[Bug c/107385] New: [asm goto] "=r" vs "+r" for outputs along indirect edges

[Bug c/107385] New: [asm goto] "=r" vs "+r" for outputs along indirect edges

[ndesaulniers at google dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

            Bug ID: 107385
           Summary: [asm goto] "=r" vs "+r" for outputs along indirect
                    edges
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ndesaulniers at google dot com
                CC: isanbard at gmail dot com
  Target Milestone: ---

I was playing around with adding support for outputs along indirect edges of
asm goto in clang.  When I was comparing codegen between various cases, I
noticed a case that looked like a bug to me:

int foo (void) {
    int x;
    asm goto ("": "=r"(x) ::: bar);
    x = 6;
bar:
    return x;
}

in gcc 12.2 with `-O2` produces:

foo:
        ret

The write to x from the inline asm should be dead along the fallthough or
default path out of the asm goto.  We should have one edge that assigns 6 to
the output.

Changing the output constraint modifier from =r to +r seems to produce the
expected code.

https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Extended-Asm mentions

> Be careful when you set output operands inside asm goto only on some possible control flow paths. If you don’t set up the output on given path and never use it on this path, it is okay. Otherwise, you should use ‘+’ constraint modifier meaning that the operand is input and output one. With this modifier you will have the correct values on all possible paths from the asm goto.

I'm not sure if that comment is alluding to this behavior, but it might be nice
to not have to pass in any prior value for output only variables?

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-10-24

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Expand is doing:

(insn 3 24 9 6 (set (reg/v:SI 82 [ <retval> ])
        (const_int 6 [0x6])) "/app/example.cpp":4:7 -1
     (nil))
(insn 9 3 14 6 (set (reg/v:SI 82 [ <retval> ])
        (reg:SI 83 [ x ])) "/app/example.cpp":3:5 -1
     (nil))


From this IR:
  <bb 2> [local count: 1073741824]:
  __asm__ goto("//" : "=r" x_2 :  :  : "bar" <L3>);
  goto <bb 4>; [35.00%]

  <bb 3> [local count: 697932184]:
<L3>:

  <bb 4> [local count: 1073741824]:
  # x_1 = PHI <x_2(3), 6(2)>
bar:

Note 6 does not need to be a constant here, it could be anything.
I suspect when the inline-asm goto is being expanded, it adds the assignment to
reg 82 on all edges going out of the inline-asm but then messing up because
there is some extra code due to the phi node.

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |jakub at gcc dot gnu.org
         Resolution|---                         |DUPLICATE

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Seems to be fixed by the PR113921 fix.  But more test coverage doesn't hurt, so
2024-02-15  Jakub Jelinek  <jakub@redhat.com>

        PR middle-end/107385
        * gcc.target/i386/pr107385.c: New test.

--- gcc/testsuite/gcc.target/i386/pr107385.c.jj 2024-01-13 00:05:00.077372302
+0100
+++ gcc/testsuite/gcc.target/i386/pr107385.c    2024-02-15 09:18:47.711260427
+0100
@@ -0,0 +1,20 @@
+/* PR middle-end/107385 */
+/* { dg-do run } */
+/* { dg-options "-O2" } */
+
+__attribute__((noipa)) int
+foo (void)
+{
+  int x;
+  asm goto ("": "=r" (x) : "0" (15), "a" (42) :: lab);
+  x = 6;
+lab:
+  return x;
+}
+
+int
+main ()
+{
+  if (foo () != 6)
+    __builtin_abort ();
+}

*** This bug has been marked as a duplicate of bug 113921 ***

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|DUPLICATE                   |FIXED

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Actually, the "a" constraint isn't needed there at all and without it it can be
used on other targets.  Like it fails on aarch64 without the patch the same.

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So
2024-02-15  Jakub Jelinek  <jakub@redhat.com>

        PR middle-end/107385
        * gcc.dg/pr107385.c: New test.

--- gcc/testsuite/gcc.dg/pr107385.c.jj  2024-01-13 00:05:00.077372302 +0100
+++ gcc/testsuite/gcc.dg/pr107385.c     2024-02-15 09:18:47.711260427 +0100
@@ -0,0 +1,20 @@
+/* PR middle-end/107385 */
+/* { dg-do run } */
+/* { dg-options "-O2" } */
+
+__attribute__((noipa)) int
+foo (void)
+{
+  int x;
+  asm goto ("": "=r" (x) : "0" (15) :: lab);
+  x = 6;
+lab:
+  return x;
+}
+
+int
+main ()
+{
+  if (foo () != 6)
+    __builtin_abort ();
+}

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #5 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:5459a9074afabf700f055fc8164f88dadb1c39b0

commit r14-9013-g5459a9074afabf700f055fc8164f88dadb1c39b0
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 15:55:25 2024 +0100

    testsuite: Add testcase for already fixed PR [PR107385]

    This testcase has been fixed by the PR113921 fix, but unlike testcase
    in there this one is not target specific.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: New test.

