[LLVMbugs] [Bug 17245] New: MachineSink Performance Issue

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=17245

            Bug ID: 17245
           Summary: MachineSink Performance Issue
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: willdtz at gmail.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 11197
  --> http://llvm.org/bugs/attachment.cgi?id=11197&action=edit
Proposed patch to address performance issue

I'm encountering bad performance behavior of the MachineSink pass,
as shown in the attached 'before.log' (MachineSink taking ~25 minutes to
execute!).

Investigation led me to create the attached patch which both avoids the n^2
behavior presently exhibited by the pass but also tries to more faithfully
implement the heuristic described in the comments of the code itself.

With this patch applied, MachineSink's execution time reduces from 1494.1
seconds down to 1.8 seconds.  I've attached the patch as well as 'after.log'
demonstrating the much improved timings.

Unfortunately the original IR is both large and something I'd rather not post,
although I've managed to reduce it into the attached 'reduced.bc' example that
has before/after timings of 10.5s/0.2s (I used bugpoint filtering on the
runtime before being greater than 10s and after being less than 1 second).

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20130914/8d72177d/attachment.html>