nvc0: implement new stream output interface
[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 "pipebuffer/pb_buffer.h"
35 #include "r600_pipe.h"
36 #include "r600_resource.h"
37 #include "r600d.h"
38 #include "r600_formats.h"
39
40 /* Copy from a full GPU texture to a transfer's staging one. */
41 static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
42 {
43 struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
44 struct pipe_resource *texture = transfer->resource;
45
46 ctx->resource_copy_region(ctx, rtransfer->staging_texture,
47 0, 0, 0, 0, texture, transfer->level,
48 &transfer->box);
49 }
50
51
52 /* Copy from a transfer's staging texture to a full GPU one. */
53 static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
54 {
55 struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
56 struct pipe_resource *texture = transfer->resource;
57 struct pipe_box sbox;
58
59 sbox.x = sbox.y = sbox.z = 0;
60 sbox.width = transfer->box.width;
61 sbox.height = transfer->box.height;
62 /* XXX that might be wrong */
63 sbox.depth = 1;
64 ctx->resource_copy_region(ctx, texture, transfer->level,
65 transfer->box.x, transfer->box.y, transfer->box.z,
66 rtransfer->staging_texture,
67 0, &sbox);
68 }
69
70 unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
71 unsigned level, unsigned layer)
72 {
73 unsigned offset = rtex->offset[level];
74
75 switch (rtex->resource.b.b.b.target) {
76 case PIPE_TEXTURE_3D:
77 case PIPE_TEXTURE_CUBE:
78 default:
79 return offset + layer * rtex->layer_size[level];
80 }
81 }
82
83 static unsigned r600_get_block_alignment(struct pipe_screen *screen,
84 enum pipe_format format,
85 unsigned array_mode)
86 {
87 struct r600_screen* rscreen = (struct r600_screen *)screen;
88 unsigned pixsize = util_format_get_blocksize(format);
89 int p_align;
90
91 switch(array_mode) {
92 case V_038000_ARRAY_1D_TILED_THIN1:
93 p_align = MAX2(8,
94 ((rscreen->tiling_info.group_bytes / 8 / pixsize)));
95 break;
96 case V_038000_ARRAY_2D_TILED_THIN1:
97 p_align = MAX2(rscreen->tiling_info.num_banks,
98 (((rscreen->tiling_info.group_bytes / 8 / pixsize)) *
99 rscreen->tiling_info.num_banks)) * 8;
100 break;
101 case V_038000_ARRAY_LINEAR_ALIGNED:
102 p_align = MAX2(64, rscreen->tiling_info.group_bytes / pixsize);
103 break;
104 case V_038000_ARRAY_LINEAR_GENERAL:
105 default:
106 p_align = rscreen->tiling_info.group_bytes / pixsize;
107 break;
108 }
109 return p_align;
110 }
111
112 static unsigned r600_get_height_alignment(struct pipe_screen *screen,
113 unsigned array_mode)
114 {
115 struct r600_screen* rscreen = (struct r600_screen *)screen;
116 int h_align;
117
118 switch (array_mode) {
119 case V_038000_ARRAY_2D_TILED_THIN1:
120 h_align = rscreen->tiling_info.num_channels * 8;
121 break;
122 case V_038000_ARRAY_1D_TILED_THIN1:
123 case V_038000_ARRAY_LINEAR_ALIGNED:
124 h_align = 8;
125 break;
126 case V_038000_ARRAY_LINEAR_GENERAL:
127 default:
128 h_align = 1;
129 break;
130 }
131 return h_align;
132 }
133
134 static unsigned r600_get_base_alignment(struct pipe_screen *screen,
135 enum pipe_format format,
136 unsigned array_mode)
137 {
138 struct r600_screen* rscreen = (struct r600_screen *)screen;
139 unsigned pixsize = util_format_get_blocksize(format);
140 int p_align = r600_get_block_alignment(screen, format, array_mode);
141 int h_align = r600_get_height_alignment(screen, array_mode);
142 int b_align;
143
144 switch (array_mode) {
145 case V_038000_ARRAY_2D_TILED_THIN1:
146 b_align = MAX2(rscreen->tiling_info.num_banks * rscreen->tiling_info.num_channels * 8 * 8 * pixsize,
147 p_align * pixsize * h_align);
148 break;
149 case V_038000_ARRAY_1D_TILED_THIN1:
150 case V_038000_ARRAY_LINEAR_ALIGNED:
151 case V_038000_ARRAY_LINEAR_GENERAL:
152 default:
153 b_align = rscreen->tiling_info.group_bytes;
154 break;
155 }
156 return b_align;
157 }
158
159 static unsigned mip_minify(unsigned size, unsigned level)
160 {
161 unsigned val;
162 val = u_minify(size, level);
163 if (level > 0)
164 val = util_next_power_of_two(val);
165 return val;
166 }
167
168 static unsigned r600_texture_get_nblocksx(struct pipe_screen *screen,
169 struct r600_resource_texture *rtex,
170 unsigned level)
171 {
172 struct pipe_resource *ptex = &rtex->resource.b.b.b;
173 unsigned nblocksx, block_align, width;
174 unsigned blocksize = util_format_get_blocksize(rtex->real_format);
175
176 if (rtex->pitch_override)
177 return rtex->pitch_override / blocksize;
178
179 width = mip_minify(ptex->width0, level);
180 nblocksx = util_format_get_nblocksx(rtex->real_format, width);
181
182 block_align = r600_get_block_alignment(screen, rtex->real_format,
183 rtex->array_mode[level]);
184 nblocksx = align(nblocksx, block_align);
185 return nblocksx;
186 }
187
188 static unsigned r600_texture_get_nblocksy(struct pipe_screen *screen,
189 struct r600_resource_texture *rtex,
190 unsigned level)
191 {
192 struct pipe_resource *ptex = &rtex->resource.b.b.b;
193 unsigned height, tile_height;
194
195 height = mip_minify(ptex->height0, level);
196 height = util_format_get_nblocksy(rtex->real_format, height);
197 tile_height = r600_get_height_alignment(screen,
198 rtex->array_mode[level]);
199
200 /* XXX Hack around an alignment issue. Less tests fail with this.
