Cheetah - SKA - PSS - Prototype Time Domain Search Pipeline
|
Some limits and constants for FLDO. More...
Some limits and constants for FLDO.
Convenience header to include all data types used in Fldo.
GPU main kernels.
GPU global device memory declarations.
Some limits and constants for the FLDO cuda based algorithm.
Limits cames from TM-CSP interface definition. The current values are taken from 4/2017 version of 300-000000-021_01_CSPMID document. A table with limits is in doc directory
k_dm is the dispersion constant (e^2/(2 * PI * m_e * c)
Limits cames from current CUDA implementation. A table with limits is in doc directory max_period comes from a pre-bin limitation we intruduced to simplify the code. To be removed.
initial assumptions: Data are read in into main CPU memory. They are divided into 64 subintegrations (blocks). At least one full block is present at the same time on main memory. Data are in unsigned char format (8 bit) Block dimension is (4096 frequency channels) X (up to 187500 sequential measures). This lead to up to 3GB. Too much, we will split either in frequency or in time.
Each thread reads and process a single frequency num_frequency = blockIdx.x and inside GPU frequency number is threadIdx.x data are read and then converted into streams of float (4bytes) in GPU main memory.