The Linaro Toolchain Working Group (TCWG) is pleased to announce the
2015.09 snapshot of the Linaro GCC 5 source package.
This monthly snapshot[1] is based on FSF GCC 5.2+svn227732 and
includes performance improvements and bug fixes backported from
mainline GCC. This snapshot contents will be part of the 2015.11
stable [1] quarterly release.
This snapshot tarball is available on:
http://snapshots.linaro.org/components/toolchain/gcc-linaro/
Interesting changes in this GCC source package snapshot include:
* Updates to GCC 5.2+svn227732
* Backport of [Bugfix] [AArch32] PR target/26702
* Backport of [Bugfix] [AArch32] PR target/26702
* Backport of [Bugfix] [AArch32] PR rtl-optimization/34503
* Backport of [Bugfix] [AArch32] PR target/64208 iwmmxt pattern
* Backport of [Bugfix] [AArch32] PR target/65924
* Backport of [Bugfix] [AArch64] PR 65770 vstN_lane on bigendian
* Backport of [Bugfix] [AArch64] PR 65375 Fix RTX cost for vector SET
* Backport of [Bugfix] [AArch64] PR target/65491: Classify V1TF
vectors as AAPCS64 short vectors rather than composite types
* Backport of [Bugfix] [AArch64] PR target/66049
* Backport of [Bugfix] [AArch64] PR 63949
* Backport of [Bugfix] PR rtl-optimization/64616 Move insns without
introducing new temporaries in loop2_invariant
* Backport of [Bugfix] PR rtl-optimization/66076
* Backport of [Bugfix] PR tree-optimization/65447
* Backport of [AArch32] Add cpu_defines.h for ARM
* Backport of [AArch32] Additional bics patterns
* Backport of [AArch32] Add support for CFI directives in fp emulation
routines for ARM
* Backport of [AArch32] Add support for crtfastmath
* Backport of [AArch32] Apply arm.h change for previous commit
* Backport of [AArch32] (*arm_subsi3_insn): Fixed redundant alternatives
* Backport of [AArch32] Fix up bootstrap and fix typo in related changelog entry
* Backport of [AArch32] Handle UNSPEC_VOLATILE in rtx costs and don't
recurse inside the unspec
* Backport of [AArch32] insns attributes and alternative cleanups
* Backport of [AArch32] Make tune params tables more self-documenting
* Backport of [AArch32] Remove vec_shr and vec_shr optabs
* Backport of [AArch32] Use uppercase for code iterator names
* Backport of [AArch64] Add alternative 'extr' pattern, calculate rtx
cost properly
* Backport of [AArch64] Add branch-cost to cpu tuning information
* Backport of [AArch64] Add __extension__ and __always_inline__ to
crypto intrinsics
* Backport of [AArch64] Add vcond(u?)didi pattern
* Backport of [AArch64] Fix aarch64_rtx_costs of PLUS/MINUS
* Backport of [AArch64] Fix Cortex-A53 shift costs
* Backport of [AArch64] Fix geniterators.sh to use standard BRE syntax in sed
* Backport of [AArch64] Fix up new line in previous commit
* Backport of [AArch64] Handle FLOAT and UNSIGNED_FLOAT in rtx costs
* Backport of [AArch64] Idiomatic 64x1 comparisons in arm_neon.h
* Backport of [AArch64] Implement -m{cpu,tune,arch}=native using only
/proc/cpuinfo
* Backport of [AArch64] In aarch64_class_max_nregs use UNITS_PER_VREG
and UNITS_PER_WORD
* Backport of [AArch64] Make aarch64_min_divisions_for_recip_mul configurable
* Backport of [AArch64] Properly cost FABD pattern
* Backport of [AArch64] Properly cost MNEG/[SU]MNEGL patterns
* Backport of [AArch64] Properly handle mvn-register and add EON+shift
pattern and cost appropriately
* Backport of [AArch64] Properly handle SHIFT ops and EXTEND in
aarch64_rtx_mult_cost
* Backport of [AArch64] Remember to cost operand 0 in FP compare-with-0.0 case
* Backport of [AArch64] Use extend_arith rtx cost appropriately