201 *
202 * The thing is depth-stencil buffers should be tiled, i.e.
203 * the alignment should be >=8. If I make them tiled, stencil starts
204 * working because it no longer overlaps with the depth buffer
205 * in memory, but texturing like drawpix-stencil breaks. */
206 if (util_format_is_depth_or_stencil(rtex->real_format) && tile_height < 8)
207 tile_height = 8;
208
209 height = align(height, tile_height);
210 return height;
211 }
212
213 static void r600_texture_set_array_mode(struct pipe_screen *screen,
214 struct r600_resource_texture *rtex,
215 unsigned level, unsigned array_mode)
216 {
217 struct pipe_resource *ptex = &rtex->resource.b.b.b;
218
219 switch (array_mode) {
220 case V_0280A0_ARRAY_LINEAR_GENERAL:
221 case V_0280A0_ARRAY_LINEAR_ALIGNED:
222 case V_0280A0_ARRAY_1D_TILED_THIN1:
223 default:
224 rtex->array_mode[level] = array_mode;
225 break;
226 case V_0280A0_ARRAY_2D_TILED_THIN1:
227 {
228 unsigned w, h, tile_height, tile_width;
229
230 tile_height = r600_get_height_alignment(screen, array_mode);
231 tile_width = r600_get_block_alignment(screen, rtex->real_format, array_mode);
232
233 w = mip_minify(ptex->width0, level);
234 h = mip_minify(ptex->height0, level);
235 if (w <= tile_width || h <= tile_height)
236 rtex->array_mode[level] = V_0280A0_ARRAY_1D_TILED_THIN1;
237 else
238 rtex->array_mode[level] = array_mode;
239 }
240 break;
241 }
242 }
243
244 static void r600_setup_miptree(struct pipe_screen *screen,
245 struct r600_resource_texture *rtex,
246 unsigned array_mode)
247 {
248 struct pipe_resource *ptex = &rtex->resource.b.b.b;
249 enum chip_class chipc = ((struct r600_screen*)screen)->chip_class;
250 unsigned size, layer_size, i, offset;
251 unsigned nblocksx, nblocksy;
252
253 for (i = 0, offset = 0; i <= ptex->last_level; i++) {
254 unsigned blocksize = util_format_get_blocksize(rtex->real_format);
255 unsigned base_align = r600_get_base_alignment(screen, rtex->real_format, array_mode);
256
257 r600_texture_set_array_mode(screen, rtex, i, array_mode);
258
259 nblocksx = r600_texture_get_nblocksx(screen, rtex, i);
260 nblocksy = r600_texture_get_nblocksy(screen, rtex, i);
261
262 if (chipc >= EVERGREEN && array_mode == V_038000_ARRAY_LINEAR_GENERAL)
263 layer_size = align(nblocksx, 64) * nblocksy * blocksize;
264 else
265 layer_size = nblocksx * nblocksy * blocksize;
266
267 if (ptex->target == PIPE_TEXTURE_CUBE) {
268 if (chipc >= R700)
269 size = layer_size * 8;
270 else
271 size = layer_size * 6;
272 }
273 else if (ptex->target == PIPE_TEXTURE_3D)
274 size = layer_size * u_minify(ptex->depth0, i);
275 else
276 size = layer_size * ptex->array_size;
277
278 /* align base image and start of miptree */
279 if ((i == 0) || (i == 1))
280 offset = align(offset, base_align);
281 rtex->offset[i] = offset;
282 rtex->layer_size[i] = layer_size;
283 rtex->pitch_in_blocks[i] = nblocksx; /* CB talks in elements */
284 rtex->pitch_in_bytes[i] = nblocksx * blocksize;
285
286 offset += size;
287 }
288 rtex->size = offset;
289 }
290
291 /* Figure out whether u_blitter will fallback to a transfer operation.
292 * If so, don't use a staging resource.
