A
Andrew Hall
Guest
Folks,
For my cryptography course I had to construct an AES core in VHDL. I
duely constructed one and have verified that it works and synthesises
fine.
Later it transpired that one of my classmates had built one with
significantly better throughput; never one to turn down a challenge I
pared down my implementation and made a composite entity with three
cores, a phase-shifted clock divider and a multiplexer on the output.
When I synthesize this (Synplify Pro 7.5.1) and look at the performance
summary I get:
Performance Summary
*******************
Worst slack in design: 955.357
Requested Estimated Requested
Estimated Clock Clock
Starting Clock Frequency Frequency Period
Period Slack Type Group
------------------------------------------------------------------------
-----------------------------------------------------------
AES|clk 1.0 MHz 22.4 MHz 1000.000
44.643 955.357 inferred Inferred_clkgroup_3
AES|clks_inferred_clock[1] 1.0 MHz 71.6 MHz 1000.000
13.965 986.035 inferred Inferred_clkgroup_0
AES|clks_inferred_clock[2] 1.0 MHz 74.2 MHz 1000.000
13.481 986.519 inferred Inferred_clkgroup_2
AES|clks_inferred_clock[3] 1.0 MHz 71.7 MHz 1000.000
13.945 986.055 inferred Inferred_clkgroup_1
========================================================================
===========================================================
I'm slightly confused as to the meaning of this, on its own my AES core
will clock at about 60Mhz. The clock divider puts out three phase
shifted clocks with a 1:2 duty cycle so I can sort of understand the
inferred clocks being able to go a bit faster (longer time to setup
etc).
However, does this really mean that my main clock will only work at
22.4Mhz?? Why so slow? This means that the individual clocks will run at
6Mhz. It's only a simple 3 register clock divider that on its own
synthesises to an expected performance of 150Mhz. All the main clock
does is drive the clock divider and the output multiplexer.
Am I reading this wrong? Is it something to do with clock buffering?
Anybody got any suggestions for how to get my multicore system to run at
a more sensible clock rate? (I'll add at this point that it is 66 layer
pipelined, I have thinned out all the logic I can find, the limiting
factor on the clock speed for the main core is a big feedback loop which
is unavoidable).
Thanks in advance guys,
Andrew.
For my cryptography course I had to construct an AES core in VHDL. I
duely constructed one and have verified that it works and synthesises
fine.
Later it transpired that one of my classmates had built one with
significantly better throughput; never one to turn down a challenge I
pared down my implementation and made a composite entity with three
cores, a phase-shifted clock divider and a multiplexer on the output.
When I synthesize this (Synplify Pro 7.5.1) and look at the performance
summary I get:
Performance Summary
*******************
Worst slack in design: 955.357
Requested Estimated Requested
Estimated Clock Clock
Starting Clock Frequency Frequency Period
Period Slack Type Group
------------------------------------------------------------------------
-----------------------------------------------------------
AES|clk 1.0 MHz 22.4 MHz 1000.000
44.643 955.357 inferred Inferred_clkgroup_3
AES|clks_inferred_clock[1] 1.0 MHz 71.6 MHz 1000.000
13.965 986.035 inferred Inferred_clkgroup_0
AES|clks_inferred_clock[2] 1.0 MHz 74.2 MHz 1000.000
13.481 986.519 inferred Inferred_clkgroup_2
AES|clks_inferred_clock[3] 1.0 MHz 71.7 MHz 1000.000
13.945 986.055 inferred Inferred_clkgroup_1
========================================================================
===========================================================
I'm slightly confused as to the meaning of this, on its own my AES core
will clock at about 60Mhz. The clock divider puts out three phase
shifted clocks with a 1:2 duty cycle so I can sort of understand the
inferred clocks being able to go a bit faster (longer time to setup
etc).
However, does this really mean that my main clock will only work at
22.4Mhz?? Why so slow? This means that the individual clocks will run at
6Mhz. It's only a simple 3 register clock divider that on its own
synthesises to an expected performance of 150Mhz. All the main clock
does is drive the clock divider and the output multiplexer.
Am I reading this wrong? Is it something to do with clock buffering?
Anybody got any suggestions for how to get my multicore system to run at
a more sensible clock rate? (I'll add at this point that it is 66 layer
pipelined, I have thinned out all the logic I can find, the limiting
factor on the clock speed for the main core is a big feedback loop which
is unavoidable).
Thanks in advance guys,
Andrew.