summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/radeon/r600_buffer_common.c
blob: 5b5c06362a72ae1d0a1bf10c4b66fc5e6af5de3c (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
/*
 * Copyright 2013 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * on the rights to use, copy, modify, merge, publish, distribute, sub
 * license, and/or sell copies of the Software, and to permit persons to whom
 * the Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * Authors:
 *      Marek Olšák
 */

#include "r600_cs.h"
#include "util/u_memory.h"
#include "util/u_upload_mgr.h"
#include <inttypes.h>
#include <stdio.h>

boolean r600_rings_is_buffer_referenced(struct r600_common_context *ctx,
					struct radeon_winsys_cs_handle *buf,
					enum radeon_bo_usage usage)
{
	if (ctx->ws->cs_is_buffer_referenced(ctx->rings.gfx.cs, buf, usage)) {
		return TRUE;
	}
	if (ctx->rings.dma.cs && ctx->rings.dma.cs->cdw &&
	    ctx->ws->cs_is_buffer_referenced(ctx->rings.dma.cs, buf, usage)) {
		return TRUE;
	}
	return FALSE;
}

void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx,
                                      struct r600_resource *resource,
                                      unsigned usage)
{
	enum radeon_bo_usage rusage = RADEON_USAGE_READWRITE;
	bool busy = false;

	if (usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
		return ctx->ws->buffer_map(resource->cs_buf, NULL, usage);
	}

	if (!(usage & PIPE_TRANSFER_WRITE)) {
		/* have to wait for the last write */
		rusage = RADEON_USAGE_WRITE;
	}

	if (ctx->rings.gfx.cs->cdw != ctx->initial_gfx_cs_size &&
	    ctx->ws->cs_is_buffer_referenced(ctx->rings.gfx.cs,
					     resource->cs_buf, rusage)) {
		if (usage & PIPE_TRANSFER_DONTBLOCK) {
			ctx->rings.gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
			return NULL;
		} else {
			ctx->rings.gfx.flush(ctx, 0, NULL);
			busy = true;
		}
	}
	if (ctx->rings.dma.cs &&
	    ctx->rings.dma.cs->cdw &&
	    ctx->ws->cs_is_buffer_referenced(ctx->rings.dma.cs,
					     resource->cs_buf, rusage)) {
		if (usage & PIPE_TRANSFER_DONTBLOCK) {
			ctx->rings.dma.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
			return NULL;
		} else {
			ctx->rings.dma.flush(ctx, 0, NULL);
			busy = true;
		}
	}

	if (busy || !ctx->ws->buffer_wait(resource->buf, 0, rusage)) {
		if (usage & PIPE_TRANSFER_DONTBLOCK) {
			return NULL;
		} else {
			/* We will be wait for the GPU. Wait for any offloaded
			 * CS flush to complete to avoid busy-waiting in the winsys. */
			ctx->ws->cs_sync_flush(ctx->rings.gfx.cs);
			if (ctx->rings.dma.cs)
				ctx->ws->cs_sync_flush(ctx->rings.dma.cs);
		}
	}

	/* Setting the CS to NULL will prevent doing checks we have done already. */
	return ctx->ws->buffer_map(resource->cs_buf, NULL, usage);
}

bool r600_init_resource(struct r600_common_screen *rscreen,
			struct r600_resource *res,
			unsigned size, unsigned alignment,
			bool use_reusable_pool)
{
	struct r600_texture *rtex = (struct r600_texture*)res;
	struct pb_buffer *old_buf, *new_buf;
	enum radeon_bo_flag flags = 0;

	switch (res->b.b.usage) {
	case PIPE_USAGE_STREAM:
		flags = RADEON_FLAG_GTT_WC;
		/* fall through */
	case PIPE_USAGE_STAGING:
		/* Transfers are likely to occur more often with these resources. */
		res->domains = RADEON_DOMAIN_GTT;
		break;
	case PIPE_USAGE_DYNAMIC:
		/* Older kernels didn't always flush the HDP cache before
		 * CS execution
		 */
		if (rscreen->info.drm_minor < 40) {
			res->domains = RADEON_DOMAIN_GTT;
			flags |= RADEON_FLAG_GTT_WC;
			break;
		}
		flags |= RADEON_FLAG_CPU_ACCESS;
		/* fall through */
	case PIPE_USAGE_DEFAULT:
	case PIPE_USAGE_IMMUTABLE:
	default:
		/* Not listing GTT here improves performance in some apps. */
		res->domains = RADEON_DOMAIN_VRAM;
		flags |= RADEON_FLAG_GTT_WC;
		break;
	}