* Backport of [AArch64] Use mov for add with large immediate
* Backport of [AArch64] Fix a couple of bugs regarding loop invariant
motion discovered by spec2k6 on aarch64
* Backport of [Musl libc] Add musl support to GCC
* Backport of [Musl libc] libitm fixes for musl support
* Backport of [Musl libc] musl libc config
* Backport of [Musl libc] mips musl support
* Backport of [Musl libc] unwind fix for musl
* Backport of [Musl libc] libstdc++, libgfortran gthr workaround for musl
* Backport of [Musl libc] fixincludes update for musl support
* Backport of [Musl libc] [AArch32] [4/13] arm musl support
* Backport of [Musl libc] [AArch64] [3/13] aarch64 musl support
* Backport of [Testsuite] [AArch32] advsimd-intrinsics.exp:
dg-do-what=compile if HW does not have Neon
* Backport of [Testsuite] [AArch32] Fix test for pr64616
* Backport of [Testsuite] [AArch32] Require Thumb2 effective target
* Backport of [Testsuite] [AArch32] Fix r222371 (PR target/26702)
* Backport of [Testsuite] Cleanup advsimd-intrinsics.exp, removing
unnecessary loop
* Backport of [Testsuite] don't clobber dg-do-what-default in
advsimd-intrinsics.exp
* Backport of [Testsuite] don't try to execute simd.exp tests on
targets without NEON
* Backport of [Testsuite] move check-gcc parallelize value into C front end
* Backport of [Testsuite] new vqmovn test
* Backport of [Testsuite] new vqmovun test
* Backport of [Testsuite] new vqrdmulh_lane test
* Backport of [Testsuite] new vqrdmulh_n test
* Backport of [Testsuite] new vqrdmulh test
* Backport of [Testsuite] new vqrshl test
* Backport of [Testsuite] new vqrshn_n test
* Backport of [Testsuite] new vqrshun_n test
* Backport of [Testsuite] new vqshl_n test
* Backport of [Testsuite] new vqshl test
* Backport of [Testsuite] new vqshlu_n test
* Backport of [Testsuite] new vqshrn_n test
* Backport of [Testsuite] new vqshrun_n test
* Backport of [Testsuite] Reinstate torture-init and torture-finalize
in advsimd-intrinsics.exp
* Backport of [Misc] Try REG_EQUAL for nonzero_bits
* Backport of [Misc] Don't reset ssa_name infor in struct iv
* Backport of [Misc] make clean' fix
* Backport of [Misc] Make vector_compare_rtx cope with VOID mode constants
* Backport of [Misc] set_nonzero_bits_and_sign_copies/combine.c
* Backport of [Misc] Expand pow (x, CONST) using square roots when possible
* Backport of [Doc] [AArch32] (ARM Options, mtune): add missing entries
* Backport of [Doc] Add missing jit and lto info.....
* Backport of [Doc] Declaring Attributes of Functions/split by target
* Backport of [Doc] reorganize (Type Attributes) and (Variable Attributes)
* Backport of [Doc] Update __atomic builtins documentation
* Backport of [Doc] Update definition location of attribute_spec in
documentation
Feedback and Support
Subscribe to the important Linaro mailing lists and join our IRC
channels to stay on top of Linaro development.
** Linaro Toolchain Development "mailing
list":http://lists.linaro.org/mailman/listinfo/linaro-toolchain
** Linaro Toolchain IRC channel on irc.freenode.net at @#linaro-tcwg@
* Bug reports should be filed in bugzilla against GCC
product:http://bugs.linaro.org/enter_bug.cgi?product=GCC
* Interested in commercial support? inquire at "Linaro
support":mailto:support@linaro.org
[1]. Stable source package releases are defined as releases where the
full Linaro Toolchain validation plan is executed.
[2]. Source package snapshots are defined when the compiler is only
put through unit-testing and full validation is not performed.
