[Weekly] 04-08 MAY 2015
* TCWG-146 - Detect smin/umin idiom (3/10) - Investigation * TCWG-110 - Investigate generating out-of-line branch tables in ARM GCC (1/10) - Resolved with ARM as questionable benefit; inline constant pools are cheap to access. * Illness (5/10) * Misc (1/10) - Conference calls == Next week == - Resolve TCWG-146 ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain
Re: [ACTIVITY] 04 May -- 08 May
On 11/05/15 15:54, Pinski, Andrew wrote: > It would be best to file a bug in glibc then and fix it there. I agree. Let me see if it is present in the trunk as well and file a bug-report. Thanks, Kugan > > From: Kugan > Sent: Sunday, May 10, 2015 10:46 PM > To: Pinski, Andrew; Linaro Toolchain > Subject: Re: [ACTIVITY] 04 May -- 08 May > > On 11/05/15 15:24, Pinski, Andrew wrote: >>> - Trunk is broken and narrowed down the failure to >>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66066 >> >> No the code is broken in SPEC ;). SPEC INT has been known to be broken C >> and they are not willing to do anything about it after the fact. This has >> been true for all versions of SPEC and one of the reason why we use >> -fno-strict-aliasing in some cases. > > Hi Andrew, > > Yes, I understand that this is non complaint code. Unfortunately it is > not the SPEC code but the code which is part of the gcc/glibc that is > showing this. I ran into this when I built the cross toolchain for > aarch64. I am sure that we are going to have this in tones of places. > > Thanks, > Kugan > ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain
Re: [ACTIVITY] 04 May -- 08 May
It would be best to file a bug in glibc then and fix it there. From: Kugan Sent: Sunday, May 10, 2015 10:46 PM To: Pinski, Andrew; Linaro Toolchain Subject: Re: [ACTIVITY] 04 May -- 08 May On 11/05/15 15:24, Pinski, Andrew wrote: >> - Trunk is broken and narrowed down the failure to >> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66066 > > No the code is broken in SPEC ;). SPEC INT has been known to be broken C and > they are not willing to do anything about it after the fact. This has been > true for all versions of SPEC and one of the reason why we use > -fno-strict-aliasing in some cases. Hi Andrew, Yes, I understand that this is non complaint code. Unfortunately it is not the SPEC code but the code which is part of the gcc/glibc that is showing this. I ran into this when I built the cross toolchain for aarch64. I am sure that we are going to have this in tones of places. Thanks, Kugan ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain
Re: [ACTIVITY] 04 May -- 08 May
On 11/05/15 15:24, Pinski, Andrew wrote: >> - Trunk is broken and narrowed down the failure to >> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66066 > > No the code is broken in SPEC ;). SPEC INT has been known to be broken C and > they are not willing to do anything about it after the fact. This has been > true for all versions of SPEC and one of the reason why we use > -fno-strict-aliasing in some cases. Hi Andrew, Yes, I understand that this is non complaint code. Unfortunately it is not the SPEC code but the code which is part of the gcc/glibc that is showing this. I ran into this when I built the cross toolchain for aarch64. I am sure that we are going to have this in tones of places. Thanks, Kugan ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain
Re: [ACTIVITY] 04 May -- 08 May
> - Trunk is broken and narrowed down the failure to > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66066 No the code is broken in SPEC ;). SPEC INT has been known to be broken C and they are not willing to do anything about it after the fact. This has been true for all versions of SPEC and one of the reason why we use -fno-strict-aliasing in some cases. From: linaro-toolchain on behalf of Kugan Sent: Sunday, May 10, 2015 9:28 PM To: Linaro Toolchain Subject: [ACTIVITY] 04 May -- 08 May == Progress == * Upstream GCC (4/10) - TCWG-779 Vector rtx costs for AArch64 ACKed to commit. - Asked to do fresh regression and benchmarking - re-based to latest trunk - Trunk is broken and narrowed down the failure to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66066 - Fixed and regression testing finished. Setting up benchmarking for spec2k before committing. * IRA (5/10) - Looked at https://bugs.linaro.org/show_bug.cgi?id=540 (TCWG-773 Median of three has unneeded register moves) - https://bugs.linaro.org/show_bug.cgi?id=1532 * Misc (1/10) - gcc-patches, gcc-bugs list - Meetings == Plan == - Continue with gcc stage1 activities - Look at register allocation ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain
[ACTIVITY] 04 May -- 08 May
== Progress == * Upstream GCC (4/10) - TCWG-779 Vector rtx costs for AArch64 ACKed to commit. - Asked to do fresh regression and benchmarking - re-based to latest trunk - Trunk is broken and narrowed down the failure to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66066 - Fixed and regression testing finished. Setting up benchmarking for spec2k before committing. * IRA (5/10) - Looked at https://bugs.linaro.org/show_bug.cgi?id=540 (TCWG-773 Median of three has unneeded register moves) - https://bugs.linaro.org/show_bug.cgi?id=1532 * Misc (1/10) - gcc-patches, gcc-bugs list - Meetings == Plan == - Continue with gcc stage1 activities - Look at register allocation ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain
[ACTIVITY] Week 19
== This week == * PR49551 (2/10) - Created new patch: http://pastebin.com/PsLTgvp7 - merge_decls() calls memcpy to copy newdecl into olddecl thus overriding DECL_COMMON in olddecl which is the source of the bug. If newdecl has common/nocommon attribute set, honor it, else DECL_COMMON would be set if both olddecl and newdecl have DECL_COMMON set. - Bootstrapped on x86, tested on x86, arm - Not yet figured out why r221297 masks this bug. * TCWG-619 (4/10) - managed to build chromium LTO on ARM, requires disabling components enabled by variable 'chromium_code'. Size of chrome executable reduces by 8 mb. - the bug: "error out of range" appears only during LTO build of components enabled by chromium_code. Results in very few instances of the same error message when LTO build for chromium_code components. - Disabling LTO in sandbox.gyp and passing -flto results in following error: /tmp/ccYJJbUK.ltrans25.ltrans.o::vtable for blink::WebDocument [clone .lto_priv.89343]: warning: relocation refers to discarded section /home/prathamesh.kulkarni/fsf-toolchain/lib/gcc/x86_64-unknown-linux-gnu/6.0.0/../../../../x86_64-unknown-linux-gnu/bin/ld.gold: error: hidden symbol '_ZN5blink11WebDocumentD1Ev' is not defined locally Also present for x86. The workaround is to pass -flto --param lto-partitions=1 as mentioned in PR57703. - gold ICE with v8 LTO build for ARM not reproducible with master branch. - Trying to use Martin Liska's scripts for gathering build stats: https://github.com/marxin/script-misc/blob/master/system_top.py https://github.com/marxin/script-misc/blob/master/vmstat_parser.py * Misc (4/10) - Exams preparation - Visa application == Next Week == - PR49551: figure out why r221297 masks the bug, write test-cases and submit patch upstream - chromium/v8 build stats - exams ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain