1/*
2 * Copyright 2013 Advanced Micro Devices, Inc.
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 *      Marek Olšák
25 */
26
27#include "r600_cs.h"
28#include "evergreen_compute.h"
29#include "compute_memory_pool.h"
30#include "util/u_memory.h"
31#include "util/u_upload_mgr.h"
32#include <inttypes.h>
33#include <stdio.h>
34
35bool r600_rings_is_buffer_referenced(struct r600_common_context *ctx,
36				     struct pb_buffer *buf,
37				     unsigned usage)
38{
39	if (ctx->ws->cs_is_buffer_referenced(&ctx->gfx.cs, buf, usage)) {
40		return true;
41	}
42	if (radeon_emitted(&ctx->dma.cs, 0) &&
43	    ctx->ws->cs_is_buffer_referenced(&ctx->dma.cs, buf, usage)) {
44		return true;
45	}
46	return false;
47}
48
49void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx,
50                                      struct r600_resource *resource,
51                                      unsigned usage)
52{
53	unsigned rusage = RADEON_USAGE_READWRITE;
54	bool busy = false;
55
56	assert(!(resource->flags & RADEON_FLAG_SPARSE));
57
58	if (usage & PIPE_MAP_UNSYNCHRONIZED) {
59		return ctx->ws->buffer_map(ctx->ws, resource->buf, NULL, usage);
60	}
61
62	if (!(usage & PIPE_MAP_WRITE)) {
63		/* have to wait for the last write */
64		rusage = RADEON_USAGE_WRITE;
65	}
66
67	if (radeon_emitted(&ctx->gfx.cs, ctx->initial_gfx_cs_size) &&
68	    ctx->ws->cs_is_buffer_referenced(&ctx->gfx.cs,
69					     resource->buf, rusage)) {
70		if (usage & PIPE_MAP_DONTBLOCK) {
71			ctx->gfx.flush(ctx, PIPE_FLUSH_ASYNC, NULL);
72			return NULL;
73		} else {
74			ctx->gfx.flush(ctx, 0, NULL);
75			busy = true;
76		}
77	}
78	if (radeon_emitted(&ctx->dma.cs, 0) &&
79	    ctx->ws->cs_is_buffer_referenced(&ctx->dma.cs,
80					     resource->buf, rusage)) {
81		if (usage & PIPE_MAP_DONTBLOCK) {
82			ctx->dma.flush(ctx, PIPE_FLUSH_ASYNC, NULL);
83			return NULL;
84		} else {
85			ctx->dma.flush(ctx, 0, NULL);
86			busy = true;
87		}
88	}
89
90	if (busy || !ctx->ws->buffer_wait(ctx->ws, resource->buf, 0, rusage)) {
91		if (usage & PIPE_MAP_DONTBLOCK) {
92			return NULL;
93		} else {
94			/* We will be wait for the GPU. Wait for any offloaded
95			 * CS flush to complete to avoid busy-waiting in the winsys. */
96			ctx->ws->cs_sync_flush(&ctx->gfx.cs);
97			if (ctx->dma.cs.priv)
98				ctx->ws->cs_sync_flush(&ctx->dma.cs);
99		}
100	}
101
102	/* Setting the CS to NULL will prevent doing checks we have done already. */
103	return ctx->ws->buffer_map(ctx->ws, resource->buf, NULL, usage);
104}
105
106void r600_init_resource_fields(struct r600_common_screen *rscreen,
107			       struct r600_resource *res,
108			       uint64_t size, unsigned alignment)
109{
110	struct r600_texture *rtex = (struct r600_texture*)res;
111
112	res->bo_size = size;
113	res->bo_alignment = alignment;
114	res->flags = 0;
115	res->texture_handle_allocated = false;
116	res->image_handle_allocated = false;
117
118	switch (res->b.b.usage) {
119	case PIPE_USAGE_STREAM:
120		res->flags = RADEON_FLAG_GTT_WC;
121		FALLTHROUGH;
122	case PIPE_USAGE_STAGING:
123		/* Transfers are likely to occur more often with these
124		 * resources. */
125		res->domains = RADEON_DOMAIN_GTT;
126		break;
127	case PIPE_USAGE_DYNAMIC:
128	case PIPE_USAGE_DEFAULT:
129	case PIPE_USAGE_IMMUTABLE:
