Multi threaded FFTW 3.1.2 on a shared memory computer - multithreading

I use FFTW 3.1.2 with Fortran to perform real to complex and complex to real FFTs. It works perfectly on one thread.
Unfortunately I have some problems when I use the multi-threaded FFTW
on a 32 CPU shared memory computer. I have two plans,
one for 9 real to complex FFT and one for 9 complex to real FFT (size
of each real field: 512*512). I use Fortran and I compile (using ifort) my
code linking to the following libraries:
-lfftw3f_threads -lfftw3f -lm -lguide -lpthread -mp
The program seems to compile correctly and the function sfftw_init_threads returns a non-zero integer value, usually 65527.
However, even though the program runs perfectly, it is slower with 2
or more threads than with one. A top command shows weird CPU load
larger than 100% (and much more larger than n_threads*100). An htop
command shows that one processor (let's say number 1) is working at a
100% load on the program, while ALL the other processors, including
number 1, are working on this very same program, at a 0% load, 0% memory and 0 TIME.
If anybody has any idea of what's going on here... thanks a lot!

This looks like it could be a synchronisation problem. You can get this type of behaviour if all threads except one are locked out e.g. by a semaphore to a library call.
How are you calling the planner? Are all your function calls correctly synchronised? Are you creating the plans in a single thread or on all threads? I assume you've read the notes on thread safety in the FFTW docs... ;)

Unless your FFTs are pretty large, the automatic multithreading in FFTW is unlikely to be a win speed wise. The synchronization overhead inside the library can dominate the computation being done. You should profile different sizes and see where the break even point is.

Related

GNU c++: how to declare that heap can be swapped out when inactive?

My code is built in GNU c++. It's heavily multi-threaded and allocates a fairly large amount of RAM. Problem is, I routinely have a dozen of different versions of the code "running" at the same time - the "running" being in quotes because only one is actually running at any given time: all the others are suspended, with all of their threads suspended.
With that many copies of the code in the RAM, I sometimes run out of memory (with disastrous consequences)... while my page file always stays at a measly 2-3% of use. I would like to make my code swappable to the page file, thus keeping space in actual RAM available for other programs as soon as I suspend it... but I have no idea of how to do that in g++.
Any idea how to do that? Many thanks!!

openmp number of threads change on Fortran code

I am using Ubuntu 14.04 x64 on a machine with Intel Xeon CPU. and I am experiencing an strange behaviour. I have a Fortran code and a lengthy part of the calculation is parallel with OpenMP. With a smaller data set (say less than 4000) everything works fine. However, when I test a data set with 90K elements, in the middle the calculation the number of threads used suddenly drops to 1 which obviously slows down the computation.
I did these checks already:
Using OMP_GET_NUM_THREADS() I monitor the number of threads during the process and it ramains the same even after system uses 1 thread.
I use a LAPACK routine for eigen value calculation inside the loop. I compiled the Lapack again on my system to make sure the libraries on my system do not do anything.
Can it be that the System changes the number of used threads from outside? If so why?
Thank you.
It looks like a load balancing problem. Try with a dynamic scheduling :
!$OMP PARALLEL SCHEDULE(DYNAMIC)

