Re: [TCWG CI] Regression caused by gcc: handle retslot in modref

[Maxim Kuvyrkov <maxim.kuvyrkov@linaro.org>]

Hi Jan,

Indeed, this seems to be fixed.  Yesterday’s build scored “1” (built binutils only), but build I triggered just now scored “2” (built binutils and bootstrap_lto) [*] .

Thanks!

[*] https://ci.linaro.org/job/tcwg_gcc_bootstrap-build-master-aarch64-bootstrap_lto/

--
Maxim Kuvyrkov
https://www.linaro.org

> On 3 Nov 2021, at 18:12, Jan Hubicka <hubicka@kam.mff.cuni.cz> wrote:
> 
>> Hi Jan,
>> 
>> Just wanted to make sure this is on your radar.
>> 
>> Your "handle retslot in modref” patch broken LTO bootstrap on aarch64 in all configurations that we test:
>> 
>> - tcwg_gcc_bootstrap/master-aarch64-bootstrap_lto
>> - tcwg_gcc_bootstrap/master-aarch64-bootstrap_profiled_lto
>> - tcwg_gcc_bootstrap/master-aarch64-bootstrap_profiled_lto_lean
> Maxim,
> I am trying lto build on other machine and lets see.  I wonder - I have
> fixed a wrong code bug related to that patch today (causing
> misopitmization of omnetpp). Is there any chance that this bug got
> fixed?
> 
> Honza
>> 
>> Regards,
>> 
>> --
>> Maxim Kuvyrkov
>> https://www.linaro.org
>> 
>>> On 30 Oct 2021, at 05:32, ci_notify@linaro.org wrote:
>>> 
>>> [TCWG CI] Regression caused by gcc: handle retslot in modref:
>>> commit b8ef019ab938471f7f877a1eee3a6374fd8a6ae9
>>> Author: Jan Hubicka <hubicka@ucw.cz>
>>> 
>>>   handle retslot in modref
>>> 
>>> Results regressed to
>>> # reset_artifacts:
>>> -10
>>> # true:
>>> 0
>>> # build_abe binutils:
>>> 1
>>> # First few build errors in logs:
>>> # 00:04:02 make[3]: [Makefile:1772: aarch64-unknown-linux-gnu/bits/largefile-config.h] Error 1 (ignored)
>>> # 00:04:02 make[3]: [Makefile:1773: aarch64-unknown-linux-gnu/bits/largefile-config.h] Error 1 (ignored)
>>> # 00:19:19 checking for suffix of object files... configure: error: in `/home/tcwg-buildslave/workspace/tcwg_gnu_13/abe/builds/aarch64-unknown-linux-gnu/aarch64-unknown-linux-gnu/gcc-gcc.git~master-stage2/aarch64-unknown-linux-gnu/libgcc':
>>> # 00:19:19 configure: error: cannot compute suffix of object files: cannot compile
>>> # 00:19:19 make[2]: *** [Makefile:19796: configure-stage2-target-libgcc] Error 1
>>> # 00:19:19 make[1]: *** [Makefile:25351: stage2-bubble] Error 2
>>> # 00:19:19 make: *** [Makefile:1019: all] Error 2
>>> 
>>> from
>>> # reset_artifacts:
>>> -10
>>> # true:
>>> 0
>>> # build_abe binutils:
>>> 1
>>> # build_abe bootstrap_lto:
>>> 2
>>> 
>>> THIS IS THE END OF INTERESTING STUFF.  BELOW ARE LINKS TO BUILDS, REPRODUCTION INSTRUCTIONS, AND THE RAW COMMIT.
