r300g: implement fast color clear
[platform/upstream/mesa.git] / src / gallium / drivers / r300 / r300_texture.c
1 /*
2  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3  * Copyright 2010 Marek Olšák <maraeo@gmail.com>
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * on the rights to use, copy, modify, merge, publish, distribute, sub
9  * license, and/or sell copies of the Software, and to permit persons to whom
10  * the Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the next
13  * paragraph) shall be included in all copies or substantial portions of the
14  * Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24 /* Always include headers in the reverse order!! ~ M. */
25 #include "r300_texture.h"
26
27 #include "r300_context.h"
28 #include "r300_reg.h"
29 #include "r300_transfer.h"
30 #include "r300_screen.h"
31 #include "r300_winsys.h"
32
33 #include "util/u_format.h"
34 #include "util/u_format_s3tc.h"
35 #include "util/u_math.h"
36 #include "util/u_memory.h"
37
38 #include "pipe/p_screen.h"
39
40 enum r300_dim {
41     DIM_WIDTH  = 0,
42     DIM_HEIGHT = 1
43 };
44
45 unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
46                                    const unsigned char *swizzle_view)
47 {
48     unsigned i;
49     unsigned char swizzle[4];
50     unsigned result = 0;
51     const uint32_t swizzle_shift[4] = {
52         R300_TX_FORMAT_R_SHIFT,
53         R300_TX_FORMAT_G_SHIFT,
54         R300_TX_FORMAT_B_SHIFT,
55         R300_TX_FORMAT_A_SHIFT
56     };
57     const uint32_t swizzle_bit[4] = {
58         R300_TX_FORMAT_X,
59         R300_TX_FORMAT_Y,
60         R300_TX_FORMAT_Z,
61         R300_TX_FORMAT_W
62     };
63
64     if (swizzle_view) {
65         /* Combine two sets of swizzles. */
66         for (i = 0; i < 4; i++) {
67             swizzle[i] = swizzle_view[i] <= UTIL_FORMAT_SWIZZLE_W ?
68                          swizzle_format[swizzle_view[i]] : swizzle_view[i];
69         }
70     } else {
71         memcpy(swizzle, swizzle_format, 4);
72     }
73
74     /* Get swizzle. */
75     for (i = 0; i < 4; i++) {
76         switch (swizzle[i]) {
77             case UTIL_FORMAT_SWIZZLE_Y:
78                 result |= swizzle_bit[1] << swizzle_shift[i];
79                 break;
80             case UTIL_FORMAT_SWIZZLE_Z:
81                 result |= swizzle_bit[2] << swizzle_shift[i];
82                 break;
83             case UTIL_FORMAT_SWIZZLE_W:
84                 result |= swizzle_bit[3] << swizzle_shift[i];
85                 break;
86             case UTIL_FORMAT_SWIZZLE_0:
87                 result |= R300_TX_FORMAT_ZERO << swizzle_shift[i];
88                 break;
89             case UTIL_FORMAT_SWIZZLE_1:
90                 result |= R300_TX_FORMAT_ONE << swizzle_shift[i];
91                 break;
92             default: /* UTIL_FORMAT_SWIZZLE_X */
93                 result |= swizzle_bit[0] << swizzle_shift[i];
94         }
95     }
96     return result;
97 }
98
99 /* Translate a pipe_format into a useful texture format for sampling.
100  *
101  * Some special formats are translated directly using R300_EASY_TX_FORMAT,
102  * but the majority of them is translated in a generic way, automatically
103  * supporting all the formats hw can support.
104  *
105  * R300_EASY_TX_FORMAT swizzles the texture.
106  * Note the signature of R300_EASY_TX_FORMAT:
107  *   R300_EASY_TX_FORMAT(B, G, R, A, FORMAT);
108  *
109  * The FORMAT specifies how the texture sampler will treat the texture, and
110  * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */
111 uint32_t r300_translate_texformat(enum pipe_format format,
112                                   const unsigned char *swizzle_view)
113 {
114     uint32_t result = 0;
115     const struct util_format_description *desc;
116     unsigned i;
117     boolean uniform = TRUE;
118     const uint32_t sign_bit[4] = {
119         R300_TX_FORMAT_SIGNED_X,
120         R300_TX_FORMAT_SIGNED_Y,
121         R300_TX_FORMAT_SIGNED_Z,
122         R300_TX_FORMAT_SIGNED_W,
123     };
124
125     desc = util_format_description(format);
126
127     /* Colorspace (return non-RGB formats directly). */
128     switch (desc->colorspace) {
129         /* Depth stencil formats.
