On Tue, Dec 23, 2014 at 09:27:32PM +0000, Daniel Sanders wrote:> clang+llvm-3.5.1-rc2-mips-linux-gnu.tar.xz: > All ok for default options and -mips32. > > clang+llvm-3.5.1-rc2-mipsel-linux-gnu.tar.xz: > All ok for default options. > It will take a couple days to get results for the -mips32 test-suite (there's no FPU or local disk on this machine so some tests take a very long time). > > clang+llvm-3.5.1-rc2-x86_64-linux-gnu-ubuntu-14.04.tar.xz (cross-compiling to Mips): > All ok for all ISA's more recent than mips32, all ABI's except big-endian N32*, and both endians. > > * I seem to have mis-stated the results for rc1 for the big-endian N32 ABI, sorry. It fails 12 tests (previously I stated 'all ok' but looking at those log files again, this was incorrect) and the failing tests seem to involve pointers passed in varargs. This is a regression from 3.5.0 since it used to pass these tests, however this appears to be a small regression that's occurred as result of a major improvement to the big-endian N32 ABI (in particular, cross-linking with GCC now works in general). I therefore think we could leave this as a known issue and make sure it's fixed in 3.6.0 and 3.5.2. > Tom: Do you agree?Yes, this sounds OK to me. -Tom> ________________________________________ > From: Daniel Sanders > Sent: 21 December 2014 12:30 > To: Hans Wennborg; Tom Stellard > Cc: Ben Pope; Nikola Smiljanic; llvmdev > Subject: RE: [LLVMdev] 3.5.1 Testing Phase II Begins > > I've uploaded clang+llvm-3.5.1-rc2-mips-linux-gnu.tar.xz and clang+llvm-3.5.1-rc2-mipsel-linux-gnu.tar.xz. The latter fails prune.m in 'make check-all' as it did last time but this is because the 'touch -a' command isn't actually changing the access timestamp. > > I've started the test-suite for both of them. > ________________________________________ > From: llvmdev-bounces at cs.uiuc.edu [llvmdev-bounces at cs.uiuc.edu] on behalf of Hans Wennborg [hans at chromium.org] > Sent: 19 December 2014 23:50 > To: Tom Stellard > Cc: Ben Pope; Nikola Smiljanic; llvmdev > Subject: Re: [LLVMdev] 3.5.1 Testing Phase II Begins > > On Fri, Dec 19, 2014 at 2:35 PM, Tom Stellard <tom at stellard.net> wrote: > > Hi, > > > > I've just tagged 3.5.1-rc2, so testing can begin. Remember to use the 3.5 > > branch of the test-suite. If you have any questions, let me know. > > I've uploaded the Windows installer to the sftp with sha1: > b56681a457954916ac9c8f0d36426cde0a21ee1d LLVM-3.5.1-rc2-win32.exe > > Build worked fine this time. The only failing lit test is > Modules/compiler_builtins.m, which I guess is PR20995. That's probably > not cause for concern, but maybe we should remove the test from the > branch for those who build from source? CC'ing Yaron who has looked at > the PR some. > > - Hans > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> From: Tom Stellard [tom at stellard.net] > Sent: 29 December 2014 15:27 > To: Daniel Sanders > Cc: Hans Wennborg; Ben Pope; Nikola Smiljanic; llvmdev > Subject: Re: [LLVMdev] 3.5.1 Testing Phase II Begins > > On Tue, Dec 23, 2014 at 09:27:32PM +0000, Daniel Sanders wrote: > > clang+llvm-3.5.1-rc2-mips-linux-gnu.tar.xz: > > All ok for default options and -mips32. > > > > clang+llvm-3.5.1-rc2-mipsel-linux-gnu.tar.xz: > > All ok for default options. > > It will take a couple days to get results for the -mips32 test-suite (there's no FPU or local disk on this machine so some tests take a very long time). > > > > clang+llvm-3.5.1-rc2-x86_64-linux-gnu-ubuntu-14.04.tar.xz (cross-compiling to Mips): > > All ok for all ISA's more recent than mips32, all ABI's except big-endian N32*, and both endians. > > > > * I seem to have mis-stated the results for rc1 for the big-endian N32 ABI, sorry. It fails 12 tests (previously I stated 'all ok' but looking at those log files again, this was incorrect) and the failing tests seem to involve pointers passed in varargs. This is a regression from 3.5.0 since it used to pass these tests, however this appears to be a small regression that's occurred as result of a major improvement to the big-endian N32 ABI (in particular, cross-linking with GCC now works in general). I therefore think we could leave this as a known issue and make sure it's fixed in 3.6.0 and 3.5.2. > > Tom: Do you agree? > > Yes, this sounds OK to me. > > -TomJust for the record: http://reviews.llvm.org/D6812 fixes most of these big-endian N32 failures but we should stick to the current plan because I haven't retested the other ABI's yet. All but one of the remaining failures is due a qemu bug. The linux user mode emulation of struct target_stat doesn't quite match any supported by glibc and causes certain stat() calls to return -1. This is fixed in my working copy of qemu and I'll upstream it soon. The last failure is that lemon's output is wrong. I don't know the cause yet.> > clang+llvm-3.5.1-rc2-mipsel-linux-gnu.tar.xz: > > All ok for default options. > > It will take a couple days to get results for the -mips32 test-suite (there's no FPU or local disk on this machine so some tests take a very long time).Quick update to this one: All ok for default and -mips32.
> Just for the record: http://reviews.llvm.org/D6812 fixes most of these big- > endian N32 failures but we should stick to the current plan because I haven't > retested the other ABI's yet. > All but one of the remaining failures is due a qemu bug. The linux user mode > emulation of struct target_stat doesn't quite match any supported by glibc and > causes certain stat() calls to return -1. > This is fixed in my working copy of qemu and I'll upstream it soon. The last > failure is that lemon's output is wrong. I don't know the cause yet.The lemon failure turned out to be caused by treating transparent unions as a normal union type and not as the type of the first member. This caused wait() to return 0 without writing an exit code to the given pointer. I'm about to commit a fix to trunk for this.