Skip to content

Platelets GPU

fomics edited this page Oct 11, 2018 · 2 revisions

Platelets GPU

The model is a time-iterative multi-scale simulation problem, in which we represent 3 minutes of physical time with a time resolution of 0.01 second (thus 30,000 iterations). The code is written in C++. The model has currently two parts: (i) Three 1D, time dependent, partial differential equations describing the transport of platelets towards the deposition surface, and (ii) the deposition surface which is a 2D array with about 500*500 cells, holding the deposited platelets. We could consider a parallelization of this part (domain decomposition). The deposition dynamics requires, at each time steps, and for each cell, to read the flux of platelet from the 1D PDE (scatter operation), then to produce a random number and to accept or reject the deposition at that cell (SIMD parallelism). The probability of deposition depends on the local neighbors of each cells (local, simple grid communications with North, South, West and East neighbors). The rejected platelets are re-injected in the 1D PDEs as a boundary condition (reduction operation).

Clone this wiki locally