r600g: trivially implement EXT_texture_shared_exponent
[mesa.git] / src / gallium / drivers / r600 / r600_texture.c
1 /*
2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Jerome Glisse
25 * Corbin Simpson
26 */
27 #include <errno.h>
28 #include <pipe/p_screen.h>
29 #include <util/u_format.h>
30 #include <util/u_format_s3tc.h>
31 #include <util/u_math.h>
32 #include <util/u_inlines.h>
33 #include <util/u_memory.h>
34 #include "state_tracker/drm_driver.h"
35 #include "pipebuffer/pb_buffer.h"
36 #include "r600_pipe.h"
37 #include "r600_resource.h"
38 #include "r600_state_inlines.h"
39 #include "r600d.h"
40 #include "r600_formats.h"
41
42 /* Copy from a full GPU texture to a transfer's staging one. */
43 static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
44 {
45 struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
46 struct pipe_resource *texture = transfer->resource;
47
48 ctx->resource_copy_region(ctx, rtransfer->staging_texture,
49 0, 0, 0, 0, texture, transfer->level,
50 &transfer->box);
51 }
52
53
54 /* Copy from a transfer's staging texture to a full GPU one. */
55 static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
56 {
57 struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
58 struct pipe_resource *texture = transfer->resource;
59 struct pipe_box sbox;
60
61 sbox.x = sbox.y = sbox.z = 0;
62 sbox.width = transfer->box.width;
63 sbox.height = transfer->box.height;
64 /* XXX that might be wrong */
65 sbox.depth = 1;
66 ctx->resource_copy_region(ctx, texture, transfer->level,
67 transfer->box.x, transfer->box.y, transfer->box.z,
68 rtransfer->staging_texture,
69 0, &sbox);
70
71 ctx->flush(ctx, NULL);
72 }
73
74 unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
75 unsigned level, unsigned layer)
76 {
77 unsigned offset = rtex->offset[level];
78
79 switch (rtex->resource.b.b.b.target) {
80 case PIPE_TEXTURE_3D:
81 case PIPE_TEXTURE_CUBE:
82 default:
83 return offset + layer * rtex->layer_size[level];
84 }
85 }
86
87 static unsigned r600_get_block_alignment(struct pipe_screen *screen,
88 enum pipe_format format,
89 unsigned array_mode)
90 {
91 struct r600_screen* rscreen = (struct r600_screen *)screen;
92 unsigned pixsize = util_format_get_blocksize(format);
93 int p_align;
94
95 switch(array_mode) {
96 case V_038000_ARRAY_1D_TILED_THIN1:
97 p_align = MAX2(8,
98 ((rscreen->tiling_info->group_bytes / 8 / pixsize)));
99 break;
100 case V_038000_ARRAY_2D_TILED_THIN1:
101 p_align = MAX2(rscreen->tiling_info->num_banks,
102 (((rscreen->tiling_info->group_bytes / 8 / pixsize)) *
103 rscreen->tiling_info->num_banks)) * 8;
104 break;
105 case V_038000_ARRAY_LINEAR_ALIGNED:
106 p_align = MAX2(64, rscreen->tiling_info->group_bytes / pixsize);
107 break;
108 case V_038000_ARRAY_LINEAR_GENERAL:
109 default:
110 p_align = rscreen->tiling_info->group_bytes / pixsize;
111 break;
112 }
113 return p_align;
114 }
115
116 static unsigned r600_get_height_alignment(struct pipe_screen *screen,
117 unsigned array_mode)
118 {
119 struct r600_screen* rscreen = (struct r600_screen *)screen;
120 int h_align;
121
122 switch (array_mode) {
123 case V_038000_ARRAY_2D_TILED_THIN1:
124 h_align = rscreen->tiling_info->num_channels * 8;
125 break;
126 case V_038000_ARRAY_1D_TILED_THIN1:
127 case V_038000_ARRAY_LINEAR_ALIGNED:
128 h_align = 8;
129 break;
130 case V_038000_ARRAY_LINEAR_GENERAL:
131 default:
132 h_align = 1;
133 break;
134 }
135 return h_align;
136 }
137
138 static unsigned r600_get_base_alignment(struct pipe_screen *screen,
139 enum pipe_format format,
140 unsigned array_mode)
141 {
142 struct r600_screen* rscreen = (struct r600_screen *)screen;
143 unsigned pixsize = util_format_get_blocksize(format);
144 int p_align = r600_get_block_alignment(screen, format, array_mode);
145 int h_align = r600_get_height_alignment(screen, array_mode);
146 int b_align;
147
148 switch (array_mode) {
149 case V_038000_ARRAY_2D_TILED_THIN1:
150 b_align = MAX2(rscreen->tiling_info->num_banks * rscreen->tiling_info->num_channels * 8 * 8 * pixsize,
151 p_align * pixsize * h_align);
152 break;