293 */
294 static boolean permit_hardware_blit(struct pipe_screen *screen,
295 const struct pipe_resource *res)
296 {
297 unsigned bind;
298
299 if (util_format_is_depth_or_stencil(res->format))
300 bind = PIPE_BIND_DEPTH_STENCIL;
301 else
302 bind = PIPE_BIND_RENDER_TARGET;
303
304 /* hackaround for S3TC */
305 if (util_format_is_compressed(res->format))
306 return TRUE;
307
308 if (!screen->is_format_supported(screen,
309 res->format,
310 res->target,
311 res->nr_samples,
312 bind))
313 return FALSE;
314
315 if (!screen->is_format_supported(screen,
316 res->format,
317 res->target,
318 res->nr_samples,
319 PIPE_BIND_SAMPLER_VIEW))
320 return FALSE;
321
322 switch (res->usage) {
323 case PIPE_USAGE_STREAM:
324 case PIPE_USAGE_STAGING:
325 return FALSE;
326
327 default:
328 return TRUE;
329 }
330 }
331
332 static boolean r600_texture_get_handle(struct pipe_screen* screen,
333 struct pipe_resource *ptex,
334 struct winsys_handle *whandle)
335 {
336 struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
337 struct r600_resource *resource = &rtex->resource;
338 struct r600_screen *rscreen = (struct r600_screen*)screen;
339
340 return rscreen->ws->buffer_get_handle(resource->buf,
341 rtex->pitch_in_bytes[0], whandle);
342 }
343
344 static void r600_texture_destroy(struct pipe_screen *screen,
345 struct pipe_resource *ptex)
346 {
347 struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
348 struct r600_resource *resource = &rtex->resource;
349
350 if (rtex->flushed_depth_texture)
351 pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
352
353 if (rtex->stencil)
354 pipe_resource_reference((struct pipe_resource **)&rtex->stencil, NULL);
355
356 pb_reference(&resource->buf, NULL);
357 FREE(rtex);
358 }
359
360 static const struct u_resource_vtbl r600_texture_vtbl =
361 {
362 r600_texture_get_handle, /* get_handle */
363 r600_texture_destroy, /* resource_destroy */
364 r600_texture_get_transfer, /* get_transfer */
365 r600_texture_transfer_destroy, /* transfer_destroy */
366 r600_texture_transfer_map, /* transfer_map */
367 u_default_transfer_flush_region,/* transfer_flush_region */
368 r600_texture_transfer_unmap, /* transfer_unmap */
369 u_default_transfer_inline_write /* transfer_inline_write */
370 };
371
372 static struct r600_resource_texture *
373 r600_texture_create_object(struct pipe_screen *screen,
374 const struct pipe_resource *base,
375 unsigned array_mode,
376 unsigned pitch_in_bytes_override,
377 unsigned max_buffer_size,
378 struct pb_buffer *buf,
379 boolean alloc_bo)
380 {
381 struct r600_resource_texture *rtex;
382 struct r600_resource *resource;
383 struct r600_screen *rscreen = (struct r600_screen*)screen;
384
385 rtex = CALLOC_STRUCT(r600_resource_texture);
386 if (rtex == NULL)
387 return NULL;
388
389 resource = &rtex->resource;
390 resource->b.b.b = *base;
391 resource->b.b.vtbl = &r600_texture_vtbl;
392 pipe_reference_init(&resource->b.b.b.reference, 1);
393 resource->b.b.b.screen = screen;
394 rtex->pitch_override = pitch_in_bytes_override;
395 rtex->real_format = base->format;
396
397 /* We must split depth and stencil into two separate buffers on Evergreen. */
398 if (!(base->flags & R600_RESOURCE_FLAG_TRANSFER) &&
399 ((struct r600_screen*)screen)->chip_class >= EVERGREEN &&
400 util_format_is_depth_and_stencil(base->format)) {
401 struct pipe_resource stencil;
402 unsigned stencil_pitch_override = 0;
403
404 switch (base->format) {
405 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
406 rtex->real_format = PIPE_FORMAT_Z24X8_UNORM;
407 break;
408 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
409 rtex->real_format = PIPE_FORMAT_X8Z24_UNORM;
410 break;
411 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
412 rtex->real_format = PIPE_FORMAT_Z32_FLOAT;
413 break;
414 default:
415 assert(0);
416 FREE(rtex);
417 return NULL;
418 }
419
420 /* Divide the pitch in bytes by 4 for stencil, because it has a smaller pixel size. */
421 if (pitch_in_bytes_override) {
422 assert(base->format == PIPE_FORMAT_Z24_UNORM_S8_UINT ||
423 base->format == PIPE_FORMAT_S8_UINT_Z24_UNORM);
424 stencil_pitch_override = pitch_in_bytes_override / 4;
425 }
426
427 /* Allocate the stencil buffer. */
428 stencil = *base;
429 stencil.format = PIPE_FORMAT_S8_UINT;
430 rtex->stencil = r600_texture_create_object(screen, &stencil, array_mode,
431 stencil_pitch_override,
432 max_buffer_size, NULL, FALSE);
