Index

1. Performance comparision of the new implementation of widesets in gm2

• the old wideset implementation used a struct with n word sized fields
• whereas the new implementation uses an ARRAY OF BYTE
  • an INCL , EXCL , AND , OR NOT etc will result in a call to the library module M2WIDESET rather than generate tree s internally as in the previous implementation
• all measurements performed on a 3Ghz AMD Ryzen 9 7950X3D 16-Core Processor

2. Performance comparision of AND operations on set sizes -O2

3. Performance comparision of AND operations on set sizes -O3

4. Performance comparision of NOT operations on set sizes -O2

5. Performance comparision of NOT operations on set sizes -O3

6. Performance comparision of INCL operations on set sizes -O2

7. Performance comparision of INCL operations on set sizes -O3

8. Performance comparision for assignment on set sizes -O2

9. Performance comparision for assignment on set sizes -O3

10. Performance comparision of OR operations on set sizes -O2

11. Performance comparision of OR operations on set sizes -O3

12. Conclusions

• the new implementation does not use word size operations yet
  • all operations use byte operands
• nevertheless AND , OR , NOT are nearly always faster in the new implementation
• the new implementation of INCL is slightly slower

13. Future work

• the set libraries need to use multi word operands rather than byte operands
• consider inlining the INCL when -O2 or -O3 is specified
• pragma for inlining selected modules needs to be implemented

Index

This document was produced using groff-1.22.