1/*
2 * Mesa 3-D graphics library
3 * Version:  7.1
4 *
5 * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25
26/**
27 * \file mipmap.c  mipmap generation and teximage resizing functions.
28 */
29
30#include "imports.h"
31#include "formats.h"
32#include "glformats.h"
33#include "mipmap.h"
34#include "mtypes.h"
35#include "teximage.h"
36#include "texobj.h"
37#include "texstore.h"
38#include "image.h"
39#include "macros.h"
40#include "../../gallium/auxiliary/util/u_format_rgb9e5.h"
41#include "../../gallium/auxiliary/util/u_format_r11g11b10f.h"
42
43
44
45static GLint
46bytes_per_pixel(GLenum datatype, GLuint comps)
47{
48   GLint b;
49
50   if (datatype == GL_UNSIGNED_INT_8_24_REV_MESA ||
51       datatype == GL_UNSIGNED_INT_24_8_MESA)
52      return 4;
53
54   b = _mesa_sizeof_packed_type(datatype);
55   assert(b >= 0);
56
57   if (_mesa_type_is_packed(datatype))
58      return b;
59   else
60      return b * comps;
61}
62
63
64/**
65 * \name Support macros for do_row and do_row_3d
66 *
67 * The macro madness is here for two reasons.  First, it compacts the code
68 * slightly.  Second, it makes it much easier to adjust the specifics of the
69 * filter to tune the rounding characteristics.
70 */
71/*@{*/
72#define DECLARE_ROW_POINTERS(t, e) \
73      const t(*rowA)[e] = (const t(*)[e]) srcRowA; \
74      const t(*rowB)[e] = (const t(*)[e]) srcRowB; \
75      const t(*rowC)[e] = (const t(*)[e]) srcRowC; \
76      const t(*rowD)[e] = (const t(*)[e]) srcRowD; \
77      t(*dst)[e] = (t(*)[e]) dstRow
78
79#define DECLARE_ROW_POINTERS0(t) \
80      const t *rowA = (const t *) srcRowA; \
81      const t *rowB = (const t *) srcRowB; \
82      const t *rowC = (const t *) srcRowC; \
83      const t *rowD = (const t *) srcRowD; \
84      t *dst = (t *) dstRow
85
86#define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
87   ((unsigned) Aj + (unsigned) Ak \
88    + (unsigned) Bj + (unsigned) Bk \
89    + (unsigned) Cj + (unsigned) Ck \
90    + (unsigned) Dj + (unsigned) Dk \
91    + 4) >> 3
92
93#define FILTER_3D(e) \
94   do { \
95      dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \
96                                rowB[j][e], rowB[k][e], \
97                                rowC[j][e], rowC[k][e], \
98                                rowD[j][e], rowD[k][e]); \
99   } while(0)
100
101#define FILTER_SUM_3D_SIGNED(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
102   (Aj + Ak \
103    + Bj + Bk \
104    + Cj + Ck \
105    + Dj + Dk \
106    + 4) / 8
107
108#define FILTER_3D_SIGNED(e) \
109   do { \
110      dst[i][e] = FILTER_SUM_3D_SIGNED(rowA[j][e], rowA[k][e], \
111                                       rowB[j][e], rowB[k][e], \
112                                       rowC[j][e], rowC[k][e], \
113                                       rowD[j][e], rowD[k][e]); \
114   } while(0)
115
116#define FILTER_F_3D(e) \
117   do { \
118      dst[i][e] = (rowA[j][e] + rowA[k][e] \
119                   + rowB[j][e] + rowB[k][e] \
120                   + rowC[j][e] + rowC[k][e] \
121                   + rowD[j][e] + rowD[k][e]) * 0.125F; \
122   } while(0)
123
124#define FILTER_HF_3D(e) \
125   do { \
126      const GLfloat aj = _mesa_half_to_float(rowA[j][e]); \
127      const GLfloat ak = _mesa_half_to_float(rowA[k][e]); \
128      const GLfloat bj = _mesa_half_to_float(rowB[j][e]); \
129      const GLfloat bk = _mesa_half_to_float(rowB[k][e]); \
130      const GLfloat cj = _mesa_half_to_float(rowC[j][e]); \
131      const GLfloat ck = _mesa_half_to_float(rowC[k][e]); \
132      const GLfloat dj = _mesa_half_to_float(rowD[j][e]); \
133      const GLfloat dk = _mesa_half_to_float(rowD[k][e]); \
134      dst[i][e] = _mesa_float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
135                                      * 0.125F); \
136   } while(0)
137/*@}*/
138
139
140/**
141 * Average together two rows of a source image to produce a single new
142 * row in the dest image.  It's legal for the two source rows to point
143 * to the same data.  The source width must be equal to either the
144 * dest width or two times the dest width.
145 * \param datatype  GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
146 * \param comps  number of components per pixel (1..4)
147 */
148static void
149do_row(GLenum datatype, GLuint comps, GLint srcWidth,
150       const GLvoid *srcRowA, const GLvoid *srcRowB,
151       GLint dstWidth, GLvoid *dstRow)
152{
153   const GLuint k0 = (srcWidth == dstWidth) ? 0 : 1;
154   const GLuint colStride = (srcWidth == dstWidth) ? 1 : 2;
155
156   ASSERT(comps >= 1);
157   ASSERT(comps <= 4);
158
159   /* This assertion is no longer valid with non-power-of-2 textures
160   assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
161   */
162
163   if (datatype == GL_UNSIGNED_BYTE && comps == 4) {
164      GLuint i, j, k;
165      const GLubyte(*rowA)[4] = (const GLubyte(*)[4]) srcRowA;
166      const GLubyte(*rowB)[4] = (const GLubyte(*)[4]) srcRowB;
167      GLubyte(*dst)[4] = (GLubyte(*)[4]) dstRow;
168      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
169           i++, j += colStride, k += colStride) {
170         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
171         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
172         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
173         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
174      }
175   }
176   else if (datatype == GL_UNSIGNED_BYTE && comps == 3) {
177      GLuint i, j, k;
178      const GLubyte(*rowA)[3] = (const GLubyte(*)[3]) srcRowA;
179      const GLubyte(*rowB)[3] = (const GLubyte(*)[3]) srcRowB;
180      GLubyte(*dst)[3] = (GLubyte(*)[3]) dstRow;
181      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
182           i++, j += colStride, k += colStride) {
183         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
184         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
185         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
186      }
187   }
188   else if (datatype == GL_UNSIGNED_BYTE && comps == 2) {
189      GLuint i, j, k;
190      const GLubyte(*rowA)[2] = (const GLubyte(*)[2]) srcRowA;
191      const GLubyte(*rowB)[2] = (const GLubyte(*)[2]) srcRowB;
192      GLubyte(*dst)[2] = (GLubyte(*)[2]) dstRow;
193      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
194           i++, j += colStride, k += colStride) {
195         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
196         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
197      }
198   }
199   else if (datatype == GL_UNSIGNED_BYTE && comps == 1) {
200      GLuint i, j, k;
201      const GLubyte *rowA = (const GLubyte *) srcRowA;
202      const GLubyte *rowB = (const GLubyte *) srcRowB;
203      GLubyte *dst = (GLubyte *) dstRow;
204      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
205           i++, j += colStride, k += colStride) {
206         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
207      }
208   }
209
210   else if (datatype == GL_BYTE && comps == 4) {
211      GLuint i, j, k;
212      const GLbyte(*rowA)[4] = (const GLbyte(*)[4]) srcRowA;
213      const GLbyte(*rowB)[4] = (const GLbyte(*)[4]) srcRowB;
214      GLbyte(*dst)[4] = (GLbyte(*)[4]) dstRow;
215      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
216           i++, j += colStride, k += colStride) {
217         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
218         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
219         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
220         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
221      }
222   }
223   else if (datatype == GL_BYTE && comps == 3) {
224      GLuint i, j, k;
225      const GLbyte(*rowA)[3] = (const GLbyte(*)[3]) srcRowA;
226      const GLbyte(*rowB)[3] = (const GLbyte(*)[3]) srcRowB;
227      GLbyte(*dst)[3] = (GLbyte(*)[3]) dstRow;
228      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
229           i++, j += colStride, k += colStride) {
230         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
231         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
232         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
233      }
234   }
235   else if (datatype == GL_BYTE && comps == 2) {
236      GLuint i, j, k;
237      const GLbyte(*rowA)[2] = (const GLbyte(*)[2]) srcRowA;
238      const GLbyte(*rowB)[2] = (const GLbyte(*)[2]) srcRowB;
239      GLbyte(*dst)[2] = (GLbyte(*)[2]) dstRow;
240      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
241           i++, j += colStride, k += colStride) {
242         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
243         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
244      }
245   }
246   else if (datatype == GL_BYTE && comps == 1) {
247      GLuint i, j, k;
248      const GLbyte *rowA = (const GLbyte *) srcRowA;
249      const GLbyte *rowB = (const GLbyte *) srcRowB;
250      GLbyte *dst = (GLbyte *) dstRow;
251      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
252           i++, j += colStride, k += colStride) {
253         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
254      }
255   }
256
257   else if (datatype == GL_UNSIGNED_SHORT && comps == 4) {
258      GLuint i, j, k;
259      const GLushort(*rowA)[4] = (const GLushort(*)[4]) srcRowA;
260      const GLushort(*rowB)[4] = (const GLushort(*)[4]) srcRowB;
261      GLushort(*dst)[4] = (GLushort(*)[4]) dstRow;
262      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
263           i++, j += colStride, k += colStride) {
264         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
265         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
266         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
267         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
268      }
269   }
270   else if (datatype == GL_UNSIGNED_SHORT && comps == 3) {
271      GLuint i, j, k;
272      const GLushort(*rowA)[3] = (const GLushort(*)[3]) srcRowA;
273      const GLushort(*rowB)[3] = (const GLushort(*)[3]) srcRowB;
274      GLushort(*dst)[3] = (GLushort(*)[3]) dstRow;
275      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
276           i++, j += colStride, k += colStride) {
277         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
278         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
279         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
280      }
281   }
282   else if (datatype == GL_UNSIGNED_SHORT && comps == 2) {
283      GLuint i, j, k;
284      const GLushort(*rowA)[2] = (const GLushort(*)[2]) srcRowA;
285      const GLushort(*rowB)[2] = (const GLushort(*)[2]) srcRowB;
286      GLushort(*dst)[2] = (GLushort(*)[2]) dstRow;
287      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
288           i++, j += colStride, k += colStride) {
289         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
290         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
291      }
292   }
293   else if (datatype == GL_UNSIGNED_SHORT && comps == 1) {
294      GLuint i, j, k;
295      const GLushort *rowA = (const GLushort *) srcRowA;
296      const GLushort *rowB = (const GLushort *) srcRowB;
297      GLushort *dst = (GLushort *) dstRow;
298      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
299           i++, j += colStride, k += colStride) {
