View previous topic :: View next topic |
Author |
Message |
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Wed Apr 07, 2021 6:08 am Post subject: TIOBE index top 20: re-emergence of Fortran |
|
|
"Another notable change is the re-emergence of Fortran in the index at 20th position, up from 34th spot a year ago. Fortran, which emerged from IBM in the 1950s, remains popular in scientific computing. Its highest ranking on Tiobe's index was 10th in 2002.
Fortran was the first commercial programming language ever, and is gaining popularity thanks to the massive need for (scientific) number crunching. Welcome back Fortran.
The top 20 programming languages this month were: C, Java, Python, C++, C#, Visual Basic, JavaScript, Assembly language, PHP, and SQL, Classic Visual Basic, Delphi/Object Pascal, Ruby, Go, Swift, R, Groovy, Perl, MATLAB, and Fortran."
https://www.tiobe.com/tiobe-index/ |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Fri Apr 09, 2021 10:36 pm Post subject: |
|
|
Noticed Fortran almost beating MATLAB which is on 19th place ?
To surpass it we need
- perfect %pl (must be publishable quality by default)
- add surfaceplot
- add 3D plot (OpenGL)
I have surfaceplot and OpenGL implemented, they are not difficult to do
Also
- perfect run speed on Polyhedron tests
- add Wininet (sockets)
- add support of OpenMP and create examples
- add support of HDF5 + examples |
|
Back to top |
|
|
wahorger
Joined: 13 Oct 2014 Posts: 1225 Location: Morrison, CO, USA
|
Posted: Fri Apr 09, 2021 11:21 pm Post subject: |
|
|
Dan you intrigued me with "surface plot". Is this a routine that uses "Z" data in addition to an X and Y to interpolate thicknesses at various positions? |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Sat Apr 10, 2021 10:28 am Post subject: |
|
|
Bill
Yes. In simple form it is like here. Done with current %GR quarter century ago. Same can be done with %PL
in more complex like here. Bottom is 3D, top - surface plot. Essentially it is 2D plot in 3D. Both done in OpenGL. Difference in programming almost zero
I published sources demo many times here. With OpenGL the only what's left is to add numbering like in the 3D picture. I modified existing OpenGL example supplied with Silverfrost Examples without even learning OpenGL
Other examples. Surface plot at the bottom
Good to add a pie chart
|
|
Back to top |
|
|
wahorger
Joined: 13 Oct 2014 Posts: 1225 Location: Morrison, CO, USA
|
Posted: Sun Apr 11, 2021 4:04 pm Post subject: |
|
|
You sent me these OpenGL examples to use for learning some of the ins/outs of OpenGL. Much appreciated. |
|
Back to top |
|
|
Robert
Joined: 29 Nov 2006 Posts: 450 Location: Manchester
|
Posted: Sun Apr 11, 2021 6:42 pm Post subject: |
|
|
I am not sure what to make of it when Assembler is in the top 20. |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Mon Apr 12, 2021 1:55 am Post subject: |
|
|
Assembler became popular lately because of extensive use of testing by everyone on the net. Beating competitors just by the factor of 1% immediately makes you the winner of universe and all others bite the dust.
Can the Polyhedron tests be disassembled to understand why gfortran and IVF sometimes beat by the factor of 2-5-10 and even 30? |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Sat Nov 11, 2023 7:43 am Post subject: |
|
|
Look at that. Fortran jumped on 12th place.
Because it is Simplest. Fastest. Best.
And on supercomputers C and Fortran are like a king and queen.
Only utter @#$% think Fortran is dead.
https://www.tiobe.com/tiobe-index/
By switching from the code written in C to Fortran i got so large boost in performance that i used just the PC and did not use supercomputers for almost a year.
That was amazing year. With this and other tricks we got the code run speed up by 5 orders of magnitude. Imagine, the code which on supercomputer 1000-2000 cores ran for two weeks ending on PC in 20 minutes! When for first time i ran this code i though it quickly crashed. But then i saw full directory of data
All that got me to realize one pity thing. Unfortunately it is impossible to satisfy humans. I got one task which grabbed all that extra power and even 100x more so that I recently returned to supercomputers again |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Wed Jan 31, 2024 12:33 pm Post subject: |
|
|
Got one more order of magnitude compared to previous 16-core AMD 5950x...Look at this thread count. Interesting that i got 1.35 times boost in speed going from 6 memory channels to 12. So appears this indeed is important.
And i made it incredibly cheap for that kind of performances, guarantee you will not believe me...
|
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Thu Aug 15, 2024 12:28 am Post subject: |
|
|
TIOBE Index for August 2024 :
Fortran steadily going to the top.
From 20th to 10th place already.
And ahead of MATLAB
/** And if Silverfrost
- made its compiler for Linux
- with MPI and OpenMP parallelization,
- with standard Makefiles and preprocessing language similar to other Fortran compilers
- with SDBG64 debugger capable to debug MPI parallel files (unbelievable, but there are no friendly graphics debuggers among all Linux compilers, they debug like Neanderthals planting PRINT into the code )
- added Clearwin for Linux and made more flashy scientific graphics examples including OpenGL
- made more freindly Plato for Linux
- optimized run speed beating at least Gfortran
- and advertised it at least ones in last 25 years
***it would be in the first 5*** |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Mon Sep 09, 2024 5:57 am Post subject: |
|
|
Run on GPU.
Colleague rewrote the program to run on NVIDIA GPUs instead of regular CPU and reported that he got the speedup 173x from just single GPU !!! When Silverfrost adopts NVIDIA CUDA? |
|
Back to top |
|
|
JohnCampbell
Joined: 16 Feb 2006 Posts: 2584 Location: Sydney
|
Posted: Wed Sep 18, 2024 6:47 am Post subject: |
|
|
Quote: | reported that he got the speedup 173x from just single GPU |
Dan,
Is this dan-talk !!
This speedup needs some explaination !
For my calculations, I can get 10 to 15x with OpenMP (100 Gflop/s), but the bottleneck has shifted to memory bandwidth.
Your example must be a unique calculation, with minimal memory footprint to achieve 173x performance gain.
Your other descriptions have referred to very large memory footprint, so a bit confusing ?
Can you give some description of the hardware and calculations.
How many local cpu's are on the GPU you are using ?
Potentially if the GPU supports intrinsic functions ?
Do GPU support AVX instructions ?
More details would be interesting. |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2865 Location: South Pole, Antarctica
|
Posted: Wed Sep 18, 2024 11:01 am Post subject: |
|
|
John, I do not know what is dan-talk - ?
I am waiting for confirmations from other people but the more i know about GPU the more i feel that such speedups are typical without any special trickery. But again, we always need double check |
|
Back to top |
|
|
|