Dense Vectors

cuLite provides specialized classes for managing dense vectors on NVIDIA GPU devices. These classes are optimized for CUDA-accelerated linear algebra operations and support both real and complex numerical types.

Available Dense Vector Types

The following classes are defined within the culite::dns namespace:

cuLite Class	Numeric Type	Precision
culite::dns::RdVector	double	Double precision real
culite::dns::RfVector	float	Single precision real
culite::dns::CdVector	cuDoubleComplex	Double precision complex
culite::dns::CfVector	cuFloatComplex	Single precision complex

Creating vectors

You can instantiate dense vectors using several initialization strategies, ranging from empty containers to pre-allocated device memory.

Default and Sized Creation

Dense vectors can be declared as empty objects or with a predefined size to trigger immediate allocation on the GPU.

Source Code

Console Output

#include <iostream> #include <culite/dense.hpp> int main() { /* * Double precision real empty vector */ culite::dns::RdVector x; std::cout << x.info("x"); /* * Single precision real vector with size 3 (uninitialized values) */ culite::dns::RfVector y(3); std::cout << y.info("y"); /* * Allocate space for x (size 5, uninitialized values) */ x = culite::dns::RdVector(5); std::cout << x.info("x"); return 0; }

==================== x ==================== Datatype............. Real Precision............ Double (64bit) Size................. 0 Values............... 0 Owner................ No =========================================== ==================== y ==================== Datatype............. Real Precision............ Single (32bit) Size................. 3 Values............... 0x504c00000 Owner................ Yes =========================================== ==================== x ==================== Datatype............. Real Precision............ Double (64bit) Size................. 5 Values............... 0x504c00200 Owner................ Yes ===========================================

Create dense vector from aux data

Creates a vector that acts as a "view" or wrapper around an existing C-style array. This avoids a deep copy of the data.

Code

Output

#include <iostream> #include <cla3p/support.hpp> #include <culite/support.hpp> #include <culite/dense.hpp> int main() { /* * Allocate space for x & y on host */ cla3p::real_t *x_host = cla3p::i_calloc_t<cla3p::real_t>(5); cla3p::real_t *y_host = cla3p::i_calloc_t<cla3p::real_t>(3); for(cla3p::uint_t i = 0; i < 5; i++) x_host[i] = i; for(cla3p::uint_t i = 0; i < 3; i++) y_host[i] = 3 - i; /* * Allocate space for x & y on device and copy data from host to device */ culite::real_t *x_device = culite::device_alloc_t<culite::real_t>(5); culite::real_t *y_device = culite::device_alloc_t<culite::real_t>(3); culite::memCopyH2D(5, x_host, x_device); culite::memCopyH2D(3, y_host, y_device); /* * Assign pointer x in vector x but do not bind * x simply hosts x, need to manually dealloc x */ culite::dns::RdVector x(5, x_device, false); std::cout << x.info("x") << x; /* * Assign pointer y in vector y and bind * y takes ownership of y, no free call for y is required */ culite::dns::RdVector y(3, y_device, true); std::cout << y.info("y") << y; /* * Free x and exit */ cla3p::i_free(x_host); cla3p::i_free(y_host); culite::device_free(x_device); // y_device is freed by y's destructor return 0; }

==================== x ==================== Datatype............. Real Precision............ Double (64bit) Size................. 5 Values............... 0x504c00000 Owner................ No =========================================== 0 0 | 0.000000e+00 1 | 1.000000e+00 2 | 2.000000e+00 3 | 3.000000e+00 4 | 4.000000e+00 ==================== y ==================== Datatype............. Real Precision............ Double (64bit) Size................. 3 Values............... 0x504c00200 Owner................ Yes =========================================== 0 0 | 3.000000e+00 1 | 2.000000e+00 2 | 1.000000e+00