>>> 
>>> This commit has regressed these CI configurations:
>>> - tcwg_gcc_bootstrap/master-aarch64-bootstrap_lto
>>> 
>>> First_bad build: https://ci.linaro.org/job/tcwg_gcc_bootstrap-bisect-master-aarch64-bootstrap_lto/3/artifact/artifacts/build-b8ef019ab938471f7f877a1eee3a6374fd8a6ae9/
>>> Last_good build: https://ci.linaro.org/job/tcwg_gcc_bootstrap-bisect-master-aarch64-bootstrap_lto/3/artifact/artifacts/build-4045d5fa42f2ee7b284977c8f2f0edc300a63e43/
>>> Baseline build: https://ci.linaro.org/job/tcwg_gcc_bootstrap-bisect-master-aarch64-bootstrap_lto/3/artifact/artifacts/build-baseline/
>>> Even more details: https://ci.linaro.org/job/tcwg_gcc_bootstrap-bisect-master-aarch64-bootstrap_lto/3/artifact/artifacts/
>>> 
>>> Reproduce builds:
>>> <cut>
>>> mkdir investigate-gcc-b8ef019ab938471f7f877a1eee3a6374fd8a6ae9
>>> cd investigate-gcc-b8ef019ab938471f7f877a1eee3a6374fd8a6ae9
>>> 
>>> # Fetch scripts
>>> git clone https://git.linaro.org/toolchain/jenkins-scripts
>>> 
>>> # Fetch manifests and test.sh script
>>> mkdir -p artifacts/manifests
>>> curl -o artifacts/manifests/build-baseline.sh https://ci.linaro.org/job/tcwg_gcc_bootstrap-bisect-master-aarch64-bootstrap_lto/3/artifact/artifacts/manifests/build-baseline.sh --fail
>>> curl -o artifacts/manifests/build-parameters.sh https://ci.linaro.org/job/tcwg_gcc_bootstrap-bisect-master-aarch64-bootstrap_lto/3/artifact/artifacts/manifests/build-parameters.sh --fail
>>> curl -o artifacts/test.sh https://ci.linaro.org/job/tcwg_gcc_bootstrap-bisect-master-aarch64-bootstrap_lto/3/artifact/artifacts/test.sh --fail
>>> chmod +x artifacts/test.sh
>>> 
>>> # Reproduce the baseline build (build all pre-requisites)
>>> ./jenkins-scripts/tcwg_gnu-build.sh @@ artifacts/manifests/build-baseline.sh
>>> 
>>> # Save baseline build state (which is then restored in artifacts/test.sh)
>>> mkdir -p ./bisect
>>> rsync -a --del --delete-excluded --exclude /bisect/ --exclude /artifacts/ --exclude /gcc/ ./ ./bisect/baseline/
>>> 
>>> cd gcc
>>> 
>>> # Reproduce first_bad build
>>> git checkout --detach b8ef019ab938471f7f877a1eee3a6374fd8a6ae9
>>> ../artifacts/test.sh
>>> 
>>> # Reproduce last_good build
>>> git checkout --detach 4045d5fa42f2ee7b284977c8f2f0edc300a63e43
>>> ../artifacts/test.sh
>>> 
>>> cd ..
>>> </cut>
>>> 
>>> Full commit (up to 1000 lines):
>>> <cut>
>>> commit b8ef019ab938471f7f877a1eee3a6374fd8a6ae9
>>> Author: Jan Hubicka <hubicka@ucw.cz>
>>> Date:   Fri Oct 29 16:01:51 2021 +0200
>>> 
>>>   handle retslot in modref
>>> 
>>>   Extend modref and tree-ssa-structalias to handle retslot flags.
>>>   Since retslot it essentially a hidden argument that is known to be write-only
>>>   we can do pretty much the same stuff as we do for regular parameters.
>>>   I plan to add static chain handling similar way.
>>> 
>>>   We do not handle IPA propagation of retslot flags (where return slot is
>>>   initialized via return slot of other function). For this ipa-prop needs
>>>   to be extended to understand retslot as well.
>>> 
>>>   Bootstrapped/regtested x86_64-linux, OK for the gimple bits?
>>> 
>>>   Honza
>>> 
>>>   gcc/ChangeLog:
>>> 
>>>           * gimple.c (gimple_call_retslot_flags): New function.
>>>           * gimple.h (gimple_call_retslot_flags): Declare.
>>>           * ipa-modref.c: Include tree-cfg.h.
>>>           (struct escape_entry): Turn parm_index to signed.
>>>           (modref_summary_lto::modref_summary_lto): Add retslot_flags.
>>>           (modref_summary::modref_summary): Initialize retslot_flags.
>>>           (struct modref_summary_lto): Likewise.
>>>           (modref_summary::useful_p): Check retslot_flags.
