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 "r600_formats.h"
28#include "r600d.h"
29
30#include <errno.h>
31#include "util/u_format_s3tc.h"
32#include "util/u_memory.h"
33
34/* Copy from a full GPU texture to a transfer's staging one. */
35static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
36{
37	struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
38	struct pipe_resource *texture = transfer->resource;
39
40	ctx->resource_copy_region(ctx, &rtransfer->staging->b.b,
41				0, 0, 0, 0, texture, transfer->level,
42				&transfer->box);
43}
44
45
46/* Copy from a transfer's staging texture to a full GPU one. */
47static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
48{
49	struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
50	struct pipe_resource *texture = transfer->resource;
51	struct pipe_box sbox;
52
53	u_box_origin_2d(transfer->box.width, transfer->box.height, &sbox);
54
55	ctx->resource_copy_region(ctx, texture, transfer->level,
56				  transfer->box.x, transfer->box.y, transfer->box.z,
57				  &rtransfer->staging->b.b,
58				  0, &sbox);
59}
60
61unsigned r600_texture_get_offset(struct r600_texture *rtex,
62					unsigned level, unsigned layer)
63{
64	return rtex->surface.level[level].offset +
65	       layer * rtex->surface.level[level].slice_size;
66}
67
68static int r600_init_surface(struct r600_screen *rscreen,
69			     struct radeon_surface *surface,
70			     const struct pipe_resource *ptex,
71			     unsigned array_mode,
72			     bool is_transfer, bool is_flushed_depth)
73{
74	const struct util_format_description *desc =
75		util_format_description(ptex->format);
76	bool is_depth, is_stencil;
77
78	is_depth = util_format_has_depth(desc);
79	is_stencil = util_format_has_stencil(desc);
80
81	surface->npix_x = ptex->width0;
82	surface->npix_y = ptex->height0;
83	surface->npix_z = ptex->depth0;
84	surface->blk_w = util_format_get_blockwidth(ptex->format);
85	surface->blk_h = util_format_get_blockheight(ptex->format);
86	surface->blk_d = 1;
87	surface->array_size = 1;
88	surface->last_level = ptex->last_level;
89
90	if (rscreen->chip_class >= EVERGREEN &&
91	    !is_transfer && !is_flushed_depth &&
92	    ptex->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
93		surface->bpe = 4; /* stencil is allocated separately on evergreen */
94	} else {
95		surface->bpe = util_format_get_blocksize(ptex->format);
96		/* align byte per element on dword */
97		if (surface->bpe == 3) {
98			surface->bpe = 4;
99		}
100	}
101
102	surface->nsamples = ptex->nr_samples ? ptex->nr_samples : 1;
103	surface->flags = 0;
104
105	switch (array_mode) {
106	case V_038000_ARRAY_1D_TILED_THIN1:
107		surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_1D, MODE);
108		break;
109	case V_038000_ARRAY_2D_TILED_THIN1:
110		surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE);
111		break;
112	case V_038000_ARRAY_LINEAR_ALIGNED:
113		surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR_ALIGNED, MODE);
114		break;
115	case V_038000_ARRAY_LINEAR_GENERAL:
116	default:
117		surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR, MODE);
118		break;
119	}
120	switch (ptex->target) {
121	case PIPE_TEXTURE_1D:
122		surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D, TYPE);
123		break;
124	case PIPE_TEXTURE_RECT:
125	case PIPE_TEXTURE_2D:
126		surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D, TYPE);
127		break;
128	case PIPE_TEXTURE_3D:
129		surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_3D, TYPE);
130		break;
131	case PIPE_TEXTURE_1D_ARRAY:
132		surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D_ARRAY, TYPE);
133		surface->array_size = ptex->array_size;
134		break;
135	case PIPE_TEXTURE_2D_ARRAY:
136		surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY, TYPE);
137		surface->array_size = ptex->array_size;
138		break;
139	case PIPE_TEXTURE_CUBE:
140		surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_CUBEMAP, TYPE);
141		break;
142	case PIPE_BUFFER:
143	default:
144		return -EINVAL;
145	}
146	if (ptex->bind & PIPE_BIND_SCANOUT) {
147		surface->flags |= RADEON_SURF_SCANOUT;
148	}
149
150	if (!is_transfer && !is_flushed_depth && is_depth) {