300         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
301      }
302   }
303
304   else if (datatype == GL_SHORT && comps == 4) {
305      GLuint i, j, k;
306      const GLshort(*rowA)[4] = (const GLshort(*)[4]) srcRowA;
307      const GLshort(*rowB)[4] = (const GLshort(*)[4]) srcRowB;
308      GLshort(*dst)[4] = (GLshort(*)[4]) dstRow;
309      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
310           i++, j += colStride, k += colStride) {
311         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
312         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
313         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
314         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
315      }
316   }
317   else if (datatype == GL_SHORT && comps == 3) {
318      GLuint i, j, k;
319      const GLshort(*rowA)[3] = (const GLshort(*)[3]) srcRowA;
320      const GLshort(*rowB)[3] = (const GLshort(*)[3]) srcRowB;
321      GLshort(*dst)[3] = (GLshort(*)[3]) dstRow;
322      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
323           i++, j += colStride, k += colStride) {
324         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
325         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
326         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
327      }
328   }
329   else if (datatype == GL_SHORT && comps == 2) {
330      GLuint i, j, k;
331      const GLshort(*rowA)[2] = (const GLshort(*)[2]) srcRowA;
332      const GLshort(*rowB)[2] = (const GLshort(*)[2]) srcRowB;
333      GLshort(*dst)[2] = (GLshort(*)[2]) dstRow;
334      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
335           i++, j += colStride, k += colStride) {
336         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
337         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
338      }
339   }
340   else if (datatype == GL_SHORT && comps == 1) {
341      GLuint i, j, k;
342      const GLshort *rowA = (const GLshort *) srcRowA;
343      const GLshort *rowB = (const GLshort *) srcRowB;
344      GLshort *dst = (GLshort *) dstRow;
345      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
346           i++, j += colStride, k += colStride) {
347         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
348      }
349   }
350
351   else if (datatype == GL_FLOAT && comps == 4) {
352      GLuint i, j, k;
353      const GLfloat(*rowA)[4] = (const GLfloat(*)[4]) srcRowA;
354      const GLfloat(*rowB)[4] = (const GLfloat(*)[4]) srcRowB;
355      GLfloat(*dst)[4] = (GLfloat(*)[4]) dstRow;
356      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
357           i++, j += colStride, k += colStride) {
358         dst[i][0] = (rowA[j][0] + rowA[k][0] +
359                      rowB[j][0] + rowB[k][0]) * 0.25F;
360         dst[i][1] = (rowA[j][1] + rowA[k][1] +
361                      rowB[j][1] + rowB[k][1]) * 0.25F;
362         dst[i][2] = (rowA[j][2] + rowA[k][2] +
363                      rowB[j][2] + rowB[k][2]) * 0.25F;
364         dst[i][3] = (rowA[j][3] + rowA[k][3] +
365                      rowB[j][3] + rowB[k][3]) * 0.25F;
366      }
367   }
368   else if (datatype == GL_FLOAT && comps == 3) {
369      GLuint i, j, k;
370      const GLfloat(*rowA)[3] = (const GLfloat(*)[3]) srcRowA;
371      const GLfloat(*rowB)[3] = (const GLfloat(*)[3]) srcRowB;
372      GLfloat(*dst)[3] = (GLfloat(*)[3]) dstRow;
373      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
374           i++, j += colStride, k += colStride) {
375         dst[i][0] = (rowA[j][0] + rowA[k][0] +
376                      rowB[j][0] + rowB[k][0]) * 0.25F;
377         dst[i][1] = (rowA[j][1] + rowA[k][1] +
378                      rowB[j][1] + rowB[k][1]) * 0.25F;
379         dst[i][2] = (rowA[j][2] + rowA[k][2] +
380                      rowB[j][2] + rowB[k][2]) * 0.25F;
381      }
382   }
383   else if (datatype == GL_FLOAT && comps == 2) {
384      GLuint i, j, k;
385      const GLfloat(*rowA)[2] = (const GLfloat(*)[2]) srcRowA;
386      const GLfloat(*rowB)[2] = (const GLfloat(*)[2]) srcRowB;
387      GLfloat(*dst)[2] = (GLfloat(*)[2]) dstRow;
388      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
389           i++, j += colStride, k += colStride) {
390         dst[i][0] = (rowA[j][0] + rowA[k][0] +
391                      rowB[j][0] + rowB[k][0]) * 0.25F;
392         dst[i][1] = (rowA[j][1] + rowA[k][1] +
393                      rowB[j][1] + rowB[k][1]) * 0.25F;
394      }
395   }
396   else if (datatype == GL_FLOAT && comps == 1) {
397      GLuint i, j, k;
398      const GLfloat *rowA = (const GLfloat *) srcRowA;
399      const GLfloat *rowB = (const GLfloat *) srcRowB;
400      GLfloat *dst = (GLfloat *) dstRow;
401      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
402           i++, j += colStride, k += colStride) {
403         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
404      }
405   }
406
407   else if (datatype == GL_HALF_FLOAT_ARB && comps == 4) {
408      GLuint i, j, k, comp;
409      const GLhalfARB(*rowA)[4] = (const GLhalfARB(*)[4]) srcRowA;
410      const GLhalfARB(*rowB)[4] = (const GLhalfARB(*)[4]) srcRowB;
411      GLhalfARB(*dst)[4] = (GLhalfARB(*)[4]) dstRow;
412      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
413           i++, j += colStride, k += colStride) {
414         for (comp = 0; comp < 4; comp++) {
415            GLfloat aj, ak, bj, bk;
416            aj = _mesa_half_to_float(rowA[j][comp]);
417            ak = _mesa_half_to_float(rowA[k][comp]);
418            bj = _mesa_half_to_float(rowB[j][comp]);
419            bk = _mesa_half_to_float(rowB[k][comp]);
420            dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
421         }
422      }
423   }
424   else if (datatype == GL_HALF_FLOAT_ARB && comps == 3) {
425      GLuint i, j, k, comp;
426      const GLhalfARB(*rowA)[3] = (const GLhalfARB(*)[3]) srcRowA;
427      const GLhalfARB(*rowB)[3] = (const GLhalfARB(*)[3]) srcRowB;
428      GLhalfARB(*dst)[3] = (GLhalfARB(*)[3]) dstRow;
429      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
430           i++, j += colStride, k += colStride) {
431         for (comp = 0; comp < 3; comp++) {
432            GLfloat aj, ak, bj, bk;
433            aj = _mesa_half_to_float(rowA[j][comp]);
434            ak = _mesa_half_to_float(rowA[k][comp]);
435            bj = _mesa_half_to_float(rowB[j][comp]);
436            bk = _mesa_half_to_float(rowB[k][comp]);
437            dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
438         }
439      }
440   }
441   else if (datatype == GL_HALF_FLOAT_ARB && comps == 2) {
442      GLuint i, j, k, comp;
443      const GLhalfARB(*rowA)[2] = (const GLhalfARB(*)[2]) srcRowA;
444      const GLhalfARB(*rowB)[2] = (const GLhalfARB(*)[2]) srcRowB;
445      GLhalfARB(*dst)[2] = (GLhalfARB(*)[2]) dstRow;
446      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
447           i++, j += colStride, k += colStride) {
448         for (comp = 0; comp < 2; comp++) {
449            GLfloat aj, ak, bj, bk;
450            aj = _mesa_half_to_float(rowA[j][comp]);
451            ak = _mesa_half_to_float(rowA[k][comp]);
452            bj = _mesa_half_to_float(rowB[j][comp]);
453            bk = _mesa_half_to_float(rowB[k][comp]);
454            dst[i][comp] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
455         }
456      }
457   }
458   else if (datatype == GL_HALF_FLOAT_ARB && comps == 1) {
459      GLuint i, j, k;
460      const GLhalfARB *rowA = (const GLhalfARB *) srcRowA;
461      const GLhalfARB *rowB = (const GLhalfARB *) srcRowB;
462      GLhalfARB *dst = (GLhalfARB *) dstRow;
463      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
464           i++, j += colStride, k += colStride) {
465         GLfloat aj, ak, bj, bk;
466         aj = _mesa_half_to_float(rowA[j]);
467         ak = _mesa_half_to_float(rowA[k]);
468         bj = _mesa_half_to_float(rowB[j]);
469         bk = _mesa_half_to_float(rowB[k]);
470         dst[i] = _mesa_float_to_half((aj + ak + bj + bk) * 0.25F);
471      }
472   }
473
474   else if (datatype == GL_UNSIGNED_INT && comps == 1) {
475      GLuint i, j, k;
476      const GLuint *rowA = (const GLuint *) srcRowA;
477      const GLuint *rowB = (const GLuint *) srcRowB;
478      GLuint *dst = (GLuint *) dstRow;
479      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
480           i++, j += colStride, k += colStride) {
481         dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
482      }
483   }
484
485   else if (datatype == GL_UNSIGNED_SHORT_5_6_5 && comps == 3) {
486      GLuint i, j, k;
487      const GLushort *rowA = (const GLushort *) srcRowA;
488      const GLushort *rowB = (const GLushort *) srcRowB;
489      GLushort *dst = (GLushort *) dstRow;
490      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
491           i++, j += colStride, k += colStride) {
492         const GLint rowAr0 = rowA[j] & 0x1f;
493         const GLint rowAr1 = rowA[k] & 0x1f;
494         const GLint rowBr0 = rowB[j] & 0x1f;
495         const GLint rowBr1 = rowB[k] & 0x1f;
496         const GLint rowAg0 = (rowA[j] >> 5) & 0x3f;
497         const GLint rowAg1 = (rowA[k] >> 5) & 0x3f;
498         const GLint rowBg0 = (rowB[j] >> 5) & 0x3f;
499         const GLint rowBg1 = (rowB[k] >> 5) & 0x3f;
500         const GLint rowAb0 = (rowA[j] >> 11) & 0x1f;
501         const GLint rowAb1 = (rowA[k] >> 11) & 0x1f;
502         const GLint rowBb0 = (rowB[j] >> 11) & 0x1f;
503         const GLint rowBb1 = (rowB[k] >> 11) & 0x1f;
504         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
505         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
506         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
507         dst[i] = (blue << 11) | (green << 5) | red;
508      }
509   }
510   else if (datatype == GL_UNSIGNED_SHORT_4_4_4_4 && comps == 4) {
511      GLuint i, j, k;
512      const GLushort *rowA = (const GLushort *) srcRowA;
513      const GLushort *rowB = (const GLushort *) srcRowB;
514      GLushort *dst = (GLushort *) dstRow;
515      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
516           i++, j += colStride, k += colStride) {
517         const GLint rowAr0 = rowA[j] & 0xf;
518         const GLint rowAr1 = rowA[k] & 0xf;
519         const GLint rowBr0 = rowB[j] & 0xf;
520         const GLint rowBr1 = rowB[k] & 0xf;
521         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
522         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
523         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
524         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
525         const GLint rowAb0 = (rowA[j] >> 8) & 0xf;
526         const GLint rowAb1 = (rowA[k] >> 8) & 0xf;
527         const GLint rowBb0 = (rowB[j] >> 8) & 0xf;
528         const GLint rowBb1 = (rowB[k] >> 8) & 0xf;
529         const GLint rowAa0 = (rowA[j] >> 12) & 0xf;
530         const GLint rowAa1 = (rowA[k] >> 12) & 0xf;
531         const GLint rowBa0 = (rowB[j] >> 12) & 0xf;
532         const GLint rowBa1 = (rowB[k] >> 12) & 0xf;
533         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
534         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
535         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
536         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
