Cumulative sum in two dimensions on array in nested loop -- CUDA implementation? - cuda

I have been thinking of how to perform this operation on CUDA using reductions, but I'm a bit at a loss as to how to accomplish it. The C code is below. The important part to keep in mind -- the variable precalculatedValue depends on both loop iterators. Also, the variable ngo is not unique to every value of m... e.g. m = 0,1,2 might have ngo = 1, whereas m = 4,5,6,7,8 could have ngo = 2, etc. I have included sizes of loop iterators in case it helps to provide better implementation suggestions.
// macro that translates 2D [i][j] array indices to 1D flattened array indices
#define idx(i,j,lda) ( (j) + ((i)*(lda)) )
int Nobs = 60480;
int NgS = 1859;
int NgO = 900;
// ngo goes from [1,900]
// rInd is an initialized (and filled earlier) as:
// rInd = new long int [Nobs];
for (m=0; m<Nobs; m++) {
ngo=rInd[m]-1;
for (n=0; n<NgS; n++) {
Aggregation[idx(n,ngo,NgO)] += precalculatedValue;
}
}
In a previous case, when precalculatedValue was only a function of the inner loop variable, I saved the values in unique array indices and added them with a parallel reduction (Thrust) after the fact. However, this case has me stumped: the values of m are not uniquely mapped to the values of ngo. Thus, I don't see a way of making this code efficient (or even workable) to use a reduction on. Any ideas are most welcome.

Just a stab...
I suspect that transposing your loops might help.
for (n=0; n<NgS; n++) {
for (m=0; m<Nobs; m++) {
ngo=rInd[m]-1;
Aggregation[idx(n,ngo,NgO)] += precalculatedValue(m,n);
}
}
The reason I did this is because idx varies more rapidly with ngo (function of m) than with n, so making m the inner loop improves coherence. Note I also made precalculatedValue a function of (m, n) because you said that it is -- this makes the pseudocode clearer.
Then, you could start by leaving the outer loop on the host, and making a kernel for the inner loop (64,480-way parallelism is enough to fill most current GPUs).
In the inner loop, just start by using an atomicAdd() to handle collisions. If they are infrequent, on Fermi GPUs performance shouldn't be too bad. In any case, you will be bandwidth bound since arithmetic intensity of this computation is low. So once this is working, measure the bandwidth you are achieving, and compare to the peak for your GPU. If you are way off, then think about optimizing further (perhaps by parallelizing the outer loop -- one iteration per thread block, and do the inner loop using some shared memory and thread cooperation optimizations).
The key: start simple, measure performance, and then decide how to optimize.
Note that this calculation looks very similar to a histogram calculation, which has similar challenges, so you might want to google for GPU histograms to see how they have been implemented.
One idea is to sort (rInd[m], m) pairs using thrust::sort_by_key() and then (since the rInd duplicates will be grouped together), you can iterate over them and do your reductions without collisions. (This is one way to do histograms.) You could even do this with thrust::reduce_by_key().