433 if (!rtex->stencil) {
434 FREE(rtex);
435 return NULL;
436 }
437 /* Proceed in creating the depth buffer. */
438 }
439
440 /* only mark depth textures the HW can hit as depth textures */
441 if (util_format_is_depth_or_stencil(rtex->real_format) && permit_hardware_blit(screen, base))
442 rtex->depth = 1;
443
444 r600_setup_miptree(screen, rtex, array_mode);
445
446 /* If we initialized separate stencil for Evergreen. place it after depth. */
447 if (rtex->stencil) {
448 unsigned stencil_align, stencil_offset;
449
450 stencil_align = r600_get_base_alignment(screen, rtex->stencil->real_format, array_mode);
451 stencil_offset = align(rtex->size, stencil_align);
452
453 for (unsigned i = 0; i <= rtex->stencil->resource.b.b.b.last_level; i++)
454 rtex->stencil->offset[i] += stencil_offset;
455
456 rtex->size = stencil_offset + rtex->stencil->size;
457 }
458
459 /* Now create the backing buffer. */
460 if (!buf && alloc_bo) {
461 struct pipe_resource *ptex = &rtex->resource.b.b.b;
462 unsigned base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
463
464 if (!r600_init_resource(rscreen, resource, rtex->size, base_align, base->bind, base->usage)) {
465 pipe_resource_reference((struct pipe_resource**)&rtex->stencil, NULL);
466 FREE(rtex);
467 return NULL;
468 }
469 } else if (buf) {
470 resource->buf = buf;
471 resource->cs_buf = rscreen->ws->buffer_get_cs_handle(buf);
472 }
473
474 if (rtex->stencil) {
475 pb_reference(&rtex->stencil->resource.buf, rtex->resource.buf);
476 rtex->stencil->resource.cs_buf = rtex->resource.cs_buf;
477 }
478 return rtex;
479 }
480
481 DEBUG_GET_ONCE_BOOL_OPTION(tiling_enabled, "R600_TILING", FALSE);
482
483 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
484 const struct pipe_resource *templ)
485 {
486 struct r600_screen *rscreen = (struct r600_screen*)screen;
487 unsigned array_mode = 0;
488
489 if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
490 !(templ->bind & PIPE_BIND_SCANOUT)) {
491 if (util_format_is_compressed(templ->format)) {
492 array_mode = V_038000_ARRAY_1D_TILED_THIN1;
493 }
494 else if (debug_get_option_tiling_enabled() &&
495 rscreen->info.drm_minor >= 9 &&
496 permit_hardware_blit(screen, templ)) {
497 array_mode = V_038000_ARRAY_2D_TILED_THIN1;
498 }
499 }
500
501 return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
502 0, 0, NULL, TRUE);
503 }
504
505 static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
506 struct pipe_resource *texture,
507 const struct pipe_surface *surf_tmpl)
508 {
509 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
510 struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
511 unsigned level = surf_tmpl->u.tex.level;
512
513 assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
514 if (surface == NULL)
515 return NULL;
516 /* XXX no offset */
517 /* offset = r600_texture_get_offset(rtex, level, surf_tmpl->u.tex.first_layer);*/
518 pipe_reference_init(&surface->base.reference, 1);
519 pipe_resource_reference(&surface->base.texture, texture);
520 surface->base.context = pipe;
521 surface->base.format = surf_tmpl->format;
522 surface->base.width = mip_minify(texture->width0, level);
523 surface->base.height = mip_minify(texture->height0, level);
524 surface->base.usage = surf_tmpl->usage;
525 surface->base.texture = texture;
526 surface->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
527 surface->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
528 surface->base.u.tex.level = level;
529
530 surface->aligned_height = r600_texture_get_nblocksy(pipe->screen,
531 rtex, level);
532 return &surface->base;
533 }
534
535 static void r600_surface_destroy(struct pipe_context *pipe,
536 struct pipe_surface *surface)
537 {
538 pipe_resource_reference(&surface->texture, NULL);
539 FREE(surface);
540 }
541
542 struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
543 const struct pipe_resource *templ,
544 struct winsys_handle *whandle)
545 {
546 struct r600_screen *rscreen = (struct r600_screen*)screen;
547 struct pb_buffer *buf = NULL;
548 unsigned stride = 0;
549 unsigned array_mode = 0;
550 enum radeon_bo_layout micro, macro;
551
552 /* Support only 2D textures without mipmaps */
553 if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) ||
554 templ->depth0 != 1 || templ->last_level != 0)
555 return NULL;
556
557 buf = rscreen->ws->buffer_from_handle(rscreen->ws, whandle, &stride);
558 if (!buf)
559 return NULL;