537         dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
538      }
539   }
540   else if (datatype == GL_UNSIGNED_SHORT_1_5_5_5_REV && comps == 4) {
541      GLuint i, j, k;
542      const GLushort *rowA = (const GLushort *) srcRowA;
543      const GLushort *rowB = (const GLushort *) srcRowB;
544      GLushort *dst = (GLushort *) dstRow;
545      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
546           i++, j += colStride, k += colStride) {
547         const GLint rowAr0 = rowA[j] & 0x1f;
548         const GLint rowAr1 = rowA[k] & 0x1f;
549         const GLint rowBr0 = rowB[j] & 0x1f;
550         const GLint rowBr1 = rowB[k] & 0x1f;
551         const GLint rowAg0 = (rowA[j] >> 5) & 0x1f;
552         const GLint rowAg1 = (rowA[k] >> 5) & 0x1f;
553         const GLint rowBg0 = (rowB[j] >> 5) & 0x1f;
554         const GLint rowBg1 = (rowB[k] >> 5) & 0x1f;
555         const GLint rowAb0 = (rowA[j] >> 10) & 0x1f;
556         const GLint rowAb1 = (rowA[k] >> 10) & 0x1f;
557         const GLint rowBb0 = (rowB[j] >> 10) & 0x1f;
558         const GLint rowBb1 = (rowB[k] >> 10) & 0x1f;
559         const GLint rowAa0 = (rowA[j] >> 15) & 0x1;
560         const GLint rowAa1 = (rowA[k] >> 15) & 0x1;
561         const GLint rowBa0 = (rowB[j] >> 15) & 0x1;
562         const GLint rowBa1 = (rowB[k] >> 15) & 0x1;
563         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
564         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
565         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
566         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
567         dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
568      }
569   }
570   else if (datatype == GL_UNSIGNED_SHORT_5_5_5_1 && comps == 4) {
571      GLuint i, j, k;
572      const GLushort *rowA = (const GLushort *) srcRowA;
573      const GLushort *rowB = (const GLushort *) srcRowB;
574      GLushort *dst = (GLushort *) dstRow;
575      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
576           i++, j += colStride, k += colStride) {
577         const GLint rowAr0 = (rowA[j] >> 11) & 0x1f;
578         const GLint rowAr1 = (rowA[k] >> 11) & 0x1f;
579         const GLint rowBr0 = (rowB[j] >> 11) & 0x1f;
580         const GLint rowBr1 = (rowB[k] >> 11) & 0x1f;
581         const GLint rowAg0 = (rowA[j] >> 6) & 0x1f;
582         const GLint rowAg1 = (rowA[k] >> 6) & 0x1f;
583         const GLint rowBg0 = (rowB[j] >> 6) & 0x1f;
584         const GLint rowBg1 = (rowB[k] >> 6) & 0x1f;
585         const GLint rowAb0 = (rowA[j] >> 1) & 0x1f;
586         const GLint rowAb1 = (rowA[k] >> 1) & 0x1f;
587         const GLint rowBb0 = (rowB[j] >> 1) & 0x1f;
588         const GLint rowBb1 = (rowB[k] >> 1) & 0x1f;
589         const GLint rowAa0 = (rowA[j] & 0x1);
590         const GLint rowAa1 = (rowA[k] & 0x1);
591         const GLint rowBa0 = (rowB[j] & 0x1);
592         const GLint rowBa1 = (rowB[k] & 0x1);
593         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
594         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
595         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
596         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
597         dst[i] = (red << 11) | (green << 6) | (blue << 1) | alpha;
598      }
599   }
600
601   else if (datatype == GL_UNSIGNED_BYTE_3_3_2 && comps == 3) {
602      GLuint i, j, k;
603      const GLubyte *rowA = (const GLubyte *) srcRowA;
604      const GLubyte *rowB = (const GLubyte *) srcRowB;
605      GLubyte *dst = (GLubyte *) dstRow;
606      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
607           i++, j += colStride, k += colStride) {
608         const GLint rowAr0 = rowA[j] & 0x3;
609         const GLint rowAr1 = rowA[k] & 0x3;
610         const GLint rowBr0 = rowB[j] & 0x3;
611         const GLint rowBr1 = rowB[k] & 0x3;
612         const GLint rowAg0 = (rowA[j] >> 2) & 0x7;
613         const GLint rowAg1 = (rowA[k] >> 2) & 0x7;
614         const GLint rowBg0 = (rowB[j] >> 2) & 0x7;
615         const GLint rowBg1 = (rowB[k] >> 2) & 0x7;
616         const GLint rowAb0 = (rowA[j] >> 5) & 0x7;
617         const GLint rowAb1 = (rowA[k] >> 5) & 0x7;
618         const GLint rowBb0 = (rowB[j] >> 5) & 0x7;
619         const GLint rowBb1 = (rowB[k] >> 5) & 0x7;
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         dst[i] = (blue << 5) | (green << 2) | red;
624      }
625   }
626
627   else if (datatype == MESA_UNSIGNED_BYTE_4_4 && comps == 2) {
628      GLuint i, j, k;
629      const GLubyte *rowA = (const GLubyte *) srcRowA;
630      const GLubyte *rowB = (const GLubyte *) srcRowB;
631      GLubyte *dst = (GLubyte *) dstRow;
632      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
633           i++, j += colStride, k += colStride) {
634         const GLint rowAr0 = rowA[j] & 0xf;
635         const GLint rowAr1 = rowA[k] & 0xf;
636         const GLint rowBr0 = rowB[j] & 0xf;
637         const GLint rowBr1 = rowB[k] & 0xf;
638         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
639         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
640         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
641         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
642         const GLint r = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
643         const GLint g = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
644         dst[i] = (g << 4) | r;
645      }
646   }
647
648   else if (datatype == GL_UNSIGNED_INT_2_10_10_10_REV && comps == 4) {
649      GLuint i, j, k;
650      const GLuint *rowA = (const GLuint *) srcRowA;
651      const GLuint *rowB = (const GLuint *) srcRowB;
652      GLuint *dst = (GLuint *) dstRow;
653      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
654           i++, j += colStride, k += colStride) {
655         const GLint rowAr0 = rowA[j] & 0x3ff;
656         const GLint rowAr1 = rowA[k] & 0x3ff;
657         const GLint rowBr0 = rowB[j] & 0x3ff;
658         const GLint rowBr1 = rowB[k] & 0x3ff;
659         const GLint rowAg0 = (rowA[j] >> 10) & 0x3ff;
660         const GLint rowAg1 = (rowA[k] >> 10) & 0x3ff;
661         const GLint rowBg0 = (rowB[j] >> 10) & 0x3ff;
662         const GLint rowBg1 = (rowB[k] >> 10) & 0x3ff;
663         const GLint rowAb0 = (rowA[j] >> 20) & 0x3ff;
664         const GLint rowAb1 = (rowA[k] >> 20) & 0x3ff;
665         const GLint rowBb0 = (rowB[j] >> 20) & 0x3ff;
666         const GLint rowBb1 = (rowB[k] >> 20) & 0x3ff;
667         const GLint rowAa0 = (rowA[j] >> 30) & 0x3;
668         const GLint rowAa1 = (rowA[k] >> 30) & 0x3;
669         const GLint rowBa0 = (rowB[j] >> 30) & 0x3;
670         const GLint rowBa1 = (rowB[k] >> 30) & 0x3;
671         const GLint red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
672         const GLint green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
673         const GLint blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
674         const GLint alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
675         dst[i] = (alpha << 30) | (blue << 20) | (green << 10) | red;
676      }
677   }
678
679   else if (datatype == GL_UNSIGNED_INT_5_9_9_9_REV && comps == 3) {
680      GLuint i, j, k;
681      const GLuint *rowA = (const GLuint*) srcRowA;
682      const GLuint *rowB = (const GLuint*) srcRowB;
683      GLuint *dst = (GLuint*)dstRow;
684      GLfloat res[3], rowAj[3], rowBj[3], rowAk[3], rowBk[3];
685      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
686           i++, j += colStride, k += colStride) {
687         rgb9e5_to_float3(rowA[j], rowAj);
688         rgb9e5_to_float3(rowB[j], rowBj);
689         rgb9e5_to_float3(rowA[k], rowAk);
690         rgb9e5_to_float3(rowB[k], rowBk);
691         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0]) * 0.25F;
692         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1]) * 0.25F;
693         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2]) * 0.25F;
694         dst[i] = float3_to_rgb9e5(res);
695      }
696   }
697
698   else if (datatype == GL_UNSIGNED_INT_10F_11F_11F_REV && comps == 3) {
699      GLuint i, j, k;
700      const GLuint *rowA = (const GLuint*) srcRowA;
701      const GLuint *rowB = (const GLuint*) srcRowB;
702      GLuint *dst = (GLuint*)dstRow;
703      GLfloat res[3], rowAj[3], rowBj[3], rowAk[3], rowBk[3];
704      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
705           i++, j += colStride, k += colStride) {
706         r11g11b10f_to_float3(rowA[j], rowAj);
707         r11g11b10f_to_float3(rowB[j], rowBj);
708         r11g11b10f_to_float3(rowA[k], rowAk);
709         r11g11b10f_to_float3(rowB[k], rowBk);
710         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0]) * 0.25F;
711         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1]) * 0.25F;
712         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2]) * 0.25F;
713         dst[i] = float3_to_r11g11b10f(res);
714      }
715   }
716
717   else if (datatype == GL_FLOAT_32_UNSIGNED_INT_24_8_REV && comps == 1) {
718      GLuint i, j, k;
719      const GLfloat *rowA = (const GLfloat *) srcRowA;
720      const GLfloat *rowB = (const GLfloat *) srcRowB;
721      GLfloat *dst = (GLfloat *) dstRow;
722      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
723           i++, j += colStride, k += colStride) {
724         dst[i*2] = (rowA[j*2] + rowA[k*2] + rowB[j*2] + rowB[k*2]) * 0.25F;
725      }
726   }
727
728   else if (datatype == GL_UNSIGNED_INT_24_8_MESA && comps == 2) {
729      GLuint i, j, k;
730      const GLuint *rowA = (const GLuint *) srcRowA;
731      const GLuint *rowB = (const GLuint *) srcRowB;
732      GLuint *dst = (GLuint *) dstRow;
733      /* note: averaging stencil values seems weird, but what else? */
734      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
735           i++, j += colStride, k += colStride) {
736         GLuint z = (((rowA[j] >> 8) + (rowA[k] >> 8) +
737                      (rowB[j] >> 8) + (rowB[k] >> 8)) / 4) << 8;
738         GLuint s = ((rowA[j] & 0xff) + (rowA[k] & 0xff) +
739                     (rowB[j] & 0xff) + (rowB[k] & 0xff)) / 4;
740         dst[i] = z | s;
741      }
742   }
743   else if (datatype == GL_UNSIGNED_INT_8_24_REV_MESA && comps == 2) {
744      GLuint i, j, k;
745      const GLuint *rowA = (const GLuint *) srcRowA;
746      const GLuint *rowB = (const GLuint *) srcRowB;
747      GLuint *dst = (GLuint *) dstRow;
748      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
749           i++, j += colStride, k += colStride) {
750         GLuint z = ((rowA[j] & 0xffffff) + (rowA[k] & 0xffffff) +
751                     (rowB[j] & 0xffffff) + (rowB[k] & 0xffffff)) / 4;
752         GLuint s = (((rowA[j] >> 24) + (rowA[k] >> 24) +
753                      (rowB[j] >> 24) + (rowB[k] >> 24)) / 4) << 24;
754         dst[i] = z | s;
755      }
756   }
757
758   else {
759      _mesa_problem(NULL, "bad format in do_row()");
760   }
761}
762
763
764/**
765 * Average together four rows of a source image to produce a single new
766 * row in the dest image.  It's legal for the two source rows to point
767 * to the same data.  The source width must be equal to either the
768 * dest width or two times the dest width.
769 *
770 * \param datatype  GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
771 *                  \c GL_FLOAT, etc.