151		surface->flags |= RADEON_SURF_ZBUFFER;
152
153		if (is_stencil) {
154			surface->flags |= RADEON_SURF_SBUFFER;
155		}
156	}
157	return 0;
158}
159
160static int r600_setup_surface(struct pipe_screen *screen,
161			      struct r600_texture *rtex,
162			      unsigned pitch_in_bytes_override)
163{
164	struct pipe_resource *ptex = &rtex->resource.b.b;
165	struct r600_screen *rscreen = (struct r600_screen*)screen;
166	unsigned i;
167	int r;
168
169	r = rscreen->ws->surface_init(rscreen->ws, &rtex->surface);
170	if (r) {
171		return r;
172	}
173	rtex->size = rtex->surface.bo_size;
174	if (pitch_in_bytes_override && pitch_in_bytes_override != rtex->surface.level[0].pitch_bytes) {
175		/* old ddx on evergreen over estimate alignment for 1d, only 1 level
176		 * for those
177		 */
178		rtex->surface.level[0].nblk_x = pitch_in_bytes_override / rtex->surface.bpe;
179		rtex->surface.level[0].pitch_bytes = pitch_in_bytes_override;
180		rtex->surface.level[0].slice_size = pitch_in_bytes_override * rtex->surface.level[0].nblk_y;
181		if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
182			rtex->surface.stencil_offset = rtex->surface.level[0].slice_size;
183		}
184	}
185	for (i = 0; i <= ptex->last_level; i++) {
186		switch (rtex->surface.level[i].mode) {
187		case RADEON_SURF_MODE_LINEAR_ALIGNED:
188			rtex->array_mode[i] = V_038000_ARRAY_LINEAR_ALIGNED;
189			break;
190		case RADEON_SURF_MODE_1D:
191			rtex->array_mode[i] = V_038000_ARRAY_1D_TILED_THIN1;
192			break;
193		case RADEON_SURF_MODE_2D:
194			rtex->array_mode[i] = V_038000_ARRAY_2D_TILED_THIN1;
195			break;
196		default:
197		case RADEON_SURF_MODE_LINEAR:
198			rtex->array_mode[i] = 0;
199			break;
200		}
201	}
202	return 0;
203}
204
205static boolean r600_texture_get_handle(struct pipe_screen* screen,
206					struct pipe_resource *ptex,
207					struct winsys_handle *whandle)
208{
209	struct r600_texture *rtex = (struct r600_texture*)ptex;
210	struct r600_resource *resource = &rtex->resource;
211	struct radeon_surface *surface = &rtex->surface;
212	struct r600_screen *rscreen = (struct r600_screen*)screen;
213
214	rscreen->ws->buffer_set_tiling(resource->buf,
215				       NULL,
216				       surface->level[0].mode >= RADEON_SURF_MODE_1D ?
217				       RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR,
218				       surface->level[0].mode >= RADEON_SURF_MODE_2D ?
219				       RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR,
220				       surface->bankw, surface->bankh,
221				       surface->tile_split,
222				       surface->stencil_tile_split,
223				       surface->mtilea,
224				       rtex->surface.level[0].pitch_bytes);
225
226	return rscreen->ws->buffer_get_handle(resource->buf,
227					      rtex->surface.level[0].pitch_bytes, whandle);
228}
229
230static void r600_texture_destroy(struct pipe_screen *screen,
231				 struct pipe_resource *ptex)
232{
233	struct r600_texture *rtex = (struct r600_texture*)ptex;
234	struct r600_resource *resource = &rtex->resource;
235
236	if (rtex->flushed_depth_texture)
237		pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
238
239	pb_reference(&resource->buf, NULL);
240	FREE(rtex);
241}
242
243static const struct u_resource_vtbl r600_texture_vtbl =
244{
245	r600_texture_get_handle,	/* get_handle */
246	r600_texture_destroy,		/* resource_destroy */
247	r600_texture_get_transfer,	/* get_transfer */
248	r600_texture_transfer_destroy,	/* transfer_destroy */
249	r600_texture_transfer_map,	/* transfer_map */
250	NULL,				/* transfer_flush_region */
251	r600_texture_transfer_unmap,	/* transfer_unmap */
252	NULL				/* transfer_inline_write */
253};
254
255/* The number of samples can be specified independently of the texture. */
256void r600_texture_get_fmask_info(struct r600_screen *rscreen,
257				 struct r600_texture *rtex,
258				 unsigned nr_samples,
259				 struct r600_fmask_info *out)
260{
261	/* FMASK is allocated pretty much like an ordinary texture.
262	 * Here we use bpe in the units of bits, not bytes. */
263	struct radeon_surface fmask = rtex->surface;
264
265	switch (nr_samples) {
266	case 2:
267		/* This should be 8,1, but we should set nsamples > 1
268		 * for the allocator to treat it as a multisample surface.
