u_gen_mipmap.c revision d35ecca5ee231c072687578642e0c22c6c0590b1
1/**************************************************************************
2 *
3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 * Copyright 2008  VMware, Inc.  All rights reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29/**
30 * @file
31 * Mipmap generation utility
32 *
33 * @author Brian Paul
34 */
35
36
37#include "pipe/p_context.h"
38#include "util/u_debug.h"
39#include "pipe/p_defines.h"
40#include "util/u_inlines.h"
41#include "pipe/p_shader_tokens.h"
42#include "pipe/p_state.h"
43
44#include "util/u_format.h"
45#include "util/u_memory.h"
46#include "util/u_draw_quad.h"
47#include "util/u_gen_mipmap.h"
48#include "util/u_simple_shaders.h"
49#include "util/u_math.h"
50#include "util/u_texture.h"
51
52#include "cso_cache/cso_context.h"
53
54
55struct gen_mipmap_state
56{
57   struct pipe_context *pipe;
58   struct cso_context *cso;
59
60   struct pipe_blend_state blend;
61   struct pipe_depth_stencil_alpha_state depthstencil;
62   struct pipe_rasterizer_state rasterizer;
63   struct pipe_sampler_state sampler;
64   struct pipe_clip_state clip;
65   struct pipe_vertex_element velem[2];
66
67   void *vs;
68   void *fs2d, *fsCube;
69
70   struct pipe_buffer *vbuf;  /**< quad vertices */
71   unsigned vbuf_slot;
72
73   float vertices[4][2][4];   /**< vertex/texcoords for quad */
74};
75
76
77
78enum dtype
79{
80   DTYPE_UBYTE,
81   DTYPE_UBYTE_3_3_2,
82   DTYPE_USHORT,
83   DTYPE_USHORT_4_4_4_4,
84   DTYPE_USHORT_5_6_5,
85   DTYPE_USHORT_1_5_5_5_REV,
86   DTYPE_UINT,
87   DTYPE_FLOAT,
88   DTYPE_HALF_FLOAT
89};
90
91
92typedef ushort half_float;
93
94
95static half_float
96float_to_half(float f)
97{
98   /* XXX fix this */
99   return 0;
100}
101
102static float
103half_to_float(half_float h)
104{
105   /* XXX fix this */
106   return 0.0f;
107}
108
109
110
111
112/**
113 * \name Support macros for do_row and do_row_3d
114 *
115 * The macro madness is here for two reasons.  First, it compacts the code
116 * slightly.  Second, it makes it much easier to adjust the specifics of the
117 * filter to tune the rounding characteristics.
118 */
119/*@{*/
120#define DECLARE_ROW_POINTERS(t, e) \
121      const t(*rowA)[e] = (const t(*)[e]) srcRowA; \
122      const t(*rowB)[e] = (const t(*)[e]) srcRowB; \
123      const t(*rowC)[e] = (const t(*)[e]) srcRowC; \
124      const t(*rowD)[e] = (const t(*)[e]) srcRowD; \
125      t(*dst)[e] = (t(*)[e]) dstRow
126
127#define DECLARE_ROW_POINTERS0(t) \
128      const t *rowA = (const t *) srcRowA; \
129      const t *rowB = (const t *) srcRowB; \
130      const t *rowC = (const t *) srcRowC; \
131      const t *rowD = (const t *) srcRowD; \
132      t *dst = (t *) dstRow
133
134#define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
135   ((unsigned) Aj + (unsigned) Ak \
136    + (unsigned) Bj + (unsigned) Bk \
137    + (unsigned) Cj + (unsigned) Ck \
138    + (unsigned) Dj + (unsigned) Dk \
139    + 4) >> 3
140
141#define FILTER_3D(e) \
142   do { \
143      dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \
144                                rowB[j][e], rowB[k][e], \
145                                rowC[j][e], rowC[k][e], \
146                                rowD[j][e], rowD[k][e]); \
147   } while(0)
148
149#define FILTER_F_3D(e) \
150   do { \
151      dst[i][e] = (rowA[j][e] + rowA[k][e] \
152                   + rowB[j][e] + rowB[k][e] \
153                   + rowC[j][e] + rowC[k][e] \
154                   + rowD[j][e] + rowD[k][e]) * 0.125F; \
155   } while(0)
156
157#define FILTER_HF_3D(e) \
158   do { \
159      const float aj = half_to_float(rowA[j][e]); \
160      const float ak = half_to_float(rowA[k][e]); \
161      const float bj = half_to_float(rowB[j][e]); \
162      const float bk = half_to_float(rowB[k][e]); \
163      const float cj = half_to_float(rowC[j][e]); \
164      const float ck = half_to_float(rowC[k][e]); \
165      const float dj = half_to_float(rowD[j][e]); \
166      const float dk = half_to_float(rowD[k][e]); \
167      dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
168                                      * 0.125F); \
169   } while(0)
170/*@}*/
171
172
173/**
174 * Average together two rows of a source image to produce a single new
175 * row in the dest image.  It's legal for the two source rows to point
176 * to the same data.  The source width must be equal to either the
177 * dest width or two times the dest width.
178 * \param datatype  GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
179 * \param comps  number of components per pixel (1..4)
180 */
181static void
182do_row(enum dtype datatype, uint comps, int srcWidth,
183       const void *srcRowA, const void *srcRowB,
184       int dstWidth, void *dstRow)
185{
186   const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
187   const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
188
189   assert(comps >= 1);
190   assert(comps <= 4);
191
192   /* This assertion is no longer valid with non-power-of-2 textures
193   assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
194   */
195
196   if (datatype == DTYPE_UBYTE && comps == 4) {
197      uint i, j, k;
198      const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA;
199      const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB;
200      ubyte(*dst)[4] = (ubyte(*)[4]) dstRow;
201      for (i = j = 0, k = k0; i < (uint) dstWidth;
202           i++, j += colStride, k += colStride) {
203         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
204         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
205         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
206         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
207      }
208   }
209   else if (datatype == DTYPE_UBYTE && comps == 3) {
210      uint i, j, k;
211      const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA;
212      const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB;
213      ubyte(*dst)[3] = (ubyte(*)[3]) dstRow;
214      for (i = j = 0, k = k0; i < (uint) dstWidth;
215           i++, j += colStride, k += colStride) {
216         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
217         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
218         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
219      }
220   }
221   else if (datatype == DTYPE_UBYTE && comps == 2) {
222      uint i, j, k;
223      const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA;
224      const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB;
225      ubyte(*dst)[2] = (ubyte(*)[2]) dstRow;
226      for (i = j = 0, k = k0; i < (uint) dstWidth;
227           i++, j += colStride, k += colStride) {
228         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
229         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
230      }
231   }
232   else if (datatype == DTYPE_UBYTE && comps == 1) {
233      uint i, j, k;
234      const ubyte *rowA = (const ubyte *) srcRowA;
235      const ubyte *rowB = (const ubyte *) srcRowB;
236      ubyte *dst = (ubyte *) dstRow;
237      for (i = j = 0, k = k0; i < (uint) dstWidth;
238           i++, j += colStride, k += colStride) {
239         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
240      }
241   }
242
243   else if (datatype == DTYPE_USHORT && comps == 4) {
244      uint i, j, k;
245      const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA;
246      const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB;
247      ushort(*dst)[4] = (ushort(*)[4]) dstRow;
248      for (i = j = 0, k = k0; i < (uint) dstWidth;
249           i++, j += colStride, k += colStride) {
250         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
251         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
