Running the code

• To compile the code run make all

• Now run ./image followed by the argument given below specific to the function

• To call Convolution with padding with matrix multiplication:

  convolution_withpadding_matrixmult [PADSIZE] MODE [matrix1.txt] [matrix1_numrows] [matrix2.txt] [matrix2_numrows]

• To call Convolution without padding with matrix multiplication:

  convolution_withoutpadding_matrixmult MODE [matrix1.txt] [matrix1_numrows] [matrix2.txt] [matrix2_numrows]

where MODE is 1 for normal multiplication, 2 for pthread, 3 for OpenBLAS, 4 for MKL.

• To call Convolution with padding with Convolution:

  convolution_withpadding_conv [PADSIZE] [matrix1.txt] [matrix1_numrows] [matrix2.txt] [matrix2_numrows]

• To call Convolution without padding with Convolution:

  convolution_withoutpadding_conv [matrix1.txt] [matrix1_numrows] [matrix2.txt] [matrix2_numrows]

• To call relu activation:

  activation_relu [matrix1.txt] [matrix1_numrows]

• To call tanh activation:

  activation_tanh [matrix1.txt] [matrix1_numrows]

• To do subsampling with maxpooling:

  subsampling_maxpooling [matrix1.txt] [matrix1_numrows]

• To do subsampling with average pooling:

  subsampling_avgpooling [matrix1.txt] [matrix1_numrows]

• To convert random float vector to probability vector using softmax:

  probability_softmax [matrix1.txt] [matrix1_numrows]

• To convert random float vector to probability vector using sigmoid

  probability_sigmoid [matrix1.txt] [matrix1_numrows]

NOTE:

• For sub-sampling (both odd and even) we have chosen the sampling matrix to be 2x2 and the stride to be 2 always.
• For Max pooling we padded the odd size matrices and for average pooling we considered the average of only two elements when we reached the edge of the matrix.
• For the convolution we inverted the kernel.

NOTE FOR SUBTASK 2:

• We have tested the time of multiplication of the toeplitz matrix with pthread, openBLAS, mkl libraries against matrix of sizes 1000,900,750 & 500 and a kernel size of 3.
• If the range of the y-axis in the png file is too small, the ps file has a elongated version.
• The times in the graphs are in microseconds measured by the chrono high resolution clock.
• This Subtask prints the time taken by the matrix multiplication for the matrix_multiplication command and prints the resultant matrix or vector.