Re: [PATCH] arm: Fix MVE intrinsics support with LTO (PR target/110268)

[Christophe Lyon <christophe.lyon@linaro.org>]

[-- Attachment #1: Type: text/plain, Size: 13004 bytes --]

On Wed, 5 Jul 2023 at 19:07, Kyrylo Tkachov <Kyrylo.Tkachov@arm.com> wrote:

> Hi Christophe,
>
> > -----Original Message-----
> > From: Christophe Lyon <christophe.lyon@linaro.org>
> > Sent: Monday, June 26, 2023 4:03 PM
> > To: gcc-patches@gcc.gnu.org; Kyrylo Tkachov <Kyrylo.Tkachov@arm.com>;
> > Richard Sandiford <Richard.Sandiford@arm.com>
> > Cc: Christophe Lyon <christophe.lyon@linaro.org>
> > Subject: [PATCH] arm: Fix MVE intrinsics support with LTO (PR
> target/110268)
> >
> > After the recent MVE intrinsics re-implementation, LTO stopped working
> > because the intrinsics would no longer be defined.
> >
> > The main part of the patch is simple and similar to what we do for
> > AArch64:
> > - call handle_arm_mve_h() from arm_init_mve_builtins to declare the
> >   intrinsics when the compiler is in LTO mode
> > - actually implement arm_builtin_decl for MVE.
> >
> > It was just a bit tricky to handle __ARM_MVE_PRESERVE_USER_NAMESPACE:
> > its value in the user code cannot be guessed at LTO time, so we always
> > have to assume that it was not defined.  The led to a few fixes in the
> > way we register MVE builtins as placeholders or not.  Without this
> > patch, we would just omit some versions of the inttrinsics when
> > __ARM_MVE_PRESERVE_USER_NAMESPACE is true. In fact, like for the C/C++
> > placeholders, we need to always keep entries for all of them to ensure
> > that we have a consistent numbering scheme.
> >
> >       2023-06-26  Christophe Lyon   <christophe.lyon@linaro.org>
> >
> >       PR target/110268
> >       gcc/
> >       * config/arm/arm-builtins.cc (arm_init_mve_builtins): Handle LTO.
> >       (arm_builtin_decl): Hahndle MVE builtins.
> >       * config/arm/arm-mve-builtins.cc (builtin_decl): New function.
> >       (add_unique_function): Fix handling of
> >       __ARM_MVE_PRESERVE_USER_NAMESPACE.
> >       (add_overloaded_function): Likewise.
> >       * config/arm/arm-protos.h (builtin_decl): New declaration.
> >
> >       gcc/testsuite/
> >       * gcc.target/arm/pr110268-1.c: New test.
> >       * gcc.target/arm/pr110268-2.c: New test.
> > ---
> >  gcc/config/arm/arm-builtins.cc            | 11 +++-
> >  gcc/config/arm/arm-mve-builtins.cc        | 61 ++++++++++++-----------
> >  gcc/config/arm/arm-protos.h               |  1 +
> >  gcc/testsuite/gcc.target/arm/pr110268-1.c | 11 ++++
> >  gcc/testsuite/gcc.target/arm/pr110268-2.c | 22 ++++++++
> >  5 files changed, 76 insertions(+), 30 deletions(-)
> >  create mode 100644 gcc/testsuite/gcc.target/arm/pr110268-1.c
> >  create mode 100644 gcc/testsuite/gcc.target/arm/pr110268-2.c
> >
> > diff --git a/gcc/config/arm/arm-builtins.cc
> b/gcc/config/arm/arm-builtins.cc
> > index 36365e40a5b..fca7dcaf565 100644
> > --- a/gcc/config/arm/arm-builtins.cc
> > +++ b/gcc/config/arm/arm-builtins.cc
> > @@ -1918,6 +1918,15 @@ arm_init_mve_builtins (void)
> >        arm_builtin_datum *d = &mve_builtin_data[i];
> >        arm_init_builtin (fcode, d, "__builtin_mve");
> >      }
> > +
> > +  if (in_lto_p)
> > +    {
> > +      arm_mve::handle_arm_mve_types_h ();
> > +      /* Under LTO, we cannot know whether
> > +      __ARM_MVE_PRESERVE_USER_NAMESPACE was defined, so assume
> > it
> > +      was not.  */
> > +      arm_mve::handle_arm_mve_h (false);
> > +    }
> >  }
> >
> >  /* Set up all the NEON builtins, even builtins for instructions that
> are not
> > @@ -2723,7 +2732,7 @@ arm_builtin_decl (unsigned code, bool initialize_p
> > ATTRIBUTE_UNUSED)
> >      case ARM_BUILTIN_GENERAL:
> >        return arm_general_builtin_decl (subcode);
> >      case ARM_BUILTIN_MVE:
> > -      return error_mark_node;
> > +      return arm_mve::builtin_decl (subcode);
> >      default:
> >        gcc_unreachable ();
> >      }
> > diff --git a/gcc/config/arm/arm-mve-builtins.cc b/gcc/config/arm/arm-mve-
> > builtins.cc
> > index 7033e41a571..e9a12f27411 100644
> > --- a/gcc/config/arm/arm-mve-builtins.cc
> > +++ b/gcc/config/arm/arm-mve-builtins.cc
> > @@ -493,6 +493,16 @@ handle_arm_mve_h (bool
> > preserve_user_namespace)
> >                                    preserve_user_namespace);
> >  }
> >
> > +/* Return the function decl with SVE function subcode CODE, or
> > error_mark_node
> > +   if no such function exists.  */
> > +tree
> > +builtin_decl (unsigned int code)
> > +{
> > +  if (code >= vec_safe_length (registered_functions))
> > +    return error_mark_node;
> > +  return (*registered_functions)[code]->decl;
> > +}
> > +
> >  /* Return true if CANDIDATE is equivalent to MODEL_TYPE for overloading
> >     purposes.  */
> >  static bool
> > @@ -849,7 +859,6 @@ function_builder::add_function (const
> > function_instance &instance,
> >      ? integer_zero_node
> >      : simulate_builtin_function_decl (input_location, name, fntype,
> >                                     code, NULL, attrs);
> > -
> >    registered_function &rfn = *ggc_alloc <registered_function> ();