252         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
253         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
254      }
255   }
256   else if (datatype == DTYPE_USHORT && comps == 3) {
257      uint i, j, k;
258      const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA;
259      const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB;
260      ushort(*dst)[3] = (ushort(*)[3]) dstRow;
261      for (i = j = 0, k = k0; i < (uint) dstWidth;
262           i++, j += colStride, k += colStride) {
263         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
264         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
265         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
266      }
267   }
268   else if (datatype == DTYPE_USHORT && comps == 2) {
269      uint i, j, k;
270      const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA;
271      const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB;
272      ushort(*dst)[2] = (ushort(*)[2]) dstRow;
273      for (i = j = 0, k = k0; i < (uint) dstWidth;
274           i++, j += colStride, k += colStride) {
275         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
276         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
277      }
278   }
279   else if (datatype == DTYPE_USHORT && comps == 1) {
280      uint i, j, k;
281      const ushort *rowA = (const ushort *) srcRowA;
282      const ushort *rowB = (const ushort *) srcRowB;
283      ushort *dst = (ushort *) dstRow;
284      for (i = j = 0, k = k0; i < (uint) dstWidth;
285           i++, j += colStride, k += colStride) {
286         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
287      }
288   }
289
290   else if (datatype == DTYPE_FLOAT && comps == 4) {
291      uint i, j, k;
292      const float(*rowA)[4] = (const float(*)[4]) srcRowA;
293      const float(*rowB)[4] = (const float(*)[4]) srcRowB;
294      float(*dst)[4] = (float(*)[4]) dstRow;
295      for (i = j = 0, k = k0; i < (uint) dstWidth;
296           i++, j += colStride, k += colStride) {
297         dst[i][0] = (rowA[j][0] + rowA[k][0] +
298                      rowB[j][0] + rowB[k][0]) * 0.25F;
299         dst[i][1] = (rowA[j][1] + rowA[k][1] +
300                      rowB[j][1] + rowB[k][1]) * 0.25F;
301         dst[i][2] = (rowA[j][2] + rowA[k][2] +
302                      rowB[j][2] + rowB[k][2]) * 0.25F;
303         dst[i][3] = (rowA[j][3] + rowA[k][3] +
304                      rowB[j][3] + rowB[k][3]) * 0.25F;
305      }
306   }
307   else if (datatype == DTYPE_FLOAT && comps == 3) {
308      uint i, j, k;
309      const float(*rowA)[3] = (const float(*)[3]) srcRowA;
310      const float(*rowB)[3] = (const float(*)[3]) srcRowB;
311      float(*dst)[3] = (float(*)[3]) dstRow;
312      for (i = j = 0, k = k0; i < (uint) dstWidth;
313           i++, j += colStride, k += colStride) {
314         dst[i][0] = (rowA[j][0] + rowA[k][0] +
315                      rowB[j][0] + rowB[k][0]) * 0.25F;
316         dst[i][1] = (rowA[j][1] + rowA[k][1] +
317                      rowB[j][1] + rowB[k][1]) * 0.25F;
318         dst[i][2] = (rowA[j][2] + rowA[k][2] +
319                      rowB[j][2] + rowB[k][2]) * 0.25F;
320      }
321   }
322   else if (datatype == DTYPE_FLOAT && comps == 2) {
323      uint i, j, k;
324      const float(*rowA)[2] = (const float(*)[2]) srcRowA;
325      const float(*rowB)[2] = (const float(*)[2]) srcRowB;
326      float(*dst)[2] = (float(*)[2]) dstRow;
327      for (i = j = 0, k = k0; i < (uint) dstWidth;
328           i++, j += colStride, k += colStride) {
329         dst[i][0] = (rowA[j][0] + rowA[k][0] +
330                      rowB[j][0] + rowB[k][0]) * 0.25F;
331         dst[i][1] = (rowA[j][1] + rowA[k][1] +
332                      rowB[j][1] + rowB[k][1]) * 0.25F;
333      }
334   }
335   else if (datatype == DTYPE_FLOAT && comps == 1) {
336      uint i, j, k;
337      const float *rowA = (const float *) srcRowA;
338      const float *rowB = (const float *) srcRowB;
339      float *dst = (float *) dstRow;
340      for (i = j = 0, k = k0; i < (uint) dstWidth;
341           i++, j += colStride, k += colStride) {
342         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
343      }
344   }
345
346#if 0
347   else if (datatype == HALF_DTYPE_FLOAT && comps == 4) {
348      uint i, j, k, comp;
349      const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA;
350      const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB;
351      half_float(*dst)[4] = (half_float(*)[4]) dstRow;
352      for (i = j = 0, k = k0; i < (uint) dstWidth;
353           i++, j += colStride, k += colStride) {
354         for (comp = 0; comp < 4; comp++) {
355            float aj, ak, bj, bk;
356            aj = half_to_float(rowA[j][comp]);
357            ak = half_to_float(rowA[k][comp]);
358            bj = half_to_float(rowB[j][comp]);
359            bk = half_to_float(rowB[k][comp]);
360            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
361         }
362      }
363   }
364   else if (datatype == DTYPE_HALF_FLOAT && comps == 3) {
365      uint i, j, k, comp;
366      const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA;
367      const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB;
368      half_float(*dst)[3] = (half_float(*)[3]) dstRow;
369      for (i = j = 0, k = k0; i < (uint) dstWidth;
370           i++, j += colStride, k += colStride) {
371         for (comp = 0; comp < 3; comp++) {
372            float aj, ak, bj, bk;
373            aj = half_to_float(rowA[j][comp]);
374            ak = half_to_float(rowA[k][comp]);
375            bj = half_to_float(rowB[j][comp]);
376            bk = half_to_float(rowB[k][comp]);
377            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
378         }
379      }
380   }
381   else if (datatype == DTYPE_HALF_FLOAT && comps == 2) {
382      uint i, j, k, comp;
383      const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA;
384      const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB;
385      half_float(*dst)[2] = (half_float(*)[2]) dstRow;
386      for (i = j = 0, k = k0; i < (uint) dstWidth;
387           i++, j += colStride, k += colStride) {
388         for (comp = 0; comp < 2; comp++) {
389            float aj, ak, bj, bk;
390            aj = half_to_float(rowA[j][comp]);
391            ak = half_to_float(rowA[k][comp]);
392            bj = half_to_float(rowB[j][comp]);
393            bk = half_to_float(rowB[k][comp]);
394            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
395         }
396      }
397   }
398   else if (datatype == DTYPE_HALF_FLOAT && comps == 1) {
399      uint i, j, k;
400      const half_float *rowA = (const half_float *) srcRowA;
401      const half_float *rowB = (const half_float *) srcRowB;
402      half_float *dst = (half_float *) dstRow;
403      for (i = j = 0, k = k0; i < (uint) dstWidth;
404           i++, j += colStride, k += colStride) {
405         float aj, ak, bj, bk;
406         aj = half_to_float(rowA[j]);
407         ak = half_to_float(rowA[k]);
408         bj = half_to_float(rowB[j]);
409         bk = half_to_float(rowB[k]);
410         dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F);
411      }
412   }
413#endif
414
415   else if (datatype == DTYPE_UINT && comps == 1) {
416      uint i, j, k;
417      const uint *rowA = (const uint *) srcRowA;
418      const uint *rowB = (const uint *) srcRowB;
419      uint *dst = (uint *) dstRow;
420      for (i = j = 0, k = k0; i < (uint) dstWidth;
421           i++, j += colStride, k += colStride) {
422         dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
423      }
424   }
425
426   else if (datatype == DTYPE_USHORT_5_6_5 && comps == 3) {
427      uint i, j, k;
428      const ushort *rowA = (const ushort *) srcRowA;
429      const ushort *rowB = (const ushort *) srcRowB;
