1 /*
2 * Copyright © 2021 Raspberry Pi
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "v3d_util.h"
25 #include "util/macros.h"
26
27 /* Choose a number of workgroups per supergroup that maximizes
28 * lane occupancy. We can pack up to 16 workgroups into a supergroup.
29 */
30 uint32_t
v3d_csd_choose_workgroups_per_supergroup(struct v3d_device_info * devinfo,bool has_subgroups,bool has_tsy_barrier,uint32_t threads,uint32_t num_wgs,uint32_t wg_size)31 v3d_csd_choose_workgroups_per_supergroup(struct v3d_device_info *devinfo,
32 bool has_subgroups,
33 bool has_tsy_barrier,
34 uint32_t threads,
35 uint32_t num_wgs,
36 uint32_t wg_size)
37 {
38 /* FIXME: subgroups may restrict supergroup packing. For now, we disable it
39 * completely if the shader uses subgroups.
40 */
41 if (has_subgroups)
42 return 1;
43
44 /* Compute maximum number of batches in a supergroup for this workgroup size.
45 * Each batch is 16 elements, and we can have up to 16 work groups in a
46 * supergroup:
47 *
48 * max_batches_per_sg = (wg_size * max_wgs_per_sg) / elements_per_batch
49 * since max_wgs_per_sg = 16 and elements_per_batch = 16, we get:
50 * max_batches_per_sg = wg_size
51 */
52 uint32_t max_batches_per_sg = wg_size;
53
54 /* QPU threads will stall at TSY barriers until the entire supergroup
55 * reaches the barrier. Limit the supergroup size to half the QPU threads
56 * available, so we can have at least 2 supergroups executing in parallel
57 * and we don't stall all our QPU threads when a supergroup hits a barrier.
58 */
59 if (has_tsy_barrier) {
60 uint32_t max_qpu_threads = devinfo->qpu_count * threads;
61 max_batches_per_sg = MIN2(max_batches_per_sg, max_qpu_threads / 2);
62 }
63 uint32_t max_wgs_per_sg = max_batches_per_sg * 16 / wg_size;
64
65 uint32_t best_wgs_per_sg = 1;
66 uint32_t best_unused_lanes = 16;
67 for (uint32_t wgs_per_sg = 1; wgs_per_sg <= max_wgs_per_sg; wgs_per_sg++) {
68 /* Don't try to pack more workgroups per supergroup than the total amount
69 * of workgroups dispatched.
70 */
71 if (wgs_per_sg > num_wgs)
72 return best_wgs_per_sg;
73
74 /* Compute wasted lines for this configuration and keep track of the
75 * config with less waste.
76 */
77 uint32_t unused_lanes = (16 - ((wgs_per_sg * wg_size) % 16)) & 0x0f;
78 if (unused_lanes == 0)
79 return wgs_per_sg;
80
81 if (unused_lanes < best_unused_lanes) {
82 best_wgs_per_sg = wgs_per_sg;
83 best_unused_lanes = unused_lanes;
84 }
85 }
86
87 return best_wgs_per_sg;
88 }
89