Related

Chisel Synchronization

Here is a description of one of the states in my state machine. What I would like to do is to go to the next state after the for loops.
is(s_multiplier){
when(ready){state := s_ready}
// Initialization of C memory to 0
for(i <- 0 to matrixSize - 1){
for(j <- 0 to matrixSize - 1){
memC.write(i + j, 0.asSInt((2 * cellSize).W))
}
}
// Objective 1 : Multiplication for the 128X128
// Objective 2 : Multiplication for the n.m and m.p size parameters given
for(i <- 0 to matrixSize - 1){
for(j <- 0 to matrixSize - 1){
sum := 0.asSInt(cellSize.W)
for(k <- 0 to matrixSize - 1){
sum = sum + memA.read(i * matrixSize + k, true.B) * memB.read(k * matrixSize + j, true.B)
}
memC.write(i * matrixSize + j, sum)
}
}
ready := true.B
}
I just created a boolean variable ready that I put to true after the loops. But as everything is supposed to be executed in parallel, I Don't think that my code is correct :/
There is a fundamental difference between writing software algorithms and using chisel to construct the hardware necessary to perform equivalent calculations.
Before discussing the matrix multiplication, consider (as a simpler example) your memory initialization operation loop. The way you have done it makes sense, but for hardware every time the inner body of the loop is executed the hardware necessary to init that memory cell is added to the hardware graph. That means you have created the necessary wires to initialize 16384 memory locations all at the same time. That a lot of wires. Not only that, it would require a memory that has 16384 write ports (you probably can't find that). Your hardware would initialize all this memory in one clock cycle, which is good, but by devoting an enormous number of gates to do so.
Typically one would initialize memory over a number of clock cycles and in this way reducing the amount of hardware required.
Similarly in the matrix multiplication section you are generating all the hardware necessary to compute a matrix multiplication in 1 clock cycle. This is great for performance but the number of multiplications required for this approach is 2,097,152 hardware multipliers plus a further large number of adders. Every * and + operation in the inner loop generates hardware. The number of gates required to multiply two 32 bit numbers is roughly 1024 gates.
The way to go about this is to figure out a way of breaking down the problem into stages. Maybe this would be module that can multiply one row by one column and sum the total. You would then need to use registers to work your way through the matrix, keeping track of the row and columns in order to compute the value at every point in the result matrix. In order to reduce the number of hardware elements you instead perform the calculation over multiple clock cycles keeping state information (indices to the rows and columns) on the progress of the calculation in registers or in memory.
There's a lot of ways to try and optimize a function this and Chisel is a great language for experimenting and testing out tactics.
Maybe you want to make the memory very wide to accommodate getting multiple cell values at once.
Maybe you will unroll your loop a bit more to compute multiple cell values at once by having more than one cell calculator.
Clever iteration strategies can optimize your memory accesses for both reading and writing.
The point is that writing hardware is not necessary harder than writing software (and Chisel helps there) but it is pretty different in the approach.
I would recommend you spend a little more time with Chisel bootcamp. The 2.3_control_flow page's section on sorting is pretty similar with respect to the discussion above. You can write a one cycle sorter but the size of the hardware to do it grows rapidly, in practice it is necessary to break the problem into pieces and spread the calculation over multiple cycles.
Good luck.

