I know that the usual method when we want to make a big math computation faster is to use multiprocessing / parallel processing: we split the job in for example 4 parts, and we let 4 CPU cores run in parallel (parallelization). This is possible for example in Python with multiprocessing module: on a 4-core CPU, it would allow to use 100% of the processing power of the computer instead of only 25% for a single-process job.
But let's say we want to make faster a non-easily-splittable computation job.
Example: we are given a number generator function generate(n) that takes the previously-generated number as input, and "it is said to have 10^20 as period". We want to check this assertion with the following pseudo-code:
a = 17
for i = 1..10^20
a = generate(a)
check if a == 17
Instead of having a computer's 4 CPU cores (3.3 Ghz) running "in parallel" with a total of 4 processes, is it possible to emulate one very fast single-core CPU of 13.2 Ghz (4*3.3) running one single process with the previous code?
Is such technique available for a desktop computer? If not, is it available on cloud computing platforms (AWS EC2, etc.)?
Single-threaded performance is extremely valuable; it's much easier to write sequential code than to explicitly expose thread-level parallelism.
If there was an easy and efficient general-purpose way to do what you're asking which works when there is no parallelism in the code, it would already be in widespread use. Either internally inside multi-core CPUs, or in software if it required higher-level / larger-scale code transformations.
Out-of-order CPUs can find and exploit instruction-level parallelism within a single thread (over short distances, like a couple hundred instructions), but you need explicit thread-level parallelism to take advantage of multiple cores.
This is similar to How does a single thread run on multiple cores? over on SoftwareEnginnering.SE, except that you've already ruled out any easy-to-find parallelism including instruction-level parallelism. (And the answer is: it doesn't. It's the hardware of a single core that finds the instruction-level parallelism in a single thread; my answer there explains some of the microarchitectural details of how that works.)
The reverse process: turning one big CPU into multiple weaker CPUs does exist, and is useful for running multiple threads which don't have much instruction-level parallelism. It's called SMT (Simultaneous MultiThreading). You've probably heard of Intel's Hyperthreading, the most widely known implementation of SMT. It trades single-threaded performance for more throughput, keeping more execution units fed with useful work more of the time. The cost of building a single wide core grows at least quadratically, which is why typical desktop CPUs don't just have a single massive core with 8-way SMT. (And note that a really wide CPU still wouldn't help with a totally dependent instruction stream, unless the generate function has some internal instruction-level parallelism.)
SMT would be good if you wanted to test 8 different generate() functions at once on a quad-core CPU. Without SMT, you could alternate in software between two generate chains in one thread, so out-of-order execution could be working on instructions from both dependency chains in parallel.
Auto-parallelization by compilers at compile time is possible for source that has some visible parallelism, but if generate(a) isn't "separable" (not the correct technical term, I think) then you're out of luck.
e.g. if it's return a + hidden_array[static_counter++]; then the compiler can use math to prove that summing chunks of the array in parallel and adding the partial sums will still give the same result.
But if there's truly a serial dependency through a (like even a simple LCG PRNG), and the software doesn't know any mathematical tricks to break the dependency or reduce it to a closed form, you're out of luck. Compilers do know tricks like sum(0..n) = n*(n+1)/2 (evaluated slightly differently to avoid integer overflow in a partial result), or a+a+a+... (n times) is a * n, but that doesn't help here.
There is a scheme studied mostly in the academy called "Thread Decomposition". It aims to do more or less what you ask about - given a single-threaded code, it tries to break it down into multiple threads in order to divide the work on a multicore system. This process can be done by a compiler (although this requires figuring out all possible side effects at compile time which is very hard), by a JIT runtime, or through HW binary-translation, but each of these methods has complicated limitations and drawbacks.
Unfortunately, other than being automated, this process has very little appeal as it can hardly match true manual parallelization done by a person how understands the code. It also doesn't simply scale performance according to the number of threads, since it usually incurs a large overhead in the form of code that has to be duplicated.
Example paper by some nice folks from UPC in Barcelona: http://ieeexplore.ieee.org/abstract/document/5260571/
Related
My question primarily applies to firestorm/icestorm (because that's the hardware I have), but I am curious about what other representative arm cores do too. Arm has strange pre- and post-incremented addressing modes. If I have (for instance) two post-incremented loads from the same register, will the second depend on the first, or is the CPU smart enough to perform them in parallel?
