A
Andreas
Hi!
I have a question about a VLIW-processor architecture.
It's quite common that you implement a specific algorithm to a
VLIW-processor, and I have seen papers about how to extract what
functional units that, for this specific algorithm, are suitable to
use in your data path.
But what if you would like to find the m most suitable functional
units for a data path that have to implement more than one algorithm?
This will of cause be dependent on the data dependencies in the
different algorithms and the operation order (for example, in this
algorithms mult is often followed up by an acc and so on).
Some tools, for example A|RT can analyze this form c-code, but this is
before scheduling, so this analysis can just as easy be done by just
looking at the code.
What I'm looking for is a tool (or feature in one of the big tools)
that after scheduling (loop unrolling and so on) can give me data
dependencies and the order of operation that have to be done in
sequence. Or some other help to, in a more scientific way, extract
what functional units that are suitable in a data path that have to
perform n different algorithms.
If anyone has seen any publications in that might help me, please tell
me!
Best regards,
Andreas Lundgren
I have a question about a VLIW-processor architecture.
It's quite common that you implement a specific algorithm to a
VLIW-processor, and I have seen papers about how to extract what
functional units that, for this specific algorithm, are suitable to
use in your data path.
But what if you would like to find the m most suitable functional
units for a data path that have to implement more than one algorithm?
This will of cause be dependent on the data dependencies in the
different algorithms and the operation order (for example, in this
algorithms mult is often followed up by an acc and so on).
Some tools, for example A|RT can analyze this form c-code, but this is
before scheduling, so this analysis can just as easy be done by just
looking at the code.
What I'm looking for is a tool (or feature in one of the big tools)
that after scheduling (loop unrolling and so on) can give me data
dependencies and the order of operation that have to be done in
sequence. Or some other help to, in a more scientific way, extract
what functional units that are suitable in a data path that have to
perform n different algorithms.
If anyone has seen any publications in that might help me, please tell
me!
Best regards,
Andreas Lundgren