Changes between Initial Version and Version 5 of Ticket #5793


Ignore:
Timestamp:
Jan 18, 2012 7:24:38 PM (4 years ago)
Author:
dterei
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #5793

    • Property Blocking changed from to 5794
    • Property Summary changed from make fibon not suck to make nofib not suck
  • Ticket #5793 – Description

    initial v5  
    1 fibon is the standard tool GHC developers use to benchmark changes to the compiler. Its a known fact though that fibon sucks. Its overall design is OK but it's had no love and care for many years and has bittrotted into near uselessness in a lot of situations.
     1Nofib is the standard tool GHC developers use to benchmark changes to the compiler. Its a known fact though that nofib sucks. Its overall design is OK but it's had no love and care for many years and has bittrotted into near uselessness in a lot of situations.
    22
    3 This task isn't to implement an awesome, really useful benchmark suite for Haskell. That would be great but is a huge undertaking. This task is just about making fibon useful again.
     3This task isn't to implement an awesome, really useful benchmark suite for Haskell. That would be great but is a huge undertaking. This task is just about making nofib useful again.
    44
    55The breakdown for this is something like:
    66
    7 1. Think and maybe fix fibon framework design. It has 'ways' which I think correspond to compilation method but more in the sense of 'dynamic' vs 'static', seems it may not suite being able to use ways for 'fasm' vs 'fllvm'. There is also the concept of 'modes' which corresponds to different benchmark input. So 'normal' and 'slow' for getting different run-times. At moment no easy way to select which benchmark groups to run, so may want to change that. I guess we should just decide, what knobs do we want to be able to easily tweak, and see how well the current design allows that.
     71. Think and maybe fix nofib framework design. It has 'ways' which I think correspond to compilation method but more in the sense of 'dynamic' vs 'static', seems it may not suite being able to use ways for 'fasm' vs 'fllvm'. There is also the concept of 'modes' which corresponds to different benchmark input. So 'normal' and 'slow' for getting different run-times. At moment no easy way to select which benchmark groups to run, so may want to change that. I guess we should just decide, what knobs do we want to be able to easily tweak, and see how well the current design allows that.
    88
    992. Fixup the runtimes for benchmarks to be significant. (A lot of them run in < 1 or < 0.1 seconds). I think trying to get all of them to run in around 10s is a good target.