[Bug bootstrap/102847] New: [12 regression] r12-4504 breaks powerpc64 build on power 7

[Bug bootstrap/102847] New: [12 regression] r12-4504 breaks powerpc64 build on power 7

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

            Bug ID: 102847
           Summary: [12 regression] r12-4504 breaks powerpc64 build on
                    power 7
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
          Assignee: unassigned at gcc dot gnu.org
          Reporter: seurer at gcc dot gnu.org
  Target Milestone: ---

g:e3ef92e79f9f1c4275a9e19652439089a310627d, r12-4504


libtool: compile:  /home/seurer/gcc/git/build/gcc-test/./gcc/xgcc
-B/home/seurer/gcc/git/build/gcc-test/./gcc/
-B/home/seurer/gcc/git/install/gcc-test/powerpc64-unknown-linux-gnu/bin/
-B/home/seurer/gcc/git/install/gcc-test/powerpc64-unknown-linux-gnu/lib/
-isystem
/home/seurer/gcc/git/install/gcc-test/powerpc64-unknown-linux-gnu/include
-isystem
/home/seurer/gcc/git/install/gcc-test/powerpc64-unknown-linux-gnu/sys-include
-DHAVE_CONFIG_H -I. -I/home/seurer/gcc/git/gcc-test/libgfortran
-iquote/home/seurer/gcc/git/gcc-test/libgfortran/io
-I/home/seurer/gcc/git/gcc-test/libgfortran/../gcc
-I/home/seurer/gcc/git/gcc-test/libgfortran/../gcc/config
-I/home/seurer/gcc/git/gcc-test/libgfortran/../libquadmath -I../.././gcc
-I/home/seurer/gcc/git/gcc-test/libgfortran/../libgcc -I../libgcc
-I/home/seurer/gcc/git/gcc-test/libgfortran/../libbacktrace -I../libbacktrace
-I../libbacktrace -std=gnu11 -Wall -Wstrict-prototypes -Wmissing-prototypes
-Wold-style-definition -Wextra -Wwrite-strings
-Werror=implicit-function-declaration -Werror=vla -fcx-fortran-rules
-ffunction-sections -fdata-sections -ffast-math -ftree-vectorize -funroll-loops
--param max-unroll-times=4 -g -O2 -MT matmul_i1.lo -MD -MP -MF
.deps/matmul_i1.Tpo -c
/home/seurer/gcc/git/gcc-test/libgfortran/generated/matmul_i1.c  -fPIC -DPIC -o
.libs/matmul_i1.o
during GIMPLE pass: vect
/home/seurer/gcc/git/gcc-test/libgfortran/generated/matmul_i1.c: In function
'matmul_i1':
/home/seurer/gcc/git/gcc-test/libgfortran/generated/matmul_i1.c:2450:1:
internal compiler error: in rs6000_builtin_vectorization_cost, at
config/rs6000/rs6000.c:5176
 2450 | matmul_i1 (gfc_array_i1 * const restrict retarray,
      | ^~~~~~~~~
0x110b9f47 rs6000_builtin_vectorization_cost
        /home/seurer/gcc/git/gcc-trunk/gcc/config/rs6000/rs6000.c:5176
0x10f13547 builtin_vectorization_cost
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.h:1520
0x10f13547 record_stmt_cost(vec<stmt_info_for_cost, va_heap, vl_ptr>*, int,
vect_cost_for_stmt, _stmt_vec_info*, tree_node*, int, vect_cost_model_location)
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-stmts.c:109
0x10f13f87 record_stmt_cost
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.h:1928
0x10f13f87 vect_get_load_cost(vec_info*, _stmt_vec_info*, int,
dr_alignment_support, int, bool, unsigned int*, unsigned int*,
vec<stmt_info_for_cost, va_heap, vl_ptr>*, vec<stmt_info_for_cost, va_heap,
vl_ptr>*, bool)
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-stmts.c:1240
0x10f3000b vect_model_load_cost
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-stmts.c:1195
0x10f3000b vectorizable_load
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-stmts.c:8783
0x10f3f55b vect_analyze_stmt(vec_info*, _stmt_vec_info*, bool*, _slp_tree*,
_slp_instance*, vec<stmt_info_for_cost, va_heap, vl_ptr>*)
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-stmts.c:10963
0x10f5f247 vect_analyze_loop_operations
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-loop.c:1779
0x10f61067 vect_analyze_loop_2
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-loop.c:2518
0x10f62e17 vect_analyze_loop(loop*, vec_info_shared*)
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-loop.c:3029
0x10fa6cbf try_vectorize_loop_1
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.c:1010
0x10fa7b9f vectorize_loops()
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.c:1249
0x10deb787 execute
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-ssa-loop.c:413



