https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #24 from marcus at mc dot pp.se ---
I can confirm that with those two patches a 3-stage build of 14.2.1 completes
with stage 2 and 3 compared as equal.
The number of failed tests with "vect" somewhere in the name is down t
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #21 from marcus at mc dot pp.se ---
For completeness, I tried building gcc 14.2.1 using
BOOT_CFLAGS="-O2 -fno-tree-vectorize"
and the build completed, successfully comparing stages 2 and 3
as equal (although I wouldn&#
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #20 from marcus at mc dot pp.se ---
Hi.
Do you need more help to fix the broken autovectorization?
Again, you don't need ARM hardware; here is how I could reproduce the
miscompilation on a PPC64 host:
---8<---
wg
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #19 from marcus at mc dot pp.se ---
If you don't have ARM hardware, the issue can be seen in qemu:
hakua:/tmp% uname -a
Linux hakua 5.4.275 #1 SMP Wed May 8 17:18:04 CEST 2024 ppc64 POWER9, altivec
supported PowerNV T2P9D01 REV
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #18 from marcus at mc dot pp.se ---
vect-iv-7.c is probably a good (small) test case to start looking at.
The source code is scalar, but gets auto-vectorized by gcc 14 at -O2.
On big endian, the result arr[] becomes
10 8 14 12 18 16
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #17 from marcus at mc dot pp.se ---
Created attachment 60561
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60561&action=edit
Real regressions (PASS -> FAIL) from 13.3 to 14
So, I diffed gcc.sum between 13.3.1 and
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #16 from marcus at mc dot pp.se ---
Created attachment 60548
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60548&action=edit
Result from check-gcc on stage1 of gcc 14.3 on aarch64_be
The tests finally completed.
S
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #15 from marcus at mc dot pp.se ---
The stage1 testsuite is still running, but I'm seeing a lot of vector test
failures:
---8<---
FAIL: gcc.dg/vect/pr53185-2.c execution test
FAIL: gcc.dg/vect/pr56826.c execution test
FAIL
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #14 from marcus at mc dot pp.se ---
Thanks. Running tests now, will get back when the results are in. :-)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #5 from marcus at mc dot pp.se ---
> So stage 1 compiler is miscompiling stage 2.
Hm, isn't it the stage 2 compiler which is miscompiling libstdc++?
It's running .../build/./gcc/xgcc, not .../build/./prev-gcc/xgcc
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #12 from marcus at mc dot pp.se ---
If it's just a lack of hardware, a regular Raspberry Pi 3 can be used.
Compiling gcc on one takes 18.5 hours though... :-)
But yeah, if it's a miscompilation of the compiler, then it&
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #10 from marcus at mc dot pp.se ---
Ok. I did a build with STAGE1_CFLAGS="-O0" STAGE1_CXXFLAGS="-O0" before
(mentioned in the Gentoo bug, but not here, sorry) and got the same error,
but I can make a new build l
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #8 from marcus at mc dot pp.se ---
Ah, now I see what you mean. Yes, that sounds right. stage1 was built using
gcc 13.3.1 and can compile libstdc++ just fine.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #6 from marcus at mc dot pp.se ---
Oh, and you asked for the make invocation as well. It's
make --output-sync=line -j2 'LDFLAGS=-Wl,-O1 -Wl,--as-needed'
LIBPATH=/usr/lib/gcc/aarch64_be-unknown-linux-gnu/14 'STAG
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #4 from marcus at mc dot pp.se ---
I added the complete build log xz compressed (it's 8 megabytes uncompressed :).
The configure arguments are:
--host=aarch64_be-unknown-linux-gnu --build=aarch64_be-unknown-linux-gnu
--prefix
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118891
--- Comment #3 from marcus at mc dot pp.se ---
Created attachment 60509
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60509&action=edit
Complete build log
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: bootstrap
Assignee: unassigned at gcc dot gnu.org
Reporter: marcus at mc dot pp.se
Target Milestone: ---
When building gcc 14 (I tried 14.2 and 14.3; 13.3 is not affected) natively on
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110606
marcus at mc dot pp.se changed:
What|Removed |Added
CC||marcus at mc dot pp.se
Assignee: unassigned at gcc dot gnu.org
Reporter: marcus at mc dot pp.se
Hi.
Please consider the following function, compiled on PPC (32 bit):
uint32_t swap32(uint32_t *in)
{
#if 1
uint8_t a[] = {
((*in) & (uint32_t)0xff00UL)>>24,
((*in) & (uint32_
19 matches
Mail list logo