130	default:
131		/* Not listing GTT here improves performance in some
132		 * apps. */
133		res->domains = RADEON_DOMAIN_VRAM;
134		res->flags |= RADEON_FLAG_GTT_WC;
135		break;
136	}
137
138	/* Tiled textures are unmappable. Always put them in VRAM. */
139	if ((res->b.b.target != PIPE_BUFFER && !rtex->surface.is_linear) ||
140	    res->flags & R600_RESOURCE_FLAG_UNMAPPABLE) {
141		res->domains = RADEON_DOMAIN_VRAM;
142		res->flags |= RADEON_FLAG_NO_CPU_ACCESS |
143			 RADEON_FLAG_GTT_WC;
144	}
145
146	/* Displayable and shareable surfaces are not suballocated. */
147	if (res->b.b.bind & (PIPE_BIND_SHARED | PIPE_BIND_SCANOUT))
148		res->flags |= RADEON_FLAG_NO_SUBALLOC; /* shareable */
149	else
150		res->flags |= RADEON_FLAG_NO_INTERPROCESS_SHARING;
151
152	if (rscreen->debug_flags & DBG_NO_WC)
153		res->flags &= ~RADEON_FLAG_GTT_WC;
154
155	/* Set expected VRAM and GART usage for the buffer. */
156	res->vram_usage = 0;
157	res->gart_usage = 0;
158
159	if (res->domains & RADEON_DOMAIN_VRAM)
160		res->vram_usage = size;
161	else if (res->domains & RADEON_DOMAIN_GTT)
162		res->gart_usage = size;
163}
164
165bool r600_alloc_resource(struct r600_common_screen *rscreen,
166			 struct r600_resource *res)
167{
168	struct pb_buffer *old_buf, *new_buf;
169
170	/* Allocate a new resource. */
171	new_buf = rscreen->ws->buffer_create(rscreen->ws, res->bo_size,
172					     res->bo_alignment,
173					     res->domains, res->flags);
174	if (!new_buf) {
175		return false;
176	}
177
178	/* Replace the pointer such that if res->buf wasn't NULL, it won't be
179	 * NULL. This should prevent crashes with multiple contexts using
180	 * the same buffer where one of the contexts invalidates it while
181	 * the others are using it. */
182	old_buf = res->buf;
183	res->buf = new_buf; /* should be atomic */
184
185	if (rscreen->info.r600_has_virtual_memory)
186		res->gpu_address = rscreen->ws->buffer_get_virtual_address(res->buf);
187	else
188		res->gpu_address = 0;
189
190	pb_reference(&old_buf, NULL);
191
192	util_range_set_empty(&res->valid_buffer_range);
193
194	/* Print debug information. */
195	if (rscreen->debug_flags & DBG_VM && res->b.b.target == PIPE_BUFFER) {
196		fprintf(stderr, "VM start=0x%"PRIX64"  end=0x%"PRIX64" | Buffer %"PRIu64" bytes\n",
197			res->gpu_address, res->gpu_address + res->buf->size,
198			res->buf->size);
199	}
200	return true;
201}
202
203void r600_buffer_destroy(struct pipe_screen *screen, struct pipe_resource *buf)
204{
205	struct r600_resource *rbuffer = r600_resource(buf);
206
207	threaded_resource_deinit(buf);
208	util_range_destroy(&rbuffer->valid_buffer_range);
209	pipe_resource_reference((struct pipe_resource**)&rbuffer->immed_buffer, NULL);
210	pb_reference(&rbuffer->buf, NULL);
211	FREE(rbuffer);
212}
213
214static bool
215r600_invalidate_buffer(struct r600_common_context *rctx,
216		       struct r600_resource *rbuffer)
217{
218	/* Shared buffers can't be reallocated. */
219	if (rbuffer->b.is_shared)
220		return false;
221
222	/* Sparse buffers can't be reallocated. */
223	if (rbuffer->flags & RADEON_FLAG_SPARSE)
224		return false;
225
226	/* In AMD_pinned_memory, the user pointer association only gets
227	 * broken when the buffer is explicitly re-allocated.