commit e3ef92e79f9f1c4275a9e19652439089a310627d (HEAD)
Author: Cl<C3><A9>ment Chigot <clement.chigot@atos.net>
Date:   Thu Oct 14 09:03:13 2021 +0200

    aix: ensure reference to __tls_get_addr is in text section.

[Bug bootstrap/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=102767

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I highly doubt r12-4504 broke bootstrap on Linux.

Also I suspect it is related to PR 102767.

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.0
          Component|bootstrap                   |target
           Keywords|                            |build, ice-on-valid-code

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
Dup.

*** This bug has been marked as a duplicate of bug 102767 ***

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

--- Comment #3 from seurer at gcc dot gnu.org ---
I reran my bisects on two different systems and both resolved to 

g:793d2549b173a0a2da6dd20ffc27acb9fd2de73e, r12-4501

as when the builds started failing.  r12-4500 worked on both systems.

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

Kewen Lin <linkw at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
   Last reconfirmed|                            |2021-10-21
     Ever confirmed|0                           |1
                 CC|                            |rguenth at gcc dot gnu.org
         Resolution|DUPLICATE                   |---

--- Comment #4 from Kewen Lin <linkw at gcc dot gnu.org> ---
(In reply to seurer from comment #3)
> I reran my bisects on two different systems and both resolved to 
> 
> g:793d2549b173a0a2da6dd20ffc27acb9fd2de73e, r12-4501
> 
> as when the builds started failing.  r12-4500 worked on both systems.

Thanks! It makes more sense, I tried my local fix for PR102767 on a Power7 BE
machine yesterday, this issue still existed, seems not an actual dup. Some
relevant commits are r11-1945 and r11-6717.

In function get_load_store_type, we just simply assign the
alignment_support_scheme to dr_unaligned_supported for VMAT_INVARIANT, without
going through the check in vect_supportable_dr_alignment (including target hook
checks).

IIUC, VMAT_INVARIANT memory accesses will end up with scalar accesses, it looks
intentional, but it's unexpected for rs6000_builtin_vectorization_cost, as
function rs6000_builtin_support_vector_misalignment says it only allows
element_size 32 and 64 for misalignment -1. In this failure, it's V16QI (aka,
element_size 8).

Not sure if it's a good idea to change *alignment_support_scheme to dr_aligned
for VMAT_INVARIANT (also for strided)?  Or force *misalignment to zero, then we
can distinguish this kind of special case in builtin_vectorization_cost hook to
avoid ICE?

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index 9cbc1af4cc9..8f527452bd0 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -1191,6 +1191,13 @@ vect_model_load_cost (vec_info *vinfo,
                                       ncopies * assumed_nunits,
                                       scalar_load, stmt_info, 0, vect_body);
     }
+  else if (memory_access_type == VMAT_INVARIANT)
+    {
+      /* Invariant loads will ideally be hoisted and splat to a vector.  */
+      prologue_cost += record_stmt_cost (cost_vec, 1,
+                                        scalar_to_vec, stmt_info, 0,
+                                        vect_prologue);
+    }
   else
     vect_get_load_cost (vinfo, stmt_info, ncopies,
                        alignment_support_scheme, misalignment, first_stmt_p,

should fix that in a better way.

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:7878981f1d44b4c9bb6c4a3137b37f06c725b73d

commit r12-4603-g7878981f1d44b4c9bb6c4a3137b37f06c725b73d
Author: Richard Biener <rguenther@suse.de>
Date:   Thu Oct 21 10:24:03 2021 +0200

    tree-optimization/102847 - properly cost VMAT_INVARIANT loads

    The following adds proper costing of VMAT_INVARIANT loads, avoiding
    to ask the target about the cost of an unsupported vector load cost
    which we won't emit anyway.