269		 * Let's set 4,2 then. */
270	case 4:
271		fmask.bpe = 4;
272		fmask.nsamples = 2;
273		break;
274	case 8:
275		fmask.bpe = 8;
276		fmask.nsamples = 4;
277		break;
278	case 16:
279		fmask.bpe = 16;
280		fmask.nsamples = 4;
281		break;
282	default:
283		R600_ERR("Invalid sample count for FMASK allocation.\n");
284		return;
285	}
286
287	/* R600-R700 errata? Anyway, this fixes colorbuffer corruption. */
288	if (rscreen->chip_class <= R700) {
289		fmask.bpe *= 2;
290	}
291
292	if (rscreen->chip_class >= EVERGREEN) {
293		fmask.bankh = nr_samples <= 4 ? 4 : 1;
294	}
295
296	if (rscreen->ws->surface_init(rscreen->ws, &fmask)) {
297		R600_ERR("Got error in surface_init while allocating FMASK.\n");
298		return;
299	}
300	assert(fmask.level[0].mode == RADEON_SURF_MODE_2D);
301
302	out->bank_height = fmask.bankh;
303	out->alignment = MAX2(256, fmask.bo_alignment);
304	out->size = (fmask.bo_size + 7) / 8;
305}
306
307static void r600_texture_allocate_fmask(struct r600_screen *rscreen,
308					struct r600_texture *rtex)
309{
310	struct r600_fmask_info fmask;
311
312	r600_texture_get_fmask_info(rscreen, rtex,
313				    rtex->resource.b.b.nr_samples, &fmask);
314
315	/* Reserve space for FMASK while converting bits back to bytes. */
316	rtex->fmask_bank_height = fmask.bank_height;
317	rtex->fmask_offset = align(rtex->size, fmask.alignment);
318	rtex->fmask_size = fmask.size;
319	rtex->size = rtex->fmask_offset + rtex->fmask_size;
320#if 0
321	printf("FMASK width=%u, height=%i, bits=%u, size=%u\n",
322	       fmask.npix_x, fmask.npix_y, fmask.bpe * fmask.nsamples, rtex->fmask_size);
323#endif
324}
325
326void r600_texture_get_cmask_info(struct r600_screen *rscreen,
327				 struct r600_texture *rtex,
328				 struct r600_cmask_info *out)
329{
330	unsigned cmask_tile_width = 8;
331	unsigned cmask_tile_height = 8;
332	unsigned cmask_tile_elements = cmask_tile_width * cmask_tile_height;
333	unsigned element_bits = 4;
334	unsigned cmask_cache_bits = 1024;
335	unsigned num_pipes = rscreen->tiling_info.num_channels;
336	unsigned pipe_interleave_bytes = rscreen->tiling_info.group_bytes;
337
338	unsigned elements_per_macro_tile = (cmask_cache_bits / element_bits) * num_pipes;
339	unsigned pixels_per_macro_tile = elements_per_macro_tile * cmask_tile_elements;
340	unsigned sqrt_pixels_per_macro_tile = sqrt(pixels_per_macro_tile);
341	unsigned macro_tile_width = util_next_power_of_two(sqrt_pixels_per_macro_tile);
342	unsigned macro_tile_height = pixels_per_macro_tile / macro_tile_width;
343
344	unsigned pitch_elements = align(rtex->surface.npix_x, macro_tile_width);
345	unsigned height = align(rtex->surface.npix_y, macro_tile_height);
346
347	unsigned base_align = num_pipes * pipe_interleave_bytes;
348	unsigned slice_bytes =
349		((pitch_elements * height * element_bits + 7) / 8) / cmask_tile_elements;
350
351	assert(macro_tile_width % 128 == 0);
352	assert(macro_tile_height % 128 == 0);
353
354	out->slice_tile_max = ((pitch_elements * height) / (128*128)) - 1;
355	out->alignment = MAX2(256, base_align);
356	out->size = rtex->surface.array_size * align(slice_bytes, base_align);
357}
358
359static void r600_texture_allocate_cmask(struct r600_screen *rscreen,
360					struct r600_texture *rtex)
361{
362	struct r600_cmask_info cmask;
363
364	r600_texture_get_cmask_info(rscreen, rtex, &cmask);
365
366	rtex->cmask_slice_tile_max = cmask.slice_tile_max;
367	rtex->cmask_offset = align(rtex->size, cmask.alignment);
368	rtex->cmask_size = cmask.size;
369	rtex->size = rtex->cmask_offset + rtex->cmask_size;
370#if 0
371	printf("CMASK: macro tile width = %u, macro tile height = %u, "
372	       "pitch elements = %u, height = %u, slice tile max = %u\n",
373	       macro_tile_width, macro_tile_height, pitch_elements, height,
374	       rtex->cmask_slice_tile_max);
375#endif
376}
377
378static struct r600_texture *
379r600_texture_create_object(struct pipe_screen *screen,
380			   const struct pipe_resource *base,
381			   unsigned pitch_in_bytes_override,
382			   struct pb_buffer *buf,
383			   boolean alloc_bo,
384			   struct radeon_surface *surface)
385{
386	struct r600_texture *rtex;
387	struct r600_resource *resource;
388	struct r600_screen *rscreen = (struct r600_screen*)screen;
389	int r;
390
391	rtex = CALLOC_STRUCT(r600_texture);
392	if (rtex == NULL)
393		return NULL;
394
395	resource = &rtex->resource;
396	resource->b.b = *base;
397	resource->b.vtbl = &r600_texture_vtbl;
398	pipe_reference_init(&resource->b.b.reference, 1);
399	resource->b.b.screen = screen;
400	rtex->pitch_override = pitch_in_bytes_override;
401
402	/* don't include stencil-only formats which we don't support for rendering */
403	rtex->is_depth = util_format_has_depth(util_format_description(rtex->resource.b.b.format));
404
405	rtex->surface = *surface;
406	r = r600_setup_surface(screen, rtex,
407			       pitch_in_bytes_override);
408	if (r) {
409		FREE(rtex);
410		return NULL;
411	}
412
413	if (base->nr_samples > 1 && !rtex->is_depth && alloc_bo) {
414		r600_texture_allocate_cmask(rscreen, rtex);
415		r600_texture_allocate_fmask(rscreen, rtex);
416	}
417
418	if (!rtex->is_depth && base->nr_samples > 1 &&
419	    (!rtex->fmask_size || !rtex->cmask_size)) {
420		FREE(rtex);
421		return NULL;
422	}
423