how to sync threads in this cuda example

I have the following rough code outline:
run a loop, millions of times
in that loop, compute values 'I's - see example of such functions below
After all 'I's have been computed, compute other values 'V's
repeat the loop
Each computation of an I or V could involve up to 20ish mathematical operations, (e.g. I1 = A + B/C * D + 1/exp(V1) - E + F + V2 etc).
There are roughly:
50 'I's
10 'V's
10 values in each I and V, i.e. they are vectors of length 10
At first I tried running a simple loop in C, with kernel calls for each time step but this was really slow. It seems like I can get the code to run faster if the main loop is in a kernel that calls other kernels. However, I'm worried about kernel call overhead (maybe I shouldn't be) so I came up with something like the following, where each I and V loop independently, with syncing between the kernels as necessary.
For reference, the variables below are hardcoded as __device__ values, but eventually I will pass some values into specific kernels to make the system interesting.
__global__ void compute_IL1()
{
int id = threadIdx.x;
//n_t = 1e6;
for (int i = 0; i < n_t; i++){
IL1[id] = gl_1*(V1[id] - El_1);
//atomic, sync, event????,
}
}
__global__ void compute_IK1()
{
int id = threadIdx.x;
for (int i = 0; i < n_t; i++){
Ik1[id] = gk_1*powf(0.75*(1-H1[id]),4)*(V1[id]-Ek_1);
//atomic, sync, event?
}
}
__global__ void compute_V1()
{
int id = threadIdx.x;
for (int i = 0; i < n_t; i++){
//wait for IL1 and Ik1 and others, but how????
V1[id] = Ik1[id]+IL1[id] + ....
//trigger the I's again
}
}
//main function
compute_IL1<<<1,10,0,s0>>>();
compute_IK1<<<1,10,0,s1>>>();
//repeat this for many 50 - 70 more kernels (Is and Vs)
So the question is, how would I sync these kernels? Is an event approach best? Is there a better paradigm to use here?
There is no sane mechanism I can think of to have multiple resident kernels synchronize without resorting to hacky atomic tricks which may well not work reliably.
If you are running blocks with 10 threads and these kernels cannot execute concurrently for correctness reasons, you are (in the best possible case) using 1/64 of the computational capacity of your device. This problem as you have described it sounds completely Ill suited to a GPU.
So, I tried a couple of approaches.
A loop with a few kernel calls, where the last kernel call is dependent on the previous ones. This can be done with cudaStreamWaitEvent which can wait for multiple events. I found this on: http://cedric-augonnet.com/declaring-dependencies-with-cudastreamwaitevent/ . Unfortunately, the kernel calls were too expensive.
Global variables between concurrent streams. The logic was pretty simple, having one thread pause until a global variable equaled the loop variable, indicating that all threads could proceed. This was then followed by a sync-threads call. Unfortunately, this did not work well.
Ultimately, I think I've settled on a nested loop, where the outer loop represents time, and the inner loop indicates which of a set instructions to run, based on dependencies. I also launched the maximum number of threads per block (1024) and broke up the vectors that needed to be processed into warps. The rough psuedocode is:
run_main<<<1,1024>>>();
__global__ void run_main(){
int warp = threadIdx.x/32;
int id = threadIdx.x - warp*32;
if (id < 10){
for (int i = 0; i < n_t; i++){
for(int j = 0; j < n_j; j++){
switch (j){
case 0:
switch(warp){
case 0:
I1[id] = a + b + c*d ...
break;
case 1:
I2[id] = f*g/h
break;
}
break;
//These things depend on case 0 OR
//we've run out of space in the first pass
//32 cases max [0 ... 31]
case 1:
switch(warp){
case 0:
V1[ID] = I1*I2+ ...
break;
case 1:
V2[ID] = ...
//syncs across the block
__syncthreads();
This design is based on my impression that each set of 32 threads runs independently but should run the same code, otherwise things can slow done significantly.
So at the end, I'm running roughly 32*10 instructions simultaneously.
Where 32 is the number of warps, and it depends on how many different values I can compute at the same time (due to dependencies) and 10 is the # of elements in each vector. This is slowed down by any imbalances in the # of computations in each warp case, since all warps need to merge before moving onto the next step (due to the syncthreads call). I'm running different parameters (parameter sweep) on top of this, so I could potentially run 3 at a time in the block, multiplied by the # of streaming processors (or whatever the official name is) on the card.
One thing I need to change is that I'm currently testing on a video card that is attached to a monitor as well. Apparently Windows will kill the kernel if it lasts for more than 5 seconds, so I need to call the kernel in chunked time steps, like once every 1e5 time steps (in my case).