430      ushort *dst = (ushort *) dstRow;
431      for (i = j = 0, k = k0; i < (uint) dstWidth;
432           i++, j += colStride, k += colStride) {
433         const int rowAr0 = rowA[j] & 0x1f;
434         const int rowAr1 = rowA[k] & 0x1f;
435         const int rowBr0 = rowB[j] & 0x1f;
436         const int rowBr1 = rowB[k] & 0x1f;
437         const int rowAg0 = (rowA[j] >> 5) & 0x3f;
438         const int rowAg1 = (rowA[k] >> 5) & 0x3f;
439         const int rowBg0 = (rowB[j] >> 5) & 0x3f;
440         const int rowBg1 = (rowB[k] >> 5) & 0x3f;
441         const int rowAb0 = (rowA[j] >> 11) & 0x1f;
442         const int rowAb1 = (rowA[k] >> 11) & 0x1f;
443         const int rowBb0 = (rowB[j] >> 11) & 0x1f;
444         const int rowBb1 = (rowB[k] >> 11) & 0x1f;
445         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
446         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
447         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
448         dst[i] = (blue << 11) | (green << 5) | red;
449      }
450   }
451   else if (datatype == DTYPE_USHORT_4_4_4_4 && comps == 4) {
452      uint i, j, k;
453      const ushort *rowA = (const ushort *) srcRowA;
454      const ushort *rowB = (const ushort *) srcRowB;
455      ushort *dst = (ushort *) dstRow;
456      for (i = j = 0, k = k0; i < (uint) dstWidth;
457           i++, j += colStride, k += colStride) {
458         const int rowAr0 = rowA[j] & 0xf;
459         const int rowAr1 = rowA[k] & 0xf;
460         const int rowBr0 = rowB[j] & 0xf;
461         const int rowBr1 = rowB[k] & 0xf;
462         const int rowAg0 = (rowA[j] >> 4) & 0xf;
463         const int rowAg1 = (rowA[k] >> 4) & 0xf;
464         const int rowBg0 = (rowB[j] >> 4) & 0xf;
465         const int rowBg1 = (rowB[k] >> 4) & 0xf;
466         const int rowAb0 = (rowA[j] >> 8) & 0xf;
467         const int rowAb1 = (rowA[k] >> 8) & 0xf;
468         const int rowBb0 = (rowB[j] >> 8) & 0xf;
469         const int rowBb1 = (rowB[k] >> 8) & 0xf;
470         const int rowAa0 = (rowA[j] >> 12) & 0xf;
471         const int rowAa1 = (rowA[k] >> 12) & 0xf;
472         const int rowBa0 = (rowB[j] >> 12) & 0xf;
473         const int rowBa1 = (rowB[k] >> 12) & 0xf;
474         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
475         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
476         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
477         const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
478         dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
479      }
480   }
481   else if (datatype == DTYPE_USHORT_1_5_5_5_REV && comps == 4) {
482      uint i, j, k;
483      const ushort *rowA = (const ushort *) srcRowA;
484      const ushort *rowB = (const ushort *) srcRowB;
485      ushort *dst = (ushort *) dstRow;
486      for (i = j = 0, k = k0; i < (uint) dstWidth;
487           i++, j += colStride, k += colStride) {
488         const int rowAr0 = rowA[j] & 0x1f;
489         const int rowAr1 = rowA[k] & 0x1f;
490         const int rowBr0 = rowB[j] & 0x1f;
491         const int rowBr1 = rowB[k] & 0x1f;
492         const int rowAg0 = (rowA[j] >> 5) & 0x1f;
493         const int rowAg1 = (rowA[k] >> 5) & 0x1f;
494         const int rowBg0 = (rowB[j] >> 5) & 0x1f;
495         const int rowBg1 = (rowB[k] >> 5) & 0x1f;
496         const int rowAb0 = (rowA[j] >> 10) & 0x1f;
497         const int rowAb1 = (rowA[k] >> 10) & 0x1f;
498         const int rowBb0 = (rowB[j] >> 10) & 0x1f;
499         const int rowBb1 = (rowB[k] >> 10) & 0x1f;
500         const int rowAa0 = (rowA[j] >> 15) & 0x1;
501         const int rowAa1 = (rowA[k] >> 15) & 0x1;
502         const int rowBa0 = (rowB[j] >> 15) & 0x1;
503         const int rowBa1 = (rowB[k] >> 15) & 0x1;
504         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
505         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
506         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
507         const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
508         dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
509      }
510   }
511   else if (datatype == DTYPE_UBYTE_3_3_2 && comps == 3) {
512      uint i, j, k;
513      const ubyte *rowA = (const ubyte *) srcRowA;
514      const ubyte *rowB = (const ubyte *) srcRowB;
515      ubyte *dst = (ubyte *) dstRow;
516      for (i = j = 0, k = k0; i < (uint) dstWidth;
517           i++, j += colStride, k += colStride) {
518         const int rowAr0 = rowA[j] & 0x3;
519         const int rowAr1 = rowA[k] & 0x3;
520         const int rowBr0 = rowB[j] & 0x3;
521         const int rowBr1 = rowB[k] & 0x3;
522         const int rowAg0 = (rowA[j] >> 2) & 0x7;
523         const int rowAg1 = (rowA[k] >> 2) & 0x7;
524         const int rowBg0 = (rowB[j] >> 2) & 0x7;
525         const int rowBg1 = (rowB[k] >> 2) & 0x7;
526         const int rowAb0 = (rowA[j] >> 5) & 0x7;
527         const int rowAb1 = (rowA[k] >> 5) & 0x7;
528         const int rowBb0 = (rowB[j] >> 5) & 0x7;
529         const int rowBb1 = (rowB[k] >> 5) & 0x7;
530         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
531         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
532         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
533         dst[i] = (blue << 5) | (green << 2) | red;
534      }
535   }
536   else {
537      debug_printf("bad format in do_row()");
538   }
539}
540
541
542/**
543 * Average together four rows of a source image to produce a single new
544 * row in the dest image.  It's legal for the two source rows to point
545 * to the same data.  The source width must be equal to either the
546 * dest width or two times the dest width.
547 *
548 * \param datatype  GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
549 *                  \c GL_FLOAT, etc.
550 * \param comps     number of components per pixel (1..4)
551 * \param srcWidth  Width of a row in the source data
552 * \param srcRowA   Pointer to one of the rows of source data
553 * \param srcRowB   Pointer to one of the rows of source data
554 * \param srcRowC   Pointer to one of the rows of source data
555 * \param srcRowD   Pointer to one of the rows of source data
556 * \param dstWidth  Width of a row in the destination data
557 * \param srcRowA   Pointer to the row of destination data
558 */
559static void
560do_row_3D(enum dtype datatype, uint comps, int srcWidth,
561          const void *srcRowA, const void *srcRowB,
562          const void *srcRowC, const void *srcRowD,
563          int dstWidth, void *dstRow)
564{
565   const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
566   const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
567   uint i, j, k;
568
569   assert(comps >= 1);
570   assert(comps <= 4);
571
572   if ((datatype == DTYPE_UBYTE) && (comps == 4)) {
573      DECLARE_ROW_POINTERS(ubyte, 4);
574
575      for (i = j = 0, k = k0; i < (uint) dstWidth;
576           i++, j += colStride, k += colStride) {
577         FILTER_3D(0);
578         FILTER_3D(1);
579         FILTER_3D(2);
580         FILTER_3D(3);
581      }
582   }
583   else if ((datatype == DTYPE_UBYTE) && (comps == 3)) {
584      DECLARE_ROW_POINTERS(ubyte, 3);
585
586      for (i = j = 0, k = k0; i < (uint) dstWidth;
587           i++, j += colStride, k += colStride) {
588         FILTER_3D(0);
589         FILTER_3D(1);
590         FILTER_3D(2);
591      }
592   }
593   else if ((datatype == DTYPE_UBYTE) && (comps == 2)) {
594      DECLARE_ROW_POINTERS(ubyte, 2);
595
596      for (i = j = 0, k = k0; i < (uint) dstWidth;
597           i++, j += colStride, k += colStride) {
598         FILTER_3D(0);
599         FILTER_3D(1);
600      }
601   }
602   else if ((datatype == DTYPE_UBYTE) && (comps == 1)) {
603      DECLARE_ROW_POINTERS(ubyte, 1);
604
605      for (i = j = 0, k = k0; i < (uint) dstWidth;