560
561 rscreen->ws->buffer_get_tiling(buf, &micro, &macro);
562
563 if (macro == RADEON_LAYOUT_TILED)
564 array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
565 else if (micro == RADEON_LAYOUT_TILED)
566 array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
567 else
568 array_mode = 0;
569
570 return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
571 stride, 0, buf, FALSE);
572 }
573
574 int r600_texture_depth_flush(struct pipe_context *ctx,
575 struct pipe_resource *texture, boolean just_create)
576 {
577 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
578 struct pipe_resource resource;
579
580 if (rtex->flushed_depth_texture)
581 goto out;
582
583 resource.target = texture->target;
584 resource.format = texture->format;
585 resource.width0 = texture->width0;
586 resource.height0 = texture->height0;
587 resource.depth0 = texture->depth0;
588 resource.array_size = texture->array_size;
589 resource.last_level = texture->last_level;
590 resource.nr_samples = texture->nr_samples;
591 resource.usage = PIPE_USAGE_DYNAMIC;
592 resource.bind = texture->bind | PIPE_BIND_DEPTH_STENCIL;
593 resource.flags = R600_RESOURCE_FLAG_TRANSFER | texture->flags;
594
595 rtex->flushed_depth_texture = (struct r600_resource_texture *)ctx->screen->resource_create(ctx->screen, &resource);
596 if (rtex->flushed_depth_texture == NULL) {
597 R600_ERR("failed to create temporary texture to hold untiled copy\n");
598 return -ENOMEM;
599 }
600
601 ((struct r600_resource_texture *)rtex->flushed_depth_texture)->is_flushing_texture = TRUE;
602 out:
603 if (just_create)
604 return 0;
605
606 /* XXX: only do this if the depth texture has actually changed:
607 */
608 r600_blit_uncompress_depth(ctx, rtex);
609 return 0;
610 }
611
612 /* Needs adjustment for pixelformat:
613 */
614 static INLINE unsigned u_box_volume( const struct pipe_box *box )
615 {
616 return box->width * box->depth * box->height;
617 };
618
619 struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
620 struct pipe_resource *texture,
621 unsigned level,
622 unsigned usage,
623 const struct pipe_box *box)
624 {
625 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
626 struct pipe_resource resource;
627 struct r600_transfer *trans;
628 int r;
629 boolean use_staging_texture = FALSE;
630
631 /* We cannot map a tiled texture directly because the data is
632 * in a different order, therefore we do detiling using a blit.
633 *
634 * Also, use a temporary in GTT memory for read transfers, as
635 * the CPU is much happier reading out of cached system memory
636 * than uncached VRAM.
637 */
638 if (R600_TEX_IS_TILED(rtex, level))
639 use_staging_texture = TRUE;
640
641 if ((usage & PIPE_TRANSFER_READ) && u_box_volume(box) > 1024)
642 use_staging_texture = TRUE;
643
644 /* XXX: Use a staging texture for uploads if the underlying BO
645 * is busy. No interface for checking that currently? so do
646 * it eagerly whenever the transfer doesn't require a readback
647 * and might block.
648 */
649 if ((usage & PIPE_TRANSFER_WRITE) &&
650 !(usage & (PIPE_TRANSFER_READ |
651 PIPE_TRANSFER_DONTBLOCK |
652 PIPE_TRANSFER_UNSYNCHRONIZED)))
653 use_staging_texture = TRUE;
654
655 if (!permit_hardware_blit(ctx->screen, texture) ||
656 (texture->flags & R600_RESOURCE_FLAG_TRANSFER))
657 use_staging_texture = FALSE;
658
659 if (use_staging_texture && (usage & PIPE_TRANSFER_MAP_DIRECTLY))
660 return NULL;
661
662 trans = CALLOC_STRUCT(r600_transfer);
663 if (trans == NULL)
664 return NULL;
665 pipe_resource_reference(&trans->transfer.resource, texture);
666 trans->transfer.level = level;
667 trans->transfer.usage = usage;
668 trans->transfer.box = *box;
669 if (rtex->depth) {
670 /* XXX: only readback the rectangle which is being mapped?
671 */
672 /* XXX: when discard is true, no need to read back from depth texture
673 */
674 r = r600_texture_depth_flush(ctx, texture, FALSE);
675 if (r < 0) {
676 R600_ERR("failed to create temporary texture to hold untiled copy\n");
677 pipe_resource_reference(&trans->transfer.resource, NULL);
678 FREE(trans);
679 return NULL;
680 }
681 trans->transfer.stride = rtex->flushed_depth_texture->pitch_in_bytes[level];
682 trans->offset = r600_texture_get_offset(rtex->flushed_depth_texture, level, box->z);
683 return &trans->transfer;
684 } else if (use_staging_texture) {
685 resource.target = PIPE_TEXTURE_2D;
686 resource.format = texture->format;
