Using OpenMP "for simd" in matrix-vector multiplication? - parallel-processing

I'm currently trying to get my matrix-vector multiplication function to compare favorably with BLAS by combining #pragma omp for with #pragma omp simd, but it's not getting any speedup improvement than if I were to just use the for construct. How do I properly vectorize the inner loop with OpenMP's SIMD construct?
vector dot(const matrix& A, const vector& x)
{
assert(A.shape(1) == x.size());
vector y = xt::zeros<double>({A.shape(0)});
int i, j;
#pragma omp parallel shared(A, x, y) private(i, j)
{
#pragma omp for // schedule(static)
for (i = 0; i < y.size(); i++) { // row major
#pragma omp simd
for (j = 0; j < x.size(); j++) {
y(i) += A(i, j) * x(j);
}
}
}
return y;
}

Your directive is incorrect because there would introduce in a race condition (on y(i)). You should use a reduction in this case. Here is an example:
vector dot(const matrix& A, const vector& x)
{
assert(A.shape(1) == x.size());
vector y = xt::zeros<double>({A.shape(0)});
int i, j;
#pragma omp parallel shared(A, x, y) private(i, j)
{
#pragma omp for // schedule(static)
for (i = 0; i < y.size(); i++) { // row major
decltype(y(0)) sum = 0;
#pragma omp simd reduction(+:sum)
for (j = 0; j < x.size(); j++) {
sum += A(i, j) * x(j);
}
y(i) += sum;
}
}
return y;
}
Note that it may not be necessary faster because some compilers are able to automatically vectorize the code (ICC for example). GCC and Clang often fail to perform (advanced) SIMD reductions automatically and such a directive help them a bit. You can check the assembly code to check how the code is vectorized or enable vectorization reports (see here for GCC).

Related

OpenMP Do I have race condition or false-sharing '?

I'm trying to write a code for matrix multiplication. As far as I understand OMP and pararel programming this code may suffer from race condition.
#pragma omp parallel
#pragma omp for
for (int k = 0; k < size; k++){
for (int i = 0; i < size; i++) {
for (int j = 0; j < size; j++) {
c[i][j] += a[i][k] * b[k][j];
}}}
Do I get rid of it if I put #pragma omp atomic before writing to c matrix or by adding private(i) to 2nd #pragma? Also is it possible to make this code false-sharing free? If yes, how ?
A race condition occurs when 2 or more threads access the same memory location and at least one of them is writing it. Line c[i][j] +=... can cause data race in your code. The solution is to reorder your nested loops (use the order of i,j,k) and you may introduce a temporary variable to calculate the dot product:
#pragma omp parallel for
for (int i = 0; i < size; i++) {
for (int j = 0; j < size; j++) {
double tmp=0; // change its type as needed
for (int k = 0; k < size; k++){
tmp += a[i][k] * b[k][j];
}
c[i][j] = tmp; //note that += was used in your original code
}
}
Note that your code will be faster if you calculate the transpose of matrix b. For more details read this.
UPDATE:
If you need to maintain the order of loops, there are 2 possibilities (but these solutions may be slower than the serial code):
Use atomic operation (i.e #pragma omp atomic). In this case false sharing also can be a problem.
If your stack is large enough to store the matrix for all threads, a better alternative is to use reduction: #pragma omp parallel for reduction(+:c[:size][:size]) (Another alternative is to do the reduction manually. In this case you can allocate the matrices used for reduction on the heap.)

OpenMP reduction on SSE2 vector

I want to compute the average of an image (3 channels of interest + 1 alpha channel we ignore here) for each channel using SSE2 intrinsics. I tried that:
__m128 average = _mm_setzero_ps();
#pragma omp parallel for reduction(+:average)
for(size_t k = 0; k < roi_out->height * roi_out->width * ch; k += ch)
{
float *in = ((float *)temp) + k;
average += _mm_load_ps(in);
}
But I get this error with GCC: user-defined reduction not found for average.
Is that possible with SSE2 ? What's wrong ?
Edit
This works:
float sum[4] = { 0.0f };
#pragma omp parallel for simd reduction(+:sum[:4])
for(size_t k = 0; k < roi_out->height * roi_out->width * ch; k += ch)
{
float *in = ((float *)temp) + k;
for (int i = 0; i < ch; ++i) sum[i] += in[i];
}
const __m128 average = _mm_load_ps(sum) / ((float)roi_out->height * roi_out->width);
You can user-define a custom reduction like this:
#pragma omp declare reduction \
(addps:__m128:omp_out+=omp_in) \
initializer(omp_priv=_mm_setzero_ps())
And then use it like:
#pragma omp parallel for reduction(addps:average)
for(size_t k = 0; k < size * ch; k += ch)
{
average += _mm_loadu_ps(data+k);
}
I think, most importantly, openmp needs to know how to get a neutral element (here _mm_setzero_ps()) for your reduction.
Full working example: https://godbolt.org/z/Fpqttc
Interesting link: http://pages.tacc.utexas.edu/~eijkhout/pcse/html/omp-reduction.html#User-definedreductions