228	 */
229	if (rbuffer->b.is_user_ptr)
230		return false;
231
232	/* Check if mapping this buffer would cause waiting for the GPU. */
233	if (r600_rings_is_buffer_referenced(rctx, rbuffer->buf, RADEON_USAGE_READWRITE) ||
234	    !rctx->ws->buffer_wait(rctx->ws, rbuffer->buf, 0, RADEON_USAGE_READWRITE)) {
235		rctx->invalidate_buffer(&rctx->b, &rbuffer->b.b);
236	} else {
237		util_range_set_empty(&rbuffer->valid_buffer_range);
238	}
239
240	return true;
241}
242
243/* Replace the storage of dst with src. */
244void r600_replace_buffer_storage(struct pipe_context *ctx,
245				 struct pipe_resource *dst,
246				 struct pipe_resource *src)
247{
248	struct r600_common_context *rctx = (struct r600_common_context *)ctx;
249	struct r600_resource *rdst = r600_resource(dst);
250	struct r600_resource *rsrc = r600_resource(src);
251	uint64_t old_gpu_address = rdst->gpu_address;
252
253	pb_reference(&rdst->buf, rsrc->buf);
254	rdst->gpu_address = rsrc->gpu_address;
255	rdst->b.b.bind = rsrc->b.b.bind;
256	rdst->flags = rsrc->flags;
257
258	assert(rdst->vram_usage == rsrc->vram_usage);
259	assert(rdst->gart_usage == rsrc->gart_usage);
260	assert(rdst->bo_size == rsrc->bo_size);
261	assert(rdst->bo_alignment == rsrc->bo_alignment);
262	assert(rdst->domains == rsrc->domains);
263
264	rctx->rebind_buffer(ctx, dst, old_gpu_address);
265}
266
267void r600_invalidate_resource(struct pipe_context *ctx,
268			      struct pipe_resource *resource)
269{
270	struct r600_common_context *rctx = (struct r600_common_context*)ctx;
271	struct r600_resource *rbuffer = r600_resource(resource);
272
273	/* We currently only do anyting here for buffers */
274	if (resource->target == PIPE_BUFFER)
275		(void)r600_invalidate_buffer(rctx, rbuffer);
276}
277
278static void *r600_buffer_get_transfer(struct pipe_context *ctx,
279				      struct pipe_resource *resource,
280                                      unsigned usage,
281                                      const struct pipe_box *box,
282				      struct pipe_transfer **ptransfer,
283				      void *data, struct r600_resource *staging,
284				      unsigned offset)
285{
286	struct r600_common_context *rctx = (struct r600_common_context*)ctx;
287	struct r600_transfer *transfer;
288
289	if (usage & TC_TRANSFER_MAP_THREADED_UNSYNC)
290		transfer = slab_zalloc(&rctx->pool_transfers_unsync);
291	else
292		transfer = slab_zalloc(&rctx->pool_transfers);
293
294	pipe_resource_reference(&transfer->b.b.resource, resource);
295	transfer->b.b.usage = usage;
296	transfer->b.b.box = *box;
297	transfer->b.b.offset = offset;
298	transfer->staging = staging;
299	*ptransfer = &transfer->b.b;
300	return data;
301}
302
303static bool r600_can_dma_copy_buffer(struct r600_common_context *rctx,
304				     unsigned dstx, unsigned srcx, unsigned size)
305{
306	bool dword_aligned = !(dstx % 4) && !(srcx % 4) && !(size % 4);
307
308	return rctx->screen->has_cp_dma ||
309	       (dword_aligned && (rctx->dma.cs.priv ||
310				  rctx->screen->has_streamout));
311
312}
313
314void *r600_buffer_transfer_map(struct pipe_context *ctx,
315                               struct pipe_resource *resource,
316                               unsigned level,
317                               unsigned usage,
318                               const struct pipe_box *box,
319                               struct pipe_transfer **ptransfer)
320{
321	struct r600_common_context *rctx = (struct r600_common_context*)ctx;
322	struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen;
323	struct r600_resource *rbuffer = r600_resource(resource);
324	uint8_t *data;
325
326	if (r600_resource(resource)->compute_global_bo) {
327		if ((data = r600_compute_global_transfer_map(ctx, resource, level, usage, box, ptransfer)))
328			return data;
329	}
330
331	assert(box->x + box->width <= resource->width0);
332
333	/* From GL_AMD_pinned_memory issues:
334	 *
335	 *     4) Is glMapBuffer on a shared buffer guaranteed to return the
336	 *        same system address which was specified at creation time?
337	 *
338	 *        RESOLVED: NO. The GL implementation might return a different
339	 *        virtual mapping of that memory, although the same physical
340	 *        page will be used.
341	 *
342	 * So don't ever use staging buffers.
