Transport Triggered Architecture Socket in VHDL

[lyonscf@gmail.com]

I'm trying to figure out the most efficient way of implementing a TTA
socket in VHDL.

I currently have a 24bit control word for each of the function units,
it looks like this:

DST sID SRC sID OPCODE
|----------------------------------------------------|
| 000000000 | 000000000 | 000000 |
|----------------------------------------------------|
| 23 - 15 | 14 - 6 | 5 - 0 |
|----------------------------------------------------|

The socket checks the control word coming along each of the 4 control
buses (as there is 4 data buses) so see if the source or destination
ID is its own and then lets the data on the appropriate bus and opcode
into the function unit.

My current implementation works in two stages, it firstly compares
each of the control inputs (I0,I1,I2,I3) coming in on the control
buses with the FUs sID , this will output a select for which bus
you're taking in from (SO), an enable (EO) and the opcode (OP). The EO
and SO go into a MUX or DEMUX depending on whether its an input or
output socket which selects from/to the data bus.

I have made this too complicated, when synthesizing it comes out for a
Maximum combinatorial path delay of 7.988ns, which is way too
expensive, especially when the point of TTA is to be simple and
therefore fast.

I'll really appreciate if anyone had any comments, suggestions,
constructive criticism, maybe I'm even going about this completely
wrong etc.

Thanks in advance!

- Colin

 

[HT-Lab]

I believe that TTA/Move is not particular FPGA friendly due to the
large/many internal busses. Here is my brief attempt to look at this
fascinating processor architecture, if only I had more time....

http://www.ht-lab.com/freecores/move/move.html

Hans
www.ht-lab.com



<lyonscf@gmail.com> wrote in message
news:1541c68a-6eb3-40ba-a290-06b03f11ae9e@d4g2000prg.googlegroups.com...

I'm trying to figure out the most efficient way of implementing a TTA
socket in VHDL.

I currently have a 24bit control word for each of the function units,
it looks like this:

DST sID SRC sID OPCODE
|----------------------------------------------------|
| 000000000 | 000000000 | 000000 |
|----------------------------------------------------|
| 23 - 15 | 14 - 6 | 5 - 0 |
|----------------------------------------------------|

The socket checks the control word coming along each of the 4 control
buses (as there is 4 data buses) so see if the source or destination
ID is its own and then lets the data on the appropriate bus and opcode
into the function unit.

My current implementation works in two stages, it firstly compares
each of the control inputs (I0,I1,I2,I3) coming in on the control
buses with the FUs sID , this will output a select for which bus
you're taking in from (SO), an enable (EO) and the opcode (OP). The EO
and SO go into a MUX or DEMUX depending on whether its an input or
output socket which selects from/to the data bus.

I have made this too complicated, when synthesizing it comes out for a
Maximum combinatorial path delay of 7.988ns, which is way too
expensive, especially when the point of TTA is to be simple and
therefore fast.

I'll really appreciate if anyone had any comments, suggestions,
constructive criticism, maybe I'm even going about this completely
wrong etc.

Thanks in advance!

- Colin