我有三组点X，Y，Z。我打算使用Eigen :: Matrix4f应用转换。我使用一个zip迭代器和一个转换运算符来做到这一点。程序会编译，但是结果只能部分正确。这篇文章的灵感来自How to modify the contents of a zip iterator
转变 A = [0 1 2; 3 4 5; 6 7 8; M = [1 2 3 4; 5 6 7 8; 9 10 11 12; 13 14 15 16]的M = A结果应为：R = [28 34 40; 68 86 104； 108138168] 但是它给出：R = [28 34 40; 208251294； 2410 2905 3400]。
正在正确修改X值。但是Y和Z值有问题。
我的代码和cmakelist如下：
#include <thrust/iterator/zip_iterator.h>
#include <thrust/execution_policy.h>
#include <thrust/copy.h>
#include <thrust/device_vector.h>
#include <Eigen/Dense>
#include <iostream>
typedef thrust::device_vector<float>::iterator FloatIterator;
typedef thrust::tuple<FloatIterator, FloatIterator, FloatIterator> FloatIteratorTuple;
typedef thrust::zip_iterator<FloatIteratorTuple> Float3Iterator;
typedef thrust::tuple<float,float,float> Float3;
struct modify_tuple
{
Eigen::Matrix4f _Mat4f;
modify_tuple(Eigen::Matrix4f Mat4f) : _Mat4f(Mat4f) { }
__host__ __device__ Float3 operator()(Float3 a) const
{
Eigen::Vector4f V(thrust::get<0>(a), thrust::get<1>(a), thrust::get<2>(a), 1.0);
V=_Mat4f*V;
Float3 res=thrust::make_tuple( V(0,0), V(1,0), V(2,0) );
return res;
}
};
int main(void)
{
thrust::device_vector<float> X(3);
thrust::device_vector<float> Y(3);
thrust::device_vector<float> Z(3);
X[0]=0, X[1]=1, X[2]=2;
Y[0]=4, Y[1]=5, Y[2]=6;
Z[0]=7, Z[1]=8, Z[2]=9;
std::cout << "X,Y,Z before transformation="<< std::endl;
thrust::copy_n(X.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Y.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Z.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
Float3Iterator P_first = thrust::make_zip_iterator(make_tuple(X.begin(), Y.begin(), Z.begin()));
Float3Iterator P_last = thrust::make_zip_iterator(make_tuple(X.end(), Y.end(), Z.end()));
Eigen::Matrix4f M;
M(0,0)= 1; M(0,1)= 2; M(0,2)= 3; M(0,3)= 4;
M(1,0)= 5; M(1,1)= 6; M(1,2)= 7; M(1,3)= 8;
M(2,0)= 9; M(2,1)= 10; M(2,2)= 11; M(2,3)= 12;
M(3,0)= 13; M(3,1)= 14; M(3,2)= 15; M(3,3)= 16;
thrust::transform(thrust::device, P_first,P_last, P_first, modify_tuple(M));
std::cout << "X, Y, Z after transformation="<< std::endl;
thrust::copy_n(X.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Y.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Z.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
return 0;
}
CMakeLists.txt
CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
FIND_PACKAGE(CUDA REQUIRED)
INCLUDE_DIRECTORIES(${CUDA_INCLUDE_DIRS})
INCLUDE_DIRECTORIES (/usr/include/eigen3)
set(
CUDA_NVCC_FLAGS
${CUDA_NVCC_FLAGS};
-O3 -gencode arch=compute_52,code=sm_52;
)
CUDA_ADD_EXECUTABLE(modify_zip_iterator_stackoverflow_ver2 modify_zip_iterator_stackoverflow_ver2.cu)
TARGET_LINK_LIBRARIES(modify_zip_iterator_stackoverflow_ver2 ${CUDA_LIBRARIES})
答案 0 :(得分：1)
可能您只需要获取最新的Eigen。
我在Fedora27上使用了CUDA 9.2，并从here获取了最新的特征。