>>>           (modref_summary_lto::useful_p): Likewise.
>>>           (modref_summary::dump): Dump retslot_flags.
>>>           (modref_summary_lto::dump): Likewise.
>>>           (struct escape_point): Add hidden_args enum.
>>>           (analyze_ssa_name_flags): Ignore return slot return;
>>>           use gimple_call_retslot_flags.
>>>           (record_escape_points): Break out from ...
>>>           (analyze_parms): ... here; handle retslot_flags.
>>>           (modref_summaries::duplicate): Duplicate retslot_flags.
>>>           (modref_summaries_lto::duplicate): Likewise.
>>>           (modref_write_escape_summary): Stream parm_index as signed.
>>>           (modref_read_escape_summary): Likewise.
>>>           (modref_write): Stream retslot_flags.
>>>           (read_section): Likewise.
>>>           (struct escape_map): Fix typo in comment.
>>>           (update_escape_summary_1): Fix whitespace.
>>>           (ipa_merge_modref_summary_after_inlining): Drop retslot_flags.
>>>           (modref_merge_call_site_flags): Merge retslot_flags.
>>>           * ipa-modref.h (struct modref_summary): Add retslot_flags.
>>>           * tree-ssa-structalias.c (handle_rhs_call): Handle retslot_flags.
>>> ---
>>> gcc/gimple.c               |  41 ++++++++++++-
>>> gcc/gimple.h               |   1 +
>>> gcc/ipa-modref.c           | 145 +++++++++++++++++++++++++++++++++------------
>>> gcc/ipa-modref.h           |   1 +
>>> gcc/tree-ssa-structalias.c |  29 ++++++---
>>> 5 files changed, 170 insertions(+), 47 deletions(-)
>>> 
>>> diff --git a/gcc/gimple.c b/gcc/gimple.c
>>> index cc7a88e822b..22dd6417d19 100644
>>> --- a/gcc/gimple.c
>>> +++ b/gcc/gimple.c
>>> @@ -1597,7 +1597,12 @@ gimple_call_arg_flags (const gcall *stmt, unsigned arg)
>>> 	  if (!node->binds_to_current_def_p ())
>>> 	    {
>>> 	      if ((modref_flags & EAF_UNUSED) && !(flags & EAF_UNUSED))
>>> -		modref_flags &= ~EAF_UNUSED;
>>> +		{
>>> +		  modref_flags &= ~EAF_UNUSED;
>>> +		  modref_flags |= EAF_NOESCAPE;
>>> +		}
>>> +	      if ((modref_flags & EAF_NOREAD) && !(flags & EAF_NOREAD))
>>> +		modref_flags &= ~EAF_NOREAD;
>>> 	      if ((modref_flags & EAF_DIRECT) && !(flags & EAF_DIRECT))
>>> 		modref_flags &= ~EAF_DIRECT;
>>> 	    }
>>> @@ -1608,6 +1613,40 @@ gimple_call_arg_flags (const gcall *stmt, unsigned arg)
>>>  return flags;
>>> }
>>> 
>>> +/* Detects argument flags for return slot on call STMT.  */
>>> +
>>> +int
>>> +gimple_call_retslot_flags (const gcall *stmt)
>>> +{
>>> +  int flags = EAF_DIRECT | EAF_NOREAD;
>>> +
>>> +  tree callee = gimple_call_fndecl (stmt);
>>> +  if (callee)
>>> +    {
>>> +      cgraph_node *node = cgraph_node::get (callee);
>>> +      modref_summary *summary = node ? get_modref_function_summary (node)
>>> +				: NULL;
>>> +
>>> +      if (summary)
>>> +	{
>>> +	  int modref_flags = summary->retslot_flags;
>>> +
>>> +	  /* We have possibly optimized out load.  Be conservative here.  */
>>> +	  if (!node->binds_to_current_def_p ())
>>> +	    {
>>> +	      if ((modref_flags & EAF_UNUSED) && !(flags & EAF_UNUSED))
>>> +		{
>>> +		  modref_flags &= ~EAF_UNUSED;
