[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [f-cpu] vhdl code for R7


Etienne LABARRE wrote:
> Hi all,
> Excuse my poor english, it's not my first language (i'm french...)

it's just a matter of ... practice ? :-)

> I'm currently coding the register set (R7 unit) for the f-cpu. I sent
> here my first version (with some YG remarks), and i whould like you to say your opinion.

so here we are back to the problem of the physical mapping of functions by a
synthesizer. As far as i remember, Etienne used a different approach from mine,
using multiplexors instead of my behavioural code with direct indices.
Who knows. One of them will maybe work ? :-)

> It works with Simili and Vanilla, but it's not perfect. It's only a
> valid model for simulation, and i think that we can perfect it.
we'll do it.

> For example, the 5-blocks organisation of R7 is
>    bit 00 to 07 = block 0 = 8 bits
>    bit 08 to 15 = block 1 = 8 bits
>    bit 16 to 31 = block 2 = 16 bits
>    bit 32 to 47 = block 3 = 16 bits
>    bit 48 to 63 = block 4 = 16 bits
you got it right.

> This organisation has no signification for me today if
> the registers aren't 64 bits wide. What's about if 128 bits ? add 4
> 16-bits blocks, or increase block width ?

We add 16-bit blocks up to 256 bits because
the loadimm instruction can manage this size : there is a 4-bit
"address" and a 16-bit immediate data so the maximum width
allowed by the instruction is 16*2^4=256. Above that, either
use constants in memory (a pointer requires 32 or 48 bits
in practice, so you can spare some size if you load a pointer
to the large data).

Above 256 bits, there is no constant but "wide" and "SIMD" operations
can take place, such as load/stores etc. so a 512-bit register has
a wide 256-bit part plus 15x16-bit "slices" and 2x8-bit.
And so on (a 1024-bit register adds a 512-bit wide slice).
But i doubt we need to care because 256-bits is already a challenge.
We can already put a ASSERT SIZE <= 256 in the early versions.

> I'm not in f-cpu team for long time, and i don't understand all the
> subject...
don't worry...

ok, now i have to integrate your work in my source tree...

> Etienne LABARRE
WHYGEE who promised to burn a F-CDROM to Etienne ...
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/