CUDA Atomic operation on array in global memory

I have a CUDA program whose kernel basically does the following.
I provide a list of n points in cartesian coordinates e.g. (x_i,y_i) in a plane of dimension dim_x * dim_y. I invoke the kernel accordingly.
For every point on this plane (x_p,y_p) I calculate by a formula the time it would take for each of those n points to reach there; given those n points are moving with a certain velocity.
I order those times in increasing order t_0,t_1,...t_n where the precision of t_i is set to 1. i.e. If t'_i=2.3453 then I would only use t_i=2.3.
Assuming the times are generated from a normal distribution I simulate the 3 quickest times to find the percentage of time those 3 points reached earliest. Hence suppose prob_0 = 0.76,prob_1=0.20 and prob_2=0.04 by a random experiment. Since t_0 reaches first most amongst the three, I also return the original index (before sorting of times) of the point. Say idx_0 = 5 (An integer).
Hence for every point on this plane I get a pair (prob,idx).
Suppose n/2 of those points are of one kind and the rest are of other. A sample image generated looks as follows.
Especially when precision of the time was set to 1 I noticed that the number of unique 3 tuples of time (t_0,t_1,t_2) was just 2.5% of the total data points i.e. number of points on the plane. This meant that most of the times the kernel was uselessly simulating when it could just use the values from previous simulations. Hence I could use a dictionary having key as 3-tuple of times and value as index and prob. Since as far as I know and tested, STL can't be accessed inside a kernel, I constructed an array of floats of size 201000000. This choice was by experimentation since none of the top 3 times exceeded 20 seconds. Hence t_0 could take any value from {0.0,0.1,0.2,...,20.0} thus having 201 choices. I could construct a key for such a dictionary like the following
Key = t_o * 10^6 + t_1 * 10^3 + t_2
As far as the value is concerned I could make it as (prob+idx). Since idx is an integer and 0.0<=prob<=1.0, I could retrieve both of those values later by
prob=dict[key]-floor(dict[key])
idx = floor(dict[key])
So now my kernel looks like the following
__global__ my_kernel(float* points,float* dict,float *p,float *i,size_t w,...){
unsigned int col = blockIdx.y*blockDim.y + threadIdx.y;
unsigned int row = blockIdx.x*blockDim.x + threadIdx.x;
//Calculate time taken for each of the points to reach a particular point on the plane
//Order the times in increasing order t_0,t_1,...,t_n
//Calculate Key = t_o * 10^6 + t_1 * 10^3 + t_2
if(dict[key]>0.0){
prob=dict[key]-floor(dict[key])
idx = floor(dict[key])
}
else{
//Simulate and find prob and idx
dict[key]=(prob+idx)
}
p[row*width+col]=prob;
i[row*width+col]=idx;
}
The result is quite similar to the original program for most points but for some it is wrong.
I am quite sure that this is due to race condition. Notice that dict was initialized with all zeroes. The basic idea would be to make the data structure "read many write once" in a particular location of the dict.
I am aware that there might be much more optimized ways of solving this problem rather than allocating so much memory. Please let me know in that case. But I would really like to understand why this particular solution is failing. In particular I would like to know how to use atomicAdd in this setting. I have failed to use it.
Unless your simulation in the else branch is very long (~100s of floating-point operations), a lookup table in global memory is likely to be slower than running the computation. Global memory access is very expensive!
In any case, there is no way to save time by "skipping work" using conditional branching. The Single Instruction, Multiple Thread architecture of a GPU means that the instructions for both sides of the branch will be executed serially, unless all of the threads in a block follow the same branch.
edit:
The fact that you are seeing a performance increase as a result of introducing the conditional branch and you didn't have any problems with deadlock suggests that all the threads in each block are always taking the same branch. I suspect that once dict starts getting populated, the performance increase will go away.
Perhaps I have misunderstood something, but if you want to calculate the probability of an event x, assuming a normal distribution and given the mean mu and standard deviation sigma, there is no need to generate a load of random numbers and approximate a Gaussian curve. You can directly calculate the probability:
p = exp(-((x - mu) * (x - mu) / (2.0f * sigma * sigma))) /
(sigma * sqrt(2.0f * M_PI));