How to distribute teams on GPU using OpenMP?

i'm trying to utilize my Nvidia Geforce GT 740M for parallel-programming using OpenMP and the clang-3.8 compiler.
When processed in parallel on the CPU, I manage to get the desired result. However, when processed on the GPU, my results are some almost random numbers.
Therefore, I figured that I'm not correctly distributing my thread teams and that there might be some data races. I guess I have to do my for-loops differently but I have no idea where the mistake could be.
#include <stdio.h>
#include <stdlib.h>
#include <omp.h>
int main(int argc, char* argv[])
{
const int n =100; float a = 3.0f; float b = 2.0f;
float *x = (float *) malloc(n * sizeof(float));
float *y = (float *) malloc(n * sizeof(float));
int i;
int j;
int k;
double start;
double end;
start = omp_get_wtime();
for (k=0; k<n; k++){
x[k] = 2.0f;
y[k] = 3.0f;
}
#pragma omp target data map(to:x[0:n]) map(tofrom:y[0:n]) map(to:i) map(to:j)
{
#pragma omp target teams
#pragma omp distribute
for(i = 0; i < n; i++) {
#pragma omp parallel for
for (j = 0; j < n; j++){
y[j] = a*x[j] + y[j];
}
}
}
end = omp_get_wtime();
printf("Work took %f seconds.\n", end - start);
free(x); free(y);
return 0;
}
I guess that it might have something to to with the Architecture of my GPU. So therefore I'm adding this:
Im fairly new to the topic, so thanks for your help :)
Yes, there is a race here. Different teams are reading and writing to the same element of the array 'y'. Perhaps you want something like this?
for(i = 0; i < n; i++) {
#pragma omp target teams distribute parallel for
for (j = 0; j < n; j++){
y[j] = a*x[j] + y[j];
}
}

openmp, for loop parallelization and critical zone error

I am new to OpenMP and I am using it to implement the Sieve of Eratosthenes, My code are:
int check_eratothenes(int *p, int pn, int n)
{
int count = 0;
bool* out = new bool[int(pow(pn, 2))];
memset(out, 0, pow(pn, 2));
#pragma omp parallel
for (int i = 0; i < n; i ++)
{
int j = floor((pn + 1) / p[i]) * p[i];
#pragma omp critical
while (j <= pow(pn, 2))
{
out[j] = 1;
j += p[i];
}
}
#pragma omp parallel
for (int i = pn+1; i < pow(pn, 2); i ++)
{
#pragma omp critical
if (out[i] == 0)
{
//cout << i << " ";
count ++;
}
}
return count;
}
But, the above OpenMP pragma is wrong. It can be complied but when it runs, it takes a lot of time to get the result, so it press CTRL + C to stop. And I felt at a loss on how to solve it . Since there are many loops and if statements.
Thanks in advance.

openmp parallel for over struct members?

Is it possible to parallelize loop over stuct members with OpenMP?
I tried the following with GCC
point_t p;
double sum;
#pragma omp parallel for private(p) reduction(+: sum)
for (p.x = 0; p.x < N; p.x++) {
for (p.y = 0; p.y < N; p.y++) {
sum += foo(p);
}
}
But that gives me a compile error
error: expected iteration declaration or initialization before ‘p
Is this a GCC bug or is it not part of the OpenMP specs?
I don't think this is allowed in OpenMP; parallel for needs to loop over a variable, not a general lvalue. Do
int x, y; // or whatever you store in a point_t
double sum;
#pragma omp parallel for reduction(+:sum)
for (x=0; x<N; x++)
for (y=0; y<N; y++) {
point_t p(x, y); // assuming C++
sum += foo(p);
}

Resources