153 case V_038000_ARRAY_1D_TILED_THIN1:
154 case V_038000_ARRAY_LINEAR_ALIGNED:
155 case V_038000_ARRAY_LINEAR_GENERAL:
156 default:
157 b_align = rscreen->tiling_info->group_bytes;
158 break;
159 }
160 return b_align;
161 }
162
163 static unsigned mip_minify(unsigned size, unsigned level)
164 {
165 unsigned val;
166 val = u_minify(size, level);
167 if (level > 0)
168 val = util_next_power_of_two(val);
169 return val;
170 }
171
172 static unsigned r600_texture_get_nblocksx(struct pipe_screen *screen,
173 struct r600_resource_texture *rtex,
174 unsigned level)
175 {
176 struct pipe_resource *ptex = &rtex->resource.b.b.b;
177 unsigned nblocksx, block_align, width;
178 unsigned blocksize = util_format_get_blocksize(ptex->format);
179
180 if (rtex->pitch_override)
181 return rtex->pitch_override / blocksize;
182
183 width = mip_minify(ptex->width0, level);
184 nblocksx = util_format_get_nblocksx(ptex->format, width);
185
186 block_align = r600_get_block_alignment(screen, ptex->format,
187 rtex->array_mode[level]);
188 nblocksx = align(nblocksx, block_align);
189 return nblocksx;
190 }
191
192 static unsigned r600_texture_get_nblocksy(struct pipe_screen *screen,
193 struct r600_resource_texture *rtex,
194 unsigned level)
195 {
196 struct pipe_resource *ptex = &rtex->resource.b.b.b;
197 unsigned height, tile_height;
198
199 height = mip_minify(ptex->height0, level);
200 height = util_format_get_nblocksy(ptex->format, height);
201 tile_height = r600_get_height_alignment(screen,
202 rtex->array_mode[level]);
203 height = align(height, tile_height);
204 return height;
205 }
206
207 static void r600_texture_set_array_mode(struct pipe_screen *screen,
208 struct r600_resource_texture *rtex,
209 unsigned level, unsigned array_mode)
210 {
211 struct pipe_resource *ptex = &rtex->resource.b.b.b;
212
213 switch (array_mode) {
214 case V_0280A0_ARRAY_LINEAR_GENERAL:
215 case V_0280A0_ARRAY_LINEAR_ALIGNED:
216 case V_0280A0_ARRAY_1D_TILED_THIN1:
217 default:
218 rtex->array_mode[level] = array_mode;
219 break;
220 case V_0280A0_ARRAY_2D_TILED_THIN1:
221 {
222 unsigned w, h, tile_height, tile_width;
223
224 tile_height = r600_get_height_alignment(screen, array_mode);
225 tile_width = r600_get_block_alignment(screen, ptex->format, array_mode);
226
227 w = mip_minify(ptex->width0, level);
228 h = mip_minify(ptex->height0, level);
229 if (w <= tile_width || h <= tile_height)
230 rtex->array_mode[level] = V_0280A0_ARRAY_1D_TILED_THIN1;
231 else
232 rtex->array_mode[level] = array_mode;
233 }
234 break;
235 }
236 }
237
238 static void r600_setup_miptree(struct pipe_screen *screen,
239 struct r600_resource_texture *rtex,
240 unsigned array_mode)
241 {
242 struct pipe_resource *ptex = &rtex->resource.b.b.b;
243 struct radeon *radeon = (struct radeon *)screen->winsys;
244 enum chip_class chipc = r600_get_family_class(radeon);
245 unsigned size, layer_size, i, offset;
246 unsigned nblocksx, nblocksy;
247
248 for (i = 0, offset = 0; i <= ptex->last_level; i++) {
249 unsigned blocksize = util_format_get_blocksize(ptex->format);
250
251 r600_texture_set_array_mode(screen, rtex, i, array_mode);
252
253 nblocksx = r600_texture_get_nblocksx(screen, rtex, i);
254 nblocksy = r600_texture_get_nblocksy(screen, rtex, i);
255
256 layer_size = nblocksx * nblocksy * blocksize;
257 if (ptex->target == PIPE_TEXTURE_CUBE) {
258 if (chipc >= R700)
259 size = layer_size * 8;
260 else
261 size = layer_size * 6;
262 }
263 else if (ptex->target == PIPE_TEXTURE_3D)
264 size = layer_size * u_minify(ptex->depth0, i);
265 else
266 size = layer_size * ptex->array_size;
267
268 /* align base image and start of miptree */
269 if ((i == 0) || (i == 1))
270 offset = align(offset, r600_get_base_alignment(screen, ptex->format, array_mode));
271 rtex->offset[i] = offset;
272 rtex->layer_size[i] = layer_size;
273 rtex->pitch_in_blocks[i] = nblocksx; /* CB talks in elements */
274 rtex->pitch_in_bytes[i] = nblocksx * blocksize;
275
276 offset += size;
277 }
278 rtex->size = offset;
279 }
280
281 /* Figure out whether u_blitter will fallback to a transfer operation.