687 resource.width0 = box->width;
688 resource.height0 = box->height;
689 resource.depth0 = 1;
690 resource.array_size = 1;
691 resource.last_level = 0;
692 resource.nr_samples = 0;
693 resource.usage = PIPE_USAGE_STAGING;
694 resource.bind = 0;
695 resource.flags = R600_RESOURCE_FLAG_TRANSFER;
696 /* For texture reading, the temporary (detiled) texture is used as
697 * a render target when blitting from a tiled texture. */
698 if (usage & PIPE_TRANSFER_READ) {
699 resource.bind |= PIPE_BIND_RENDER_TARGET;
700 }
701 /* For texture writing, the temporary texture is used as a sampler
702 * when blitting into a tiled texture. */
703 if (usage & PIPE_TRANSFER_WRITE) {
704 resource.bind |= PIPE_BIND_SAMPLER_VIEW;
705 }
706 /* Create the temporary texture. */
707 trans->staging_texture = ctx->screen->resource_create(ctx->screen, &resource);
708 if (trans->staging_texture == NULL) {
709 R600_ERR("failed to create temporary texture to hold untiled copy\n");
710 pipe_resource_reference(&trans->transfer.resource, NULL);
711 FREE(trans);
712 return NULL;
713 }
714
715 trans->transfer.stride =
716 ((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
717 if (usage & PIPE_TRANSFER_READ) {
718 r600_copy_to_staging_texture(ctx, trans);
719 /* Always referenced in the blit. */
720 r600_flush(ctx, NULL, 0);
721 }
722 return &trans->transfer;
723 }
724 trans->transfer.stride = rtex->pitch_in_bytes[level];
725 trans->transfer.layer_stride = rtex->layer_size[level];
726 trans->offset = r600_texture_get_offset(rtex, level, box->z);
727 return &trans->transfer;
728 }
729
730 void r600_texture_transfer_destroy(struct pipe_context *ctx,
731 struct pipe_transfer *transfer)
732 {
733 struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
734 struct pipe_resource *texture = transfer->resource;
735 struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
736
737 if (rtransfer->staging_texture) {
738 if (transfer->usage & PIPE_TRANSFER_WRITE) {
739 r600_copy_from_staging_texture(ctx, rtransfer);
740 }
741 pipe_resource_reference(&rtransfer->staging_texture, NULL);
742 }
743
744 if (rtex->depth && !rtex->is_flushing_texture) {
745 if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtex->flushed_depth_texture)
746 r600_blit_push_depth(ctx, rtex);
747 }
748
749 pipe_resource_reference(&transfer->resource, NULL);
750 FREE(transfer);
751 }
752
753 void* r600_texture_transfer_map(struct pipe_context *ctx,
754 struct pipe_transfer* transfer)
755 {
756 struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
757 struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
758 struct pb_buffer *buf;
759 enum pipe_format format = transfer->resource->format;
760 unsigned offset = 0;
761 char *map;
762
763 if (rtransfer->staging_texture) {
764 buf = ((struct r600_resource *)rtransfer->staging_texture)->buf;
765 } else {
766 struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
767
768 if (rtex->flushed_depth_texture)
769 buf = ((struct r600_resource *)rtex->flushed_depth_texture)->buf;
770 else
771 buf = ((struct r600_resource *)transfer->resource)->buf;
772
773 offset = rtransfer->offset +
774 transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
775 transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
776 }
777
778 if (!(map = rctx->ws->buffer_map(buf, rctx->ctx.cs, transfer->usage))) {
779 return NULL;
780 }
781
782 return map + offset;
783 }
784
785 void r600_texture_transfer_unmap(struct pipe_context *ctx,
786 struct pipe_transfer* transfer)
787 {
788 struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
789 struct r600_pipe_context *rctx = (struct r600_pipe_context*)ctx;
790 struct pb_buffer *buf;
791
792 if (rtransfer->staging_texture) {
793 buf = ((struct r600_resource *)rtransfer->staging_texture)->buf;
794 } else {
795 struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
796
797 if (rtex->flushed_depth_texture) {
798 buf = ((struct r600_resource *)rtex->flushed_depth_texture)->buf;
799 } else {
800 buf = ((struct r600_resource *)transfer->resource)->buf;
801 }
802 }
803 rctx->ws->buffer_unmap(buf);
804 }
805
806 void r600_init_surface_functions(struct r600_pipe_context *r600)
807 {
808 r600->context.create_surface = r600_create_surface;
809 r600->context.surface_destroy = r600_surface_destroy;
810 }
811
812 static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
813 const unsigned char *swizzle_view)
814 {
815 unsigned i;
