[Bug c++/102581] New: ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2

[Bug c++/102581] New: ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

            Bug ID: 102581
           Summary: ice in forced_merge, at ipa-modref-tree.h:352 with
                    -fno-strict-aliasing and -O2
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dcb314 at hotmail dot com
  Target Milestone: ---

Created attachment 51542
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51542&action=edit
gzipped C++ source code

The attached C++ source code, with compiler flags -fno-strict-aliasing
and -O2, does this:

/home/dcb35/rpmbuild/BUILD/Vulkan-Tools-sdk-1.2.189.0/vulkaninfo/vulkaninfo.cpp:1082:1:
internal compiler error: in forced_merge, at ipa-modref-tree.h:352

The bug first occurs sometime before git hash 2fcfc03459a907c0,
from about a month ago.

I will try to reduce this source code downto somethine sensible.

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |marxin at gcc dot gnu.org
           Keywords|                            |ice-on-valid-code
            Summary|ice in forced_merge, at     |[12 Regression] ice in
                   |ipa-modref-tree.h:352 with  |forced_merge, at
                   |-fno-strict-aliasing and    |ipa-modref-tree.h:352 with
                   |-O2                         |-fno-strict-aliasing and
                   |                            |-O2
   Target Milestone|---                         |7.3
          Component|c++                         |ipa

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
      /* We assume that containment and lossless merging
         was tested earlier.  */
      gcc_checking_assert (!contains (a) && !a.contains (*this)
                           && !merge (a, record_adjustments));

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

--- Comment #2 from David Binderman <dcb314 at hotmail dot com> ---
git blame says:

f5ff3a8ed4ca (Jan Hubicka   2021-08-28 20:57:08 +0200  349)       /* We assume
that containment and lossless merging
f5ff3a8ed4ca (Jan Hubicka   2021-08-28 20:57:08 +0200  350)      was tested
earlier.  */
f5ff3a8ed4ca (Jan Hubicka   2021-08-28 20:57:08 +0200  351)      
gcc_checking_assert (!contains (a) && !a.contains (*this)
f5ff3a8ed4ca (Jan Hubicka   2021-08-28 20:57:08 +0200  352)                    
   && !merge (a, record_adjustments));

Adding Jan for his best advice.

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-10-04
     Ever confirmed|0                           |1
            Summary|[12 Regression] ice in      |[12 Regression] ice in
                   |forced_merge, at            |forced_merge, at
                   |ipa-modref-tree.h:352 with  |ipa-modref-tree.h:352 with
                   |-fno-strict-aliasing and    |-fno-strict-aliasing and
                   |-O2                         |-O2 since
                   |                            |r12-3202-gf5ff3a8ed4ca9173
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=102557
             Status|UNCONFIRMED                 |NEW

--- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
Confirmed, similar to PR102557.

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|7.3                         |12.0
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |marxin at gcc dot gnu.org

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

--- Comment #4 from David Binderman <dcb314 at hotmail dot com> ---
Reduced C++ source code, after a pretty hefty four hours of reduction, is

enum VkStructureType {
  VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_EXT,
  VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR
} typedef VkPhysicalDeviceSparseProperties;
struct VkPhysicalDeviceProperties {
  int apiVersion;
  VkPhysicalDeviceSparseProperties sparseProperties;
};
typedef struct {
  VkStructureType sType;
  int *pPhysicalDevices;
} VkPhysicalDeviceFeatures2;
typedef struct VkPhysicalDeviceProperties2 {
  VkStructureType sType;
  void *pNext;
} VkPhysicalDeviceMemoryProperties2;
struct VulkanVersion {
  int major;
  int minor;
  int patch;
};
int make_vulkan_version_version;
VulkanVersion make_vulkan_version() {
  return {make_vulkan_version_version, make_vulkan_version_version,
          make_vulkan_version_version};
}
struct AppGpu {
  int &inst;
  int id;
  int *phys_device = nullptr;
  VulkanVersion api_version{};
  VkPhysicalDeviceProperties props{};
  VkPhysicalDeviceProperties2 props2{};
  int memory_props{};
  VkPhysicalDeviceMemoryProperties2 memory_props2{};
  int features{};
  VkPhysicalDeviceFeatures2 features2{};
  int *dev = nullptr;
  int enabled_features{};
  int AppGpu_phys_device;
  int AppGpu_inst;
  AppGpu() : inst(AppGpu_inst), id() {
    api_version = make_vulkan_version();
    props2.sType = memory_props2.sType = features2.sType =
        VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR;
  }
};
main() { AppGpu(); }

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

--- Comment #5 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to David Binderman from comment #4)
> Reduced C++ source code, after a pretty hefty four hours of reduction, is

Heh, my reduction has just finished as well, resulting in the same source code
as yours.

