Let's say that I have a variable x.
x = 0
I then spawn some number of threads, and each of them may or may not run the following expression WITHOUT the use of atomics.
x |= 1
After all threads have joined with my main thread, the main thread branches on the value.
if(x) { ... } else { ... }
Is it possible for there to be a race condition in this situation? My thoughts say no, because it doesn't seem to matter whether or not a thread is interrupted by another thread between reading and writing 'x' (in both cases, either 'x == 1', or 'x == 1'). That said, I want to make sure I'm not missing something stupid obvious or ridiculously subtle.
Also, if you happen to provide an answer to the contrary, please provide an instruction-by-instruction example!
Context:
I'm trying to, in OpenCL, have my threads indicate the presence or absence of a feature among any of their work-items. If any of the threads indicate the presence of the feature, my host ought to be able to branch on the result. I'm thinking of using the above method. If you guys have a better suggestion, that works too!
Detail:
I'm trying to add early-exit to my OpenCL radix-sort implementation, to skip radix passes if the data is banded (i.e. 'x' above would be x[RADIX] and I'd have all work groups, right after partial reduction of the data, indicate presence or absence of elements in the RADIX bins via 'x').
It may work within a work-group. You will need to insert a barrier before testing x. I'm not sure it will be faster than using atomic increments.
It will not work across several work-groups. Imagine you have 1000 work-groups to run on 20 cores. Typically, only a small number of work-groups can be resident on a single core, for example 4, meaning only 80 work-groups can be in flight inside the GPU at a given time. Once a work-group is done executing, it is retired, and another one is started. Halting a kernel in the middle of execution to wait for all 1000 work-groups to reach the same point is impossible.
Related
I am trying to develop the following lock free code (c++11):
int val_max;
std::array<std::atomic<int>, 255> vector;
if (vector[i] > val_max) {
val_max =vector[i];
}
The problem is that when there are many threads (128 threads), the result is not correct, because if for example val_max=1, and three threads ( with vector[i] = 5, 15 and 20) will execute the code in the next line, and it will be a data race..
So, I don't know the best way to solve the problem with the available functions in c++11 (I cannot use mutex, locks), protecting the whole code.
Any suggestions? Thanks in advance!
You need to describe the bigger problem you need to solve, and why you want multiple threads for this.
If you have a lot of data and want to find the maximum, and you want to split that problem, you're doing it wrong. If all threads try to access a shared maximum, not only is it hard to get right, but by the time you have it right, you have fully serialized accesses, thus making the entire thing an exercise in adding complexity and thread overhead to a serial program.
The correct way to make it parallel is to give every thread a chunk of the array to work on (and the array members are not atomics), for which the thread calculates a local maximum, then when all threads are done have one thread find the maximum of the individual results.
Do an atomic fetch of val_max.
If the value fetched is greater than or equal to vector[i], stop, you are done.
Do an atomic compare exchange -- compare val_max to the value you read in step 1 and exchange it for the value of vector[i] if it compares.
If the compare succeeded, stop, you are done.
Go to step 1, you raced with another thread that made forward progress.
I've seen many questions scattered across the Internet about branch divergence, and how to avoid it. However, even after reading dozens of articles on how CUDA works, I can't seem to see how avoiding branch divergence helps in most cases. Before anyone jumps on on me with claws outstretched, allow me to describe what I consider to be "most cases".
It seems to me that most instances of branch divergence involve a number of truly distinct blocks of code. For example, we have the following scenario:
if (A):
foo(A)
else:
bar(B)
If we have two threads that encounter this divergence, thread 1 will execute first, taking path A. Following this, thread 2 will take path B. In order to remove the divergence, we might change the block above to read like this:
foo(A)
bar(B)
Assuming it is safe to call foo(A) on thread 2 and bar(B) on thread 1, one might expect performance to improve. However, here's the way I see it:
In the first case, threads 1 and 2 execute in serial. Call this two clock cycles.
In the second case, threads 1 and 2 execute foo(A) in parallel, then execute bar(B) in parallel. This still looks to me like two clock cycles, the difference is that in the former case, if foo(A) involves a read from memory, I imagine thread 2 can begin execution during that latency, which results in latency hiding. If this is the case, the branch divergent code is faster.
You're assuming (at least it's the example you give and the only reference you make) that the only way to avoid branch divergence is to allow all threads to execute all the code.
In that case I agree there's not much difference.
