Use precomputed gaussian for filter function
This commit is contained in:
parent
50dfa98f51
commit
26df83affc
2 changed files with 111 additions and 42 deletions
|
@ -3,6 +3,7 @@
|
||||||
#include <random>
|
#include <random>
|
||||||
#include <cassert>
|
#include <cassert>
|
||||||
#include <iostream>
|
#include <iostream>
|
||||||
|
#include <memory>
|
||||||
|
|
||||||
#ifndef NDEBUG
|
#ifndef NDEBUG
|
||||||
# include <cstdio>
|
# include <cstdio>
|
||||||
|
@ -58,8 +59,10 @@ std::vector<bool> dither::blue_noise(int width, int height, int threads) {
|
||||||
|
|
||||||
int filter_size = (width + height) / 2;
|
int filter_size = (width + height) / 2;
|
||||||
|
|
||||||
|
std::unique_ptr<std::vector<float>> precomputed = std::make_unique<std::vector<float>>(internal::precompute_gaussian(filter_size));
|
||||||
|
|
||||||
internal::compute_filter(pbp, width, height, count, filter_size,
|
internal::compute_filter(pbp, width, height, count, filter_size,
|
||||||
filter_out, threads);
|
filter_out, precomputed.get(), threads);
|
||||||
internal::write_filter(filter_out, width, "filter_out_start.pgm");
|
internal::write_filter(filter_out, width, "filter_out_start.pgm");
|
||||||
while(true) {
|
while(true) {
|
||||||
//#ifndef NDEBUG
|
//#ifndef NDEBUG
|
||||||
|
@ -69,7 +72,7 @@ std::vector<bool> dither::blue_noise(int width, int height, int threads) {
|
||||||
//#endif
|
//#endif
|
||||||
// get filter values
|
// get filter values
|
||||||
internal::compute_filter(pbp, width, height, count, filter_size,
|
internal::compute_filter(pbp, width, height, count, filter_size,
|
||||||
filter_out, threads);
|
filter_out, precomputed.get(), threads);
|
||||||
|
|
||||||
#ifndef NDEBUG
|
#ifndef NDEBUG
|
||||||
// for(int i = 0; i < count; ++i) {
|
// for(int i = 0; i < count; ++i) {
|
||||||
|
@ -112,7 +115,7 @@ std::vector<bool> dither::blue_noise(int width, int height, int threads) {
|
||||||
|
|
||||||
// get filter values again
|
// get filter values again
|
||||||
internal::compute_filter(pbp, width, height, count, filter_size,
|
internal::compute_filter(pbp, width, height, count, filter_size,
|
||||||
filter_out, threads);
|
filter_out, precomputed.get(), threads);
|
||||||
|
|
||||||
// get second buffer's min
|
// get second buffer's min
|
||||||
int second_min;
|
int second_min;
|
||||||
|
@ -146,7 +149,7 @@ std::vector<bool> dither::blue_noise(int width, int height, int threads) {
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
internal::compute_filter(pbp, width, height, count, filter_size,
|
internal::compute_filter(pbp, width, height, count, filter_size,
|
||||||
filter_out, threads);
|
filter_out, precomputed.get(), threads);
|
||||||
internal::write_filter(filter_out, width, "filter_out_final.pgm");
|
internal::write_filter(filter_out, width, "filter_out_final.pgm");
|
||||||
|
|
||||||
//#ifndef NDEBUG
|
//#ifndef NDEBUG
|
||||||
|
|
|
@ -32,6 +32,19 @@ namespace internal {
|
||||||
return std::exp(-(x*x + y*y)/(2*mu_squared));
|
return std::exp(-(x*x + y*y)/(2*mu_squared));
|
||||||
}
|
}
|
||||||
|
|
||||||
|
inline std::vector<float> precompute_gaussian(int size) {
|
||||||
|
std::vector<float> precomputed;
|
||||||
|
precomputed.reserve(size * size);
|
||||||
|
|
||||||
|
for(int i = 0; i < size * size; ++i) {
