[Bug ipa/92550] [10 Regression] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[Bug ipa/92550] [10 Regression] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

--- Comment #3 from Martin Jambor <jamborm at gcc dot gnu.org> ---
Almost certainly started with new IPA-SRA (r275982 or as we now call
it gcc-10-3311-gff6686d2e5f).  I looked at dumps from a cross-compiler
and the funny bit is, however, that new IPA-SRA simply does nothing.

That is not as it should be.  Because foo is not versionable, the pass
does not even look at it and then cannot do anything because it has
not seen a call to get_a.  But of course it should still analyze
outgoing calls to allow IPA-SRA of callees.

But that is merely a missed optimization, not this miscompilation.  I
looks almost as if it was simply the expand of misaligned structure
copy that is broken on (this?) strict-aliasing target.  I also believe
the test case does not successfuly run when compiled with earlier
revisions and option -fno-ipa-sra.

[Bug target/92550] [10 Regression] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

Martin Jambor <jamborm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|ipa                         |target

--- Comment #4 from Martin Jambor <jamborm at gcc dot gnu.org> ---
Not an IPA issue.

[Bug target/92550] [10 Regression] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bernd.edlinger at hotmail dot de

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Just a wild guess that this might be fallout of all the param/expand reorg
around strict alignment correctness for arm.

John, are you set up to bisect the testcase with an additional -fno-ipa-sra
flag?

[Bug target/92550] [10 Regression] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Might also be interesting to try other strict-align targets with -fno-ipa-sra.
But - this might also be a speciality of the callee-copy ABI of hpux.

[Bug target/92550] [10 Regression] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

--- Comment #7 from dave.anglin at bell dot net ---
On 2020-04-09 1:53 p.m., rguenth at gcc dot gnu.org wrote:
> Might also be interesting to try other strict-align targets with -fno-ipa-sra.
> But - this might also be a speciality of the callee-copy ABI of hpux.
I agree. The test doesn't fail on linux which we changed to caller copy.

[Bug target/92550] [10 Regression] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

--- Comment #8 from dave.anglin at bell dot net ---
On 2020-04-09 1:52 p.m., rguenth at gcc dot gnu.org wrote:
> John, are you set up to bisect the testcase with an additional -fno-ipa-sra
> flag?
I can bisect but not in a very automated way.

[Bug testsuite/92550] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |testsuite
            Summary|[10 Regression] FAIL:       |FAIL:
                   |gcc.dg/ipa/ipa-sra-8.c      |gcc.dg/ipa/ipa-sra-8.c
                   |execution test              |execution test

--- Comment #9 from John David Anglin <danglin at gcc dot gnu.org> ---
With -fno-ipa-sra, the test always fails since it was first introduced.

The test passes if I change the argument of get_a from SS to SSS.  Then,
it handles the misaligned argument:

get_a:
        .PROC
        .CALLINFO FRAME=0,NO_CALLS
        .ENTRY
        .stabn  68,0,14,L$M0001-L$FBB0001
L$M0001:
        ldb 0(%r26),%r19
        zdep %r19,7,8,%r19
        ldb 1(%r26),%r20
        zdep %r20,15,16,%r20
        ldb 2(%r26),%r28
        or %r20,%r19,%r20
        zdep %r28,23,24,%r28
        ldb 3(%r26),%r19
        or %r28,%r20,%r28
        .stabn  68,0,17,L$M0002-L$FBB0001
L$M0002:
        bv %r0(%r2)
        or %r19,%r28,%r28
        .EXIT
        .PROCEND

[Bug testsuite/92550] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

--- Comment #10 from John David Anglin <danglin at gcc dot gnu.org> ---
Created attachment 48257
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48257&action=edit
Patch to fix argument alignment

[Bug testsuite/92550] FAIL: gcc.dg/ipa/ipa-sra-8.c execution test

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92550

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Target Milestone|10.0                        |---
   Last reconfirmed|                            |2020-04-14
           Keywords|wrong-code                  |
             Status|UNCONFIRMED                 |NEW
           Priority|P4                          |P3

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
Indeed an obvious bug in the testcase, prone to fail on any strict-align
target.