDepth;
1969      GLint border;
1970      GLint slice;
1971      GLubyte **srcMaps, **dstMaps;
1972      GLboolean success = GL_TRUE;
1973
1974      /* get src image parameters */
1975      srcImage = _mesa_select_tex_image(texObj, target, level);
1976      assert(srcImage);
1977      srcWidth = srcImage->Width;
1978      srcHeight = srcImage->Height;
1979      srcDepth = srcImage->Depth;
1980      border = srcImage->Border;
1981
1982      /* get dest gl_texture_image */
1983      dstImage = _mesa_select_tex_image(texObj, target, level + 1);
1984      if (!dstImage) {
1985         break;
1986      }
1987      dstWidth = dstImage->Width;
1988      dstHeight = dstImage->Height;
1989      dstDepth = dstImage->Depth;
1990
1991      if (target == GL_TEXTURE_1D_ARRAY) {
1992         srcDepth = srcHeight;
1993         dstDepth = dstHeight;
1994         srcHeight = 1;
1995         dstHeight = 1;
1996      }
1997
1998      /* Map src texture image slices */
1999      srcMaps = calloc(srcDepth, sizeof(GLubyte *));
2000      if (srcMaps) {
2001         for (slice = 0; slice < srcDepth; slice++) {
2002            st_MapTextureImage(ctx, srcImage, slice,
2003                               0, 0, srcWidth, srcHeight,
2004                               GL_MAP_READ_BIT,
2005                               &srcMaps[slice], &srcRowStride);
2006            if (!srcMaps[slice]) {
2007               success = GL_FALSE;
2008               break;
2009            }
2010         }
2011      }
2012      else {
2013         success = GL_FALSE;
2014      }
2015
2016      /* Map dst texture image slices */
2017      dstMaps = calloc(dstDepth, sizeof(GLubyte *));
2018      if (dstMaps) {
2019         for (slice = 0; slice < dstDepth; slice++) {
2020            st_MapTextureImage(ctx, dstImage, slice,
2021                               0, 0, dstWidth, dstHeight,
2022                               GL_MAP_WRITE_BIT,
2023                               &dstMaps[slice], &dstRowStride);
2024            if (!dstMaps[slice]) {
2025               success = GL_FALSE;
2026               break;
2027            }
2028         }
2029      }
2030      else {
2031         success = GL_FALSE;
2032      }
2033
2034      if (success) {
2035         /* generate one mipmap level (for 1D/2D/3D/array/etc texture) */
2036         _mesa_generate_mipmap_level(target, datatype, comps, border,
2037                                     srcWidth, srcHeight, srcDepth,
2038                                     (const GLubyte **) srcMaps, srcRowStride,
2039                                     dstWidth, dstHeight, dstDepth,
2040                                     dstMaps, dstRowStride);
2041      }
2042
2043      /* Unmap src image slices */
2044      if (srcMaps) {
2045         for (slice = 0; slice < srcDepth; slice++) {
2046            if (srcMaps[slice]) {
2047               st_UnmapTextureImage(ctx, srcImage, slice);
2048            }
2049         }
2050         free(srcMaps);
2051      }
2052
2053      /* Unmap dst image slices */
2054      if (dstMaps) {
2055         for (slice = 0; slice < dstDepth; slice++) {
2056            if (dstMaps[slice]) {
2057               st_UnmapTextureImage(ctx, dstImage, slice);
2058            }
2059         }
2060         free(dstMaps);
2061      }
2062
2063      if (!success) {
2064         _mesa_error(ctx, GL_OUT_OF_MEMORY, "mipmap generation");
2065         break;
2066      }
2067   } /* loop over mipmap levels */
2068}
2069
2070
2071static void
2072generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
2073                           struct gl_texture_object *texObj,
2074                           struct gl_texture_image *srcImage,
2075                           GLuint maxLevel)
2076{
2077   GLuint level;
2078   mesa_format temp_format;
2079   GLint components;
2080   GLuint temp_src_row_stride, temp_src_img_stride; /* in bytes */
2081   GLubyte *temp_src = NULL, *temp_dst = NULL;
2082   GLenum temp_datatype;
2083   GLenum temp_base_format;
2084   GLubyte **temp_src_slices = NULL, **temp_dst_slices = NULL;
2085
2086   /* only two types of compressed textures at this time */
2087   assert(texObj->Target == GL_TEXTURE_2D ||
2088          texObj->Target == GL_TEXTURE_2D_ARRAY ||
2089          texObj->Target == GL_TEXTURE_CUBE_MAP ||
2090          texObj->Target == GL_TEXTURE_CUBE_MAP_ARRAY);
2091
2092   /*
2093    * Choose a format for the temporary, uncompressed base image.
