25 const Result &causal_input,
26 const Result &non_causal_input,
37 output.allocate_texture(transposed_domain);
50 const Result &non_causal_input,
55 output.allocate_texture(transposed_domain);
60 float4 filter_output = float4(causal_input.load_pixel<Color>(texel)) +
61 float4(non_causal_input.load_pixel<Color>(texel));
65 output.store_pixel(int2(texel.y, texel.x), Color(filter_output));
82 const Result &causal_input,
83 const Result &non_causal_input,
86 if (context.use_gpu()) {
104 context.cache_manager().deriche_gaussian_coefficients.get(context, sigma);
107 "causal_feedforward_coefficients",
110 "non_causal_feedforward_coefficients",
117 "non_causal_boundary_coefficient",
141 Result &non_causal_output,
145 context.cache_manager().deriche_gaussian_coefficients.get(context, sigma);
149 const float4 non_causal_feedforward_coefficients =
float4(
153 const float non_causal_boundary_coefficient =
float(
174 int y = invocation.y;
179 bool is_causal = invocation.x == 0;
180 float4 feedforward_coefficients = is_causal ? causal_feedforward_coefficients :
181 non_causal_feedforward_coefficients;
182 float boundary_coefficient = is_causal ? causal_boundary_coefficient :
183 non_causal_boundary_coefficient;
188 int2 boundary_texel = is_causal ?
int2(0,
y) :
int2(width - 1,
y);
191 input_boundary, input_boundary, input_boundary, input_boundary, input_boundary};
198 float4 output_boundary = input_boundary * boundary_coefficient;
200 output_boundary, output_boundary, output_boundary, output_boundary, output_boundary};
202 for (
int x = 0;
x < width;
x++) {
211 outputs[0] =
float4(0.0f);
212 int first_input_index = is_causal ? 0 : 1;
214 outputs[0] += feedforward_coefficients[i] * inputs[first_input_index + i];
215 outputs[0] -= feedback_coefficients[i] * outputs[i + 1];
231 inputs[i] = inputs[i - 1];
238 outputs[i] = outputs[i - 1];
249 if (context.use_gpu()) {
250 blur_pass_gpu(context, input, causal_result, non_causal_result, sigma);
253 blur_pass_cpu(context, input, causal_result, non_causal_result, sigma);
272 const bool extend_bounds)
275 "Deriche filter is slower and less accurate than direct convolution for sigma "
276 "values less 3. Use direct convolution blur instead.");
278 "Deriche filter is not accurate nor numerically stable for sigma values larger "
279 "than 32. Use Van Vliet filter instead.");
283 Result padded_input = context.create_result(input.
type());
286 Result horizontal_pass_result = context.create_result(input.
type());
287 blur_pass(context, padded_input, horizontal_pass_result, sigma.x);
290 Result padded_horizontal_pass_result = context.create_result(input.
type());
292 horizontal_pass_result,
293 padded_horizontal_pass_result,
294 int2(padding_size.y, 0),
296 horizontal_pass_result.
release();
299 padded_horizontal_pass_result.
release();
302 Result horizontal_pass_result = context.create_result(input.
type());
303 blur_pass(context, input, horizontal_pass_result, sigma.x);
305 horizontal_pass_result.
release();
#define BLI_assert_msg(a, msg)
double non_causal_boundary_coefficient() const
const double4 & causal_feedforward_coefficients() const
double causal_boundary_coefficient() const
const double4 & feedback_coefficients() const
const double4 & non_causal_feedforward_coefficients() const
Domain transposed() const
void store_pixel(const int2 &texel, const T &pixel_value)
void allocate_texture(const Domain domain, const bool from_pool=true, const std::optional< ResultStorageType > storage_type=std::nullopt)
void unbind_as_texture() const
static ResultType type(gpu::TextureFormat format)
void bind_as_texture(gpu::Shader *shader, const char *texture_name) const
T load_pixel(const int2 &texel) const
void unbind_as_image() const
void bind_as_image(gpu::Shader *shader, const char *image_name, bool read=false) const
static void sum_causal_and_non_causal_results_cpu(const Result &causal_input, const Result &non_causal_input, Result &output)
void compute_dispatch_threads_at_least(gpu::Shader *shader, int2 threads_range, int2 local_size=int2(16))
static void sum_causal_and_non_causal_results(Context &context, const Result &causal_input, const Result &non_causal_input, Result &output)
static float2 compute_radius_from_sigma(const float2 sigma)
static void blur_pass_cpu(Context &context, const Result &input, Result &causal_output, Result &non_causal_output, const float sigma)
ColorSceneLinear4f< eAlpha::Premultiplied > Color
void pad(Context &context, const Result &input, Result &output, const int2 size, const PaddingMethod padding_method)
void parallel_for(const int2 range, const Function &function)
static void blur_pass_gpu(Context &context, const Result &input, Result &causal_result, Result &non_causal_result, const float sigma)
void deriche_gaussian_blur(Context &context, const Result &input, Result &output, const float2 &sigma, const bool extend_bounds)
static void sum_causal_and_non_causal_results_gpu(Context &context, const Result &causal_input, const Result &non_causal_input, Result &output)
static void blur_pass(Context &context, const Result &input, Result &output, const float sigma)
T reduce_max(const VecBase< T, Size > &a)
void GPU_shader_uniform_4fv(gpu::Shader *sh, const char *name, const float data[4])
void GPU_shader_bind(gpu::Shader *shader, const gpu::shader::SpecializationConstants *constants_state=nullptr)
void GPU_shader_uniform_1f(gpu::Shader *sh, const char *name, float value)
KERNEL_FEATURE_LIGHT_PASSES KERNEL_FEATURE_PATH_GUIDING float