>>> +		  modref_flags |= EAF_NOESCAPE;
>>> +		}
>>> +	    }
>>> +	  if (dbg_cnt (ipa_mod_ref_pta))
>>> +	    flags |= modref_flags;
>>> +	}
>>> +    }
>>> +  return flags;
>>> +}
>>> +
>>> /* Detects return flags for the call STMT.  */
>>> 
>>> int
>>> diff --git a/gcc/gimple.h b/gcc/gimple.h
>>> index 303623b3ced..23a124ec769 100644
>>> --- a/gcc/gimple.h
>>> +++ b/gcc/gimple.h
>>> @@ -1589,6 +1589,7 @@ gimple_seq gimple_seq_copy (gimple_seq);
>>> bool gimple_call_same_target_p (const gimple *, const gimple *);
>>> int gimple_call_flags (const gimple *);
>>> int gimple_call_arg_flags (const gcall *, unsigned);
>>> +int gimple_call_retslot_flags (const gcall *);
>>> int gimple_call_return_flags (const gcall *);
>>> bool gimple_call_nonnull_result_p (gcall *);
>>> tree gimple_call_nonnull_arg (gcall *);
>>> diff --git a/gcc/ipa-modref.c b/gcc/ipa-modref.c
>>> index 0bbec8df0a2..3539cb43d11 100644
>>> --- a/gcc/ipa-modref.c
>>> +++ b/gcc/ipa-modref.c
>>> @@ -86,6 +86,7 @@ along with GCC; see the file COPYING3.  If not see
>>> #include "stringpool.h"
>>> #include "tree-ssanames.h"
>>> #include "attribs.h"
>>> +#include "tree-cfg.h"
>>> 
>>> 
>>> namespace {
>>> @@ -133,7 +134,7 @@ static fnspec_summaries_t *fnspec_summaries = NULL;
>>> struct escape_entry
>>> {
>>>  /* Parameter that escapes at a given call.  */
>>> -  unsigned int parm_index;
>>> +  int parm_index;
>>>  /* Argument it escapes to.  */
>>>  unsigned int arg;
>>>  /* Minimal flags known about the argument.  */
>>> @@ -269,7 +270,7 @@ static GTY(()) fast_function_summary <modref_summary_lto *, va_gc>
>>> /* Summary for a single function which this pass produces.  */
>>> 
>>> modref_summary::modref_summary ()
>>> -  : loads (NULL), stores (NULL), writes_errno (NULL)
>>> +  : loads (NULL), stores (NULL), retslot_flags (0), writes_errno (false)
>>> {
>>> }
>>> 
>>> @@ -322,6 +323,8 @@ modref_summary::useful_p (int ecf_flags, bool check_flags)
>>>  if (check_flags && eaf_flags_useful_p (arg_flags, ecf_flags))
>>>    return true;
>>>  arg_flags.release ();
>>> +  if (check_flags && remove_useless_eaf_flags (retslot_flags, ecf_flags, false))
>>> +    return true;
>>>  if (ecf_flags & ECF_CONST)
>>>    return false;
>>>  if (loads && !loads->every_base)
>>> @@ -363,6 +366,7 @@ struct GTY(()) modref_summary_lto
>>>  modref_records_lto *loads;
>>>  modref_records_lto *stores;
>>>  auto_vec<eaf_flags_t> GTY((skip)) arg_flags;
>>> +  eaf_flags_t retslot_flags;
>>>  bool writes_errno;
>>> 
>>>  modref_summary_lto ();
>>> @@ -374,7 +378,7 @@ struct GTY(()) modref_summary_lto
>>> /* Summary for a single function which this pass produces.  */
>>> 
>>> modref_summary_lto::modref_summary_lto ()
>>> -  : loads (NULL), stores (NULL), writes_errno (NULL)
>>> +  : loads (NULL), stores (NULL), retslot_flags (0), writes_errno (false)
>>> {
>>> }
>>> 
>>> @@ -400,6 +404,8 @@ modref_summary_lto::useful_p (int ecf_flags, bool check_flags)
>>>  if (check_flags && eaf_flags_useful_p (arg_flags, ecf_flags))
>>>    return true;
>>>  arg_flags.release ();
>>> +  if (check_flags && remove_useless_eaf_flags (retslot_flags, ecf_flags, false))