772 * \param comps     number of components per pixel (1..4)
773 * \param srcWidth  Width of a row in the source data
774 * \param srcRowA   Pointer to one of the rows of source data
775 * \param srcRowB   Pointer to one of the rows of source data
776 * \param srcRowC   Pointer to one of the rows of source data
777 * \param srcRowD   Pointer to one of the rows of source data
778 * \param dstWidth  Width of a row in the destination data
779 * \param srcRowA   Pointer to the row of destination data
780 */
781static void
782do_row_3D(GLenum datatype, GLuint comps, GLint srcWidth,
783          const GLvoid *srcRowA, const GLvoid *srcRowB,
784          const GLvoid *srcRowC, const GLvoid *srcRowD,
785          GLint dstWidth, GLvoid *dstRow)
786{
787   const GLuint k0 = (srcWidth == dstWidth) ? 0 : 1;
788   const GLuint colStride = (srcWidth == dstWidth) ? 1 : 2;
789   GLuint i, j, k;
790
791   ASSERT(comps >= 1);
792   ASSERT(comps <= 4);
793
794   if ((datatype == GL_UNSIGNED_BYTE) && (comps == 4)) {
795      DECLARE_ROW_POINTERS(GLubyte, 4);
796
797      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
798           i++, j += colStride, k += colStride) {
799         FILTER_3D(0);
800         FILTER_3D(1);
801         FILTER_3D(2);
802         FILTER_3D(3);
803      }
804   }
805   else if ((datatype == GL_UNSIGNED_BYTE) && (comps == 3)) {
806      DECLARE_ROW_POINTERS(GLubyte, 3);
807
808      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
809           i++, j += colStride, k += colStride) {
810         FILTER_3D(0);
811         FILTER_3D(1);
812         FILTER_3D(2);
813      }
814   }
815   else if ((datatype == GL_UNSIGNED_BYTE) && (comps == 2)) {
816      DECLARE_ROW_POINTERS(GLubyte, 2);
817
818      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
819           i++, j += colStride, k += colStride) {
820         FILTER_3D(0);
821         FILTER_3D(1);
822      }
823   }
824   else if ((datatype == GL_UNSIGNED_BYTE) && (comps == 1)) {
825      DECLARE_ROW_POINTERS(GLubyte, 1);
826
827      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
828           i++, j += colStride, k += colStride) {
829         FILTER_3D(0);
830      }
831   }
832   else if ((datatype == GL_BYTE) && (comps == 4)) {
833      DECLARE_ROW_POINTERS(GLbyte, 4);
834
835      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
836           i++, j += colStride, k += colStride) {
837         FILTER_3D_SIGNED(0);
838         FILTER_3D_SIGNED(1);
839         FILTER_3D_SIGNED(2);
840         FILTER_3D_SIGNED(3);
841      }
842   }
843   else if ((datatype == GL_BYTE) && (comps == 3)) {
844      DECLARE_ROW_POINTERS(GLbyte, 3);
845
846      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
847           i++, j += colStride, k += colStride) {
848         FILTER_3D_SIGNED(0);
849         FILTER_3D_SIGNED(1);
850         FILTER_3D_SIGNED(2);
851      }
852   }
853   else if ((datatype == GL_BYTE) && (comps == 2)) {
854      DECLARE_ROW_POINTERS(GLbyte, 2);
855
856      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
857           i++, j += colStride, k += colStride) {
858         FILTER_3D_SIGNED(0);
859         FILTER_3D_SIGNED(1);
860       }
861   }
862   else if ((datatype == GL_BYTE) && (comps == 1)) {
863      DECLARE_ROW_POINTERS(GLbyte, 1);
864
865      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
866           i++, j += colStride, k += colStride) {
867         FILTER_3D_SIGNED(0);
868      }
869   }
870   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 4)) {
871      DECLARE_ROW_POINTERS(GLushort, 4);
872
873      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
874           i++, j += colStride, k += colStride) {
875         FILTER_3D(0);
876         FILTER_3D(1);
877         FILTER_3D(2);
878         FILTER_3D(3);
879      }
880   }
881   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 3)) {
882      DECLARE_ROW_POINTERS(GLushort, 3);
883
884      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
885           i++, j += colStride, k += colStride) {
886         FILTER_3D(0);
887         FILTER_3D(1);
888         FILTER_3D(2);
889      }
890   }
891   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 2)) {
892      DECLARE_ROW_POINTERS(GLushort, 2);
893
894      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
895           i++, j += colStride, k += colStride) {
896         FILTER_3D(0);
897         FILTER_3D(1);
898      }
899   }
900   else if ((datatype == GL_UNSIGNED_SHORT) && (comps == 1)) {
901      DECLARE_ROW_POINTERS(GLushort, 1);
902
903      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
904           i++, j += colStride, k += colStride) {
905         FILTER_3D(0);
906      }
907   }
908   else if ((datatype == GL_SHORT) && (comps == 4)) {
909      DECLARE_ROW_POINTERS(GLshort, 4);
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         FILTER_3D(3);
917      }
918   }
919   else if ((datatype == GL_SHORT) && (comps == 3)) {
920      DECLARE_ROW_POINTERS(GLshort, 3);
921
922      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
923           i++, j += colStride, k += colStride) {
924         FILTER_3D(0);
925         FILTER_3D(1);
926         FILTER_3D(2);
927      }
928   }
929   else if ((datatype == GL_SHORT) && (comps == 2)) {
930      DECLARE_ROW_POINTERS(GLshort, 2);
931
932      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
933           i++, j += colStride, k += colStride) {
934         FILTER_3D(0);
935         FILTER_3D(1);
936      }
937   }
938   else if ((datatype == GL_SHORT) && (comps == 1)) {
939      DECLARE_ROW_POINTERS(GLshort, 1);
940
941      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
942           i++, j += colStride, k += colStride) {
943         FILTER_3D(0);
944      }
945   }
946   else if ((datatype == GL_FLOAT) && (comps == 4)) {
947      DECLARE_ROW_POINTERS(GLfloat, 4);
948
949      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
950           i++, j += colStride, k += colStride) {
951         FILTER_F_3D(0);
952         FILTER_F_3D(1);
953         FILTER_F_3D(2);
954         FILTER_F_3D(3);
955      }
956   }
957   else if ((datatype == GL_FLOAT) && (comps == 3)) {
958      DECLARE_ROW_POINTERS(GLfloat, 3);
959
960      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
961           i++, j += colStride, k += colStride) {
962         FILTER_F_3D(0);
963         FILTER_F_3D(1);
964         FILTER_F_3D(2);
965      }
966   }
967   else if ((datatype == GL_FLOAT) && (comps == 2)) {
968      DECLARE_ROW_POINTERS(GLfloat, 2);
969
970      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
971           i++, j += colStride, k += colStride) {
972         FILTER_F_3D(0);
973         FILTER_F_3D(1);
974      }
975   }
976   else if ((datatype == GL_FLOAT) && (comps == 1)) {
977      DECLARE_ROW_POINTERS(GLfloat, 1);
978
979      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
980           i++, j += colStride, k += colStride) {
981         FILTER_F_3D(0);
982      }
983   }
984   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 4)) {
985      DECLARE_ROW_POINTERS(GLhalfARB, 4);
986
987      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
988           i++, j += colStride, k += colStride) {
989         FILTER_HF_3D(0);
990         FILTER_HF_3D(1);
991         FILTER_HF_3D(2);
992         FILTER_HF_3D(3);
993      }
994   }
995   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 3)) {
996      DECLARE_ROW_POINTERS(GLhalfARB, 3);
997
998      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
999           i++, j += colStride, k += colStride) {
1000         FILTER_HF_3D(0);
1001         FILTER_HF_3D(1);
1002         FILTER_HF_3D(2);
1003      }
1004   }
1005   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 2)) {
1006      DECLARE_ROW_POINTERS(GLhalfARB, 2);
1007
1008      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1009           i++, j += colStride, k += colStride) {
1010         FILTER_HF_3D(0);
1011         FILTER_HF_3D(1);
1012      }
1013   }
1014   else if ((datatype == GL_HALF_FLOAT_ARB) && (comps == 1)) {
1015      DECLARE_ROW_POINTERS(GLhalfARB, 1);
1016
1017      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1018           i++, j += colStride, k += colStride) {
1019         FILTER_HF_3D(0);
1020      }
1021   }
1022   else if ((datatype == GL_UNSIGNED_INT) && (comps == 1)) {
1023      const GLuint *rowA = (const GLuint *) srcRowA;
1024      const GLuint *rowB = (const GLuint *) srcRowB;
1025      const GLuint *rowC = (const GLuint *) srcRowC;
1026      const GLuint *rowD = (const GLuint *) srcRowD;
1027      GLfloat *dst = (GLfloat *) dstRow;
1028
1029      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1030           i++, j += colStride, k += colStride) {
1031         const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k])
1032                               + ((uint64_t) rowB[j] + (uint64_t) rowB[k])
1033                               + ((uint64_t) rowC[j] + (uint64_t) rowC[k])
1034                               + ((uint64_t) rowD[j] + (uint64_t) rowD[k]));
1035         dst[i] = (GLfloat)((double) tmp * 0.125);
1036      }
1037   }
1038   else if ((datatype == GL_UNSIGNED_SHORT_5_6_5) && (comps == 3)) {
1039      DECLARE_ROW_POINTERS0(GLushort);
1040
1041      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1042           i++, j += colStride, k += colStride) {
1043         const GLint rowAr0 = rowA[j] & 0x1f;
1044         const GLint rowAr1 = rowA[k] & 0x1f;
1045         const GLint rowBr0 = rowB[j] & 0x1f;
1046         const GLint rowBr1 = rowB[k] & 0x1f;
1047         const GLint rowCr0 = rowC[j] & 0x1f;
1048         const GLint rowCr1 = rowC[k] & 0x1f;
1049         const GLint rowDr0 = rowD[j] & 0x1f;
1050         const GLint rowDr1 = rowD[k] & 0x1f;
1051         const GLint rowAg0 = (rowA[j] >> 5) & 0x3f;
1052         const GLint rowAg1 = (rowA[k] >> 5) & 0x3f;
1053         const GLint rowBg0 = (rowB[j] >> 5) & 0x3f;
1054         const GLint rowBg1 = (rowB[k] >> 5) & 0x3f;
1055         const GLint rowCg0 = (rowC[j] >> 5) & 0x3f;
1056         const GLint rowCg1 = (rowC[k] >> 5) & 0x3f;
1057         const GLint rowDg0 = (rowD[j] >> 5) & 0x3f;
1058         const GLint rowDg1 = (rowD[k] >> 5) & 0x3f;
1059         const GLint rowAb0 = (rowA[j] >> 11) & 0x1f;
1060         const GLint rowAb1 = (rowA[k] >> 11) & 0x1f;
1061         const GLint rowBb0 = (rowB[j] >> 11) & 0x1f;
1062         const GLint rowBb1 = (rowB[k] >> 11) & 0x1f;
1063         const GLint rowCb0 = (rowC[j] >> 11) & 0x1f;
1064         const GLint rowCb1 = (rowC[k] >> 11) & 0x1f;
1065         const GLint rowDb0 = (rowD[j] >> 11) & 0x1f;
1066         const GLint rowDb1 = (rowD[k] >> 11) & 0x1f;
1067         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1068                                       rowCr0, rowCr1, rowDr0, rowDr1);
1069         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1070                                       rowCg0, rowCg1, rowDg0, rowDg1);
1071         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1072                                       rowCb0, rowCb1, rowDb0, rowDb1);
1073         dst[i] = (b << 11) | (g << 5) | r;
1074      }
1075   }
1076   else if ((datatype == GL_UNSIGNED_SHORT_4_4_4_4) && (comps == 4)) {
1077      DECLARE_ROW_POINTERS0(GLushort);
1078
1079      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1080           i++, j += colStride, k += colStride) {
