[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 Jeffrey A. Law changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #21 from Jeffrey A. Law --- Per c#19.
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 Richard Biener changed: What|Removed |Added Priority|P3 |P2
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 --- Comment #20 from Jeffrey A. Law --- THe question I'm really trying to answer is whether or not the bug is limited to gcc-4.9.x. If so, then we can just close this as 4.9.x are dead branches. It appears that's the case, but I'm doing a verification step to be sure.
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 --- Comment #19 from wilco at gcc dot gnu.org --- (In reply to Jeffrey A. Law from comment #18) > I've been able to reproduce this under arm-qemu. > > Start by bootstrapping and installing gcc-4.9.4. Then use that gcc-4.9.4 to > build gcc-5.3.0 and you'll get the comparison failure. > > Based on various comments I strongly suspect this is a bug in gcc-4.9.4. Yes, the (In reply to Jeffrey A. Law from comment #18) > I've been able to reproduce this under arm-qemu. > > Start by bootstrapping and installing gcc-4.9.4. Then use that gcc-4.9.4 to > build gcc-5.3.0 and you'll get the comparison failure. > > Based on various comments I strongly suspect this is a bug in gcc-4.9.4. Yes we know the bug is still in 4.9.4, but that branch has been closed. It's fixed everywhere else. Simple workarounds for 4.9.4 are -mneon-for-64bits, -marm or -mfpu=vfpv4.
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 Jeffrey A. Law changed: What|Removed |Added CC||law at redhat dot com --- Comment #18 from Jeffrey A. Law --- I've been able to reproduce this under arm-qemu. Start by bootstrapping and installing gcc-4.9.4. Then use that gcc-4.9.4 to build gcc-5.3.0 and you'll get the comparison failure. Based on various comments I strongly suspect this is a bug in gcc-4.9.4.
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 Jeffrey A. Law changed: What|Removed |Added Status|WAITING |NEW
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 wilco at gcc dot gnu.org changed: What|Removed |Added CC||wilco at gcc dot gnu.org --- Comment #17 from wilco at gcc dot gnu.org --- I've backported PR78041 to GCC5 branch so it's worth trying again to see whether this is now fixed?
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 Richard Earnshaw changed: What|Removed |Added Status|UNCONFIRMED |WAITING Last reconfirmed||2017-01-20 Ever confirmed|0 |1 --- Comment #16 from Richard Earnshaw --- May be related to PR78041. Can you test with latest trunk, or latests gcc-6 branch?
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 --- Comment #15 from Ludovic Courtès --- Hi Jakub, Thanks for the suggestions. Unfortunately I cannot offer to rebuild everything with dumps right now; I'll see if I can do something later, no promise. Hopefully, with the details and faulty commit I gave earlier, people can more easily reproduce and pinpoint the problem.
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 Jakub Jelinek changed: What|Removed |Added CC||jakub at gcc dot gnu.org --- Comment #14 from Jakub Jelinek --- One can use --param min-nondebug-insn-uid=1 or something similar to make RTL dumps easier to compare; while the insns surrounded by debug insns still have different numbers for previous and/or next insn, if there are larger blocks of insns without debug insns in between them, there shouldn't be that many differences. First of all, can you reproduce the bug just with -fcompare-debug added to the list of options to compile say real.c (or whatever other file that shows the differences)? If yes, then do -fdump-tree-all -da and then try to find the first dump with differences. If not, then most likely some stage is miscompiled.
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 --- Comment #13 from Ludovic Courtès --- Ping!
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 --- Comment #12 from Ludovic Courtès --- (In reply to Ludovic Courtès from comment #10) > In fact, the revert only needs to be done on the initial > pseudo-cross-compiler that is used to build the native compiler (I'm > attaching a PDF of the dependency graph, where the initial > pseudo-cross-compiler as described at > http://linuxfromscratch.org/lfs/view/stable/chapter05/gcc-pass1.html is > marked as "gcc-cross-boot0", and the final native compiler is simply "gcc"). > > To summarize, if we build 4.9.4 with a cross-4.9.3, that's fine; if we build > 4.9.4 with a cross-4.9.4, we get the bootstrap comparison failure. The faulty commit is r231177 (commit f6ab85b7049a03962ea98924d00802da357a1ad3 in the Git mirror). If we take 4.9.4 and revert r231177 in the "gcc-cross-boot0" compiler, then the final gcc builds fine (no bootstrap comparison failure). Note that the revert needs not be applied to the final gcc, only to gcc-cross-boot0. (For reference, we applied the revert here: http://git.savannah.gnu.org/cgit/guix.git/commit/?id=e7e43727ce7b3426a31b2f50b035a5b0aba61d52 .) Any idea what's going on here?
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 --- Comment #11 from Ludovic Courtès --- Created attachment 39587 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=39587=edit Dependency graph of the native compiler
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 --- Comment #10 from Ludovic Courtès --- In fact, the revert only needs to be done on the initial pseudo-cross-compiler that is used to build the native compiler (I'm attaching a PDF of the dependency graph, where the initial pseudo-cross-compiler as described at http://linuxfromscratch.org/lfs/view/stable/chapter05/gcc-pass1.html is marked as "gcc-cross-boot0", and the final native compiler is simply "gcc"). To summarize, if we build 4.9.4 with a cross-4.9.3, that's fine; if we build 4.9.4 with a cross-4.9.4, we get the bootstrap comparison failure. That would suggest a code generation issue that somehow propagates from the initial cross-compiler up to stage2 of the compiler we're building.
[Bug target/71399] [5/6/7 Regression] 5.3.0 bootstrap comparison failure on arm-linux-gnueabihf
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71399 Andrew Pinski changed: What|Removed |Added Keywords||build, wrong-code Component|bootstrap |target Known to work||4.9.3 Target Milestone|--- |5.5 Summary|GCC 5.3.0 bootstrap |[5/6/7 Regression] 5.3.0 |comparison failure on |bootstrap comparison |arm-linux-gnueabihf |failure on ||arm-linux-gnueabihf Known to fail||4.9.3, 5.3.0
