[libre-riscv-dev] [hw-dev] Re: 6600-style out-of-order scoreboard designs (ariane)

[Luke Kenneth Casson Leighton]

hiya mitch, just a quick update on progress.

in my random-wandering modifications i noticed that when the read
hazards are entirely clear (which can be when each FU is idle) the big
NOR readable signal(s) can be *true*.  this then causes all of the
read priority pickers to go into permanent-selection of the highest
priority FU, even if it's idle.

does the addition to the Computation Unit of a Read_Release_Request
signal look about right to you?  it's mirrored on the write release
request (except without the timing delay).

secondly, i finally figured out the significance of the major-minor
clock cycles on the FU-Regs Dependency Cell.  holy cow is it elegant,
efficient, and scary at the same time.

the key that i previously missed is that when the clock is LOW, the
"Reset" on the SR Latch is *disabled*, allowing the incoming Go_Read
to be ANDed with the Q state of the SRLatch, to generate the "Register
Select" signal for the brief [half] cycle needed to tell the Register
File "you can write now" (or read).

on the major clock cycle (half a clock cycle later), with the Go_Read
still being asserted, the latch will be reset because the combination
of the AND of the clock line plus the Go_Read will cause the Reset of
the Latch.

emulating this in "modern" design i believe is going to need something
insane like 30 gates: three DFFs, 4 MUXes at least.  where the 6600
has only 6 gates, one of those being a 3-input AND so a total of only
around 13 transistors.

just incredible.

l.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: comp_unit_req_rel.jpg
Type: image/jpeg
Size: 29818 bytes
Desc: not available
URL: <http://lists.libre-riscv.org/pipermail/libre-riscv-dev/attachments/20190520/ae041496/attachment-0003.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: group_pick_rd_rel.jpg
Type: image/jpeg
Size: 28567 bytes
Desc: not available
URL: <http://lists.libre-riscv.org/pipermail/libre-riscv-dev/attachments/20190520/ae041496/attachment-0004.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 2019-05-15_16-25.jpg
Type: image/jpeg
Size: 187705 bytes
Desc: not available
URL: <http://lists.libre-riscv.org/pipermail/libre-riscv-dev/attachments/20190520/ae041496/attachment-0005.jpg>