double perf_accumulate(const compute::vector<T>& data,
const size_t trials,
compute::command_queue& queue)
{
perf_timer t;
for(size_t trial = 0; trial < trials; trial++){
t.start();
compute::accumulate(data.begin(), data.end(), T(0), queue);
queue.finish();
t.stop();
}
return t.min_time();
}
void saxpy(const int num, bool gen = true, int iter = 0)
{
static compute::device gpu;
static compute::context context;
static compute::command_queue queue;
static compute::vector<T> x;
static compute::vector<T> y;
static compute::vector<T> res;
static T alpha = 3.5;
using compute::lambda::_1;
using compute::lambda::_2;
if (gen) {
gpu = compute::system::default_device();
context = compute::context(gpu);
queue = compute::command_queue(context, gpu);
x = compute::vector<T>(num, context);
std::vector<T> h_x(num);
std::generate(h_x.begin(), h_x.end(), rand);
compute::copy(h_x.begin(), h_x.end(), x.begin(), queue);
y = compute::vector<T>(num, context);
std::vector<T> h_y(num);
std::generate(h_y.begin(), h_y.end(), rand);
compute::copy(h_y.begin(), h_y.end(), y.begin(), queue);
res = compute::vector<T>(num, context);
queue.finish();
}
for (int i = 0; i < iter; i++) {
compute::transform(x.begin(), x.end(),
y.begin(), res.begin(),
alpha * _1 + _2,
queue);
}
queue.finish();
}
BOOST_CHECK_CLOSE(float(result[1]), 16.0f, 1e-4f);
BOOST_CHECK_CLOSE(float(result[2]), 81.0f, 1e-4f);
BOOST_CHECK_CLOSE(float(result[3]), 256.0f, 1e-4f);
}
BOOST_AUTO_TEST_CASE(transform_custom_function)
{
float data[] = { 9.0f, 7.0f, 5.0f, 3.0f };
bc::vector<float> vector(data, data + 4, queue);
BOOST_COMPUTE_FUNCTION(float, pow3add4, (float x),
{
return pow(x, 3.0f) + 4.0f;
});
bc::vector<float> result(4, context);
bc::transform(vector.begin(),
vector.end(),
result.begin(),
pow3add4,
queue);
queue.finish();
BOOST_CHECK_CLOSE(float(result[0]), 733.0f, 1e-4f);
BOOST_CHECK_CLOSE(float(result[1]), 347.0f, 1e-4f);
BOOST_CHECK_CLOSE(float(result[2]), 129.0f, 1e-4f);
BOOST_CHECK_CLOSE(float(result[3]), 31.0f, 1e-4f);
}
BOOST_AUTO_TEST_CASE(extract_vector_component)
{
using bc::int2_;
{
using boost::compute::int_;
using boost::compute::float_;
int n = 1024;
std::vector<int_> host_keys(n);
std::vector<float_> host_values(n);
for(int i = 0; i < n; i++){
host_keys[i] = n - i;
host_values[i] = (n - i) / 2.f;
}
BOOST_COMPUTE_FUNCTION(bool, sort_int, (int_ a, int_ b),
{
return a < b;
});
compute::vector<int_> keys(host_keys.begin(), host_keys.end(), queue);
compute::vector<float_> values(host_values.begin(), host_values.end(), queue);
BOOST_CHECK(compute::is_sorted(keys.begin(), keys.end(), sort_int, queue) == false);
BOOST_CHECK(compute::is_sorted(values.begin(), values.end(), queue) == false);
compute::sort_by_key(keys.begin(), keys.end(), values.begin(), sort_int, queue);
BOOST_CHECK(compute::is_sorted(keys.begin(), keys.end(), sort_int, queue) == true);
BOOST_CHECK(compute::is_sorted(values.begin(), values.end(), queue) == true);
}
BOOST_AUTO_TEST_SUITE_END()
#include <boost/compute/types/pair.hpp>
#include "check_macros.hpp"
#include "context_setup.hpp"
namespace compute = boost::compute;
BOOST_AUTO_TEST_CASE(add_three)
{
BOOST_COMPUTE_FUNCTION(int, add_three, (int),
{
return _1 + 3;
});
int data[] = { 1, 2, 3, 4 };
compute::vector<int> vector(data, data + 4, queue);
compute::transform(
vector.begin(), vector.end(), vector.begin(), add_three, queue
);
CHECK_RANGE_EQUAL(int, 4, vector, (4, 5, 6, 7));
}
BOOST_AUTO_TEST_CASE(sum_odd_values)
{
BOOST_COMPUTE_FUNCTION(int, add_odd_value, (int, int),
{
if(_2 & 1){
return _1 + _2;
}
else {
// store the eigen 4x4 matrix as a float16
bcl::float16_ M =
bcl::eigen_matrix4f_to_float16(matrix);
// returns the result of M*x
BOOST_COMPUTE_CLOSURE(Eigen::Vector4f, transform4x4, (const Eigen::Vector4f x), (M),
{
float4 r;
r.x = dot(M.s048c, x);
r.y = dot(M.s159d, x);
r.z = dot(M.s26ae, x);
r.w = dot(M.s37bf, x);
return r;
});
bcl::vector<Eigen::Vector4f> vectors(4, context);
bcl::vector<Eigen::Vector4f> results(4, context);
bcl::copy(host_vectors.begin(), host_vectors.end(), vectors.begin(), queue);
