3 ms·

Why stop there? Move the ALU to the RAM

rcxdude 2y ago
There are some proposals of that form: basically if you have a RAM interface that can support commands like "fill this range of memory with this pattern" or "add this constant to all values in this range" it can help speed up a decent amount of code, but also drastically reduce power consumption, as the memory interface is a pretty significant source of power drain on mobile systems. It does significantly complicate the interfaces involved, though, so it would require a lot of effort to implement.
- sroussey 2y ago
  https://www.tomshardware.com/tech-industry/neo-semiconductor-develops-3d-dram-with-ai-processing https://www.tomshardware.com/tech-industry/neo-semiconductor...
weinzierl 2y ago
This has been tried. I think the IBM System/360 could do arithmetic directly in RAM.
- _a_a_a_ 2y ago
  err, any refs for that? I really doubt it but could well be wrong. I'm not aware of any mainstream arch that can/could do that, that I can think of.
  - weinzierl 2y ago
    At least the IBM System/370 (not 360 as I said above) had assembly instructions that took memory addresses directly without a register or immediate being in the instruction. Here is an example of such an instruction from the "AP, Add Packed (Decimal) The data string located at the storage address specified by operand-2 (b2+d2) is added to the data string located at the storage address specified by operand-1 (b1+d1). Operand-2 remains unchanged." http://www.simotime.com/asmins01.htm#AP http://www.simotime.com/asmins01.htm#AP Now, how this was actually done in hardware I don't know, but I would not be surprised if hidden registers were involved. So, maybe this is not really an example that fits the original premise of this subthread, but I think it is interesting nevertheless.
vasco 2y ago
We've been doing the reverse and bringing the RAM to the ALU.
dzaima 2y ago
There's already an x86-64 extension for this: RAO-INT (remote atomics).

There are some proposals of that form: basically if you have a RAM interface that can support commands like "fill this range of memory with this pattern" or "add this constant to all values in this range" it can help speed up a decent amount of code, but also drastically reduce power consumption, as the memory interface is a pretty significant source of power drain on mobile systems. It does significantly complicate the interfaces involved, though, so it would require a lot of effort to implement.

sroussey 2y ago
https://www.tomshardware.com/tech-industry/neo-semiconductor-develops-3d-dram-with-ai-processing https://www.tomshardware.com/tech-industry/neo-semiconductor...

weinzierl 2y ago

This has been tried. I think the IBM System/360 could do arithmetic directly in RAM.

_a_a_a_ 2y ago
err, any refs for that? I really doubt it but could well be wrong. I'm not aware of any mainstream arch that can/could do that, that I can think of.
- weinzierl 2y ago
  At least the IBM System/370 (not 360 as I said above) had assembly instructions that took memory addresses directly without a register or immediate being in the instruction. Here is an example of such an instruction from the "AP, Add Packed (Decimal) The data string located at the storage address specified by operand-2 (b2+d2) is added to the data string located at the storage address specified by operand-1 (b1+d1). Operand-2 remains unchanged." http://www.simotime.com/asmins01.htm#AP http://www.simotime.com/asmins01.htm#AP Now, how this was actually done in hardware I don't know, but I would not be surprised if hidden registers were involved. So, maybe this is not really an example that fits the original premise of this subthread, but I think it is interesting nevertheless.

vasco 2y ago

We've been doing the reverse and bringing the RAM to the ALU.

dzaima 2y ago

There's already an x86-64 extension for this: RAO-INT (remote atomics).