Hacker News Reader

Comment by dmitrygr | Hacker News Reader

I’m all for new soft cores. However, do we really need to pollute the name space? Microblaze is already a name of an architecture that somebody might want to Google.

Just call it AMDcoreV or some such thing

View on HN

Correct, microblaze is the name of an architecture AMD owns, and this is compatible with, per the article.

> It allows developers to leverage the open-source RISC-V software ecosystem, is hardware compatible with the classic MicroBlaze processor

What does "hardware compatible" mean here though? It runs on the same FPGA hardware, but has unrelated CPU architecture?

I assume it means it's essentially pin-compatible, like, say, a PCIe GPU upgrade would be. Still needs new software. But has the same HDMI and PCIe and the same low-level protocols.

Soft cores don't have their own pins. I think "hardware compatible" means that it runs on the same FPGAs as microblaze arm with the same toolchain product suite.

"Pins", in this case, means the logical ports used to interface one module within an FPGA to another. Think AMBA bus, clock(s), interrupt lines, etc. An FPGA SOC design that currently throws a bunch of different IPs (modules) together to do... something... could do the same with this core thrown in to replace a (classic) Microblaze with only a resynthesis on the hardware side, and a recompile on the software side.

It has the same interface to the rest of the design -- the buses and control signals etc. You can drop the new RISC-V soft core straight into an existing design using Microblaze and change only the program code.

CPU is just one part. You're probably adding your own coprocessor and whatever other IO to the FPGA. If the CPU uses a different interconnect/protocol design, you'll have to redesign and debug your entire interface which will be very expensive.

Meanwhile, microblaze is a proprietary architecture. AMD puts a few hundred thousand dollars of dev time into it here or there, but most software stacks will have tons of stuff not at all designed for that ISA. If you can switch to RISC-V, you get access to an ecosystem where tons of companies are all pouring multi-millions in every year.

EDIT: a great example is that microBlaze was added to LLVM in 2010, but then removed in 2013 because nobody was free to maintain it.

Now AMD comes along and says you can swap out the CPU ISA to RISC-V without needing any hardware redesigns and then with a simple recompile, you can start using that massive ecosystem instead of spending man-years writing your own stuff which saves your company lots of money.

Hardware compatible doesn’t mean very much in this case. It’s a different ISA so you’ll need a different tool chain and recompilation (possibly rewrite given that assembly is very common in deeply embedded processors). It’ll also have different LUT usage, etc.

It means it's designed as a direct drop in replacent for a traditional microblaze ISA code..sure software must be recompiled, but hardware interfaces match what was there before for peripheral access, debug, instruction offload, etc. This makes the transition to using RISC-v much more seamless if you were already using microblaze

It is the same interface facing your HDL. Whereas on the software side, it is RISC-V, the ISA that's rapidly growing the strongest ecosystem.

Just a recompile away, except this time, you don't have to deal with poor, proprietary, bespoke, low quality tooling, but you can instead use reputable industry standard tools like gcc, binutils, llvm.

I'm pretty sure the "bespoke" tooling was just older versions of those industry standard tools because of poor upstreaming of architecture support. And there are probably many, very reasonable, justifications for that having been the case... the least of which is shipping toolchains with their own design tools.

RISC-V doesn't seem like the "strongest" ISA ecosystem, unless you mean among soft core CPUs?

Among 32-bit microcontroller-class architectures, modern toolchain support for rv32imc is second only to armv7m IMHO. Nobody wants to maintain support for any other vendor-specific ISAs.

Strongest of any ISA after ARM and x86 at this stage, I should think.

I can't really see AMD/Xilinx or Intel/Altera offering a free 486SX soft core, even though they legally could. It would I expect use a lot more LUTs, and be slower.

Licensing Cortex-M1 doesn't seem likely either.

Beides that there are ARM softcores available on Xilinx...

https://www.arm.com/resources/free-arm-cortex-m-on-fpga

>unless you mean among soft core CPUs

Note that RISC-V hard cores in FPGAs are already a reality, including Microsemi PolarFire as well as GOWIN GW2N.

Not long from now, I expect every FPGA vendor to switch over. The ones that already offer soft cores won't take long.

>RISC-V doesn't seem like the "strongest" ISA ecosystem

RISC-V is rapidly building the strongest ecosystem.

It already is the strongest besides x86 and ARM, and the position of these two isn't, by any means, secure.

This is the new Xilinx Microblaze. They went from an AVR-like instruction set to RV32.

...as in the MicroBlaze architecture owned by AMD by way of its acquisition of Xilinx?

At least it's "MicroBlaze V" which is better than no affix.

or simplify it as AMD V core to make it easier to say.

Not to be confused with AMD-V, the existing virtualisation extension (subtly different from and incompatible with Intel's VT-x) for its x86 CPUs.

"AMD V core" would only work (ignoring AMD-V virtualization assisting x86 extension) if AMD only ever planned to make one RISC-V core.

I am looking forward to Zen-V.

Reply on news.ycombinator.com