topic Re: Unable to utilize all threads - Core i5-1135G7 in Intel® oneAPI Math Kernel Library

Unable to utilize all threads - Core i5-1135G7

muzamil-py — Sun, 07 Jan 2024 18:01:05 GMT

Hi everyone, I am facing a problem.

Problem:
I installed OneAPI Base Toolkit on my Dell Inspiron 13 7306 which is having Core i5-1135G7. I am running both Windows and Linux (dual boot) on it. First I installed OneAPI Base Toolkit on Linux. After which I compiled and executed the following code but I am unable to use Max Number of Cores / Threads i.e 4 Cores and 8 Threads. In my code I have used

"int numThreads = mkl_get_max_threads();" which returns 1.

I have used:

"mkl_set_num_threads(4);" to increase the number of threads but it does not increase the cores.

I have tested it on Windows as well and same issue.

Code:

#include <iostream>

#include <mkl.h>

#include <chrono>

int main() {

mkl_set_num_threads(4);

int numThreads = mkl_get_max_threads();

std::cout << "Number of threads: " << numThreads << std::endl;

// Matrix dimensions

const int m = 10000;

const int n = 10000;

const int k = 10000;

// Allocate memory for matrices A, B, and C

double *A = (double*)mkl_malloc(m * k * sizeof(double), 64); // 64-byte alignment

double *B = (double*)mkl_malloc(k * n * sizeof(double), 64);

double *C = (double*)mkl_malloc(m * n * sizeof(double), 64);

// Initialize matrices A and B with random values

for (int i = 0; i < m * k; ++i)

A[i] = static_cast<double>(rand()) / RAND_MAX;

for (int i = 0; i < k * n; ++i)

B[i] = static_cast<double>(rand()) / RAND_MAX;

auto start = std::chrono::high_resolution_clock::now();

// Perform matrix multiplication using Intel MKL

mkl_set_num_threads(4);

cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, m, n, k, 1.0, A, k, B, n, 0.0, C, n);

auto end = std::chrono::high_resolution_clock::now();

auto duration = std::chrono::duration_cast<std::chrono::seconds>(end - start);

// Print the result (printing a small part for illustration)

const int printSize = 5;

std::cout << "\nMatrix A: \n";

for (int i = 0; i < printSize; ++i) {

for (int j = 0; j < printSize; ++j)

std::cout << A[i * m + j] << " ";

std::cout << "\n";

}

std::cout << "\nMatrix B: \n";

for (int i = 0; i < printSize; ++i) {

for (int j = 0; j < printSize; ++j)

std::cout << B[i * m + j] << " ";

std::cout << "\n";

}

std::cout << "Result (showing first " << printSize << " elements):\n";

for (int i = 0; i < printSize; ++i) {

for (int j = 0; j < printSize; ++j)

std::cout << C[i * n + j] << " ";

std::cout << "\n";

}

std::cout << "Execution Time: " << duration.count() << " seconds" << std::endl;

// Deallocate memory

mkl_free(A);

mkl_free(B);

mkl_free(C);

return 0;

}

Compile: g++ -o code code.cpp -I /opt/intel/oneapi/mkl/latest/include/ -L /opt/intel/oneapi/mkl/latest/lib/ -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread -lm -ldl

Execute: ./code

Is this a bug or it is meant to work like this?

Thanks.

Re: Unable to utilize all threads - Core i5-1135G7

Gennady_F_Intel — Mon, 08 Jan 2024 06:14:43 GMT

You have to link this test against -lmkl_intel_thread instead of -lmkl_sequential.

Re: Unable to utilize all threads - Core i5-1135G7

muzamil-py — Mon, 08 Jan 2024 13:54:05 GMT

SOLVED!

Thank you @Gennady_F_Intel , your answer lead towards the solution.