1081         const GLint rowAr0 = rowA[j] & 0xf;
1082         const GLint rowAr1 = rowA[k] & 0xf;
1083         const GLint rowBr0 = rowB[j] & 0xf;
1084         const GLint rowBr1 = rowB[k] & 0xf;
1085         const GLint rowCr0 = rowC[j] & 0xf;
1086         const GLint rowCr1 = rowC[k] & 0xf;
1087         const GLint rowDr0 = rowD[j] & 0xf;
1088         const GLint rowDr1 = rowD[k] & 0xf;
1089         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
1090         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
1091         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
1092         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
1093         const GLint rowCg0 = (rowC[j] >> 4) & 0xf;
1094         const GLint rowCg1 = (rowC[k] >> 4) & 0xf;
1095         const GLint rowDg0 = (rowD[j] >> 4) & 0xf;
1096         const GLint rowDg1 = (rowD[k] >> 4) & 0xf;
1097         const GLint rowAb0 = (rowA[j] >> 8) & 0xf;
1098         const GLint rowAb1 = (rowA[k] >> 8) & 0xf;
1099         const GLint rowBb0 = (rowB[j] >> 8) & 0xf;
1100         const GLint rowBb1 = (rowB[k] >> 8) & 0xf;
1101         const GLint rowCb0 = (rowC[j] >> 8) & 0xf;
1102         const GLint rowCb1 = (rowC[k] >> 8) & 0xf;
1103         const GLint rowDb0 = (rowD[j] >> 8) & 0xf;
1104         const GLint rowDb1 = (rowD[k] >> 8) & 0xf;
1105         const GLint rowAa0 = (rowA[j] >> 12) & 0xf;
1106         const GLint rowAa1 = (rowA[k] >> 12) & 0xf;
1107         const GLint rowBa0 = (rowB[j] >> 12) & 0xf;
1108         const GLint rowBa1 = (rowB[k] >> 12) & 0xf;
1109         const GLint rowCa0 = (rowC[j] >> 12) & 0xf;
1110         const GLint rowCa1 = (rowC[k] >> 12) & 0xf;
1111         const GLint rowDa0 = (rowD[j] >> 12) & 0xf;
1112         const GLint rowDa1 = (rowD[k] >> 12) & 0xf;
1113         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1114                                       rowCr0, rowCr1, rowDr0, rowDr1);
1115         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1116                                       rowCg0, rowCg1, rowDg0, rowDg1);
1117         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1118                                       rowCb0, rowCb1, rowDb0, rowDb1);
1119         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1120                                       rowCa0, rowCa1, rowDa0, rowDa1);
1121
1122         dst[i] = (a << 12) | (b << 8) | (g << 4) | r;
1123      }
1124   }
1125   else if ((datatype == GL_UNSIGNED_SHORT_1_5_5_5_REV) && (comps == 4)) {
1126      DECLARE_ROW_POINTERS0(GLushort);
1127
1128      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1129           i++, j += colStride, k += colStride) {
1130         const GLint rowAr0 = rowA[j] & 0x1f;
1131         const GLint rowAr1 = rowA[k] & 0x1f;
1132         const GLint rowBr0 = rowB[j] & 0x1f;
1133         const GLint rowBr1 = rowB[k] & 0x1f;
1134         const GLint rowCr0 = rowC[j] & 0x1f;
1135         const GLint rowCr1 = rowC[k] & 0x1f;
1136         const GLint rowDr0 = rowD[j] & 0x1f;
1137         const GLint rowDr1 = rowD[k] & 0x1f;
1138         const GLint rowAg0 = (rowA[j] >> 5) & 0x1f;
1139         const GLint rowAg1 = (rowA[k] >> 5) & 0x1f;
1140         const GLint rowBg0 = (rowB[j] >> 5) & 0x1f;
1141         const GLint rowBg1 = (rowB[k] >> 5) & 0x1f;
1142         const GLint rowCg0 = (rowC[j] >> 5) & 0x1f;
1143         const GLint rowCg1 = (rowC[k] >> 5) & 0x1f;
1144         const GLint rowDg0 = (rowD[j] >> 5) & 0x1f;
1145         const GLint rowDg1 = (rowD[k] >> 5) & 0x1f;
1146         const GLint rowAb0 = (rowA[j] >> 10) & 0x1f;
1147         const GLint rowAb1 = (rowA[k] >> 10) & 0x1f;
1148         const GLint rowBb0 = (rowB[j] >> 10) & 0x1f;
1149         const GLint rowBb1 = (rowB[k] >> 10) & 0x1f;
1150         const GLint rowCb0 = (rowC[j] >> 10) & 0x1f;
1151         const GLint rowCb1 = (rowC[k] >> 10) & 0x1f;
1152         const GLint rowDb0 = (rowD[j] >> 10) & 0x1f;
1153         const GLint rowDb1 = (rowD[k] >> 10) & 0x1f;
1154         const GLint rowAa0 = (rowA[j] >> 15) & 0x1;
1155         const GLint rowAa1 = (rowA[k] >> 15) & 0x1;
1156         const GLint rowBa0 = (rowB[j] >> 15) & 0x1;
1157         const GLint rowBa1 = (rowB[k] >> 15) & 0x1;
1158         const GLint rowCa0 = (rowC[j] >> 15) & 0x1;
1159         const GLint rowCa1 = (rowC[k] >> 15) & 0x1;
1160         const GLint rowDa0 = (rowD[j] >> 15) & 0x1;
1161         const GLint rowDa1 = (rowD[k] >> 15) & 0x1;
1162         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1163                                       rowCr0, rowCr1, rowDr0, rowDr1);
1164         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1165                                       rowCg0, rowCg1, rowDg0, rowDg1);
1166         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1167                                       rowCb0, rowCb1, rowDb0, rowDb1);
1168         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1169                                       rowCa0, rowCa1, rowDa0, rowDa1);
1170
1171         dst[i] = (a << 15) | (b << 10) | (g << 5) | r;
1172      }
1173   }
1174   else if ((datatype == GL_UNSIGNED_SHORT_5_5_5_1) && (comps == 4)) {
1175      DECLARE_ROW_POINTERS0(GLushort);
1176
1177      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1178           i++, j += colStride, k += colStride) {
1179         const GLint rowAr0 = (rowA[j] >> 11) & 0x1f;
1180         const GLint rowAr1 = (rowA[k] >> 11) & 0x1f;
1181         const GLint rowBr0 = (rowB[j] >> 11) & 0x1f;
1182         const GLint rowBr1 = (rowB[k] >> 11) & 0x1f;
1183         const GLint rowCr0 = (rowC[j] >> 11) & 0x1f;
1184         const GLint rowCr1 = (rowC[k] >> 11) & 0x1f;
1185         const GLint rowDr0 = (rowD[j] >> 11) & 0x1f;
1186         const GLint rowDr1 = (rowD[k] >> 11) & 0x1f;
1187         const GLint rowAg0 = (rowA[j] >> 6) & 0x1f;
1188         const GLint rowAg1 = (rowA[k] >> 6) & 0x1f;
1189         const GLint rowBg0 = (rowB[j] >> 6) & 0x1f;
1190         const GLint rowBg1 = (rowB[k] >> 6) & 0x1f;
1191         const GLint rowCg0 = (rowC[j] >> 6) & 0x1f;
1192         const GLint rowCg1 = (rowC[k] >> 6) & 0x1f;
1193         const GLint rowDg0 = (rowD[j] >> 6) & 0x1f;
1194         const GLint rowDg1 = (rowD[k] >> 6) & 0x1f;
1195         const GLint rowAb0 = (rowA[j] >> 1) & 0x1f;
1196         const GLint rowAb1 = (rowA[k] >> 1) & 0x1f;
1197         const GLint rowBb0 = (rowB[j] >> 1) & 0x1f;
1198         const GLint rowBb1 = (rowB[k] >> 1) & 0x1f;
1199         const GLint rowCb0 = (rowC[j] >> 1) & 0x1f;
1200         const GLint rowCb1 = (rowC[k] >> 1) & 0x1f;
1201         const GLint rowDb0 = (rowD[j] >> 1) & 0x1f;
1202         const GLint rowDb1 = (rowD[k] >> 1) & 0x1f;
1203         const GLint rowAa0 = (rowA[j] & 0x1);
1204         const GLint rowAa1 = (rowA[k] & 0x1);
1205         const GLint rowBa0 = (rowB[j] & 0x1);
1206         const GLint rowBa1 = (rowB[k] & 0x1);
1207         const GLint rowCa0 = (rowC[j] & 0x1);
1208         const GLint rowCa1 = (rowC[k] & 0x1);
1209         const GLint rowDa0 = (rowD[j] & 0x1);
1210         const GLint rowDa1 = (rowD[k] & 0x1);
1211         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1212                                       rowCr0, rowCr1, rowDr0, rowDr1);
1213         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1214                                       rowCg0, rowCg1, rowDg0, rowDg1);
1215         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1216                                       rowCb0, rowCb1, rowDb0, rowDb1);
1217         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1218                                       rowCa0, rowCa1, rowDa0, rowDa1);
1219
1220         dst[i] = (r << 11) | (g << 6) | (b << 1) | a;
1221      }
1222   }
1223   else if ((datatype == GL_UNSIGNED_BYTE_3_3_2) && (comps == 3)) {
1224      DECLARE_ROW_POINTERS0(GLubyte);
1225
1226      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1227           i++, j += colStride, k += colStride) {
1228         const GLint rowAr0 = rowA[j] & 0x3;
1229         const GLint rowAr1 = rowA[k] & 0x3;
1230         const GLint rowBr0 = rowB[j] & 0x3;
1231         const GLint rowBr1 = rowB[k] & 0x3;
1232         const GLint rowCr0 = rowC[j] & 0x3;
1233         const GLint rowCr1 = rowC[k] & 0x3;
1234         const GLint rowDr0 = rowD[j] & 0x3;
1235         const GLint rowDr1 = rowD[k] & 0x3;
1236         const GLint rowAg0 = (rowA[j] >> 2) & 0x7;
1237         const GLint rowAg1 = (rowA[k] >> 2) & 0x7;
1238         const GLint rowBg0 = (rowB[j] >> 2) & 0x7;
1239         const GLint rowBg1 = (rowB[k] >> 2) & 0x7;
1240         const GLint rowCg0 = (rowC[j] >> 2) & 0x7;
1241         const GLint rowCg1 = (rowC[k] >> 2) & 0x7;
1242         const GLint rowDg0 = (rowD[j] >> 2) & 0x7;
1243         const GLint rowDg1 = (rowD[k] >> 2) & 0x7;
1244         const GLint rowAb0 = (rowA[j] >> 5) & 0x7;
1245         const GLint rowAb1 = (rowA[k] >> 5) & 0x7;
1246         const GLint rowBb0 = (rowB[j] >> 5) & 0x7;
1247         const GLint rowBb1 = (rowB[k] >> 5) & 0x7;
1248         const GLint rowCb0 = (rowC[j] >> 5) & 0x7;
1249         const GLint rowCb1 = (rowC[k] >> 5) & 0x7;
1250         const GLint rowDb0 = (rowD[j] >> 5) & 0x7;
1251         const GLint rowDb1 = (rowD[k] >> 5) & 0x7;
1252         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1253                                       rowCr0, rowCr1, rowDr0, rowDr1);
1254         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1255                                       rowCg0, rowCg1, rowDg0, rowDg1);
1256         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1257                                       rowCb0, rowCb1, rowDb0, rowDb1);
1258         dst[i] = (b << 5) | (g << 2) | r;
1259      }
1260   }
1261   else if (datatype == MESA_UNSIGNED_BYTE_4_4 && comps == 2) {
1262      DECLARE_ROW_POINTERS0(GLubyte);
1263
1264      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1265           i++, j += colStride, k += colStride) {
1266         const GLint rowAr0 = rowA[j] & 0xf;
1267         const GLint rowAr1 = rowA[k] & 0xf;
1268         const GLint rowBr0 = rowB[j] & 0xf;
1269         const GLint rowBr1 = rowB[k] & 0xf;
1270         const GLint rowCr0 = rowC[j] & 0xf;
1271         const GLint rowCr1 = rowC[k] & 0xf;
1272         const GLint rowDr0 = rowD[j] & 0xf;
1273         const GLint rowDr1 = rowD[k] & 0xf;
1274         const GLint rowAg0 = (rowA[j] >> 4) & 0xf;
1275         const GLint rowAg1 = (rowA[k] >> 4) & 0xf;
1276         const GLint rowBg0 = (rowB[j] >> 4) & 0xf;
1277         const GLint rowBg1 = (rowB[k] >> 4) & 0xf;
1278         const GLint rowCg0 = (rowC[j] >> 4) & 0xf;
1279         const GLint rowCg1 = (rowC[k] >> 4) & 0xf;
1280         const GLint rowDg0 = (rowD[j] >> 4) & 0xf;
1281         const GLint rowDg1 = (rowD[k] >> 4) & 0xf;
1282         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1283                                       rowCr0, rowCr1, rowDr0, rowDr1);
1284         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1285                                       rowCg0, rowCg1, rowDg0, rowDg1);
1286         dst[i] = (g << 4) | r;
1287      }
1288   }
1289   else if ((datatype == GL_UNSIGNED_INT_2_10_10_10_REV) && (comps == 4)) {
1290      DECLARE_ROW_POINTERS0(GLuint);
1291
1292      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1293           i++, j += colStride, k += colStride) {