bcl::transform(
vectors.begin(), vectors.end(), results.begin(), transform4x4, queue
);
std::vector<Eigen::Vector4f> host_results(4);
bcl::copy(results.begin(), results.end(), host_results.begin(), queue);
BOOST_CHECK((matrix * host_vectors[0]) == host_results[0]);
BOOST_CHECK((matrix * host_vectors[1]) == host_results[1]);
BOOST_CHECK((matrix * host_vectors[2]) == host_results[2]);
BOOST_CHECK((matrix * host_vectors[3]) == host_results[3]);
BOOST_AUTO_TEST_CASE(sort_custom_struct)
{
// function to compare particles by their x-coordinate
BOOST_COMPUTE_FUNCTION(bool, sort_by_x, (Particle a, Particle b),
{
return a.x < b.x;
});
std::vector<Particle> particles;
particles.push_back(Particle(0.1f, 0.f));
particles.push_back(Particle(-0.4f, 0.f));
particles.push_back(Particle(10.0f, 0.f));
particles.push_back(Particle(0.001f, 0.f));
boost::compute::vector<Particle> vector(4, context);
boost::compute::copy(particles.begin(), particles.end(), vector.begin(), queue);
BOOST_CHECK_EQUAL(vector.size(), size_t(4));
BOOST_CHECK(
boost::compute::is_sorted(vector.begin(), vector.end(),
sort_by_x, queue) == false
);
boost::compute::sort(vector.begin(), vector.end(), sort_by_x, queue);
BOOST_CHECK(
boost::compute::is_sorted(vector.begin(), vector.end(),
sort_by_x, queue) == true
);
boost::compute::copy(vector.begin(), vector.end(), particles.begin(), queue);
BOOST_CHECK_CLOSE(particles[0].x, -0.4f, 0.1);
BOOST_CHECK_CLOSE(particles[1].x, 0.001f, 0.1);
// See http://boostorg.github.com/compute for more information.
//---------------------------------------------------------------------------//
#define BOOST_TEST_MODULE TestTabulate
#include <boost/test/unit_test.hpp>
#include <boost/compute/system.hpp>
#include <boost/compute/function.hpp>
#include <boost/compute/algorithm/copy_n.hpp>
#include <boost/compute/container/vector.hpp>
#include <boost/compute/experimental/tabulate.hpp>
#include "check_macros.hpp"
#include "context_setup.hpp"
namespace compute = boost::compute;
BOOST_AUTO_TEST_CASE(tabulate_negative_int)
{
BOOST_COMPUTE_FUNCTION(int, negate, (int x),
{
return -x;
});
compute::vector<int> vector(10, context);
compute::experimental::tabulate(vector.begin(), vector.end(), negate, queue);
CHECK_RANGE_EQUAL(int, 10, vector, (0, -1, -2, -3, -4, -5, -6, -7, -8, -9));
}
BOOST_AUTO_TEST_SUITE_END()
std::partial_sum(test.begin(), test.end(),
test.begin(), std::multiplies<int>());
BOOST_CHECK_EQUAL_COLLECTIONS(host_vector.begin(), host_vector.end(),
test.begin(), test.end());
}
BOOST_AUTO_TEST_CASE(inclusive_scan_int_custom_function)
{
BOOST_COMPUTE_FUNCTION(int, multi, (int x, int y),
{
return x * y * 2;
});
int data[] = { 1, 2, 1, 2, 3 };
bc::vector<int> vector(data, data + 5, queue);
BOOST_CHECK_EQUAL(vector.size(), size_t(5));
bc::vector<int> result(5, context);
BOOST_CHECK_EQUAL(result.size(), size_t(5));
// inclusive scan
bc::inclusive_scan(vector.begin(), vector.end(), result.begin(),
multi, queue);
CHECK_RANGE_EQUAL(int, 5, result, (1, 4, 8, 32, 192));
// in-place inclusive scan
CHECK_RANGE_EQUAL(int, 5, vector, (1, 2, 1, 2, 3));
bc::inclusive_scan(vector.begin(), vector.end(), vector.begin(),
multi, queue);
CHECK_RANGE_EQUAL(int, 5, vector, (1, 4, 8, 32, 192));
#include "check_macros.hpp"
#include "context_setup.hpp"
namespace compute = boost::compute;
BOOST_AUTO_TEST_CASE(add_two)
{
int two = 2;
BOOST_COMPUTE_CLOSURE(int, add_two, (int x), (two),
{
return x + two;
});
int data[] = { 1, 2, 3, 4 };
compute::vector<int> vector(data, data + 4, queue);
compute::transform(
vector.begin(), vector.end(), vector.begin(), add_two, queue
);
CHECK_RANGE_EQUAL(int, 4, vector, (3, 4, 5, 6));
}
BOOST_AUTO_TEST_CASE(add_two_and_pi)
{
int two = 2;
float pi = 3.14f;
BOOST_COMPUTE_CLOSURE(float, add_two_and_pi, (float x), (two, pi),
{
return x + two + pi;
});
static void resize( boost::compute::vector< T, A > &x , const boost::compute::vector< T, A > &y )
{
x.resize( y.size() );
}
static bool same_size( const boost::compute::vector< T, A > &x , const boost::compute::vector< T, A > &y )
{
return x.size() == y.size();
}
