search

jonysy / parenchyma

An extensible HPC framework for CUDA, OpenCL and native CPU.
75 stars 4 forks source link

"Lingua franca" #17

Open jonysy opened 7 years ago

jonysy commented 7 years ago

TODO..

Reference Table

OpenCL CUDA Parenchyma Collenchyma
program module bundle binary
?? ?? buffer memory
kernel function function ??
work-group block .. n/a
work-item thread .. n/a
compute unit streaming multiprocessor compute unit (CPU core) compute unit
compute device CUDA-enabled GPU (non-CPU) device hardware
Host CPU CPU CPU??
processing element streaming processor .. n/a
work-group thread block .. n/a
kernel execution instance kernel grid?? .. n/a
context context context device??
command queues CUDA streams channels ??
CL_DEVICE_TYPE ?? device kind hardware type

note:

*A streaming multiprocessor (*CUDA*), or a compute unit (OpenCL*), has 8 streaming processors. OpenCL contexts can have more than one device. CUDA contexts are tied to single device ??. More than one OpenCL command queue can be associated with a single device. More than one CUDA streams can be associated with a single context. Certain GPU devices may not support double precision.

symbols:

symbol meaning
?? not sure
.. TODO