424	/* Now create the backing buffer. */
425	if (!buf && alloc_bo) {
426		unsigned base_align = rtex->surface.bo_alignment;
427		unsigned usage = R600_TEX_IS_TILED(rtex, 0) ? PIPE_USAGE_STATIC : base->usage;
428
429		if (!r600_init_resource(rscreen, resource, rtex->size, base_align, base->bind, usage)) {
430			FREE(rtex);
431			return NULL;
432		}
433	} else if (buf) {
434		resource->buf = buf;
435		resource->cs_buf = rscreen->ws->buffer_get_cs_handle(buf);
436		resource->domains = RADEON_DOMAIN_GTT | RADEON_DOMAIN_VRAM;
437	}
438
439	if (rtex->cmask_size) {
440		/* Initialize the cmask to 0xCC (= compressed state). */
441		char *ptr = rscreen->ws->buffer_map(resource->cs_buf, NULL, PIPE_TRANSFER_WRITE);
442		memset(ptr + rtex->cmask_offset, 0xCC, rtex->cmask_size);
443		rscreen->ws->buffer_unmap(resource->cs_buf);
444	}
445	return rtex;
446}
447
448struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
449						const struct pipe_resource *templ)
450{
451	struct r600_screen *rscreen = (struct r600_screen*)screen;
452	struct radeon_surface surface;
453	unsigned array_mode = 0;
454	int r;
455
456	if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER)) {
457		if (templ->flags & R600_RESOURCE_FLAG_FORCE_TILING) {
458			array_mode = V_038000_ARRAY_2D_TILED_THIN1;
459		} else if (!(templ->bind & PIPE_BIND_SCANOUT) &&
460		    templ->usage != PIPE_USAGE_STAGING &&
461		    templ->usage != PIPE_USAGE_STREAM) {
462			array_mode = V_038000_ARRAY_2D_TILED_THIN1;
463		} else if (util_format_is_compressed(templ->format)) {
464			array_mode = V_038000_ARRAY_1D_TILED_THIN1;
465		}
466	}
467
468	/* XXX tiling is broken for the 422 formats */
469	if (util_format_description(templ->format)->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED)
470		array_mode = V_038000_ARRAY_LINEAR_ALIGNED;
471
472	r = r600_init_surface(rscreen, &surface, templ, array_mode,
473			      templ->flags & R600_RESOURCE_FLAG_TRANSFER,
474			      templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH);
475	if (r) {
476		return NULL;
477	}
478	r = rscreen->ws->surface_best(rscreen->ws, &surface);
479	if (r) {
480		return NULL;
481	}
482	return (struct pipe_resource *)r600_texture_create_object(screen, templ,
483								  0, NULL, TRUE, &surface);
484}
485
486static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
487						struct pipe_resource *texture,
488						const struct pipe_surface *templ)
489{
490	struct r600_texture *rtex = (struct r600_texture*)texture;
491	struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
492	unsigned level = templ->u.tex.level;
493
494	assert(templ->u.tex.first_layer == templ->u.tex.last_layer);
495	if (surface == NULL)
496		return NULL;
497	pipe_reference_init(&surface->base.reference, 1);
498	pipe_resource_reference(&surface->base.texture, texture);
499	surface->base.context = pipe;
500	surface->base.format = templ->format;
501	surface->base.width = rtex->surface.level[level].npix_x;
502	surface->base.height = rtex->surface.level[level].npix_y;
503	surface->base.usage = templ->usage;
504	surface->base.u = templ->u;
505	return &surface->base;
506}
507
508static void r600_surface_destroy(struct pipe_context *pipe,
509				 struct pipe_surface *surface)
510{
511	struct r600_surface *surf = (struct r600_surface*)surface;
512	pipe_resource_reference((struct pipe_resource**)&surf->cb_buffer_fmask, NULL);
513	pipe_resource_reference((struct pipe_resource**)&surf->cb_buffer_cmask, NULL);
514	pipe_resource_reference(&surface->texture, NULL);
515	FREE(surface);
516}
517
518struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
519					       const struct pipe_resource *templ,
520					       struct winsys_handle *whandle)
521{
522	struct r600_screen *rscreen = (struct r600_screen*)screen;
523	struct pb_buffer *buf = NULL;
524	unsigned stride = 0;
525	unsigned array_mode = 0;
526	enum radeon_bo_layout micro, macro;
527	struct radeon_surface surface;
528	int r;
529
530	/* Support only 2D textures without mipmaps */
531	if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) ||
532	      templ->depth0 != 1 || templ->last_level != 0)
533		return NULL;
534
535	buf = rscreen->ws->buffer_from_handle(rscreen->ws, whandle, &stride);
536	if (!buf)
537		return NULL;
538
539	rscreen->ws->buffer_get_tiling(buf, &micro, &macro,
540				       &surface.bankw, &surface.bankh,
541				       &surface.tile_split,
542				       &surface.stencil_tile_split,
543				       &surface.mtilea);
544
545	if (macro == RADEON_LAYOUT_TILED)
546		array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
547	else if (micro == RADEON_LAYOUT_TILED)
548		array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
549	else
550		array_mode = 0;
551
552	r = r600_init_surface(rscreen, &surface, templ, array_mode, false, false);
553	if (r) {
554		return NULL;
555	}
556	return (struct pipe_resource *)r600_texture_create_object(screen, templ,
557								  stride, buf, FALSE, &surface);
558}
559
560bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
561				     struct pipe_resource *texture,
562				     struct r600_texture **staging)
563{
564	struct r600_texture *rtex = (struct r600_texture*)texture;
565	struct pipe_resource resource;
566	struct r600_texture **flushed_depth_texture = staging ?
