Understanding the PyCUDA memory model with matrix manipulation
A PyCUDA program, to make the most of available resources, should respect the rules dictated by the structure and the internal organization of the SM that imposes constraints on the performance of the thread. In particular, the knowledge and correct use of the various types of memory that the GPU makes available is fundamental in order to achieve maximum efficiency in the programs. In the CUDA-capable GPU card, there are four types of memories, which are defined, as follows:
- Registers: In this, a register is allocated for each thread. This can only access its register but not the registers of other threads, even if they belong to the same block.
- The shared memory: Here, each block has ...
Get Python Parallel Programming Cookbook now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