	if (res->b.b.target == PIPE_BUFFER &&
	    res->b.b.flags & (PIPE_RESOURCE_FLAG_MAP_PERSISTENT |
			      PIPE_RESOURCE_FLAG_MAP_COHERENT)) {
		/* Use GTT for all persistent mappings with older kernels,
		 * because they didn't always flush the HDP cache before CS
		 * execution.
		 *
		 * Write-combined CPU mappings are fine, the kernel ensures all CPU
		 * writes finish before the GPU executes a command stream.
		 */
		if (rscreen->info.drm_minor < 40)
			res->domains = RADEON_DOMAIN_GTT;
		else if (res->domains & RADEON_DOMAIN_VRAM)
			flags |= RADEON_FLAG_CPU_ACCESS;
	}

	/* Tiled textures are unmappable. Always put them in VRAM. */
	if (res->b.b.target != PIPE_BUFFER &&
	    rtex->surface.level[0].mode >= RADEON_SURF_MODE_1D) {
		res->domains = RADEON_DOMAIN_VRAM;
		flags &= ~RADEON_FLAG_CPU_ACCESS;
		flags |= RADEON_FLAG_NO_CPU_ACCESS;
	}

	/* Allocate a new resource. */
	new_buf = rscreen->ws->buffer_create(rscreen->ws, size, alignment,
					     use_reusable_pool,
					     res->domains, flags);
	if (!new_buf) {
		return false;
	}

	/* Replace the pointer such that if res->buf wasn't NULL, it won't be
	 * NULL. This should prevent crashes with multiple contexts using
	 * the same buffer where one of the contexts invalidates it while
	 * the others are using it. */
	old_buf = res->buf;
	res->cs_buf = rscreen->ws->buffer_get_cs_handle(new_buf); /* should be atomic */
	res->buf = new_buf; /* should be atomic */

	if (rscreen->info.r600_virtual_address)
		res->gpu_address = rscreen->ws->buffer_get_virtual_address(res->cs_buf);
	else
		res->gpu_address = 0;

	pb_reference(&old_buf, NULL);

	util_range_set_empty(&res->valid_buffer_range);
	res->TC_L2_dirty = false;

	if (rscreen->debug_flags & DBG_VM && res->b.b.target == PIPE_BUFFER) {
		fprintf(stderr, "VM start=0x%"PRIX64"  end=0x%"PRIX64" | Buffer %u bytes\n",
			res->gpu_address, res->gpu_address + res->buf->size,
			res->buf->size);
	}
	return true;
}

static void r600_buffer_destroy(struct pipe_screen *screen,
				struct pipe_resource *buf)
{
	struct r600_resource *rbuffer = r600_resource(buf);

	util_range_destroy(&rbuffer->valid_buffer_range);
	pb_reference(&rbuffer->buf, NULL);
	FREE(rbuffer);
}

static void *r600_buffer_get_transfer(struct pipe_context *ctx,
				      struct pipe_resource *resource,
                                      unsigned level,
                                      unsigned usage,
                                      const struct pipe_box *box,
				      struct pipe_transfer **ptransfer,
				      void *data, struct r600_resource *staging,
				      unsigned offset)
{
	struct r600_common_context *rctx = (struct r600_common_context*)ctx;
	struct r600_transfer *transfer = util_slab_alloc(&rctx->pool_transfers);

	transfer->transfer.resource = resource;
	transfer->transfer.level = level;
	transfer->transfer.usage = usage;
	transfer->transfer.box = *box;
	transfer->transfer.stride = 0;
	transfer->transfer.layer_stride = 0;
	transfer->offset = offset;
	transfer->staging = staging;
	*ptransfer = &transfer->transfer;
	return data;
}

static bool r600_can_dma_copy_buffer(struct r600_common_context *rctx,
				     unsigned dstx, unsigned srcx, unsigned size)
{
	bool dword_aligned = !(dstx % 4) && !(srcx % 4) && !(size % 4);

	return rctx->screen->has_cp_dma ||
	       (dword_aligned && (rctx->rings.dma.cs ||
				  rctx->screen->has_streamout));