130          * Swizzles are added in r300_merge_textures_and_samplers. */
131         case UTIL_FORMAT_COLORSPACE_ZS:
132             switch (format) {
133                 case PIPE_FORMAT_Z16_UNORM:
134                     return R300_TX_FORMAT_X16;
135                 case PIPE_FORMAT_X8Z24_UNORM:
136                 case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
137                     return R500_TX_FORMAT_Y8X24;
138                 default:
139                     return ~0; /* Unsupported. */
140             }
141
142         /* YUV formats. */
143         case UTIL_FORMAT_COLORSPACE_YUV:
144             result |= R300_TX_FORMAT_YUV_TO_RGB;
145
146             switch (format) {
147                 case PIPE_FORMAT_UYVY:
148                     return R300_EASY_TX_FORMAT(X, Y, Z, ONE, YVYU422) | result;
149                 case PIPE_FORMAT_YUYV:
150                     return R300_EASY_TX_FORMAT(X, Y, Z, ONE, VYUY422) | result;
151                 default:
152                     return ~0; /* Unsupported/unknown. */
153             }
154
155         /* Add gamma correction. */
156         case UTIL_FORMAT_COLORSPACE_SRGB:
157             result |= R300_TX_FORMAT_GAMMA;
158             break;
159
160         default:
161             switch (format) {
162                 /* Same as YUV but without the YUR->RGB conversion. */
163                 case PIPE_FORMAT_R8G8_B8G8_UNORM:
164                     return R300_EASY_TX_FORMAT(X, Y, Z, ONE, YVYU422) | result;
165                 case PIPE_FORMAT_G8R8_G8B8_UNORM:
166                     return R300_EASY_TX_FORMAT(X, Y, Z, ONE, VYUY422) | result;
167                 default:;
168             }
169     }
170
171     result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view);
172
173     /* S3TC formats. */
174     if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
175         if (!util_format_s3tc_enabled) {
176             return ~0; /* Unsupported. */
177         }
178
179         switch (format) {
180             case PIPE_FORMAT_DXT1_RGB:
181             case PIPE_FORMAT_DXT1_RGBA:
182             case PIPE_FORMAT_DXT1_SRGB:
183             case PIPE_FORMAT_DXT1_SRGBA:
184                 return R300_TX_FORMAT_DXT1 | result;
185             case PIPE_FORMAT_DXT3_RGBA:
186             case PIPE_FORMAT_DXT3_SRGBA:
187                 return R300_TX_FORMAT_DXT3 | result;
188             case PIPE_FORMAT_DXT5_RGBA:
189             case PIPE_FORMAT_DXT5_SRGBA:
190                 return R300_TX_FORMAT_DXT5 | result;
191             default:
192                 return ~0; /* Unsupported/unknown. */
193         }
194     }
195
196     /* Add sign. */
197     for (i = 0; i < desc->nr_channels; i++) {
198         if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
199             result |= sign_bit[i];
200         }
201     }
202
203     /* This is truly a special format.
204      * It stores R8G8 and B is computed using sqrt(1 - R^2 - G^2)
205      * in the sampler unit. Also known as D3DFMT_CxV8U8. */
206     if (format == PIPE_FORMAT_R8G8Bx_SNORM) {
207         return R300_TX_FORMAT_CxV8U8 | result;
208     }
209
210     /* RGTC formats. */
211     if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
212         switch (format) {
213             case PIPE_FORMAT_RGTC1_UNORM:
214             case PIPE_FORMAT_RGTC1_SNORM:
215                 return R500_TX_FORMAT_ATI1N | result;
216             case PIPE_FORMAT_RGTC2_UNORM:
217             case PIPE_FORMAT_RGTC2_SNORM:
218                 return R400_TX_FORMAT_ATI2N | result;
219             default:
220                 return ~0; /* Unsupported/unknown. */
221         }
222     }
223
224     /* See whether the components are of the same size. */
225     for (i = 1; i < desc->nr_channels; i++) {
226         uniform = uniform && desc->channel[0].size == desc->channel[i].size;
227     }
228
229     /* Non-uniform formats. */
230     if (!uniform) {
231         switch (desc->nr_channels) {
232             case 3:
233                 if (desc->channel[0].size == 5 &&
234                     desc->channel[1].size == 6 &&
235                     desc->channel[2].size == 5) {
236                     return R300_TX_FORMAT_Z5Y6X5 | result;
237                 }
238                 if (desc->channel[0].size == 5 &&
239                     desc->channel[1].size == 5 &&
240                     desc->channel[2].size == 6) {
241                     return R300_TX_FORMAT_Z6Y5X5 | result;
242                 }
243                 return ~0; /* Unsupported/unknown. */
244
245             case 4:
246                 if (desc->channel[0].size == 5 &&
247                     desc->channel[1].size == 5 &&
248                     desc->channel[2].size == 5 &&
249                     desc->channel[3].size == 1) {
250                     return R300_TX_FORMAT_W1Z5Y5X5 | result;
251                 }
252                 if (desc->channel[0].size == 10 &&
253                     desc->channel[1].size == 10 &&
254                     desc->channel[2].size == 10 &&
255                     desc->channel[3].size == 2) {
256                     return R300_TX_FORMAT_W2Z10Y10X10 | result;
257                 }
258         }
259         return ~0; /* Unsupported/unknown. */
260     }
261
262     /* And finally, uniform formats. */
263     switch (desc->channel[0].type) {
264         case UTIL_FORMAT_TYPE_UNSIGNED:
265         case UTIL_FORMAT_TYPE_SIGNED:
266             if (!desc->channel[0].normalized &&
267                 desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
268                 return ~0;
269             }
270
271             switch (desc->channel[0].size) {
272                 case 4:
273                     switch (desc->nr_channels) {
274                         case 2:
275                             return R300_TX_FORMAT_Y4X4 | result;
276                         case 4:
277                             return R300_TX_FORMAT_W4Z4Y4X4 | result;
278                     }
279                     return ~0;
280
281                 case 8:
282                     switch (desc->nr_channels) {
283                         case 1:
284                             return R300_TX_FORMAT_X8 | result;
285                         case 2:
286                             return R300_TX_FORMAT_Y8X8 | result;
287                         case 4:
288                             return R300_TX_FORMAT_W8Z8Y8X8 | result;
289                     }
290                     return ~0;
291
292                 case 16:
293                     switch (desc->nr_channels) {
294                         case 1:
295                             return R300_TX_FORMAT_X16 | result;
296                         case 2:
297                             return R300_TX_FORMAT_Y16X16 | result;
298                         case 4:
299                             return R300_TX_FORMAT_W16Z16Y16X16 | result;
300                     }
301             }
302             return ~0;
303
304         case UTIL_FORMAT_TYPE_FLOAT:
305             switch (desc->channel[0].size) {
306                 case 16:
307                     switch (desc->nr_channels) {
308                         case 1:
309                             return R300_TX_FORMAT_16F | result;
310                         case 2:
311                             return R300_TX_FORMAT_16F_16F | result;
312                         case 4:
313                             return R300_TX_FORMAT_16F_16F_16F_16F | result;
314                     }
315                     return ~0;
316
317                 case 32:
318                     switch (desc->nr_channels) {
319                         case 1:
320                             return R300_TX_FORMAT_32F | result;
321                         case 2:
322                             return R300_TX_FORMAT_32F_32F | result;
323                         case 4:
324                             return R300_TX_FORMAT_32F_32F_32F_32F | result;
325                     }
326             }
327     }
328
329     return ~0; /* Unsupported/unknown. */
330 }
331
332 uint32_t r500_tx_format_msb_bit(enum pipe_format format)
333 {
334     switch (format) {
335         case PIPE_FORMAT_RGTC1_UNORM:
336         case PIPE_FORMAT_RGTC1_SNORM:
337         case PIPE_FORMAT_X8Z24_UNORM:
338         case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
339             return R500_TXFORMAT_MSB;
340         default:
341             return 0;
342     }
343 }
344
345 /* Buffer formats. */
346
347 /* Colorbuffer formats. This is the unswizzled format of the RB3D block's
348  * output. For the swizzling of the targets, check the shader's format. */
349 static uint32_t r300_translate_colorformat(enum pipe_format format)
350 {
351     switch (format) {
352         /* 8-bit buffers. */
353         case PIPE_FORMAT_A8_UNORM:
354         case PIPE_FORMAT_I8_UNORM:
355         case PIPE_FORMAT_L8_UNORM:
356         case PIPE_FORMAT_R8_UNORM:
357         case PIPE_FORMAT_R8_SNORM:
358             return R300_COLOR_FORMAT_I8;
359
360         /* 16-bit buffers. */
361         case PIPE_FORMAT_B5G6R5_UNORM:
362             return R300_COLOR_FORMAT_RGB565;
363
364         case PIPE_FORMAT_B5G5R5A1_UNORM:
365         case PIPE_FORMAT_B5G5R5X1_UNORM:
366             return R300_COLOR_FORMAT_ARGB1555;
367
368         case PIPE_FORMAT_B4G4R4A4_UNORM:
369         case PIPE_FORMAT_B4G4R4X4_UNORM:
370             return R300_COLOR_FORMAT_ARGB4444;
371
372         /* 32-bit buffers. */
373         case PIPE_FORMAT_B8G8R8A8_UNORM:
374         case PIPE_FORMAT_B8G8R8X8_UNORM:
375         case PIPE_FORMAT_A8R8G8B8_UNORM:
376         case PIPE_FORMAT_X8R8G8B8_UNORM:
377         case PIPE_FORMAT_A8B8G8R8_UNORM:
378         case PIPE_FORMAT_R8G8B8A8_SNORM:
379         case PIPE_FORMAT_X8B8G8R8_UNORM:
380         case PIPE_FORMAT_R8G8B8X8_UNORM:
381         case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
382             return R300_COLOR_FORMAT_ARGB8888;
383
384         case PIPE_FORMAT_R10G10B10A2_UNORM:
385         case PIPE_FORMAT_R10G10B10X2_SNORM:
386         case PIPE_FORMAT_B10G10R10A2_UNORM:
387         case PIPE_FORMAT_R10SG10SB10SA2U_NORM:
388             return R500_COLOR_FORMAT_ARGB2101010;  /* R5xx-only? */
389
390         /* 64-bit buffers. */
391         case PIPE_FORMAT_R16G16B16A16_UNORM:
392         case PIPE_FORMAT_R16G16B16A16_SNORM:
393         case PIPE_FORMAT_R16G16B16A16_FLOAT:
394             return R300_COLOR_FORMAT_ARGB16161616;
395
396         /* 128-bit buffers. */
397         case PIPE_FORMAT_R32G32B32A32_FLOAT:
398             return R300_COLOR_FORMAT_ARGB32323232;
399
400         /* YUV buffers. */
401         case PIPE_FORMAT_UYVY:
402             return R300_COLOR_FORMAT_YVYU;
403         case PIPE_FORMAT_YUYV:
404             return R300_COLOR_FORMAT_VYUY;
405         default:
406             return ~0; /* Unsupported. */
407     }
408 }
409
410 /* Depthbuffer and stencilbuffer. Thankfully, we only support two flavors. */
411 static uint32_t r300_translate_zsformat(enum pipe_format format)
412 {
413     switch (format) {
414         /* 16-bit depth, no stencil */
415         case PIPE_FORMAT_Z16_UNORM:
416             return R300_DEPTHFORMAT_16BIT_INT_Z;
417         /* 24-bit depth, ignored stencil */
418         case PIPE_FORMAT_X8Z24_UNORM:
419         /* 24-bit depth, 8-bit stencil */
420         case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
421             return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
422         default:
423             return ~0; /* Unsupported. */
424     }
425 }
426
427 /* Shader output formats. This is essentially the swizzle from the shader
428  * to the RB3D block.
429  *
430  * Note that formats are stored from C3 to C0. */
431 static uint32_t r300_translate_out_fmt(enum pipe_format format)
432 {
433     uint32_t modifier = 0;
434     unsigned i;
435     const struct util_format_description *desc;
436     static const uint32_t sign_bit[4] = {
437         R300_OUT_SIGN(0x1),
438         R300_OUT_SIGN(0x2),
439         R300_OUT_SIGN(0x4),
440         R300_OUT_SIGN(0x8),
441     };
442
443     desc = util_format_description(format);
444
445     /* Specifies how the shader output is written to the fog unit. */
446     if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT) {
447         if (desc->channel[0].size == 32) {
448             modifier |= R300_US_OUT_FMT_C4_32_FP;
449         } else {
450             modifier |= R300_US_OUT_FMT_C4_16_FP;
451         }
452     } else {
453         if (desc->channel[0].size == 16) {
454             modifier |= R300_US_OUT_FMT_C4_16;
455         } else {
456             /* C4_8 seems to be used for the formats whose pixel size
457              * is <= 32 bits. */
458             modifier |= R300_US_OUT_FMT_C4_8;
459         }
460     }
461
462     /* Add sign. */
463     for (i = 0; i < 4; i++)
464         if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
465             modifier |= sign_bit[i];
466         }
467
468     /* Add swizzles and return. */
469     switch (format) {
470         /* 8-bit outputs.