But avoiding branch divergence probably has more to do with algorithm re-structuring at a higher level than just the addition or removal of some if statements and making code "safe" to execute in all threads.
I'll offer up one example. Suppose I know that odd threads will need to handle the blue component of a pixel and even threads will need to handle the green component:
#define N 2 // number of pixel components
#define BLUE 0
#define GREEN 1
// pixel order: px0BL px0GR px1BL px1GR ...
if (threadIdx.x & 1) foo(pixel(N*threadIdx.x+BLUE));
else bar(pixel(N*threadIdx.x+GREEN));
This means that every alternate thread is taking a given path, whether it be foo or bar. So now my warp takes twice as long to execute.
However, if I rearrange my pixel data so that the color components are contiguous perhaps in chunks of 32 pixels:
BL0 BL1 BL2 ... GR0 GR1 GR2 ...
I can write similar code:
if (threadIdx.x & 32) foo(pixel(threadIdx.x));
else bar(pixel(threadIdx.x));
It still looks like I have the possibility for divergence. But since the divergence happens on warp boundaries, a give warp executes either the if path or the else path, so no actual divergence occurs.
This is a trivial example, and probably stupid, but it illustrates that there may be ways to work around warp divergence that don't involve running all the code of all the divergent paths.
I'm faced with parallelizing an algorithm which in its serial implementation examines the six faces of a cube of array locations within a much larger three dimensional array. (That is, select an array element, and then define a cube or cuboid around that element 'n' elements distant in x, y, and z, bounded by the bounds of the array.
Each work unit looks something like this (Fortran pseudocode; the serial algorithm is in Fortran):
do n1=nlo,nhi
do o1=olo,ohi
if (somecondition(n1,o1) .eq. .TRUE.) then
retval =.TRUE.
RETURN
endif
end do
end do
Or C pseudocode:
for (n1=nlo,n1<=nhi,n++) {
for (o1=olo,o1<=ohi,o++) {
if(somecondition(n1,o1)!=0) {
return (bool)true;
}
}
}
There are six work units like this in the total algorithm, where the 'lo' and 'hi' values generally range between 10 and 300.
What I think would be best would be to schedule six or more threads of execution, round-robin if there aren't that many CPU cores, ideally with the loops executing in parallel, with the goal the same as the serial algorithm: somecondition() becomes True, execution among all the threads must immediately stop and a value of True set in a shared location.
What techniques exist in a Windows compiler to facilitate parallelizing tasks like this? Obviously, I need a master thread which waits on a semaphore or the completion of the worker threads, so there is a need for nesting and signaling, but my experience with OpenMP is introductory at this point.
Are there message passing mechanisms in OpenMP?
EDIT: If the highest difference between "nlo" and "nhi" or "olo" and "ohi" is eight to ten, that would imply no more than 64 to 100 iterations for this nested loop, and no more than 384 to 600 iterations for the six work units together. Based on that, is it worth parallelizing at all?
Would it be better to parallelize the loop over the array elements and leave this algorithm serial, with multiple threads running the algorithm on different array elements? I'm thinking this from your comment "The time consumption comes from the fact that every element in the array must be tested like this. The arrays commonly have between four million and twenty million elements." The design of implementing the parallelelization of the array elements is also flexible in terms of the number threads. Unless there is a reason that the array elements have to be checked in some order?
It seems that the portion that you are showing us doesn't take that long to execute so making it take less clock time by making it parallel might not be easy ... there is always some overhead to multiple threads, and if there is not much time to gain, parallel code might not be faster.
One possibility is to use OpenMP to parallelize over the 6 loops -- declare logical :: array(6), allow each loop to run to completion, and then retval = any(array). Then you can check this value and return outside the parallelized loop. Add a schedule(dynamic) to the parallel do statement if you do this. Or, have a separate !$omp parallel and then put !$omp do schedule(dynamic) ... !$omp end do nowait around each of the 6 loops.
Or, you can follow the good advice by #M.S.B. and parallelize the outermost loop over the whole array. The problem here is that you cannot have a RETURN inside a parallel loop -- so label the second outermost loop (the largest one within the parallel part), and EXIT that loop -- smth like
retval = .FALSE.
