[Cryptography] Speculation considered harmful?

[Ryan Carboni]

I don't know how there is still enough demand to continue development on
Itanium, which ought to be the real question.

The obvious answer to me is (optional) x86 hardware emulation, and stricter
virtualization, with eDRAM buffers and barrel multiprocessing. Technically
MS-DOS is a virtual machine in the same sense that a burrito is a sandwich.
The difference between a proper software and hardware implementation is
only parsing speed.

Arguably the only part of a CPU that needs to be implemented in silicon is
the ALU, everything else can be implemented in a FPGA. Allows for even more
product binning ( a bit further than agner's idea
http://www.agner.org/optimize/blog/read.php?i=421#421 ). This is amazing:
https://github.com/cliffordwolf/picorv32 . 2000 slices for something as
fast as a Pentium III or IV.

Regardless, synthetic benchmarks are pointless if no one focuses on the
amount of cycles spent waiting for data.
https://gist.github.com/jboner/2841832
Forty million cycles to read from disk.

I have no idea how many branches are computed typically, but apparently it
is around a dozen? http://www.agner.org/optimize/microarchitecture.pdf
Seems like a waste in power, to get around the endless spurts and stops.


Ideally Google or Amazon would appoint Agner Fog as CPU Czar. They both
have the spare cash to actually build an entire chip fab.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.metzdowd.com/pipermail/cryptography/attachments/20180122/4df08a6f/attachment.html>