471          * COLORFORMAT_I8 stores the C2 component. */
472         case PIPE_FORMAT_A8_UNORM:
473             return modifier | R300_C2_SEL_A;
474         case PIPE_FORMAT_I8_UNORM:
475         case PIPE_FORMAT_L8_UNORM:
476         case PIPE_FORMAT_R8_UNORM:
477         case PIPE_FORMAT_R8_SNORM:
478             return modifier | R300_C2_SEL_R;
479
480         /* BGRA outputs. */
481         case PIPE_FORMAT_B5G6R5_UNORM:
482         case PIPE_FORMAT_B5G5R5A1_UNORM:
483         case PIPE_FORMAT_B5G5R5X1_UNORM:
484         case PIPE_FORMAT_B4G4R4A4_UNORM:
485         case PIPE_FORMAT_B4G4R4X4_UNORM:
486         case PIPE_FORMAT_B8G8R8A8_UNORM:
487         case PIPE_FORMAT_B8G8R8X8_UNORM:
488         case PIPE_FORMAT_B10G10R10A2_UNORM:
489             return modifier |
490                 R300_C0_SEL_B | R300_C1_SEL_G |
491                 R300_C2_SEL_R | R300_C3_SEL_A;
492
493         /* ARGB outputs. */
494         case PIPE_FORMAT_A8R8G8B8_UNORM:
495         case PIPE_FORMAT_X8R8G8B8_UNORM:
496             return modifier |
497                 R300_C0_SEL_A | R300_C1_SEL_R |
498                 R300_C2_SEL_G | R300_C3_SEL_B;
499
500         /* ABGR outputs. */
501         case PIPE_FORMAT_A8B8G8R8_UNORM:
502         case PIPE_FORMAT_X8B8G8R8_UNORM:
503             return modifier |
504                 R300_C0_SEL_A | R300_C1_SEL_B |
505                 R300_C2_SEL_G | R300_C3_SEL_R;
506
507         /* RGBA outputs. */
508         case PIPE_FORMAT_R8G8B8X8_UNORM:
509         case PIPE_FORMAT_R8G8B8A8_SNORM:
510         case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
511         case PIPE_FORMAT_R10G10B10A2_UNORM:
512         case PIPE_FORMAT_R10G10B10X2_SNORM:
513         case PIPE_FORMAT_R10SG10SB10SA2U_NORM:
514         case PIPE_FORMAT_R16G16B16A16_UNORM:
515         case PIPE_FORMAT_R16G16B16A16_SNORM:
516         case PIPE_FORMAT_R16G16B16A16_FLOAT:
517         case PIPE_FORMAT_R32G32B32A32_FLOAT:
518             return modifier |
519                 R300_C0_SEL_R | R300_C1_SEL_G |
520                 R300_C2_SEL_B | R300_C3_SEL_A;
521
522         default:
523             return ~0; /* Unsupported. */
524     }
525 }
526
527 boolean r300_is_colorbuffer_format_supported(enum pipe_format format)
528 {
529     return r300_translate_colorformat(format) != ~0 &&
530            r300_translate_out_fmt(format) != ~0;
531 }
532
533 boolean r300_is_zs_format_supported(enum pipe_format format)
534 {
535     return r300_translate_zsformat(format) != ~0;
536 }
537
538 boolean r300_is_sampler_format_supported(enum pipe_format format)
539 {
540     return r300_translate_texformat(format, 0) != ~0;
541 }
542
543 static void r300_texture_setup_immutable_state(struct r300_screen* screen,
544                                                struct r300_texture* tex)
545 {
546     struct r300_texture_format_state* f = &tex->tx_format;
547     struct pipe_resource *pt = &tex->b.b;
548     boolean is_r500 = screen->caps.is_r500;
549
550     /* Set sampler state. */
551     f->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) |
552                  R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff);
553
554     if (tex->uses_pitch) {
555         /* rectangles love this */
556         f->format0 |= R300_TX_PITCH_EN;
557         f->format2 = (tex->hwpitch[0] - 1) & 0x1fff;
558     } else {
559         /* power of two textures (3D, mipmaps, and no pitch) */
560         f->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth0) & 0xf);
561     }
562
563     f->format1 = 0;
564     if (pt->target == PIPE_TEXTURE_CUBE) {
565         f->format1 |= R300_TX_FORMAT_CUBIC_MAP;
566     }
567     if (pt->target == PIPE_TEXTURE_3D) {
568         f->format1 |= R300_TX_FORMAT_3D;
569     }
570
571     /* large textures on r500 */
572     if (is_r500)
573     {
574         if (pt->width0 > 2048) {
575             f->format2 |= R500_TXWIDTH_BIT11;
576         }
577         if (pt->height0 > 2048) {
578             f->format2 |= R500_TXHEIGHT_BIT11;
579         }
580     }
581
582     f->tile_config = R300_TXO_MACRO_TILE(tex->macrotile) |
583                      R300_TXO_MICRO_TILE(tex->microtile);
584 }
585
586 static void r300_texture_setup_fb_state(struct r300_screen* screen,
587                                         struct r300_texture* tex)
588 {
589     unsigned i;
590
591     /* Set framebuffer state. */
592     if (util_format_is_depth_or_stencil(tex->b.b.format)) {
593         for (i = 0; i <= tex->b.b.last_level; i++) {
594             tex->fb_state.pitch[i] =
595                 tex->hwpitch[i] |
596                 R300_DEPTHMACROTILE(tex->mip_macrotile[i]) |
597                 R300_DEPTHMICROTILE(tex->microtile);
598         }
599         tex->fb_state.format = r300_translate_zsformat(tex->b.b.format);
600     } else {
601         for (i = 0; i <= tex->b.b.last_level; i++) {
602             tex->fb_state.pitch[i] =
603                 tex->hwpitch[i] |
604                 r300_translate_colorformat(tex->b.b.format) |
605                 R300_COLOR_TILE(tex->mip_macrotile[i]) |
606                 R300_COLOR_MICROTILE(tex->microtile);
607         }
608         tex->fb_state.format = r300_translate_out_fmt(tex->b.b.format);