606           i++, j += colStride, k += colStride) {
607         FILTER_3D(0);
608      }
609   }
610   else if ((datatype == DTYPE_USHORT) && (comps == 4)) {
611      DECLARE_ROW_POINTERS(ushort, 4);
612
613      for (i = j = 0, k = k0; i < (uint) dstWidth;
614           i++, j += colStride, k += colStride) {
615         FILTER_3D(0);
616         FILTER_3D(1);
617         FILTER_3D(2);
618         FILTER_3D(3);
619      }
620   }
621   else if ((datatype == DTYPE_USHORT) && (comps == 3)) {
622      DECLARE_ROW_POINTERS(ushort, 3);
623
624      for (i = j = 0, k = k0; i < (uint) dstWidth;
625           i++, j += colStride, k += colStride) {
626         FILTER_3D(0);
627         FILTER_3D(1);
628         FILTER_3D(2);
629      }
630   }
631   else if ((datatype == DTYPE_USHORT) && (comps == 2)) {
632      DECLARE_ROW_POINTERS(ushort, 2);
633
634      for (i = j = 0, k = k0; i < (uint) dstWidth;
635           i++, j += colStride, k += colStride) {
636         FILTER_3D(0);
637         FILTER_3D(1);
638      }
639   }
640   else if ((datatype == DTYPE_USHORT) && (comps == 1)) {
641      DECLARE_ROW_POINTERS(ushort, 1);
642
643      for (i = j = 0, k = k0; i < (uint) dstWidth;
644           i++, j += colStride, k += colStride) {
645         FILTER_3D(0);
646      }
647   }
648   else if ((datatype == DTYPE_FLOAT) && (comps == 4)) {
649      DECLARE_ROW_POINTERS(float, 4);
650
651      for (i = j = 0, k = k0; i < (uint) dstWidth;
652           i++, j += colStride, k += colStride) {
653         FILTER_F_3D(0);
654         FILTER_F_3D(1);
655         FILTER_F_3D(2);
656         FILTER_F_3D(3);
657      }
658   }
659   else if ((datatype == DTYPE_FLOAT) && (comps == 3)) {
660      DECLARE_ROW_POINTERS(float, 3);
661
662      for (i = j = 0, k = k0; i < (uint) dstWidth;
663           i++, j += colStride, k += colStride) {
664         FILTER_F_3D(0);
665         FILTER_F_3D(1);
666         FILTER_F_3D(2);
667      }
668   }
669   else if ((datatype == DTYPE_FLOAT) && (comps == 2)) {
670      DECLARE_ROW_POINTERS(float, 2);
671
672      for (i = j = 0, k = k0; i < (uint) dstWidth;
673           i++, j += colStride, k += colStride) {
674         FILTER_F_3D(0);
675         FILTER_F_3D(1);
676      }
677   }
678   else if ((datatype == DTYPE_FLOAT) && (comps == 1)) {
679      DECLARE_ROW_POINTERS(float, 1);
680
681      for (i = j = 0, k = k0; i < (uint) dstWidth;
682           i++, j += colStride, k += colStride) {
683         FILTER_F_3D(0);
684      }
685   }
686   else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 4)) {
687      DECLARE_ROW_POINTERS(half_float, 4);
688
689      for (i = j = 0, k = k0; i < (uint) dstWidth;
690           i++, j += colStride, k += colStride) {
691         FILTER_HF_3D(0);
692         FILTER_HF_3D(1);
693         FILTER_HF_3D(2);
694         FILTER_HF_3D(3);
695      }
696   }
697   else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 3)) {
698      DECLARE_ROW_POINTERS(half_float, 4);
699
700      for (i = j = 0, k = k0; i < (uint) dstWidth;
701           i++, j += colStride, k += colStride) {
702         FILTER_HF_3D(0);
703         FILTER_HF_3D(1);
704         FILTER_HF_3D(2);
705      }
706   }
707   else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 2)) {
708      DECLARE_ROW_POINTERS(half_float, 4);
709
710      for (i = j = 0, k = k0; i < (uint) dstWidth;
711           i++, j += colStride, k += colStride) {
712         FILTER_HF_3D(0);
713         FILTER_HF_3D(1);
714      }
715   }
716   else if ((datatype == DTYPE_HALF_FLOAT) && (comps == 1)) {
717      DECLARE_ROW_POINTERS(half_float, 4);
718
719      for (i = j = 0, k = k0; i < (uint) dstWidth;
720           i++, j += colStride, k += colStride) {
721         FILTER_HF_3D(0);
722      }
723   }
724   else if ((datatype == DTYPE_UINT) && (comps == 1)) {
725      const uint *rowA = (const uint *) srcRowA;
726      const uint *rowB = (const uint *) srcRowB;
727      const uint *rowC = (const uint *) srcRowC;
728      const uint *rowD = (const uint *) srcRowD;
729      float *dst = (float *) dstRow;
730
731      for (i = j = 0, k = k0; i < (uint) dstWidth;
732           i++, j += colStride, k += colStride) {
733         const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k])
734                               + ((uint64_t) rowB[j] + (uint64_t) rowB[k])
735                               + ((uint64_t) rowC[j] + (uint64_t) rowC[k])
736                               + ((uint64_t) rowD[j] + (uint64_t) rowD[k]));
737         dst[i] = (float)((double) tmp * 0.125);
738      }
739   }
740   else if ((datatype == DTYPE_USHORT_5_6_5) && (comps == 3)) {
741      DECLARE_ROW_POINTERS0(ushort);
742
743      for (i = j = 0, k = k0; i < (uint) dstWidth;
744           i++, j += colStride, k += colStride) {
745         const int rowAr0 = rowA[j] & 0x1f;
746         const int rowAr1 = rowA[k] & 0x1f;
747         const int rowBr0 = rowB[j] & 0x1f;
748         const int rowBr1 = rowB[k] & 0x1f;
749         const int rowCr0 = rowC[j] & 0x1f;
750         const int rowCr1 = rowC[k] & 0x1f;
751         const int rowDr0 = rowD[j] & 0x1f;
752         const int rowDr1 = rowD[k] & 0x1f;
753         const int rowAg0 = (rowA[j] >> 5) & 0x3f;
754         const int rowAg1 = (rowA[k] >> 5) & 0x3f;
755         const int rowBg0 = (rowB[j] >> 5) & 0x3f;
756         const int rowBg1 = (rowB[k] >> 5) & 0x3f;
757         const int rowCg0 = (rowC[j] >> 5) & 0x3f;
758         const int rowCg1 = (rowC[k] >> 5) & 0x3f;
759         const int rowDg0 = (rowD[j] >> 5) & 0x3f;
760         const int rowDg1 = (rowD[k] >> 5) & 0x3f;
761         const int rowAb0 = (rowA[j] >> 11) & 0x1f;
762         const int rowAb1 = (rowA[k] >> 11) & 0x1f;
763         const int rowBb0 = (rowB[j] >> 11) & 0x1f;
764         const int rowBb1 = (rowB[k] >> 11) & 0x1f;
765         const int rowCb0 = (rowC[j] >> 11) & 0x1f;
766         const int rowCb1 = (rowC[k] >> 11) & 0x1f;
767         const int rowDb0 = (rowD[j] >> 11) & 0x1f;
768         const int rowDb1 = (rowD[k] >> 11) & 0x1f;
769         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
770                                       rowCr0, rowCr1, rowDr0, rowDr1);
771         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
772                                       rowCg0, rowCg1, rowDg0, rowDg1);
773         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
774                                       rowCb0, rowCb1, rowDb0, rowDb1);
775         dst[i] = (b << 11) | (g << 5) | r;
776      }
777   }
778   else if ((datatype == DTYPE_USHORT_4_4_4_4) && (comps == 4)) {
779      DECLARE_ROW_POINTERS0(ushort);
780
781      for (i = j = 0, k = k0; i < (uint) dstWidth;
782           i++, j += colStride, k += colStride) {
783         const int rowAr0 = rowA[j] & 0xf;
784         const int rowAr1 = rowA[k] & 0xf;
785         const int rowBr0 = rowB[j] & 0xf;
786         const int rowBr1 = rowB[k] & 0xf;
787         const int rowCr0 = rowC[j] & 0xf;
788         const int rowCr1 = rowC[k] & 0xf;
789         const int rowDr0 = rowD[j] & 0xf;
790         const int rowDr1 = rowD[k] & 0xf;
791         const int rowAg0 = (rowA[j] >> 4) & 0xf;
792         const int rowAg1 = (rowA[k] >> 4) & 0xf;
793         const int rowBg0 = (rowB[j] >> 4) & 0xf;
794         const int rowBg1 = (rowB[k] >> 4) & 0xf;
795         const int rowCg0 = (rowC[j] >> 4) & 0xf;
796         const int rowCg1 = (rowC[k] >> 4) & 0xf;
797         const int rowDg0 = (rowD[j] >> 4) & 0xf;
798         const int rowDg1 = (rowD[k] >> 4) & 0xf;
799         const int rowAb0 = (rowA[j] >> 8) & 0xf;
800         const int rowAb1 = (rowA[k] >> 8) & 0xf;
801         const int rowBb0 = (rowB[j] >> 8) & 0xf;
802         const int rowBb1 = (rowB[k] >> 8) & 0xf;
803         const int rowCb0 = (rowC[j] >> 8) & 0xf;
804         const int rowCb1 = (rowC[k] >> 8) & 0xf;