AFAIK the exact behaviour of the M1 execution units is mainly undocumented. Still, there is certainly a dependency chain in this case. In fact, it would be very hard to break it and the design of modern processors make this even harder: the decoders, execution units, schedulers are distinct units and it would be insane to dynamically adapt the scheduling based on the instructions executed in parallel by execution units so to be able to break the chain in this particular case. Not to mention that instructions are pipelined and it generally takes few cycles for them to be committed. Furthermore, the time of the instructions is variable based on the fetched memory location. Finally, even this would be the case, the Firestorm documents does not mention such a feedback loop (see below for the links). Another possible solution for a processor to optimize such a pattern is to fuse the microinstructions so to combine the increment and add more parallelism but this is pretty complex to do for a relatively small improvement and there is no evidence showing Firestorm can do that so far (see here for more information about Firestorm instruction fusion/elimitation).
The M1 big cores (Apple's Firestorm) are designed to be massively parallel. They have 6 ALUs per core so they can execute a lot instructions in parallel on each core (possibly at the expense of a higher latency). However, this design tends to require a lot more transistors than current mainstream x86 Intel/AMD alternative (Alderlake/XX-Cove architecture put aside). Thus, the cores operate at a significantly lower frequency so to keep the energy consumption low. This means dependency chains are significantly more expensive on such an architecture compared to others unless there are enough independent instructions to be execute in parallel on the critical path. For more information about how CPUs works please thread Modern Microprocessors - A 90-Minute Guide!. For more information about the M1 processors and especially the Firestorm architecture, please read this deep analysis.
Note that Icestorm cores are designed to be energy efficient so they are far less parallel and thus having a dependency chain should be less critical on such a core. Still, having less dependency is often a good idea.
As for other ARM processors, recent core architecture are not as parallel as Firestorm. For example, the Cortex-A77 and Neoverse V1 have "only" 4 ALUs (which is already quite good). One need to also care about the latency of each instruction actually used in a given code. This information is available on the ARM website and AFAIK not yet published for Apple processors (one need to benchmark the instructions).
As for the pre VS post increment, I expect them to take the same time (same latency and throughput), especially on big cores like Firestorm (that try to reduce the latency of most frequent instruction at the expense of more transistors). However, the actual scheduling of the instruction for a given code can cause one to be slower than the other if the latency is not hidden by other instructions.
I received an answer to this on IRC: such usage will be fairly fast (makes sense when you consider it corresponds to typical looping patterns; good if the loop-carried dependency doesn't hurt too much), but it is still better to avoid it if possible, as it takes up rename bandwidth.
What could be a typical or real problem for using parallel programming? It can be quite challenging to implement. On the internet they explain how to use it but not why.
Performance is the most common reason to use parallel programming. But: Not all programs will become faster by using parallel programming. In most cases your algorithm consists of parts that are parallelizable and parts, that are inherently sequential. You always have to reason about the potential performance gain of using parallel programming. In some cases the overhead for using it will actually make your program slower. Have a look at Amdahl's law to learn more about the potential performance improvements you can reach.
If you only want some examples of usage of parallel computations: There are some classes of algorithms that are inherently parallel, see this article the dwarfs of berkeley
Another reason for using a multithreaded application architecture is it's responsiveness. There are certain functions which block program execution for a certain amount of time, i.e. reads from files, network, waiting for user inputs, etc. While waiting like this does not consume CPU power, it often blocks or slows program flow.
Using threads in such case is simply a good practice to make the code clearer. Instead of using (often complex or unintuitive) checks for inputs, integrating those checks into program flow, manual switching between handling input and other tasks, a programmer may choose to use threads and let one thread wait for input, and the other i.e. to perform calculations.
In other words, multiple threads sometimes allow for better use of different resources at your computer's disposal: network, disk, input devices or simply monitor.
Generalization: using multiple threads (including parallel data processing) is advisable when the speed and responsiveness gains outweigh the synchronization costs and work required to parallelize the application.
The reason why there is increased interest in parallel programming is partly because the hardware we use today is more parallel. (multicore processors, many-core GPU). To fully benefit from this hardware you need to program in parallel.
Interestingly, parallel processing also improves battery life:
Having 4 cores at 1Ghz draws less power than one single core at 4Ghz.
A phone with a multicore CPU will try to run as much tasks as possible simultaneously, so it can turn off the CPU when all work is done. This is sometimes called "the rush to idle".
Now, some programs are more easy parallelize than others. You should not randomly try to parallelize your entire code base. But it can be a useful excersise to do so even if there is no business reason: then you will be more ready the day when you really need it.
There are very few problems which can't be solved more quickly by a parallel program than by a serial program. There are very few computers which do not have multiple processing units.
I conclude, therefore, that you should use parallel programming all the time.
I am tunning my GEMM code and comparing with Eigen and MKL. I have a system with four physical cores. Until now I have used the default number of threads from OpenMP (eight on my system). I assumed this would be at least as good as four threads. However, I discovered today that if I run Eigen and my own GEMM code on a large dense matrix (1000x1000) I get better performance using four threads instead of eight. The efficiency jumped from 45% to 65%. I think this can be also seen in this plot
https://plafrim.bordeaux.inria.fr/doku.php?id=people:guenneba
The difference is quite substantial. However, the performance is much less stable. The performance jumps around quit a bit each iteration both with Eigen and my own GEMM code. I'm surprised that Hyperthreading makes the performance so much worse. I guess this is not not a question. It's an unexpected observation which I'm hoping to find feedback on.
I see that not using hyper threading is also suggested here.
How to speed up Eigen library's matrix product?
I do have a question regarding measuring max performance. What I do now is run CPUz and look at the frequency as I'm running my GEMM code and then use that number in my code (4.3 GHz on one overclocked system I use). Can I trust this number for all threads? How do I know the frequency per thread to determine the maximum? How to I properly account for turbo boost?
The purpose of hyperthreading is to improve CPU usage for code exhibiting high latency. Hyperthreading masks this latency by treating two threads at once thus having more instruction level parallelism.
However, a well written matrix product kernel exhibits an excellent instruction level parallelism and thus exploits nearly 100% of the CPU ressources. Therefore there is no room for a second "hyper" thread, and the overhead of its management can only decrease the overall performance.
Unless I've missed something, always possible, your CPU has one clock shared by all its components so if you measure it's rate at 4.3GHz (or whatever) then that's the rate of all the components for which it makes sense to figure out a rate. Imagine the chaos if this were not so, some cores running at one rate, others at another rate; the shared components (eg memory access) would become unmanageable.
As to hyperthreading actually worsening the performance of your matrix multiplication, I'm not surprised. After all, hyperthreading is a poor-person's parallelisation technique, duplicating instruction pipelines but not functional units. Once you've got your code screaming along pushing your n*10^6 contiguous memory locations through the FPUs a context switch in response to a pipeline stall isn't going to help much. At best the other pipeline will scream along for a while before another context switch robs you of useful clock cycles, at worst all the careful arrangement of data in the memory hierarchy will be horribly mangled at each switch.
Hyperthreading is designed not for parallel numeric computational speed but for improving the performance of a much more general workload; we use general-purpose CPUs in high-performance computing not because we want hyperthreading but because all the specialist parallel numeric CPUs have gone the way of all flesh.
As a provider of multithreaded concurrency services, I have explored how hyperthreading affects performance under a variety of conditions. I have found that with software that limits its own high-utilization threads to no more that the actual physical processors available, the presence or absence of HT makes very little difference. Software that attempts to use more threads than that for heavy computational work, is likely unaware that it is doing so, relying on merely the total processor count (which doubles under HT), and predictably runs more slowly. Perhaps the largest benefit that enabling HT may provide, is that you can max out all physical processors, without bringing the rest of the system to a crawl. Without HT, software often has to leave one CPU free to keep the host system running normally. Hyperthreads are just more switchable threads, they are not additional processors.
I'm creating a multi-threaded application in Linux. here is the scenario:
Suppose I am having x instance of a class BloomFilter and I have some y GB of data(greater than memory available). I need to test membership for this y GB of data in each of the bloom filter instance. It is pretty much clear that parallel programming will help to speed up the task moreover since I am only reading the data so it can be shared across all processes or threads.