>>> +    return true;
>>>  if (ecf_flags & ECF_CONST)
>>>    return false;
>>>  if (loads && !loads->every_base)
>>> @@ -608,6 +614,11 @@ modref_summary::dump (FILE *out)
>>> 	    dump_eaf_flags (out, arg_flags[i]);
>>> 	  }
>>>    }
>>> +  if (retslot_flags)
>>> +    {
>>> +      fprintf (out, "  Retslot flags:");
>>> +      dump_eaf_flags (out, retslot_flags);
>>> +    }
>>> }
>>> 
>>> /* Dump summary.  */
>>> @@ -630,6 +641,11 @@ modref_summary_lto::dump (FILE *out)
>>> 	    dump_eaf_flags (out, arg_flags[i]);
>>> 	  }
>>>    }
>>> +  if (retslot_flags)
>>> +    {
>>> +      fprintf (out, "  Retslot flags:");
>>> +      dump_eaf_flags (out, retslot_flags);
>>> +    }
>>> }
>>> 
>>> /* Get function summary for FUNC if it exists, return NULL otherwise.  */
>>> @@ -1396,6 +1412,11 @@ namespace {
>>> 
>>> struct escape_point
>>> {
>>> +  /* Extra hidden args we keep track of.  */
>>> +  enum hidden_args
>>> +  {
>>> +    retslot_arg = -1
>>> +  };
>>>  /* Value escapes to this call.  */
>>>  gcall *call;
>>>  /* Argument it escapes to.  */
>>> @@ -1705,7 +1726,11 @@ analyze_ssa_name_flags (tree name, vec<modref_lattice> &lattice, int depth,
>>> 	 Returning name counts as an use by tree-ssa-structalias.c  */
>>>      if (greturn *ret = dyn_cast <greturn *> (use_stmt))
>>> 	{
>>> -	  if (gimple_return_retval (ret) == name)
>>> +	  /* Returning through return slot is seen as memory write earlier.  */
>>> +	  if (DECL_RESULT (current_function_decl)
>>> +	      && DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
>>> +	    ;
>>> +	  else if (gimple_return_retval (ret) == name)
>>> 	    lattice[index].merge (~(EAF_UNUSED | EAF_NOT_RETURNED));
>>> 	  else if (memory_access_to (gimple_return_retval (ret), name))
>>> 	    {
>>> @@ -1748,7 +1773,7 @@ analyze_ssa_name_flags (tree name, vec<modref_lattice> &lattice, int depth,
>>> 		     may make LHS to escape.  See PR 98499.  */
>>> 		  if (gimple_call_return_slot_opt_p (call)
>>> 		      && TREE_ADDRESSABLE (TREE_TYPE (gimple_call_lhs (call))))
>>> -		    lattice[index].merge (EAF_NOREAD | EAF_DIRECT);
>>> +		    lattice[index].merge (gimple_call_retslot_flags (call));
>>> 		}
>>> 
>>> 	      /* We do not track accesses to the static chain (we could)
>>> @@ -1777,7 +1802,7 @@ analyze_ssa_name_flags (tree name, vec<modref_lattice> &lattice, int depth,
>>> 			  lattice[index].merge (call_flags);
>>> 			else
>>> 			  lattice[index].add_escape_point (call, i,
>>> -			     				   call_flags, true);
>>> +							   call_flags, true);
>>> 		      }
>>> 		    if (!ignore_retval)
>>> 		      merge_call_lhs_flags (call, i, index, false,
>>> @@ -1912,6 +1937,29 @@ analyze_ssa_name_flags (tree name, vec<modref_lattice> &lattice, int depth,
>>>  lattice[index].known = true;
>>> }
>>> 
>>> +/* Record escape points of PARM_INDEX according to LATTICE.  */
>>> +
>>> +static void
>>> +record_escape_points (modref_lattice &lattice, int parm_index, int flags)
>>> +{
>>> +  if (lattice.escape_points.length ())
>>> +    {
>>> +      escape_point *ep;
>>> +      unsigned int ip;
