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
|
/* SPDX-License-Identifier: GPL-2.0-or-later
* Copyright 2011 Blender Foundation. */
#include "COM_GaussianYBlurOperation.h"
#include "COM_OpenCLDevice.h"
namespace blender::compositor {
GaussianYBlurOperation::GaussianYBlurOperation() : GaussianBlurBaseOperation(eDimension::Y)
{
}
void *GaussianYBlurOperation::initialize_tile_data(rcti * /*rect*/)
{
lock_mutex();
if (!sizeavailable_) {
update_gauss();
}
void *buffer = get_input_operation(0)->initialize_tile_data(nullptr);
unlock_mutex();
return buffer;
}
void GaussianYBlurOperation::init_execution()
{
GaussianBlurBaseOperation::init_execution();
init_mutex();
if (sizeavailable_ && execution_model_ == eExecutionModel::Tiled) {
float rad = max_ff(size_ * data_.sizey, 0.0f);
filtersize_ = min_ii(ceil(rad), MAX_GAUSSTAB_RADIUS);
gausstab_ = BlurBaseOperation::make_gausstab(rad, filtersize_);
#ifdef BLI_HAVE_SSE2
gausstab_sse_ = BlurBaseOperation::convert_gausstab_sse(gausstab_, filtersize_);
#endif
}
}
void GaussianYBlurOperation::update_gauss()
{
if (gausstab_ == nullptr) {
update_size();
float rad = max_ff(size_ * data_.sizey, 0.0f);
rad = min_ff(rad, MAX_GAUSSTAB_RADIUS);
filtersize_ = min_ii(ceil(rad), MAX_GAUSSTAB_RADIUS);
gausstab_ = BlurBaseOperation::make_gausstab(rad, filtersize_);
#ifdef BLI_HAVE_SSE2
gausstab_sse_ = BlurBaseOperation::convert_gausstab_sse(gausstab_, filtersize_);
#endif
}
}
void GaussianYBlurOperation::execute_pixel(float output[4], int x, int y, void *data)
{
float ATTR_ALIGN(16) color_accum[4] = {0.0f, 0.0f, 0.0f, 0.0f};
float multiplier_accum = 0.0f;
MemoryBuffer *input_buffer = (MemoryBuffer *)data;
const rcti &input_rect = input_buffer->get_rect();
float *buffer = input_buffer->get_buffer();
int bufferwidth = input_buffer->get_width();
int bufferstartx = input_rect.xmin;
int bufferstarty = input_rect.ymin;
int xmin = max_ii(x, input_rect.xmin);
int ymin = max_ii(y - filtersize_, input_rect.ymin);
int ymax = min_ii(y + filtersize_ + 1, input_rect.ymax);
int index;
int step = get_step();
const int buffer_indexx = ((xmin - bufferstartx) * 4);
#ifdef BLI_HAVE_SSE2
__m128 accum_r = _mm_load_ps(color_accum);
for (int ny = ymin; ny < ymax; ny += step) {
index = (ny - y) + filtersize_;
int bufferindex = buffer_indexx + ((ny - bufferstarty) * 4 * bufferwidth);
const float multiplier = gausstab_[index];
__m128 reg_a = _mm_load_ps(&buffer[bufferindex]);
reg_a = _mm_mul_ps(reg_a, gausstab_sse_[index]);
accum_r = _mm_add_ps(accum_r, reg_a);
multiplier_accum += multiplier;
}
_mm_store_ps(color_accum, accum_r);
#else
for (int ny = ymin; ny < ymax; ny += step) {
index = (ny - y) + filtersize_;
int bufferindex = buffer_indexx + ((ny - bufferstarty) * 4 * bufferwidth);
const float multiplier = gausstab_[index];
madd_v4_v4fl(color_accum, &buffer[bufferindex], multiplier);
multiplier_accum += multiplier;
}
#endif
mul_v4_v4fl(output, color_accum, 1.0f / multiplier_accum);
}
void GaussianYBlurOperation::execute_opencl(OpenCLDevice *device,
MemoryBuffer *output_memory_buffer,
cl_mem cl_output_buffer,
MemoryBuffer **input_memory_buffers,
std::list<cl_mem> *cl_mem_to_clean_up,
std::list<cl_kernel> * /*cl_kernels_to_clean_up*/)
{
cl_kernel gaussian_yblur_operation_kernel = device->COM_cl_create_kernel(
"gaussian_yblur_operation_kernel", nullptr);
cl_int filter_size = filtersize_;
cl_mem gausstab = clCreateBuffer(device->get_context(),
CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR,
sizeof(float) * (filtersize_ * 2 + 1),
gausstab_,
nullptr);
device->COM_cl_attach_memory_buffer_to_kernel_parameter(gaussian_yblur_operation_kernel,
0,
1,
cl_mem_to_clean_up,
input_memory_buffers,
input_program_);
device->COM_cl_attach_output_memory_buffer_to_kernel_parameter(
gaussian_yblur_operation_kernel, 2, cl_output_buffer);
device->COM_cl_attach_memory_buffer_offset_to_kernel_parameter(
gaussian_yblur_operation_kernel, 3, output_memory_buffer);
clSetKernelArg(gaussian_yblur_operation_kernel, 4, sizeof(cl_int), &filter_size);
device->COM_cl_attach_size_to_kernel_parameter(gaussian_yblur_operation_kernel, 5, this);
clSetKernelArg(gaussian_yblur_operation_kernel, 6, sizeof(cl_mem), &gausstab);
device->COM_cl_enqueue_range(gaussian_yblur_operation_kernel, output_memory_buffer, 7, this);
clReleaseMemObject(gausstab);
}
void GaussianYBlurOperation::deinit_execution()
{
GaussianBlurBaseOperation::deinit_execution();
if (gausstab_) {
MEM_freeN(gausstab_);
gausstab_ = nullptr;
}
#ifdef BLI_HAVE_SSE2
if (gausstab_sse_) {
MEM_freeN(gausstab_sse_);
gausstab_sse_ = nullptr;
}
#endif
deinit_mutex();
}
bool GaussianYBlurOperation::determine_depending_area_of_interest(
rcti *input, ReadBufferOperation *read_operation, rcti *output)
{
rcti new_input;
if (!sizeavailable_) {
rcti size_input;
size_input.xmin = 0;
size_input.ymin = 0;
size_input.xmax = 5;
size_input.ymax = 5;
NodeOperation *operation = this->get_input_operation(1);
if (operation->determine_depending_area_of_interest(&size_input, read_operation, output)) {
return true;
}
}
{
if (sizeavailable_ && gausstab_ != nullptr) {
new_input.xmax = input->xmax;
new_input.xmin = input->xmin;
new_input.ymax = input->ymax + filtersize_ + 1;
new_input.ymin = input->ymin - filtersize_ - 1;
}
else {
new_input.xmax = this->get_width();
new_input.xmin = 0;
new_input.ymax = this->get_height();
new_input.ymin = 0;
}
return NodeOperation::determine_depending_area_of_interest(&new_input, read_operation, output);
}
}
} // namespace blender::compositor
|
