| Age | Commit message (Collapse) | Author |
|---|
|
may come handy if you're brave enough to try this at home.
|
|
updated offset constraint values accordingly.
Another important change is that the core selector is now multicycle. The I/O
logic runs at 45 MHz, while all the cores and the core selector run at 90 MHz.
The job of the core selector is to distribute the write data (STM32 -> FPGA) to
the particular core being addressed and to select the read data (FPGA -> STM32)
from the core being addressed.
Timing issues arising from the distribution of write commands are mitigated by
replication of data and control signals. Each core has its individual set of
chip select, address and write data registers, they can be placed somewhere
inbetween the core itself and the selector and thus the critical timing paths
become twice shorter.
Readback logic has a huge multiplexor that selects the read data from the core
being addressed and then forwards it into the FMC bus arbiter. Since the FMC
arbiter operates at 45 MHz (twice slower, than the readback multiplexor), it
makes sense to give the multiplexor two 90 MHz clock cycles to select the
value, since the arbiter waits one 45 MHz clock cycle before sampling the
readback data. This is achieved by applying FROM-TO constraint. Note the two
gotches here (took me some time to figure out):
1) you attach the TNM or TNM_NET... well, okay, there's the third potential
gotcha here, since only one of the two works with I/O pads, read the CDG for
more details, but luckily, it's not out case, phew. So, you attach a TNM_NET
to a net, and ISE will follow the net and attach the TNM to **the next
register being driven by this net**. This is somewhat non-obvious: say, you
have a flip-flop called 'something_dout<0>' and it's output net is naturally
also called 'something_dout<0>', attaching TNM_NET to this net doesn't
apply the TNM to the flip-flop, it instead follows the net and attaches the
TNM to all the load flip-flops on the net. For FROM-TO to work as expected
we have to apply TNM_NET to all the output "read data" nets of all the
cores, so that they could be traced into the multiplexor. Note, that for a
multicycle multiplexor you actually need two FROM-TO constraints: one for
the data and one for the control signal.
2) Applying FROM-TO affects both the setup and the hold check in ISE. This is
different from Vivado, when you have to individually specify the setup and
the hold checks.
|
|
Okay, here's the story. Xilinx synthesis tool ("XST") is smart in the sense,
that it detects all the registers with equivalent behaviour and then removes
all of them, but one, and connects all loads to this one flip-flop. This works
fine most of the time and usually even saves some resources, but for our
particular design it was starting to cause just too many problems.
The reason is that ModExp* cores exploit the parallel nature of an FPGA,
for example, the ModExpNG instantiates four copies of the modular multiplier
internally. Those multipliers all operate the same way (but on different data,
of course), so all their internal signals such as, say, clock enables and word
counters are the same. XST happily throws away all the internals from three
multipliers, leaves only one instance of control signals and then the map and
place&route tools start struggling for hours fusing this all together. Turning
off equivalent register removal entirely leads to excessive resource
consumption, so the optimal solution would be to selectively turn it off only
for those tricky places where several copies of control signals are actually
required to meet timing. The problem is that according to Xilinx' docs (UG687
v14.5, p. 363) "quivalent_register_removal = no" inline constraint can be
applied to entire modules, not only individual registers, but I was unable to
get this to work, XST seems to just ignore it. This may have been fixed in
Vivado though, haven't tried yet.
Another potential solution is to prepend every register declaration inside the
modular multiplier with this constraint, but that would look just ugly. One
trick I've seen somewhere is to `define a new 'keep_equivalent_reg' "keyword"
to be '"quivalent_register_removal = no" reg' and tweak register declarations
accordingly, that seems to looks somewhat less ugly, don't know.
Yet another way around might be to use the "max_fanout" constraint instead.
Say there're eight DSP slices per multiplier (thirty two DSP slices total since
there're four multiplier instances). In theory we can constrain their clock
enable fanout to not exceed 8. The problem is that XST will first throw away
three of the clock enables, and then gradually add them back to limit each
clock enable fanout to 8. This way there's no guarantee, that the first clock
enable will be routed to all the eight DSP slices in the first multiplier, it
can be routed to DSP slices in the three remaining multipliers as well, since
XST will try to just limit the fanout. It's difficult to predict how the
place&route tools will handle this.
Anyways, the current slice consumption with 2x ModExpA7 and 1x ModExpNG is ~40%,
and the timing situation is very good (the very first phase of place and
route already has zero setup time violations, yay!). With global equivalent
register removal turned on, utilization drops to ~35%, but timing is impossible
to meet even on the highest map and place&route effort setting. I believe the
best way forward is to just keep global removal disabled for now. We may
revisit this in the future, say, if we decide to generate a custom dedicated
RSA-only signer bitstream with as many core instances as possible. Then every
register will count, but I suspect we won't get away with just re-enabling
global equivalent register removal alone, likely some floorplanning will be
required too at least.
|
|
|
|
|
|
|
|
* 45 MHz (aka "io_clk") is the I/O clock for the FMC bus
* 90 MHz (aka "sys_clk") is the system clock for all the cores
* 180 MHz (aka "core_clk") is the high-speed clock for high-performance cores
|
|
|
|
|
|
The original version of this file appears to have been attempting to
do this, but got the grotty details wrong.
|
|
|
|
|
|
|
|
2. Enabled multi-threading for MAP and PAR, the corresponding switch is -mt.
MAP supports -mt off|2, PAR supports -mt off|2|3|4. Please revert back to
-mt off if the build system has only two cores.
|
|
|
|
(which he only committed on fmc_clk, and I was only looking at master).
But I moved the curly brackets from Makefile to xilinx.mk, because
a) Makefile shouldn't need to know the picky details of xst option
syntax, and
b) xst will throw an uninformative error if called with '-vlgincdir '
versus '-vlgincdir {}', if vlgincdir isn't defined in Makefile.
|
|
(which he only committed on fmc_clk, and I was only looking at master).
But I moved the curly brackets from Makefile to xilinx.mk, because
a) Makefile shouldn't need to know the picky details of xst option
syntax, and
b) xst will throw an uninformative error if called with '-vlgincdir '
versus '-vlgincdir {}', if vlgincdir isn't defined in Makefile.
|
|
|
|
|
|
|
|
designed for.
|
|
|
|
The original version of this file appears to have been attempting to
do this, but got the grotty details wrong.
|
|
|
|
|
|
|
|
2. Enabled multi-threading for MAP and PAR, the corresponding switch is -mt.
MAP supports -mt off|2, PAR supports -mt off|2|3|4. Please revert back to
-mt off if the build system has only two cores.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
larger fmc address bus.
|
|
|
|
|
|
|
|
|
|
|
|
but gives Pavel and Fredrik a place to put their stuff.
|
