From time to time, I write inline asm code to speed up a number cruncher. I always try to avoid writing to memory, but I must write temporary arrays to memory as there isn't enough register space. So, my question is this, are there any plans on adding an array register to Intel CPUs?
Something on the order of 64k bytes with an acronym of RX or ERX, depending on whether entries are referenced 16 or 32 bits, would be good. The array register doesn't have to have a slew of instructions, just Read Item, Write Item, Zero All, and possibly Read/Write Block of Items from/to memory. A fast temporary array storage on the CPU would shed lots of clock cycles from code that doesn't need to save a list of values between calls.
Has anyone else noticed this bottleneck? Has this been proposed before? Am I wasting my time by asking these questions?
I think that it is best for you to post this on the Software Community Website as well.
Possibly someone will reply you here as well.
All the best,