How to combine thrust comparisons using placeholders? - cuda

I have a device_vector of float values A of size N. I also have a float value V for comparison. Depending on an input value I need to extract indices of A for which the values are > OR < OR == V.
I use the following code but it seems cumbersome. Is there a more concise way to do it?
void detect_indices_lesser_greater_equal_to_value(thrust::device_vector<float> S, float value,
int criterion, thrust::device_vector<int>& indices)
{
int N=S.size();
int size=N;
if(criterion==0) // criterion =0 => equal
{
thrust::device_vector<int>::iterator end = thrust::copy_if(thrust::device,thrust::make_counting_iterator(0),
thrust::make_counting_iterator(N),
S.begin(),
indices.begin(),
thrust::placeholders::_1 == value);
size = end-indices.begin();
}
if(criterion==1) // criterion =1 => less
{
thrust::device_vector<int>::iterator end = thrust::copy_if(thrust::device,thrust::make_counting_iterator(0),
thrust::make_counting_iterator(N),
S.begin(),
indices.begin(),
thrust::placeholders::_1 < value);
size = end-indices.begin();
}
if(criterion==2) // criterion =2 => greater
{
thrust::device_vector<int>::iterator end = thrust::copy_if(thrust::device,thrust::make_counting_iterator(0),
thrust::make_counting_iterator(N),
S.begin(),
indices.begin(),
thrust::placeholders::_1 > value);
size = end-indices.begin();
}
indices.resize(size);
}

This can be done with two thrust::partition operations. Partitioning is pretty simple: everything that results in a true predicate is moved to the left side of the input vector. Everything else is moved to the right. Here's a simple example:
$ cat t22.cu
#include <thrust/partition.h>
#include <thrust/copy.h>
#include <thrust/device_vector.h>
typedef float mt;
using namespace thrust::placeholders;
int main(){
const mt pval = 4;
mt data[] = {1,3,7,4,5,2,4,3,9};
const int ds = sizeof(data)/sizeof(data[0]);
thrust::device_vector<mt> d(data, data+ds);
auto end1 = thrust::partition(d.begin(), d.end(), _1<pval);
auto end2 = thrust::partition(end1, d.end(), _1==pval);
std::cout << "less than pval:" << std::endl;
thrust::copy(d.begin(), end1, std::ostream_iterator<mt>(std::cout,","));
std::cout << std::endl << "equal to pval:" << std::endl;
thrust::copy(end1, end2, std::ostream_iterator<mt>(std::cout,","));
std::cout << std::endl << "greater than pval:" << std::endl;
thrust::copy(end2, d.end(), std::ostream_iterator<mt>(std::cout,","));
std::cout << std::endl;
}
$ nvcc -o t22 t22.cu
$ ./t22
less than pval:
1,3,2,3,
equal to pval:
4,4,
greater than pval:
7,5,9,
$
If you require that the ordering in the 3 resultant sub-vectors be the same as the original input ordering, you could use the thrust::stable_partition variant.
(Note that in your question you refer to float quantities, but your example code uses <int> iterators. However the above code can work with either by modifying the typedef).

