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
|
/*
* Copyright 2021 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#pragma once
CCL_NAMESPACE_BEGIN
/* Given an array of states, build an array of indices for which the states
* are active.
*
* Shared memory requirement is `sizeof(int) * (number_of_warps + 1)`. */
#include "util/atomic.h"
#ifdef __HIP__
# define GPU_PARALLEL_ACTIVE_INDEX_DEFAULT_BLOCK_SIZE 1024
#else
# define GPU_PARALLEL_ACTIVE_INDEX_DEFAULT_BLOCK_SIZE 512
#endif
#ifdef __KERNEL_METAL__
struct ActiveIndexContext {
ActiveIndexContext(int _thread_index,
int _global_index,
int _threadgroup_size,
int _simdgroup_size,
int _simd_lane_index,
int _simd_group_index,
int _num_simd_groups,
threadgroup int *_simdgroup_offset)
: thread_index(_thread_index),
global_index(_global_index),
blocksize(_threadgroup_size),
ccl_gpu_warp_size(_simdgroup_size),
thread_warp(_simd_lane_index),
warp_index(_simd_group_index),
num_warps(_num_simd_groups),
warp_offset(_simdgroup_offset)
{
}
const int thread_index, global_index, blocksize, ccl_gpu_warp_size, thread_warp, warp_index,
num_warps;
threadgroup int *warp_offset;
template<uint blocksizeDummy, typename IsActiveOp>
void active_index_array(const uint num_states,
ccl_global int *indices,
ccl_global int *num_indices,
IsActiveOp is_active_op)
{
const uint state_index = global_index;
#else
template<uint blocksize, typename IsActiveOp>
__device__ void gpu_parallel_active_index_array(const uint num_states,
ccl_global int *indices,
ccl_global int *num_indices,
IsActiveOp is_active_op)
{
extern ccl_gpu_shared int warp_offset[];
const uint thread_index = ccl_gpu_thread_idx_x;
const uint thread_warp = thread_index % ccl_gpu_warp_size;
const uint warp_index = thread_index / ccl_gpu_warp_size;
const uint num_warps = blocksize / ccl_gpu_warp_size;
const uint state_index = ccl_gpu_block_idx_x * blocksize + thread_index;
#endif
/* Test if state corresponding to this thread is active. */
const uint is_active = (state_index < num_states) ? is_active_op(state_index) : 0;
/* For each thread within a warp compute how many other active states precede it. */
const uint thread_offset = popcount(ccl_gpu_ballot(is_active) &
ccl_gpu_thread_mask(thread_warp));
/* Last thread in warp stores number of active states for each warp. */
if (thread_warp == ccl_gpu_warp_size - 1) {
warp_offset[warp_index] = thread_offset + is_active;
}
ccl_gpu_syncthreads();
/* Last thread in block converts per-warp sizes to offsets, increments global size of
* index array and gets offset to write to. */
if (thread_index == blocksize - 1) {
/* TODO: parallelize this. */
int offset = 0;
for (int i = 0; i < num_warps; i++) {
int num_active = warp_offset[i];
warp_offset[i] = offset;
offset += num_active;
}
const uint block_num_active = warp_offset[warp_index] + thread_offset + is_active;
warp_offset[num_warps] = atomic_fetch_and_add_uint32(num_indices, block_num_active);
}
ccl_gpu_syncthreads();
/* Write to index array. */
if (is_active) {
const uint block_offset = warp_offset[num_warps];
indices[block_offset + warp_offset[warp_index] + thread_offset] = state_index;
}
}
#ifdef __KERNEL_METAL__
}; /* end class ActiveIndexContext */
/* inject the required thread params into a struct, and redirect to its templated member function
*/
# define gpu_parallel_active_index_array \
ActiveIndexContext(metal_local_id, \
metal_global_id, \
metal_local_size, \
simdgroup_size, \
simd_lane_index, \
simd_group_index, \
num_simd_groups, \
simdgroup_offset) \
.active_index_array
#endif
CCL_NAMESPACE_END
|