282 * If so, don't use a staging resource.
283 */
284 static boolean permit_hardware_blit(struct pipe_screen *screen,
285 const struct pipe_resource *res)
286 {
287 unsigned bind;
288
289 if (util_format_is_depth_or_stencil(res->format))
290 bind = PIPE_BIND_DEPTH_STENCIL;
291 else
292 bind = PIPE_BIND_RENDER_TARGET;
293
294 /* hackaround for S3TC */
295 if (util_format_is_compressed(res->format))
296 return TRUE;
297
298 if (!screen->is_format_supported(screen,
299 res->format,
300 res->target,
301 res->nr_samples,
302 bind))
303 return FALSE;
304
305 if (!screen->is_format_supported(screen,
306 res->format,
307 res->target,
308 res->nr_samples,
309 PIPE_BIND_SAMPLER_VIEW))
310 return FALSE;
311
312 return TRUE;
313 }
314
315 static boolean r600_texture_get_handle(struct pipe_screen* screen,
316 struct pipe_resource *ptex,
317 struct winsys_handle *whandle)
318 {
319 struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
320 struct r600_resource *resource = &rtex->resource;
321 struct radeon *radeon = (struct radeon *)screen->winsys;
322
323 return r600_bo_get_winsys_handle(radeon, resource->bo,
324 rtex->pitch_in_bytes[0], whandle);
325 }
326
327 static void r600_texture_destroy(struct pipe_screen *screen,
328 struct pipe_resource *ptex)
329 {
330 struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
331 struct r600_resource *resource = &rtex->resource;
332 struct radeon *radeon = (struct radeon *)screen->winsys;
333
334 if (rtex->flushed_depth_texture)
335 pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
336
337 if (resource->bo) {
338 r600_bo_reference(radeon, &resource->bo, NULL);
339 }
340 FREE(rtex);
341 }
342
343 static const struct u_resource_vtbl r600_texture_vtbl =
344 {
345 r600_texture_get_handle, /* get_handle */
346 r600_texture_destroy, /* resource_destroy */
347 r600_texture_get_transfer, /* get_transfer */
348 r600_texture_transfer_destroy, /* transfer_destroy */
349 r600_texture_transfer_map, /* transfer_map */
350 u_default_transfer_flush_region,/* transfer_flush_region */
351 r600_texture_transfer_unmap, /* transfer_unmap */
352 u_default_transfer_inline_write /* transfer_inline_write */
353 };
354
355 static struct r600_resource_texture *
356 r600_texture_create_object(struct pipe_screen *screen,
357 const struct pipe_resource *base,
358 unsigned array_mode,
359 unsigned pitch_in_bytes_override,
360 unsigned max_buffer_size,
361 struct r600_bo *bo)
362 {
363 struct r600_resource_texture *rtex;
364 struct r600_resource *resource;
365 struct radeon *radeon = (struct radeon *)screen->winsys;
366
367 rtex = CALLOC_STRUCT(r600_resource_texture);
368 if (rtex == NULL)
369 return NULL;
370
371 resource = &rtex->resource;
372 resource->b.b.b = *base;
373 resource->b.b.vtbl = &r600_texture_vtbl;
374 pipe_reference_init(&resource->b.b.b.reference, 1);
375 resource->b.b.b.screen = screen;
376 resource->bo = bo;
377 rtex->pitch_override = pitch_in_bytes_override;
378 /* only mark depth textures the HW can hit as depth textures */
379 if (util_format_is_depth_or_stencil(base->format) && permit_hardware_blit(screen, base))
380 rtex->depth = 1;
381
382 r600_setup_miptree(screen, rtex, array_mode);
383
384 resource->size = rtex->size;
385
386 if (!resource->bo) {
387 struct pipe_resource *ptex = &rtex->resource.b.b.b;
388 int base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
389
390 resource->bo = r600_bo(radeon, rtex->size, base_align, base->bind, base->usage);
391 if (!resource->bo) {
392 FREE(rtex);
393 return NULL;
394 }
395 }
396 return rtex;
397 }
398
399 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
400 const struct pipe_resource *templ)
401 {
402 unsigned array_mode = 0;
403 static int force_tiling = -1;
404
405 /* Would like some magic "get_bool_option_once" routine.