609     }
610 }
611
612 void r300_texture_reinterpret_format(struct pipe_screen *screen,
613                                      struct pipe_resource *tex,
614                                      enum pipe_format new_format)
615 {
616     struct r300_screen *r300screen = r300_screen(screen);
617
618     SCREEN_DBG(r300screen, DBG_TEX,
619         "r300: texture_reinterpret_format: %s -> %s\n",
620         util_format_short_name(tex->format),
621         util_format_short_name(new_format));
622
623     tex->format = new_format;
624
625     r300_texture_setup_fb_state(r300_screen(screen), r300_texture(tex));
626 }
627
628 unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
629                                  unsigned zslice, unsigned face)
630 {
631     unsigned offset = tex->offset[level];
632
633     switch (tex->b.b.target) {
634         case PIPE_TEXTURE_3D:
635             assert(face == 0);
636             return offset + zslice * tex->layer_size[level];
637
638         case PIPE_TEXTURE_CUBE:
639             assert(zslice == 0);
640             return offset + face * tex->layer_size[level];
641
642         default:
643             assert(zslice == 0 && face == 0);
644             return offset;
645     }
646 }
647
648 /* Returns the number of pixels that the texture should be aligned to
649  * in the given dimension. */
650 static unsigned r300_get_pixel_alignment(struct r300_texture *tex,
651                                          enum r300_buffer_tiling macrotile,
652                                          enum r300_dim dim)
653 {
654     static const unsigned table[2][5][3][2] =
655     {
656         {
657     /* Macro: linear    linear    linear
658        Micro: linear    tiled  square-tiled */
659             {{ 32, 1}, { 8,  4}, { 0,  0}}, /*   8 bits per pixel */
660             {{ 16, 1}, { 8,  2}, { 4,  4}}, /*  16 bits per pixel */
661             {{  8, 1}, { 4,  2}, { 0,  0}}, /*  32 bits per pixel */
662             {{  4, 1}, { 0,  0}, { 2,  2}}, /*  64 bits per pixel */
663             {{  2, 1}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
664         },
665         {
666     /* Macro: tiled     tiled     tiled
667        Micro: linear    tiled  square-tiled */
668             {{256, 8}, {64, 32}, { 0,  0}}, /*   8 bits per pixel */
669             {{128, 8}, {64, 16}, {32, 32}}, /*  16 bits per pixel */
670             {{ 64, 8}, {32, 16}, { 0,  0}}, /*  32 bits per pixel */
671             {{ 32, 8}, { 0,  0}, {16, 16}}, /*  64 bits per pixel */
672             {{ 16, 8}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
673         }
674     };
675     static const unsigned aa_block[2] = {4, 8};
676     unsigned res = 0;
677     unsigned pixsize = util_format_get_blocksize(tex->b.b.format);
678
679     assert(macrotile <= R300_BUFFER_TILED);
680     assert(tex->microtile <= R300_BUFFER_SQUARETILED);
681     assert(pixsize <= 16);
682     assert(dim <= DIM_HEIGHT);
683
684     if (tex->b.b.nr_samples > 1) {
685         /* Multisampled textures have their own alignment scheme. */
686         if (pixsize == 4)
687             res = aa_block[dim];
688     } else {
689         /* Standard alignment. */
690         res = table[macrotile][util_logbase2(pixsize)][tex->microtile][dim];
691     }
692
693     assert(res);
694     return res;
695 }
696
697 /* Return true if macrotiling should be enabled on the miplevel. */
698 static boolean r300_texture_macro_switch(struct r300_texture *tex,
699                                          unsigned level,
700                                          boolean rv350_mode,
701                                          enum r300_dim dim)
702 {
703     unsigned tile, texdim;
704
705     tile = r300_get_pixel_alignment(tex, R300_BUFFER_TILED, dim);
706     if (dim == DIM_WIDTH) {
707         texdim = u_minify(tex->b.b.width0, level);
708     } else {
709         texdim = u_minify(tex->b.b.height0, level);
710     }
711
712     /* See TX_FILTER1_n.MACRO_SWITCH. */
713     if (rv350_mode) {
714         return texdim >= tile;
715     } else {
716         return texdim > tile;
717     }
718 }
719
720 /**
721  * Return the stride, in bytes, of the texture images of the given texture
722  * at the given level.
723  */
724 unsigned r300_texture_get_stride(struct r300_screen* screen,
725                                  struct r300_texture* tex, unsigned level)
726 {
727     unsigned tile_width, width, stride;
728
729     if (tex->stride_override)
730         return tex->stride_override;
731
732     /* Check the level. */
733     if (level > tex->b.b.last_level) {
734         SCREEN_DBG(screen, DBG_TEX, "%s: level (%u) > last_level (%u)\n",
735                    __FUNCTION__, level, tex->b.b.last_level);
736         return 0;
737     }
738
739     width = u_minify(tex->b.b.width0, level);
740
741     if (util_format_is_plain(tex->b.b.format)) {
742         tile_width = r300_get_pixel_alignment(tex, tex->mip_macrotile[level],
743                                               DIM_WIDTH);
744         width = align(width, tile_width);
745
746         stride = util_format_get_stride(tex->b.b.format, width);
747
748         /* Some IGPs need a minimum stride of 64 bytes, hmm...
