search

ObjectPascal-Community / 1brc-ObjectPascal

1️⃣🐝🏎️ The 1 Billion Row Challenge in Object Pascal
MIT License
54 stars 22 forks source link

Update README.md #170

Closed EagleAglow closed 6 months ago

EagleAglow commented 6 months ago

Version 3.2 uses 8 threads on my PC, not sure about the LINUX side.

gcarreno commented 6 months ago

Hey Brian(@EagleAglow),

CRAP!! Mea culpa !! I read your message somewhere and completely forgot to correct that!!

Cheers, Gus

EagleAglow commented 6 months ago

No worries, but I AM confused by the results. My first attempt (1 thread) was about 16 minutes on my PC and 6 on the testbed). My last version was about 2.5 minutes on my PC (8 threads), so I was hoping for the same time ratio.

gcarreno commented 6 months ago

Hey Brian(@EagleAglow),

Humm, that is suss AF, indeed!!

I'm running your binary with these switches: -i [[input]], where [[input]] get's replaced by the input file.

Am I missing something here?

If not, I'm guessing that you're running your test in either Win32 or Win64. A lot of people have told me that the code that comes out of the Linux compiler is a bit shit when compared with :scream: Win32. I'm still baffled by the fact that some Delphi experts are touting the feats of the Win32 version when we're in a 64b world :shrug:

We need to get to the bottom of this, nonetheless!!

Cheers, Gus

EagleAglow commented 6 months ago

Right, my runs are in 64bit windows. 4 cores, Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz with about 10GB free memory.

The only switch is that a command line without an output file uses STDOUT.

I was wondering if the compiler for LINUX disliked Ryzen architecture, but my ignorance is probably showing.

gcarreno commented 6 months ago

Hey Brian(@EagleAglow),

I was wondering if the compiler for LINUX disliked Ryzen architecture, but my ignorance is probably showing.

Yours and mine!! I have absolutely no clue what's going on with the Delphi compiler for Linux :shrug:

I was quite hopeful that this would not be a thing, but alas, looks like we do have a bit of an issue with performance when coming from the Linux compiler.

Cheers, Gus