Ensure all critical (benchmarking) data is logged - TCWG-349 [1/10]
* Added logging of several factors
* Documented what we do and don't log
Noise control experiments on Juno - TCWG-349 [4/10]
* Rediscovered that my Juno is an r0, not an r1
* Rebuilt target image in a carefully scripted way
* Fiddled about with differences between local and LAVA targets
* Ran a couple of trials to test infrastructure
Connect preparation - [2/10]
* General sorting out of tickets et al
* Final pass through benchmarking presentation
Misc - [3/10]
=Plan=
Merge 'logging' branch
Final pass through logging documentation
Run some noise experiments on Juno
If time, work on Jenkins benchmarking jobs
== Progress ==
LLDB development
-- Progress to add support for un-alinged watchpoints on AArch64
[TCWG-367] [5/10]
-- Truncating watched bits wasnt accepted as a solution, will look
into issues during connect.
-- Trying to figure out a way to use multiple watchpoint slots for a
single watchpoint.
Miscellaneous [1/10]
-- Meetings, emails, discussions etc.
-- Travel preparations
Holiday 10 - 11 September 2015 - Travelling to US
== Plan ==
Holiday 14 - 18 September 2015
Connect 21 - 25 September 2015
Holiday 28 September - 5th October 2015
Hi,
I've tried probably a dozen different ABE build command lines with various combinations of cpu/arch/tune on the latest tree and on the 2014.09 release and am unable to build for the armv6/1136J-S.
Before bothering the list, I scoured the gzipped archive back through 2013, and there is little mention of any problems OR successes with armv6. I am wondering if armv6 is supported, and if so, what is the secret sauce ABE command line to build successfully for the 1136J-S.
Some of the builds I tried were missing tarball components, others fail later on such as the one below that gets through building gcc and bombs building the eabi.
--------------------
'abe.sh -target arm-linux-gnueabi -build all -set cpu=arm1136j-s' produces the following, see further below for the gcc version data:
arm-linux-gnueabi-gcc ../sysdeps/arm/aeabi_memclr.c -c -std=gnu99
-fgnu89-inline -O2 -Wall -Werror -Winline -Wno-error=undef -Wundef
-Wwrite-strings -fmerge-all-constants -frounding-math -g
-Wstrict-prototypes -I../include
-I/home/billd/_build/builds/x86_64-unknown-linux-gnu/arm-linux-gnueabi/glibc.git~release-2.21-master/csu
-I/home/billd/_build/builds/x86_64-unknown-linux-gnu/arm-linux-gnueabi/glibc.git~release-2.21-master
-I../sysdeps/unix/sysv/linux/arm -I../sysdeps/arm/nptl
-I../sysdeps/unix/sysv/linux/include -I../sysdeps/unix/sysv/linux
-I../sysdeps/nptl -I../sysdeps/pthread -I../sysdeps/gnu
-I../sysdeps/unix/inet -I../sysdeps/unix/sysv -I../sysdeps/unix/arm
-I../sysdeps/unix -I../sysdeps/posix -I../sysdeps/arm/armv6
-I../sysdeps/arm/include -I../sysdeps/arm -I../sysdeps/wordsize-32
-I../sysdeps/ieee754/flt-32 -I../sysdeps/ieee754/dbl-64
-I../sysdeps/ieee754 -I../sysdeps/generic -I.. -I../libio -I. -nostdinc
-isystem
/home/billd/_build/builds/destdir/x86_64-unknown-linux-gnu/lib/gcc/arm-linux-gnueabi/5.1.1/include
-isystem
/home/billd/_build/builds/destdir/x86_64-unknown-linux-gnu/lib/gcc/arm-linux-gnueabi/5.1.1/include-fixed
-isystem /home/billd/_build/sysroots/arm-linux-gnueabi/usr/include
-D_LIBC_REENTRANT -include
/home/billd/_build/builds/x86_64-unknown-linux-gnu/arm-linux-gnueabi/glibc.git~release-2.21-master/libc-modules.h
-DMODULE_NAME=libc -include ../include/libc-symbols.h -o
/home/billd/_build/builds/x86_64-unknown-linux-gnu/arm-linux-gnueabi/glibc.git~release-2.21-master/csu/aeabi_memclr.o
-MD -MP -MF
/home/billd/_build/builds/x86_64-unknown-linux-gnu/arm-linux-gnueabi/glibc.git~release-2.21-master/csu/aeabi_memclr.o.dt
-MT
/home/billd/_build/builds/x86_64-unknown-linux-gnu/arm-linux-gnueabi/glibc.git~release-2.21-master/csu/aeabi_memclr.o
/tmp/ccTT8UgL.s: Assembler messages:
/tmp/ccTT8UgL.s:492: Error: lo register required -- `add
pc,r3,#(0xffff0fc0-0xffff0fff)'
/tmp/ccTT8UgL.s:490: Error: invalid immediate: -61441 is out of range
make[2]: ***
[/home/billd/_build/builds/x86_64-unknown-linux-gnu/arm-linux-gnueabi/glibc.git~release-2.21-master/csu/libc-start.o]
Error 1
--- here is the gcc -v output for the build above. ----
./arm-linux-gnueabi-gcc -v
Using built-in specs.