805         const int rowDb0 = (rowD[j] >> 8) & 0xf;
806         const int rowDb1 = (rowD[k] >> 8) & 0xf;
807         const int rowAa0 = (rowA[j] >> 12) & 0xf;
808         const int rowAa1 = (rowA[k] >> 12) & 0xf;
809         const int rowBa0 = (rowB[j] >> 12) & 0xf;
810         const int rowBa1 = (rowB[k] >> 12) & 0xf;
811         const int rowCa0 = (rowC[j] >> 12) & 0xf;
812         const int rowCa1 = (rowC[k] >> 12) & 0xf;
813         const int rowDa0 = (rowD[j] >> 12) & 0xf;
814         const int rowDa1 = (rowD[k] >> 12) & 0xf;
815         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
816                                       rowCr0, rowCr1, rowDr0, rowDr1);
817         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
818                                       rowCg0, rowCg1, rowDg0, rowDg1);
819         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
820                                       rowCb0, rowCb1, rowDb0, rowDb1);
821         const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
822                                       rowCa0, rowCa1, rowDa0, rowDa1);
823
824         dst[i] = (a << 12) | (b << 8) | (g << 4) | r;
825      }
826   }
827   else if ((datatype == DTYPE_USHORT_1_5_5_5_REV) && (comps == 4)) {
828      DECLARE_ROW_POINTERS0(ushort);
829
830      for (i = j = 0, k = k0; i < (uint) dstWidth;
831           i++, j += colStride, k += colStride) {
832         const int rowAr0 = rowA[j] & 0x1f;
833         const int rowAr1 = rowA[k] & 0x1f;
834         const int rowBr0 = rowB[j] & 0x1f;
835         const int rowBr1 = rowB[k] & 0x1f;
836         const int rowCr0 = rowC[j] & 0x1f;
837         const int rowCr1 = rowC[k] & 0x1f;
838         const int rowDr0 = rowD[j] & 0x1f;
839         const int rowDr1 = rowD[k] & 0x1f;
840         const int rowAg0 = (rowA[j] >> 5) & 0x1f;
841         const int rowAg1 = (rowA[k] >> 5) & 0x1f;
842         const int rowBg0 = (rowB[j] >> 5) & 0x1f;
843         const int rowBg1 = (rowB[k] >> 5) & 0x1f;
844         const int rowCg0 = (rowC[j] >> 5) & 0x1f;
845         const int rowCg1 = (rowC[k] >> 5) & 0x1f;
846         const int rowDg0 = (rowD[j] >> 5) & 0x1f;
847         const int rowDg1 = (rowD[k] >> 5) & 0x1f;
848         const int rowAb0 = (rowA[j] >> 10) & 0x1f;
849         const int rowAb1 = (rowA[k] >> 10) & 0x1f;
850         const int rowBb0 = (rowB[j] >> 10) & 0x1f;
851         const int rowBb1 = (rowB[k] >> 10) & 0x1f;
852         const int rowCb0 = (rowC[j] >> 10) & 0x1f;
853         const int rowCb1 = (rowC[k] >> 10) & 0x1f;
854         const int rowDb0 = (rowD[j] >> 10) & 0x1f;
855         const int rowDb1 = (rowD[k] >> 10) & 0x1f;
856         const int rowAa0 = (rowA[j] >> 15) & 0x1;
857         const int rowAa1 = (rowA[k] >> 15) & 0x1;
858         const int rowBa0 = (rowB[j] >> 15) & 0x1;
859         const int rowBa1 = (rowB[k] >> 15) & 0x1;
860         const int rowCa0 = (rowC[j] >> 15) & 0x1;
861         const int rowCa1 = (rowC[k] >> 15) & 0x1;
862         const int rowDa0 = (rowD[j] >> 15) & 0x1;
863         const int rowDa1 = (rowD[k] >> 15) & 0x1;
864         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
865                                       rowCr0, rowCr1, rowDr0, rowDr1);
866         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
867                                       rowCg0, rowCg1, rowDg0, rowDg1);
868         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
869                                       rowCb0, rowCb1, rowDb0, rowDb1);
870         const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
871                                       rowCa0, rowCa1, rowDa0, rowDa1);
872
873         dst[i] = (a << 15) | (b << 10) | (g << 5) | r;
874      }
875   }
876   else if ((datatype == DTYPE_UBYTE_3_3_2) && (comps == 3)) {
877      DECLARE_ROW_POINTERS0(ushort);
878
879      for (i = j = 0, k = k0; i < (uint) dstWidth;
880           i++, j += colStride, k += colStride) {
881         const int rowAr0 = rowA[j] & 0x3;
882         const int rowAr1 = rowA[k] & 0x3;
883         const int rowBr0 = rowB[j] & 0x3;
884         const int rowBr1 = rowB[k] & 0x3;
885         const int rowCr0 = rowC[j] & 0x3;
886         const int rowCr1 = rowC[k] & 0x3;
887         const int rowDr0 = rowD[j] & 0x3;
888         const int rowDr1 = rowD[k] & 0x3;
889         const int rowAg0 = (rowA[j] >> 2) & 0x7;
890         const int rowAg1 = (rowA[k] >> 2) & 0x7;
891         const int rowBg0 = (rowB[j] >> 2) & 0x7;
892         const int rowBg1 = (rowB[k] >> 2) & 0x7;
893         const int rowCg0 = (rowC[j] >> 2) & 0x7;
894         const int rowCg1 = (rowC[k] >> 2) & 0x7;
895         const int rowDg0 = (rowD[j] >> 2) & 0x7;
896         const int rowDg1 = (rowD[k] >> 2) & 0x7;
897         const int rowAb0 = (rowA[j] >> 5) & 0x7;
898         const int rowAb1 = (rowA[k] >> 5) & 0x7;
899         const int rowBb0 = (rowB[j] >> 5) & 0x7;
900         const int rowBb1 = (rowB[k] >> 5) & 0x7;
901         const int rowCb0 = (rowC[j] >> 5) & 0x7;
902         const int rowCb1 = (rowC[k] >> 5) & 0x7;
903         const int rowDb0 = (rowD[j] >> 5) & 0x7;
904         const int rowDb1 = (rowD[k] >> 5) & 0x7;
905         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
906                                       rowCr0, rowCr1, rowDr0, rowDr1);
907         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
908                                       rowCg0, rowCg1, rowDg0, rowDg1);
909         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
910                                       rowCb0, rowCb1, rowDb0, rowDb1);
911         dst[i] = (b << 5) | (g << 2) | r;
912      }
913   }
914   else {
915      debug_printf("bad format in do_row_3D()");
916   }
917}
918
919
920
921static void
922format_to_type_comps(enum pipe_format pformat,
923                     enum dtype *datatype, uint *comps)
924{
925   /* XXX I think this could be implemented in terms of the pf_*() functions */
926   switch (pformat) {
927   case PIPE_FORMAT_B8G8R8A8_UNORM:
928   case PIPE_FORMAT_B8G8R8X8_UNORM:
929   case PIPE_FORMAT_A8R8G8B8_UNORM:
930   case PIPE_FORMAT_X8R8G8B8_UNORM:
931   case PIPE_FORMAT_A8B8G8R8_SRGB:
932   case PIPE_FORMAT_X8B8G8R8_SRGB:
933   case PIPE_FORMAT_B8G8R8A8_SRGB:
934   case PIPE_FORMAT_B8G8R8X8_SRGB:
935   case PIPE_FORMAT_A8R8G8B8_SRGB:
936   case PIPE_FORMAT_X8R8G8B8_SRGB:
937   case PIPE_FORMAT_R8G8B8_SRGB:
938      *datatype = DTYPE_UBYTE;
939      *comps = 4;
940      return;
941   case PIPE_FORMAT_B5G5R5A1_UNORM:
942      *datatype = DTYPE_USHORT_1_5_5_5_REV;
943      *comps = 4;
944      return;
945   case PIPE_FORMAT_B4G4R4A4_UNORM:
946      *datatype = DTYPE_USHORT_4_4_4_4;
947      *comps = 4;
948      return;
949   case PIPE_FORMAT_B5G6R5_UNORM:
950      *datatype = DTYPE_USHORT_5_6_5;
951      *comps = 3;
952      return;
953   case PIPE_FORMAT_L8_UNORM:
954   case PIPE_FORMAT_L8_SRGB:
955   case PIPE_FORMAT_A8_UNORM:
956   case PIPE_FORMAT_I8_UNORM:
957      *datatype = DTYPE_UBYTE;
958      *comps = 1;
959      return;
960   case PIPE_FORMAT_L8A8_UNORM:
961   case PIPE_FORMAT_L8A8_SRGB:
962      *datatype = DTYPE_UBYTE;
963      *comps = 2;
964      return;
965   default:
966      assert(0);
967      *datatype = DTYPE_UBYTE;
968      *comps = 0;
969      break;
970   }
971}
972
973
974static void
975reduce_1d(enum pipe_format pformat,
976          int srcWidth, const ubyte *srcPtr,
977          int dstWidth, ubyte *dstPtr)
978{
979   enum dtype datatype;
980   uint comps;
981
982   format_to_type_comps(pformat, &datatype, &comps);
983
984   /* we just duplicate the input row, kind of hack, saves code */
985   do_row(datatype, comps,
986          srcWidth, srcPtr, srcPtr,
987          dstWidth, dstPtr);
988}
989
990
991/**
992 * Strides are in bytes.  If zero, it'll be computed as width * bpp.
