Anyone could introduce how the opencl memory model is mapping to Intel CPU device and HD Graphics?
Global Memory:
Global/Constant Memory/Data Cache:
Local Memory: this should be SLM (part of L3 cache) in HD graphics
Private Memory: this should be register files in HD graphics.
Thanks!
Link Copied
On HD graphics:
Local memory is mapped to SLM (low latency)
Global/Constant is mapped to Main memory (graphics L3 cache)
Private data is mapped to the graphics register file but in some cases may be mapped to main memory (for large arrays etc.)
Hope this answers your question.
Thanks,
Raghu
For more complete information about compiler optimizations, see our Optimization Notice.