>>> +      cgraph_node *node = cgraph_node::get (current_function_decl);
>>> +
>>> +      FOR_EACH_VEC_ELT (lattice.escape_points, ip, ep)
>>> +	if ((ep->min_flags & flags) != flags)
>>> +	  {
>>> +	    cgraph_edge *e = node->get_edge (ep->call);
>>> +	    struct escape_entry ee = {parm_index, ep->arg,
>>> +				      ep->min_flags, ep->direct};
>>> +
>>> +	    escape_summaries->get_create (e)->esc.safe_push (ee);
>>> +	  }
>>> +    }
>>> +}
>>> +
>>> /* Determine EAF flags for function parameters.  */
>>> 
>>> static void
>>> @@ -1921,16 +1969,22 @@ analyze_parms (modref_summary *summary, modref_summary_lto *summary_lto,
>>>  unsigned int parm_index = 0;
>>>  unsigned int count = 0;
>>>  int ecf_flags = flags_from_decl_or_type (current_function_decl);
>>> +  tree retslot = NULL;
>>> 
>>>  /* For novops functions we have nothing to gain by EAF flags.  */
>>>  if (ecf_flags & ECF_NOVOPS)
>>>    return;
>>> 
>>> +  /* If there is return slot, look up its SSA name.  */
>>> +  if (DECL_RESULT (current_function_decl)
>>> +      && DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
>>> +    retslot = ssa_default_def (cfun, DECL_RESULT (current_function_decl));
>>> +
>>>  for (tree parm = DECL_ARGUMENTS (current_function_decl); parm;
>>>       parm = TREE_CHAIN (parm))
>>>    count++;
>>> 
>>> -  if (!count)
>>> +  if (!count && !retslot)
>>>    return;
>>> 
>>>  auto_vec<modref_lattice> lattice;
>>> @@ -1984,24 +2038,24 @@ analyze_parms (modref_summary *summary, modref_summary_lto *summary_lto,
>>> 		summary_lto->arg_flags.safe_grow_cleared (count, true);
>>> 	      summary_lto->arg_flags[parm_index] = flags;
>>> 	    }
>>> -	  if (lattice[SSA_NAME_VERSION (name)].escape_points.length ())
>>> -	    {
>>> -	      escape_point *ep;
>>> -	      unsigned int ip;
>>> -	      cgraph_node *node = cgraph_node::get (current_function_decl);
>>> -
>>> -	      gcc_checking_assert (ipa);
>>> -	      FOR_EACH_VEC_ELT
>>> -		  (lattice[SSA_NAME_VERSION (name)].escape_points, ip, ep)
>>> -		if ((ep->min_flags & flags) != flags)
>>> -		  {
>>> -		    cgraph_edge *e = node->get_edge (ep->call);
>>> -		    struct escape_entry ee = {parm_index, ep->arg,
>>> -		     			      ep->min_flags, ep->direct};
>>> +	  record_escape_points (lattice[SSA_NAME_VERSION (name)],
>>> +				parm_index, flags);
>>> +	}
>>> +    }
>>> +  if (retslot)
>>> +    {
>>> +      analyze_ssa_name_flags (retslot, lattice, 0, ipa);
>>> +      int flags = lattice[SSA_NAME_VERSION (retslot)].flags;
>>> 
>>> -		    escape_summaries->get_create (e)->esc.safe_push (ee);
>>> -		  }
>>> -	    }
>>> +      flags = remove_useless_eaf_flags (flags, ecf_flags, false);
>>> +      if (flags)
>>> +	{
>>> +	  if (summary)
>>> +	    summary->retslot_flags = flags;
>>> +	  if (summary_lto)
>>> +	    summary_lto->retslot_flags = flags;
>>> +	  record_escape_points (lattice[SSA_NAME_VERSION (retslot)],
>>> +				escape_point::retslot_arg, flags);
>>> 	}
>>>    }
>>>  if (ipa)
>>> @@ -2287,6 +2341,7 @@ modref_summaries::duplicate (cgraph_node *, cgraph_node *dst,
>>>  dst_data->writes_errno = src_data->writes_errno;