993 */
994static void
995reduce_2d(enum pipe_format pformat,
996          int srcWidth, int srcHeight,
997          int srcRowStride, const ubyte *srcPtr,
998          int dstWidth, int dstHeight,
999          int dstRowStride, ubyte *dstPtr)
1000{
1001   enum dtype datatype;
1002   uint comps;
1003   const int bpt = util_format_get_blocksize(pformat);
1004   const ubyte *srcA, *srcB;
1005   ubyte *dst;
1006   int row;
1007
1008   format_to_type_comps(pformat, &datatype, &comps);
1009
1010   if (!srcRowStride)
1011      srcRowStride = bpt * srcWidth;
1012
1013   if (!dstRowStride)
1014      dstRowStride = bpt * dstWidth;
1015
1016   /* Compute src and dst pointers */
1017   srcA = srcPtr;
1018   if (srcHeight > 1)
1019      srcB = srcA + srcRowStride;
1020   else
1021      srcB = srcA;
1022   dst = dstPtr;
1023
1024   for (row = 0; row < dstHeight; row++) {
1025      do_row(datatype, comps,
1026             srcWidth, srcA, srcB,
1027             dstWidth, dst);
1028      srcA += 2 * srcRowStride;
1029      srcB += 2 * srcRowStride;
1030      dst += dstRowStride;
1031   }
1032}
1033
1034
1035static void
1036reduce_3d(enum pipe_format pformat,
1037          int srcWidth, int srcHeight, int srcDepth,
1038          int srcRowStride, const ubyte *srcPtr,
1039          int dstWidth, int dstHeight, int dstDepth,
1040          int dstRowStride, ubyte *dstPtr)
1041{
1042   const int bpt = util_format_get_blocksize(pformat);
1043   const int border = 0;
1044   int img, row;
1045   int bytesPerSrcImage, bytesPerDstImage;
1046   int bytesPerSrcRow, bytesPerDstRow;
1047   int srcImageOffset, srcRowOffset;
1048   enum dtype datatype;
1049   uint comps;
1050
1051   format_to_type_comps(pformat, &datatype, &comps);
1052
1053   bytesPerSrcImage = srcWidth * srcHeight * bpt;
1054   bytesPerDstImage = dstWidth * dstHeight * bpt;
1055
1056   bytesPerSrcRow = srcWidth * bpt;
1057   bytesPerDstRow = dstWidth * bpt;
1058
1059   /* Offset between adjacent src images to be averaged together */
1060   srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage;
1061
1062   /* Offset between adjacent src rows to be averaged together */
1063   srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt;
1064
1065   /*
1066    * Need to average together up to 8 src pixels for each dest pixel.
1067    * Break that down into 3 operations:
1068    *   1. take two rows from source image and average them together.
1069    *   2. take two rows from next source image and average them together.
1070    *   3. take the two averaged rows and average them for the final dst row.
1071    */
1072
1073   /*
1074   printf("mip3d %d x %d x %d  ->  %d x %d x %d\n",
1075          srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
1076   */
1077
1078   for (img = 0; img < dstDepth; img++) {
1079      /* first source image pointer, skipping border */
1080      const ubyte *imgSrcA = srcPtr
1081         + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border
1082         + img * (bytesPerSrcImage + srcImageOffset);
1083      /* second source image pointer, skipping border */
1084      const ubyte *imgSrcB = imgSrcA + srcImageOffset;
1085      /* address of the dest image, skipping border */
1086      ubyte *imgDst = dstPtr
1087         + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border
1088         + img * bytesPerDstImage;
1089
1090      /* setup the four source row pointers and the dest row pointer */
1091      const ubyte *srcImgARowA = imgSrcA;
1092      const ubyte *srcImgARowB = imgSrcA + srcRowOffset;
1093      const ubyte *srcImgBRowA = imgSrcB;
1094      const ubyte *srcImgBRowB = imgSrcB + srcRowOffset;
1095      ubyte *dstImgRow = imgDst;
1096
1097      for (row = 0; row < dstHeight; row++) {
1098         do_row_3D(datatype, comps, srcWidth,
1099                   srcImgARowA, srcImgARowB,
1100                   srcImgBRowA, srcImgBRowB,
1101                   dstWidth, dstImgRow);
1102
1103         /* advance to next rows */
1104         srcImgARowA += bytesPerSrcRow + srcRowOffset;
1105         srcImgARowB += bytesPerSrcRow + srcRowOffset;
1106         srcImgBRowA += bytesPerSrcRow + srcRowOffset;
1107         srcImgBRowB += bytesPerSrcRow + srcRowOffset;
1108         dstImgRow += bytesPerDstRow;
1109      }
1110   }
1111}
1112
1113
1114
1115
1116static void
1117make_1d_mipmap(struct gen_mipmap_state *ctx,
1118               struct pipe_texture *pt,
1119               uint face, uint baseLevel, uint lastLevel)
1120{
1121   struct pipe_context *pipe = ctx->pipe;
1122   const uint zslice = 0;
1123   uint dstLevel;
1124
1125   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1126      const uint srcLevel = dstLevel - 1;
1127      struct pipe_transfer *srcTrans, *dstTrans;
1128      void *srcMap, *dstMap;
1129
1130      srcTrans = pipe->get_tex_transfer(pipe, pt, face, srcLevel, zslice,
1131                                          PIPE_TRANSFER_READ, 0, 0,
1132                                          u_minify(pt->width0, srcLevel),
1133                                          u_minify(pt->height0, srcLevel));
1134      dstTrans = pipe->get_tex_transfer(pipe, pt, face, dstLevel, zslice,
1135                                          PIPE_TRANSFER_WRITE, 0, 0,
1136                                          u_minify(pt->width0, dstLevel),
1137                                          u_minify(pt->height0, dstLevel));
1138
1139      srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1140      dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1141
1142      reduce_1d(pt->format,
1143                srcTrans->width, srcMap,
1144                dstTrans->width, dstMap);
1145
1146      pipe->transfer_unmap(pipe, srcTrans);
1147      pipe->transfer_unmap(pipe, dstTrans);
1148
1149      pipe->tex_transfer_destroy(pipe, srcTrans);
1150      pipe->tex_transfer_destroy(pipe, dstTrans);
1151   }
1152}
1153
1154
1155static void
1156make_2d_mipmap(struct gen_mipmap_state *ctx,
1157               struct pipe_texture *pt,
1158               uint face, uint baseLevel, uint lastLevel)
1159{
1160   struct pipe_context *pipe = ctx->pipe;
1161   const uint zslice = 0;
1162   uint dstLevel;
1163
1164   assert(util_format_get_blockwidth(pt->format) == 1);
1165   assert(util_format_get_blockheight(pt->format) == 1);
1166
1167   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1168      const uint srcLevel = dstLevel - 1;
1169      struct pipe_transfer *srcTrans, *dstTrans;
1170      ubyte *srcMap, *dstMap;
1171
1172      srcTrans = pipe->get_tex_transfer(pipe, pt, face, srcLevel, zslice,
1173                                          PIPE_TRANSFER_READ, 0, 0,
