huytd/matrix-gpu

Matrix Multiplication on GPU

Matrix Multiplication on GPU

There are 2 programs: cpu and gpu. Both of them does the same thing is multiply two 1000 x 1000 matrices.

The algorithm to multiply is simple:

for (int row = 0; row < MATRIX_SIZE; row++) {
    for (int col = 0; col < MATRIX_SIZE; col++) {
      ulong sum = 0;
      for (int k = 0; k < MATRIX_SIZE; k++) {
        sum += A[row * MATRIX_SIZE + k] * B[k * MATRIX_SIZE + col];
      }
      C[row * MATRIX_SIZE + col] = sum;
    }
  }

The GPU program uses OpenCL.

How to run

On macOS, compile with Make:

make

Them run the CPU program with:

./cpu

Run the GPU program with:

./gpu

Compare the result

To compare the result of the two methods, do the following steps:

Step 1: Run CPU code and save the output to output_cpu file:

./cpu > output_cpu

Step 2: Run GPU code and save the output to output_gpu file:

./gpu > output_gpu

Step 3: Use diff to compare the two results to make sure outputs are the same:

diff output_cpu output_gpu

The only different is the running time. For example:

< Runtime: 4.849347ms
---
> Runtime: 0.001779ms

matrix-gpu's Issues

Adding Linux support

Now the project only build-able on macOS. Need to support Linux as well.