406 */
407 if (force_tiling == -1) {
408 #if 0
409 /* reenable when 2D tiling is fixed better */
410 struct r600_screen *rscreen = (struct r600_screen *)screen;
411 if (r600_get_minor_version(rscreen->radeon) >= 9)
412 force_tiling = debug_get_bool_option("R600_TILING", TRUE);
413 #endif
414 force_tiling = debug_get_bool_option("R600_TILING", FALSE);
415 }
416
417 if (force_tiling && permit_hardware_blit(screen, templ)) {
418 if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
419 !(templ->bind & PIPE_BIND_SCANOUT)) {
420 array_mode = V_038000_ARRAY_2D_TILED_THIN1;
421 }
422 }
423
424 if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
425 util_format_is_compressed(templ->format))
426 array_mode = V_038000_ARRAY_1D_TILED_THIN1;
427
428 return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
429 0, 0, NULL);
430
431 }
432
433 static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
434 struct pipe_resource *texture,
435 const struct pipe_surface *surf_tmpl)
436 {
437 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
438 struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
439 unsigned level = surf_tmpl->u.tex.level;
440
441 assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
442 if (surface == NULL)
443 return NULL;
444 /* XXX no offset */
445 /* offset = r600_texture_get_offset(rtex, level, surf_tmpl->u.tex.first_layer);*/
446 pipe_reference_init(&surface->base.reference, 1);
447 pipe_resource_reference(&surface->base.texture, texture);
448 surface->base.context = pipe;
449 surface->base.format = surf_tmpl->format;
450 surface->base.width = mip_minify(texture->width0, level);
451 surface->base.height = mip_minify(texture->height0, level);
452 surface->base.usage = surf_tmpl->usage;
453 surface->base.texture = texture;
454 surface->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
455 surface->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
456 surface->base.u.tex.level = level;
457
458 surface->aligned_height = r600_texture_get_nblocksy(pipe->screen,
459 rtex, level);
460 return &surface->base;
461 }
462
463 static void r600_surface_destroy(struct pipe_context *pipe,
464 struct pipe_surface *surface)
465 {
466 pipe_resource_reference(&surface->texture, NULL);
467 FREE(surface);
468 }
469
470
471 struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
472 const struct pipe_resource *templ,
473 struct winsys_handle *whandle)
474 {
475 struct radeon *rw = (struct radeon*)screen->winsys;
476 struct r600_bo *bo = NULL;
477 unsigned array_mode = 0;
478
479 /* Support only 2D textures without mipmaps */
480 if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) ||
481 templ->depth0 != 1 || templ->last_level != 0)
482 return NULL;
483
484 bo = r600_bo_handle(rw, whandle->handle, &array_mode);
485 if (bo == NULL) {
486 return NULL;
487 }
488
489 return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
490 whandle->stride,
491 0,
492 bo);
493 }
494
495 int r600_texture_depth_flush(struct pipe_context *ctx,
496 struct pipe_resource *texture, boolean just_create)
497 {
498 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
499 struct pipe_resource resource;
500
501 if (rtex->flushed_depth_texture)
502 goto out;
503
504 resource.target = PIPE_TEXTURE_2D;
505 resource.format = texture->format;
506 resource.width0 = texture->width0;
507 resource.height0 = texture->height0;
508 resource.depth0 = 1;
509 resource.array_size = 1;
510 resource.last_level = texture->last_level;
511 resource.nr_samples = 0;
512 resource.usage = PIPE_USAGE_DYNAMIC;
513 resource.bind = 0;
514 resource.flags = R600_RESOURCE_FLAG_TRANSFER;
515
516 resource.bind |= PIPE_BIND_DEPTH_STENCIL;
517
518 rtex->flushed_depth_texture = (struct r600_resource_texture *)ctx->screen->resource_create(ctx->screen, &resource);
519 if (rtex->flushed_depth_texture == NULL) {
520 R600_ERR("failed to create temporary texture to hold untiled copy\n");
521 return -ENOMEM;
522 }
523
524 ((struct r600_resource_texture *)rtex->flushed_depth_texture)->is_flushing_texture = TRUE;
525 out:
526 if (just_create)
527 return 0;
528
529 /* XXX: only do this if the depth texture has actually changed:
530 */
531 r600_blit_uncompress_depth(ctx, rtex);
532 return 0;
533 }
534
535 /* Needs adjustment for pixelformat:
536 */
537 static INLINE unsigned u_box_volume( const struct pipe_box *box )
538 {
539 return box->width * box->depth * box->height;
540 };
541
542 struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
543 struct pipe_resource *texture,
544 unsigned level,
545 unsigned usage,
546 const struct pipe_box *box)
547 {
548 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
549 struct pipe_resource resource;
550 struct r600_transfer *trans;
551 int r;
552 boolean use_staging_texture = FALSE;
553
554 /* We cannot map a tiled texture directly because the data is
555 * in a different order, therefore we do detiling using a blit.