Benefit of splitting a big CUDA kernel and using dynamic parallelism

I have a big kernel in which an initial state is evolved using different techniques. That is, I have a loop in the kernel, in this loop a certain predicate is evaluated on the current state and on the result of this predicate, a certain action is taken.
The kernel needs a bit of temporary data and shared memory, but since it is big it uses 63 registers and the occupancy is very very low.
I would like to split the kernel in many little kernels, but every block is totally independent from the others and I (think I) can't use a single thread on the host code to launch multiple small kernels.
I am not sure if streams are adequate for this kind of work, I never used them, but since I have the option to use the dynamic parallelism, I would like if that is a good option to implement this kind of job.
Is it fast to launch a kernel from a kernel?
Do I need to copy data in global memory to make them available to a sub-kernel?
If I split my big kernel in many little ones, and leave the first kernel with a main loop which calls the required kernel when necessary (which allows me to move temporary variables in every sub-kernel), will help me increase the occupancy?
I know it is a bit generic question, but I do not know this technology and I would like if it fits my case or if streams are better.
EDIT:
To provide some other details, you can imagine my kernel to have this kind of structure:
__global__ void kernel(int *sampleData, int *initialData) {
__shared__ int systemState[N];
__shared__ int someTemp[N * 3];
__shared__ int time;
int tid = ...;
systemState[tid] = initialData[tid];
while (time < TIME_END) {
bool c = calc_something(systemState);
if (c)
break;
someTemp[tid] = do_something(systemState);
c = do_check(someTemp);
if (__syncthreads_or(c))
break;
sample(sampleData, systemState);
if (__syncthreads_and(...)) {
do_something(systemState);
sync();
time += some_increment(systemState);
}
else {
calcNewTemp(someTemp, systemState);
sync();
do_something_else(someTemp, systemState);
time += some_other_increment(someTemp, systemState);
}
}
do_some_stats();
}
this is to show you that there is a main loop, that there are temporary data which are used somewhere and not in other points, that there are shared data, synchronization points, etc.
Threads are used to compute vectorial data, while there is, ideally, one single loop in each block (well, of course it is not true, but logically it is)... One "big flow" for each block.
Now, I am not sure about how to use streams in this case... Where is the "big loop"? On the host I guess... But how do I coordinate, from a single loop, all the blocks? This is what leaves me most dubious. May I use streams from different host threads (One thread per block)?
I am less dubious about dynamic parallelism, because I could easily keep the big loop running, but I am not sure if I could have advantages here.
I have benefitted from dynamic parallelism for solving an interpolation problem of the form:
int i = threadIdx.x + blockDim.x * blockIdx.x;
for(int m=0; m<(2*K+1); m++) {
PP1 = calculate_PP1(i,m);
phi_cap1 = calculate_phi_cap1(i,m);
for(int n=0; n<(2*K+1); n++) {
PP2 = calculate_PP2(i,m);
phi_cap2 = calculate_phi_cap2(i,n);
atomicAdd(&result[PP1][PP2],data[i]*phi_cap1*phi_cap2); } } }
where K=6. In this interpolation problem, the computation of each addend is independent of the others, so I have split them in a (2K+1)x(2K+1) kernel.
From my (possibly incomplete) experience, dynamic parallelism will help if you have a few number of independent iterations. For larger number of iterations, perhaps you could end up by calling the child kernel several times and so you should check if the overhead in kernel launch will be the limiting factor.

What is an effective way to filter numbers according to certain number-ranges using CUDA?

I have a lot of random floating point numbers residing in global GPU memory. I also have "buckets" that specify ranges of numbers they will accept and a capacity of numbers they will accept.
ie:
numbers: -2 0 2 4
buckets(size=1): [-2, 0], [1, 5]
I want to run a filtration process that yields me
filtered_nums: -2 2
(where filtered_nums can be a new block of memory)
But every approach I take runs into a huge overhead of trying to synchronize threads across bucket counters. If I try to use a single-thread, the algorithm completes successfully, but takes frighteningly long (over 100 times slower than generating the numbers in the first place).
What I am asking for is a general high-level, efficient, as-simple-as-possible approach algorithm that you would use to filter these numbers.
edit
I will be dealing with 10 buckets and half a million numbers. Where all the numbers fall into exactly 1 of the 10 bucket ranges. Each bucket will hold 43000 elements. (There are excess elements, since the objective is to fill every bucket, and many numbers will be discarded).
2nd edit
It's important to point out that the buckets do not have to be stored individually. The objective is just to discard elements that would not fit into a bucket.
You can use thrust::remove_copy_if
struct within_limit
{
__host__ __device__
bool operator()(const int x)
{
return (x >=lo && x < hi);
}
};
thrust::remove_copy_if(input, input + N, result, within_limit());
You will have to replace lo and hi with constants for each bin..
I think you can templatize the kernel, but then again you will have to instantiate the template with actual constants. I can't see an easy way at it, but I may be missing something.
If you are willing to look at third party libraries, arrayfire may offer an easier solution.
array I = array(N, input, afDevice);
float **Res = (float **)malloc(sizeof(float *) * nbins);
for(int i = 0; i < nbins; i++) {
array res = where(I >= lo[i] && I < hi[i]);
Res[i] = res.device<float>();
}