343	 */
344	if (rbuffer->b.is_user_ptr)
345		usage |= PIPE_MAP_PERSISTENT;
346
347	/* See if the buffer range being mapped has never been initialized,
348	 * in which case it can be mapped unsynchronized. */
349	if (!(usage & (PIPE_MAP_UNSYNCHRONIZED |
350		       TC_TRANSFER_MAP_NO_INFER_UNSYNCHRONIZED)) &&
351	    usage & PIPE_MAP_WRITE &&
352	    !rbuffer->b.is_shared &&
353	    !util_ranges_intersect(&rbuffer->valid_buffer_range, box->x, box->x + box->width)) {
354		usage |= PIPE_MAP_UNSYNCHRONIZED;
355	}
356
357	/* If discarding the entire range, discard the whole resource instead. */
358	if (usage & PIPE_MAP_DISCARD_RANGE &&
359	    box->x == 0 && box->width == resource->width0) {
360		usage |= PIPE_MAP_DISCARD_WHOLE_RESOURCE;
361	}
362
363	if (usage & PIPE_MAP_DISCARD_WHOLE_RESOURCE &&
364	    !(usage & (PIPE_MAP_UNSYNCHRONIZED |
365		       TC_TRANSFER_MAP_NO_INVALIDATE))) {
366		assert(usage & PIPE_MAP_WRITE);
367
368		if (r600_invalidate_buffer(rctx, rbuffer)) {
369			/* At this point, the buffer is always idle. */
370			usage |= PIPE_MAP_UNSYNCHRONIZED;
371		} else {
372			/* Fall back to a temporary buffer. */
373			usage |= PIPE_MAP_DISCARD_RANGE;
374		}
375	}
376
377	if ((usage & PIPE_MAP_DISCARD_RANGE) &&
378	    !(rscreen->debug_flags & DBG_NO_DISCARD_RANGE) &&
379	    ((!(usage & (PIPE_MAP_UNSYNCHRONIZED |
380			 PIPE_MAP_PERSISTENT)) &&
381	      r600_can_dma_copy_buffer(rctx, box->x, 0, box->width)) ||
382	     (rbuffer->flags & RADEON_FLAG_SPARSE))) {
383		assert(usage & PIPE_MAP_WRITE);
384
385		/* Check if mapping this buffer would cause waiting for the GPU.
386		 */
387		if (rbuffer->flags & RADEON_FLAG_SPARSE ||
388		    r600_rings_is_buffer_referenced(rctx, rbuffer->buf, RADEON_USAGE_READWRITE) ||
389		    !rctx->ws->buffer_wait(rctx->ws, rbuffer->buf, 0, RADEON_USAGE_READWRITE)) {
390			/* Do a wait-free write-only transfer using a temporary buffer. */
391			unsigned offset;
392			struct r600_resource *staging = NULL;
393
394			u_upload_alloc(ctx->stream_uploader, 0,
395                                       box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT),
396				       rctx->screen->info.tcc_cache_line_size,
397				       &offset, (struct pipe_resource**)&staging,
398                                       (void**)&data);
399
400			if (staging) {
401				data += box->x % R600_MAP_BUFFER_ALIGNMENT;
402				return r600_buffer_get_transfer(ctx, resource, usage, box,
403								ptransfer, data, staging, offset);
404			} else if (rbuffer->flags & RADEON_FLAG_SPARSE) {
405				return NULL;
406			}
407		} else {
408			/* At this point, the buffer is always idle (we checked it above). */
409			usage |= PIPE_MAP_UNSYNCHRONIZED;
410		}
411	}
412	/* Use a staging buffer in cached GTT for reads. */
413	else if (((usage & PIPE_MAP_READ) &&
414		  !(usage & PIPE_MAP_PERSISTENT) &&
415		  (rbuffer->domains & RADEON_DOMAIN_VRAM ||
416		   rbuffer->flags & RADEON_FLAG_GTT_WC) &&
417		  r600_can_dma_copy_buffer(rctx, 0, box->x, box->width)) ||
418		 (rbuffer->flags & RADEON_FLAG_SPARSE)) {
419		struct r600_resource *staging;
420
421		assert(!(usage & TC_TRANSFER_MAP_THREADED_UNSYNC));
422		staging = (struct r600_resource*) pipe_buffer_create(
423				ctx->screen, 0, PIPE_USAGE_STAGING,
424				box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT));
425		if (staging) {
426			/* Copy the VRAM buffer to the staging buffer. */
427			rctx->dma_copy(ctx, &staging->b.b, 0,
428				       box->x % R600_MAP_BUFFER_ALIGNMENT,
