CALLOC FREE weirdness

gistfile1.c

#include<stdlib.h>

// number of floats that fit in L1 cache
#define L1_SIZE 1024
#define L1_MATRIX_ALLOT 512
// number of floats that fit in L2 cache
#define L2_SIZE 8192
// number of floats that fit in L2 cache
#define L3_SIZE 8192

void transpose( int n, int blocksize, float *dst, float *src ) {
    int i,j,k,m;
    /* TO DO: implement blocking (two more loops) */
    for ( k = 0; k < n; k+=blocksize ) {
      for (m = 0; m < n; m+=blocksize ) {
        if (k+blocksize-1 < n) {
          if (m+blocksize-1 < n) {
            for( i = 0; i < blocksize; i++ )
                for( j = 0; j < blocksize; j++ )
                    dst[(j+m)+(i+k)*n] = src[(i+k)+(j+m)*n];
          }
          else {
            for( i = 0; i < blocksize; i++ )
                for( j = m; j < n; j++ )
                    dst[j+(i+k)*n] = src[(i+k)+j*n];
          }
        }
        else {
          if (m+blocksize-1 < n) {
            for( i = k; i < n; i++ )
                for( j = 0; j < blocksize; j++ )
                    dst[(j+m)+i*n] = src[i+(j+m)*n];
          }
          else {
            for( i = k; i < n; i++ )
                for( j = m; j < n; j++ )
                    dst[j+i*n] = src[i+j*n];
          }
        }
      }
    }
}


/* This routine performs a sgemm operation
 *  C := C + A * B
 * where A, B, and C are lda-by-lda matrices stored in column-major format.
 * On exit, A and B maintain their input values. */    
void square_sgemm (int n, float* A, float* B, float* C)
{
  float *A_trans = (float*)calloc(n, sizeof(float));
  printf("%d\n", A_trans);
  transpose(n, 32, A_trans, A);
  printf("%d\n", A_trans);
  /* For each row i of A */
  for (int i = 0; i < n; ++i) {
    /* For each column j of B */
    for (int j = 0; j < n; ++j) 
    {
      /* Compute C(i,j) */
      float cij = C[i+j*n];
      for( int k = 0; k < n; k++ )
        cij += A[i+k*n] * B[k+j*n];
      C[i+j*n] = cij;
    }
  }
  free(A_trans);
}