Understanding cpu frequency, thread selection and more

With a 1270v3 and a single thread app I'm at the end of performance but when I watch monitoring tools like atop I don't understand how this whole stuff works. I tried to find a nice article about this sort of topic but they either have been explained in a language I don't understand or are not about the stuff I would like to know. I hope it is alright to ask this kind of stuff here.
From my understanding a single-thread app does only use one thread for all/most of the work. So the performance is defined by the single-thread power of the CPU.
A moment before I wrote this question I played around with CPU-frequency and noticed that although there are only two instances of the app running the usage is shared across all cores.
So I assume that the thread jumps around between these cores.
So I set the CPU scaling to performance with cpufreq-set -g performance. The result was that all CPU cores/threads stayed at about 2GHz like it was before besides one that is permanently on 3.5GHz (100%). As I only changed the scaling for one core, why is the usage still shared across all cores? I mean the app is running at about 300%, why doesn't it stick to the CPU core with the 100%?
Furthermore as I noticed that only one of the CPU's got scaled up I looked into the help page and found -r which should scale all cores with the performance settings. Unfortunately nothing does change. (Is this a bug in Ubuntu 1404?) So I used -c with the number 8 (8 threads) -> didn't work. 4 -> works but only scales 2 cores out of 8. 7 -> scaled 4 cores. So I'm wondering, does this not support hyper-threading or is the whole program that buggy?
However as I understand it, the CPU's with the max frequency together with the thread jump around in the monitoring tools as they display the average the usage, which than looks like shared. Did I figure this right?
Would forcing one cpu to 3.5GHz and forcing the app to this core improve performance or is all the stuff I'm wondering about only about avg calculation between the data they show each second.
If so am I right that I should run best with cpufreq-set -c 7 -g performance if power consumption doesn't matter?
Thanks for reading so far, I hope you have a moment to help me understand the whole thing.
Atop example screenshots:
http://i.imgur.com/VFEBvLx.png
http://i.imgur.com/cBKOnJM.png
http://i.imgur.com/bgQfwfU.png
I believe a lot of your confusion has to do with the fuzzy mapping of the capabilities of cpufreq to the actual capabilities of the hardware.
Here’s a description of what is taking place on the HW and in the OS.
A processor is a collection of cores on the same silicon substrate. The cores are what we used to call CPUs with some enhancements. CPUs now have the capability of running multiple HW threads (hyperthreading), each hardware thread being equivalent to one of the old type CPUs. Putting this all together, the 1270v3 is a quad core (if I recall correctly), meaning it has 4 cores on the same silicon substrate. Each core can support two HW threads, each HW thread being equivalent to what the OS calls a CPU (and I’ll call a virtual CPU). So from the OS perspective, the 1270v3 has 8 (virtual) CPUs.
The OS doesn’t see packages, cores or HW threads. It sees CPUs, and to it there appear to be 8 of them.
To further complicate the issue, modern processors have various HW supporting power saving states called P-states, C-states and package C-states. Why do I mention these? It’s because they are intimately associated with the frequency of the processor. And cpufreq professes to provide the user with control over the processor’s frequency.
Now, I’m not familiar with cpufreq outside of reading the manpage and other material on the web. From my reading, it has a lot of idiosyncrasies, so I’ll talk about what it is doing from a broad perspective.
In a general sense, cpufreq has a lot of generic capability that may or may not be supported by the HW or the kernel. This is confusing because it looks like the functionality is there but then things don’t happen as you would expect. For example, cpufreq gives the impression that you can set each CPU’s frequency independently. In reality, on a hyperthreaded system, two “CPUs” are associated with each core and must have the same frequency.
A lot of the functionality that cpufreq is supposed to control is associated with the power efficiency characteristics of the processor, but again, its mapping to the processor’s actual hardware capabilities is incomplete and misleading. Though cpufreq seems to allow you to set max and min frequencies, the processor hardware doesn’t work this way. In modern Intel processors, such as the 1270v3, there are something called P-states. These P-states are frequency-voltage pairs that slow down a processor’s frequency (and drop its voltage) to reduce the processor’s power consumption at the cost of the application taking longer to run. These frequency-voltage pairings aren’t arbitrary though cpufreq gives the impression that they are.
What does this all mean? In addition to the thread migration issues that the commenter mentioned, cpufreq isn’t going to behave the way you expect because it appears to have capability that it actually doesn’t, and the capability that it does actually have maps only roughly to the actual capabilities of the HW and OS.
I embedded some further comments in your text.
With a 1270v3 and a single thread app I'm at the end of performance but when I watch monitoring tools like atop I don't understand how this whole stuff works. I tried to find a nice article about this sort of topic but they either have been explained in a language I don't understand or are not about the stuff I would like to know. I hope it is alright to ask this kind of stuff here.
From my understanding a single-thread app does only use one thread for all/most of the work. [Yes, but this doesn’t mean that the thread is locked to a specific virtual CPU or core.] So the performance is defined by the single-thread power of the CPU. [It’s not that simple. The OS migrates threads around, it has its own maintenance processes, etc] A moment before I wrote this question I played around with CPU-frequency and noticed that although there are only two instances of the app running the usage is shared across all cores. So I assume that the thread jumps around between these cores. So I set the CPU scaling to performance with cpufreq-set -g performance. The result was that all CPU cores/threads stayed at about 2GHz like it was before besides one that is permanently on 3.5GHz (100%). As I only changed the scaling for one core, why is the usage still shared across all cores? I mean the app is running at about 300%, why doesn't it stick to the CPU core with the 100%? [Since I can’t see what you are observing, I don’t really understand what you are asking.]
Furthermore as I noticed that only one of the CPU's got scaled up I looked into the help page and found -r which should scale all cores with the performance settings. Unfortunately nothing does change. (Is this a bug in Ubuntu 1404?) So I used -c with the number 8 (8 threads) -> didn't work. 4 -> works but only scales 2 cores out of 8. 7 -> scaled 4 cores. [I haven’t used cpufreq so can’t directly speak to its behavior, but the manpage implies that “-c ” refers to a specific virtual CPU and not the number of virtual CPUs.] So I'm wondering, does this not support hyper-threading or is the whole program that buggy? [Again, I’m not sure from your explanation what you are doing, but the n->n/2 behavior makes sense to me. You can change the frequency of a core but since each core has two hyperthreads/virtual CPUs, two of those virtual CPUs must scale together.]
However as I understand it, the CPU's with the max frequency together with the thread jump around in the monitoring tools as they display the average the usage, which than looks like shared. Did I figure this right? [Again, I’m not sure what you are observing. Both physically and in atop, the CPU designation should not change, meaning CPU001 will always refer to the same virtual CPU. The core with the max frequency shouldn’t physically jump around, though the user thread may. Something to note is that monitoring tools can be pretty heavy users of the CPU. This heavy usage can make figuring out your processor usage difficult if it causes threads to jump around to different virtual CPUs.]
Would forcing one cpu to 3.5GHz and forcing the app to this core improve performance or is all the stuff I'm wondering about only about avg calculation between the data they show each second. [I found a pretty good explanation of atop with a lot of helpful screen shots: http://www.unixmen.com/linux-basics-monitor-system-resources-processes-using-atop/] If so am I right that I should run best with cpufreq-set -c 7 -g performance if power consumption doesn't matter? [It all depends upon what other processes are running on your system. If your system is mostly idle except for your processes, then forcing a core to a certain frequency won’t make a difference. [I’m suspicious of what a “governor” does. The language appears to refer to power-efficiency/performance (“balanced”, “powersave”, “performance”, etc) but the details don’t match the capability of today’s hardware.]
Thanks for reading so far, I hope you have a moment to help me

