Fig. 3.

GPU multi-threading structure schema. a Data/operations on GPU global memory. Dotted boxes represent virtual warps. b In GPU block shared memory, each blue shaded box represents one GPU block. Each curved line represents one thread. c Stored data-structure to compute mutual information for each pair of genes. One node in the tree represents un-normalized MI calculated from one quartz after each partitioning, stored in each block shared memory. d Network after collecting all pairwise mutual information