1174                                          u_minify(pt->width0, srcLevel),
1175                                          u_minify(pt->height0, srcLevel));
1176      dstTrans = pipe->get_tex_transfer(pipe, pt, face, dstLevel, zslice,
1177                                          PIPE_TRANSFER_WRITE, 0, 0,
1178                                          u_minify(pt->width0, dstLevel),
1179                                          u_minify(pt->height0, dstLevel));
1180
1181      srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1182      dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1183
1184      reduce_2d(pt->format,
1185                srcTrans->width, srcTrans->height,
1186                srcTrans->stride, srcMap,
1187                dstTrans->width, dstTrans->height,
1188                dstTrans->stride, dstMap);
1189
1190      pipe->transfer_unmap(pipe, srcTrans);
1191      pipe->transfer_unmap(pipe, dstTrans);
1192
1193      pipe->tex_transfer_destroy(pipe, srcTrans);
1194      pipe->tex_transfer_destroy(pipe, dstTrans);
1195   }
1196}
1197
1198
1199static void
1200make_3d_mipmap(struct gen_mipmap_state *ctx,
1201               struct pipe_texture *pt,
1202               uint face, uint baseLevel, uint lastLevel)
1203{
1204#if 0
1205   struct pipe_context *pipe = ctx->pipe;
1206   struct pipe_screen *screen = pipe->screen;
1207   uint dstLevel, zslice = 0;
1208
1209   assert(util_format_get_blockwidth(pt->format) == 1);
1210   assert(util_format_get_blockheight(pt->format) == 1);
1211
1212   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1213      const uint srcLevel = dstLevel - 1;
1214      struct pipe_transfer *srcTrans, *dstTrans;
1215      ubyte *srcMap, *dstMap;
1216
1217      srcTrans = pipe->get_tex_transfer(pipe, pt, face, srcLevel, zslice,
1218                                          PIPE_TRANSFER_READ, 0, 0,
1219                                          u_minify(pt->width0, srcLevel),
1220                                          u_minify(pt->height0, srcLevel));
1221      dstTrans = pipe->get_tex_transfer(pipe, pt, face, dstLevel, zslice,
1222                                          PIPE_TRANSFER_WRITE, 0, 0,
1223                                          u_minify(pt->width0, dstLevel),
1224                                          u_minify(pt->height0, dstLevel));
1225
1226      srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
1227      dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
1228
1229      reduce_3d(pt->format,
1230                srcTrans->width, srcTrans->height,
1231                srcTrans->stride, srcMap,
1232                dstTrans->width, dstTrans->height,
1233                dstTrans->stride, dstMap);
1234
1235      pipe->transfer_unmap(pipe, srcTrans);
1236      pipe->transfer_unmap(pipe, dstTrans);
1237
1238      pipe->tex_transfer_destroy(pipe, srcTrans);
1239      pipe->tex_transfer_destroy(pipe, dstTrans);
1240   }
1241#else
1242   (void) reduce_3d;
1243#endif
1244}
1245
1246
1247static void
1248fallback_gen_mipmap(struct gen_mipmap_state *ctx,
1249                    struct pipe_texture *pt,
1250                    uint face, uint baseLevel, uint lastLevel)
1251{
1252   switch (pt->target) {
1253   case PIPE_TEXTURE_1D:
1254      make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1255      break;
1256   case PIPE_TEXTURE_2D:
1257   case PIPE_TEXTURE_CUBE:
1258      make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1259      break;
1260   case PIPE_TEXTURE_3D:
1261      make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel);
1262      break;
1263   default:
1264      assert(0);
1265   }
1266}
1267
1268
1269/**
1270 * Create a mipmap generation context.
1271 * The idea is to create one of these and re-use it each time we need to
1272 * generate a mipmap.
1273 */
1274struct gen_mipmap_state *
1275util_create_gen_mipmap(struct pipe_context *pipe,
1276                       struct cso_context *cso)
1277{
1278   struct gen_mipmap_state *ctx;
1279   uint i;
1280
1281   ctx = CALLOC_STRUCT(gen_mipmap_state);
1282   if (!ctx)
1283      return NULL;
1284
1285   ctx->pipe = pipe;
1286   ctx->cso = cso;
1287
1288   /* disabled blending/masking */
1289   memset(&ctx->blend, 0, sizeof(ctx->blend));
1290   ctx->blend.rt[0].colormask = PIPE_MASK_RGBA;
1291
1292   /* no-op depth/stencil/alpha */
1293   memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
1294
1295   /* rasterizer */
1296   memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
1297   ctx->rasterizer.front_winding = PIPE_WINDING_CW;
1298   ctx->rasterizer.cull_mode = PIPE_WINDING_NONE;
1299   ctx->rasterizer.gl_rasterization_rules = 1;
1300
1301   /* sampler state */
1302   memset(&ctx->sampler, 0, sizeof(ctx->sampler));
1303   ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1304   ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1305   ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1306   ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
1307   ctx->sampler.normalized_coords = 1;
1308
1309   /* vertex elements state */
1310   memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
1311   for (i = 0; i < 2; i++) {
1312      ctx->velem[i].src_offset = i * 4 * sizeof(float);
1313      ctx->velem[i].instance_divisor = 0;
1314      ctx->velem[i].vertex_buffer_index = 0;
1315      ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
1316   }
1317
1318   /* vertex shader - still needed to specify mapping from fragment
1319    * shader input semantics to vertex elements
1320    */
1321   {
1322      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
1323                                      TGSI_SEMANTIC_GENERIC };
1324      const uint semantic_indexes[] = { 0, 0 };
1325      ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
1326                                                    semantic_indexes);
1327   }
1328
1329   /* fragment shader */
1330   ctx->fs2d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_2D);
1331   ctx->fsCube = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_CUBE);
1332
1333   /* vertex data that doesn't change */
1334   for (i = 0; i < 4; i++) {
1335      ctx->vertices[i][0][2] = 0.0f; /* z */
1336      ctx->vertices[i][0][3] = 1.0f; /* w */
1337      ctx->vertices[i][1][3] = 1.0f; /* q */
1338   }
1339
1340   /* Note: the actual vertex buffer is allocated as needed below */
1341
1342   return ctx;
1343}
1344
1345
1346/**
1347 * Get next "slot" of vertex space in the vertex buffer.
1348 * We're allocating one large vertex buffer and using it piece by piece.
