Modifying zip iterator with eigen::Matrix gives errenous results - cuda

Question

I have three set of points X,Y,Z. I intend to apply a transform using Eigen::Matrix4f. I use a zip iterator and a transform operator to do it. The program compiles however the result is only partially correct. this post is inspired by How to modify the contents of a zip iterator
The transformation of
A= [0 1 2;3 4 5;6 7 8; 1 1 1] with M=[1 2 3 4;5 6 7 8;9 10 11 12;13 14 15 16] using M*A should result to: R=[28 34 40; 68 86 104; 108 138 168]
However it gives: R=[28 34 40; 208 251 294; 2410 2905 3400].
The X values are being correctly modified. However the Y & Z values are faulty.
My code and cmakelists is as below:
#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})

Answer 1

Probably you just need to get the latest Eigen.
I used CUDA 9.2 on Fedora27, and grabbed the latest eigen from here.
Then I compiled and ran your code as follows:
$ 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,
$
The output doesn't match what you are expecting in your question, but what you are expecting is not correct either.
The first tuple provided to your functor as a result of dereferencing your zip iterator will be (X[0],Y[0],Z[0]), which is (0,4,7). Your functor then converts that to (0,4,7,1) and does a matrix-vector multiplication with your M matrix. The first row inner product is given by 0*1+4*2+7*3+1*4, which sum is 33. The second row inner product is given by 0*5+4*6+7*7+1*8, which sum is 81. The third row inner product is given by 0*9+4*10+7*11+1*12, which sum is 129. You can see this sequence 33,81,129 is exactly the first column of the output above.
The second tuple provided to your functor as a result of dereferencing your zip iterator will be (X[1],Y[1],Z[1]), which is (1,5,8). Your functor then converts that to (1,5,8,1) and does a matrix-vector multiplication with your M matrix. The first row inner product is given by 1*1+5*2+8*3+1*4 which sum is 39. The second row innner product is given by 1*5+5*6+8*7+1*8 which sum is 99. The third row inner product is given by 1*9+5*10+8*11+1*12 which sum is 159. You can see this sequence 39,99,159 is exactly the second column of the output above.
I haven't done the corresponding arithmetic for the 3rd column of the output, but I don't think it is wrong.
Here's a modification of your code, demonstrating the correctness of the results, doing the arithmetic in Eigen host code:
$ 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;
thrust::host_vector<float> hX = X;
thrust::host_vector<float> hY = Y;
thrust::host_vector<float> hZ = Z;
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;
Eigen::Vector4f hV;
hV(0) = hX[0];
hV(1) = hY[0];
hV(2) = hZ[0];
hV(3) = 1;
hV = M*hV;
std::cout << "column 0:" << std::endl;
std::cout << hV;
std::cout << std::endl;
hV(0) = hX[1];
hV(1) = hY[1];
hV(2) = hZ[1];
hV(3) = 1;
hV = M*hV;
std::cout << "column 1:" << std::endl;
std::cout << hV;
std::cout << std::endl;
hV(0) = hX[2];
hV(1) = hY[2];
hV(2) = hZ[2];
hV(3) = 1;
hV = M*hV;
std::cout << "column 2:" << std::endl;
std::cout << hV;
std::cout << std::endl;
return 0;
}
$ nvcc -std=c++11 -I/home/bob/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,
column 0:
33
81
129
177
column 1:
39
99
159
219
column 2:
45
117
189
261
$