749          * This doesn't seem to apply to tiled textures, according to r300c. */
750         if (!tex->microtile && !tex->mip_macrotile[level] &&
751             (screen->caps.family == CHIP_FAMILY_RS600 ||
752              screen->caps.family == CHIP_FAMILY_RS690 ||
753              screen->caps.family == CHIP_FAMILY_RS740)) {
754             return stride < 64 ? 64 : stride;
755         }
756
757         /* The alignment to 32 bytes is sort of implied by the layout... */
758         return stride;
759     } else {
760         return align(util_format_get_stride(tex->b.b.format, width), 32);
761     }
762 }
763
764 static unsigned r300_texture_get_nblocksy(struct r300_texture* tex,
765                                           unsigned level)
766 {
767     unsigned height, tile_height;
768
769     height = u_minify(tex->b.b.height0, level);
770
771     if (util_format_is_plain(tex->b.b.format)) {
772         tile_height = r300_get_pixel_alignment(tex, tex->mip_macrotile[level],
773                                                DIM_HEIGHT);
774         height = align(height, tile_height);
775
776         /* This is needed for the kernel checker, unfortunately. */
777         height = util_next_power_of_two(height);
778     }
779
780     return util_format_get_nblocksy(tex->b.b.format, height);
781 }
782
783 static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen,
784                                            struct r300_texture *tex)
785 {
786     /* The kernels <= 2.6.34-rc4 compute the size of mipmapped 3D textures
787      * incorrectly. This is a workaround to prevent CS from being rejected. */
788
789     unsigned i, size;
790
791     if (!screen->rws->get_value(screen->rws, R300_VID_DRM_2_3_0) &&
792         tex->b.b.target == PIPE_TEXTURE_3D &&
793         tex->b.b.last_level > 0) {
794         size = 0;
795
796         for (i = 0; i <= tex->b.b.last_level; i++) {
797             size += r300_texture_get_stride(screen, tex, i) *
798                     r300_texture_get_nblocksy(tex, i);
799         }
800
801         size *= tex->b.b.depth0;
802         tex->size = size;
803     }
804 }
805
806 static void r300_setup_miptree(struct r300_screen* screen,
807                                struct r300_texture* tex)
808 {
809     struct pipe_resource* base = &tex->b.b;
810     unsigned stride, size, layer_size, nblocksy, i;
811     boolean rv350_mode = screen->caps.is_rv350;
812
813     SCREEN_DBG(screen, DBG_TEXALLOC,
814         "r300: Making miptree for texture, format %s\n",
815         util_format_short_name(base->format));
816
817     for (i = 0; i <= base->last_level; i++) {
818         /* Let's see if this miplevel can be macrotiled. */
819         tex->mip_macrotile[i] =
820             (tex->macrotile == R300_BUFFER_TILED &&
821              r300_texture_macro_switch(tex, i, rv350_mode, DIM_WIDTH) &&
822              r300_texture_macro_switch(tex, i, rv350_mode, DIM_HEIGHT)) ?
823              R300_BUFFER_TILED : R300_BUFFER_LINEAR;
824
825         stride = r300_texture_get_stride(screen, tex, i);
826         nblocksy = r300_texture_get_nblocksy(tex, i);
827         layer_size = stride * nblocksy;
828
829         if (base->nr_samples) {
830             layer_size *= base->nr_samples;
831         }
832
833         if (base->target == PIPE_TEXTURE_CUBE)
834             size = layer_size * 6;
835         else
836             size = layer_size * u_minify(base->depth0, i);
837
838         tex->offset[i] = tex->size;
839         tex->size = tex->offset[i] + size;
840         tex->layer_size[i] = layer_size;
841         tex->pitch[i] = stride / util_format_get_blocksize(base->format);
842         tex->hwpitch[i] =
843                 tex->pitch[i] * util_format_get_blockwidth(base->format);
844
845         SCREEN_DBG(screen, DBG_TEXALLOC, "r300: Texture miptree: Level %d "
846                 "(%dx%dx%d px, pitch %d bytes) %d bytes total, macrotiled %s\n",
847                 i, u_minify(base->width0, i), u_minify(base->height0, i),
848                 u_minify(base->depth0, i), stride, tex->size,
849                 tex->mip_macrotile[i] ? "TRUE" : "FALSE");
850     }
851 }
852
853 static void r300_setup_flags(struct r300_texture* tex)
854 {
855     tex->uses_pitch = !util_is_power_of_two(tex->b.b.width0) ||
856                       !util_is_power_of_two(tex->b.b.height0) ||
857                       tex->stride_override;
858 }
859
860 static void r300_setup_tiling(struct pipe_screen *screen,
861                               struct r300_texture *tex)
862 {
863     struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys;
864     enum pipe_format format = tex->b.b.format;
865     boolean rv350_mode = r300_screen(screen)->caps.is_rv350;
866     boolean is_zb = util_format_is_depth_or_stencil(format);
867     boolean dbg_no_tiling = SCREEN_DBG_ON(r300_screen(screen), DBG_NO_TILING);
868
869     if (!util_format_is_plain(format)) {
870         return;
871     }
872
873     /* If height == 1, disable microtiling except for zbuffer. */
874     if (!is_zb && (tex->b.b.height0 == 1 || dbg_no_tiling)) {
