I've filed http://www.llvm.org/bugs/show_bug.cgi?id=1912 for the optimizer bug. Evan On Jan 10, 2008, at 11:28 AM, Evan Cheng wrote:> Hi, > > We are getting closer. > > 1. In Makefile, all the references to CFLAGS should be CPPFLAGS > instead. > 2. filterdiff.sh uses sed -re. This causes a problem on Mac OS X where > -E means using extended regular expression, not -r. > > sed: illegal option -- r > usage: sed script [-Ealn] [-i extension] [file ...] > sed [-Ealn] [-i extension] [-e script] ... [-f > script_file] ... [file ...] > > Can this be changed? > > 3. This triggers a optimizer bug: > /Users/echeng/LLVM/llvm/Release/bin/opt -std-compile-opts -time-passes > -info-output-file=/Volumes/Muggles/LLVM/llvm/projects/llvm-test/ > MultiSource/Applications/ClamAV/Out\ > put/clamscan.linked.bc.info Output/clamscan.linked.rbc -o Output/ > clamscan.linked.bc -f > Assertion failed: (getLoopLatch() && "Loop latch is missing"), > function verifyLoop, file /Volumes/Muggles/LLVM/llvm/include/llvm/ > Analysis/LoopInfo.h, line 517. > > I'll file a bug on this. > > Evan > > > On Jan 10, 2008, at 2:41 AM, Török Edwin wrote: > >> Török Edwin wrote: >>> Yes, I have uploaded the .tar.gz here: >>> http://edwintorok.googlepages.com/ClamAV-srcflat.tar.gz >>> [inputs dir contains some symlink, place ClamAV dir in >>> llvm/projects/llvm-test/MultiSource/Applications to make >>> links point to right place] >>> >>> >> >> Hi, >> >> Because llvm bug #1730 got fixed, this testcase can run under with >> the >> JIT on x86-64 :). >> I have attached the updated filterdiff.sh script (JIT has extra file >> descriptor open), and now all tests pass. >> >> I've also uploaded the new tarball here: >> http://edwintorok.googlepages.com/ClamAV-srcflat2.tar.gz >> >> TEST-PASS: compile /MultiSource/Applications/ClamAV/clamscan >> TEST-RESULT-compile: Total Execution Time: 6.0010 seconds (6.7158 >> wall >> clock) >> >> TEST-RESULT-compile: 1728712 Output/clamscan.llvm.bc >> >> >> TEST-RESULT-nat-time: program 12.880000 >> >> TEST-PASS: llc /MultiSource/Applications/ClamAV/clamscan >> TEST-RESULT-llc: Total Execution Time: 7.3758 seconds (7.7852 wall >> clock) >> TEST-RESULT-llc-time: program 13.940000 >> >> >> TEST-PASS: jit /MultiSource/Applications/ClamAV/clamscan >> TEST-RESULT-jit-time: program 18.320000 >> >> TEST-RESULT-jit-comptime: >> >> TEST-PASS: cbe /MultiSource/Applications/ClamAV/clamscan >> TEST-RESULT-cbe-time: program 12.750000 >> >> >> Best regards, >> --Edwin >> <filterdiff.sh>_______________________________________________ >> LLVM Developers mailing list >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
Devang has fixed the bug. I've added ClamAV to the testsuite. Thanks! Evan On Jan 13, 2008, at 7:33 PM, Evan Cheng wrote:> I've filed > http://www.llvm.org/bugs/show_bug.cgi?id=1912 > for the optimizer bug. > > Evan > > On Jan 10, 2008, at 11:28 AM, Evan Cheng wrote: > >> Hi, >> >> We are getting closer. >> >> 1. In Makefile, all the references to CFLAGS should be CPPFLAGS >> instead. >> 2. filterdiff.sh uses sed -re. This causes a problem on Mac OS X >> where >> -E means using extended regular expression, not -r. >> >> sed: illegal option -- r >> usage: sed script [-Ealn] [-i