[LLVMbugs] [Bug 7328] New: miscompare of SPEC CPU2006's 453.povray

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=7328

           Summary: miscompare of SPEC CPU2006's 453.povray
           Product: dragonegg
           Version: 2.7
          Platform: PC
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: New Bugs
        AssignedTo: baldrick at free.fr
        ReportedBy: kenneth.hoste at ugent.be
                CC: llvmbugs at cs.uiuc.edu


Compiling the 453.povray SPEC CPU2006 benchmark with dragonegg-2.7 -O2 (with
gcc-4.5.0) makes the SPEC framework complain about a miscompile:

  Running (#1) 453.povray test base x86-64_llvm-2.7_dragonegg_O2 default
/local/SPEC_CPU2006_v1.1_src/bin/specinvoke -d
/local/SPEC_CPU2006_v1.1_src/benchspec/CPU2006/453.povray/run/run_base_test_x86-64_llvm-2.7_dragonegg_O2.0000
-e speccmds.err -o speccmds.stdout -f speccmds.cmd -C
/local/SPEC_CPU2006_v1.1_src/bin/specinvoke -E -d
/local/SPEC_CPU2006_v1.1_src/benchspec/CPU2006/453.povray/run/run_base_test_x86-64_llvm-2.7_dragonegg_O2.0000
-c 1 -e compare.err -o compare.stdout -f compare.cmd

*** Miscompare of SPEC-benchmark.log; for details see
   
/local/SPEC_CPU2006_v1.1_src/benchspec/CPU2006/453.povray/run/run_base_test_x86-64_llvm-2.7_dragonegg_O2.0000/SPEC-benchmark.log.mis

*** Miscompare of SPEC-benchmark.tga; for details see
   
/local/SPEC_CPU2006_v1.1_src/benchspec/CPU2006/453.povray/run/run_base_test_x86-64_llvm-2.7_dragonegg_O2.0000/SPEC-benchmark.tga.mis


I've looked into the problem a bit further, and the difference in output files
stems from a failing internal check of the benchmark. I've found this in
SPEC-benchmark.log (which shouldn't be there):

"Memory leakage detected, see file 'Memory.Log' for list"

In Memory.Log, I find these:

File:     bbox.cpp  Line:1431  Size:40
File:     bbox.cpp  Line:1441  Size:24
(sizes vary for Line:1441)

This problem also occurs when povray is compiled with llvm-gcc-4.2 -O2.
The problem doesn't occur when -O0 is used, or when the benchmark is built with
gcc-4.5.0 -O2 without dragonegg, which makes me conclude the problem lies
within LLVM.

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.