429				       0, 0, resource, 0, box);
430
431			data = r600_buffer_map_sync_with_rings(rctx, staging,
432							       usage & ~PIPE_MAP_UNSYNCHRONIZED);
433			if (!data) {
434				r600_resource_reference(&staging, NULL);
435				return NULL;
436			}
437			data += box->x % R600_MAP_BUFFER_ALIGNMENT;
438
439			return r600_buffer_get_transfer(ctx, resource, usage, box,
440							ptransfer, data, staging, 0);
441		} else if (rbuffer->flags & RADEON_FLAG_SPARSE) {
442			return NULL;
443		}
444	}
445
446	data = r600_buffer_map_sync_with_rings(rctx, rbuffer, usage);
447	if (!data) {
448		return NULL;
449	}
450	data += box->x;
451
452	return r600_buffer_get_transfer(ctx, resource, usage, box,
453					ptransfer, data, NULL, 0);
454}
455
456static void r600_buffer_do_flush_region(struct pipe_context *ctx,
457					struct pipe_transfer *transfer,
458				        const struct pipe_box *box)
459{
460	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
461	struct r600_resource *rbuffer = r600_resource(transfer->resource);
462
463	if (rtransfer->staging) {
464		struct pipe_resource *dst, *src;
465		unsigned soffset;
466		struct pipe_box dma_box;
467
468		dst = transfer->resource;
469		src = &rtransfer->staging->b.b;
470		soffset = rtransfer->b.b.offset + box->x % R600_MAP_BUFFER_ALIGNMENT;
471
472		u_box_1d(soffset, box->width, &dma_box);
473
474		/* Copy the staging buffer into the original one. */
475		ctx->resource_copy_region(ctx, dst, 0, box->x, 0, 0, src, 0, &dma_box);
476	}
477
478	util_range_add(&rbuffer->b.b, &rbuffer->valid_buffer_range, box->x,
479		       box->x + box->width);
480}
481
482void r600_buffer_flush_region(struct pipe_context *ctx,
483			      struct pipe_transfer *transfer,
484			      const struct pipe_box *rel_box)
485{
486	unsigned required_usage = PIPE_MAP_WRITE |
487				  PIPE_MAP_FLUSH_EXPLICIT;
488
489	if (r600_resource(transfer->resource)->compute_global_bo)
490		return;
491
492	if ((transfer->usage & required_usage) == required_usage) {
493		struct pipe_box box;
494
495		u_box_1d(transfer->box.x + rel_box->x, rel_box->width, &box);
496		r600_buffer_do_flush_region(ctx, transfer, &box);
497	}
498}
499
500void r600_buffer_transfer_unmap(struct pipe_context *ctx,
501				struct pipe_transfer *transfer)
502{
503	struct r600_common_context *rctx = (struct r600_common_context*)ctx;
504	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
505	struct r600_resource *rtransferr = r600_resource(transfer->resource);
506
507	if (rtransferr->compute_global_bo && !rtransferr->b.is_user_ptr) {
508		r600_compute_global_transfer_unmap(ctx, transfer);
509		return;
510	}
511
512	if (transfer->usage & PIPE_MAP_WRITE &&
513	    !(transfer->usage & PIPE_MAP_FLUSH_EXPLICIT))
514		r600_buffer_do_flush_region(ctx, transfer, &transfer->box);
515
516	r600_resource_reference(&rtransfer->staging, NULL);
517	assert(rtransfer->b.staging == NULL); /* for threaded context only */
518	pipe_resource_reference(&transfer->resource, NULL);
519
520	/* Don't use pool_transfers_unsync. We are always in the driver
521	 * thread. */
522	slab_free(&rctx->pool_transfers, transfer);
523}
524
525void r600_buffer_subdata(struct pipe_context *ctx,
526			 struct pipe_resource *buffer,
527			 unsigned usage, unsigned offset,
528			 unsigned size, const void *data)
529{
530	struct pipe_transfer *transfer = NULL;
531	struct pipe_box box;
532	uint8_t *map = NULL;
533
534	usage |= PIPE_MAP_WRITE;
535
536	if (!(usage & PIPE_MAP_DIRECTLY))
537		usage |= PIPE_MAP_DISCARD_RANGE;
538