extension] [file ...] >> sed [-Ealn] [-i extension] [-e script] ... [-f >> script_file] ... [file ...] >> >> Can this be changed? >> >> 3. This triggers a optimizer bug: >> /Users/echeng/LLVM/llvm/Release/bin/opt -std-compile-opts -time- >> passes >> -info-output-file=/Volumes/Muggles/LLVM/llvm/projects/llvm-test/ >> MultiSource/Applications/ClamAV/Out\ >> put/clamscan.linked.bc.info Output/clamscan.linked.rbc -o Output/ >> clamscan.linked.bc -f >> Assertion failed: (getLoopLatch() && "Loop latch is missing"), >> function verifyLoop, file /Volumes/Muggles/LLVM/llvm/include/llvm/ >> Analysis/LoopInfo.h, line 517. >> >> I'll file a bug on this. >> >> Evan >> >> >> On Jan 10, 2008, at 2:41 AM, Török Edwin wrote: >> >>> Török Edwin wrote: >>>> Yes, I have uploaded the .tar.gz here: >>>> http://edwintorok.googlepages.com/ClamAV-srcflat.tar.gz >>>> [inputs dir contains some symlink, place ClamAV dir in >>>> llvm/projects/llvm-test/MultiSource/Applications to make >>>> links point to right place] >>>> >>>> >>> >>> Hi, >>> >>> Because llvm bug #1730 got fixed, this testcase can run under with >>> the >>> JIT on x86-64 :). >>> I have attached the updated filterdiff.sh script (JIT has extra file >>> descriptor open), and now all tests pass. >>> >>> I've also uploaded the new tarball here: >>> http://edwintorok.googlepages.com/ClamAV-srcflat2.tar.gz >>> >>> TEST-PASS: compile /MultiSource/Applications/ClamAV/clamscan >>> TEST-RESULT-compile: Total Execution Time: 6.0010 seconds (6.7158 >>> wall >>> clock) >>> >>> TEST-RESULT-compile: 1728712 Output/clamscan.llvm.bc >>> >>> >>> TEST-RESULT-nat-time: program 12.880000 >>> >>> TEST-PASS: llc /MultiSource/Applications/ClamAV/clamscan >>> TEST-RESULT-llc: Total Execution Time: 7.3758 seconds (7.7852 wall >>> clock) >>> TEST-RESULT-llc-time: program 13.940000 >>> >>> >>> TEST-PASS: jit /MultiSource/Applications/ClamAV/clamscan >>> TEST-RESULT-jit-time: program 18.320000 >>> >>> TEST-RESULT-jit-comptime: >>> >>> TEST-PASS: cbe /MultiSource/Applications/ClamAV/clamscan >>> TEST-RESULT-cbe-time: program 12.750000 >>> >>> >>> Best regards, >>> --Edwin >>> <filterdiff.sh>_______________________________________________ >>> LLVM Developers mailing list >>> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu >>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >> >> >> _______________________________________________ >> LLVM Developers mailing list >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
Evan Cheng wrote:> Devang has fixed the bug. I've added ClamAV to the testsuite. > > Thanks! >Thanks a lot! I noticed that the runtime is ~0.27 seconds, if you want it to be higher (to better highlight differences between native, llc, jit, cbe), you can add some more files to inputs/. For example symlinks to already existing large binary files in the llvm-testsuite, such as: large.pcm, mei16v2.m2v, tune. Best regards, --Edwin
Possibly Parallel Threads
- [LLVMdev] Adding ClamAV to the llvm testsuite (long)
- [LLVMdev] Adding ClamAV to the llvm testsuite (long)
- [LLVMdev] Adding ClamAV to the llvm testsuite (long)
- [LLVMdev] Adding ClamAV to the llvm testsuite (long)
- [LLVMdev] Adding ClamAV to the llvm testsuite (long)