Related

boost::property_tree::ptree accessing array's first complex element

My JSON is this:
{
"apps":[
{
"id":"x",
"val":"y",
}
]
}
I can get id's value "x" by looping, and exiting when it.first is id:
for (const ptree::value_type &app : root.get_child("apps"))
{
for (const ptree::value_type &it : app.second) {
if (it.first == "id") {
std::cout << it.second.get_value<std::string>().c_str() << std::endl;
}
}
}
What I want, however, is to get the id's value by something like this:
std::cout << root.get<std::string>("apps[0].id").c_str() << std::endl;
Of course this displays nothing to me, for I am probably using wrong syntax accessing 1st element of the apps array. It might be that this has to be done in a different way all together.
I have found only this ugly and dangerous method:
std::cout << root.get_child("apps").begin()->second.begin()->second.get_value<std::string>().c_str() << std::endl;
I cannot really use it this way as it won't throw an exception when the array is empty, it will core dump!
Below is the whole program to make it easier for any one who wants to help:
#include <boost/property_tree/ptree.hpp>
#include <boost/property_tree/json_parser.hpp>
using boost::property_tree::ptree;
int main()
{
std::stringstream ss("{\"apps\":[{\"id\":\"x\",\"val\":\"y\"}]}");
ptree root;
read_json(ss, root);
for (const ptree::value_type &app : root.get_child("apps"))
{
for (const ptree::value_type &it : app.second) {
if (it.first == "id") {
std::cout << it.second.get_value<std::string>().c_str() << std::endl;
}
}
}
std::cout << root.get_child("apps").begin()->second.begin()->second.get_value<std::string>().c_str() << std::endl;
return 0;
}
As the docs say, array elements are nodes with "" keys.
If you're after the first element, you're in luck:
root.get("apps..id", "")
The .. in the path selects the first empty key
Live On Coliru
#include <boost/property_tree/json_parser.hpp>
#include <iostream>
using boost::property_tree::ptree;
int main() {
std::stringstream ss(R"({"apps":[{"id":"x","val":"y"}]})");
ptree root;
read_json(ss, root);
std::cout << root.get("apps..id", "") << "\n";
}
BONUS
If you need to address elements other than the first, write a helper function. This would be a good start:
#include <string>
#include <stdexcept> // std::out_of_range
template <typename Tree>
Tree query(Tree& pt, typename Tree::path_type path) {
if (path.empty())
return pt;
auto const head = path.reduce();
auto subscript = head.find('[');
auto name = head.substr(0, subscript);
auto index = std::string::npos != subscript && head.back() == ']'
? std::stoul(head.substr(subscript+1))
: 0u;
auto matches = pt.equal_range(name);
if (matches.first==matches.second)
throw std::out_of_range("name:" + name);
for (; matches.first != matches.second && index; --index)
++matches.first;
if (index || matches.first==matches.second)
throw std::out_of_range("index:" + head);
return query(matches.first->second, path);
}
Here's some live tests using it:
Live On Coliru
#include <boost/property_tree/json_parser.hpp>
#include <iostream>
using boost::property_tree::ptree;
int main() {
std::stringstream ss(R"({
"apps": [
{
"id": "x",
"val": "y",
"id": "hidden duplicate"
},
{
"id": "a",
"val": "b"
}
]
})");
ptree root;
read_json(ss, root);
for (auto path : {
"apps..id", "apps.[0].id", // (equivalent)
//
"apps.[0].id[]", // invalid
"apps.[0].id[0]", // x
"apps.[0].id[1]", // hidden duplicate
"apps.[1].id", // a
"apps.[1].id[0]", // a
"apps.[1].id[1]", // out of range
"apps.[2].id", // out of range
"drinks" // huh, no drinks at the foo bar
}) try {
std::cout << "Path '" << path << "' -> ";
std::cout << query(root, path).get_value<std::string>() << "\n";
} catch(std::exception const& e) {
std::cout << "Error: " << e.what() << "\n";
}
}
Prints:
Path 'apps..id' -> x
Path 'apps.[0].id' -> x
Path 'apps.[0].id[]' -> Error: stoul
Path 'apps.[0].id[0]' -> x
Path 'apps.[0].id[1]' -> hidden duplicate
Path 'apps.[1].id' -> a
Path 'apps.[1].id[0]' -> a
Path 'apps.[1].id[1]' -> Error: index:id[1]
Path 'apps.[2].id' -> Error: index:[2]
Path 'drinks' -> Error: name:drinks

Cuda Thrust - max vec3

When i want to perform a reduction on an array of float i usually do the following :
float res = *thrust::max_element(thrust::device,
thrust::device_ptr<float>(dDensities),
thrust::device_ptr<float>(dDensities+numParticles)
);
However what i would like to do now is pretty much the same thing on a vec3 (the glm library type) array :
float res = *thrust::max_element(thrust::device,
thrust::device_ptr<glm::vec3>(dDensities),
thrust::device_ptr<glm::vec3>(dDensities+numParticles)
);
As you can see, this has no sense because the '<' operator is not defined on. But i would like to get the maximum vec3 based on his length :
len = sqrtf(v.x*v.x + v.y*v.y + v.z*v.z);
Is that possible ?
Yes, its possible. You may want to read the thrust quickstart guide if you're not already familiar with it.
If you review the thrust extrema documentation, you'll note that thrust::max_element comes in several different varieties (as do most thrust algorithms). One of these accepts a binary comparison functor. We can define a comparison functor which will do what you want.
Here's a trivial worked example:
$ cat t134.cu
#include <thrust/extrema.h>
#include <thrust/device_ptr.h>
#include <glm/glm.hpp>
#include <iostream>
struct comp
{
template <typename T>
__host__ __device__
bool operator()(T &t1, T &t2){
return ((t1.x*t1.x+t1.y*t1.y+t1.z*t1.z) < (t2.x*t2.x+t2.y*t2.y+t2.z*t2.z));
}
};
int main(){
int numParticles = 3;
glm::vec3 d[numParticles];
d[0].x = 0; d[0].y = 0; d[0].z = 0;
d[1].x = 2; d[1].y = 2; d[1].z = 2;
d[2].x = 1; d[2].y = 1; d[2].z = 1;
glm::vec3 *dDensities;
cudaMalloc(&dDensities, numParticles*sizeof(glm::vec3));
cudaMemcpy(dDensities, d, numParticles*sizeof(glm::vec3), cudaMemcpyHostToDevice);
glm::vec3 res = *thrust::max_element(thrust::device,
thrust::device_ptr<glm::vec3>(dDensities),
thrust::device_ptr<glm::vec3>(dDensities+numParticles),
comp()
);
std::cout << "max element x: " << res.x << " y: " << res.y << " z: " << res.z << std::endl;
}
$ nvcc -arch=sm_61 -o t134 t134.cu
$ ./t134
max element x: 2 y: 2 z: 2
$

How can a Eigen matrix be written to file in CSV format?

