0

I'm looking at IBM's Developing PowerPC Embedded Application Binary (EABI) Compliant Programs, in particular at table 4 on page 7.

enter image description here

The benchmark results are 88kDhry/sec without using SDA, and only 77kDhry/sec with SDA. I would have expected using SDA not only reduce code size but also improve performance because access to variables only needs two instead of three instructions. Can somebody explain the numbers in the table?

What am I missing?

Igor Skochinsky
  • 24,629
  • 2
  • 72
  • 109
Jens
  • 69,818
  • 15
  • 125
  • 179
  • I would expect it to be faster too. Could be a simple thing like accidental swapping of the values. Since we don't have access to the test program binaries and environment, I can only suggest to try your own tests. – Igor Skochinsky Jan 31 '13 at 17:04
  • I find so few runs of a dhrystone (or any benchmark) to be almost useless. Need to compile with different levels of optimization. Add or rearrange code (I normally add or remove nops) in the startup code or early in the binary to adjust where the code hits the cache lines, etc. Use different compilers, the same source code on the same machine on the same day should have a good width on dhrystone performance results, any single number is no better than pulling the number out of thin air and reporting that. and dhrystones only matter if dhrystone is your primary application. – old_timer Jan 31 '13 at 18:06
  • Would the downvoter please care to state what is wrong with my question? – Jens Jan 31 '13 at 21:33
  • Compile it and diff the assembly code? – tc. Feb 20 '13 at 02:41
  • I don't have the IBM High C compiler. (and I don't believe in programming by experimentation :-) – Jens Feb 20 '13 at 07:51

0 Answers0