Re: DCompute target: Intel to Introduce New CPU-FPGA Hybrid Chip Supported by Acceleration Stack

2017-10-21 Thread Nicholas Wilson via Digitalmars-d-announce

On Friday, 20 October 2017 at 20:41:24 UTC, Nordlöw wrote:
Ever since I first tried programming in VHDL and realized that 
it, at that time, was far too unproductive for my taste, I've 
been waiting for the software and FPGA programming models to 
unite...


What kinds of simplifications (over OpenCL) can and will 
DCompute offer in this regard?


Over OpenCL: Same benefits it does over standard C
* a not completely insane interface
* generality and parameterisation
* reusability
* ...

As to how much of the total power of the FPGA you can use from D 
compared to VHDL remains to be seen, although it will be 
interesting to see how well this can cooperate with Luís' DHDL.




Re: DCompute target: Intel to Introduce New CPU-FPGA Hybrid Chip Supported by Acceleration Stack

2017-10-21 Thread Ola Fosheim Grøstad via Digitalmars-d-announce
But I also get this feeling that Intel do this as an 
anti-competitive monopolistic. Basically preventing ARM and AMD 
from partnering with Altera. So it could be more hostile than 
friendly…


The buy up might not make sense business wise in terms of new 
products. But it could make a lot of sense to keep other 
competing products off the table to keep the pricing of Xeons up… 
A pessimistic view, perhaps… but Intel has a history…




Re: DCompute target: Intel to Introduce New CPU-FPGA Hybrid Chip Supported by Acceleration Stack

2017-10-21 Thread Ola Fosheim Grøstad via Digitalmars-d-announce

On Friday, 20 October 2017 at 20:41:24 UTC, Nordlöw wrote:
Ever since I first tried programming in VHDL and realized that 
it, at that time, was far too unproductive for my taste, I've 
been waiting for the software and FPGA programming models to 
unite...


What kinds of simplifications (over OpenCL) can and will 
DCompute offer in this regard?


I don't know. It is an interesting development, but I've got a 
feeling that you have to address the hardware very specifically 
to get worthwhile performance. My gut feeling is that 
abstractions would be a bad idea…


So maybe this will be best suited for very narrow domains where 
you can rely on third party libraries (e.g. statistical signal 
processing and other fields) or narrow applications that can 
afford to tune very carefully to the underlying hardware.




Re: DCompute target: Intel to Introduce New CPU-FPGA Hybrid Chip Supported by Acceleration Stack

2017-10-20 Thread Nordlöw via Digitalmars-d-announce

On Friday, 20 October 2017 at 20:41:24 UTC, Nordlöw wrote:

https://github.com/ldc-developers/ldc/commits/dcompute


or rather

https://github.com/libmir/dcompute