Re: [isa-dev] Digest for isa-dev@groups.riscv.org - 2 updates in 1 topic

[Robert Lipe]

I think if you're a customer of the vendor (Andes? Nuclei?) that provided the basis of P, P matters a lot. Packed SIMD fills a slightly different niche than V, but there's certainly overlap.

IMO, much more interest in the industry is swirling around V. I don't think P is dying as much as it is 'paving the cowpaths' of existing art from that one vendor and, perhaps, making it more usable to others.

If shipping silicon and product inertia didn't exist and everyone got to the starting line and the finishing line at the same time, there probably wouldn't be two, but here we are. You can perceive this as either a bad thing (ack! fragmentation! different is bad) or a good thing (yay! Competing specs! and implementation choice!) but it's just a thing that RISC-V has at this point.

In just a couple of weeks at the RISC-V love-in in San Jose, I suspect we'll see a flurry of spec activity and press releases. Some of those will turn into shipping products that a commoner can actually buy.

RJL

On Sun, Nov 6, 2022 at 5:47 AM <isa...@groups.riscv.org> wrote:

isa...@groups.riscv.org

Google Groups

Topic digest
View all topics

• Is P-extension dying or abandoned ? - 2 Updates

Is P-extension dying or abandoned ?

Anonymous Anonymous <privat...@gmail.com>: Nov 05 11:26PM -0700 It seems that everyone talks about the Vector-extension......

Tommy Murphy <tommy_...@hotmail.com>: Nov 06 11:08AM https://lists.riscv.org/g/tech-p-ext

Back to top

You received this digest because you're subscribed to updates for this group. You can change your settings on the group membership page. To unsubscribe from this group and stop receiving emails from it send an email to isa-dev+u...@groups.riscv.org.

[Philipp Tomsich]

On Mon, 7 Nov 2022 at 11:10, Robert Lipe <rober...@gmail.com> wrote:

I think if you're a customer of the vendor (Andes? Nuclei?) that provided the basis of P, P matters a lot. Packed SIMD fills a slightly different niche than V, but there's certainly overlap.

IMO, much more interest in the industry is swirling around V. I don't think P is dying as much as it is 'paving the cowpaths' of existing art from that one vendor and, perhaps, making it more usable to others.

If shipping silicon and product inertia didn't exist and everyone got to the starting line and the finishing line at the same time, there probably wouldn't be two, but here we are. You can perceive this as either a bad thing (ack! fragmentation! different is bad) or a good thing (yay! Competing specs! and implementation choice!) but it's just a thing that RISC-V has at this point.

In just a couple of weeks at the RISC-V love-in in San Jose, I suspect we'll see a flurry of spec activity and press releases. Some of those will turn into shipping products that a commoner can actually buy.

If vendors have implemented the previous P specification, we have a blueprint (from the V 0.7.1 discussion) on how this can still be supported upstream as a non-standard (i.e., it overlaps standard opcode space) custom extension (e.g., X-<vendor>-P).  The onus of developing the upstream support for X-P and maintaining it will fall on the proponents of X-P.

In my personal opinion, P (and X-P) fill a different space from V ... so there should be a motivation to the develop and upstream support.

Thanks,
Philipp.

[kr...@sifive.com]

P and V are quite different.

P cannot scale up to V space, and V cannot scale all the way down to P
space. This is why we have both.

P is not part of RVA profiles.

Krste

| --
| You received this message because you are subscribed to the Google Groups
| "RISC-V ISA Dev" group.
| To unsubscribe from this group and stop receiving emails from it, send an
| email to isa-dev+u...@groups.riscv.org.
| To view this discussion on the web visit
| https://groups.google.com/a/groups.riscv.org/d/msgid/isa-dev/CAAeLtUCxZ%3D03nVGc_2kugS-ezoyZZqNe55_O1sjk8FU_1c%2BhnQ%40mail.gmail.com
| .