TalkChess.com

I wonder why they didn't at least compress the data? a simple LZ compressor can reduce the net from 21 to 13.6MB and with
a simple byte-slicing filter (data is 16-bit, so simply group even and odd bytes together) to 12.5MB, that's 60% of the original
still trivial to decompress and also very fast

mar wrote: ↑Mon Aug 31, 2020 3:28 am I wonder why they didn't at least compress the data? a simple LZ compressor can reduce the net from 21 to 13.6MB and with
a simple byte-slicing filter (data is 16-bit, so simply group even and odd bytes together) to 12.5MB, that's 60% of the original
still trivial to decompress and also very fast

Because it is even more trivial to just zip the executable.

corres wrote: ↑Mon Aug 31, 2020 2:34 am

corres wrote: ↑Sun Aug 30, 2020 11:29 pm Concerning the compilation of SF+NNUE I mentioned the SF developer forgot to name the Windows compiler what appropriate to their "simple" Make file and to SF+NNUE non-embedded.
I think if you, Mr. "syzygy" publish this information, there would be many people who would thank you for help.

This is my text without sarcasm.
So which Windows compiler is good for SF+NNUE without embedding??
Excluding Visual Studio.

If you want to compile SF, install MSYS2.
https://github.com/glinscott/fishtest/w ... on-Windows
Instead of executing the "bash script", just do this (cd src; make help, etc.):
https://github.com/official-stockfish/S ... he-sources

Here is the problem that is probably unique for me, I have lots of machines of different kinds.
I have Intel and Amd and 20 year old machines and brand new machines.
Now, I also get all of Mike's stuff like Bluefish, etc. So multiply that machine hardware type by the number of build types.
Let's add in Corchess, Crystal, Cfish and similar things, because I simply can't live without them.

Yes, I can probably hand edit all the makefiles and figure out all the changes I need to make with each and every new build (and just saving the old makefile does not work, because makefiles change as often as the code does)

Here is the strange thing. If I build the SF version before embedding and the nnue file is not there it is automatically downloaded. Why can't they do the same thing when I am using it and not building it?
Or just have a batch file that looks for the file and if it does not see it call wget

I know I am a big whiner. But if you had the same perverse chess tendencies that I do, you would understand

Why not have a packaged version with the bin file and one without

I guess that the makefile will become much more complex if they double the build types.
And I only do profile builds so if they have a normal build without the binary that won't help me any.
I suspect that the bundling of the binary is a complex process, but I did not look into it yet.

syzygy wrote: ↑Mon Aug 31, 2020 3:49 am

corres wrote: ↑Mon Aug 31, 2020 2:34 am

corres wrote: ↑Sun Aug 30, 2020 11:29 pm Concerning the compilation of SF+NNUE I mentioned the SF developer forgot to name the Windows compiler what appropriate to their "simple" Make file and to SF+NNUE non-embedded.
I think if you, Mr. "syzygy" publish this information, there would be many people who would thank you for help.

This is my text without sarcasm.
So which Windows compiler is good for SF+NNUE without embedding??
Excluding Visual Studio.

If you want to compile SF, install MSYS2.
https://github.com/glinscott/fishtest/w ... on-Windows
Instead of executing the "bash script", just do this (cd src; make help, etc.):
https://github.com/official-stockfish/S ... he-sources

Many thanks for you and for Glinscott.
I did not know this spring.
For my luckiness I found a resolving to the present time with msys 1.0 and mingw, and I can reach the same speed as the official release has, but I am happy to know the general solution.
So many thanks for it.

corres wrote: ↑Mon Aug 31, 2020 2:34 am So which Windows compiler is good for SF+NNUE without embedding??
Excluding Visual Studio.

Fishtest is 1-click environment setup

https://github.com/glinscott/fishtest/w ... dows#msys2

AndrewGrant wrote: ↑Sun Aug 30, 2020 12:04 pm There is utility in packaging them together. The vast majority of Stockfish users are not people who frequent any forums like this.

If you really want to be ahead of the curve, just compile your own Stockfishs yourself.

There is a glitch in that, at least for me: the dev compile is 4 times as fast as mine. OK, I admit am totally new to compiling.

Gabor Szots wrote: ↑Mon Aug 31, 2020 12:17 pm

AndrewGrant wrote: ↑Sun Aug 30, 2020 12:04 pm There is utility in packaging them together. The vast majority of Stockfish users are not people who frequent any forums like this.

If you really want to be ahead of the curve, just compile your own Stockfishs yourself.

There is a glitch in that, at least for me: the dev compile is 4 times as fast as mine. OK, I admit am totally new to compiling.

That would be because you selected a suboptimal make target. It is far more complex now that there are a dozen different targets to account for the vector maths that some processors have. If you told me your CPU, I could probably look at the makefile and tell you the right one, but I've not followed all the changes