567			staging : &rtex->flushed_depth_texture;
568
569	if (!staging && rtex->flushed_depth_texture)
570		return true; /* it's ready */
571
572	resource.target = texture->target;
573	resource.format = texture->format;
574	resource.width0 = texture->width0;
575	resource.height0 = texture->height0;
576	resource.depth0 = texture->depth0;
577	resource.array_size = texture->array_size;
578	resource.last_level = texture->last_level;
579	resource.nr_samples = texture->nr_samples;
580	resource.usage = staging ? PIPE_USAGE_STAGING : PIPE_USAGE_STATIC;
581	resource.bind = texture->bind & ~PIPE_BIND_DEPTH_STENCIL;
582	resource.flags = texture->flags | R600_RESOURCE_FLAG_FLUSHED_DEPTH;
583
584	if (staging)
585		resource.flags |= R600_RESOURCE_FLAG_TRANSFER;
586
587	*flushed_depth_texture = (struct r600_texture *)ctx->screen->resource_create(ctx->screen, &resource);
588	if (*flushed_depth_texture == NULL) {
589		R600_ERR("failed to create temporary texture to hold flushed depth\n");
590		return false;
591	}
592
593	(*flushed_depth_texture)->is_flushing_texture = TRUE;
594	return true;
595}
596
597struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
598						struct pipe_resource *texture,
599						unsigned level,
600						unsigned usage,
601						const struct pipe_box *box)
602{
603	struct r600_context *rctx = (struct r600_context*)ctx;
604	struct r600_texture *rtex = (struct r600_texture*)texture;
605	struct pipe_resource resource;
606	struct r600_transfer *trans;
607	boolean use_staging_texture = FALSE;
608
609	/* We cannot map a tiled texture directly because the data is
610	 * in a different order, therefore we do detiling using a blit.
611	 *
612	 * Also, use a temporary in GTT memory for read transfers, as
613	 * the CPU is much happier reading out of cached system memory
614	 * than uncached VRAM.
615	 */
616	if (R600_TEX_IS_TILED(rtex, level)) {
617		use_staging_texture = TRUE;
618	}
619
620	/* Use a staging texture for uploads if the underlying BO is busy. */
621	if (!(usage & PIPE_TRANSFER_READ) &&
622	    (rctx->ws->cs_is_buffer_referenced(rctx->cs, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) ||
623	     rctx->ws->buffer_is_busy(rtex->resource.buf, RADEON_USAGE_READWRITE))) {
624		use_staging_texture = TRUE;
625	}
626
627	if (texture->flags & R600_RESOURCE_FLAG_TRANSFER) {
628		use_staging_texture = FALSE;
629	}
630
631	if (use_staging_texture && (usage & PIPE_TRANSFER_MAP_DIRECTLY)) {
632		return NULL;
633	}
634
635	trans = CALLOC_STRUCT(r600_transfer);
636	if (trans == NULL)
637		return NULL;
638	pipe_resource_reference(&trans->transfer.resource, texture);
639	trans->transfer.level = level;
640	trans->transfer.usage = usage;
641	trans->transfer.box = *box;
642	if (rtex->is_depth) {
643		/* XXX: only readback the rectangle which is being mapped?
644		*/
645		/* XXX: when discard is true, no need to read back from depth texture
646		*/
647		struct r600_texture *staging_depth;
648
649		if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
650			R600_ERR("failed to create temporary texture to hold untiled copy\n");
651			pipe_resource_reference(&trans->transfer.resource, NULL);
652			FREE(trans);
653			return NULL;
654		}
655
656		r600_blit_decompress_depth(ctx, rtex, staging_depth,
657					   level, level,
658					   box->z, box->z + box->depth - 1,
659					   0, 0);
660
661		trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes;
662		trans->offset = r600_texture_get_offset(staging_depth, level, box->z);
663		trans->staging = (struct r600_resource*)staging_depth;
664		return &trans->transfer;
665	} else if (use_staging_texture) {
666		resource.target = PIPE_TEXTURE_2D;
667		resource.format = texture->format;
668		resource.width0 = box->width;
669		resource.height0 = box->height;
670		resource.depth0 = 1;
671		resource.array_size = 1;
672		resource.last_level = 0;
673		resource.nr_samples = 0;
674		resource.usage = PIPE_USAGE_STAGING;
675		resource.bind = 0;
676		resource.flags = R600_RESOURCE_FLAG_TRANSFER;
677		/* For texture reading, the temporary (detiled) texture is used as
678		 * a render target when blitting from a tiled texture. */
679		if (usage & PIPE_TRANSFER_READ) {
680			resource.bind |= PIPE_BIND_RENDER_TARGET;
681		}
682		/* For texture writing, the temporary texture is used as a sampler
683		 * when blitting into a tiled texture. */
684		if (usage & PIPE_TRANSFER_WRITE) {
685			resource.bind |= PIPE_BIND_SAMPLER_VIEW;
686		}
687		/* Create the temporary texture. */
688		trans->staging = (struct r600_resource*)ctx->screen->resource_create(ctx->screen, &resource);
689		if (trans->staging == NULL) {