}

static void *r600_buffer_transfer_map(struct pipe_context *ctx,
                                      struct pipe_resource *resource,
                                      unsigned level,
                                      unsigned usage,
                                      const struct pipe_box *box,
                                      struct pipe_transfer **ptransfer)
{
	struct r600_common_context *rctx = (struct r600_common_context*)ctx;
	struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen;
        struct r600_resource *rbuffer = r600_resource(resource);
        uint8_t *data;

	assert(box->x + box->width <= resource->width0);

	/* See if the buffer range being mapped has never been initialized,
	 * in which case it can be mapped unsynchronized. */
	if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
	    usage & PIPE_TRANSFER_WRITE &&
	    !util_ranges_intersect(&rbuffer->valid_buffer_range, box->x, box->x + box->width)) {
		usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
	}

	/* If discarding the entire range, discard the whole resource instead. */
	if (usage & PIPE_TRANSFER_DISCARD_RANGE &&
	    box->x == 0 && box->width == resource->width0) {
		usage |= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE;
	}

	if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE &&
	    !(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
		assert(usage & PIPE_TRANSFER_WRITE);

		/* Check if mapping this buffer would cause waiting for the GPU. */
		if (r600_rings_is_buffer_referenced(rctx, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
		    !rctx->ws->buffer_wait(rbuffer->buf, 0, RADEON_USAGE_READWRITE)) {
			rctx->invalidate_buffer(&rctx->b, &rbuffer->b.b);
		}
		/* At this point, the buffer is always idle. */
		usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
	}
	else if ((usage & PIPE_TRANSFER_DISCARD_RANGE) &&
		 !(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
		 !(rscreen->debug_flags & DBG_NO_DISCARD_RANGE) &&
		 r600_can_dma_copy_buffer(rctx, box->x, 0, box->width)) {
		assert(usage & PIPE_TRANSFER_WRITE);

		/* Check if mapping this buffer would cause waiting for the GPU. */
		if (r600_rings_is_buffer_referenced(rctx, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
		    !rctx->ws->buffer_wait(rbuffer->buf, 0, RADEON_USAGE_READWRITE)) {
			/* Do a wait-free write-only transfer using a temporary buffer. */
			unsigned offset;
			struct r600_resource *staging = NULL;

			u_upload_alloc(rctx->uploader, 0, box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT),
				       &offset, (struct pipe_resource**)&staging, (void**)&data);

			if (staging) {
				data += box->x % R600_MAP_BUFFER_ALIGNMENT;
				return r600_buffer_get_transfer(ctx, resource, level, usage, box,
								ptransfer, data, staging, offset);
			} else {
				return NULL; /* error, shouldn't occur though */
			}
		}
		/* At this point, the buffer is always idle (we checked it above). */
		usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
	}
	/* Using a staging buffer in GTT for larger reads is much faster. */
	else if ((usage & PIPE_TRANSFER_READ) &&
		 !(usage & PIPE_TRANSFER_WRITE) &&
		 rbuffer->domains == RADEON_DOMAIN_VRAM &&
		 r600_can_dma_copy_buffer(rctx, 0, box->x, box->width)) {
		struct r600_resource *staging;

		staging = (struct r600_resource*) pipe_buffer_create(
				ctx->screen, PIPE_BIND_TRANSFER_READ, PIPE_USAGE_STAGING,
				box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT));
		if (staging) {
			/* Copy the VRAM buffer to the staging buffer. */
			rctx->dma_copy(ctx, &staging->b.b, 0,
				       box->x % R600_MAP_BUFFER_ALIGNMENT,
				       0, 0, resource, level, box);

			data = r600_buffer_map_sync_with_rings(rctx, staging, PIPE_TRANSFER_READ);
			data += box->x % R600_MAP_BUFFER_ALIGNMENT;

