Bug 100623 - [11 Regression] wrong code with -Os -fno-dce -fno-defer-pop -fno-forward-propagate -flive-range-shrinkage -fno-rerun-cse-after-loop -mno-push-args since r10-7515-g2c0fa3ecf70d199a
Summary: [11 Regression] wrong code with -Os -fno-dce -fno-defer-pop -fno-forward-prop...
Status: NEW
Alias: None
Product: gcc
Classification: Unclassified
Component: target (show other bugs)
Version: 12.0
: P2 normal
Target Milestone: 11.5
Assignee: Not yet assigned to anyone
URL:
Keywords: ra, wrong-code
Depends on:
Blocks:
 
Reported: 2021-05-16 14:46 UTC by Zdenek Sojka
Modified: 2024-03-11 03:39 UTC (History)
3 users (show)

See Also:
Host: x86_64-pc-linux-gnu
Target: x86_64-pc-linux-gnu
Build:
Known to work:
Known to fail: 10.3.1, 11.1.1, 11.2.0, 12.0
Last reconfirmed: 2021-05-17 00:00:00

Attachments
reduced testcase (from openssl sources) (325 bytes, text/plain)
2021-05-16 14:46 UTC, Zdenek Sojka 		Details
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Zdenek Sojka 2021-05-16 14:46:35 UTC 
Created attachment 50819 [details]
reduced testcase (from openssl sources)

Output:
$ x86_64-pc-linux-gnu-gcc -Os -fno-dce -fno-defer-pop -fno-forward-propagate -flive-range-shrinkage -fno-rerun-cse-after-loop -mno-push-args testcase.c
$ ./a.out 
Aborted

bn_add_words() is called with r == NULL

$ x86_64-pc-linux-gnu-gcc -v
Using built-in specs.
COLLECT_GCC=/repo/gcc-trunk/binary-latest/bin/x86_64-pc-linux-gnu-gcc
COLLECT_LTO_WRAPPER=/repo/gcc-trunk/binary-trunk-r12-815-20210516001637-g1ba3f0ef546-checking-yes-rtl-df-extra-amd64/bin/../libexec/gcc/x86_64-pc-linux-gnu/12.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: /repo/gcc-trunk//configure --enable-languages=c,c++ --enable-valgrind-annotations --disable-nls --enable-checking=yes,rtl,df,extra --with-cloog --with-ppl --with-isl --build=x86_64-pc-linux-gnu --host=x86_64-pc-linux-gnu --target=x86_64-pc-linux-gnu --with-ld=/usr/bin/x86_64-pc-linux-gnu-ld --with-as=/usr/bin/x86_64-pc-linux-gnu-as --disable-libstdcxx-pch --prefix=/repo/gcc-trunk//binary-trunk-r12-815-20210516001637-g1ba3f0ef546-checking-yes-rtl-df-extra-amd64
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 12.0.0 20210516 (experimental) (GCC)
Comment 1 Martin Liška 2021-05-17 08:05:19 UTC 
Confirmed, started with r10-7515-g2c0fa3ecf70d199a.
Comment 2 Andrew Pinski 2022-01-29 23:48:46 UTC 
Looks to be fixed on the trunk.
Comment 3 Andrew Pinski 2022-01-29 23:55:14 UTC 
There seems to be some IR difference on the gimple level which might be making this latent.
In GCC 11.2.0 we have (bn_add_words) :
  _2 = n_3 & -4;
...
  if (_2 != 0)

While on the trunk we have:
  _14 = (unsigned int) n_3;
  if (_14 > 3)

This does look like a valid transformation knowing that n is subtracted by 4.
Comment 4 Andrew Pinski 2022-01-29 23:59:33 UTC 
(In reply to Andrew Pinski from comment #3)
> There seems to be some IR difference on the gimple level which might be
> making this latent.

But that is not where the problem is located really as far as I can tell, the assembly for bn_mul_recursive is different too (and it looks like the difference in the assembly for bn_add_words is only different in the comparison and nothing else [no other code changes]).
Comment 5 Martin Liška 2022-01-31 14:38:27 UTC 
It's fixed on master with r12-5944-ga7acb6dca941db2b. Which is a revision that changes LRA costs.
Comment 6 Jakub Jelinek 2022-06-28 10:45:03 UTC 
GCC 10.4 is being released, retargeting bugs to GCC 10.5.
Comment 7 Richard Biener 2023-07-07 10:39:58 UTC 
GCC 10 branch is being closed.
Comment 8 Jeffrey A. Law 2024-03-11 03:39:11 UTC 
Fixed in gcc-12 and newer.  Adjusting regression markers.