COLLECT_GCC=./arm-linux-gnueabi-gcc
COLLECT_LTO_WRAPPER=/home/billd/_build/builds/destdir/x86_64-unknown-linux-gnu/libexec/gcc/arm-linux-gnueabi/5.1.1/lto-wrapper
Target: arm-linux-gnueabi
Configured with:
'/home/billd/_build/snapshots/gcc.git~linaro-gcc-5-branch/configure'
SHELL=/bin/bash --with-bugurl=https://bugs.linaro.org
--with-mpc=/home/billd/_build/builds/destdir/x86_64-unknown-linux-gnu
--with-mpfr=/home/billd/_build/builds/destdir/x86_64-unknown-linux-gnu
--with-gmp=/home/billd/_build/builds/destdir/x86_64-unknown-linux-gnu
--with-gnu-as --with-gnu-ld --disable-libstdcxx-pch --disable-libmudflap
--with-cpu=arm1136j-s --with-cloog=no --with-ppl=no --with-isl=no
--disable-nls --enable-c99 --with-fpu=vfpv3-d16 --with-float=softfp
--with-mode=thumb --disable-multilib --enable-multiarch --disable-libssp
--disable-libquadmath --disable-threads --without-headers --with-newlib
--disable-libmudflap --disable-bootstrap --disable-decimal-float
--disable-libgomp --disable-libatomic --disable-libsanitizer
--disable-plugins --disable-libitm MAKEINFO=echo --enable-languages=c
--with-sysroot=/home/billd/_build/builds/sysroot-arm-linux-gnueabi
--disable-shared --with-glibc-version=2.18
--build=x86_64-unknown-linux-gnu --host=x86_64-unknown-linux-gnu
--target=arm-linux-gnueabi
--prefix=/home/billd/_build/builds/destdir/x86_64-unknown-linux-gnu
Thread model: single
gcc version 5.1.1 20150608 (Linaro GCC 5.1-2015.06-1~dev)
== Progress ==
o Linaro GCC validation (8/10)
* Finished on-going backports validation
* Still dealing with jenkins/infra instability
* Prepared branch merge
* Continue on my new reviewing/validation tool
o Misc (2/10)
* Various meetings
* AArch64 libunwind patch review
== Plan ==
o Finish branch merge
o GCC snapshot
o Travel to SFO'15
== This week ==
* TCWG-316 -Exploit vector multiply by scalar instructions when multiple
scalars are used as
coefficients in a loop (5/10)
- Wrote patterns to allow combine pass to mergenon-standard multiply
by lane patterns.
* Bugzilla 57195 (mode iterator bug) blocked compiling new pattern (1/10)
- Updated patch based on upstream comments and re-sent upstream
* TCWG-77 - Transform end of loop conditions to min_expr (1/10)
- Writing dejagnu test case.
