summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/radeonsi/gfx10_query.c
blob: 8584b2af505ebed7f16310f8d4041e60fd835a9f (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
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
/*
 * Copyright 2018 Advanced Micro Devices, Inc.
 * All Rights Reserved.
 *
 * 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.
 */

#include <stddef.h>

#include "si_pipe.h"
#include "si_query.h"
#include "util/u_memory.h"
#include "util/u_suballoc.h"
#include "sid.h"

/**
 * The query buffer is written to by ESGS NGG shaders with statistics about
 * generated and (streamout-)emitted primitives.
 *
 * The context maintains a ring of these query buffers, and queries simply
 * point into the ring, allowing an arbitrary number of queries to be active
 * without additional GPU cost.
 */
struct gfx10_sh_query_buffer {
	struct list_head list;
	struct si_resource *buf;
	unsigned refcount;

	/* Offset into the buffer in bytes; points at the first un-emitted entry. */
	unsigned head;
};

/* Memory layout of the query buffer. Must be kept in sync with shaders
 * (including QBO shaders) and should be aligned to cachelines.
 *
 * The somewhat awkward memory layout is for compatibility with the
 * SET_PREDICATION packet, which also means that we're setting the high bit
 * of all those values unconditionally.
 */
struct gfx10_sh_query_buffer_mem {
	struct {
		uint64_t generated_primitives_start_dummy;
		uint64_t emitted_primitives_start_dummy;
		uint64_t generated_primitives;
		uint64_t emitted_primitives;
	} stream[4];
	uint32_t fence; /* bottom-of-pipe fence: set to ~0 when draws have finished */
	uint32_t pad[31];
};

/* Shader-based queries. */
struct gfx10_sh_query {
	struct si_query b;

	struct gfx10_sh_query_buffer *first;
	struct gfx10_sh_query_buffer *last;
	unsigned first_begin;
	unsigned last_end;

	unsigned stream;
};

static void emit_shader_query(struct si_context *sctx)
{
	assert(!LIST_IS_EMPTY(&sctx->shader_query_buffers));

	struct gfx10_sh_query_buffer *qbuf = list_last_entry(&sctx->shader_query_buffers,
							     struct gfx10_sh_query_buffer, list);
	qbuf->head += sizeof(struct gfx10_sh_query_buffer_mem);
}

static void gfx10_release_query_buffers(struct si_context *sctx,
					struct gfx10_sh_query_buffer *first,
					struct gfx10_sh_query_buffer *last)
{
	while (first) {
		struct gfx10_sh_query_buffer *qbuf = first;
		if (first != last)
			first = LIST_ENTRY(struct gfx10_sh_query_buffer, qbuf->list.next, list);
		else
			first = NULL;

		qbuf->refcount--;
		if (qbuf->refcount)
			continue;

		if (qbuf->list.next == &sctx->shader_query_buffers)
			continue; /* keep the most recent buffer; it may not be full yet */
		if (qbuf->list.prev == &sctx->shader_query_buffers)
			continue; /* keep the oldest buffer for recycling */

		LIST_DEL(&qbuf->list);
		si_resource_reference(&qbuf->buf, NULL);
		FREE(qbuf);
	}
}

static bool gfx10_alloc_query_buffer(struct si_context *sctx)
{
	if (si_is_atom_dirty(sctx, &sctx->atoms.s.shader_query))
		return true;

	struct gfx10_sh_query_buffer *qbuf = NULL;

	if (!LIST_IS_EMPTY(&sctx->shader_query_buffers)) {
		qbuf = list_last_entry(&sctx->shader_query_buffers,
				       struct gfx10_sh_query_buffer, list);
		if (qbuf->head + sizeof(struct gfx10_sh_query_buffer_mem) <= qbuf->buf->b.b.width0)
			goto success;

		qbuf = list_first_entry(&sctx->shader_query_buffers,
				        struct gfx10_sh_query_buffer, list);
		if (!qbuf->refcount &&
		    !si_rings_is_buffer_referenced(sctx, qbuf->buf->buf, RADEON_USAGE_READWRITE) &&
		    sctx->ws->buffer_wait(qbuf->buf->buf, 0, RADEON_USAGE_READWRITE)) {
			/* Can immediately re-use the oldest buffer */
			LIST_DEL(&qbuf->list);
		} else {
			qbuf = NULL;
		}
	}

	if (!qbuf) {
		qbuf = CALLOC_STRUCT(gfx10_sh_query_buffer);
		if (unlikely(!qbuf))
			return false;

		struct si_screen *screen = sctx->screen;
		unsigned buf_size = MAX2(sizeof(struct gfx10_sh_query_buffer_mem),
					 screen->info.min_alloc_size);
		qbuf->buf = si_resource(
			pipe_buffer_create(&screen->b, 0, PIPE_USAGE_STAGING, buf_size));
		if (unlikely(!qbuf->buf)) {
			FREE(qbuf);
			return false;
		}
	}

