[Bug tree-optimization/114079] New: [14 regression] fftw fails tests again with -O3 -march=znver2 -m32

[Bug tree-optimization/114079] New: [14 regression] fftw fails tests again with -O3 -march=znver2 -m32

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114079

            Bug ID: 114079
           Summary: [14 regression] fftw fails tests again with -O3
                    -march=znver2 -m32
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: sjames at gcc dot gnu.org
  Target Milestone: ---

It's not the same as PR113059, although it's the same initial symptoms :(

```
wget https://www.fftw.org/fftw-3.3.10.tar.gz && tar xvf fftw-3.3.10.tar.xz
./configure CFLAGS="-O3 -march=znver2 -m32 -ggdb3"
make -j$(nproc)
make check -j$(nproc) # should fail, probably on tests/bench --verify
okd6912o01 (you can extract the verify param from gdb)
```

```
$ libtool --mode=execute gdb --args /home/sam/data/fftw/fftw-3.3.10/tests/bench
--verify okd6912o01
Reading symbols from /home/sam/data/fftw/fftw-3.3.10/tests/bench...
(gdb) r
Starting program: /home/sam/data/fftw/fftw-3.3.10/tests/bench --verify
okd6912o01
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib64/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
r2cf_32 (R0=0xffff9ee0, R1=0xffff9ff0, Cr=0xffff9ee0, Ci=<optimized out>,
rs=<optimized out>, csr=0x566c2080, csi=0x0, v=<optimized out>, ivs=1, ovs=1)
at r2cf_32.c:490
490                         Ci[WS(csi, 8)] = T29 - T28;
(gdb) bt
#0  r2cf_32 (R0=0xffff9ee0, R1=0xffff9ff0, Cr=0xffff9ee0, Ci=<optimized out>,
rs=<optimized out>, csr=0x566c2080, csi=0x0, v=<optimized out>, ivs=1, ovs=1)
at r2cf_32.c:490
#1  0x5663e323 in dobatch_r2hc (ego=0x566c1780, I=0x566b30b0, O=0x566b30b0,
buf=0xffff9ee0, batchsz=1) at direct-r2c.c:91
#2  0x5663e44d in iterate (ego=0x566c1780, I=<optimized out>, O=<optimized
out>, dobatch=<optimized out>) at direct-r2c.c:142
#3  0x565a86e5 in fftw_rdft_solve (ego_=0x566c1780, p_=0x566c1c50) at
solve.c:29
#4  0x56563d8b in measure (pln=<optimized out>, p=<optimized out>, iter=1) at
timer.c:136
#5  fftw_measure_execution_time (plnr=0x566a9710, pln=0x566c1780, p=0x566c1c50)
at timer.c:159
#6  0x56561053 in evaluate_plan (ego=ego@entry=0x566a9710,
pln=pln@entry=0x566c1780, p=p@entry=0x566c1c50) at planner.c:460
#7  0x565620a3 in search0 (ego=ego@entry=0x566a9710, p=p@entry=0x566c1c50,
slvndx=slvndx@entry=0xffffc2bc, flagsp=<optimized out>) at planner.c:529
#8  0x56562467 in search (ego=0x566a9710, p=0x566c1c50, slvndx=0xffffc2bc,
flagsp=0xffffc2c0) at planner.c:600
#9  mkplan (ego=<optimized out>, p=<optimized out>) at planner.c:711
#10 0x56562f00 in fftw_mkplan_d (ego=0x566a9710, p=0x566c1c50) at planner.c:970
#11 0x5663f3eb in mkcldw (ego_=0x566ac3a0, kind=R2HC00, r=32, m=216, ms=1, v=1,
vs=0, mstart=0, mcount=109, IO=0x566b30b0, plnr=0x566a9710) at
hc2hc-direct.c:205
#12 0x565a26d0 in mkplan (ego_=0x566ac3a0, p_=0x566c1830, plnr=0x566a9710) at
hc2hc.c:142
#13 0x56562151 in invoke_solver (ego=0x566a9710, p=<optimized out>,