690			R600_ERR("failed to create temporary texture to hold untiled copy\n");
691			pipe_resource_reference(&trans->transfer.resource, NULL);
692			FREE(trans);
693			return NULL;
694		}
695
696		trans->transfer.stride =
697			((struct r600_texture *)trans->staging)->surface.level[0].pitch_bytes;
698		if (usage & PIPE_TRANSFER_READ) {
699			r600_copy_to_staging_texture(ctx, trans);
700			/* Always referenced in the blit. */
701			r600_flush(ctx, NULL, 0);
702		}
703		return &trans->transfer;
704	}
705	trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
706	trans->transfer.layer_stride = rtex->surface.level[level].slice_size;
707	trans->offset = r600_texture_get_offset(rtex, level, box->z);
708	return &trans->transfer;
709}
710
711void r600_texture_transfer_destroy(struct pipe_context *ctx,
712				   struct pipe_transfer *transfer)
713{
714	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
715	struct pipe_resource *texture = transfer->resource;
716	struct r600_texture *rtex = (struct r600_texture*)texture;
717
718	if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtransfer->staging) {
719		if (rtex->is_depth) {
720			ctx->resource_copy_region(ctx, texture, transfer->level,
721						  transfer->box.x, transfer->box.y, transfer->box.z,
722						  &rtransfer->staging->b.b, transfer->level,
723						  &transfer->box);
724		} else {
725			r600_copy_from_staging_texture(ctx, rtransfer);
726		}
727	}
728
729	if (rtransfer->staging)
730		pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL);
731
732	pipe_resource_reference(&transfer->resource, NULL);
733	FREE(transfer);
734}
735
736void* r600_texture_transfer_map(struct pipe_context *ctx,
737				struct pipe_transfer* transfer)
738{
739	struct r600_context *rctx = (struct r600_context *)ctx;
740	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
741	struct radeon_winsys_cs_handle *buf;
742	struct r600_texture *rtex =
743			(struct r600_texture*)transfer->resource;
744	enum pipe_format format = transfer->resource->format;
745	unsigned offset = 0;
746	char *map;
747
748	if ((transfer->resource->bind & PIPE_BIND_GLOBAL) && transfer->resource->target == PIPE_BUFFER) {
749		return r600_compute_global_transfer_map(ctx, transfer);
750	}
751
752	if (rtransfer->staging) {
753		buf = ((struct r600_resource *)rtransfer->staging)->cs_buf;
754	} else {
755		buf = ((struct r600_resource *)transfer->resource)->cs_buf;
756	}
757
758	if (rtex->is_depth || !rtransfer->staging)
759		offset = rtransfer->offset +
760			transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
761			transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
762
763	if (!(map = rctx->ws->buffer_map(buf, rctx->cs, transfer->usage))) {
764		return NULL;
765	}
766
767	return map + offset;
768}
769
770void r600_texture_transfer_unmap(struct pipe_context *ctx,
771				 struct pipe_transfer* transfer)
772{
773	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
774	struct r600_context *rctx = (struct r600_context*)ctx;
775	struct radeon_winsys_cs_handle *buf;
776
777	if ((transfer->resource->bind & PIPE_BIND_GLOBAL) && transfer->resource->target == PIPE_BUFFER) {
778		return r600_compute_global_transfer_unmap(ctx, transfer);
779	}
780
781	if (rtransfer->staging) {
782		buf = ((struct r600_resource *)rtransfer->staging)->cs_buf;
783	} else {
784		buf = ((struct r600_resource *)transfer->resource)->cs_buf;
785	}
786	rctx->ws->buffer_unmap(buf);
787}
788
789void r600_init_surface_functions(struct r600_context *r600)
790{
791	r600->context.create_surface = r600_create_surface;
792	r600->context.surface_destroy = r600_surface_destroy;
793}
794
795static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
796		const unsigned char *swizzle_view)
797{
798	unsigned i;
799	unsigned char swizzle[4];
800	unsigned result = 0;
801	const uint32_t swizzle_shift[4] = {
802		16, 19, 22, 25,
803	};
804	const uint32_t swizzle_bit[4] = {
805		0, 1, 2, 3,
806	};
807
808	if (swizzle_view) {
809		util_format_compose_swizzles(swizzle_format, swizzle_view, swizzle);
810	} else {
811		memcpy(swizzle, swizzle_format, 4);
812	}
813
814	/* Get swizzle. */
815	for (i = 0; i < 4; i++) {
816		switch (swizzle[i]) {
817		case UTIL_FORMAT_SWIZZLE_Y:
818			result |= swizzle_bit[1] << swizzle_shift[i];
819			break;
820		case UTIL_FORMAT_SWIZZLE_Z:
821			result |= swizzle_bit[2] << swizzle_shift[i];
822			break;
823		case UTIL_FORMAT_SWIZZLE_W:
824			result |= swizzle_bit[3] << swizzle_shift[i];
825			break;
826		case UTIL_FORMAT_SWIZZLE_0:
827			result |= V_038010_SQ_SEL_0 << swizzle_shift[i];
828			break;