Memory Debugging

Currently I analyze a C++ application and its memory consumption. Checking the memory consumption of the process before and after a certain function call is possible. However, it seems that, for technical reasons or for better efficiency the OS (Linux) assigns not only the required number of bytes but always a few more which can be consumed later by the application. This makes it hard to analyze the memory behavior of the application.
Is there a workaround? Can one switch Linux to a mode where it assigns just the required number of bytes/pages?
if you use malloc/new, the allocator will always alloc a little more bytes than you requested , as it needs some room to do its housekeeping, also it may need to align the bytes on pages boundaries. The amount of supplementary bytes allocated is implementation dependent.
you can consider to use tools such as gperftools (google) to monitor the memory used.
I wanted to check a process for memory leeks some years ago.
What I did was the following: I wrote a very small debugger (it is easier than it sounds) that simply set breakpoints to malloc(), free(), mmap(), ... and similar functions (I did that under Windows but under Linux it is simpler - I did it in Linux for another purpose!).
Whenever a breakpoint was reached I logged the function arguments and continued program execution...
By processing the logfile (semi-automated) I could find memory leaks.
Disadvantage: It is not possible to debug the program using another debugger in parallel.

the output of my fortran code is killed , any suggestion?

I'm trying to run a code on ssh that works perfect for a smaller mesh , but since the new mesh is much bigger i used ifort command to compile it,
ifort -mcmodel=medium -i-dynamic -otest.out*.f
and it complies but when i run it , the output is:
killed
i know that problem is from memory, does anyone know if there's any way to run it?
how can i understand where in code cause memory problem?
Thanks
shadi
From the ifort command line, I think you are running on Linux.
Seeing "killed" as output is generally the result of Linux's Out Of Memory killer (OOM) getting involved to prevent an impending crash (because it's common practice for applications to ask for more memory then they need requests for more memory than is currently available are accepted - check for "Out of Memory: Killed process [PID] [process name]" in the system log files). The OOM killer is generally pretty good at disposing of the application responsible for using all the memory, so the place to start is your applications memory usage.
The first thing to do is try and estimate (even if it's only roughly) how much memory you expect your application to use. One approach is to guestimate the size of the major arrays and multiply them by the number of bits needed per element. Another approach is to think about how you would expect the memory use to grow with mesh size. You can study this by experiment (run with different mesh sizes, measure the memory use and extrapolate) or from one measurement and knowledge of how the major array scale. It may be that you are asking for much more memory then you have on the machine: and the solution to this is probably to get a access to bigger computer. (Or you could try and find an alternative algorithm which uses less memory.)
If their is a memory leak you should see more memory use than expected, even for the smaller mesh size. If this is the case, valgrind should help. Moving from static to dynamic storage probably isn't going to help here - I would expect to see a segmentation fault if you were just exceeding the available space on the stack.
try using valgrind. i tried it to find memory leaks in my fortran code with good success.
http://valgrind.org/

Resources