Suppose I have a double Eigen matrix and I want to write it to a csv file. I find the way of writing into a file in raw format but I need commas between entries. Here is the code I foudn for simple writing.
void writeToCSVfile(string name, MatrixXd matrix)
{
ofstream file(name.c_str());
if (file.is_open())
{
file << matrix << '\n';
//file << "m" << '\n' << colm(matrix) << '\n';
}
}
Using format is a bit more concise:
// define the format you want, you only need one instance of this...
const static IOFormat CSVFormat(StreamPrecision, DontAlignCols, ", ", "\n");
...
void writeToCSVfile(string name, MatrixXd matrix)
{
ofstream file(name.c_str());
file << matrix.format(CSVFormat);
}
Here is what I came up;
void writeToCSVfile(string name, MatrixXd matrix)
{
ofstream file(name.c_str());
for(int i = 0; i < matrix.rows(); i++){
for(int j = 0; j < matrix.cols(); j++){
string str = lexical_cast<std::string>(matrix(i,j));
if(j+1 == matrix.cols()){
file<<str;
}else{
file<<str<<',';
}
}
file<<'\n';
}
}
This is the MWE to the solution given by Partha Lal.
// eigen2csv.cpp
#include <Eigen/Dense>
#include <iostream>
#include <fstream>
// define the format you want, you only need one instance of this...
// see https://eigen.tuxfamily.org/dox/structEigen_1_1IOFormat.html
const static Eigen::IOFormat CSVFormat(Eigen::StreamPrecision, Eigen::DontAlignCols, ", ", "\n");
// writing functions taking Eigen types as parameters,
// see https://eigen.tuxfamily.org/dox/TopicFunctionTakingEigenTypes.html
template <typename Derived>
void writeToCSVfile(std::string name, const Eigen::MatrixBase<Derived>& matrix)
{
std::ofstream file(name.c_str());
file << matrix.format(CSVFormat);
// file.close() is not necessary,
// desctructur closes file, see https://en.cppreference.com/w/cpp/io/basic_ofstream
}
int main()
{
Eigen::MatrixXd vals = Eigen::MatrixXd::Random(10, 3);
writeToCSVfile("test.csv", vals);
}
Compile with g++ eigen2csv.cpp -I<EigenIncludePath>.

