[LLVMbugs] [Bug 22630] New: Long execution time for large IR module

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=22630

            Bug ID: 22630
           Summary: Long execution time for large IR module
           Product: libraries
           Version: 3.5
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Global Analyses
          Assignee: unassignedbugs at nondot.org
          Reporter: djones at xtreme-eda.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 13903
  --> http://llvm.org/bugs/attachment.cgi?id=13903&action=edit
Timing report

I have a rather large (71MB) IR assembly file (includes debug metadata) that
takes 100 seconds to optimize and generate x86-64 code for. (Host is AMD K10
running at 2.7-3.3GHz.)

If I run:

/tools/llvm/3.5.1dbg/bin/llc -O2  LMtb_ethernet.ll -o opt.o -time-passes

the pass timing report (attached) accounts for about 40 seconds of run time,
even though the command takes about 100 seconds.

In comparison, clang takes only 4-6 seconds to compile my largest C++ file,
which results in 61MB of LLVM IR assembly (again, including debug).

I'm not sure what's actually taking up the time. If I run opt instead of llc,
opt will finish in about 40 seconds.  Is this a case of LLVM backend behaving
reasonably on very large output, or is there something pathological in either
my code or LLVM that one of us can fix?

It should be noted that the file I am trying to compile is by no means
considered "large" in my industry.

-- 
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/20150219/be1b79c3/attachment.html>