SNAPPY fast builds

[guest59697]

hello, I did a small build of Google Snappy for Delphi and FPC (windows for now, will do linux objects too).
After a lot of environment settings and many compilers, I have done llvm builds.
Here you can found the lib and some benchmarks http://www.dellapasqua.com/snappy64/
thank you and best regards
Roberto

[Thaddy]

Hi Roberto! long time saw you here

I had a first look and - since I am in a bet - my only available computer at home is a Raspberrypi 3.
Sadly, linux armhf is missing Can I help?

Also: I saw some home baked C standard library routines e.g. in snappyarm.pas.
In FreePascal you can simply link against libc.

[guest59697]

hello Thaddy, nice to meet you again!

I have updated the snappy page with zlib vs snappy http://www.dellapasqua.com/snappy64/
with json files (from ecommerce ordering system), the ratio is similar, but the speed is 10x
this can save a lot of cpu and bandwidth on a typical client-server scenario
I did a lot of study over many algorithms with a general purpose license (see http://mattmahoney.net/dc/text.html), at the end a good combination of coding and toolchains permits those very good results through snappy (and for free)

probably I'll not provide linux static libraries...who uses linux knows how to gcc a library and call it from fpc :-P

cu !
btw. it's brook web framework suitable for fastcgi high demanding web server?
lighttpd+cassandra+brook
(this lib compiles? https://pascassa.codeplex.com/)

[shobits1]

I was about to ask for comparison between snappy vs SynLZ (from mormot) but found it here: http://synopse.info/forum/viewtopic.php?id=3460&p=1 for anyone interested  .

[guest59697]

I wait llvm 3.9 and will do builds for win32, win64, osx32, android, ios32, ios64
I'm seeking ways to make vectorization of loops thorugh llvm

[guest59697]

Hello,

I have done a build using llvm 4 with polly optimizer, speed gain is negligible with large data to compress and something more with uncompress, anyway is under dev, so consider it a test...
Instead, using a basic loop using TParallel.For in my I7 I get near 7-8GB/s (this can really show a multithreaded scenario)

If you like take a look at www.dellapasqua.com/snappy64

cheers
Roberto