Opened 10 years ago

Closed 10 years ago

Last modified 7 years ago

#1000 closed task (fixed)

Refactor HPC support

Reported by: simonmar Owned by: AndyGill
Priority: normal Milestone: 6.8.1
Component: Compiler Version: 6.7
Keywords: Cc: andy@…
Operating System: Unknown/Multiple Architecture: Unknown/Multiple
Type of failure: None/Unknown Test Case:
Blocked By: Blocking:
Related Tickets: Differential Rev(s):
Wiki Page:

Description

So we don't forget: as per discussion on cvs-ghc (http://www.haskell.org/pipermail/cvs-ghc/2006-November/032434.html), look into refactoring the HPC support to use a PrimOp instead of a Note in Core.

Change History (7)

comment:1 Changed 10 years ago by AndyGill

Owner: set to AndyGill
Status: newassigned

The plan is to attempt to change to use of Note to hold ticks into using primops to represent ticks internally.

Simon suggested looking at the FCallId, as an example of how to have a primop like thing, but also have the primop "saturated with constants".

The will be done in late Nov or early Dec, 2006.

comment:2 Changed 10 years ago by AndyGill

Version: 6.66.7

comment:3 Changed 10 years ago by AndyGill

Resolution: fixed
Status: assignedclosed

Completed this change. We no longer use Note, rather use

  case tick<mod,n> of
    _ -> ...

The generated code is the same boolean-wise as the old tick method.

  • In some cases we generate better CAFs.
  • The execution runtime is faster with -O2. (~50% faster on raytracer)
  • The compile time is slower with -O2. (~50% slower, perhaps because of new optimizations kicking in)

comment:4 Changed 9 years ago by igloo

Milestone: 6.8 branch6.8.1

comment:5 Changed 8 years ago by simonmar

Architecture: UnknownUnknown/Multiple

comment:6 Changed 8 years ago by simonmar

Operating System: UnknownUnknown/Multiple

comment:7 Changed 7 years ago by simonmar

difficulty: Moderate (1 day)Moderate (less than a day)
Note: See TracTickets for help on using tickets.