I used -lmkl_intel_thread intead of -lmkl_sequential. So my compile command was:
g++ -o code code.cpp -I/opt/intel/oneapi/mkl/latest/include -L/opt/intel/oneapi/mkl/latest/lib -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lpthread -lm -ldl

BUT it was giving a new error which was as follows:

/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_critical_with_hint'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_critical'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_init_8'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_float8_add'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_next_4'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_master'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_fini'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_init_4u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_reduce'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_flush'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_next_8u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_init_8u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_init_4'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_fork_call'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_next_8'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_push_num_threads'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_ordered'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_omp_task_alloc'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_init_8u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_barrier'

/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_cmplx8_add'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_omp_task'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_global_thread_num'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_init_4'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_critical'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_serialized_parallel'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_fixed4_add'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_omp_task_with_deps'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_reduce_nowait'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_float4_max'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_master'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_single'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_fini_8'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_init_4u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_init_8'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_fixed8_add'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_cmplx4_add'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_ok_to_fork'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_single'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_next_4u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_ordered'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_reduce_nowait'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_serialized_parallel'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_float4_add'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_float8_max'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_reduce'
collect2: error: ld returned 1 exit status

So here are some changes I made which resolved this error:
1. Used icpx -fsycl as my compiler.

2. Added -liomp5 flag. Reference: Intel MKL Thread

After this my code compiled successfully and executed on max threads or any specified number of threads.

My Final Build Command for the same code is now:

icpx -fsycl -o code code.cpp -I/opt/intel/oneapi/mkl/latest/include -L/opt/intel/oneapi/mkl/latest/lib -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lm -ldl -liomp5 -lpthread

Re: Unable to utilize all threads - Core i5-1135G7

muzamil-py — Mon, 08 Jan 2024 14:09:03 GMT

Thank you @Gennady_F_Intel , your answer led towards solution.

I replaced -lmkl_sequential with -lmkl_intel_thread but I was still getting an error.

Error:

/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_critical_with_hint'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_critical'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_init_8'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_atomic_float8_add'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_next_4'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_master'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_fini'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_init_4u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_end_reduce'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_flush'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_next_8u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_init_8u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_for_static_init_4'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_fork_call'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_next_8'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_push_num_threads'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_ordered'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_omp_task_alloc'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_dispatch_init_8u'
/usr/bin/ld: /opt/intel/oneapi/mkl/latest/lib/libmkl_intel_thread.so: undefined reference to `__kmpc_barrier'

To solve this:

I used icpx compiler instead of g++.
I added -liomp5 flag. (Reference to this change: Intel MKL Thread)

After appending these changes my code successfully compiled and executed with multiple threads.

My final build command now is:

icpx -fsycl -o test test.cpp -I/opt/intel/oneapi/mkl/latest/include -L/opt/intel/oneapi/mkl/latest/lib -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lm -ldl -liomp5 -lpthread

Re: Unable to utilize all threads - Core i5-1135G7

muzamil-py — Mon, 08 Jan 2024 14:10:22 GMT

@Gennady_F_Intel I found a solution but upon posting it, it is being marked spam. Tried posting 2 times.

Re: Unable to utilize all threads - Core i5-1135G7

muzamil-py — Mon, 08 Jan 2024 14:20:38 GMT

Thank you @Gennady_F_Intel , your answer led towards solution.

I used -lmkl_intel_thread intead of -lmkl_sequential. So my compile command was:
g++ -o code code.cpp -I/opt/intel/oneapi/mkl/latest/include -L/opt/intel/oneapi/mkl/latest/lib -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lpthread -lm -ldl

I replaced -lmkl_sequential with -lmkl_intel_thread but I was still getting an error.

Error: collect2: error: ld returned 1 exit status (Screenshot Attached.)

To solve this:

I used icpx compiler instead of g++.
I added -liomp5 flag. (Reference to this change: Intel MKL Thread)

After appending these changes my code successfully compiled and executed with multiple threads.

My final build command now is:

icpx -fsycl -o test test.cpp -I/opt/intel/oneapi/mkl/latest/include -L/opt/intel/oneapi/mkl/latest/lib -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lm -ldl -liomp5 -lpthread