>>>  if (src_data->arg_flags.length ())
>>>    dst_data->arg_flags = src_data->arg_flags.copy ();
>>> +  dst_data->retslot_flags = src_data->retslot_flags;
>>> }
>>> 
>>> /* Called when new clone is inserted to callgraph late.  */
>>> @@ -2312,6 +2367,7 @@ modref_summaries_lto::duplicate (cgraph_node *, cgraph_node *,
>>>  dst_data->writes_errno = src_data->writes_errno;
>>>  if (src_data->arg_flags.length ())
>>>    dst_data->arg_flags = src_data->arg_flags.copy ();
>>> +  dst_data->retslot_flags = src_data->retslot_flags;
>>> }
>>> 
>>> namespace
>>> @@ -2551,7 +2607,7 @@ modref_write_escape_summary (struct bitpack_d *bp, escape_summary *esum)
>>>  escape_entry *ee;
>>>  FOR_EACH_VEC_ELT (esum->esc, i, ee)
>>>    {
>>> -      bp_pack_var_len_unsigned (bp, ee->parm_index);
>>> +      bp_pack_var_len_int (bp, ee->parm_index);
>>>      bp_pack_var_len_unsigned (bp, ee->arg);
>>>      bp_pack_var_len_unsigned (bp, ee->min_flags);
>>>      bp_pack_value (bp, ee->direct, 1);
>>> @@ -2571,7 +2627,7 @@ modref_read_escape_summary (struct bitpack_d *bp, cgraph_edge *e)
>>>  for (unsigned int i = 0; i < n; i++)
>>>    {
>>>      escape_entry ee;
>>> -      ee.parm_index = bp_unpack_var_len_unsigned (bp);
>>> +      ee.parm_index = bp_unpack_var_len_int (bp);
>>>      ee.arg = bp_unpack_var_len_unsigned (bp);
>>>      ee.min_flags = bp_unpack_var_len_unsigned (bp);
>>>      ee.direct = bp_unpack_value (bp, 1);
>>> @@ -2628,6 +2684,7 @@ modref_write ()
>>> 	  streamer_write_uhwi (ob, r->arg_flags.length ());
>>> 	  for (unsigned int i = 0; i < r->arg_flags.length (); i++)
>>> 	    streamer_write_uhwi (ob, r->arg_flags[i]);
>>> +	  streamer_write_uhwi (ob, r->retslot_flags);
>>> 
>>> 	  write_modref_records (r->loads, ob);
>>> 	  write_modref_records (r->stores, ob);
>>> @@ -2724,6 +2781,11 @@ read_section (struct lto_file_decl_data *file_data, const char *data,
>>> 	  if (modref_sum_lto)
>>> 	    modref_sum_lto->arg_flags.quick_push (flags);
>>> 	}
>>> +      eaf_flags_t flags = streamer_read_uhwi (&ib);
>>> +      if (modref_sum)
>>> +	modref_sum->retslot_flags = flags;
>>> +      if (modref_sum_lto)
>>> +	modref_sum_lto->retslot_flags = flags;
>>>      read_modref_records (&ib, data_in,
>>> 			   modref_sum ? &modref_sum->loads : NULL,
>>> 			   modref_sum_lto ? &modref_sum_lto->loads : NULL);
>>> @@ -3098,7 +3160,7 @@ struct escape_map
>>>  bool direct;
>>> };
>>> 
>>> -/* Update escape map fo E.  */
>>> +/* Update escape map for E.  */
>>> 
>>> static void
>>> update_escape_summary_1 (cgraph_edge *e,
>>> @@ -3117,7 +3179,10 @@ update_escape_summary_1 (cgraph_edge *e,
>>>    {
>>>      unsigned int j;
>>>      struct escape_map *em;
>>> -      if (ee->parm_index >= map.length ())
>>> +      /* TODO: We do not have jump functions for return slots, so we
>>> +	 never propagate them to outer function.  */
>>> +      if (ee->parm_index >= (int)map.length ()
>>> +	  || ee->parm_index < 0)
>>> 	continue;
>>>      FOR_EACH_VEC_ELT (map[ee->parm_index], j, em)
>>> 	{
>>> @@ -3125,7 +3190,7 @@ update_escape_summary_1 (cgraph_edge *e,
>>> 	  if (ee->direct && !em->direct)