1294         const GLint rowAr0 = rowA[j] & 0x3ff;
1295         const GLint rowAr1 = rowA[k] & 0x3ff;
1296         const GLint rowBr0 = rowB[j] & 0x3ff;
1297         const GLint rowBr1 = rowB[k] & 0x3ff;
1298         const GLint rowCr0 = rowC[j] & 0x3ff;
1299         const GLint rowCr1 = rowC[k] & 0x3ff;
1300         const GLint rowDr0 = rowD[j] & 0x3ff;
1301         const GLint rowDr1 = rowD[k] & 0x3ff;
1302         const GLint rowAg0 = (rowA[j] >> 10) & 0x3ff;
1303         const GLint rowAg1 = (rowA[k] >> 10) & 0x3ff;
1304         const GLint rowBg0 = (rowB[j] >> 10) & 0x3ff;
1305         const GLint rowBg1 = (rowB[k] >> 10) & 0x3ff;
1306         const GLint rowCg0 = (rowC[j] >> 10) & 0x3ff;
1307         const GLint rowCg1 = (rowC[k] >> 10) & 0x3ff;
1308         const GLint rowDg0 = (rowD[j] >> 10) & 0x3ff;
1309         const GLint rowDg1 = (rowD[k] >> 10) & 0x3ff;
1310         const GLint rowAb0 = (rowA[j] >> 20) & 0x3ff;
1311         const GLint rowAb1 = (rowA[k] >> 20) & 0x3ff;
1312         const GLint rowBb0 = (rowB[j] >> 20) & 0x3ff;
1313         const GLint rowBb1 = (rowB[k] >> 20) & 0x3ff;
1314         const GLint rowCb0 = (rowC[j] >> 20) & 0x3ff;
1315         const GLint rowCb1 = (rowC[k] >> 20) & 0x3ff;
1316         const GLint rowDb0 = (rowD[j] >> 20) & 0x3ff;
1317         const GLint rowDb1 = (rowD[k] >> 20) & 0x3ff;
1318         const GLint rowAa0 = (rowA[j] >> 30) & 0x3;
1319         const GLint rowAa1 = (rowA[k] >> 30) & 0x3;
1320         const GLint rowBa0 = (rowB[j] >> 30) & 0x3;
1321         const GLint rowBa1 = (rowB[k] >> 30) & 0x3;
1322         const GLint rowCa0 = (rowC[j] >> 30) & 0x3;
1323         const GLint rowCa1 = (rowC[k] >> 30) & 0x3;
1324         const GLint rowDa0 = (rowD[j] >> 30) & 0x3;
1325         const GLint rowDa1 = (rowD[k] >> 30) & 0x3;
1326         const GLint r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
1327                                       rowCr0, rowCr1, rowDr0, rowDr1);
1328         const GLint g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
1329                                       rowCg0, rowCg1, rowDg0, rowDg1);
1330         const GLint b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
1331                                       rowCb0, rowCb1, rowDb0, rowDb1);
1332         const GLint a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
1333                                       rowCa0, rowCa1, rowDa0, rowDa1);
1334
1335         dst[i] = (a << 30) | (b << 20) | (g << 10) | r;
1336      }
1337   }
1338
1339   else if (datatype == GL_UNSIGNED_INT_5_9_9_9_REV && comps == 3) {
1340      DECLARE_ROW_POINTERS0(GLuint);
1341
1342      GLfloat res[3];
1343      GLfloat rowAj[3], rowBj[3], rowCj[3], rowDj[3];
1344      GLfloat rowAk[3], rowBk[3], rowCk[3], rowDk[3];
1345
1346      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1347           i++, j += colStride, k += colStride) {
1348         rgb9e5_to_float3(rowA[j], rowAj);
1349         rgb9e5_to_float3(rowB[j], rowBj);
1350         rgb9e5_to_float3(rowC[j], rowCj);
1351         rgb9e5_to_float3(rowD[j], rowDj);
1352         rgb9e5_to_float3(rowA[k], rowAk);
1353         rgb9e5_to_float3(rowB[k], rowBk);
1354         rgb9e5_to_float3(rowC[k], rowCk);
1355         rgb9e5_to_float3(rowD[k], rowDk);
1356         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0] +
1357                   rowCj[0] + rowCk[0] + rowDj[0] + rowDk[0]) * 0.125F;
1358         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1] +
1359                   rowCj[1] + rowCk[1] + rowDj[1] + rowDk[1]) * 0.125F;
1360         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2] +
1361                   rowCj[2] + rowCk[2] + rowDj[2] + rowDk[2]) * 0.125F;
1362         dst[i] = float3_to_rgb9e5(res);
1363      }
1364   }
1365
1366   else if (datatype == GL_UNSIGNED_INT_10F_11F_11F_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         r11g11b10f_to_float3(rowA[j], rowAj);
1376         r11g11b10f_to_float3(rowB[j], rowBj);
1377         r11g11b10f_to_float3(rowC[j], rowCj);
1378         r11g11b10f_to_float3(rowD[j], rowDj);
1379         r11g11b10f_to_float3(rowA[k], rowAk);
1380         r11g11b10f_to_float3(rowB[k], rowBk);
1381         r11g11b10f_to_float3(rowC[k], rowCk);
1382         r11g11b10f_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_r11g11b10f(res);
1390      }
1391   }
1392
1393   else if (datatype == GL_FLOAT_32_UNSIGNED_INT_24_8_REV && comps == 1) {
1394      DECLARE_ROW_POINTERS(GLfloat, 2);
1395
1396      for (i = j = 0, k = k0; i < (GLuint) dstWidth;
1397           i++, j += colStride, k += colStride) {
1398         FILTER_F_3D(0);
1399      }
1400   }
1401
1402   else {
1403      _mesa_problem(NULL, "bad format in do_row()");
1404   }
1405}
1406
1407
1408/*
1409 * These functions generate a 1/2-size mipmap image from a source image.
1410 * Texture borders are handled by copying or averaging the source image's
1411 * border texels, depending on the scale-down factor.
1412 */
1413
1414static void
1415make_1d_mipmap(GLenum datatype, GLuint comps, GLint border,
1416               GLint srcWidth, const GLubyte *srcPtr,
1417               GLint dstWidth, GLubyte *dstPtr)
1418{
1419   const GLint bpt = bytes_per_pixel(datatype, comps);
1420   const GLubyte *src;
1421   GLubyte *dst;
1422
1423   /* skip the border pixel, if any */
1424   src = srcPtr + border * bpt;
1425   dst = dstPtr + border * bpt;
1426
1427   /* we just duplicate the input row, kind of hack, saves code */
1428   do_row(datatype, comps, srcWidth - 2 * border, src, src,
1429          dstWidth - 2 * border, dst);
1430
1431   if (border) {
1432      /* copy left-most pixel from source */
1433      assert(dstPtr);
1434      assert(srcPtr);
1435      memcpy(dstPtr, srcPtr, bpt);
1436      /* copy right-most pixel from source */
1437      memcpy(dstPtr + (dstWidth - 1) * bpt,
1438             srcPtr + (srcWidth - 1) * bpt,
1439             bpt);
1440   }
1441}
1442
1443
1444static void
1445make_2d_mipmap(GLenum datatype, GLuint comps, GLint border,
1446               GLint srcWidth, GLint srcHeight,
1447	       const GLubyte *srcPtr, GLint srcRowStride,
1448               GLint dstWidth, GLint dstHeight,
1449	       GLubyte *dstPtr, GLint dstRowStride)
1450{
1451   const GLint bpt = bytes_per_pixel(datatype, comps);
1452   const GLint srcWidthNB = srcWidth - 2 * border;  /* sizes w/out border */
1453   const GLint dstWidthNB = dstWidth - 2 * border;
1454   const GLint dstHeightNB = dstHeight - 2 * border;
1455   const GLubyte *srcA, *srcB;
1456   GLubyte *dst;
1457   GLint row, srcRowStep;
1458
1459   /* Compute src and dst pointers, skipping any border */
1460   srcA = srcPtr + border * ((srcWidth + 1) * bpt);
1461   if (srcHeight > 1 && srcHeight > dstHeight) {
1462      /* sample from two source rows */
1463      srcB = srcA + srcRowStride;
1464      srcRowStep = 2;
1465   }
1466   else {
1467      /* sample from one source row */
1468      srcB = srcA;
1469      srcRowStep = 1;
1470   }
1471
1472   dst = dstPtr + border * ((dstWidth + 1) * bpt);
1473
1474   for (row = 0; row < dstHeightNB; row++) {
1475      do_row(datatype, comps, srcWidthNB, srcA, srcB,
1476             dstWidthNB, dst);
1477      srcA += srcRowStep * srcRowStride;
1478      srcB += srcRowStep * srcRowStride;
1479      dst += dstRowStride;
1480   }
1481
1482   /* This is ugly but probably won't be used much */
1483   if (border > 0) {
1484      /* fill in dest border */
1485      /* lower-left border pixel */
1486      assert(dstPtr);
1487      assert(srcPtr);
1488      memcpy(dstPtr, srcPtr, bpt);
1489      /* lower-right border pixel */
1490      memcpy(dstPtr + (dstWidth - 1) * bpt,
1491             srcPtr + (srcWidth - 1) * bpt, bpt);
1492      /* upper-left border pixel */
1493      memcpy(dstPtr + dstWidth * (dstHeight - 1) * bpt,
1494             srcPtr + srcWidth * (srcHeight - 1) * bpt, bpt);
1495      /* upper-right border pixel */
1496      memcpy(dstPtr + (dstWidth * dstHeight - 1) * bpt,
1497             srcPtr + (srcWidth * srcHeight - 1) * bpt, bpt);
1498      /* lower border */
1499      do_row(datatype, comps, srcWidthNB,
1500             srcPtr + bpt,
1501             srcPtr + bpt,
1502             dstWidthNB, dstPtr + bpt);
1503      /* upper border */
1504      do_row(datatype, comps, srcWidthNB,
1505             srcPtr + (srcWidth * (srcHeight - 1) + 1) * bpt,
1506             srcPtr + (srcWidth * (srcHeight - 1) + 1) * bpt,
1507             dstWidthNB,
1508             dstPtr + (dstWidth * (dstHeight - 1) + 1) * bpt);
1509      /* left and right borders */
1510      if (srcHeight == dstHeight) {
1511         /* copy border pixel from src to dst */
1512         for (row = 1; row < srcHeight; row++) {
1513            memcpy(dstPtr + dstWidth * row * bpt,
1514                   srcPtr + srcWidth * row * bpt, bpt);
1515            memcpy(dstPtr + (dstWidth * row + dstWidth - 1) * bpt,
1516                   srcPtr + (srcWidth * row + srcWidth - 1) * bpt, bpt);
1517         }
1518      }
1519      else {
1520         /* average two src pixels each dest pixel */
1521         for (row = 0; row < dstHeightNB; row += 2) {
1522            do_row(datatype, comps, 1,
1523                   srcPtr + (srcWidth * (row * 2 + 1)) * bpt,
1524                   srcPtr + (srcWidth * (row * 2 + 2)) * bpt,
1525                   1, dstPtr + (dstWidth * row + 1) * bpt);
1526            do_row(datatype, comps, 1,
1527                   srcPtr + (srcWidth * (row * 2 + 1) + srcWidth - 1) * bpt,
1528                   srcPtr + (srcWidth * (row * 2 + 2) + srcWidth - 1) * bpt,
1529                   1, dstPtr + (dstWidth * row + 1 + dstWidth - 1) * bpt);
1530         }
1531      }
1532   }
1533}
1534
1535
1536static void
1537make_3d_mipmap(GLenum datatype, GLuint comps, GLint border,
1538               GLint srcWidth, GLint srcHeight, GLint srcDepth,
1539               const GLubyte **srcPtr, GLint srcRowStride,
1540               GLint dstWidth, GLint dstHeight, GLint dstDepth,
1541               GLubyte **dstPtr, GLint dstRowStride)
1542{
1543   const GLint bpt = bytes_per_pixel(datatype, comps);
1544   const GLint srcWidthNB = srcWidth - 2 * border;  /* sizes w/out border */
1545   const GLint srcDepthNB = srcDepth - 2 * border;
1546   const GLint dstWidthNB = dstWidth - 2 * border;
1547   const GLint dstHeightNB = dstHeight - 2 * border;
1548   const GLint dstDepthNB = dstDepth - 2 * border;
1549   GLint img, row;
1550   GLint bytesPerSrcImage, bytesPerDstImage;
1551   GLint bytesPerSrcRow, bytesPerDstRow;
1552   GLint srcImageOffset, srcRowOffset;
1553
1554   (void) srcDepthNB; /* silence warnings */
1555
1556
1557   bytesPerSrcImage = srcWidth * srcHeight * bpt;
1558   bytesPerDstImage = dstWidth * dstHeight * bpt;
1559
1560   bytesPerSrcRow = srcWidth * bpt;
1561   bytesPerDstRow = dstWidth * bpt;
1562
1563   /* Offset between adjacent src images to be averaged together */
1564   srcImageOffset = (srcDepth == dstDepth) ? 0 : 1;
1565
1566   /* Offset between adjacent src rows to be averaged together */
1567   srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt;
1568
1569   /*
1570    * Need to average together up to 8 src pixels for each dest pixel.
1571    * Break that down into 3 operations:
1572    *   1. take two rows from source image and average them together.
1573    *   2. take two rows from next source image and average them together.
1574    *   3. take the two averaged rows and average them for the final dst row.