|
||||||
|
auto xy = oneToTwo(i, size);
|
||||||
|
precomputed.push_back(gaussian(
|
||||||
|
(float)xy.first - size / 2.0f, (float)xy.second - size / 2.0f));
|
||||||
|
}
|
||||||
|
|
||||||
|
return precomputed;
|
||||||
|
}
|
||||||
|
|
||||||
inline float filter(
|
inline float filter(
|
||||||
const std::vector<bool>& pbp,
|
const std::vector<bool>& pbp,
|
||||||
int x, int y,
|
int x, int y,
|
||||||
|
@ -46,8 +59,7 @@ namespace internal {
|
||||||
int q_prime = (height + filter_size / 2 + y - q) % height;
|
int q_prime = (height + filter_size / 2 + y - q) % height;
|
||||||
for(int p = 0; p < filter_size; ++p) {
|
for(int p = 0; p < filter_size; ++p) {
|
||||||
int p_prime = (width + filter_size / 2 + x - p) % width;
|
int p_prime = (width + filter_size / 2 + x - p) % width;
|
||||||
bool pbp_value = pbp[twoToOne(p_prime, q_prime, width)];
|
if(pbp[twoToOne(p_prime, q_prime, width)]) {
|
||||||
if(pbp_value) {
|
|
||||||
sum += gaussian((float)p - filter_size/2.0f, (float)q - filter_size/2.0f);
|
sum += gaussian((float)p - filter_size/2.0f, (float)q - filter_size/2.0f);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -56,15 +68,46 @@ namespace internal {
|
||||||
return sum;
|
return sum;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
inline float filter_with_precomputed(
|
||||||
|
const std::vector<bool>& pbp,
|
||||||
|
int x, int y,
|
||||||
|
int width, int height, int filter_size,
|
||||||
|
const std::vector<float> &precomputed) {
|
||||||
|
float sum = 0.0f;
|
||||||
|
|
||||||
|
for(int q = 0; q < filter_size; ++q) {
|
||||||
|
int q_prime = (height + filter_size / 2 + y - q) % height;
|
||||||
|
for(int p = 0; p < filter_size; ++p) {
|
||||||
|
int p_prime = (width + filter_size / 2 + x - p) % width;
|
||||||
|
if(pbp[twoToOne(p_prime, q_prime, width)]) {
|
||||||
|
sum += precomputed[twoToOne(p, q, filter_size)];
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
return sum;
|
||||||
|
}
|
||||||
|
|
||||||
inline void compute_filter(
|
inline void compute_filter(
|
||||||
const std::vector<bool> &pbp, int width, int height,
|
const std::vector<bool> &pbp, int width, int height,
|
||||||
int count, int filter_size, std::vector<float> &filter_out,
|
int count, int filter_size, std::vector<float> &filter_out,
|
||||||
|
const std::vector<float> *precomputed = nullptr,
|
||||||
int threads = 1) {
|
int threads = 1) {
|
||||||
if(threads == 1) {
|
if(threads == 1) {
|
||||||
for(int y = 0; y < height; ++y) {
|
if(precomputed) {
|
||||||
for(int x = 0; x < width; ++x) {
|
for(int y = 0; y < height; ++y) {
|
||||||
filter_out[internal::twoToOne(x, y, width)] =
|
for(int x = 0; x < width; ++x) {
|
||||||
internal::filter(pbp, x, y, width, height, filter_size);
|
filter_out[internal::twoToOne(x, y, width)] =
|
||||||
|
internal::filter_with_precomputed(
|
||||||
|
pbp, x, y, width, height, filter_size, *precomputed);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
} else {
|
||||||
|
for(int y = 0; y < height; ++y) {
|
||||||
|
for(int x = 0; x < width; ++x) {
|
||||||
|
filter_out[internal::twoToOne(x, y, width)] =
|
||||||
|
internal::filter(pbp, x, y, width, height, filter_size);
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
|
@ -74,39 +117,62 @@ namespace internal {