556 *
557 * Also, use a temporary in GTT memory for read transfers, as
558 * the CPU is much happier reading out of cached system memory
559 * than uncached VRAM.
560 */
561 if (R600_TEX_IS_TILED(rtex, level))
562 use_staging_texture = TRUE;
563
564 if ((usage & PIPE_TRANSFER_READ) && u_box_volume(box) > 1024)
565 use_staging_texture = TRUE;
566
567 /* XXX: Use a staging texture for uploads if the underlying BO
568 * is busy. No interface for checking that currently? so do
569 * it eagerly whenever the transfer doesn't require a readback
570 * and might block.
571 */
572 if ((usage & PIPE_TRANSFER_WRITE) &&
573 !(usage & (PIPE_TRANSFER_READ |
574 PIPE_TRANSFER_DONTBLOCK |
575 PIPE_TRANSFER_UNSYNCHRONIZED)))
576 use_staging_texture = TRUE;
577
578 if (!permit_hardware_blit(ctx->screen, texture) ||
579 (texture->flags & R600_RESOURCE_FLAG_TRANSFER))
580 use_staging_texture = FALSE;
581
582 trans = CALLOC_STRUCT(r600_transfer);
583 if (trans == NULL)
584 return NULL;
585 pipe_resource_reference(&trans->transfer.resource, texture);
586 trans->transfer.level = level;
587 trans->transfer.usage = usage;
588 trans->transfer.box = *box;
589 if (rtex->depth) {
590 /* XXX: only readback the rectangle which is being mapped?
591 */
592 /* XXX: when discard is true, no need to read back from depth texture
593 */
594 r = r600_texture_depth_flush(ctx, texture, FALSE);
595 if (r < 0) {
596 R600_ERR("failed to create temporary texture to hold untiled copy\n");
597 pipe_resource_reference(&trans->transfer.resource, NULL);
598 FREE(trans);
599 return NULL;
600 }
601 trans->transfer.stride = rtex->flushed_depth_texture->pitch_in_bytes[level];
602 trans->offset = r600_texture_get_offset(rtex->flushed_depth_texture, level, box->z);
603 return &trans->transfer;
604 } else if (use_staging_texture) {
605 resource.target = PIPE_TEXTURE_2D;
606 resource.format = texture->format;
607 resource.width0 = box->width;
608 resource.height0 = box->height;
609 resource.depth0 = 1;
610 resource.array_size = 1;
611 resource.last_level = 0;
612 resource.nr_samples = 0;
613 resource.usage = PIPE_USAGE_STAGING;
614 resource.bind = 0;
615 resource.flags = R600_RESOURCE_FLAG_TRANSFER;
616 /* For texture reading, the temporary (detiled) texture is used as
617 * a render target when blitting from a tiled texture. */
618 if (usage & PIPE_TRANSFER_READ) {
619 resource.bind |= PIPE_BIND_RENDER_TARGET;
620 }
621 /* For texture writing, the temporary texture is used as a sampler
622 * when blitting into a tiled texture. */
623 if (usage & PIPE_TRANSFER_WRITE) {
624 resource.bind |= PIPE_BIND_SAMPLER_VIEW;
625 }
626 /* Create the temporary texture. */
627 trans->staging_texture = ctx->screen->resource_create(ctx->screen, &resource);
628 if (trans->staging_texture == NULL) {
629 R600_ERR("failed to create temporary texture to hold untiled copy\n");
630 pipe_resource_reference(&trans->transfer.resource, NULL);
631 FREE(trans);
632 return NULL;
633 }
634
635 trans->transfer.stride =
636 ((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
637 if (usage & PIPE_TRANSFER_READ) {
638 r600_copy_to_staging_texture(ctx, trans);
639 /* Always referenced in the blit. */
640 ctx->flush(ctx, NULL);
641 }
642 return &trans->transfer;
643 }
644 trans->transfer.stride = rtex->pitch_in_bytes[level];
645 trans->transfer.layer_stride = rtex->layer_size[level];
646 trans->offset = r600_texture_get_offset(rtex, level, box->z);
647 return &trans->transfer;
648 }
649
650 void r600_texture_transfer_destroy(struct pipe_context *ctx,
651 struct pipe_transfer *transfer)
652 {
