aboutsummaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/freedreno/a4xx/fd4_query.c
blob: 3ae3971dc5fb74ac08c2563236911fde07d9bc22 (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
/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */

/*
 * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
 *
 * 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
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS 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:
 *    Rob Clark <robclark@freedesktop.org>
 */

#include "freedreno_query_hw.h"
#include "freedreno_context.h"
#include "freedreno_util.h"

#include "fd4_query.h"
#include "fd4_context.h"
#include "fd4_draw.h"
#include "fd4_format.h"


struct fd_rb_samp_ctrs {
	uint64_t ctr[16];
};

/*
 * Occlusion Query:
 *
 * OCCLUSION_COUNTER and OCCLUSION_PREDICATE differ only in how they
 * interpret results
 */

static struct fd_hw_sample *
occlusion_get_sample(struct fd_batch *batch, struct fd_ringbuffer *ring)
{
	struct fd_hw_sample *samp =
			fd_hw_sample_init(batch, sizeof(struct fd_rb_samp_ctrs));

	/* low bits of sample addr should be zero (since they are control
	 * flags in RB_SAMPLE_COUNT_CONTROL):
	 */
	debug_assert((samp->offset & 0x3) == 0);

	/* Set RB_SAMPLE_COUNT_ADDR to samp->offset plus value of
	 * HW_QUERY_BASE_REG register:
	 */
	OUT_PKT3(ring, CP_SET_CONSTANT, 3);
	OUT_RING(ring, CP_REG(REG_A4XX_RB_SAMPLE_COUNT_CONTROL) | 0x80000000);
	OUT_RING(ring, HW_QUERY_BASE_REG);
	OUT_RING(ring, A4XX_RB_SAMPLE_COUNT_CONTROL_COPY |
			samp->offset);

	OUT_PKT3(ring, CP_DRAW_INDX_OFFSET, 3);
	OUT_RING(ring, DRAW4(DI_PT_POINTLIST_PSIZE, DI_SRC_SEL_AUTO_INDEX,
						INDEX4_SIZE_32_BIT, USE_VISIBILITY));
	OUT_RING(ring, 1);             /* NumInstances */
	OUT_RING(ring, 0);             /* NumIndices */

	fd_event_write(batch, ring, ZPASS_DONE);

	return samp;
}

static uint64_t
count_samples(const struct fd_rb_samp_ctrs *start,
		const struct fd_rb_samp_ctrs *end)
{
	return end->ctr[0] - start->ctr[0];
}

static void
occlusion_counter_accumulate_result(struct fd_context *ctx,
		const void *start, const void *end,
		union pipe_query_result *result)
{
	uint64_t n = count_samples(start, end);
	result->u64 += n;
}

static void
occlusion_predicate_accumulate_result(struct fd_context *ctx,
		const void *start, const void *end,
		union pipe_query_result *result)
{
	uint64_t n = count_samples(start, end);
	result->b |= (n > 0);
}

/*
 * Time Elapsed Query:
 *
 * Note: we could in theory support timestamp queries, but they
 * won't give sensible results for tilers.
 */

static void
time_elapsed_enable(struct fd_context *ctx, struct fd_ringbuffer *ring)
{
	/* Right now, the assignment of countable to counter register is
	 * just hard coded.  If we start exposing more countables than we
	 * have counters, we will need to be more clever.
	 */
	fd_wfi(ctx->batch, ring);
	OUT_PKT0(ring, REG_A4XX_CP_PERFCTR_CP_SEL_0, 1);
	OUT_RING(ring, CP_ALWAYS_COUNT);
}

static struct fd_hw_sample *
time_elapsed_get_sample(struct fd_batch *batch, struct fd_ringbuffer *ring)
{
	struct fd_hw_sample *samp = fd_hw_sample_init(batch, sizeof(uint64_t));

	/* use unused part of vsc_size_mem as scratch space, to avoid
	 * extra allocation:
	 */
	struct fd_bo *scratch_bo = fd4_context(batch->ctx)->vsc_size_mem;
	const int sample_off = 128;
	const int addr_off = sample_off + 8;

	debug_assert(batch->ctx->screen->max_freq > 0);

	/* Basic issue is that we need to read counter value to a relative
	 * destination (with per-tile offset) rather than absolute dest
	 * addr.  But there is no pm4 packet that can do that.  This is
	 * where it would be *really* nice if we could write our own fw
	 * since afaict implementing the sort of packet we need would be
	 * trivial.
	 *
	 * Instead, we:
	 * (1) CP_REG_TO_MEM to do a 64b copy of counter to scratch buffer
	 * (2) CP_MEM_WRITE to write per-sample offset to scratch buffer
	 * (3) CP_REG_TO_MEM w/ accumulate flag to add the per-tile base
	 *     address to the per-sample offset in the scratch buffer
	 * (4) CP_MEM_TO_REG to copy resulting address from steps #2 and #3
	 *     to CP_ME_NRT_ADDR
	 * (5) CP_MEM_TO_REG's to copy saved counter value from scratch
	 *     buffer to CP_ME_NRT_DATA to trigger the write out to query
	 *     result buffer
	 *
	 * Straightforward, right?
	 *
	 * Maybe could swap the order of things in the scratch buffer to
	 * put address first, and copy back to CP_ME_NRT_ADDR+DATA in one
	 * shot, but that's really just polishing a turd..
	 */

	fd_wfi(batch, ring);