|
||||||
int active_count = 0;
|
int active_count = 0;
|
||||||
std::mutex cv_mutex;
|
std::mutex cv_mutex;
|
||||||
std::condition_variable cv;
|
std::condition_variable cv;
|
||||||
for(int i = 0; i < count; ++i) {
|
if(precomputed) {
|
||||||
{
|
for(int i = 0; i < count; ++i) {
|
||||||
std::unique_lock lock(cv_mutex);
|
{
|
||||||
active_count += 1;
|
std::unique_lock lock(cv_mutex);
|
||||||
}
|
active_count += 1;
|
||||||
std::thread t([] (int *ac, std::mutex *cvm,
|
}
|
||||||
std::condition_variable *cv, int i,
|
std::thread t([] (int *ac, std::mutex *cvm,
|
||||||
const std::vector<bool> *pbp, int width,
|
std::condition_variable *cv, int i,
|
||||||
int height, int filter_size,
|
const std::vector<bool> *pbp, int width,
|
||||||
std::vector<float> *fout) {
|
int height, int filter_size,
|
||||||
int x, y;
|
std::vector<float> *fout,
|
||||||
std::tie(x, y) = internal::oneToTwo(i, width);
|
const std::vector<float> *precomputed) {
|
||||||
(*fout)[i] = internal::filter(
|
int x, y;
|
||||||
*pbp, x, y, width, height, filter_size);
|
std::tie(x, y) = internal::oneToTwo(i, width);
|
||||||
std::unique_lock lock(*cvm);
|
(*fout)[i] = internal::filter_with_precomputed(
|
||||||
*ac -= 1;
|
*pbp, x, y, width, height, filter_size, *precomputed);
|
||||||
cv->notify_all();
|
std::unique_lock lock(*cvm);
|
||||||
},
|
*ac -= 1;
|
||||||
&active_count, &cv_mutex, &cv, i, &pbp, width, height,
|
cv->notify_all();
|
||||||
filter_size, &filter_out);
|
},
|
||||||
t.detach();
|
&active_count, &cv_mutex, &cv, i, &pbp, width, height,
|
||||||
|
filter_size, &filter_out, precomputed);
|
||||||
|
t.detach();
|
||||||
|
|
||||||
std::unique_lock lock(cv_mutex);
|
std::unique_lock lock(cv_mutex);
|
||||||
while(active_count >= threads) {
|
while(active_count >= threads) {
|
||||||
#ifndef NDEBUG
|
cv.wait_for(lock, std::chrono::seconds(1));
|
||||||
// std::cout << "0, active_count = " << active_count
|
}
|
||||||
// << ", pre wait_for" << std::endl;
|
}
|
||||||
#endif
|
} else {
|
||||||
cv.wait_for(lock, std::chrono::seconds(1));
|
for(int i = 0; i < count; ++i) {
|
||||||
#ifndef NDEBUG
|
{
|
||||||
// std::cout << "0, active_count = " << active_count
|
std::unique_lock lock(cv_mutex);
|
||||||
// << ", post wait_for" << std::endl;
|
active_count += 1;
|
||||||
#endif
|
}
|
||||||
|
std::thread t([] (int *ac, std::mutex *cvm,
|
||||||
|
std::condition_variable *cv, int i,
|
||||||
|
const std::vector<bool> *pbp, int width,
|
||||||
|
int height, int filter_size,
|
||||||
|
std::vector<float> *fout) {
|
||||||
|
int x, y;
|
||||||
|
std::tie(x, y) = internal::oneToTwo(i, width);
|
||||||
|
(*fout)[i] = internal::filter(
|
||||||
|
*pbp, x, y, width, height, filter_size);
|
||||||
|
std::unique_lock lock(*cvm);
|
||||||
|
*ac -= 1;
|
||||||
|
cv->notify_all();
|
||||||
|
},
|
||||||
|
&active_count, &cv_mutex, &cv, i, &pbp, width, height,
|
||||||
|
filter_size, &filter_out);
|
||||||
|
t.detach();
|
||||||
|
|
||||||
|
std::unique_lock lock(cv_mutex);
|
||||||
|
while(active_count >= threads) {
|
||||||
|
cv.wait_for(lock, std::chrono::seconds(1));
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
std::unique_lock lock(cv_mutex);
|
std::unique_lock lock(cv_mutex);
|
||||||
|
|
Loading…
Reference in a new issue