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|marxin at gcc dot gnu.org          |hubicka at gcc dot gnu.org

--- Comment #6 from Martin Liška <marxin at gcc dot gnu.org> ---
I'm not familiar enough with the code, leaving to Honza.

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

--- Comment #7 from Jan Hubicka <hubicka at ucw dot cz> ---
Hi,
the problem is that we assume that merge is symmetric (merging a to b
succeeds if and only if merging b to a succeeds). There was one
symetrical path missing in the (fancy and bit ugly) logic on what we can
merge losslessly.

I also added sanity check for this assumption.  So bugs like this should
fire more often.

Honza

diff --git a/gcc/ipa-modref-tree.h b/gcc/ipa-modref-tree.h
index 6a9ed5ce54b..61b4b16dbed 100644
--- a/gcc/ipa-modref-tree.h
+++ b/gcc/ipa-modref-tree.h
@@ -234,6 +234,17 @@ struct GTY(()) modref_access_node
                      record_adjustments);
              return true;
            }
+         if (known_size_p (a.size)
+             && (!known_size_p (size) || known_lt (size, a.size)))
+           {
+             if (((known_size_p (max_size) || known_size_p (a.max_size))
+                  && !known_eq (max_size, a.max_size))
+                  || !known_eq (offset1, aoffset1))
+               return false;
+             update (new_parm_offset, offset1, size, max_size,
+                     record_adjustments);
+             return true;
+           }
          /* If sizes are same, we can extend the interval.  */
          if ((known_size_p (size) || known_size_p (a.size))
              && !known_eq (size, a.size))
@@ -618,6 +629,7 @@ private:
            found = true;
          if (!found && n->merge (*a, false))
            found = restart = true;
+         gcc_checking_assert (found || !a->merge (*n, false));
          if (found)
            {
              accesses->unordered_remove (i);

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

--- Comment #8 from Jan Hubicka <hubicka at ucw dot cz> ---
Actually, this is shorter patch - we already should notice that one
range is contained in other, but we give up too early.

Honza

diff --git a/gcc/ipa-modref-tree.h b/gcc/ipa-modref-tree.h
index 6a9ed5ce54b..8e9b89b3e2c 100644
--- a/gcc/ipa-modref-tree.h
+++ b/gcc/ipa-modref-tree.h
@@ -110,8 +110,11 @@ struct GTY(()) modref_access_node
               if (!a.parm_offset_known)
                 return false;
               /* Accesses are never below parm_offset, so look
-                 for smaller offset.  */
-              if (!known_le (parm_offset, a.parm_offset))
+                 for smaller offset.
+                 If access ranges are known still allow merging
+                 when bit offsets comparsion passes.  */
+              if (!known_le (parm_offset, a.parm_offset)
+                  && !range_info_useful_p ())
                 return false;
               aoffset_adj = (a.parm_offset - parm_offset)
                             << LOG2_BITS_PER_UNIT;
@@ -618,6 +621,7 @@ private:
            found = true;
          if (!found && n->merge (*a, false))
            found = restart = true;
+         gcc_checking_assert (found || !a->merge (*n, false));
          if (found)
            {
              accesses->unordered_remove (i);

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jan Hubicka <hubicka@gcc.gnu.org>:

https://gcc.gnu.org/g:44b61586d8640b79e78cfdb6a555200ccee8df77

commit r12-4227-g44b61586d8640b79e78cfdb6a555200ccee8df77
Author: Jan Hubicka <jh@suse.cz>
Date:   Thu Oct 7 15:26:01 2021 +0200

    Fix access node merging

    gcc/ChangeLog:

    2021-10-07  Jan Hubicka  <hubicka@ucw.cz>

            PR ipa/102581
            * ipa-modref-tree.h (modref_access_node::contains_p): Handle
offsets
            better.
            (modref_access_node::try_merge_with): Add sanity check that there
            are no redundant entries in the list.

    gcc/testsuite/ChangeLog:

    2021-10-07  Jan Hubicka  <hubicka@ucw.cz>

            * g++.dg/torture/pr102581.C: New test.

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

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

--- Comment #10 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Fixed.

[Bug ipa/102581] [12 Regression] ice in forced_merge, at ipa-modref-tree.h:352 with -fno-strict-aliasing and -O2 since r12-3202-gf5ff3a8ed4ca9173

[ian at airs dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102581

Ian Lance Taylor <ian at airs dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ian at airs dot com

--- Comment #11 from Ian Lance Taylor <ian at airs dot com> ---
*** Bug 102474 has been marked as a duplicate of this bug. ***