	/* copy sample counter _LO and _HI to scratch: */
	OUT_PKT3(ring, CP_REG_TO_MEM, 2);
	OUT_RING(ring, CP_REG_TO_MEM_0_REG(REG_A4XX_RBBM_PERFCTR_CP_0_LO) |
			CP_REG_TO_MEM_0_64B |
			CP_REG_TO_MEM_0_CNT(2-1)); /* write 2 regs to mem */
	OUT_RELOCW(ring, scratch_bo, sample_off, 0, 0);

	/* ok... here we really *would* like to use the CP_SET_CONSTANT
	 * mode which can add a constant to value in reg2 and write to
	 * reg1... *but* that only works for banked/context registers,
	 * and CP_ME_NRT_DATA isn't one of those.. so we need to do some
	 * CP math to the scratch buffer instead:
	 *
	 * (note first 8 bytes are counter value, use offset 0x8 for
	 * address calculation)
	 */

	/* per-sample offset to scratch bo: */
	OUT_PKT3(ring, CP_MEM_WRITE, 2);
	OUT_RELOCW(ring, scratch_bo, addr_off, 0, 0);
	OUT_RING(ring, samp->offset);

	/* now add to that the per-tile base: */
	OUT_PKT3(ring, CP_REG_TO_MEM, 2);
	OUT_RING(ring, CP_REG_TO_MEM_0_REG(HW_QUERY_BASE_REG) |
			CP_REG_TO_MEM_0_ACCUMULATE |
			CP_REG_TO_MEM_0_CNT(1-1));       /* readback 1 regs */
	OUT_RELOCW(ring, scratch_bo, addr_off, 0, 0);

	/* now copy that back to CP_ME_NRT_ADDR: */
	OUT_PKT3(ring, CP_MEM_TO_REG, 2);
	OUT_RING(ring, REG_A4XX_CP_ME_NRT_ADDR);
	OUT_RELOC(ring, scratch_bo, addr_off, 0, 0);

	/* and finally, copy sample from scratch buffer to CP_ME_NRT_DATA
	 * to trigger the write to result buffer
	 */
	OUT_PKT3(ring, CP_MEM_TO_REG, 2);
	OUT_RING(ring, REG_A4XX_CP_ME_NRT_DATA);
	OUT_RELOC(ring, scratch_bo, sample_off, 0, 0);

	/* and again to get the value of the _HI reg from scratch: */
	OUT_PKT3(ring, CP_MEM_TO_REG, 2);
	OUT_RING(ring, REG_A4XX_CP_ME_NRT_DATA);
	OUT_RELOC(ring, scratch_bo, sample_off + 0x4, 0, 0);

	/* Sigh.. */

	return samp;
}

static void
time_elapsed_accumulate_result(struct fd_context *ctx,
		const void *start, const void *end,
		union pipe_query_result *result)
{
	uint64_t n = *(uint64_t *)end - *(uint64_t *)start;
	/* max_freq is in Hz, convert cycle count to ns: */
	result->u64 += n * 1000000000 / ctx->screen->max_freq;
}

static void
timestamp_accumulate_result(struct fd_context *ctx,
		const void *start, const void *end,
		union pipe_query_result *result)
{
	/* just return the value from fist tile: */
	if (result->u64 != 0)
		return;
	uint64_t n = *(uint64_t *)start;
	/* max_freq is in Hz, convert cycle count to ns: */
	result->u64 = n * 1000000000 / ctx->screen->max_freq;
}

static const struct fd_hw_sample_provider occlusion_counter = {
		.query_type = PIPE_QUERY_OCCLUSION_COUNTER,
		.active = FD_STAGE_DRAW,
		.get_sample = occlusion_get_sample,
		.accumulate_result = occlusion_counter_accumulate_result,
};

static const struct fd_hw_sample_provider occlusion_predicate = {
		.query_type = PIPE_QUERY_OCCLUSION_PREDICATE,
		.active = FD_STAGE_DRAW,
		.get_sample = occlusion_get_sample,
		.accumulate_result = occlusion_predicate_accumulate_result,
};

static const struct fd_hw_sample_provider time_elapsed = {
		.query_type = PIPE_QUERY_TIME_ELAPSED,
		.active = FD_STAGE_DRAW | FD_STAGE_CLEAR,
		.enable = time_elapsed_enable,
		.get_sample = time_elapsed_get_sample,
		.accumulate_result = time_elapsed_accumulate_result,
};

/* NOTE: timestamp query isn't going to give terribly sensible results
 * on a tiler.  But it is needed by qapitrace profile heatmap.  If you
 * add in a binning pass, the results get even more non-sensical.  So
 * we just return the timestamp on the first tile and hope that is
 * kind of good enough.
 */
static const struct fd_hw_sample_provider timestamp = {
		.query_type = PIPE_QUERY_TIMESTAMP,
		.active = FD_STAGE_ALL,
		.enable = time_elapsed_enable,
		.get_sample = time_elapsed_get_sample,
		.accumulate_result = timestamp_accumulate_result,
};

void fd4_query_context_init(struct pipe_context *pctx)
{
	fd_hw_query_register_provider(pctx, &occlusion_counter);
	fd_hw_query_register_provider(pctx, &occlusion_predicate);
	fd_hw_query_register_provider(pctx, &time_elapsed);
	fd_hw_query_register_provider(pctx, &timestamp);
}