Blender  V3.3
pass_accessor_gpu.cpp
Go to the documentation of this file.
1 /* SPDX-License-Identifier: Apache-2.0
2  * Copyright 2011-2022 Blender Foundation */
3 
5 
6 #include "device/queue.h"
7 #include "session/buffers.h"
8 #include "util/log.h"
9 
11 
13  const PassAccessInfo &pass_access_info,
14  float exposure,
15  int num_samples)
16  : PassAccessor(pass_access_info, exposure, num_samples), queue_(queue)
17 
18 {
19 }
20 
21 /* --------------------------------------------------------------------
22  * Kernel execution.
23  */
24 
26  const RenderBuffers *render_buffers,
27  const BufferParams &buffer_params,
28  const Destination &destination) const
29 {
30  KernelFilmConvert kfilm_convert;
31  init_kernel_film_convert(&kfilm_convert, buffer_params, destination);
32 
33  const int work_size = buffer_params.window_width * buffer_params.window_height;
34 
35  const int destination_stride = destination.stride != 0 ? destination.stride :
36  buffer_params.window_width;
37 
38  const int offset = buffer_params.window_x * buffer_params.pass_stride +
39  buffer_params.window_y * buffer_params.stride * buffer_params.pass_stride;
40 
41  if (destination.d_pixels) {
42  DCHECK_EQ(destination.stride, 0) << "Custom stride for float destination is not implemented.";
43 
44  DeviceKernelArguments args(&kfilm_convert,
45  &destination.d_pixels,
46  &render_buffers->buffer.device_pointer,
47  &work_size,
48  &buffer_params.window_width,
49  &offset,
50  &buffer_params.stride,
51  &destination.offset,
52  &destination_stride);
53 
54  queue_->enqueue(kernel, work_size, args);
55  }
56  if (destination.d_pixels_half_rgba) {
57  const DeviceKernel kernel_half_float = static_cast<DeviceKernel>(kernel + 1);
58 
59  DeviceKernelArguments args(&kfilm_convert,
60  &destination.d_pixels_half_rgba,
61  &render_buffers->buffer.device_pointer,
62  &work_size,
63  &buffer_params.window_width,
64  &offset,
65  &buffer_params.stride,
66  &destination.offset,
67  &destination_stride);
68 
69  queue_->enqueue(kernel_half_float, work_size, args);
70  }
71 
73 }
74 
75 /* --------------------------------------------------------------------
76  * Pass accessors.
77  */
78 
79 #define DEFINE_PASS_ACCESSOR(pass, kernel_pass) \
80  void PassAccessorGPU::get_pass_##pass(const RenderBuffers *render_buffers, \
81  const BufferParams &buffer_params, \
82  const Destination &destination) const \
83  { \
84  run_film_convert_kernels( \
85  DEVICE_KERNEL_FILM_CONVERT_##kernel_pass, render_buffers, buffer_params, destination); \
86  }
87 
88 /* Float (scalar) passes. */
89 DEFINE_PASS_ACCESSOR(depth, DEPTH);
91 DEFINE_PASS_ACCESSOR(sample_count, SAMPLE_COUNT);
93 
94 /* Float3 passes. */
95 DEFINE_PASS_ACCESSOR(light_path, LIGHT_PATH);
97 
98 /* Float4 passes. */
99 DEFINE_PASS_ACCESSOR(motion, MOTION);
100 DEFINE_PASS_ACCESSOR(cryptomatte, CRYPTOMATTE);
101 DEFINE_PASS_ACCESSOR(shadow_catcher, SHADOW_CATCHER);
102 DEFINE_PASS_ACCESSOR(shadow_catcher_matte_with_shadow, SHADOW_CATCHER_MATTE_WITH_SHADOW);
103 DEFINE_PASS_ACCESSOR(combined, COMBINED);
105 
106 #undef DEFINE_PASS_ACCESSOR
107 
float float4[4]
int pass_stride
Definition: buffers.h:93
int stride
Definition: buffers.h:90
int window_y
Definition: buffers.h:79
int window_height
Definition: buffers.h:81
int window_width
Definition: buffers.h:80
int window_x
Definition: buffers.h:78
virtual bool synchronize()=0
virtual bool enqueue(DeviceKernel kernel, const int work_size, DeviceKernelArguments const &args)=0
DeviceQueue * queue_
void run_film_convert_kernels(DeviceKernel kernel, const RenderBuffers *render_buffers, const BufferParams &buffer_params, const Destination &destination) const
PassAccessorGPU(DeviceQueue *queue, const PassAccessInfo &pass_access_info, float exposure, int num_samples)
virtual void init_kernel_film_convert(KernelFilmConvert *kfilm_convert, const BufferParams &buffer_params, const Destination &destination) const
device_vector< float > buffer
Definition: buffers.h:159
device_ptr device_pointer
#define CCL_NAMESPACE_END
Definition: cuda/compat.h:9
SyclQueue * queue
SyclQueue void void size_t num_bytes SyclQueue void const char void *memory_device_pointer KernelContext int kernel
ccl_gpu_kernel_postfix ccl_global const int ccl_global float const int work_size
ccl_gpu_kernel_postfix ccl_global float int int int int float bool int offset
ccl_gpu_kernel_postfix ccl_global float int int int int ccl_global const float int int int int int int int int int int int int num_samples
DeviceKernel
#define DCHECK_EQ(a, b)
Definition: log.h:64
#define DEFINE_PASS_ACCESSOR(pass, kernel_pass)
@ FLOAT4
@ FLOAT3
@ FLOAT