https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88284

--- Comment #8 from Michael_S <already5chosen at yahoo dot com> ---
(In reply to sandra from comment #7)
> While Intel has revived the "Altera" name, the Nios II processor is still
> listed as discontinued.  I see they are offering ARM-based FPGA products
> again instead.
> 

Arm is used for "hard" relatively high-performance core loosely integrated with
fabric part of the chip. It is not new. 
First, hard Arm core appeared in Escalibour (not sure about spelling) that was
an extremely unsuccessful Altera chip of very early 00s.
It reappeared in much more successful HPS variant of Cyclone-5, still few years
before Intel acquisition. Since than, all new families (there were too few)
featured both variant with HPS and variant without it.

Nios2 is used in different role - soft core. Naturally, being part of the main
fabric, it is very tightly integrated in it. Performance is much lower, but if
you want, you can have more cores than you have on HPS side. Think of it as
microcontroller within FPGA rather than as application processor.

ARM also has this type of core - Cortex-M1. But licensing policies of Arm are
such that for great majority of Altera-based designs Cortex-M1 is unattractive.
The same applies to Xilinx (owned by AMD). The same does *not* apply to
Microsemi (owned by Microchop).

> For many years Altera (and later Intel) had a business relationship with
> CodeSourcery/Mentor Graphics/Siemens to provide long-term support for the
> GNU toolchain on Nios II, but that contract ended a year ago, Siemens
> decided to get out of the compiler services business entirely, and both
> Chung-Lin and I changed jobs shortly afterwards and are now at BayLibre. 
> The real issue with continuing upstream toolchain support is that we lost
> our test configuration; the boards were returned to Intel and even if our
> right to use the simulators and support software they'd provided us with
> didn't end when the contract did, it certainly didn't transfer to BayLibre. 
> BayLibre sales did talk to Intel earlier this year and they were still not
> interested in Nios II support, so we basically had no choice but to pull the
> plug.

Sad to hear.
Comparatively to non-compressed variety of RV32, Nios2 is better architecture.
Not much better, but better nonetheless.
More importantly, migration to RV32 has cost for us, customers. Both in
development time and in potential bugs. Cost can be acceptable when you get
back at least some improvements. In this case there are nothing we are getting
back.

Reply via email to