RaspberryBASIC.org Forum

[jalih]

Pascal was my first "REAL" language (via Delphi 6 on Windows a long time ago).

You seem to be using semicolons  more like C than Pascal... I personally have not programmed in Pascal after elementary school but have sometimes used Modula-2 or Component Pascal.

[AIR]

I use swiftc to compile.

Right, which by default compiles with debugging enabled which of course will slow things down.

I did a test on my Mac, compiling using swiftc.  The first run with no optimization, the second with optimization:

[riveraa@mini ~/tmp/swift] $ time ./debug
r LEN:  999986
Front:  ZYXWVUTSRQPONMLKJIHGFEDCBA
Back:   ZYXWVUTSRQPONMLKJIHGFEDCBA
UBVal:  100000

real    0m0.977s
user    0m0.961s
sys    0m0.011s

[riveraa@mini ~/tmp/swift] $ time ./release
r LEN:  999986
Front:  ZYXWVUTSRQPONMLKJIHGFEDCBA
Back:   ZYXWVUTSRQPONMLKJIHGFEDCBA
UBVal:  100000

real    0m0.326s
user    0m0.313s
sys    0m0.009s

Optimized compile:  swiftc -O main.swift -o release

AIR.

[jalih]

I timed following 8th code on ROCK64 (should be about the same as RPI4 speedwise):

Code: [Select]

1000000 constant LIMIT

a:new 0 a:push var, a
0 b:new true b:writable var, s
0 b:new true b:writable var, t


: iterate
  s @ "" 2 pick n:1- 26 n:mod 65 n:+ s:+ b:append
  b:len 26 n:< not if
    t @ swap b:append drop
    0 b:new true b:writable s !
  else
    drop
  then
  a:push ;

: app:main
  a @ ' iterate 1 LIMIT loop
  t @ b:rev >s s:len "r LEN: %d\n" s:strfmt .
  dup 26 s:lsub "Front: %s\n" s:strfmt .
  26 s:rsub "Back:  %s\n" s:strfmt .
  LIMIT a:@ nip "UBVal: %d\n" s:strfmt .
  bye ;

Result was:

Code: [Select]

root@DietPi:~/Downloads# time /opt/8th/bin/rpi64/8th r3.8th
r LEN: 999986
Front: ZYXWVUTSRQPONMLKJIHGFEDCBA
Back:  ZYXWVUTSRQPONMLKJIHGFEDCBA
UBVal: 1000000

real 0m2,270s
user 0m2,160s
sys 0m0,100s
Memory usage was:

Code: [Select]

{"fault":0,"isrss":0,"rss":66452,"load15":0.30000,"ixrss":0,"load1":0.56000,"idrss":0,"swap":0,"load5":0.47000}

So, it's quite a lot faster than 40 seconds you have on the chart...

[AIR]

Nice, Jalih!

One thing I should point out:  I had installed DietPi on my Raspberry Pi 4 when I first got it, and my benchmarks were quite a bit slower than Raspbian installed on the same Pi, even when I overclocked it.

I didn't go too deep into determining why that was, in both cases I was running minimal installs instead of the full blown install, so it wasn't about additional resources being consumed by GUI processes since I don't use Linux like that (old school console/ssh access is all I need!  LOL).

AIR.

[jalih]

One thing I should point out:  I had installed DietPi on my Raspberry Pi 4 when I first got it, and my benchmarks were quite a bit slower than Raspbian installed on the same Pi, even when I overclocked it.

I didn't test many distros, DietPi worked and was clean enough! Performance with 8th is actually a lot better in most tasks with ROCK64 than on my six core desktop PC running Windows! It seems Windows version of the 8th is the slowest version of all!     

[John]

Jalif,

I need a binary for your latest submission if you are ready to release your code.

[John]

AIR,
I used -O with your last submission. Swift generates a main even if you use the -o option. There was about a second improvement over non-optimized.

[John]

I'm hoping a FreeBasic fan sees this and submits an optimized version.

[John]

Can we move this thread to the RaspberryBASIC forum? It isn't a blog and I'm wasting a lot of time being a messenger. I approved AIR's membership so he is free to post there. I'm sadden that @ejolson, heater, hippy and others didn't bother to join us. I miss the relationship.

[AIR]

Ever since you approved me, this is what I get when logging in:

  An Error Has Occurred!    Your session timed out while posting.  Please go back and try again.

[John]

Use the toolbar login method and it should save your cookie.

This seems to be happening with all the sites I host. Sorry!

[AIR]

John, what timing do you get with the following?

Code: Text

1. // Swift 5.1.2 - 1mil3.swift
2.  
3. var s = ""
4. var t = ""
5. var a = [Int](repeating: 0, count: 1000001)
6. var i = 0
7.  
8. for x in 1...1000000 {
9.   let c = Character(UnicodeScalar(((x - 1) % 26)+65)!)
10.   s.append(c)
11.   i += 1
12.   a[x] = x
13.   if i == 26 {
14.     t.append(String(s.reversed()))
15.     s = ""
16.     i = 0
17.   }
18. }
19.  
20. print("r LEN: ", t.count)
21. print("Front:  \(t.prefix(26))")
22. print("Back:   \(t.suffix(26))")
23. print("UBVal: ", a[100000])
24.  

AIR.

[John]

No improvement.


pi@RPi4B:~/swift-dev/examples $ swiftc -O 1mil3-4.swift
pi@RPi4B:~/swift-dev/examples $ timex ./main
r LEN:  999986
Front:  ZYXWVUTSRQPONMLKJIHGFEDCBA
Back:   ZYXWVUTSRQPONMLKJIHGFEDCBA
UBVal:  100000
23.33user 0.05system 0:23.51elapsed 99%CPU (0avgtext+0avgdata 12024maxresident)k
6368inputs+0outputs (25major+1877minor)pagefaults 0swaps
pi@RPi4B:~/swift-dev/examples $



You can download my Swift 5.1.2 RPi build here.
https://raspberrybasic.org/download/swift-5.1.2_armv7.tgz

[AIR]

The String object in Swift.org's RPI version of Swift is brain dead, on top of being slow.

I downloaded and installed the RPI deb from their site, tried a few things, but it reinforces the fact that I will never use Swift for any real code.  I'm about to uninstall it now.

AIR.

[John]

My 5.1.2 of Swift was built from the latest Apple sources and native on the RPi 4B. The latest binary I was able to find for the RPi was 5.1.1 which has the compare bug I found.

I didn't see any RPi distributions on the swift.org site. Where did you get the .deb from and what version was it?