875         return;
876     }
877
878     /* Set microtiling. */
879     switch (util_format_get_blocksize(format)) {
880         case 1:
881         case 4:
882             tex->microtile = R300_BUFFER_TILED;
883             break;
884
885         case 2:
886         case 8:
887             if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) {
888                 tex->microtile = R300_BUFFER_SQUARETILED;
889             }
890             break;
891     }
892
893     if (dbg_no_tiling) {
894         return;
895     }
896
897     /* Set macrotiling. */
898     if (r300_texture_macro_switch(tex, 0, rv350_mode, DIM_WIDTH) &&
899         r300_texture_macro_switch(tex, 0, rv350_mode, DIM_HEIGHT)) {
900         tex->macrotile = R300_BUFFER_TILED;
901     }
902 }
903
904 static unsigned r300_texture_is_referenced(struct pipe_context *context,
905                                          struct pipe_resource *texture,
906                                          unsigned face, unsigned level)
907 {
908     struct r300_context *r300 = r300_context(context);
909     struct r300_texture *rtex = (struct r300_texture *)texture;
910
911     if (r300->rws->is_buffer_referenced(r300->rws, rtex->buffer, R300_REF_CS))
912         return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
913
914     return PIPE_UNREFERENCED;
915 }
916
917 static void r300_texture_destroy(struct pipe_screen *screen,
918                                  struct pipe_resource* texture)
919 {
920     struct r300_texture* tex = (struct r300_texture*)texture;
921     struct r300_winsys_screen *rws = (struct r300_winsys_screen *)texture->screen->winsys;
922
923     rws->buffer_reference(rws, &tex->buffer, NULL);
924     FREE(tex);
925 }
926
927 static boolean r300_texture_get_handle(struct pipe_screen* screen,
928                                        struct pipe_resource *texture,
929                                        struct winsys_handle *whandle)
930 {
931     struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys;
932     struct r300_texture* tex = (struct r300_texture*)texture;
933
934     if (!tex) {
935         return FALSE;
936     }
937
938     return rws->buffer_get_handle(rws, tex->buffer, whandle,
939                     r300_texture_get_stride(r300_screen(screen), tex, 0));
940 }
941
942 struct u_resource_vtbl r300_texture_vtbl = 
943 {
944    r300_texture_get_handle,           /* get_handle */
945    r300_texture_destroy,              /* resource_destroy */
946    r300_texture_is_referenced,        /* is_resource_referenced */
947    r300_texture_get_transfer,         /* get_transfer */
948    r300_texture_transfer_destroy,     /* transfer_destroy */
949    r300_texture_transfer_map,         /* transfer_map */
950    u_default_transfer_flush_region,   /* transfer_flush_region */
951    r300_texture_transfer_unmap,       /* transfer_unmap */
952    u_default_transfer_inline_write    /* transfer_inline_write */
953 };
954
955 /* Create a new texture. */
956 struct pipe_resource* r300_texture_create(struct pipe_screen* screen,
957                                           const struct pipe_resource* base)
958 {
959     struct r300_texture* tex = CALLOC_STRUCT(r300_texture);
960     struct r300_screen* rscreen = r300_screen(screen);
961     struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys;
962
963     if (!tex) {
964         return NULL;
965     }
966
967     /* Refuse to create a texture with size 0. */
968     if (!base->width0 ||
969         (!base->height0 && (base->target == PIPE_TEXTURE_2D ||
970                             base->target == PIPE_TEXTURE_CUBE)) ||
971         (!base->depth0 && base->target == PIPE_TEXTURE_3D)) {
972         fprintf(stderr, "r300: texture_create: "
973                 "Got invalid texture dimensions: %ix%ix%i\n",
974                 base->width0, base->height0, base->depth0);
975         FREE(tex);
976         return NULL;
977     }
978
979     tex->b.b = *base;
980     tex->b.vtbl = &r300_texture_vtbl;
981     pipe_reference_init(&tex->b.b.reference, 1);
982     tex->b.b.screen = screen;
983
984     r300_setup_flags(tex);
985     if (!(base->flags & R300_RESOURCE_FLAG_TRANSFER) &&
986         !(base->bind & PIPE_BIND_SCANOUT)) {
987         r300_setup_tiling(screen, tex);
988     }
989     r300_setup_miptree(rscreen, tex);
990     r300_texture_3d_fix_mipmapping(rscreen, tex);
991     r300_texture_setup_immutable_state(rscreen, tex);
992     r300_texture_setup_fb_state(rscreen, tex);
993
994     SCREEN_DBG(rscreen, DBG_TEX,
995                "r300: texture_create: Macro: %s, Micro: %s, Pitch: %i, "
996                "Dim: %ix%ix%i, LastLevel: %i, Size: %i, Format: %s\n",
997                tex->macrotile ? "YES" : " NO",
998                tex->microtile ? "YES" : " NO",
999                tex->hwpitch[0],
1000                base->width0, base->height0, base->depth0, base->last_level,
1001                tex->size,
1002                util_format_short_name(base->format));
1003
1004     tex->domain = base->flags & R300_RESOURCE_FLAG_TRANSFER ?