	/* The buffer is currently unused by the GPU. Initialize it.
	 *
	 * We need to set the high bit of all the primitive counters for
	 * compatibility with the SET_PREDICATION packet.
	 */
	uint64_t *results = sctx->ws->buffer_map(qbuf->buf->buf, NULL,
						 PIPE_TRANSFER_WRITE |
						 PIPE_TRANSFER_UNSYNCHRONIZED);
	assert(results);

	for (unsigned i = 0, e = qbuf->buf->b.b.width0 / sizeof(struct gfx10_sh_query_buffer_mem);
	     i < e; ++i) {
		for (unsigned j = 0; j < 16; ++j)
			results[32 * i + j] = (uint64_t)1 << 63;
		results[32 * i + 16] = 0;
	}

	LIST_ADDTAIL(&qbuf->list, &sctx->shader_query_buffers);
	qbuf->head = 0;
	qbuf->refcount = sctx->num_active_shader_queries;

success:;
	struct pipe_shader_buffer sbuf;
	sbuf.buffer = &qbuf->buf->b.b;
	sbuf.buffer_offset = qbuf->head;
	sbuf.buffer_size = sizeof(struct gfx10_sh_query_buffer_mem);
	si_set_rw_shader_buffer(sctx, GFX10_GS_QUERY_BUF, &sbuf);

	si_mark_atom_dirty(sctx, &sctx->atoms.s.shader_query);
	return true;
}

static void gfx10_sh_query_destroy(struct si_context *sctx, struct si_query *rquery)
{
	struct gfx10_sh_query *query = (struct gfx10_sh_query *)rquery;
	gfx10_release_query_buffers(sctx, query->first, query->last);
	FREE(query);
}

static bool gfx10_sh_query_begin(struct si_context *sctx, struct si_query *rquery)
{
	struct gfx10_sh_query *query = (struct gfx10_sh_query *)rquery;

	gfx10_release_query_buffers(sctx, query->first, query->last);
	query->first = query->last = NULL;

	if (unlikely(!gfx10_alloc_query_buffer(sctx)))
		return false;

	query->first = list_last_entry(&sctx->shader_query_buffers,
				       struct gfx10_sh_query_buffer, list);
	query->first_begin = query->first->head;

	sctx->num_active_shader_queries++;
	query->first->refcount++;

	return true;
}

static bool gfx10_sh_query_end(struct si_context *sctx, struct si_query *rquery)
{
	struct gfx10_sh_query *query = (struct gfx10_sh_query *)rquery;

	if (unlikely(!query->first))
		return false; /* earlier out of memory error */

	query->last = list_last_entry(&sctx->shader_query_buffers,
				      struct gfx10_sh_query_buffer, list);
	query->last_end = query->last->head;

	/* Signal the fence of the previous chunk */
	if (query->last_end != 0) {
		uint64_t fence_va = query->last->buf->gpu_address;
		fence_va += query->last_end - sizeof(struct gfx10_sh_query_buffer_mem);
		fence_va += offsetof(struct gfx10_sh_query_buffer_mem, fence);
		si_cp_release_mem(sctx, sctx->gfx_cs,
				  V_028A90_BOTTOM_OF_PIPE_TS, 0,
				  EOP_DST_SEL_MEM, EOP_INT_SEL_NONE,
				  EOP_DATA_SEL_VALUE_32BIT,
				  query->last->buf, fence_va, 0xffffffff,
				  PIPE_QUERY_GPU_FINISHED);
	}

	sctx->num_active_shader_queries--;

	if (sctx->num_active_shader_queries > 0) {
		gfx10_alloc_query_buffer(sctx);
	} else {
		si_set_rw_shader_buffer(sctx, GFX10_GS_QUERY_BUF, NULL);