!$omp parallel do default(private) shared(BIGARRAY,retval) schedule(dynamic,1)
do k=1,NN
if(.not. retval) then
outer2: do j=1,NN
do i=1,NN
! --- your loop #1
do n1=nlo,nhi
do o1=olo,ohi
if (somecondition(BIGARRAY(i,j,k),n1,o1)) then
retval =.TRUE.
exit outer2
endif
end do
end do
! --- your loops #2 ... #6 go here
end do
end do outer2
end if
end do
!$omp end parallel do
[edit: the if statement is there presuming that you need to find out if there is at least one element like that in the big array. If you need to figure the condition for every element, you can similarly either add a dummy loop exit or goto, skipping the rest of the processing for that element. Again, use schedule(dynamic) or schedule(guided).]
As a separate point, you might also want to check if it may be a good idea to go through the innermost loop by some larger step (depending on float size), compute a vector of logicals on each iteration and then aggregate the results, eg. smth like if(count(somecondition(x(o1:o1+step,n1,k)))>0); in this case the compiler may be able to vectorize somecondition.
I believe you can do what you want with the task construct introduced in OpenMP 3; Intel Fortran supports tasking in OpenMP. I don't use tasks often so I won't offer you any wonky pseudocode.
You already mentioned the obvious way to stop all threads as soon as any thread finds the ending condition: have each check some shared variable which gives the status of the ending condition, thereby determining whether to break out of the loops. Obviously this is an overhead, so if you decide to take this approach I would suggest a few things:
Use atomics to check the ending condition, this avoids expensive memory flushing as just the variable in question is flushed. Move to OpenMP 3.1, there are some new atomic operations supported.
Check infrequently, maybe like once per outer iteration. You should only be parallelizing large cases to overcome the overhead of multithreading.
This one is optional, but you can try adding compiler hints, e.g. if you expect a certain condition to be false most of the time, the compiler will optimize the code accordingly.
Another (somewhat dirty) approach is to use shared variables for the loop ranges for each thread, maybe use a shared array where index n is for thread n. When one thread finds the ending condition, it changes the loop ranges of all the other threads so that they stop. You'll need the appropriate memory synchronization. Basically the overhead has now moved from checking a dummy variable to synchronizing/checking loop conditions. Again probably not so good to do this frequently, so maybe use shared outer loop variables and private inner loop variables.
On another note, this reminds me of the classic polling versus interrupt problem. Unfortunately I don't think OpenMP supports interrupts where you can send some kind of kill signal to each thread.
There are hacking work-arounds like using a child process for just this parallel work and invoking the operating system scheduler to emulate interrupts, however this is rather tricky to get correct and would make your code extremely unportable.
Update in response to comment:
Try something like this:
char shared_var = 0;
#pragma omp parallel
{
//you should have some method for setting loop ranges for each thread
for (n1=nlo; n1<=nhi; n1++) {
for (o1=olo; o1<=ohi; o1++) {
if (somecondition(n1,o1)!=0) {
#pragma omp atomic write
shared_var = 1; //done marker, this will also trigger the other break below
break; //could instead use goto to break out of both loops in 1 go
}
}
#pragma omp atomic read
private_var = shared_var;
if (private_var!=0) break;
}
}
A suitable parallel approach might be, to let each worker examine a part of the overall problem, exactly as in the serial case and use a local (non-shared) variable for the result (retval). Finally do a reduction over all workers on these local variables into a shared overall result.
I have a function that I use to look up a value based on an index. The value takes some time to calculate, so I want to do it with ParallelMap, and references another similar such function that returns a list of expressions, also based on an index.
However, when I set it all up in a seemingly reasonable fashion, I see some very bizarre behaviour. First, I see that the function appears to work, albeit very slowly. For large indexes, however, the processor activity in Taskmangler stays entirely at zero for an extended period of time (i.e. 2-4 minutes) where all instances of Mathematica are seemingly inert. Then, without the slightest blip of CPU use, a result appears. Is this another case of Mathematica spukhafte Fernwirkung?
That is, I want to create a variable/function that stores an expression, here a list of integers (ListOfInts), and then on the parallel workers I want to perform some function on that expression (here I apply a set of replacement rules and take the Min). I want the result of that function to also be indexed by the same index under another variable/function (IndexedFunk), whose result is then available back on the main instance of Mathematica:
(*some arbitrary rules that will convert some of the integers to negative values:*)
rulez=Dispatch[Thread[Rule[Range[222],-Range[222]]]];
maxIndex = 333;
Clear[ListOfInts]
Scan[(ListOfInts[#]=RandomInteger[{1,999},55])&,Range[maxIndex ]]
(*just for safety's sake:*)
DistributeDefinitions[rulez, ListOfInts]
Clear[IndexedFunk]
(*I believe I have to have at least one value of IndexedFunk defined before I Share the definition to the workers:*)
IndexedFunk[1]=Min[ListOfInts[1]]/.rulez
(*... and this should let me retrieve the values back on the primary instance of MMA:*)
SetSharedFunction[IndexedFunk]
(*Now, here is the mysterious part: this just sits there on my multiprocessor machine for many minutes until suddenly a result appears. If I up maxIndex to say 99999 (and of course re-execute the above code again) then the effect can more clearly be seen.*)
AbsoluteTiming[Short[ParallelMap[(IndexedFunk[#]=Min[ListOfInts[#]/.rulez])&, Range[maxIndex]]]]
I believe this is some bug, but then I am still trying to figure out Mathematica Parallel, so I can't be too confident in this conclusion. Despite its being depressingly slow, it is nonetheless impressive in its ability to perform calculations without actually requiring a CPU to do so.
I thought perhaps it was due to whatever communications protocol is being used between the master and slave processes, perhaps it is so slow that it just appears that the processors are doing nothing when if fact they are just waiting to send the next bit of some definition or other. In which case I thought ParallelMap[..., Method->"CoarsestGrained"] would be of some use. But no, that doesn't work neither.
A question: "Am I doing something obviously wrong, or is this a bug?"
I am afraid you are. The problem is with the shared definition of a variable. Mathematica maintains a single coherent value in all copies of the variable across kernels, and therefore that variable becomes a single point of huge contention. CPU is idle because kernels line up to the queue waiting for the variable IndexedFunk, and most time is spent in interprocess or inter-machine communication. Go figure.
By the way, there is no function SetSharedDefinition in any Mathematica version I know of. You probably intended to write SetSharedVariable. But remove that evil call anyway! To avoid contention, return results from the parallelized computation as a list of pairs, and then assemble them into downvalues of your variable at the main kernel:
Clear[IndexedFunk]
Scan[(IndexedFunk[#[[1]]] = #[[2]]) &,
ParallelMap[{#, Min[ListOfInts[#] /. rulez]} &, Range[maxIndex]]
]
ParallelMap takes care of distributing definition automagically, so the call to DistributeDefinitions is superfluous. (As a minor note, it is not correct as written, omitting the maxIndex variable, but the omission is automatically taken care of by ParallelMap in this particular case.)
EDIT, NB!: The automatic distribution applies only to the version 8 of Mathematica. Thanks #MikeHoneychurch for the correction.
What is atomic in C/C++ programming ?
I just visited the dearly cppreference.com (well I don't take the title for granted but wait for my story to finish), and the home changed to describe some of the C++0x/C++1x (let's call it C+++, okay ?) new features.
There was a mysterious and never seen by my zombie programmer's eye, the new <atomic>.
I guess its purpose is not to program atomic bombs or black holes (but I highly doubt this could have ANY connection with black holes, I don't know how those 2 words slipped here), but I'd like to know something:
What is the purpose of this feature ? Is it a type ? A function ? Is it a data container ? Is it related to threads ? May it have some relation with python's "import antigravity" ? I mean, we are programming here, we're not bloody physicist or semanticists !
Atomic refers to something which is not divisible.
An atomic expression is one that is actually executed by a single operation.
For example a++ is not atomic, since to exec it you need first to get the value of a, then to sum 1 to it, then to store the result into a.
Reading the value of an int should instead be atomic.
Atomic-ness is important in shared-memory parallel computations (eg: when using threads): because it tells you that an expression will give you the result you're expecting no matter what the other threads are doing.
AFAIK you could use atomic functions to create your own semaphores etc. The name atomic came from atom, you cant break it smaller, so those function calls can't be "broken apart" and paused by the operating system. This is for thread programming.
Is intended for multithreading. It avoids you to have concurrent threads mix operations. An atomic operation is an indivisible operation. You can’t observe such an operation half-done from any thread in the system; it’s either done or not done. With an atomic operation you cannot get a data race between threads. In a real world analogy you will use atomic not for physics but for semaphores and other traffic signals on roads. Cars will be threads, roads will be rules, locations will be data. Semaphores will be atomic. You don't need semaphores when there is only one car on all roads, right?