1005                   R300_DOMAIN_GTT :
1006                   R300_DOMAIN_VRAM | R300_DOMAIN_GTT;
1007
1008     tex->buffer = rws->buffer_create(rws, 2048, base->bind, tex->domain,
1009                                      tex->size);
1010
1011     if (!tex->buffer) {
1012         FREE(tex);
1013         return NULL;
1014     }
1015
1016     rws->buffer_set_tiling(rws, tex->buffer,
1017             tex->pitch[0] * util_format_get_blocksize(tex->b.b.format),
1018             tex->microtile,
1019             tex->macrotile);
1020
1021     return (struct pipe_resource*)tex;
1022 }
1023
1024 /* Not required to implement u_resource_vtbl, consider moving to another file:
1025  */
1026 struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
1027                                           struct pipe_resource* texture,
1028                                           unsigned face,
1029                                           unsigned level,
1030                                           unsigned zslice,
1031                                           unsigned flags)
1032 {
1033     struct r300_texture* tex = r300_texture(texture);
1034     struct r300_surface* surface = CALLOC_STRUCT(r300_surface);
1035
1036     if (surface) {
1037         uint32_t stride, offset, tile_height;
1038
1039         pipe_reference_init(&surface->base.reference, 1);
1040         pipe_resource_reference(&surface->base.texture, texture);
1041         surface->base.format = texture->format;
1042         surface->base.width = u_minify(texture->width0, level);
1043         surface->base.height = u_minify(texture->height0, level);
1044         surface->base.usage = flags;
1045         surface->base.zslice = zslice;
1046         surface->base.face = face;
1047         surface->base.level = level;
1048
1049         surface->buffer = tex->buffer;
1050
1051         /* Prefer VRAM if there are multiple domains to choose from. */
1052         surface->domain = tex->domain;
1053         if (surface->domain & R300_DOMAIN_VRAM)
1054             surface->domain &= ~R300_DOMAIN_GTT;
1055
1056         surface->offset = r300_texture_get_offset(tex, level, zslice, face);
1057         surface->pitch = tex->fb_state.pitch[level];
1058         surface->format = tex->fb_state.format;
1059
1060         /* Parameters for the CBZB clear. */
1061         surface->cbzb_width = align(surface->base.width, 64);
1062
1063         /* Height must be aligned to the size of a tile. */
1064         tile_height = r300_get_pixel_alignment(tex, tex->mip_macrotile[level],
1065                                                DIM_HEIGHT);
1066         surface->cbzb_height = align((surface->base.height + 1) / 2,
1067                                      tile_height);
1068
1069         /* Offset must be aligned to 2K and must point at the beginning
1070          * of a scanline. */
1071         stride = r300_texture_get_stride(r300_screen(screen), tex, level);
1072         offset = surface->offset + stride * surface->cbzb_height;
1073         surface->cbzb_midpoint_offset = offset & ~2047;
1074
1075         surface->cbzb_pitch = surface->pitch & 0x1ffffc;
1076
1077         if (util_format_get_blocksizebits(surface->base.format) == 32)
1078             surface->cbzb_format = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
1079         else
1080             surface->cbzb_format = R300_DEPTHFORMAT_16BIT_INT_Z;
1081
1082         SCREEN_DBG(r300_screen(screen), DBG_TEX,
1083                    "CBZB Dim: %ix%i, Misalignment: %i, Macro: %s\n",
1084                    surface->cbzb_width, surface->cbzb_height,
1085                    offset & 2047,
1086                    tex->mip_macrotile[level] ? "YES" : " NO");
1087     }
1088
1089     return &surface->base;
1090 }
1091
1092 /* Not required to implement u_resource_vtbl, consider moving to another file:
1093  */
1094 void r300_tex_surface_destroy(struct pipe_surface* s)
1095 {
1096     pipe_resource_reference(&s->texture, NULL);
1097     FREE(s);
1098 }
1099
1100 struct pipe_resource*
1101 r300_texture_from_handle(struct pipe_screen* screen,
1102                           const struct pipe_resource* base,
1103                           struct winsys_handle *whandle)
1104 {
1105     struct r300_winsys_screen *rws = (struct r300_winsys_screen*)screen->winsys;
1106     struct r300_screen* rscreen = r300_screen(screen);
1107     struct r300_winsys_buffer *buffer;
1108     struct r300_texture* tex;
1109     unsigned stride;
1110     boolean override_zb_flags;
1111
1112     /* Support only 2D textures without mipmaps */
1113     if (base->target != PIPE_TEXTURE_2D ||
1114         base->depth0 != 1 ||
1115         base->last_level != 0) {
1116         return NULL;
1117     }
1118
1119     buffer = rws->buffer_from_handle(rws, whandle, &stride);
1120     if (!buffer) {
1121         return NULL;
1122     }
1123
1124     tex = CALLOC_STRUCT(r300_texture);
1125     if (!tex) {
1126         return NULL;
1127     }
1128
1129     tex->b.b = *base;
1130     tex->b.vtbl = &r300_texture_vtbl;
1131     pipe_reference_init(&tex->b.b.reference, 1);
1132     tex->b.b.screen = screen;
1133     tex->domain = R300_DOMAIN_VRAM;
1134
1135     tex->stride_override = stride;
1136
1137     /* one ref already taken */
1138     tex->buffer = buffer;
1139
1140     rws->buffer_get_tiling(rws, buffer, &tex->microtile, &tex->macrotile);
1141     r300_setup_flags(tex);
1142     SCREEN_DBG(rscreen, DBG_TEX,
1143                "r300: texture_from_handle: Macro: %s, Micro: %s, "
1144                "Pitch: % 4i, Dim: %ix%i, Format: %s\n",
1145                tex->macrotile ? "YES" : " NO",
1146                tex->microtile ? "YES" : " NO",
1147                stride / util_format_get_blocksize(base->format),
1148                base->width0, base->height0,
1149                util_format_short_name(base->format));
1150
1151     /* Enforce microtiled zbuffer. */
1152     override_zb_flags = util_format_is_depth_or_stencil(base->format) &&
1153                         tex->microtile == R300_BUFFER_LINEAR;
1154
1155     if (override_zb_flags) {
1156         switch (util_format_get_blocksize(base->format)) {
1157             case 4:
1158                 tex->microtile = R300_BUFFER_TILED;
1159                 break;
1160
1161             case 2:
1162                 if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) {
1163                     tex->microtile = R300_BUFFER_SQUARETILED;
1164                     break;
1165                 }
1166                 /* Pass through. */
1167
1168             default:
1169                 override_zb_flags = FALSE;
1170         }
1171     }
1172
1173     r300_setup_miptree(rscreen, tex);
1174     r300_texture_setup_immutable_state(rscreen, tex);
1175     r300_texture_setup_fb_state(rscreen, tex);
1176
1177     if (override_zb_flags) {
1178         rws->buffer_set_tiling(rws, tex->buffer,
1179                 tex->pitch[0] * util_format_get_blocksize(tex->b.b.format),
1180                 tex->microtile,
1181                 tex->macrotile);
1182     }
1183     return (struct pipe_resource*)tex;
1184 }