653 struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
654 struct pipe_resource *texture = transfer->resource;
655 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
656
657 if (rtransfer->staging_texture) {
658 if (transfer->usage & PIPE_TRANSFER_WRITE) {
659 r600_copy_from_staging_texture(ctx, rtransfer);
660 }
661 pipe_resource_reference(&rtransfer->staging_texture, NULL);
662 }
663
664 if (rtex->depth && !rtex->is_flushing_texture) {
665 if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtex->flushed_depth_texture)
666 r600_blit_push_depth(ctx, rtex);
667 }
668
669 pipe_resource_reference(&transfer->resource, NULL);
670 FREE(transfer);
671 }
672
673 void* r600_texture_transfer_map(struct pipe_context *ctx,
674 struct pipe_transfer* transfer)
675 {
676 struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
677 struct r600_bo *bo;
678 enum pipe_format format = transfer->resource->format;
679 struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
680 unsigned offset = 0;
681 unsigned usage = 0;
682 char *map;
683
684 if (rtransfer->staging_texture) {
685 bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
686 } else {
687 struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
688
689 if (rtex->flushed_depth_texture)
690 bo = ((struct r600_resource *)rtex->flushed_depth_texture)->bo;
691 else
692 bo = ((struct r600_resource *)transfer->resource)->bo;
693
694 offset = rtransfer->offset +
695 transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
696 transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
697 }
698
699 if (transfer->usage & PIPE_TRANSFER_WRITE) {
700 usage |= PB_USAGE_CPU_WRITE;
701
702 if (transfer->usage & PIPE_TRANSFER_DISCARD) {
703 }
704
705 if (transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT) {
706 }
707 }
708
709 if (transfer->usage & PIPE_TRANSFER_READ) {
710 usage |= PB_USAGE_CPU_READ;
711 }
712
713 if (transfer->usage & PIPE_TRANSFER_DONTBLOCK) {
714 usage |= PB_USAGE_DONTBLOCK;
715 }
716
717 if (transfer->usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
718 usage |= PB_USAGE_UNSYNCHRONIZED;
719 }
720
721 map = r600_bo_map(radeon, bo, usage, ctx);
722 if (!map) {
723 return NULL;
724 }
725
726 return map + offset;
727 }
728
729 void r600_texture_transfer_unmap(struct pipe_context *ctx,
730 struct pipe_transfer* transfer)
731 {
732 struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
733 struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
734 struct r600_bo *bo;
735
736 if (rtransfer->staging_texture) {
737 bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
738 } else {
739 struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
740
741 if (rtex->flushed_depth_texture) {
742 bo = ((struct r600_resource *)rtex->flushed_depth_texture)->bo;
743 } else {
744 bo = ((struct r600_resource *)transfer->resource)->bo;
745 }
746 }
747 r600_bo_unmap(radeon, bo);
748 }
749
750 void r600_init_surface_functions(struct r600_pipe_context *r600)
751 {
752 r600->context.create_surface = r600_create_surface;
753 r600->context.surface_destroy = r600_surface_destroy;
754 }
755
756 static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
757 const unsigned char *swizzle_view)
758 {
759 unsigned i;
760 unsigned char swizzle[4];
761 unsigned result = 0;
762 const uint32_t swizzle_shift[4] = {
763 16, 19, 22, 25,
764 };
765 const uint32_t swizzle_bit[4] = {
766 0, 1, 2, 3,
767 };
768
769 if (swizzle_view) {
770 /* Combine two sets of swizzles. */
771 for (i = 0; i < 4; i++) {
772 swizzle[i] = swizzle_view[i] <= UTIL_FORMAT_SWIZZLE_W ?