1349 */
1350static unsigned
1351get_next_slot(struct gen_mipmap_state *ctx)
1352{
1353   const unsigned max_slots = 4096 / sizeof ctx->vertices;
1354
1355   if (ctx->vbuf_slot >= max_slots)
1356      util_gen_mipmap_flush( ctx );
1357
1358   if (!ctx->vbuf) {
1359      ctx->vbuf = pipe_buffer_create(ctx->pipe->screen,
1360                                     32,
1361                                     PIPE_BUFFER_USAGE_VERTEX,
1362                                     max_slots * sizeof ctx->vertices);
1363   }
1364
1365   return ctx->vbuf_slot++ * sizeof ctx->vertices;
1366}
1367
1368
1369static unsigned
1370set_vertex_data(struct gen_mipmap_state *ctx,
1371                enum pipe_texture_target tex_target,
1372                uint face)
1373{
1374   unsigned offset;
1375
1376   /* vert[0].position */
1377   ctx->vertices[0][0][0] = -1.0f; /*x*/
1378   ctx->vertices[0][0][1] = -1.0f; /*y*/
1379
1380   /* vert[1].position */
1381   ctx->vertices[1][0][0] = 1.0f;
1382   ctx->vertices[1][0][1] = -1.0f;
1383
1384   /* vert[2].position */
1385   ctx->vertices[2][0][0] = 1.0f;
1386   ctx->vertices[2][0][1] = 1.0f;
1387
1388   /* vert[3].position */
1389   ctx->vertices[3][0][0] = -1.0f;
1390   ctx->vertices[3][0][1] = 1.0f;
1391
1392   /* Setup vertex texcoords.  This is a little tricky for cube maps. */
1393   if (tex_target == PIPE_TEXTURE_CUBE) {
1394      static const float st[4][2] = {
1395         {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
1396      };
1397
1398      util_map_texcoords2d_onto_cubemap(face, &st[0][0], 2,
1399                                        &ctx->vertices[0][1][0], 8);
1400   }
1401   else {
1402      /* 1D/2D */
1403      ctx->vertices[0][1][0] = 0.0f; /*s*/
1404      ctx->vertices[0][1][1] = 0.0f; /*t*/
1405      ctx->vertices[0][1][2] = 0.0f; /*r*/
1406
1407      ctx->vertices[1][1][0] = 1.0f;
1408      ctx->vertices[1][1][1] = 0.0f;
1409      ctx->vertices[1][1][2] = 0.0f;
1410
1411      ctx->vertices[2][1][0] = 1.0f;
1412      ctx->vertices[2][1][1] = 1.0f;
1413      ctx->vertices[2][1][2] = 0.0f;
1414
1415      ctx->vertices[3][1][0] = 0.0f;
1416      ctx->vertices[3][1][1] = 1.0f;
1417      ctx->vertices[3][1][2] = 0.0f;
1418   }
1419
1420   offset = get_next_slot( ctx );
1421
1422   pipe_buffer_write_nooverlap(ctx->pipe->screen, ctx->vbuf,
1423                               offset, sizeof(ctx->vertices), ctx->vertices);
1424
1425   return offset;
1426}
1427
1428
1429
1430/**
1431 * Destroy a mipmap generation context
1432 */
1433void
1434util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
1435{
1436   struct pipe_context *pipe = ctx->pipe;
1437
1438   pipe->delete_vs_state(pipe, ctx->vs);
1439   pipe->delete_fs_state(pipe, ctx->fs2d);
1440   pipe->delete_fs_state(pipe, ctx->fsCube);
1441
1442   pipe_buffer_reference(&ctx->vbuf, NULL);
1443
1444   FREE(ctx);
1445}
1446
1447
1448
1449/* Release vertex buffer at end of frame to avoid synchronous
1450 * rendering.
1451 */
1452void util_gen_mipmap_flush( struct gen_mipmap_state *ctx )
1453{
1454   pipe_buffer_reference(&ctx->vbuf, NULL);
1455   ctx->vbuf_slot = 0;
1456}
1457
1458
1459/**
1460 * Generate mipmap images.  It's assumed all needed texture memory is
1461 * already allocated.
1462 *
1463 * \param pt  the texture to generate mipmap levels for
1464 * \param face  which cube face to generate mipmaps for (0 for non-cube maps)
1465 * \param baseLevel  the first mipmap level to use as a src
1466 * \param lastLevel  the last mipmap level to generate
1467 * \param filter  the minification filter used to generate mipmap levels with
1468 * \param filter  one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST
1469 */
1470void
1471util_gen_mipmap(struct gen_mipmap_state *ctx,
1472                struct pipe_texture *pt,
1473                uint face, uint baseLevel, uint lastLevel, uint filter)
1474{
1475   struct pipe_context *pipe = ctx->pipe;
1476   struct pipe_screen *screen = pipe->screen;
1477   struct pipe_framebuffer_state fb;
1478   void *fs = (pt->target == PIPE_TEXTURE_CUBE) ? ctx->fsCube : ctx->fs2d;
1479   uint dstLevel;
1480   uint zslice = 0;
1481   uint offset;
1482
1483   /* The texture object should have room for the levels which we're
1484    * about to generate.
1485    */
1486   assert(lastLevel <= pt->last_level);
1487
1488   /* If this fails, why are we here? */
1489   assert(lastLevel > baseLevel);
1490
1491   assert(filter == PIPE_TEX_FILTER_LINEAR ||
1492          filter == PIPE_TEX_FILTER_NEAREST);
1493
1494   /* check if we can render in the texture's format */
1495   if (!screen->is_format_supported(screen, pt->format, PIPE_TEXTURE_2D,
1496                                    PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) {
1497      fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
1498      return;
1499   }
1500
1501   /* save state (restored below) */
1502   cso_save_blend(ctx->cso);
1503   cso_save_depth_stencil_alpha(ctx->cso);
1504   cso_save_rasterizer(ctx->cso);
1505   cso_save_samplers(ctx->cso);
1506   cso_save_sampler_textures(ctx->cso);
1507   cso_save_framebuffer(ctx->cso);
1508   cso_save_fragment_shader(ctx->cso);
1509   cso_save_vertex_shader(ctx->cso);
1510   cso_save_viewport(ctx->cso);
1511   cso_save_clip(ctx->cso);
1512   cso_save_vertex_elements(ctx->cso);
1513
1514   /* bind our state */
1515   cso_set_blend(ctx->cso, &ctx->blend);
1516   cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
1517   cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
1518   cso_set_clip(ctx->cso, &ctx->clip);
1519   cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
1520
1521   cso_set_fragment_shader_handle(ctx->cso, fs);
1522   cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
1523
1524   /* init framebuffer state */
1525   memset(&fb, 0, sizeof(fb));
1526   fb.nr_cbufs = 1;
1527
1528   /* set min/mag to same filter for faster sw speed */
1529   ctx->sampler.mag_img_filter = filter;
1530   ctx->sampler.min_img_filter = filter;
1531
1532   /*
1533    * XXX for small mipmap levels, it may be faster to use the software
1534    * fallback path...
1535    */
1536   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
1537      const uint srcLevel = dstLevel - 1;
1538      struct pipe_viewport_state vp;
1539
1540      struct pipe_surface *surf =
1541         screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
1542                                 PIPE_BUFFER_USAGE_GPU_WRITE);
1543
1544      /*
1545       * Setup framebuffer / dest surface
1546       */
1547      fb.cbufs[0] = surf;
1548      fb.width = u_minify(pt->width0, dstLevel);
1549      fb.height = u_minify(pt->height0, dstLevel);
1550      cso_set_framebuffer(ctx->cso, &fb);
1551
1552      /* viewport */
1553      vp.scale[0] = 0.5f * fb.width;
1554      vp.scale[1] = 0.5f * fb.height;
1555      vp.scale[2] = 1.0f;
1556      vp.scale[3] = 1.0f;
1557      vp.translate[0] = 0.5f * fb.width;
1558      vp.translate[1] = 0.5f * fb.height;
1559      vp.translate[2] = 0.0f;
1560      vp.translate[3] = 0.0f;
1561      cso_set_viewport(ctx->cso, &vp);
1562
1563      /*
1564       * Setup sampler state
1565       * Note: we should only have to set the min/max LOD clamps to ensure
1566       * we grab texels from the right mipmap level.  But some hardware
1567       * has trouble with min clamping so we also set the lod_bias to
1568       * try to work around that.
1569       */
1570      ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel;
1571      ctx->sampler.lod_bias = (float) srcLevel;
1572      cso_single_sampler(ctx->cso, 0, &ctx->sampler);
1573      cso_single_sampler_done(ctx->cso);
1574
1575      cso_set_sampler_textures(ctx->cso, 1, &pt);
1576
1577      /* quad coords in clip coords */
1578      offset = set_vertex_data(ctx,
1579                               pt->target,
1580                               face);
1581
1582      util_draw_vertex_buffer(ctx->pipe,
1583                              ctx->vbuf,
1584                              offset,
1585                              PIPE_PRIM_TRIANGLE_FAN,
1586                              4,  /* verts */
1587                              2); /* attribs/vert */
1588
1589      pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
1590
1591      /* need to signal that the texture has changed _after_ rendering to it */
1592      pipe_surface_reference( &surf, NULL );
1593   }
1594
1595   /* restore state we changed */
1596   cso_restore_blend(ctx->cso);
1597   cso_restore_depth_stencil_alpha(ctx->cso);
1598   cso_restore_rasterizer(ctx->cso);
1599   cso_restore_samplers(ctx->cso);
1600   cso_restore_sampler_textures(ctx->cso);
1601   cso_restore_framebuffer(ctx->cso);
1602   cso_restore_fragment_shader(ctx->cso);
1603   cso_restore_vertex_shader(ctx->cso);
1604   cso_restore_viewport(ctx->cso);
1605   cso_restore_clip(ctx->cso);
1606   cso_restore_vertex_elements(ctx->cso);
1607}
1608