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #6 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:407b04b59f712ba41d1bcfbf86eba68c52e7917f

commit r13-8329-g407b04b59f712ba41d1bcfbf86eba68c52e7917f
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 15:55:25 2024 +0100

    testsuite: Add testcase for already fixed PR [PR107385]

    This testcase has been fixed by the PR113921 fix, but unlike testcase
    in there this one is not target specific.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: New test.

    (cherry picked from commit 5459a9074afabf700f055fc8164f88dadb1c39b0)

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #7 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:d2ebd5948bd21f54fdbc5ea99e391be59d0af64c

commit r12-10159-gd2ebd5948bd21f54fdbc5ea99e391be59d0af64c
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 15:55:25 2024 +0100

    testsuite: Add testcase for already fixed PR [PR107385]

    This testcase has been fixed by the PR113921 fix, but unlike testcase
    in there this one is not target specific.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: New test.

    (cherry picked from commit 5459a9074afabf700f055fc8164f88dadb1c39b0)

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #8 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:a6005ed41111f1200ce59501cfadd2d8f558ad90

commit r11-11240-ga6005ed41111f1200ce59501cfadd2d8f558ad90
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 15:55:25 2024 +0100

    testsuite: Add testcase for already fixed PR [PR107385]

    This testcase has been fixed by the PR113921 fix, but unlike testcase
    in there this one is not target specific.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: New test.

    (cherry picked from commit 5459a9074afabf700f055fc8164f88dadb1c39b0)

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #9 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:0d5d1c75f5c68b6064640c3154ae5f4c0b464905

commit r14-9017-g0d5d1c75f5c68b6064640c3154ae5f4c0b464905
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 20:04:01 2024 +0100

    testsuite: Require lra effective target for pr107385.c

    Old reload doesn't support asm goto with output operands.
    We have lra effective target (though, strangely it returns
    0 just for 2 targets out of at least 16 targets with no LRA support),
    so this patch uses it, similarly how it is done in other asm goto
    tests with output operands.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: Require lra effective target.

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #10 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:2354ccb493ddca6d2e1af9cc76ad45367932f1e2

commit r13-8330-g2354ccb493ddca6d2e1af9cc76ad45367932f1e2
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 20:04:01 2024 +0100

    testsuite: Require lra effective target for pr107385.c

    Old reload doesn't support asm goto with output operands.
    We have lra effective target (though, strangely it returns
    0 just for 2 targets out of at least 16 targets with no LRA support),
    so this patch uses it, similarly how it is done in other asm goto
    tests with output operands.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: Require lra effective target.

    (cherry picked from commit 0d5d1c75f5c68b6064640c3154ae5f4c0b464905)

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #11 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:7c81fed21c355565182e00bed58375e2f3dcf01c

commit r12-10160-g7c81fed21c355565182e00bed58375e2f3dcf01c
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 20:04:01 2024 +0100

    testsuite: Require lra effective target for pr107385.c

    Old reload doesn't support asm goto with output operands.
    We have lra effective target (though, strangely it returns
    0 just for 2 targets out of at least 16 targets with no LRA support),
    so this patch uses it, similarly how it is done in other asm goto
    tests with output operands.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: Require lra effective target.

    (cherry picked from commit 0d5d1c75f5c68b6064640c3154ae5f4c0b464905)

[Bug middle-end/107385] [asm goto] "=r" vs "+r" for outputs along indirect edges

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107385

--- Comment #12 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:da93a3d0976ad42aa6b4f5742119f3189ea21400

commit r11-11241-gda93a3d0976ad42aa6b4f5742119f3189ea21400
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Feb 15 20:04:01 2024 +0100

    testsuite: Require lra effective target for pr107385.c

    Old reload doesn't support asm goto with output operands.
    We have lra effective target (though, strangely it returns
    0 just for 2 targets out of at least 16 targets with no LRA support),
    so this patch uses it, similarly how it is done in other asm goto
    tests with output operands.

    2024-02-15  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/107385
            * gcc.dg/pr107385.c: Require lra effective target.

    (cherry picked from commit 0d5d1c75f5c68b6064640c3154ae5f4c0b464905)