4 ms·

Will they go the IP troll route in the future after having established quiet a IP portfolio for RISCV.

klelatti 3y ago
It's interesting that that they are arguing strongly (vs Qualcomm) for retaining an approach in the ISA that probably requires macro-op fusion - whilst holding at least one patent that relates to macro-op fusion.
- camel-cdr 3y ago
  I don't think the argument is that much about macro-op fusion vs cracking, as both sides agree that highperformance implementations are doable and not hinderes much buy both. I think the real dispute is how the opcode space and code density will evolve when more and more extension will need to be added. Will more 32 opcode space and 64 bit instructions but no 16 and no 48 bit instructions in the longterm be a better choice than fewer 32 bit insrructions, but 16/48/64 bit instructions?
  - klelatti 3y ago
    That's not my interpretation.
- fidotron 3y ago
  Have to wonder if the business strategy here is to get Qualcomm to buy what remains for access to these patents thinking Qualcomm will then change their position over the macro op fusion, which would be a classic move by people detached from technical realities.
- hajile 3y ago
  Neither side is correct IMO. Qualcomm wants 32-bit only instructions, but that leaves out many use cases that require longer instructions. On the flip side, a 48-bit instruction takes FIVE bits just to specify the length which is a massive amount of overhead. I think 64-bit instructions which contain between 1 and 4 instructions (not VLIW as no ILP is specified) would be a better solution offering more compressed instructions and much lower length encoding overhead for 48 and 64-bit instructions while still leaving a potential opening for very long instructions composed of several 64-bit groups. This beats out the Qualcomm proposal in terms of cache, code density, fusion, jumping, and many more without losing any of the benefits of compressed instructions (in fact, it increases the compression benefits).

It's interesting that that they are arguing strongly (vs Qualcomm) for retaining an approach in the ISA that probably requires macro-op fusion - whilst holding at least one patent that relates to macro-op fusion.

camel-cdr 3y ago
I don't think the argument is that much about macro-op fusion vs cracking, as both sides agree that highperformance implementations are doable and not hinderes much buy both. I think the real dispute is how the opcode space and code density will evolve when more and more extension will need to be added. Will more 32 opcode space and 64 bit instructions but no 16 and no 48 bit instructions in the longterm be a better choice than fewer 32 bit insrructions, but 16/48/64 bit instructions?
- klelatti 3y ago
  That's not my interpretation.
fidotron 3y ago
Have to wonder if the business strategy here is to get Qualcomm to buy what remains for access to these patents thinking Qualcomm will then change their position over the macro op fusion, which would be a classic move by people detached from technical realities.
hajile 3y ago
Neither side is correct IMO. Qualcomm wants 32-bit only instructions, but that leaves out many use cases that require longer instructions. On the flip side, a 48-bit instruction takes FIVE bits just to specify the length which is a massive amount of overhead. I think 64-bit instructions which contain between 1 and 4 instructions (not VLIW as no ILP is specified) would be a better solution offering more compressed instructions and much lower length encoding overhead for 48 and 64-bit instructions while still leaving a potential opening for very long instructions composed of several 64-bit groups. This beats out the Qualcomm proposal in terms of cache, code density, fusion, jumping, and many more without losing any of the benefits of compressed instructions (in fact, it increases the compression benefits).