Properly aligns caching tiles in CUDA and OpenCL diffusion code, fixing bugs that produced partial diffusion for most grid sizes (issues #96, #112, and #118)
Drastically reduces the number of host-to-device data transfers in OpenAcc diffusion code, dropping runtimes from 45× to 4× that of CUDA (issue #49)
Converts static to dynamic allocation of the shared tile cache for CUDA and OpenCL diffusion codes, making the code much more flexible (issue #113)
Renames variables from (col,row) to (x,y), reflecting the physical nature of the work (issue #114)
The documentation has not yet been updated (issue #110).
Details about this contribution:
This PR resolves the following issues:
The documentation has not yet been updated (issue #110).
This change is