539	u_box_1d(offset, size, &box);
540	map = r600_buffer_transfer_map(ctx, buffer, 0, usage, &box, &transfer);
541	if (!map)
542		return;
543
544	memcpy(map, data, size);
545	r600_buffer_transfer_unmap(ctx, transfer);
546}
547
548static struct r600_resource *
549r600_alloc_buffer_struct(struct pipe_screen *screen,
550			 const struct pipe_resource *templ)
551{
552	struct r600_resource *rbuffer;
553
554	rbuffer = MALLOC_STRUCT(r600_resource);
555
556	rbuffer->b.b = *templ;
557	rbuffer->b.b.next = NULL;
558	pipe_reference_init(&rbuffer->b.b.reference, 1);
559	rbuffer->b.b.screen = screen;
560
561	threaded_resource_init(&rbuffer->b.b, false);
562
563	rbuffer->buf = NULL;
564	rbuffer->bind_history = 0;
565	rbuffer->immed_buffer = NULL;
566	rbuffer->compute_global_bo = false;
567	util_range_init(&rbuffer->valid_buffer_range);
568	return rbuffer;
569}
570
571struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
572					 const struct pipe_resource *templ,
573					 unsigned alignment)
574{
575	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
576	struct r600_resource *rbuffer = r600_alloc_buffer_struct(screen, templ);
577
578	r600_init_resource_fields(rscreen, rbuffer, templ->width0, alignment);
579
580	if (templ->flags & PIPE_RESOURCE_FLAG_SPARSE)
581		rbuffer->flags |= RADEON_FLAG_SPARSE;
582
583	if (!r600_alloc_resource(rscreen, rbuffer)) {
584		FREE(rbuffer);
585		return NULL;
586	}
587	return &rbuffer->b.b;
588}
589
590struct pipe_resource *r600_aligned_buffer_create(struct pipe_screen *screen,
591						 unsigned flags,
592						 unsigned usage,
593						 unsigned size,
594						 unsigned alignment)
595{
596	struct pipe_resource buffer;
597
598	memset(&buffer, 0, sizeof buffer);
599	buffer.target = PIPE_BUFFER;
600	buffer.format = PIPE_FORMAT_R8_UNORM;
601	buffer.bind = 0;
602	buffer.usage = usage;
603	buffer.flags = flags;
604	buffer.width0 = size;
605	buffer.height0 = 1;
606	buffer.depth0 = 1;
607	buffer.array_size = 1;
608	return r600_buffer_create(screen, &buffer, alignment);
609}
610
611struct pipe_resource *
612r600_buffer_from_user_memory(struct pipe_screen *screen,
613			     const struct pipe_resource *templ,
614			     void *user_memory)
615{
616	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
617	struct radeon_winsys *ws = rscreen->ws;
618	struct r600_resource *rbuffer;
619
620	if ((templ->bind & PIPE_BIND_GLOBAL) &&
621	    (templ->bind & PIPE_BIND_COMPUTE_RESOURCE)) {
622		rbuffer = r600_resource(r600_compute_global_buffer_create(screen, templ));
623		((struct r600_resource_global *)rbuffer)->chunk->real_buffer = rbuffer;
624	} else {
625		rbuffer = r600_alloc_buffer_struct(screen, templ);
626	}
627
628	rbuffer->domains = RADEON_DOMAIN_GTT;
629	rbuffer->flags = 0;
630	rbuffer->b.is_user_ptr = true;
631	util_range_add(&rbuffer->b.b, &rbuffer->valid_buffer_range, 0, templ->width0);
632	util_range_add(&rbuffer->b.b, &rbuffer->b.valid_buffer_range, 0, templ->width0);
633
634	/* Convert a user pointer to a buffer. */
635	rbuffer->buf = ws->buffer_from_ptr(ws, user_memory, templ->width0,
636	                                   templ->usage == PIPE_USAGE_IMMUTABLE? RADEON_FLAG_READ_ONLY : 0);
637	if (!rbuffer->buf) {
638		FREE(rbuffer);
639		return NULL;
640	}
641
642	if (rscreen->info.r600_has_virtual_memory)
643		rbuffer->gpu_address =
644			ws->buffer_get_virtual_address(rbuffer->buf);
645	else
646		rbuffer->gpu_address = 0;
647
648	rbuffer->vram_usage = 0;
649	rbuffer->gart_usage = templ->width0;
650
651	return &rbuffer->b.b;
652}
653