    2021-10-21  Richard Biener  <rguenther@suse.de>

            PR tree-optimization/102847
            * tree-vect-stmts.c (vect_model_load_cost): Explicitely
            handle VMAT_INVARIANT as a splat in the prologue.

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed hopefully.

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

--- Comment #8 from Kewen Lin <linkw at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #5)
> diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
> index 9cbc1af4cc9..8f527452bd0 100644
> --- a/gcc/tree-vect-stmts.c
> +++ b/gcc/tree-vect-stmts.c
> @@ -1191,6 +1191,13 @@ vect_model_load_cost (vec_info *vinfo,
>                                        ncopies * assumed_nunits,
>                                        scalar_load, stmt_info, 0, vect_body);
>      }
> +  else if (memory_access_type == VMAT_INVARIANT)
> +    {
> +      /* Invariant loads will ideally be hoisted and splat to a vector.  */
> +      prologue_cost += record_stmt_cost (cost_vec, 1,
> +                                        scalar_to_vec, stmt_info, 0,
> +                                        vect_prologue);
> +    }
>    else
>      vect_get_load_cost (vinfo, stmt_info, ncopies,
>                         alignment_support_scheme, misalignment, first_stmt_p,
> 
> should fix that in a better way.

Nice, I confirmed this change can fix the issue.  Thanks!

For costing, it seems we need one more scalar_load costing for vect_prologue?
or here we aggressively consider scalar_to_vec is also for load_and_splat?

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

--- Comment #9 from rguenther at suse dot de <rguenther at suse dot de> ---
On Thu, 21 Oct 2021, linkw at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847
> 
> --- Comment #8 from Kewen Lin <linkw at gcc dot gnu.org> ---
> (In reply to Richard Biener from comment #5)
> > diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
> > index 9cbc1af4cc9..8f527452bd0 100644
> > --- a/gcc/tree-vect-stmts.c
> > +++ b/gcc/tree-vect-stmts.c
> > @@ -1191,6 +1191,13 @@ vect_model_load_cost (vec_info *vinfo,
> >                                        ncopies * assumed_nunits,
> >                                        scalar_load, stmt_info, 0, vect_body);
> >      }
> > +  else if (memory_access_type == VMAT_INVARIANT)
> > +    {
> > +      /* Invariant loads will ideally be hoisted and splat to a vector.  */
> > +      prologue_cost += record_stmt_cost (cost_vec, 1,
> > +                                        scalar_to_vec, stmt_info, 0,
> > +                                        vect_prologue);
> > +    }
> >    else
> >      vect_get_load_cost (vinfo, stmt_info, ncopies,
> >                         alignment_support_scheme, misalignment, first_stmt_p,
> > 
> > should fix that in a better way.
> 
> Nice, I confirmed this change can fix the issue.  Thanks!
> 
> For costing, it seems we need one more scalar_load costing for vect_prologue?
> or here we aggressively consider scalar_to_vec is also for load_and_splat?

Yeah, true.  Let me follow up.

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:8516963942b0a8f06fa8f5ebd7c37f9edea151ac

commit r12-4604-g8516963942b0a8f06fa8f5ebd7c37f9edea151ac
Author: Richard Biener <rguenther@suse.de>
Date:   Thu Oct 21 12:10:20 2021 +0200

    tree-optimization/102847 - adjust VMAT_INVARIANT load costing

    This adds the missing scalar load cost in the prologue.

    2021-10-21  Richard Biener  <rguenther@suse.de>

            PR tree-optimization/102847
            * tree-vect-stmts.c (vect_model_load_cost): Add the scalar
            load cost in the prologue for VMAT_INVARIANT.

[Bug target/102847] [12 regression] r12-4504 breaks powerpc64 build on power 7

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102847

--- Comment #11 from seurer at gcc dot gnu.org ---
The build is indeed once again working on power 7.  Thanks!