Now I am confused about which one to use Cilk, Cilk++ or OpenMP(which one is better). Also I am confused about which one to go for Multithreading or Multiprocessing
Cilk Plus is the current implementation of Cilk by Intel.
They both are multithreaded environment, i.e., multiple threads are spawned during execution.
If you are new to parallel programming probably OpenMP is better for you since it allows an easier parallelization of already developed sequential code. Do you already have a sequential version of your code?
OpenMP uses pragma to instruct the compiler which portions of the code has to run in parallel. If I understand your problem correctly you probably need something like this:
#pragma omp parallel for firstprivate(array_of_bloom_filters)
for i in DATA:
check(i,array_of_bloom_filters);
the instances of different bloom filters are replicated in every thread in order to avoid contention while data is shared among thread.
update:
The paper actually consider an application which is very unbalanced, i.e., different taks (allocated on different thread) may incur in very different workload. Citing the paper that you mentioned "a highly unbalanced task graph that challenges scheduling,
load balancing, termination detection, and task coarsening strategies". Consider that in order to balance computation among threads it is necessary to reduce the task size and therefore increase the time spent in synchronizations.
In other words, good load balancing comes always at a cost. The description of your problem is not very detailed but it seems to me that the problem you have is quite balanced. If this is not the case then go for Cilk, its work stealing approach its probably the best solution for unbalanced workloads.
At the time this was posted, Intel was putting a lot of effort into boosting Cilk(tm) Plus; more recently, some effort has been diverted toward OpenMP 4.0.
It's difficult in general to contrast OpenMP with Cilk(tm) Plus.
If it's not possible to distribute work evenly across threads, one would likely set schedule(runtime) in an OpenMP version, and then at run time try various values of environment variable, such as OMP_SCHEDULE=guided, OMP_SCHEDULE=dynamic,2 or OMP_SCHEDULE=auto. Those are the closest OpenMP analogies to the way Cilk(tm) Plus work stealing works.
Some sparse matrix functions in Intel MKL library do actually scan the job first and determine how much to allocate to each thread so as to balance work. For this method to be useful, the time spent in serial scanning and allocating has to be of lower order than the time spent in parallel work.
Work-stealing, or dynamic scheduling, may lose much of the potential advantage of OpenMP in promoting cache locality by pinning threads with cache locality e.g. by OMP_PROC_BIND=close.
Poor cache locality becomes a bigger issue on a NUMA architecture where it may lead to significant time spent on remote memory access.
Both OpenMP and Cilk(tm) Plus have facilities for switching between serial and parallel execution.
Almost all of the CUDA exemplar code describes doing near-atomic operations on large data sets. What kind of practical limitations are the to the size of a problem each thread can do?
For example, I have another question open at the minute that involves per-thread matrix solving. Is this kind of thing too large to put within each thread?
CUDA is a data parallel programming model for what is effectively an SIMD architecture, so obviously it isn't as flexible as a general purpose multithreaded or MIMD architecture. Certainly kernels can be a lot more complex than simple arithmetic operations.
In my own work I use CUDA a lot for solving partial differential equations (so the finite element, finite difference and finite volume methods), which every thread processes a cell or element from a discretised continuum. In that sort of calculation, there are a lot of FLOPs per thread per cell/element.
The key area to be mindful of is branch divergence. Because it is an SIMD architecture under the hood, code where there is a lot of branching within a warp of threads (which is effectively the SIMD width), will suffer performance penalties. But branch divergence and code complexity need not be synonymous, you can write very "branchy" and "loopy" code which will run well, as long as threads within any given warp don't diverge too often. In FLOP and IOP heavy algorithms, that is usually not too hard to achieve.
I just want to reiterate talonmies and say that there is no real limit to the "size" of a kernel in number of operations. As long as the computation is parallel, CUDA will be effective!
As far a practical considerations, I would just add a few small notes
long running kernels can timeout, depending on os (or when profiling with cudaProf). You might have to change a setting somewhere to increase maximum kernel execution time.
long running kernels on systems without a dedicated gpu can freeze the display (interrupting ui).
warps are executed asynchronously - one warp can access memory while another performs arithmetic in order to use clock cycles effectively. long running kernels might benefit more from attention to this kind of optimization. i'm not really sure about this last one.