CS 2734 Computer Organization II

Read pages 199-209 (Section 4.5.2 - 4.5.4) of Tanenbaum

• Implementation note on Laboratory 11. If you are on a Sparc 5, you can make the display look better by editing the Xdefaults1 file in your home directory to contain the line:

         XTango*geometry:                =576x700
  

• Development of a MULD M instruction implemented in MIC-2:

         sum = 0;
         for (i = 0; i < 15; i++) {
            if (high bit of a is 1)
                sum += b;
            sum += sum;
            a += a;
         }
         if (high bit of a is 1) 
            sum += b; 
  

• Comparison of the vertical and horizontal implementation of the MULD

• Methods of speeding up the instruction cycle:
  • Allow the number of subcycles in a machine cycle to vary.
  • Increase the flexibility of the microjumps (e.g. have a single bit in the microinstruction to indicate that NZVC should be OR'd with the low bits of the address field to provide a jump table).
  • Might allow opcode bits to be OR'd with address field to provide a jump table.
  • Pipelining - the real problem is with the branch.
    • Could stall.
    • Could squash
    • Could predict (static is done by the compiler, dynamic is done at run time).
    • Determine whether the branch will be taken early in the pipeline.
    • Do both paths in parallel.
  • Quick look at microprogramming of two real machines:
    • Intel 8088
    • Motorola 68000

Revision Date: 4/2/97 at 9 am