2094    * Then, get number of components, choose temporary image datatype,
2095    * and get base format.
2096    */
2097   temp_format = _mesa_get_uncompressed_format(srcImage->TexFormat);
2098
2099   components = _mesa_format_num_components(temp_format);
2100
2101   switch (_mesa_get_format_datatype(srcImage->TexFormat)) {
2102   case GL_FLOAT:
2103      temp_datatype = GL_FLOAT;
2104      break;
2105   case GL_SIGNED_NORMALIZED:
2106      /* Revisit this if we get compressed formats with >8 bits per component */
2107      temp_datatype = GL_BYTE;
2108      break;
2109   default:
2110      temp_datatype = GL_UNSIGNED_BYTE;
2111   }
2112
2113   temp_base_format = _mesa_get_format_base_format(temp_format);
2114
2115
2116   /* allocate storage for the temporary, uncompressed image */
2117   temp_src_row_stride = _mesa_format_row_stride(temp_format, srcImage->Width);
2118   temp_src_img_stride = _mesa_format_image_size(temp_format, srcImage->Width,
2119                                                 srcImage->Height, 1);
2120   temp_src = malloc(temp_src_img_stride * srcImage->Depth);
2121
2122   /* Allocate storage for arrays of slice pointers */
2123   temp_src_slices = malloc(srcImage->Depth * sizeof(GLubyte *));
2124   temp_dst_slices = malloc(srcImage->Depth * sizeof(GLubyte *));
2125
2126   if (!temp_src || !temp_src_slices || !temp_dst_slices) {
2127      _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
2128      goto end;
2129   }
2130
2131   /* decompress base image to the temporary src buffer */
2132   {
2133      /* save pixel packing mode */
2134      struct gl_pixelstore_attrib save = ctx->Pack;
2135      /* use default/tight packing parameters */
2136      ctx->Pack = ctx->DefaultPacking;
2137
2138      /* Get the uncompressed image */
2139      assert(srcImage->Level == texObj->Attrib.BaseLevel);
2140      st_GetTexSubImage(ctx,
2141                        0, 0, 0,
2142                        srcImage->Width, srcImage->Height,
2143                        srcImage->Depth,
2144                        temp_base_format, temp_datatype,
2145                        temp_src, srcImage);
2146      /* restore packing mode */
2147      ctx->Pack = save;
2148   }
2149
2150   for (level = texObj->Attrib.BaseLevel; level < maxLevel; level++) {
2151      /* generate image[level+1] from image[level] */
2152      const struct gl_texture_image *srcImage;
2153      struct gl_texture_image *dstImage;
2154      GLint srcWidth, srcHeight, srcDepth;
2155      GLint dstWidth, dstHeight, dstDepth;
2156      GLint border;
2157      GLuint temp_dst_row_stride, temp_dst_img_stride; /* in bytes */
2158      GLint i;
2159
2160      /* get src image parameters */
2161      srcImage = _mesa_select_tex_image(texObj, target, level);
2162      assert(srcImage);
2163      srcWidth = srcImage->Width;
2164      srcHeight = srcImage->Height;
2165      srcDepth = srcImage->Depth;
2166      border = srcImage->Border;
2167
2168      /* get dest gl_texture_image */
2169      dstImage = _mesa_select_tex_image(texObj, target, level + 1);
2170      if (!dstImage) {
2171         break;
2172      }
2173      dstWidth = dstImage->Width;
2174      dstHeight = dstImage->Height;
2175      dstDepth = dstImage->Depth;
2176
2177      /* Compute dst image strides and alloc memory on first iteration */
2178      temp_dst_row_stride = _mesa_format_row_stride(temp_format, dstWidth);
2179      temp_dst_img_stride = _mesa_format_image_size(temp_format, dstWidth,
2180                                                    dstHeight, 1);
2181      if (!temp_dst) {
2182         temp_dst = malloc(temp_dst_img_stride * dstDepth);
2183         if (!temp_dst) {
2184            _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
2185            goto end;
2186         }
2187      }
2188
2189      /* for 2D arrays, setup array[depth] of slice pointers */
2190      for (i = 0; i < srcDepth; i++) {
2191         temp_src_slices[i] = temp_src + temp_src_img_stride * i;
2192      }
2193      for (i = 0; i < dstDepth; i++) {
2194         temp_dst_slices[i] = temp_dst + temp_dst_img_stride * i;
2195      }
2196
2197      /* Rescale src image to dest image.
