At 08:49 PM 5/04/2019 +0000, you wrote:
hat's a really interesting problem, and the government (NSA) wanted this badly and
they asked Seymour Cray to create a specific instruction for this and they called it
Anybody know the why and how it is useful?
I am deep in matrix math books and 'classification algorithms' in statistics math,
looking into electronics reliability WCCA, so this is an interesting topic.
If we're considering hardware solutions, then the best way is to build a simple I/O
device, with a writeable latch for the data word, fed as address into some nonvolatile
memory like a big EPROM or flash, the output of which can be read via a port. Fill the NV
memory with the required lookup table (derived by some code written in anything. BASIC for
So the required code is just one write and one read.
See the size chart: https://en.wikipedia.org/wiki/EPROM
The biggest EPROM made was ST M27C322, 32 Mbit, 2Mx16. 21 Address bits, 16 data bits,
80nS access time.
Wasteful though, since the result only needs 5 bits.
For an 8 bit result, 27C4001, 512K x 8 bit, 19 Address bits.
If the argument is only a 16bit word, then use a 27512. 64K x 8 bit