816 unsigned char swizzle[4];
817 unsigned result = 0;
818 const uint32_t swizzle_shift[4] = {
819 16, 19, 22, 25,
820 };
821 const uint32_t swizzle_bit[4] = {
822 0, 1, 2, 3,
823 };
824
825 if (swizzle_view) {
826 util_format_compose_swizzles(swizzle_format, swizzle_view, swizzle);
827 } else {
828 memcpy(swizzle, swizzle_format, 4);
829 }
830
831 /* Get swizzle. */
832 for (i = 0; i < 4; i++) {
833 switch (swizzle[i]) {
834 case UTIL_FORMAT_SWIZZLE_Y:
835 result |= swizzle_bit[1] << swizzle_shift[i];
836 break;
837 case UTIL_FORMAT_SWIZZLE_Z:
838 result |= swizzle_bit[2] << swizzle_shift[i];
839 break;
840 case UTIL_FORMAT_SWIZZLE_W:
841 result |= swizzle_bit[3] << swizzle_shift[i];
842 break;
843 case UTIL_FORMAT_SWIZZLE_0:
844 result |= V_038010_SQ_SEL_0 << swizzle_shift[i];
845 break;
846 case UTIL_FORMAT_SWIZZLE_1:
847 result |= V_038010_SQ_SEL_1 << swizzle_shift[i];
848 break;
849 default: /* UTIL_FORMAT_SWIZZLE_X */
850 result |= swizzle_bit[0] << swizzle_shift[i];
851 }
852 }
853 return result;
854 }
855
856 /* texture format translate */
857 uint32_t r600_translate_texformat(struct pipe_screen *screen,
858 enum pipe_format format,
859 const unsigned char *swizzle_view,
860 uint32_t *word4_p, uint32_t *yuv_format_p)
861 {
862 uint32_t result = 0, word4 = 0, yuv_format = 0;
863 const struct util_format_description *desc;
864 boolean uniform = TRUE;
865 static int r600_enable_s3tc = -1;
866
867 int i;
868 const uint32_t sign_bit[4] = {
869 S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED),
870 S_038010_FORMAT_COMP_Y(V_038010_SQ_FORMAT_COMP_SIGNED),
871 S_038010_FORMAT_COMP_Z(V_038010_SQ_FORMAT_COMP_SIGNED),
872 S_038010_FORMAT_COMP_W(V_038010_SQ_FORMAT_COMP_SIGNED)
873 };
874 desc = util_format_description(format);
875
876 word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view);
877
878 /* Colorspace (return non-RGB formats directly). */
879 switch (desc->colorspace) {
880 /* Depth stencil formats */
881 case UTIL_FORMAT_COLORSPACE_ZS:
882 switch (format) {
883 case PIPE_FORMAT_Z16_UNORM:
884 result = FMT_16;
885 goto out_word4;
886 case PIPE_FORMAT_X24S8_UINT:
887 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
888 case PIPE_FORMAT_Z24X8_UNORM:
889 case PIPE_FORMAT_Z24_UNORM_S8_UINT:
890 result = FMT_8_24;
891 goto out_word4;
892 case PIPE_FORMAT_S8X24_UINT:
893 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
894 case PIPE_FORMAT_X8Z24_UNORM:
895 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
896 result = FMT_24_8;
897 goto out_word4;
898 case PIPE_FORMAT_S8_UINT:
899 result = FMT_8;
900 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
901 goto out_word4;
902 case PIPE_FORMAT_Z32_FLOAT:
903 result = FMT_32_FLOAT;
904 goto out_word4;
905 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
906 result = FMT_X24_8_32_FLOAT;
907 goto out_word4;
908 default:
909 goto out_unknown;
910 }
911
912 case UTIL_FORMAT_COLORSPACE_YUV:
913 yuv_format |= (1 << 30);
914 switch (format) {
915 case PIPE_FORMAT_UYVY:
916 case PIPE_FORMAT_YUYV:
917 default:
918 break;
919 }
920 goto out_unknown; /* TODO */
921
922 case UTIL_FORMAT_COLORSPACE_SRGB:
923 word4 |= S_038010_FORCE_DEGAMMA(1);
924 break;
925
926 default:
927 break;
928 }
929
930 if (r600_enable_s3tc == -1) {
931 struct r600_screen *rscreen = (struct r600_screen *)screen;
932 if (rscreen->info.drm_minor >= 9)
933 r600_enable_s3tc = 1;
934 else
935 r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
936 }
937
938 if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
939 if (!r600_enable_s3tc)
940 goto out_unknown;
941
942 switch (format) {
943 case PIPE_FORMAT_RGTC1_SNORM:
944 case PIPE_FORMAT_LATC1_SNORM:
945 word4 |= sign_bit[0];
946 case PIPE_FORMAT_RGTC1_UNORM:
947 case PIPE_FORMAT_LATC1_UNORM:
948 result = FMT_BC4;
949 goto out_word4;
950 case PIPE_FORMAT_RGTC2_SNORM:
951 case PIPE_FORMAT_LATC2_SNORM:
952 word4 |= sign_bit[0] | sign_bit[1];
953 case PIPE_FORMAT_RGTC2_UNORM:
954 case PIPE_FORMAT_LATC2_UNORM:
955 result = FMT_BC5;
956 goto out_word4;
957 default:
958 goto out_unknown;
959 }
960 }
961
962 if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
963
964 if (!r600_enable_s3tc)
965 goto out_unknown;
966
967 if (!util_format_s3tc_enabled) {
968 goto out_unknown;
969 }
970
971 switch (format) {
972 case PIPE_FORMAT_DXT1_RGB:
973 case PIPE_FORMAT_DXT1_RGBA:
974 case PIPE_FORMAT_DXT1_SRGB:
975 case PIPE_FORMAT_DXT1_SRGBA:
976 result = FMT_BC1;
977 goto out_word4;
978 case PIPE_FORMAT_DXT3_RGBA:
979 case PIPE_FORMAT_DXT3_SRGBA:
980 result = FMT_BC2;
981 goto out_word4;
982 case PIPE_FORMAT_DXT5_RGBA:
983 case PIPE_FORMAT_DXT5_SRGBA:
984 result = FMT_BC3;
985 goto out_word4;
986 default:
987 goto out_unknown;
988 }
989 }
990
991 if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) {
992 result = FMT_5_9_9_9_SHAREDEXP;
993 goto out_word4;
994 } else if (format == PIPE_FORMAT_R11G11B10_FLOAT) {
995 result = FMT_10_11_11_FLOAT;
996 goto out_word4;
997 }
998
999
1000 for (i = 0; i < desc->nr_channels; i++) {
1001 if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
1002 word4 |= sign_bit[i];
1003 }
1004 }
1005
1006 /* R8G8Bx_SNORM - TODO CxV8U8 */
1007
1008 /* See whether the components are of the same size. */
1009 for (i = 1; i < desc->nr_channels; i++) {
1010 uniform = uniform && desc->channel[0].size == desc->channel[i].size;
1011 }
1012
1013 /* Non-uniform formats. */
1014 if (!uniform) {
1015 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
1016 desc->channel[0].pure_integer)
1017 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
1018 switch(desc->nr_channels) {
1019 case 3:
1020 if (desc->channel[0].size == 5 &&
1021 desc->channel[1].size == 6 &&
1022 desc->channel[2].size == 5) {
1023 result = FMT_5_6_5;
1024 goto out_word4;
1025 }
1026 goto out_unknown;
1027 case 4:
1028 if (desc->channel[0].size == 5 &&
1029 desc->channel[1].size == 5 &&
1030 desc->channel[2].size == 5 &&
1031 desc->channel[3].size == 1) {
1032 result = FMT_1_5_5_5;
1033 goto out_word4;
1034 }
1035 if (desc->channel[0].size == 10 &&
1036 desc->channel[1].size == 10 &&
1037 desc->channel[2].size == 10 &&
1038 desc->channel[3].size == 2) {
1039 result = FMT_2_10_10_10;
1040 goto out_word4;
1041 }
1042 goto out_unknown;
1043 }
1044 goto out_unknown;
1045 }
1046
1047 /* Find the first non-VOID channel. */
1048 for (i = 0; i < 4; i++) {
1049 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
1050 break;
1051 }
1052 }
1053
1054 if (i == 4)
1055 goto out_unknown;
1056
1057 /* uniform formats */
1058 switch (desc->channel[i].type) {
1059 case UTIL_FORMAT_TYPE_UNSIGNED:
1060 case UTIL_FORMAT_TYPE_SIGNED:
1061 #if 0
1062 if (!desc->channel[i].normalized &&
1063 desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
1064 goto out_unknown;
1065 }
1066 #endif
1067 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
1068 desc->channel[i].pure_integer)
1069 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
1070
1071 switch (desc->channel[i].size) {
1072 case 4:
1073 switch (desc->nr_channels) {
1074 case 2:
1075 result = FMT_4_4;
1076 goto out_word4;
1077 case 4:
1078 result = FMT_4_4_4_4;
1079 goto out_word4;
1080 }
1081 goto out_unknown;
1082 case 8:
1083 switch (desc->nr_channels) {
1084 case 1:
1085 result = FMT_8;
1086 goto out_word4;
1087 case 2:
1088 result = FMT_8_8;
1089 goto out_word4;
1090 case 4:
1091 result = FMT_8_8_8_8;
1092 goto out_word4;
1093 }
1094 goto out_unknown;
1095 case 16:
1096 switch (desc->nr_channels) {
1097 case 1:
1098 result = FMT_16;
1099 goto out_word4;
1100 case 2:
1101 result = FMT_16_16;
1102 goto out_word4;
1103 case 4:
1104 result = FMT_16_16_16_16;
1105 goto out_word4;
1106 }
1107 goto out_unknown;
1108 case 32:
1109 switch (desc->nr_channels) {
1110 case 1:
1111 result = FMT_32;
1112 goto out_word4;
1113 case 2:
1114 result = FMT_32_32;
1115 goto out_word4;
1116 case 4:
1117 result = FMT_32_32_32_32;
1118 goto out_word4;
1119 }
1120 }
1121 goto out_unknown;
1122
1123 case UTIL_FORMAT_TYPE_FLOAT:
1124 switch (desc->channel[i].size) {
1125 case 16:
1126 switch (desc->nr_channels) {
1127 case 1:
1128 result = FMT_16_FLOAT;
1129 goto out_word4;
1130 case 2:
1131 result = FMT_16_16_FLOAT;
1132 goto out_word4;
1133 case 4:
1134 result = FMT_16_16_16_16_FLOAT;
1135 goto out_word4;
1136 }
1137 goto out_unknown;
1138 case 32:
1139 switch (desc->nr_channels) {
1140 case 1:
1141 result = FMT_32_FLOAT;
1142 goto out_word4;
1143 case 2:
1144 result = FMT_32_32_FLOAT;
1145 goto out_word4;
1146 case 4:
1147 result = FMT_32_32_32_32_FLOAT;
1148 goto out_word4;
1149 }
1150 }
1151 goto out_unknown;
1152 }
1153
1154 out_word4:
1155 if (word4_p)
1156 *word4_p = word4;
1157 if (yuv_format_p)
1158 *yuv_format_p = yuv_format;
1159 return result;
1160 out_unknown:
1161 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
1162 return ~0;
1163 }