		/* If a query_begin is followed by a query_end without a draw
		 * in-between, we need to clear the atom to ensure that the
		 * next query_begin will re-initialize the shader buffer. */
		si_set_atom_dirty(sctx, &sctx->atoms.s.shader_query, false);
	}

	return true;
}

static void gfx10_sh_query_add_result(struct gfx10_sh_query *query,
				      struct gfx10_sh_query_buffer_mem *qmem,
				      union pipe_query_result *result)
{
	static const uint64_t mask = ((uint64_t)1 << 63) - 1;

	switch (query->b.type) {
	case PIPE_QUERY_PRIMITIVES_EMITTED:
		result->u64 += qmem->stream[query->stream].emitted_primitives & mask;
		break;
	case PIPE_QUERY_PRIMITIVES_GENERATED:
		result->u64 += qmem->stream[query->stream].generated_primitives & mask;
		break;
	case PIPE_QUERY_SO_STATISTICS:
		result->so_statistics.num_primitives_written +=
			qmem->stream[query->stream].emitted_primitives & mask;
		result->so_statistics.primitives_storage_needed +=
			qmem->stream[query->stream].generated_primitives & mask;
		break;
	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
		result->b |= qmem->stream[query->stream].emitted_primitives !=
			     qmem->stream[query->stream].generated_primitives;
		break;
	case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
		for (unsigned stream = 0; stream < SI_MAX_STREAMS; ++stream) {
			result->b |= qmem->stream[query->stream].emitted_primitives !=
				     qmem->stream[query->stream].generated_primitives;
		}
		break;
	default:
		assert(0);
	}
}

static bool gfx10_sh_query_get_result(struct si_context *sctx, struct si_query *rquery,
				      bool wait, union pipe_query_result *result)
{
	struct gfx10_sh_query *query = (struct gfx10_sh_query *)rquery;

	util_query_clear_result(result, query->b.type);

	if (unlikely(!query->first))
		return false; /* earlier out of memory error */
	assert(query->last);

	for (struct gfx10_sh_query_buffer *qbuf = query->last;;
	     qbuf = LIST_ENTRY(struct gfx10_sh_query_buffer, qbuf->list.prev, list)) {
		unsigned usage = PIPE_TRANSFER_READ |
				 (wait ? 0 : PIPE_TRANSFER_DONTBLOCK);
		void *map;

		if (rquery->b.flushed)
			map = sctx->ws->buffer_map(qbuf->buf->buf, NULL, usage);
		else
			map = si_buffer_map_sync_with_rings(sctx, qbuf->buf, usage);

		if (!map)
			return false;

		unsigned results_begin = 0;
		unsigned results_end = qbuf->head;
		if (qbuf == query->first)
			results_begin = query->first_begin;
		if (qbuf == query->last)
			results_end = query->last_end;

		while (results_begin != results_end) {
			struct gfx10_sh_query_buffer_mem *qmem = map + results_begin;
			results_begin += sizeof(*qmem);

			gfx10_sh_query_add_result(query, qmem, result);
		}

		if (qbuf == query->first)
			break;
	}

	return true;
}

static void gfx10_sh_query_get_result_resource(struct si_context *sctx,
					       struct si_query *rquery,
					       bool wait,
					       enum pipe_query_value_type result_type,
					       int index,
					       struct pipe_resource *resource,
					       unsigned offset)
{
	struct gfx10_sh_query *query = (struct gfx10_sh_query *)rquery;
	struct si_qbo_state saved_state = {};
	struct pipe_resource *tmp_buffer = NULL;
	unsigned tmp_buffer_offset = 0;

	if (!sctx->sh_query_result_shader) {
		sctx->sh_query_result_shader = gfx10_create_sh_query_result_cs(sctx);
		if (!sctx->sh_query_result_shader)
			return;
	}

	if (query->first != query->last) {
		u_suballocator_alloc(sctx->allocator_zeroed_memory, 16, 16,
				     &tmp_buffer_offset, &tmp_buffer);
		if (!tmp_buffer)
			return;
	}

	si_save_qbo_state(sctx, &saved_state);

	/* Pre-fill the constants configuring the shader behavior. */
	struct {
		uint32_t config;
		uint32_t offset;
		uint32_t chain;
		uint32_t result_count;
	} consts;
	struct pipe_constant_buffer constant_buffer = {};

	if (index >= 0) {
		switch (query->b.type) {
		case PIPE_QUERY_PRIMITIVES_GENERATED:
			consts.offset = sizeof(uint32_t) * query->stream;
			consts.config = 0;
			break;
		case PIPE_QUERY_PRIMITIVES_EMITTED:
			consts.offset = sizeof(uint32_t) * (4 + query->stream);
			consts.config = 0;
			break;
		case PIPE_QUERY_SO_STATISTICS:
			consts.offset = sizeof(uint32_t) * (4 * index + query->stream);
			consts.config = 0;
			break;
		case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
			consts.offset = sizeof(uint32_t) * query->stream;
			consts.config = 2;
			break;
		case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
			consts.offset = 0;
			consts.config = 3;
			break;
		default: unreachable("bad query type");
		}
	} else {
		/* Check result availability. */
		consts.offset = 0;
		consts.config = 1;
	}

	if (result_type == PIPE_QUERY_TYPE_I64 || result_type == PIPE_QUERY_TYPE_U64)
		consts.config |= 8;

	constant_buffer.buffer_size = sizeof(consts);
	constant_buffer.user_buffer = &consts;

