News for hackers. Stuff that matters.
Re:Guy does demoscene tricks to ESP32 powered display (2022) by majke on Friday October 04, 2024 @11:55AM (#41742697)
ESP32 uses the Xtensa instruction set. As far as I understand this is a custom instruction set, which is surprisingly popular in low power applications.

Investigating this lead me to an interesting polish company called semihalf who were recently acquired by google.

https://www.youtube.com/watch?v=qLh1FOcGysY

Apparently they added some Xtensa support to clang. I guess that means xtensa is very important in the android context.

Re:Guy does demoscene tricks to ESP32 powered display (2022) by philsnow on Friday October 04, 2024 @01:36PM (#41743650)
There are ESP32s that use RISC-V, see https://www.espressif.com/en/products/socs/esp32-c3
Re:Guy does demoscene tricks to ESP32 powered display (2022) by jsheard on Friday October 04, 2024 @01:44PM (#41743730)
All of their models released after a certain point are RISC-V, though there's still gaps in the capabilities of their RISC-V range (e.g. no LCD controller or USB OTG) which means it can't cover all of the Xtensa models use-cases yet.
Re:Guy does demoscene tricks to ESP32 powered display (2022) by robjwells on Friday October 04, 2024 @02:37PM (#41744278)
Xtensa is from Tensilica (now Cadence) but, as sibling comments point out, new Espressif designs use RISC-V. There's a bit of an overlap point: the ESP32-S2 and ESP32-S3 are recent designs that use Xtensa, and they were released at roughly the same time as the ESP32-C3 (I believe the first Espressif RISC-V part).

Here's the ISA summary for Xtensa LX as used by the ESP32: https://www.cadence.com/content/dam/cadence-www/global/en_US...

The S2 & S3 are more powerful and capable than the currently available Espressif RISC-V microcontrollers. Perhaps it's best to think of them marking the summit or climax of Espressif's Xtensa parts. But the future clearly lies with RISC-V.

The Xtensa support in LLVM/Clang appears to have resulted (at least in part) from Espressif's active support for Rust on their microcontrollers. The recent Rust 1.81 release merged support for the ESP32, ESP32-S2 and ESP32-S3 into upstream rustc. It's great to see a microcontroller vendor directly support the use of Rust on its parts, and not just its "new" parts.

Edit: To be clear, the ESP-IDF framework is a huge C project, so Espressif working on LLVM support for Xtensa is clearly _not just_ for Rust. And "easy mode" Rust-on-Espressif (std support!) relies on the ESP-IDF so merging their patches into LLVM is beneficial all-round.

Re:Guy does demoscene tricks to ESP32 powered display (2022) by ithkuil on Friday October 04, 2024 @03:03PM (#41744554)
Xtensa is an interesting "instruction set" because it's not one single instruction set but it's a "parametric instruction set" (hence the name which hints at "extensible")

When instantiating an xtensa core you can choose how many registers you want, whether you want to have register windows or not, and many other options.

If you want a compiler that targets a given instance of that architecture you have to tailor it to those parameters.

Re:Guy does demoscene tricks to ESP32 powered display (2022) by amelius on Friday October 04, 2024 @07:06PM (#41746344)
Sounds nice in theory, and it makes sense to some extent, but in practice that leaves you with a plethora of different instruction sets, so you'll have binary incompatibility issues all the time. Instead of downloading that "x86_64" package, you now have to download xtensa_par1_par2_par3_par4 etc.
Re:Guy does demoscene tricks to ESP32 powered display (2022) by ithkuil on Saturday October 05, 2024 @02:31AM (#41748101)
BTW I wrote a gdb-protocol speaking software debugger component for the esp8266 (before xtensa wrote/published their own)

I had to reverse engineer a lot of details (number of registers, Layout of the stack frame in various kinds of interrupts and traps etc) because xtensa didn't publish the actual parameters of their xtensa instantiation