> >    rfn.instance = instance;
> >    rfn.decl = decl;
> > @@ -889,15 +898,12 @@ function_builder::add_unique_function (const
> > function_instance &instance,
> >    gcc_assert (!*rfn_slot);
> >    *rfn_slot = &rfn;
> >
> > -  /* Also add the non-prefixed non-overloaded function, if the user
> > namespace
> > -     does not need to be preserved.  */
> > -  if (!preserve_user_namespace)
> > -    {
> > -      char *noprefix_name = get_name (instance, false, false);
> > -      tree attrs = get_attributes (instance);
> > -      add_function (instance, noprefix_name, fntype, attrs,
> requires_float,
> > -                 false, false);
> > -    }
> > +  /* Also add the non-prefixed non-overloaded function, as placeholder
> > +     if the user namespace does not need to be preserved.  */
> > +  char *noprefix_name = get_name (instance, false, false);
> > +  attrs = get_attributes (instance);
> > +  add_function (instance, noprefix_name, fntype, attrs, requires_float,
> > +             false, preserve_user_namespace);
> >
> >    /* Also add the function under its overloaded alias, if we want
> >       a separate decl for each instance of an overloaded function.  */
> > @@ -905,20 +911,17 @@ function_builder::add_unique_function (const
> > function_instance &instance,
> >    if (strcmp (name, overload_name) != 0)
> >      {
> >        /* Attribute lists shouldn't be shared.  */
> > -      tree attrs = get_attributes (instance);
> > +      attrs = get_attributes (instance);
> >        bool placeholder_p = !(m_direct_overloads ||
> force_direct_overloads);
> >        add_function (instance, overload_name, fntype, attrs,
> >                   requires_float, false, placeholder_p);
> >
> > -      /* Also add the non-prefixed overloaded function, if the user
> namespace
> > -      does not need to be preserved.  */
> > -      if (!preserve_user_namespace)
> > -     {
> > -       char *noprefix_overload_name = get_name (instance, false, true);
> > -       tree attrs = get_attributes (instance);
> > -       add_function (instance, noprefix_overload_name, fntype, attrs,
> > -                     requires_float, false, placeholder_p);
> > -     }
> > +      /* Also add the non-prefixed overloaded function, as placeholder
> > +      if the user namespace does not need to be preserved.  */
> > +      char *noprefix_overload_name = get_name (instance, false, true);
> > +      attrs = get_attributes (instance);
> > +      add_function (instance, noprefix_overload_name, fntype, attrs,
> > +                 requires_float, false, preserve_user_namespace ||
> > placeholder_p);
> >      }
> >
> >    obstack_free (&m_string_obstack, name);
> > @@ -948,15 +951,15 @@ function_builder::add_overloaded_function (const
> > function_instance &instance,
> >       = add_function (instance, name, m_overload_type, NULL_TREE,
> >                       requires_float, true, m_direct_overloads);
> >        m_overload_names.put (name, &rfn);
> > -      if (!preserve_user_namespace)
> > -     {
> > -       char *noprefix_name = get_name (instance, false, true);
> > -       registered_function &noprefix_rfn
> > -         = add_function (instance, noprefix_name, m_overload_type,
> > -                         NULL_TREE, requires_float, true,
> > -                         m_direct_overloads);
> > -       m_overload_names.put (noprefix_name, &noprefix_rfn);
> > -     }
> > +
> > +      /* Also add the non-prefixed function, as placeholder if the
> > +      user namespace does not need to be preserved.  */
> > +      char *noprefix_name = get_name (instance, false, true);
> > +      registered_function &noprefix_rfn
> > +     = add_function (instance, noprefix_name, m_overload_type,
> > +                     NULL_TREE, requires_float, true,
> > +                     preserve_user_namespace || m_direct_overloads);
> > +      m_overload_names.put (noprefix_name, &noprefix_rfn);
> >      }
> >  }
> >
> > diff --git a/gcc/config/arm/arm-protos.h b/gcc/config/arm/arm-protos.h
> > index 7d73c66a15d..6186921011e 100644
> > --- a/gcc/config/arm/arm-protos.h
> > +++ b/gcc/config/arm/arm-protos.h
> > @@ -232,6 +232,7 @@ const unsigned int ARM_BUILTIN_CLASS = (1 <<
> > ARM_BUILTIN_SHIFT) - 1;
> >  namespace arm_mve {
> >    void handle_arm_mve_types_h ();
> >    void handle_arm_mve_h (bool);
> > +  tree builtin_decl (unsigned);
> >    tree resolve_overloaded_builtin (location_t, unsigned int,
> >                                  vec<tree, va_gc> *);
> >    bool check_builtin_call (location_t, vec<location_t>, unsigned int,
> > diff --git a/gcc/testsuite/gcc.target/arm/pr110268-1.c
> > b/gcc/testsuite/gcc.target/arm/pr110268-1.c
> > new file mode 100644
> > index 00000000000..d133011343c
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/arm/pr110268-1.c
> > @@ -0,0 +1,11 @@
> > +/* { dg-do link }  */
> > +/* { dg-require-effective-target arm_v8_1m_mve_ok } */
> > +/* { dg-add-options arm_v8_1m_mve } */
> > +/* { dg-additional-options "-O2 -flto -specs=rdimon.specs" } */
>
> IIRC rdimon.specs is a thing coming out of libgloss/newlib so may not be
> valid for some other libcs?
>

