2 Example: Matrix-Matrix Multiplication

Start by considering the FORTRAN coding of the BLAS routine dgemm

For this routine, I have gone through the following steps:

• Write a driver routine that queries matrix sizes, creates the matrices, fills them with random numbers, and calls the BLAS routine. The files are given by
  • driver1.c
  • Makefile
  To time and compute MFLOPS, one needs the routine that calls the clock
  • clock.c
• Next, I took this routine, and created one that only performs the operation

  

  but still in FORTRAN. The result is the routine dgemm_notrans_notrans, given in

  • dgemm_notrans_notrans.f
• A translation to C, dgemm_notrans_notrans_c, is given in
  • dgemm_notrans_notrans_c.c

Notice that by typing

yes | rm *.o
make driver

one creates a version that does not use an optimizer, where typing

yes | rm *.o
make driver3

creates a version that uses optimization level -O 3.

Some performance numbers (in MFLOPS):

place HR here