829		case UTIL_FORMAT_SWIZZLE_1:
830			result |= V_038010_SQ_SEL_1 << swizzle_shift[i];
831			break;
832		default: /* UTIL_FORMAT_SWIZZLE_X */
833			result |= swizzle_bit[0] << swizzle_shift[i];
834		}
835	}
836	return result;
837}
838
839/* texture format translate */
840uint32_t r600_translate_texformat(struct pipe_screen *screen,
841				  enum pipe_format format,
842				  const unsigned char *swizzle_view,
843				  uint32_t *word4_p, uint32_t *yuv_format_p)
844{
845	uint32_t result = 0, word4 = 0, yuv_format = 0;
846	const struct util_format_description *desc;
847	boolean uniform = TRUE;
848	static int r600_enable_s3tc = -1;
849	bool is_srgb_valid = FALSE;
850
851	int i;
852	const uint32_t sign_bit[4] = {
853		S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED),
854		S_038010_FORMAT_COMP_Y(V_038010_SQ_FORMAT_COMP_SIGNED),
855		S_038010_FORMAT_COMP_Z(V_038010_SQ_FORMAT_COMP_SIGNED),
856		S_038010_FORMAT_COMP_W(V_038010_SQ_FORMAT_COMP_SIGNED)
857	};
858	desc = util_format_description(format);
859
860	word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view);
861
862	/* Colorspace (return non-RGB formats directly). */
863	switch (desc->colorspace) {
864		/* Depth stencil formats */
865	case UTIL_FORMAT_COLORSPACE_ZS:
866		switch (format) {
867		case PIPE_FORMAT_Z16_UNORM:
868			result = FMT_16;
869			goto out_word4;
870		case PIPE_FORMAT_X24S8_UINT:
871			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
872		case PIPE_FORMAT_Z24X8_UNORM:
873		case PIPE_FORMAT_Z24_UNORM_S8_UINT:
874			result = FMT_8_24;
875			goto out_word4;
876		case PIPE_FORMAT_S8X24_UINT:
877			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
878		case PIPE_FORMAT_X8Z24_UNORM:
879		case PIPE_FORMAT_S8_UINT_Z24_UNORM:
880			result = FMT_24_8;
881			goto out_word4;
882		case PIPE_FORMAT_S8_UINT:
883			result = FMT_8;
884			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
885			goto out_word4;
886		case PIPE_FORMAT_Z32_FLOAT:
887			result = FMT_32_FLOAT;
888			goto out_word4;
889		case PIPE_FORMAT_X32_S8X24_UINT:
890			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
891		case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
892			result = FMT_X24_8_32_FLOAT;
893			goto out_word4;
894		default:
895			goto out_unknown;
896		}
897
898	case UTIL_FORMAT_COLORSPACE_YUV:
899		yuv_format |= (1 << 30);
900		switch (format) {
901		case PIPE_FORMAT_UYVY:
902		case PIPE_FORMAT_YUYV:
903		default:
904			break;
905		}
906		goto out_unknown; /* XXX */
907
908	case UTIL_FORMAT_COLORSPACE_SRGB:
909		word4 |= S_038010_FORCE_DEGAMMA(1);
910		break;
911
912	default:
913		break;
914	}
915
916	if (r600_enable_s3tc == -1) {
917		struct r600_screen *rscreen = (struct r600_screen *)screen;
918		if (rscreen->info.drm_minor >= 9)
919			r600_enable_s3tc = 1;
920		else
921			r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
922	}
923
924	if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
925		if (!r600_enable_s3tc)
926			goto out_unknown;
927
928		switch (format) {
929		case PIPE_FORMAT_RGTC1_SNORM:
930		case PIPE_FORMAT_LATC1_SNORM:
931			word4 |= sign_bit[0];
932		case PIPE_FORMAT_RGTC1_UNORM:
933		case PIPE_FORMAT_LATC1_UNORM:
934			result = FMT_BC4;
935			goto out_word4;
936		case PIPE_FORMAT_RGTC2_SNORM:
937		case PIPE_FORMAT_LATC2_SNORM:
938			word4 |= sign_bit[0] | sign_bit[1];
939		case PIPE_FORMAT_RGTC2_UNORM:
940		case PIPE_FORMAT_LATC2_UNORM:
941			result = FMT_BC5;
942			goto out_word4;
943		default:
944			goto out_unknown;
945		}
946	}
947
948	if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
949
950		if (!r600_enable_s3tc)
951			goto out_unknown;
952
953		if (!util_format_s3tc_enabled) {
954			goto out_unknown;
955		}
956
957		switch (format) {
958		case PIPE_FORMAT_DXT1_RGB:
959		case PIPE_FORMAT_DXT1_RGBA:
960		case PIPE_FORMAT_DXT1_SRGB:
961		case PIPE_FORMAT_DXT1_SRGBA:
962			result = FMT_BC1;
963			is_srgb_valid = TRUE;
964			goto out_word4;
965		case PIPE_FORMAT_DXT3_RGBA:
966		case PIPE_FORMAT_DXT3_SRGBA:
967			result = FMT_BC2;
968			is_srgb_valid = TRUE;
969			goto out_word4;
970		case PIPE_FORMAT_DXT5_RGBA:
971		case PIPE_FORMAT_DXT5_SRGBA:
972			result = FMT_BC3;
973			is_srgb_valid = TRUE;
974			goto out_word4;
975		default:
976			goto out_unknown;
977		}
978	}
979
980	if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) {
981		switch (format) {
982		case PIPE_FORMAT_R8G8_B8G8_UNORM:
983		case PIPE_FORMAT_G8R8_B8R8_UNORM:
984			result = FMT_GB_GR;
985			goto out_word4;
986		case PIPE_FORMAT_G8R8_G8B8_UNORM:
987		case PIPE_FORMAT_R8G8_R8B8_UNORM:
988			result = FMT_BG_RG;
989			goto out_word4;
990		default:
991			goto out_unknown;
992		}
993	}
994
995	if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) {
996		result = FMT_5_9_9_9_SHAREDEXP;
997		goto out_word4;
998	} else if (format == PIPE_FORMAT_R11G11B10_FLOAT) {
999		result = FMT_10_11_11_FLOAT;
1000		goto out_word4;
1001	}
1002
1003
1004	for (i = 0; i < desc->nr_channels; i++) {
1005		if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
1006			word4 |= sign_bit[i];
1007		}
1008	}
1009
1010	/* R8G8Bx_SNORM - XXX CxV8U8 */
1011
1012	/* See whether the components are of the same size. */
1013	for (i = 1; i < desc->nr_channels; i++) {
1014		uniform = uniform && desc->channel[0].size == desc->channel[i].size;
1015	}
1016
1017	/* Non-uniform formats. */
1018	if (!uniform) {
1019		if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
1020		    desc->channel[0].pure_integer)
1021			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
1022		switch(desc->nr_channels) {
1023		case 3:
1024			if (desc->channel[0].size == 5 &&
1025			    desc->channel[1].size == 6 &&
1026			    desc->channel[2].size == 5) {
1027				result = FMT_5_6_5;
1028				goto out_word4;
1029			}
1030			goto out_unknown;
1031		case 4:
1032			if (desc->channel[0].size == 5 &&
1033			    desc->channel[1].size == 5 &&
1034			    desc->channel[2].size == 5 &&
1035			    desc->channel[3].size == 1) {
1036				result = FMT_1_5_5_5;
1037				goto out_word4;
1038			}
1039			if (desc->channel[0].size == 10 &&
1040			    desc->channel[1].size == 10 &&
1041			    desc->channel[2].size == 10 &&
1042			    desc->channel[3].size == 2) {
1043				result = FMT_2_10_10_10;
1044				goto out_word4;
1045			}
1046			goto out_unknown;
1047		}
1048		goto out_unknown;
1049	}
1050
1051	/* Find the first non-VOID channel. */
1052	for (i = 0; i < 4; i++) {
1053		if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
1054			break;
1055		}
1056	}
1057
1058	if (i == 4)
1059		goto out_unknown;
1060
1061	/* uniform formats */
1062	switch (desc->channel[i].type) {
1063	case UTIL_FORMAT_TYPE_UNSIGNED:
1064	case UTIL_FORMAT_TYPE_SIGNED:
1065#if 0
1066		if (!desc->channel[i].normalized &&
1067		    desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
1068			goto out_unknown;
1069		}
1070#endif
1071		if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
1072		    desc->channel[i].pure_integer)
1073			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
1074
1075		switch (desc->channel[i].size) {
1076		case 4:
1077			switch (desc->nr_channels) {
1078			case 2:
1079				result = FMT_4_4;
1080				goto out_word4;
1081			case 4:
1082				result = FMT_4_4_4_4;
1083				goto out_word4;
1084			}
1085			goto out_unknown;
1086		case 8:
1087			switch (desc->nr_channels) {
1088			case 1:
1089				result = FMT_8;
1090				goto out_word4;
1091			case 2:
1092				result = FMT_8_8;
1093				goto out_word4;
1094			case 4:
1095				result = FMT_8_8_8_8;
1096				is_srgb_valid = TRUE;
1097				goto out_word4;
1098			}
1099			goto out_unknown;
1100		case 16:
1101			switch (desc->nr_channels) {
1102			case 1:
1103				result = FMT_16;
1104				goto out_word4;
1105			case 2:
1106				result = FMT_16_16;
1107				goto out_word4;
1108			case 4:
1109				result = FMT_16_16_16_16;
1110				goto out_word4;
1111			}
1112			goto out_unknown;
1113		case 32:
1114			switch (desc->nr_channels) {
1115			case 1:
1116				result = FMT_32;
1117				goto out_word4;
1118			case 2:
1119				result = FMT_32_32;
1120				goto out_word4;
1121			case 4:
1122				result = FMT_32_32_32_32;
1123				goto out_word4;
1124			}
1125		}
1126		goto out_unknown;
1127
1128	case UTIL_FORMAT_TYPE_FLOAT:
1129		switch (desc->channel[i].size) {
1130		case 16:
1131			switch (desc->nr_channels) {
1132			case 1:
1133				result = FMT_16_FLOAT;
1134				goto out_word4;
1135			case 2:
1136				result = FMT_16_16_FLOAT;
1137				goto out_word4;
1138			case 4:
1139				result = FMT_16_16_16_16_FLOAT;
1140				goto out_word4;
1141			}
1142			goto out_unknown;
1143		case 32:
1144			switch (desc->nr_channels) {
1145			case 1:
1146				result = FMT_32_FLOAT;
1147				goto out_word4;
1148			case 2:
1149				result = FMT_32_32_FLOAT;
1150				goto out_word4;
1151			case 4:
1152				result = FMT_32_32_32_32_FLOAT;
1153				goto out_word4;
1154			}
1155		}
1156		goto out_unknown;
1157	}
1158
1159out_word4:
1160
1161	if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB && !is_srgb_valid)
1162		return ~0;
1163	if (word4_p)
1164		*word4_p = word4;
1165	if (yuv_format_p)
1166		*yuv_format_p = yuv_format;
1167	return result;
1168out_unknown:
1169	/* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
1170	return ~0;
1171}
1172