773 swizzle_format[swizzle_view[i]] : swizzle_view[i];
774 }
775 } else {
776 memcpy(swizzle, swizzle_format, 4);
777 }
778
779 /* Get swizzle. */
780 for (i = 0; i < 4; i++) {
781 switch (swizzle[i]) {
782 case UTIL_FORMAT_SWIZZLE_Y:
783 result |= swizzle_bit[1] << swizzle_shift[i];
784 break;
785 case UTIL_FORMAT_SWIZZLE_Z:
786 result |= swizzle_bit[2] << swizzle_shift[i];
787 break;
788 case UTIL_FORMAT_SWIZZLE_W:
789 result |= swizzle_bit[3] << swizzle_shift[i];
790 break;
791 case UTIL_FORMAT_SWIZZLE_0:
792 result |= V_038010_SQ_SEL_0 << swizzle_shift[i];
793 break;
794 case UTIL_FORMAT_SWIZZLE_1:
795 result |= V_038010_SQ_SEL_1 << swizzle_shift[i];
796 break;
797 default: /* UTIL_FORMAT_SWIZZLE_X */
798 result |= swizzle_bit[0] << swizzle_shift[i];
799 }
800 }
801 return result;
802 }
803
804 /* texture format translate */
805 uint32_t r600_translate_texformat(struct pipe_screen *screen,
806 enum pipe_format format,
807 const unsigned char *swizzle_view,
808 uint32_t *word4_p, uint32_t *yuv_format_p)
809 {
810 uint32_t result = 0, word4 = 0, yuv_format = 0;
811 const struct util_format_description *desc;
812 boolean uniform = TRUE;
813 static int r600_enable_s3tc = -1;
814
815 int i;
816 const uint32_t sign_bit[4] = {
817 S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED),
818 S_038010_FORMAT_COMP_Y(V_038010_SQ_FORMAT_COMP_SIGNED),
819 S_038010_FORMAT_COMP_Z(V_038010_SQ_FORMAT_COMP_SIGNED),
820 S_038010_FORMAT_COMP_W(V_038010_SQ_FORMAT_COMP_SIGNED)
821 };
822 desc = util_format_description(format);
823
824 word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view);
825
826 /* Colorspace (return non-RGB formats directly). */
827 switch (desc->colorspace) {
828 /* Depth stencil formats */
829 case UTIL_FORMAT_COLORSPACE_ZS:
830 switch (format) {
831 case PIPE_FORMAT_Z16_UNORM:
832 result = FMT_16;
833 goto out_word4;
834 case PIPE_FORMAT_X24S8_USCALED:
835 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
836 case PIPE_FORMAT_Z24X8_UNORM:
837 case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
838 result = FMT_8_24;
839 goto out_word4;
840 case PIPE_FORMAT_S8X24_USCALED:
841 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
842 case PIPE_FORMAT_X8Z24_UNORM:
843 case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
844 result = FMT_24_8;
845 goto out_word4;
846 case PIPE_FORMAT_S8_USCALED:
847 result = FMT_8;
848 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
849 goto out_word4;
850 default:
851 goto out_unknown;
852 }
853
854 case UTIL_FORMAT_COLORSPACE_YUV:
855 yuv_format |= (1 << 30);
856 switch (format) {
857 case PIPE_FORMAT_UYVY:
858 case PIPE_FORMAT_YUYV:
859 default:
860 break;
861 }
862 goto out_unknown; /* TODO */
863
864 case UTIL_FORMAT_COLORSPACE_SRGB:
865 word4 |= S_038010_FORCE_DEGAMMA(1);
866 break;
867
868 default:
869 break;
870 }
871
872 if (r600_enable_s3tc == -1) {
873 struct r600_screen *rscreen = (struct r600_screen *)screen;
874 if (r600_get_minor_version(rscreen->radeon) >= 9)
875 r600_enable_s3tc = 1;
876 else
877 r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
878 }
879
880 if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
881 if (!r600_enable_s3tc)
882 goto out_unknown;
883
884 switch (format) {
885 case PIPE_FORMAT_RGTC1_SNORM:
886 case PIPE_FORMAT_LATC1_SNORM:
887 word4 |= sign_bit[0];
888 case PIPE_FORMAT_RGTC1_UNORM:
889 case PIPE_FORMAT_LATC1_UNORM:
890 result = FMT_BC4;
891 goto out_word4;
892 case PIPE_FORMAT_RGTC2_SNORM:
893 case PIPE_FORMAT_LATC2_SNORM:
894 word4 |= sign_bit[0] | sign_bit[1];
895 case PIPE_FORMAT_RGTC2_UNORM:
896 case PIPE_FORMAT_LATC2_UNORM:
897 result = FMT_BC5;
898 goto out_word4;
899 default:
900 goto out_unknown;
901 }
902 }
903
904 if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
905
906 if (!r600_enable_s3tc)
907 goto out_unknown;
908
909 if (!util_format_s3tc_enabled) {
910 goto out_unknown;
911 }
912
913 switch (format) {
914 case PIPE_FORMAT_DXT1_RGB:
915 case PIPE_FORMAT_DXT1_RGBA:
916 case PIPE_FORMAT_DXT1_SRGB:
917 case PIPE_FORMAT_DXT1_SRGBA:
918 result = FMT_BC1;
919 goto out_word4;
920 case PIPE_FORMAT_DXT3_RGBA:
921 case PIPE_FORMAT_DXT3_SRGBA:
922 result = FMT_BC2;
923 goto out_word4;
924 case PIPE_FORMAT_DXT5_RGBA:
925 case PIPE_FORMAT_DXT5_SRGBA:
926 result = FMT_BC3;
927 goto out_word4;
928 default:
929 goto out_unknown;
930 }
931 }
932
933 if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) {
934 result = FMT_5_9_9_9_SHAREDEXP;
935 goto out_word4;
936 }
937
938
939 for (i = 0; i < desc->nr_channels; i++) {
940 if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
941 word4 |= sign_bit[i];
942 }
943 }
944
945 /* R8G8Bx_SNORM - TODO CxV8U8 */
946
947 /* See whether the components are of the same size. */
948 for (i = 1; i < desc->nr_channels; i++) {
949 uniform = uniform && desc->channel[0].size == desc->channel[i].size;
950 }
951
952 /* Non-uniform formats. */
953 if (!uniform) {
954 switch(desc->nr_channels) {
955 case 3:
956 if (desc->channel[0].size == 5 &&
957 desc->channel[1].size == 6 &&
958 desc->channel[2].size == 5) {
959 result = FMT_5_6_5;
960 goto out_word4;
961 }
962 goto out_unknown;
963 case 4:
964 if (desc->channel[0].size == 5 &&
965 desc->channel[1].size == 5 &&
966 desc->channel[2].size == 5 &&
967 desc->channel[3].size == 1) {
968 result = FMT_1_5_5_5;
969 goto out_word4;
970 }
971 if (desc->channel[0].size == 10 &&
972 desc->channel[1].size == 10 &&
973 desc->channel[2].size == 10 &&
974 desc->channel[3].size == 2) {
975 result = FMT_2_10_10_10;
976 goto out_word4;
977 }
978 goto out_unknown;
979 }
980 goto out_unknown;
981 }
982
983 /* Find the first non-VOID channel. */
984 for (i = 0; i < 4; i++) {
985 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
986 break;
987 }
988 }
989
990 if (i == 4)
991 goto out_unknown;
992
993 /* uniform formats */
994 switch (desc->channel[i].type) {
995 case UTIL_FORMAT_TYPE_UNSIGNED:
996 case UTIL_FORMAT_TYPE_SIGNED:
997 if (!desc->channel[i].normalized &&
998 desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
999 goto out_unknown;
1000 }
1001
1002 switch (desc->channel[i].size) {
1003 case 4:
1004 switch (desc->nr_channels) {
1005 case 2:
1006 result = FMT_4_4;
1007 goto out_word4;
1008 case 4:
1009 result = FMT_4_4_4_4;
1010 goto out_word4;
1011 }
1012 goto out_unknown;
1013 case 8:
1014 switch (desc->nr_channels) {
1015 case 1:
1016 result = FMT_8;
1017 goto out_word4;
1018 case 2:
1019 result = FMT_8_8;
1020 goto out_word4;
1021 case 4:
1022 result = FMT_8_8_8_8;
1023 goto out_word4;
1024 }
1025 goto out_unknown;
1026 case 16:
1027 switch (desc->nr_channels) {
1028 case 1:
1029 result = FMT_16;
1030 goto out_word4;
1031 case 2:
1032 result = FMT_16_16;
1033 goto out_word4;
1034 case 4:
1035 result = FMT_16_16_16_16;
1036 goto out_word4;
1037 }
1038 goto out_unknown;
1039 case 32:
1040 switch (desc->nr_channels) {
1041 case 1:
1042 result = FMT_32;
1043 goto out_word4;
1044 case 2:
1045 result = FMT_32_32;
1046 goto out_word4;
1047 case 4:
1048 result = FMT_32_32_32_32;
1049 goto out_word4;
1050 }
1051 }
1052 goto out_unknown;
1053
1054 case UTIL_FORMAT_TYPE_FLOAT:
1055 switch (desc->channel[i].size) {
1056 case 16:
1057 switch (desc->nr_channels) {
1058 case 1:
1059 result = FMT_16_FLOAT;
1060 goto out_word4;
1061 case 2:
1062 result = FMT_16_16_FLOAT;
1063 goto out_word4;
1064 case 4:
1065 result = FMT_16_16_16_16_FLOAT;
1066 goto out_word4;
1067 }
1068 goto out_unknown;
1069 case 32:
1070 switch (desc->nr_channels) {
1071 case 1:
1072 result = FMT_32_FLOAT;
1073 goto out_word4;
1074 case 2:
1075 result = FMT_32_32_FLOAT;
1076 goto out_word4;
1077 case 4:
1078 result = FMT_32_32_32_32_FLOAT;
1079 goto out_word4;
1080 }
1081 }
1082
1083 }
1084 out_word4:
1085 if (word4_p)
1086 *word4_p = word4;
1087 if (yuv_format_p)
1088 *yuv_format_p = yuv_format;
1089 return result;
1090 out_unknown:
1091 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
1092 return ~0;
1093 }