s=<optimized out>, nflags=<optimized out>) at planner.c:486
#14 search0 (ego=ego@entry=0x566a9710, p=p@entry=0x566c1830,
slvndx=slvndx@entry=0xffffc4ec, flagsp=<optimized out>) at planner.c:529
#15 0x56562467 in search (ego=0x566a9710, p=0x566c1830, slvndx=0xffffc4ec,
flagsp=0xffffc4f0) at planner.c:600
#16 mkplan (ego=<optimized out>, p=<optimized out>) at planner.c:711
#17 0x56562f00 in fftw_mkplan_d (ego=0x566a9710, p=0x566c1830) at planner.c:970
#18 0x5662af0f in mkplan (ego_=0x566b3080, p_=0x566c0da0, plnr=0x566a9710) at
reodft010e-r2hc.c:357
#19 0x56562151 in invoke_solver (ego=0x566a9710, p=<optimized out>,
s=<optimized out>, nflags=<optimized out>) at planner.c:486
#20 search0 (ego=ego@entry=0x566a9710, p=p@entry=0x566c0da0,
slvndx=slvndx@entry=0xffffc6cc, flagsp=<optimized out>) at planner.c:529
#21 0x565628a8 in search (ego=0x566a9710, p=0x566c0da0, slvndx=0xffffc6cc,
flagsp=0xffffc6d0) at planner.c:600
#22 mkplan (ego=<optimized out>, p=<optimized out>) at planner.c:711
#23 0x5655ea4a in mkplan0 (plnr=0x566a9710, flags=1, prb=<optimized out>,
hash_info=0, wisdom_state=WISDOM_NORMAL) at apiplan.c:42
#24 mkplan (plnr=0x566a9710, flags=1, prb=<optimized out>, hash_info=0) at
apiplan.c:56
#25 fftw_mkapiplan (sign=<optimized out>, flags=<optimized out>, prb=<optimized
out>) at apiplan.c:124
#26 0x56560464 in fftw_plan_many_r2r (rank=1, n=0xffffc860, howmany=1,
in=0x5668e680, inembed=<optimized out>, istride=1, idist=1, out=0x5669bf00,
onembed=0x0, ostride=1, odist=1,
    kind=0xffffc86c, flags=1) at plan-many-r2r.c:40
#27 0x5656060a in fftw_plan_r2r (rank=1, n=0xffffc860, in=0x5668e680,
out=0x5669bf00, kind=0xffffc86c, flags=1) at plan-r2r.c:26
#28 0x565604ac in fftw_plan_r2r_1d (n=<optimized out>, in=0x5668e680,
out=0x5669bf00, kind=<optimized out>, flags=1) at plan-r2r-1d.c:25
#29 0x5655bc19 in mkplan_r2r (p=0x5668e080, flags=1) at bench.c:451
#30 mkplan (p=0x5668e080, flags=1) at bench.c:525
#31 0x5655d980 in setup (p=0x5668e080) at fftw-bench.c:243
#32 0x56642d6b in verify (param=0xffffcce2 "okd6912o01", rounds=10, tol=1e-10)
at verify.c:55
#33 0x5663ff99 in bench_main (argc=<optimized out>, argv=<optimized out>) at
bench-main.c:107
#34 0x56559337 in main (argc=3, argv=0xffffca94) at main.c:38
```

[Bug tree-optimization/114079] [14 regression] fftw fails tests again with -O3 -march=znver2 -m32

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114079

--- Comment #1 from Sam James <sjames at gcc dot gnu.org> ---
I'll bisect it now.

[Bug tree-optimization/114079] [14 regression] fftw fails tests again with -O3 -march=znver2 -m32

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114079

Sam James <sjames at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #2 from Sam James <sjames at gcc dot gnu.org> ---
Actually, I think the test is really flaky (sometimes it pases, but it'll fail
if you try it a few times). If I checkout Jakub's fix from PR113059 in a loop,
it still fails. It's surely the same thing then?

*** This bug has been marked as a duplicate of bug 113059 ***