2198       * This will loop over the slices of a 2D array.
2199       */
2200      _mesa_generate_mipmap_level(target, temp_datatype, components, border,
2201                                  srcWidth, srcHeight, srcDepth,
2202                                  (const GLubyte **) temp_src_slices,
2203                                  temp_src_row_stride,
2204                                  dstWidth, dstHeight, dstDepth,
2205                                  temp_dst_slices, temp_dst_row_stride);
2206
2207      /* The image space was allocated above so use glTexSubImage now */
2208      st_TexSubImage(ctx, 2, dstImage,
2209                     0, 0, 0, dstWidth, dstHeight, dstDepth,
2210                     temp_base_format, temp_datatype,
2211                     temp_dst, &ctx->DefaultPacking);
2212
2213      /* swap src and dest pointers */
2214      {
2215         GLubyte *temp = temp_src;
2216         temp_src = temp_dst;
2217         temp_dst = temp;
2218         temp_src_row_stride = temp_dst_row_stride;
2219         temp_src_img_stride = temp_dst_img_stride;
2220      }
2221   } /* loop over mipmap levels */
2222
2223end:
2224   free(temp_src);
2225   free(temp_dst);
2226   free(temp_src_slices);
2227   free(temp_dst_slices);
2228}
2229
2230/**
2231 * Automatic mipmap generation.
2232 * This is the fallback/default function for mipmap generation.
2233 * Generate a complete set of mipmaps from texObj's BaseLevel image.
2234 * Stop at texObj's MaxLevel or when we get to the 1x1 texture.
2235 * For cube maps, target will be one of
2236 * GL_TEXTURE_CUBE_MAP_POSITIVE/NEGATIVE_X/Y/Z; never GL_TEXTURE_CUBE_MAP.
2237 */
2238void
2239_mesa_generate_mipmap(struct gl_context *ctx, GLenum target,
2240                      struct gl_texture_object *texObj)
2241{
2242   struct gl_texture_image *srcImage;
2243   GLint maxLevel;
2244
2245   assert(texObj);
2246   srcImage = _mesa_select_tex_image(texObj, target, texObj->Attrib.BaseLevel);
2247   assert(srcImage);
2248
2249   maxLevel = _mesa_max_texture_levels(ctx, texObj->Target) - 1;
2250   assert(maxLevel >= 0);  /* bad target */
2251
2252   maxLevel = MIN2(maxLevel, texObj->Attrib.MaxLevel);
2253
2254   _mesa_prepare_mipmap_levels(ctx, texObj, texObj->Attrib.BaseLevel, maxLevel);
2255
2256   if (_mesa_is_format_compressed(srcImage->TexFormat)) {
2257      generate_mipmap_compressed(ctx, target, texObj, srcImage, maxLevel);
2258   } else {
2259      generate_mipmap_uncompressed(ctx, target, texObj, srcImage, maxLevel);
2260   }
2261}
2262