Subject: Re: [boost] [SORT] Parallel Algorithms
From: Francisco JosÃ© Tapia (fjtapia_at_[hidden])
Date: 2015-04-06 12:52:50
About to change the name and use intarray, it's ok , I am changing now. I
was looking the code of randomgen and I didn't understood how is running.
With the explanation about the operator >>, I understand and I will use in
the test implementation.
About Android, I recommend you to see the Shield consoles from Nvidia. The
Sw inside these consoles is comparable with any Linux workstation.
But, perhaps we have not the correct perspective about this. We don't
pretend to say to the programmers what must use, and what not. We pretend
provide tools, in order to simplify their work, and use or not is their
In the actual implementation of parallel sort algorithms in TBB or with
OpenMP, the way for to select only a part of the HW Threads is not clear.
Recently Ezchip announces a processor with 100 ARM cores. Our
implementation have a parameter for to select the number of HW threads to
use, by default use all. But perhaps want to use only 40 HW threads, or a
percent of the HW threads of the machine where is running the program.
In the future, we must involve the GPU, but, in my opinion, first, we
must obtain fast and robust parallel algorithms over the CPU.