Benchmarks

Benchmarks 1A and 1B both perform the same function. We chose to implement it first with our mtm (memory-to-memory copy) operation, and second (1B) with lw and sw. This will allow us to compare them and speculate as to how effective mtm is. Since the RTL for mem required no changes to the datapath and only minimal changes to the control, it is unlikely that removing it would change our maximum clock speed. Thus, it is fair to directly compare the running times of Benchmarks 1A and 1B.

Performance Data

We calculated the following values for each benchmark:

• LOC = lines of code
  • Visible instructions in ASM file
  • Not counting pseudo-instructions, labels, or comments
  • Not counting the "done" placeholder instruction to be executed after the benchmark
• I = total instructions executed
• CYC = total clock cycles
• CPI = average cycles per instruction
• RT = total running time

These calculations are based on a clock speed of 9.853 MHz, the maximum clock speed reported by the Xilinx Foundation Series software.
The calculations were performed in Excel 2000

The mtm instruction saved 1024 cycles (0.104 ms) when Benchmark 1, as compared to using lw and sw.