voffka wrote: ↑Fri Apr 16, 2021 1:10 pm
Hello there Alex,
It would be nice to compile Igel natively for ARM for maximum performance of Igel and avoid emulation. Let me do some more digging and see how to do it using gcc, I would appreciate if you run my builds as a test as unfortunately I don't have ARM devices at all.
I will ping you over PM when I have something.
P.S. You are running Windows ARM natively? or you M1? If on M1, wouldn't it be better to compile natively for Apple Silicon?
It's a little difficult to explain: Using Parallels Desktop for M1, Linux is native (as I cannot run Intel x64 engines but Windows 10 Is the contrary) I try to be more clear:
I run chess engines Intel x64 popcnt sse3 or sse41 on Windows 10 ARM64 under Parallels Desktop 16.5 for Mac M1 (not natives ARM64 but INTERPRETED) . Igel 2.6.0 Intel x64 for modern computers runs perfectly, Igel 3.0.0 not due to BM2 and AVX2 only builds. Id like to have the latest version for my tournaments on Windows 10.
On Ubuntu Linux 20.04 ARM64 (NATIVE but always under Parallels Desktop M1) Igel 3.0.0 (compiled by jsriver) with only 4 CPUs/8 calculates 1600-2000-4000 knps on starting position, middle game and endgame. Apple Silicon M1 has also neuronal engines inside, if you could use them for NNUE maybe could be very fast. On Mac Big M1 Sur Igel 2.9.0 NNUE compiled natIvely by Acepoint works perfectly.
https://www.google.com/url?sa=t&source= ... rPTm9H3ChH
I'm available for all the esperiments you need on the 3 operating systems on Mac mini M1 8-256 GB
Thank you very much for your support, I'd like to contribute to improve Igel.
Regards, AlexChess