Ha right. In my mind MVE implies arm-eabi target, which implies newlib so I
overlooked this.


> Would it make sense to also require a newlib effective target?
>
I didn't know about this one ;-)

Revisiting this, I realized that -specs=rdimon.specs does not belong here,
in fact I updated my dejagnu setup to add this (required) option
automagically behind the scenes (via set_board_info ldflags), thus removing
it from dg-additional-options.

In fact there's no reason to require newlib, we just need to be able to
link. In principle we could have linux+glibc on an MVE target.
Unfortunately, it seems there's no such effective-target yet, the closest
being arm_arch_v8_1m_main_multilib.
However this one also implies being able to execute the resulting binary,
which may not be possible (actually it isn't in my current setup due to a
bug/feature in qemu).

So I hacked additions to target-supports.exp, to add:
check_effective_target_arm_arch_FUNC_link
check_effective_target_FUNC_link
to our existing list, and they call "check_no_compiler_messages ...
executable" instead of check_runtime.

And this means the testcases now have:
/* { dg-require-effective-target arm_arch_v8_1m_main_link } */ /* Make sure
we have suitable multilibs to link successfully.  */
/* { dg-require-effective-target arm_v8_1m_mve_ok } */

Ideally we should perhaps also increase the list of effective targets that
cover MVE, such that we also have "_multilib" etc but that seems a bit
overkill?
OTOH, the discrepancy above might be confusing (arm_arch_v8_1m_main
vs arm_v8_1m_mve).

So.... are you OK with a separate patch to
add check_effective_target_arm_arch_FUNC_link
and check_effective_target_FUNC_link and this one updated with
the dg-require-effective-target as above?

Too bad things are so complicated with the many flavours we have ;-)

Thanks,

Christophe

Ok otherwise.
> Thanks,
> Kyrill
>
> > +
> > +#include <arm_mve.h>
> > +
> > +int main(int argc, char* argv[])
> > +{
> > +  return vaddvq(__arm_vdupq_n_s8 (argc));
> > +}
> > diff --git a/gcc/testsuite/gcc.target/arm/pr110268-2.c
> > b/gcc/testsuite/gcc.target/arm/pr110268-2.c
> > new file mode 100644
> > index 00000000000..ad03fb37793
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/arm/pr110268-2.c
> > @@ -0,0 +1,22 @@
> > +/* { dg-do link }  */
> > +/* { dg-require-effective-target arm_v8_1m_mve_ok } */
> > +/* { dg-add-options arm_v8_1m_mve } */
> > +/* { dg-additional-options "-O2 -flto -specs=rdimon.specs" } */
> > +
> > +/* Check MVE intrinsics with LTO with
> > __ARM_MVE_PRESERVE_USER_NAMESPACE and a
> > +   user-overridden intrinsic.  */
> > +
> > +#define __ARM_MVE_PRESERVE_USER_NAMESPACE
> > +#include <arm_mve.h>
> > +
> > +int global_int;
> > +int32_t vaddvq(int8x16_t x)
> > +{
> > +  return global_int + __arm_vgetq_lane_s8 (x, 0);
> > +}
> > +
> > +int main(int argc, char* argv[])
> > +{
> > +  global_int = argc;
> > +  return vaddvq(__arm_vdupq_n_s8 (argc));
> > +}
> > --
> > 2.34.1
>
>