	/* Pre-fill the SSBOs and grid. */
	struct pipe_shader_buffer ssbo[3];
	struct pipe_grid_info grid = {};

	ssbo[1].buffer = tmp_buffer;
	ssbo[1].buffer_offset = tmp_buffer_offset;
	ssbo[1].buffer_size = 16;

	ssbo[2] = ssbo[1];

	sctx->b.bind_compute_state(&sctx->b, sctx->sh_query_result_shader);

	grid.block[0] = 1;
	grid.block[1] = 1;
	grid.block[2] = 1;
	grid.grid[0] = 1;
	grid.grid[1] = 1;
	grid.grid[2] = 1;

	struct gfx10_sh_query_buffer *qbuf = query->first;
	for (;;) {
		unsigned begin = qbuf == query->first ? query->first_begin : 0;
		unsigned end = qbuf == query->last ? query->last_end : qbuf->buf->b.b.width0;
		if (!end)
			continue;

		ssbo[0].buffer = &qbuf->buf->b.b;
		ssbo[0].buffer_offset = begin;
		ssbo[0].buffer_size = end - begin;

		consts.result_count = (end - begin) / sizeof(struct gfx10_sh_query_buffer_mem);
		consts.chain = 0;
		if (qbuf != query->first)
			consts.chain |= 1;
		if (qbuf != query->last)
			consts.chain |= 2;

		if (qbuf == query->last) {
			ssbo[2].buffer = resource;
			ssbo[2].buffer_offset = offset;
			ssbo[2].buffer_size = 8;
		}

		sctx->b.set_constant_buffer(&sctx->b, PIPE_SHADER_COMPUTE, 0, &constant_buffer);
		sctx->b.set_shader_buffers(&sctx->b, PIPE_SHADER_COMPUTE, 0, 3, ssbo, 0x6);

		if (wait) {
			uint64_t va;

			/* Wait for result availability. Wait only for readiness
			 * of the last entry, since the fence writes should be
			 * serialized in the CP.
			 */
			va = qbuf->buf->gpu_address;
			va += end - sizeof(struct gfx10_sh_query_buffer_mem);
			va += offsetof(struct gfx10_sh_query_buffer_mem, fence);

			si_cp_wait_mem(sctx, sctx->gfx_cs, va, 0x00000001, 0x00000001, 0);
		}

		sctx->b.launch_grid(&sctx->b, &grid);
		sctx->flags |= SI_CONTEXT_CS_PARTIAL_FLUSH;

		if (qbuf == query->last)
			break;
		qbuf = LIST_ENTRY(struct gfx10_sh_query_buffer, qbuf->list.next, list);
	}

	si_restore_qbo_state(sctx, &saved_state);
	pipe_resource_reference(&tmp_buffer, NULL);
}

static const struct si_query_ops gfx10_sh_query_ops = {
	.destroy = gfx10_sh_query_destroy,
	.begin = gfx10_sh_query_begin,
	.end = gfx10_sh_query_end,
	.get_result = gfx10_sh_query_get_result,
	.get_result_resource = gfx10_sh_query_get_result_resource,
};

struct pipe_query *gfx10_sh_query_create(struct si_screen *screen,
					 enum pipe_query_type query_type,
					 unsigned index)
{
	struct gfx10_sh_query *query = CALLOC_STRUCT(gfx10_sh_query);
	if (unlikely(!query))
		return NULL;

	query->b.ops = &gfx10_sh_query_ops;
	query->b.type = query_type;
	query->stream = index;

	return (struct pipe_query *)query;
}

void gfx10_init_query(struct si_context *sctx)
{
	LIST_INITHEAD(&sctx->shader_query_buffers);
	sctx->atoms.s.shader_query.emit = emit_shader_query;
}

void gfx10_destroy_query(struct si_context *sctx)
{
	while (!LIST_IS_EMPTY(&sctx->shader_query_buffers)) {
		struct gfx10_sh_query_buffer *qbuf =
			list_first_entry(&sctx->shader_query_buffers,
					 struct gfx10_sh_query_buffer, list);
		LIST_DEL(&qbuf->list);

		assert(!qbuf->refcount);
		si_resource_reference(&qbuf->buf, NULL);
		FREE(qbuf);
	}
}