Fair enough... The following tests under MultiSource fail: Benchmarks/Olden/power Benchmarks/OptimizerEval Benchmarks/Ptrdist/ks Benchmarks/MallocBench/perl Applications/sgefa However, they also fail in the exact same way without my change. OptimizerEval appears to be non-deterministic; it produces different output each time it runs. Everything else passes. I also compared a few *.s files generated by the tests, with and without my change, to see if everything is as I expected. I tried running SingleSource also, and ran into a few problems. Some of the tests failed because they cannot include alloca.h. This may be a linux dependency. Anyway, again I saw the same results with or without my change. I cannot run qmtest at all. I don't seem to have it or any way of building it. There's no rush to commit this. If you don't feel comfortable doing so, it can wait until Chris gets back. And as it's my first submission, you shouldn't feel comfortable :) Jeff On Sat, 28 Aug 2004 22:20:57 -0700 Reid Spencer <reid at x10sys.com> wrote:> Jeff, > > Chris isn't likely to respond to this for a while as he's on vacation. > I'll take a look at it and will commit it if it looks good. Since code > gen isn't my specialty, could you increase my comfort level a little by > giving me some examples of the test results you got when testing your > patches? Ideally, I'd like to see some of the test/Programs/MultiSource > programs working and generating correct code with this path. > > Thanks, > > Reid.
On Sun, 2004-08-29 at 16:01, Jeff Cohen wrote:> Fair enough... The following tests under MultiSource fail: > > Benchmarks/Olden/power > Benchmarks/OptimizerEval > Benchmarks/Ptrdist/ks > Benchmarks/MallocBench/perl > Applications/sgefa > > However, they also fail in the exact same way without my change. > OptimizerEval appears to be non-deterministic; it produces different > output each time it runs. Everything else passes. I also compared a > few *.s files generated by the tests, with and without my change, to see > if everything is as I expected.I know for certain that sgefa, perl, and power are XFAIL tests. Not sure about the other two.> > I tried running SingleSource also, and ran into a few problems. Some of > the tests failed because they cannot include alloca.h. This may be a > linux dependency. Anyway, again I saw the same results with or without > my change.Which platform are you compiling on?> I cannot run qmtest at all. I don't seem to have it or any way of > building it.qmtest is an external package. You need version 2.03. I suggest you read: http://llvm.cs.uiuc.edu/docs/TestingGuide.html In the requirements section there's a link to the QMTest software.> > There's no rush to commit this. If you don't feel comfortable doing so, > it can wait until Chris gets back. And as it's my first submission, you > shouldn't feel comfortable :)On the other hand, two weeks is a long time to wait for your first submission :) I'm going to review the code, if it looks good I'll test it, if that looks okay, I'll commit it and we can see what it does to the nightly tester. Thanks for your willingness to help, Jeff. Reid.> > Jeff > > On Sat, 28 Aug 2004 22:20:57 -0700 > Reid Spencer <reid at x10sys.com> wrote: > > > Jeff, > > > > Chris isn't likely to respond to this for a while as he's on vacation. > > I'll take a look at it and will commit it if it looks good. Since code > > gen isn't my specialty, could you increase my comfort level a little by > > giving me some examples of the test results you got when testing your > > patches? Ideally, I'd like to see some of the test/Programs/MultiSource > > programs working and generating correct code with this path. > > > > Thanks, > > > > Reid. > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev-------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: This is a digitally signed message part URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20040829/e7e85deb/attachment.sig>
Jeff, I tried the Benchmarks/Olden/power, Benchmarks/OptimizerEval, and Benchmarks/Ptrdist/ks tests. They all worked with your patches. I suggest you update your tree :) The changes also survived all the Feature and Regression tests on Linux. So, your changes are committed. Thanks for the patches! Reid. On Sun, 2004-08-29 at 16:01, Jeff Cohen wrote:> Fair enough... The following tests under MultiSource fail: > > Benchmarks/Olden/power > Benchmarks/OptimizerEval > Benchmarks/Ptrdist/ks > Benchmarks/MallocBench/perl > Applications/sgefa > > However, they also fail in the exact same way without my change. > OptimizerEval appears to be non-deterministic; it produces different > output each time it runs. Everything else passes. I also compared a > few *.s files generated by the tests, with and without my change, to see > if everything is as I expected. > > I tried running SingleSource also, and ran into a few problems. Some of > the tests failed because they cannot include alloca.h. This may be a > linux dependency. Anyway, again I saw the same results with or without > my change. > > I cannot run qmtest at all. I don't seem to have it or any way of > building it. > > There's no rush to commit this. If you don't feel comfortable doing so, > it can wait until Chris