			return r600_buffer_get_transfer(ctx, resource, level, usage, box,
							ptransfer, data, staging, 0);
		}
	}

	data = r600_buffer_map_sync_with_rings(rctx, rbuffer, usage);
	if (!data) {
		return NULL;
	}
	data += box->x;

	return r600_buffer_get_transfer(ctx, resource, level, usage, box,
					ptransfer, data, NULL, 0);
}

static void r600_buffer_transfer_unmap(struct pipe_context *ctx,
				       struct pipe_transfer *transfer)
{
	struct r600_common_context *rctx = (struct r600_common_context*)ctx;
	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
	struct r600_resource *rbuffer = r600_resource(transfer->resource);

	if (rtransfer->staging) {
		if (rtransfer->transfer.usage & PIPE_TRANSFER_WRITE) {
			struct pipe_resource *dst, *src;
			unsigned soffset, doffset, size;
			struct pipe_box box;

			dst = transfer->resource;
			src = &rtransfer->staging->b.b;
			size = transfer->box.width;
			doffset = transfer->box.x;
			soffset = rtransfer->offset + transfer->box.x % R600_MAP_BUFFER_ALIGNMENT;

			u_box_1d(soffset, size, &box);

			/* Copy the staging buffer into the original one. */
			rctx->dma_copy(ctx, dst, 0, doffset, 0, 0, src, 0, &box);
		}
		pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL);
	}

	if (transfer->usage & PIPE_TRANSFER_WRITE) {
		util_range_add(&rbuffer->valid_buffer_range, transfer->box.x,
			       transfer->box.x + transfer->box.width);
	}
	util_slab_free(&rctx->pool_transfers, transfer);
}

static const struct u_resource_vtbl r600_buffer_vtbl =
{
	NULL,				/* get_handle */
	r600_buffer_destroy,		/* resource_destroy */
	r600_buffer_transfer_map,	/* transfer_map */
	NULL,				/* transfer_flush_region */
	r600_buffer_transfer_unmap,	/* transfer_unmap */
	NULL				/* transfer_inline_write */
};

static struct r600_resource *
r600_alloc_buffer_struct(struct pipe_screen *screen,
			 const struct pipe_resource *templ)
{
	struct r600_resource *rbuffer;

	rbuffer = MALLOC_STRUCT(r600_resource);

	rbuffer->b.b = *templ;
	pipe_reference_init(&rbuffer->b.b.reference, 1);
	rbuffer->b.b.screen = screen;
	rbuffer->b.vtbl = &r600_buffer_vtbl;
	rbuffer->buf = NULL;
	rbuffer->TC_L2_dirty = false;
	util_range_init(&rbuffer->valid_buffer_range);
	return rbuffer;
}

struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
					 const struct pipe_resource *templ,
					 unsigned alignment)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
	struct r600_resource *rbuffer = r600_alloc_buffer_struct(screen, templ);

	if (!r600_init_resource(rscreen, rbuffer, templ->width0, alignment, TRUE)) {
		FREE(rbuffer);
		return NULL;
	}
	return &rbuffer->b.b;
}

struct pipe_resource *
r600_buffer_from_user_memory(struct pipe_screen *screen,
			     const struct pipe_resource *templ,
			     void *user_memory)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
	struct radeon_winsys *ws = rscreen->ws;
	struct r600_resource *rbuffer = r600_alloc_buffer_struct(screen, templ);

	rbuffer->domains = RADEON_DOMAIN_GTT;
	util_range_add(&rbuffer->valid_buffer_range, 0, templ->width0);

	/* Convert a user pointer to a buffer. */
	rbuffer->buf = ws->buffer_from_ptr(ws, user_memory, templ->width0);
	if (!rbuffer->buf) {
		FREE(rbuffer);
		return NULL;
	}

	rbuffer->cs_buf = ws->buffer_get_cs_handle(rbuffer->buf);

	if (rscreen->info.r600_virtual_address)
		rbuffer->gpu_address =
			ws->buffer_get_virtual_address(rbuffer->cs_buf);
	else
		rbuffer->gpu_address = 0;

	return &rbuffer->b.b;
}