odeint streaming observer and related questions

I have a system of 4 coupled equations to solve and a parameter Gamma[i] to iterate over. Since I am quite new to C++, my code is a very rudimentary. If it looks sophisticated and elegant in certain parts, it is only because I have adapted code from the author of odeint. :)
This question is related to (http://stackoverflow.com/questions/12060111/using-odeint-function-definition/12066958#comment16253600_12066958) but not exactly the same. Please do not delete this. :(
Questions have been inserted between the lines of code.
#include <iostream>
#include <iterator>
#include <algorithm>
#include <boost/numeric/odeint.hpp>
#include <cmath>
#include <vector>
#include <fstream>
#include <iomanip>
using namespace std;
using namespace boost::numeric::odeint;
class NLI_class {
private:
double gamma;
public:
NLI_class (double r) : gamma(r) {}
void operator()( vector<double> &u , vector<double> &du , double z ) {
du[0] = u[0]*u[1]*cos(u[3]); //u1
du[1] = -u[0]*u[0]*cos(u[3]); //u2
du[2] = gamma * (2/(u[0]*u[0]) - 1/(u[1]*u[1])); //theta
du[3] = gamma * (1.0/(u[0]*u[0])); //phi1
du[4] = gamma * (1.0/(u[1]*u[1])); //phi2;
}
};
Question #1:
In my original program, I had something like this to pipe the output to a csv file:
inline void save(vector<double>& v, string filename)
{
ofstream output(filename);
for(int i=0;i<v.size();++i){
output << setprecision(64) << v[i] << endl;
}
}
How do I adapt streaming_observer to do what my save() does? Basically, I want to generate .csv files for each iteration i. At this point, I am doing it the ugly way, i.e compiling everything, opening a windows command prompt and then piping the exe output to a text file. This generates one big file with all iterations thrown in there.
This becomes very painful to analyze for a large number of iterations.
struct streaming_observer {
std::ostream &m_out;
streaming_observer( std::ostream &out ) : m_out( out ) {}
void operator()( const vector<double> &x , double t ) const
{
m_out << t;
for( size_t i=0 ; i < x.size() ; ++i )
m_out << "\t" << x[i];
m_out << "\n";
}
};
int main(){
vector<double> x( 5 );
vector<double> Gamma;
vector<double>delta;
const double pi=acos(-1.0);
short delta_n=5;
const double delta_step=(2*pi)/delta_n;
const double dz = 0.01;
const double zeta = 3.0;
const double theta_initial=0.0;
const double u20=tanh(zeta);
const double u10=sqrt(1.0-(u20*u20));
double d=0.0;
double G=0.0;
for(int i=0;i<=delta_n;i++){
//When i=0, the d=0.0 and G=0.0 are pushed into the vector.
delta.push_back(d);
Gamma.push_back(G);
// Compute delta and Gamma
d=d+delta_step;
G=-u10*u10*u20*sin(theta_initial+d);
}
save(delta,"delta.csv");
save(Gamma,"Gamma.csv");
Question#2:
The results I get here do not agree with what I get with what I get using a simple explicit Euler method. Hence, I would like to see the RK4 coefficients (preferably dump them to a file) or the intermediate steps. How can I get this information?
//Numeric Integration
for (unsigned i = 0; i < Gamma.size(); ++i) {
x[0] = u10;
x[1] = u20;
x[2] = 0.0;
x[3] = 0.0;
x[4] = 0.0;
NLI_class nli_obj(Gamma[i]);
integrate_const( runge_kutta4< vector<double > >(), nli_obj, x , 0.0 , 3.0 , dz,streaming_observer( std::cout ) );
}
}
Thank you for all those who helped!
Edit:
Is there some way to get a running error estimate? Note that u[0]*u[0]+u[1]*u[1]=1 at all times.
Question #1 :
I do not understand exactly what kind of output you need. But if you want to write the result after each iteration you can implement an output observer like this:
struct output_observer
{
string filename_;
size_t count_;
output_observer( const string &filename ) : filename_( filename ) , count_( 0 ) { }
void operator()( const state_type &x , time_type dt )
{
char fn[512] = "";
sprintf( fn , "%s_%04lu.csv" , filename_.c_str() , count_ );
ofstream fout( fn );
for( size_t i=0 ; i<x.size() ; ++i ) fout << x[i] << "\n";
++count_;
}
};
You can apply this observer simply by
integrate_const( runge_kutta4< vector<double > >() , nli_obj , x ,
0.0 , 3.0 , dz , output_observer( "filename" ) );
Is this the desired functionality?
Question #2 :
It is not possible to see the intermediate e steps of runge_kutta4. The coefficients are the standard ones for the classical Runge-Kutta method: http://en.wikipedia.org/wiki/Runge%E2%80%93Kutta_methods
Question #3 :
odeint has several error steppers, which estimate the error made during one step. You can use for example the Runge_Kutta Cash Karp algorithm;
runge_kutta_cash_karp54< state_type > rk;
state_type xerr;
rk.do_step( nli_obj , x , t , xerr );
which makes ONE step and estimates the error and writes the error result in xerr.

Getting html attributes from DOM trees ( libxml )

I'm using this program to display a list of all html tags in a given file:
#include <cstdio>
#include <libxml/HTMLparser.h>
#include <libxml/tree.h>
#include <iostream>
#include <cstring>
using namespace std;
static void
print_element_names(htmlNodePtr a_node)
{
htmlNodePtr cur_node = NULL;
for (cur_node = a_node; cur_node!=NULL; cur_node = cur_node->next) {
printf("node type: Element, name: %s\n", cur_node->name);
print_element_names(cur_node->children);
}
}
int main(int argc, char **argv) {
htmlDocPtr doc;
htmlNodePtr root_node;
doc = htmlReadFile(argv[1], NULL, 0);
root_node = xmlDocGetRootElement(doc);
print_element_names(root_node);
xmlFreeDoc(doc);
xmlCleanupParser();
return 0;
}
How do I get it to display the attributes as well (eg. href="something" for <a>)?
It seems there is no such field:
zajec#linux-lbnn:~/Prog_zesp> g++ `xml2-config --cflags --libs` -o tester tester.cpp
tester.cpp: In function ‘void print_element_names(xmlNode*)’:
tester.cpp:17: error: ‘struct _xmlNode’ has no member named ‘attributes’
=== EDIT ===
If I do something like this:
if (strcmp((char *)cur_node->name, "a")==0) {
cout << cur_node->properties->name << endl;
I get the name of the attribute - "href"
If I go one step further:
if (strcmp((char *)cur_node->name, "a")==0) {
cout << cur_node->properties->children->name << endl;
I get "text", but not the actual link.