gets back. And as it's my first submission, you > shouldn't feel comfortable :) > > Jeff > > On Sat, 28 Aug 2004 22:20:57 -0700 > Reid Spencer <reid at x10sys.com> wrote: > > > Jeff, > > > > Chris isn't likely to respond to this for a while as he's on vacation. > > I'll take a look at it and will commit it if it looks good. Since code > > gen isn't my specialty, could you increase my comfort level a little by > > giving me some examples of the test results you got when testing your > > patches? Ideally, I'd like to see some of the test/Programs/MultiSource > > programs working and generating correct code with this path. > > > > Thanks, > > > > Reid. > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev-------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: This is a digitally signed message part URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20040829/7711002d/attachment.sig>
Jeff, Thought you might like to know. Your changes caused a decrease in the native code size of executables in the nightly test. Have a look at: http://llvm.x10sys.com/testresults and, in particular: http://llvm.x10sys.com/testresults/running_Olden_machcode_large.png TYMLTK Reid. On Sun, 2004-08-29 at 16:01, Jeff Cohen wrote:> Fair enough... The following tests under MultiSource fail: > > Benchmarks/Olden/power > Benchmarks/OptimizerEval > Benchmarks/Ptrdist/ks > Benchmarks/MallocBench/perl > Applications/sgefa > > However, they also fail in the exact same way without my change. > OptimizerEval appears to be non-deterministic; it produces different > output each time it runs. Everything else passes. I also compared a > few *.s files generated by the tests, with and without my change, to see > if everything is as I expected. > > I tried running SingleSource also, and ran into a few problems. Some of > the tests failed because they cannot include alloca.h. This may be a > linux dependency. Anyway, again I saw the same results with or without > my change. > > I cannot run qmtest at all. I don't seem to have it or any way of > building it. > > There's no rush to commit this. If you don't feel comfortable doing so, > it can wait until Chris gets back. And as it's my first submission, you > shouldn't feel comfortable :) > > Jeff > > On Sat, 28 Aug 2004 22:20:57 -0700 > Reid Spencer <reid at x10sys.com> wrote: > > > Jeff, > > > > Chris isn't likely to respond to this for a while as he's on vacation. > > I'll take a look at it and will commit it if it looks good. Since code > > gen isn't my specialty, could you increase my comfort level a little by > > giving me some examples of the test results you got when testing your > > patches? Ideally, I'd like to see some of the test/Programs/MultiSource > > programs working and generating correct code with this path. > > > > Thanks, > > > > Reid. > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev-------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: This is a digitally signed message part URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20040830/ab95e0cc/attachment.sig>
On Sun, 29 Aug 2004 16:19:21 -0700 Reid Spencer <reid at x10sys.com> wrote:> On Sun, 2004-08-29 at 16:01, Jeff Cohen wrote: > > I tried running SingleSource also, and ran into a few problems. Some of > > the tests failed because they cannot include alloca.h. This may be a > > linux dependency. Anyway, again I saw the same results with or without > > my change. > > Which platform are you compiling on?FreeBSD 5.2.1> > I cannot run qmtest at all. I don't seem to have it or any way of > > building it. > > qmtest is an external package. You need version 2.03. I suggest you > read: > > http://llvm.cs.uiuc.edu/docs/TestingGuide.html > > In the requirements section there's a link to the QMTest software.I somehow missed that document...
On Sun, 29 Aug 2004 17:14:18 -0700 Reid Spencer <reid at x10sys.com> wrote:> Jeff, > > I tried the Benchmarks/Olden/power, Benchmarks/OptimizerEval, and > Benchmarks/Ptrdist/ks tests. They all worked with your patches. I > suggest you update your tree :)I was using the 1.3 release, though I did pull the current code from CVS to verify no other changes were made to those files. Unfortunately this machine is too underpowered to frequently rebuild LLVM. It was collecting dust in the closet until I decided to put FreeBSD on it. For the record it's a 64MB 300Mhz Pentium II and it takes 30-35 hours to build LLVM from scratch. It's possible some of my test problems are due to LLVM generating X86 code that's not Pentium II-friendly. If LLVM was Windows-friendly I'd build it on my much more powerful Windows system, but relying on cygwin means combining the worst of both Unix and Windows. When FreeBSD 5 goes production in a few months I plan to put it on a more powerful system.> The changes also survived all the Feature and Regression tests on Linux. > So, your changes are committed. > > Thanks for the patches!You're welcome!> Reid.
Maybe Matching Threads
- [LLVMdev] Optimization opportunity
- [LLVMdev] Optimization opportunity
- [LLVMdev] Optimization opportunity
- [LLVMdev] 2.2 Prerelease available for testing
- Compare test-suite benchmarks performance complied without TBAA, with default TBAA and with new TBAA struct path