然后我按如下所示编译并运行了您的代码：
$ cat t21.cu
#include <thrust/iterator/zip_iterator.h>
#include <thrust/execution_policy.h>
#include <thrust/copy.h>
#include <thrust/device_vector.h>
#include <Eigen/Dense>
#include <iostream>
typedef thrust::device_vector<float>::iterator FloatIterator;
typedef thrust::tuple<FloatIterator, FloatIterator, FloatIterator> FloatIteratorTuple;
typedef thrust::zip_iterator<FloatIteratorTuple> Float3Iterator;
typedef thrust::tuple<float,float,float> Float3;
struct modify_tuple
{
Eigen::Matrix4f _Mat4f;
modify_tuple(Eigen::Matrix4f Mat4f) : _Mat4f(Mat4f) { }
__host__ __device__ Float3 operator()(Float3 a) const
{
Eigen::Vector4f V(thrust::get<0>(a), thrust::get<1>(a), thrust::get<2>(a), 1.0);
V=_Mat4f*V;
Float3 res=thrust::make_tuple( V(0,0), V(1,0), V(2,0) );
return res;
}
};
int main(void)
{
thrust::device_vector<float> X(3);
thrust::device_vector<float> Y(3);
thrust::device_vector<float> Z(3);
X[0]=0, X[1]=1, X[2]=2;
Y[0]=4, Y[1]=5, Y[2]=6;
Z[0]=7, Z[1]=8, Z[2]=9;
std::cout << "X,Y,Z before transformation="<< std::endl;
thrust::copy_n(X.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Y.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Z.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
Float3Iterator P_first = thrust::make_zip_iterator(make_tuple(X.begin(), Y.begin(), Z.begin()));
Float3Iterator P_last = thrust::make_zip_iterator(make_tuple(X.end(), Y.end(), Z.end()));
Eigen::Matrix4f M;
M(0,0)= 1; M(0,1)= 2; M(0,2)= 3; M(0,3)= 4;
M(1,0)= 5; M(1,1)= 6; M(1,2)= 7; M(1,3)= 8;
M(2,0)= 9; M(2,1)= 10; M(2,2)= 11; M(2,3)= 12;
M(3,0)= 13; M(3,1)= 14; M(3,2)= 15; M(3,3)= 16;
thrust::transform(thrust::device, P_first,P_last, P_first, modify_tuple(M));
std::cout << "X, Y, Z after transformation="<< std::endl;
thrust::copy_n(X.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Y.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
thrust::copy_n(Z.begin(), 3, std::ostream_iterator<float>(std::cout, ","));
std::cout << std::endl;
return 0;
}
$ nvcc -std=c++11 -I/path/to/eigen/eigen-eigen-71546f1a9f0c t21.cu -o t21 --expt-relaxed-constexpr
$ ./t21
X,Y,Z before transformation=
0,1,2,
4,5,6,
7,8,9,
X, Y, Z after transformation=
33,39,45,
81,99,117,
129,159,189,
$
输出与您在问题中的期望不符，但是您期望的也不正确。
由于取消对zip迭代器的引用而提供给函子的第一个元组为(X[0],Y[0],Z[0])，即(0,4,7)。然后，您的函子将其转换为(0,4,7,1)，并与您的M矩阵进行矩阵向量乘法。第一行内部乘积由0*1+4*2+7*3+1*4给出，其总和为33。第二行内部乘积由0*5+4*6+7*7+1*8给出，其总和为81。第三行内部积由{{1} }，其总和为129。您可以看到此序列33,81,129恰好是上面输出的第一列。
由于取消对zip迭代器的引用而提供给函子的第二个元组将是0*9+4*10+7*11+1*12，即(X[1],Y[1],Z[1])。然后，您的函子将其转换为(1,5,8)，并与您的(1,5,8,1)矩阵进行矩阵向量乘法。第一行内部乘积由M给出，总和为39。第二行内部乘积由1*1+5*2+8*3+1*4给出，总和为99。第三行内部乘积由1*5+5*6+8*7+1*8给出，其中sum是159。您可以看到此序列39,99,159恰好是上面输出的第二列。
我没有对输出的第三列进行相应的算术运算，但我认为这是不对的。
这里是对代码的修改，展示了结果的正确性，并使用Eigen宿主代码进行了算术运算：
1*9+5*10+8*11+1*12