- Writing function to check for min expr support on target
* Misc (1/10)
- Conference calls
* Holiday (2/10)
== Next week ==
- Complete TCWG-77 and send upstream
- Continued investigation into TCWG-316
== This week ==
* TCWG-80 (1/10)
- PRE and dead-store elimination ipa pass (WIP upstream) already
handles optimization
* TCWG-120 (2/10)
- Investigating 3 possible approaches:
a) fold arm_andsi3_insn/arm_cmpsi_insn to
zeroextractsi_compare0_scratch/andsi3_compare0_scratch
b) Undo cse in test conditions, and modify arm_rtx_costs to fold
arm_andsi3_insn/arm_cmpsi_insn to
zeroextractsi_compare0_scratch/andsi3_compare0_scratch
c) Expand directly to zeroextractsi_compare0_scratch/andsi3_compare0_scratch
- Not sure whether the generated assembly is better (in terms of
speed) than with trunk.
asm diff at -O1: http://pastebin.com/yXBHHkhM
* TCWG-72 (2/10)
- Rebased Kugan's patch
* TCWG-299 (3/10)
- Simple workaround for PR65837 - configure gcc with --with-fpu=neon
- Firefox trunk doesn't build with gcc for arm, using apt-get source firefox
- LTO build fails with out-of-memory on my laptop in qemu-arm chroot
* TCWG-319 (1/10)
- Benchmarking setup with Bernie on Juno for running SPEC.
- SPEC Runs without error on Juno-{a53,a57} and APM
* Misc (1/10)
- Meetings
== Next Week ==
- Continue with TCWG-120, TCWG-72, TCWG-319
Hi!
The pre-built version of the stable version of Linaro Toolchain (Linaro
GDB 2015.02-3) for Windows is shipped with GDB 7.8-2014.09-1-git. GDB
was built with the following options:
(gdb) show configuration
This GDB was configured as follows:
configure --host=i686-w64-mingw32 --target=arm-linux-gnueabihf
--with-auto-load-dir=$debugdir:$datadir/auto-load
--with-auto-load-safe-path=$debugdir:$datadir/auto-load
--without-expat
--with-gdb-datadir=/home/buildslave/workspace/BinaryRelease/label/hetzner/target/arm-linux-gnueabihf/_build/builds/destdir/i686-w64-mingw32/share/gdb
(relocatable)
--with-jit-reader-dir=/home/buildslave/workspace/BinaryRelease/label/hetzner/target/arm-linux-gnueabihf/_build/builds/destdir/i686-w64-mingw32/lib/gdb
(relocatable)
--without-libunwind-ia64
--without-lzma
--without-guile
--with-separate-debug-dir=/home/buildslave/workspace/BinaryRelease/label/hetzner/target/arm-linux-gnueabihf/_build/builds/destdir/i686-w64-mingw32/lib/debug
(relocatable)
--without-zlib
--without-babeltrace
As you can see in the output GDB was built without expat support. This
is a major problem for widely used micro controllers like ARM Cortex A8
(e.g. used in Beaglebone Black), because the technical description of
this microcontroller is loaded out of xml files (arm-with-neon.xml part
of GDB). It's also possible to do that manually with "set tdesc filename
<xml-file>, before connecting to the target.
Without having the expat option enabled the g package, which is sent
from gdbserver during establishing a connection to gdb, can't be parsed
correctly. The result is in the first step a warning because of the
missing xml support and later on a error regarding the unexpected
content of the g package:
warning: Can not parse XML target description; XML support was disabled
at compile time
Remote 'g' packet reply is too long:
00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000f0fcffbe00000000407afdb6300000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
Old GDB versions of Linaro Toolchain like 7.6.1-2013.10 doesn't have
that problem because they were compiled with enabled expat support.
In my opinion there can be two root causes regarding that missing expat
support:
a) Expat support was explicitly disabled with --without-expat option
during the built of GDB.
b) On the build machine there wasn't a expat library available when gdb
was built.
Best regards,
Andreas Schmidl
== Progress ==
* Libraries (1/10)
- Working on getting the libc++ tests green
* Maintenance (6/10)
- Working with Vinicius on the gnueabi memcpy issue
- More TargetTuple/TargetParser discussions
- More sanitizers discussions, patch reviews
* Background (3/10)
- Code review, meetings, discussions, general support, etc.
- Connect stuff
== Plan ==
* Look a bit more on the library issues, try to upstream the bot changes
* Check instability in SciMark2
* Change HD from a D01
* Continue discussion about the sanitizer's multiple VMA problem
* Prepare for Connect, travel arrangements, etc.
