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
|
/* SPDX-License-Identifier: Apache-2.0
* Copyright 2011-2022 Blender Foundation */
#pragma once
#include "kernel/types.h"
#include "util/atomic.h"
CCL_NAMESPACE_BEGIN
/* Control Flow
*
* Utilities for control flow between kernels. The implementation is different between CPU and
* GPU devices. For the latter part of the logic is handled on the host side with wavefronts.
*
* There is a main path for regular path tracing camera for path tracing. Shadows for next
* event estimation branch off from this into their own path, that may be computed in
* parallel while the main path continues. Additionally, shading kernels are sorted using
* a key for coherence.
*
* Each kernel on the main path must call one of these functions. These may not be called
* multiple times from the same kernel.
*
* integrator_path_init(kg, state, next_kernel)
* integrator_path_next(kg, state, current_kernel, next_kernel)
* integrator_path_terminate(kg, state, current_kernel)
*
* For the shadow path similar functions are used, and again each shadow kernel must call
* one of them, and only once.
*/
ccl_device_forceinline bool integrator_path_is_terminated(ConstIntegratorState state)
{
return INTEGRATOR_STATE(state, path, queued_kernel) == 0;
}
ccl_device_forceinline bool integrator_shadow_path_is_terminated(ConstIntegratorShadowState state)
{
return INTEGRATOR_STATE(state, shadow_path, queued_kernel) == 0;
}
#ifdef __KERNEL_GPU__
ccl_device_forceinline void integrator_path_init(KernelGlobals kg,
IntegratorState state,
const DeviceKernel next_kernel)
{
atomic_fetch_and_add_uint32(&kernel_integrator_state.queue_counter->num_queued[next_kernel], 1);
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
}
ccl_device_forceinline void integrator_path_next(KernelGlobals kg,
IntegratorState state,
const DeviceKernel current_kernel,
const DeviceKernel next_kernel)
{
atomic_fetch_and_sub_uint32(&kernel_integrator_state.queue_counter->num_queued[current_kernel],
1);
atomic_fetch_and_add_uint32(&kernel_integrator_state.queue_counter->num_queued[next_kernel], 1);
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
}
ccl_device_forceinline void integrator_path_terminate(KernelGlobals kg,
IntegratorState state,
const DeviceKernel current_kernel)
{
atomic_fetch_and_sub_uint32(&kernel_integrator_state.queue_counter->num_queued[current_kernel],
1);
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = 0;
}
ccl_device_forceinline IntegratorShadowState integrator_shadow_path_init(
KernelGlobals kg, IntegratorState state, const DeviceKernel next_kernel, const bool is_ao)
{
IntegratorShadowState shadow_state = atomic_fetch_and_add_uint32(
&kernel_integrator_state.next_shadow_path_index[0], 1);
atomic_fetch_and_add_uint32(&kernel_integrator_state.queue_counter->num_queued[next_kernel], 1);
INTEGRATOR_STATE_WRITE(shadow_state, shadow_path, queued_kernel) = next_kernel;
return shadow_state;
}
ccl_device_forceinline void integrator_shadow_path_next(KernelGlobals kg,
IntegratorShadowState state,
const DeviceKernel current_kernel,
const DeviceKernel next_kernel)
{
atomic_fetch_and_sub_uint32(&kernel_integrator_state.queue_counter->num_queued[current_kernel],
1);
atomic_fetch_and_add_uint32(&kernel_integrator_state.queue_counter->num_queued[next_kernel], 1);
INTEGRATOR_STATE_WRITE(state, shadow_path, queued_kernel) = next_kernel;
}
ccl_device_forceinline void integrator_shadow_path_terminate(KernelGlobals kg,
IntegratorShadowState state,
const DeviceKernel current_kernel)
{
atomic_fetch_and_sub_uint32(&kernel_integrator_state.queue_counter->num_queued[current_kernel],
1);
INTEGRATOR_STATE_WRITE(state, shadow_path, queued_kernel) = 0;
}
# ifdef __KERNEL_SORT_PARTITIONING__
/* Sort first by truncated state index (for good locality), then by key (for good coherence). */
# define INTEGRATOR_SORT_KEY(key, state) \
(key + kernel_data.max_shaders * (state / kernel_integrator_state.sort_partition_divisor))
# else
# define INTEGRATOR_SORT_KEY(key, state) (key)
# endif
ccl_device_forceinline void integrator_path_init_sorted(KernelGlobals kg,
IntegratorState state,
const DeviceKernel next_kernel,
const uint32_t key)
{
const int key_ = INTEGRATOR_SORT_KEY(key, state);
atomic_fetch_and_add_uint32(&kernel_integrator_state.queue_counter->num_queued[next_kernel], 1);
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
INTEGRATOR_STATE_WRITE(state, path, shader_sort_key) = key_;
atomic_fetch_and_add_uint32(&kernel_integrator_state.sort_key_counter[next_kernel][key_], 1);
}
ccl_device_forceinline void integrator_path_next_sorted(KernelGlobals kg,
IntegratorState state,
const DeviceKernel current_kernel,
const DeviceKernel next_kernel,
const uint32_t key)
{
const int key_ = INTEGRATOR_SORT_KEY(key, state);
atomic_fetch_and_sub_uint32(&kernel_integrator_state.queue_counter->num_queued[current_kernel],
1);
atomic_fetch_and_add_uint32(&kernel_integrator_state.queue_counter->num_queued[next_kernel], 1);
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
INTEGRATOR_STATE_WRITE(state, path, shader_sort_key) = key_;
atomic_fetch_and_add_uint32(&kernel_integrator_state.sort_key_counter[next_kernel][key_], 1);
}
#else
ccl_device_forceinline void integrator_path_init(KernelGlobals kg,
IntegratorState state,
const DeviceKernel next_kernel)
{
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
}
ccl_device_forceinline void integrator_path_init_sorted(KernelGlobals kg,
IntegratorState state,
const DeviceKernel next_kernel,
const uint32_t key)
{
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
(void)key;
}
ccl_device_forceinline void integrator_path_next(KernelGlobals kg,
IntegratorState state,
const DeviceKernel current_kernel,
const DeviceKernel next_kernel)
{
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
(void)current_kernel;
}
ccl_device_forceinline void integrator_path_terminate(KernelGlobals kg,
IntegratorState state,
const DeviceKernel current_kernel)
{
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = 0;
(void)current_kernel;
}
ccl_device_forceinline void integrator_path_next_sorted(KernelGlobals kg,
IntegratorState state,
const DeviceKernel current_kernel,
const DeviceKernel next_kernel,
const uint32_t key)
{
INTEGRATOR_STATE_WRITE(state, path, queued_kernel) = next_kernel;
(void)key;
(void)current_kernel;
}
ccl_device_forceinline IntegratorShadowState integrator_shadow_path_init(
KernelGlobals kg, IntegratorState state, const DeviceKernel next_kernel, const bool is_ao)
{
IntegratorShadowState shadow_state = (is_ao) ? &state->ao : &state->shadow;
INTEGRATOR_STATE_WRITE(shadow_state, shadow_path, queued_kernel) = next_kernel;
return shadow_state;
}
ccl_device_forceinline void integrator_shadow_path_next(KernelGlobals kg,
IntegratorShadowState state,
const DeviceKernel current_kernel,
const DeviceKernel next_kernel)
{
INTEGRATOR_STATE_WRITE(state, shadow_path, queued_kernel) = next_kernel;
(void)current_kernel;
}
ccl_device_forceinline void integrator_shadow_path_terminate(KernelGlobals kg,
IntegratorShadowState state,
const DeviceKernel current_kernel)
{
INTEGRATOR_STATE_WRITE(state, shadow_path, queued_kernel) = 0;
(void)current_kernel;
}
#endif
CCL_NAMESPACE_END
|