1575    */
1576
1577   /*
1578   printf("mip3d %d x %d x %d  ->  %d x %d x %d\n",
1579          srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
1580   */
1581
1582   for (img = 0; img < dstDepthNB; img++) {
1583      /* first source image pointer, skipping border */
1584      const GLubyte *imgSrcA = srcPtr[img * 2 + border]
1585         + bytesPerSrcRow * border + bpt * border;
1586      /* second source image pointer, skipping border */
1587      const GLubyte *imgSrcB = srcPtr[img * 2 + srcImageOffset + border]
1588         + bytesPerSrcRow * border + bpt * border;
1589
1590      /* address of the dest image, skipping border */
1591      GLubyte *imgDst = dstPtr[img + border]
1592         + bytesPerDstRow * border + bpt * border;
1593
1594      /* setup the four source row pointers and the dest row pointer */
1595      const GLubyte *srcImgARowA = imgSrcA;
1596      const GLubyte *srcImgARowB = imgSrcA + srcRowOffset;
1597      const GLubyte *srcImgBRowA = imgSrcB;
1598      const GLubyte *srcImgBRowB = imgSrcB + srcRowOffset;
1599      GLubyte *dstImgRow = imgDst;
1600
1601      for (row = 0; row < dstHeightNB; row++) {
1602         do_row_3D(datatype, comps, srcWidthNB,
1603                   srcImgARowA, srcImgARowB,
1604                   srcImgBRowA, srcImgBRowB,
1605                   dstWidthNB, dstImgRow);
1606
1607         /* advance to next rows */
1608         srcImgARowA += bytesPerSrcRow + srcRowOffset;
1609         srcImgARowB += bytesPerSrcRow + srcRowOffset;
1610         srcImgBRowA += bytesPerSrcRow + srcRowOffset;
1611         srcImgBRowB += bytesPerSrcRow + srcRowOffset;
1612         dstImgRow += bytesPerDstRow;
1613      }
1614   }
1615
1616
1617   /* Luckily we can leverage the make_2d_mipmap() function here! */
1618   if (border > 0) {
1619      /* do front border image */
1620      make_2d_mipmap(datatype, comps, 1,
1621                     srcWidth, srcHeight, srcPtr[0], srcRowStride,
1622                     dstWidth, dstHeight, dstPtr[0], dstRowStride);
1623      /* do back border image */
1624      make_2d_mipmap(datatype, comps, 1,
1625                     srcWidth, srcHeight, srcPtr[srcDepth - 1], srcRowStride,
1626                     dstWidth, dstHeight, dstPtr[dstDepth - 1], dstRowStride);
1627
1628      /* do four remaining border edges that span the image slices */
1629      if (srcDepth == dstDepth) {
1630         /* just copy border pixels from src to dst */
1631         for (img = 0; img < dstDepthNB; img++) {
1632            const GLubyte *src;
1633            GLubyte *dst;
1634
1635            /* do border along [img][row=0][col=0] */
1636            src = srcPtr[img * 2];
1637            dst = dstPtr[img];
1638            memcpy(dst, src, bpt);
1639
1640            /* do border along [img][row=dstHeight-1][col=0] */
1641            src = srcPtr[img * 2] + (srcHeight - 1) * bytesPerSrcRow;
1642            dst = dstPtr[img] + (dstHeight - 1) * bytesPerDstRow;
1643            memcpy(dst, src, bpt);
1644
1645            /* do border along [img][row=0][col=dstWidth-1] */
1646            src = srcPtr[img * 2] + (srcWidth - 1) * bpt;
1647            dst = dstPtr[img] + (dstWidth - 1) * bpt;
1648            memcpy(dst, src, bpt);
1649
1650            /* do border along [img][row=dstHeight-1][col=dstWidth-1] */
1651            src = srcPtr[img * 2] + (bytesPerSrcImage - bpt);
1652            dst = dstPtr[img] + (bytesPerDstImage - bpt);
1653            memcpy(dst, src, bpt);
1654         }
1655      }
1656      else {
1657         /* average border pixels from adjacent src image pairs */
1658         ASSERT(srcDepthNB == 2 * dstDepthNB);
1659         for (img = 0; img < dstDepthNB; img++) {
1660            const GLubyte *srcA, *srcB;
1661            GLubyte *dst;
1662
1663            /* do border along [img][row=0][col=0] */
1664            srcA = srcPtr[img * 2 + 0];
1665            srcB = srcPtr[img * 2 + srcImageOffset];
1666            dst = dstPtr[img];
1667            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1668
1669            /* do border along [img][row=dstHeight-1][col=0] */
1670            srcA = srcPtr[img * 2 + 0]
1671               + (srcHeight - 1) * bytesPerSrcRow;
1672            srcB = srcPtr[img * 2 + srcImageOffset]
1673               + (srcHeight - 1) * bytesPerSrcRow;
1674            dst = dstPtr[img] + (dstHeight - 1) * bytesPerDstRow;
1675            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1676
1677            /* do border along [img][row=0][col=dstWidth-1] */
1678            srcA = srcPtr[img * 2 + 0] + (srcWidth - 1) * bpt;
1679            srcB = srcPtr[img * 2 + srcImageOffset] + (srcWidth - 1) * bpt;
1680            dst = dstPtr[img] + (dstWidth - 1) * bpt;
1681            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1682
1683            /* do border along [img][row=dstHeight-1][col=dstWidth-1] */
1684            srcA = srcPtr[img * 2 + 0] + (bytesPerSrcImage - bpt);
1685            srcB = srcPtr[img * 2 + srcImageOffset] + (bytesPerSrcImage - bpt);
1686            dst = dstPtr[img] + (bytesPerDstImage - bpt);
1687            do_row(datatype, comps, 1, srcA, srcB, 1, dst);
1688         }
1689      }
1690   }
1691}
1692
1693
1694/**
1695 * Down-sample a texture image to produce the next lower mipmap level.
1696 * \param comps  components per texel (1, 2, 3 or 4)
1697 * \param srcData  array[slice] of pointers to source image slices
1698 * \param dstData  array[slice] of pointers to dest image slices
1699 * \param srcRowStride  stride between source rows, in bytes
1700 * \param dstRowStride  stride between destination rows, in bytes
1701 */
1702void
1703_mesa_generate_mipmap_level(GLenum target,
1704                            GLenum datatype, GLuint comps,
1705                            GLint border,
1706                            GLint srcWidth, GLint srcHeight, GLint srcDepth,
1707                            const GLubyte **srcData,
1708                            GLint srcRowStride,
1709                            GLint dstWidth, GLint dstHeight, GLint dstDepth,
1710                            GLubyte **dstData,
1711                            GLint dstRowStride)
1712{
1713   int i;
1714
1715   switch (target) {
1716   case GL_TEXTURE_1D:
1717      make_1d_mipmap(datatype, comps, border,
1718                     srcWidth, srcData[0],
1719                     dstWidth, dstData[0]);
1720      break;
1721   case GL_TEXTURE_2D:
1722   case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB:
1723   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB:
1724   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB:
1725   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB:
1726   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB:
1727   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB:
1728      make_2d_mipmap(datatype, comps, border,
1729                     srcWidth, srcHeight, srcData[0], srcRowStride,
1730                     dstWidth, dstHeight, dstData[0], dstRowStride);
1731      break;
1732   case GL_TEXTURE_3D:
1733      make_3d_mipmap(datatype, comps, border,
1734                     srcWidth, srcHeight, srcDepth,
1735                     srcData, srcRowStride,
1736                     dstWidth, dstHeight, dstDepth,
1737                     dstData, dstRowStride);
1738      break;
1739   case GL_TEXTURE_1D_ARRAY_EXT:
1740      assert(srcHeight == 1);
1741      assert(dstHeight == 1);
1742      for (i = 0; i < dstDepth; i++) {
1743	 make_1d_mipmap(datatype, comps, border,
1744			srcWidth, srcData[i],
1745			dstWidth, dstData[i]);
1746      }
1747      break;
1748   case GL_TEXTURE_2D_ARRAY_EXT:
1749      for (i = 0; i < dstDepth; i++) {
1750	 make_2d_mipmap(datatype, comps, border,
1751			srcWidth, srcHeight, srcData[i], srcRowStride,
1752			dstWidth, dstHeight, dstData[i], dstRowStride);
1753      }
1754      break;
1755   case GL_TEXTURE_RECTANGLE_NV:
1756   case GL_TEXTURE_EXTERNAL_OES:
1757      /* no mipmaps, do nothing */
1758      break;
1759   default:
1760      _mesa_problem(NULL, "bad tex target in _mesa_generate_mipmaps");
1761      return;
1762   }
1763}
1764
1765
1766/**
1767 * compute next (level+1) image size
1768 * \return GL_FALSE if no smaller size can be generated (eg. src is 1x1x1 size)
1769 */
1770static GLboolean
1771next_mipmap_level_size(GLenum target, GLint border,
1772                       GLint srcWidth, GLint srcHeight, GLint srcDepth,
1773                       GLint *dstWidth, GLint *dstHeight, GLint *dstDepth)
1774{
1775   if (srcWidth - 2 * border > 1) {
1776      *dstWidth = (srcWidth - 2 * border) / 2 + 2 * border;
1777   }
1778   else {
1779      *dstWidth = srcWidth; /* can't go smaller */
1780   }
1781
1782   if ((srcHeight - 2 * border > 1) &&
1783       (target != GL_TEXTURE_1D_ARRAY_EXT)) {
1784      *dstHeight = (srcHeight - 2 * border) / 2 + 2 * border;
1785   }
1786   else {
1787      *dstHeight = srcHeight; /* can't go smaller */
1788   }
1789
1790   if ((srcDepth - 2 * border > 1) &&
1791       (target != GL_TEXTURE_2D_ARRAY_EXT)) {
1792      *dstDepth = (srcDepth - 2 * border) / 2 + 2 * border;
1793   }
1794   else {
1795      *dstDepth = srcDepth; /* can't go smaller */
1796   }
1797
1798   if (*dstWidth == srcWidth &&
1799       *dstHeight == srcHeight &&
1800       *dstDepth == srcDepth) {
1801      return GL_FALSE;
1802   }
1803   else {
1804      return GL_TRUE;
1805   }
1806}
1807
1808
1809/**
1810 * Helper function for mipmap generation.
1811 * Make sure the specified destination mipmap level is the right size/format
1812 * for mipmap generation.  If not, (re) allocate it.
1813 * \return GL_TRUE if successful, GL_FALSE if mipmap generation should stop
1814 */
1815GLboolean
1816_mesa_prepare_mipmap_level(struct gl_context *ctx,
1817                           struct gl_texture_object *texObj, GLuint level,
1818                           GLsizei width, GLsizei height, GLsizei depth,
1819                           GLsizei border, GLenum intFormat, gl_format format)
1820{
1821   const GLuint numFaces = _mesa_num_tex_faces(texObj->Target);
1822   GLuint face;
1823
1824   if (texObj->Immutable) {
1825      /* The texture was created with glTexStorage() so the number/size of
1826       * mipmap levels is fixed and the storage for all images is already
1827       * allocated.
1828       */
1829      if (!texObj->Image[0][level]) {
1830         /* No more levels to create - we're done */
1831         return GL_FALSE;
1832      }
1833      else {
1834         /* Nothing to do - the texture memory must have already been
1835          * allocated to the right size so we're all set.
1836          */
1837         return GL_TRUE;
1838      }
1839   }
1840
1841   for (face = 0; face < numFaces; face++) {
1842      struct gl_texture_image *dstImage;
1843      GLenum target;
1844
1845      if (numFaces == 1)
1846         target = texObj->Target;
1847      else
1848         target = GL_TEXTURE_CUBE_MAP_POSITIVE_X + face;
1849
1850      dstImage = _mesa_get_tex_image(ctx, texObj, target, level);
1851      if (!dstImage) {
1852         /* out of memory */
1853         return GL_FALSE;
1854      }
1855
1856      if (dstImage->Width != width ||
1857          dstImage->Height != height ||
1858          dstImage->Depth != depth ||
1859          dstImage->Border != border ||
1860          dstImage->InternalFormat != intFormat ||
1861          dstImage->TexFormat != format) {
1862         /* need to (re)allocate image */
1863         ctx->Driver.FreeTextureImageBuffer(ctx, dstImage);
1864
1865         _mesa_init_teximage_fields(ctx, dstImage,
1866                                    width, height, depth,
1867                                    border, intFormat, format);
1868
1869         ctx->Driver.AllocTextureImageBuffer(ctx, dstImage);
1870
1871         /* in case the mipmap level is part of an FBO: */
1872         _mesa_update_fbo_texture(ctx, texObj, face, level);
1873
1874         ctx->NewState |= _NEW_TEXTURE;
1875      }
1876   }
1877
1878   return GL_TRUE;
1879}
1880
1881
1882static void
1883generate_mipmap_uncompressed(struct gl_context *ctx, GLenum target,
1884			     struct gl_texture_object *texObj,
1885			     const struct gl_texture_image *srcImage,
1886			     GLuint maxLevel)
1887{
1888   GLint level;
1889   GLenum datatype;
1890   GLuint comps;
1891
1892   _mesa_format_to_type_and_comps(srcImage->TexFormat, &datatype, &comps);
1893
1894   for (level = texObj->BaseLevel; level < maxLevel; level++) {
1895      /* generate image[level+1] from image[level] */
1896      struct gl_texture_image *srcImage, *dstImage;
1897      GLint srcRowStride, dstRowStride;
1898      GLint srcWidth, srcHeight, srcDepth;
1899      GLint dstWidth, dstHeight, dstDepth;
1900      GLint border;
1901      GLint slice;
1902      GLboolean nextLevel;
1903      GLubyte **srcMaps, **dstMaps;
1904      GLboolean success = GL_TRUE;
1905
1906      /* get src image parameters */
1907      srcImage = _mesa_select_tex_image(ctx, texObj, target, level);
1908      ASSERT(srcImage);
1909      srcWidth = srcImage->Width;
1910      srcHeight = srcImage->Height;
1911      srcDepth = srcImage->Depth;
1912      border = srcImage->Border;
1913
1914      nextLevel = next_mipmap_level_size(target, border,
1915                                         srcWidth, srcHeight, srcDepth,
1916                                         &dstWidth, &dstHeight, &dstDepth);
1917      if (!nextLevel)
1918         return;
1919
1920      if (!_mesa_prepare_mipmap_level(ctx, texObj, level + 1,
1921                                      dstWidth, dstHeight, dstDepth,
1922                                      border, srcImage->InternalFormat,
1923                                      srcImage->TexFormat)) {
1924         return;
1925      }
1926
1927      /* get dest gl_texture_image */
1928      dstImage = _mesa_get_tex_image(ctx, texObj, target, level + 1);
1929      if (!dstImage) {
1930         _mesa_error(ctx, GL_OUT_OF_MEMORY, "generating mipmaps");
1931         return;
1932      }
1933
1934      if (target == GL_TEXTURE_1D_ARRAY) {
1935	 srcDepth = srcHeight;
1936	 dstDepth = dstHeight;
1937	 srcHeight = 1;
1938	 dstHeight = 1;
1939      }
1940
1941      /* Map src texture image slices */
1942      srcMaps = (GLubyte **) calloc(srcDepth, sizeof(GLubyte *));
1943      if (srcMaps) {
1944         for (slice = 0; slice < srcDepth; slice++) {
1945            ctx->Driver.MapTextureImage(ctx, srcImage, slice,
1946                                        0, 0, srcWidth, srcHeight,
1947                                        GL_MAP_READ_BIT,
1948                                        &srcMaps[slice], &srcRowStride);
1949            if (!srcMaps[slice]) {
1950               success = GL_FALSE;
1951               break;
1952            }
1953         }
1954      }
1955      else {
1956         success = GL_FALSE;
1957      }
1958
1959      /* Map dst texture image slices */
1960      dstMaps = (GLubyte **) calloc(dstDepth, sizeof(GLubyte *));
1961      if (dstMaps) {
1962         for (slice = 0; slice < dstDepth; slice++) {
1963            ctx->Driver.MapTextureImage(ctx, dstImage, slice,
1964                                        0, 0, dstWidth, dstHeight,
1965                                        GL_MAP_WRITE_BIT,
1966                                        &dstMaps[slice], &dstRowStride);
1967            if (!dstMaps[slice]) {
1968               success = GL_FALSE;
1969               break;
1970            }
1971         }
1972      }
1973      else {
1974         success = GL_FALSE;
1975      }
1976
1977      if (success) {
1978         /* generate one mipmap level (for 1D/2D/3D/array/etc texture) */
1979         _mesa_generate_mipmap_level(target, datatype, comps, border,
1980                                     srcWidth, srcHeight, srcDepth,
1981                                     (const GLubyte **) srcMaps, srcRowStride,
1982                                     dstWidth, dstHeight, dstDepth,
1983                                     dstMaps, dstRowStride);
1984      }
1985
1986      /* Unmap src image slices */
1987      if (srcMaps) {
1988         for (slice = 0; slice < srcDepth; slice++) {
1989            if (srcMaps[slice]) {
1990               ctx->Driver.UnmapTextureImage(ctx, srcImage, slice);
1991            }
1992         }
1993         free(srcMaps);
1994      }
1995
1996      /* Unmap dst image slices */
1997      if (dstMaps) {
1998         for (slice = 0; slice < dstDepth; slice++) {
1999            if (dstMaps[slice]) {
2000               ctx->Driver.UnmapTextureImage(ctx, dstImage, slice);
2001            }
2002         }
2003         free(dstMaps);
2004      }
2005
2006      if (!success) {
2007         _mesa_error(ctx, GL_OUT_OF_MEMORY, "mipmap generation");
2008         break;
2009      }
2010   } /* loop over mipmap levels */
2011}
2012
2013
2014static void
2015generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
2016			   struct gl_texture_object *texObj,
2017			   struct gl_texture_image *srcImage,
2018			   GLuint maxLevel)
2019{
2020   GLint level;
2021   gl_format temp_format;
2022   GLint components;
2023   GLuint temp_src_stride; /* in bytes */
2024   GLubyte *temp_src = NULL, *temp_dst = NULL;
2025   GLenum temp_datatype;
2026   GLenum temp_base_format;
2027
2028   /* only two types of compressed textures at this time */
2029   assert(texObj->Target == GL_TEXTURE_2D ||
2030	  texObj->Target == GL_TEXTURE_CUBE_MAP_ARB);
2031
2032   /*
2033    * Choose a format for the temporary, uncompressed base image.
2034    * Then, get number of components, choose temporary image datatype,
2035    * and get base format.
2036    */
2037   temp_format = _mesa_get_uncompressed_format(srcImage->TexFormat);
2038
2039   components = _mesa_format_num_components(temp_format);
2040
2041   /* Revisit this if we get compressed formats with >8 bits per component */
2042   if (_mesa_get_format_datatype(srcImage->TexFormat)
2043       == GL_SIGNED_NORMALIZED) {
2044      temp_datatype = GL_BYTE;
2045   }
2046   else {
2047      temp_datatype = GL_UNSIGNED_BYTE;
2048   }
2049
2050   temp_base_format = _mesa_get_format_base_format(temp_format);
2051
2052
2053   /* allocate storage for the temporary, uncompressed image */
2054   /* 20 extra bytes, just be safe when calling last FetchTexel */
2055   temp_src_stride = _mesa_format_row_stride(temp_format, srcImage->Width);
2056   temp_src = (GLubyte *) malloc(temp_src_stride * srcImage->Height + 20);
2057   if (!temp_src) {
2058      _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
2059      return;
2060   }
2061
2062   /* decompress base image to the temporary */
2063   {
2064      /* save pixel packing mode */
2065      struct gl_pixelstore_attrib save = ctx->Pack;
2066      /* use default/tight packing parameters */
2067      ctx->Pack = ctx->DefaultPacking;
2068
2069      /* Get the uncompressed image */
2070      assert(srcImage->Level == texObj->BaseLevel);
2071      ctx->Driver.GetTexImage(ctx,
2072                              temp_base_format, temp_datatype,
2073                              temp_src, srcImage);
2074      /* restore packing mode */
2075      ctx->Pack = save;
2076   }
2077
2078
2079   for (level = texObj->BaseLevel; level < maxLevel; level++) {
2080      /* generate image[level+1] from image[level] */
2081      const struct gl_texture_image *srcImage;
2082      struct gl_texture_image *dstImage;
2083      GLint srcWidth, srcHeight, srcDepth;
2084      GLint dstWidth, dstHeight, dstDepth;
2085      GLint border;
2086      GLboolean nextLevel;
2087      GLuint temp_dst_stride; /* in bytes */
2088
2089      /* get src image parameters */
2090      srcImage = _mesa_select_tex_image(ctx, texObj, target, level);
2091      ASSERT(srcImage);
2092      srcWidth = srcImage->Width;
2093      srcHeight = srcImage->Height;
2094      srcDepth = srcImage->Depth;
2095      border = srcImage->Border;
2096
2097      nextLevel = next_mipmap_level_size(target, border,
2098                                         srcWidth, srcHeight, srcDepth,
2099                                         &dstWidth, &dstHeight, &dstDepth);
2100      if (!nextLevel)
2101	 break;
2102
2103      temp_dst_stride = _mesa_format_row_stride(temp_format, dstWidth);
2104      if (!temp_dst) {
2105	 temp_dst = (GLubyte *) malloc(temp_dst_stride * dstHeight);
2106	 if (!temp_dst) {
2107	    _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
2108	    break;
2109	 }
2110      }
2111
2112      /* get dest gl_texture_image */
2113      dstImage = _mesa_get_tex_image(ctx, texObj, target, level + 1);
2114      if (!dstImage) {
2115         _mesa_error(ctx, GL_OUT_OF_MEMORY, "generating mipmaps");
2116         free(temp_dst);
2117         return;
2118      }
2119
2120      /* rescale src image to dest image */
2121      _mesa_generate_mipmap_level(target, temp_datatype, components, border,
2122                                  srcWidth, srcHeight, srcDepth,
2123                                  (const GLubyte **) &temp_src,
2124                                  temp_src_stride,
2125                                  dstWidth, dstHeight, dstDepth,
2126                                  &temp_dst, temp_dst_stride);
2127
2128      if (!_mesa_prepare_mipmap_level(ctx, texObj, level + 1,
2129                                      dstWidth, dstHeight, dstDepth,
2130                                      border, srcImage->InternalFormat,
2131                                      srcImage->TexFormat)) {
2132         free(temp_dst);
2133         return;
2134      }
2135
2136      /* The image space was allocated above so use glTexSubImage now */
2137      ctx->Driver.TexSubImage(ctx, 2, dstImage,
2138                              0, 0, 0, dstWidth, dstHeight, 1,
2139                              temp_base_format, temp_datatype,
2140                              temp_dst, &ctx->DefaultPacking);
2141
2142      /* swap src and dest pointers */
2143      {
2144	 GLubyte *temp = temp_src;
2145	 temp_src = temp_dst;
2146	 temp_dst = temp;
2147	 temp_src_stride = temp_dst_stride;
2148      }
2149   } /* loop over mipmap levels */
2150
2151   free(temp_src);
2152   free(temp_dst);
2153}
2154
2155/**
2156 * Automatic mipmap generation.
2157 * This is the fallback/default function for ctx->Driver.GenerateMipmap().
2158 * Generate a complete set of mipmaps from texObj's BaseLevel image.
2159 * Stop at texObj's MaxLevel or when we get to the 1x1 texture.
2160 * For cube maps, target will be one of
2161 * GL_TEXTURE_CUBE_MAP_POSITIVE/NEGATIVE_X/Y/Z; never GL_TEXTURE_CUBE_MAP.
2162 */
2163void
2164_mesa_generate_mipmap(struct gl_context *ctx, GLenum target,
2165                      struct gl_texture_object *texObj)
2166{
2167   struct gl_texture_image *srcImage;
2168   GLint maxLevel;
2169
2170   ASSERT(texObj);
2171   srcImage = _mesa_select_tex_image(ctx, texObj, target, texObj->BaseLevel);
2172   ASSERT(srcImage);
2173
2174   maxLevel = _mesa_max_texture_levels(ctx, texObj->Target) - 1;
2175   ASSERT(maxLevel >= 0);  /* bad target */
2176
2177   maxLevel = MIN2(maxLevel, texObj->MaxLevel);
2178
2179   if (_mesa_is_format_compressed(srcImage->TexFormat)) {
2180      generate_mipmap_compressed(ctx, target, texObj, srcImage, maxLevel);
2181   } else {
2182      generate_mipmap_uncompressed(ctx, target, texObj, srcImage, maxLevel);
2183   }
2184}
2185