>>> 	    min_flags = deref_flags (min_flags, ignore_stores);
>>> 	  struct escape_entry entry = {em->parm_index, ee->arg,
>>> -	    			       ee->min_flags,
>>> +				       ee->min_flags,
>>> 				       ee->direct & em->direct};
>>> 	  sum->esc.safe_push (entry);
>>> 	}
>>> @@ -3245,7 +3310,11 @@ ipa_merge_modref_summary_after_inlining (cgraph_edge *edge)
>>>    FOR_EACH_VEC_ELT (sum->esc, i, ee)
>>>      {
>>> 	bool needed = false;
>>> -	if (to_info && to_info->arg_flags.length () > ee->parm_index)
>>> +	/* TODO: We do not have jump functions for return slots, so we
>>> +	   never propagate them to outer function.  */
>>> +	if (ee->parm_index < 0)
>>> +	  continue;
>>> +	if (to_info && (int)to_info->arg_flags.length () > ee->parm_index)
>>> 	  {
>>> 	    int flags = callee_info
>>> 			&& callee_info->arg_flags.length () > ee->arg
>>> @@ -3259,7 +3328,7 @@ ipa_merge_modref_summary_after_inlining (cgraph_edge *edge)
>>> 	    if (to_info->arg_flags[ee->parm_index])
>>> 	      needed = true;
>>> 	  }
>>> -	if (to_info_lto && to_info_lto->arg_flags.length () > ee->parm_index)
>>> +	if (to_info_lto && (int)to_info_lto->arg_flags.length () > ee->parm_index)
>>> 	  {
>>> 	    int flags = callee_info_lto
>>> 			&& callee_info_lto->arg_flags.length () > ee->arg
>>> @@ -3798,29 +3867,31 @@ modref_merge_call_site_flags (escape_summary *sum,
>>>      if (flags_lto & EAF_NOESCAPE)
>>> 	flags_lto |= EAF_NODIRECTESCAPE;
>>>      if (!(flags & EAF_UNUSED)
>>> -	  && cur_summary && ee->parm_index < cur_summary->arg_flags.length ())
>>> +	  && cur_summary && ee->parm_index < (int)cur_summary->arg_flags.length ())
>>> 	{
>>> -	  int f = cur_summary->arg_flags[ee->parm_index];
>>> +	  eaf_flags_t &f = ee->parm_index == escape_point::retslot_arg
>>> +			   ? cur_summary->retslot_flags
>>> +			   : cur_summary->arg_flags[ee->parm_index];
>>> 	  if ((f & flags) != f)
>>> 	    {
>>> 	      f = remove_useless_eaf_flags
>>> 			 (f & flags, ecf_flags,
>>> 			  VOID_TYPE_P (TREE_TYPE (TREE_TYPE (caller))));
>>> -	      cur_summary->arg_flags[ee->parm_index] = f;
>>> 	      changed = true;
>>> 	    }
>>> 	}
>>>      if (!(flags_lto & EAF_UNUSED)
>>> 	  && cur_summary_lto
>>> -	  && ee->parm_index < cur_summary_lto->arg_flags.length ())
>>> +	  && ee->parm_index < (int)cur_summary_lto->arg_flags.length ())
>>> 	{
>>> -	  int f = cur_summary_lto->arg_flags[ee->parm_index];
>>> +	  eaf_flags_t &f = ee->parm_index == escape_point::retslot_arg
>>> +			   ? cur_summary_lto->retslot_flags
>>> +			   : cur_summary_lto->arg_flags[ee->parm_index];
>>> 	  if ((f & flags_lto) != f)
>>> 	    {
>>> 	      f = remove_useless_eaf_flags
>>> 			 (f & flags_lto, ecf_flags,
>>> 			  VOID_TYPE_P (TREE_TYPE (TREE_TYPE (caller))));
>>> -	      cur_summary_lto->arg_flags[ee->parm_index] = f;
>>> 	      changed = true;
>>> 	    }
>>> 	}
>>> diff --git a/gcc/ipa-modref.h b/gcc/ipa-modref.h
>>> index 5afa3aa439f..a4db27471eb 100644
>>> --- a/gcc/ipa-modref.h
>>> +++ b/gcc/ipa-modref.h
>>> @@ -31,6 +31,7 @@ struct GTY(()) modref_summary
>>>  modref_records *loads;
>>>  modref_records *stores;
>>>  auto_vec<eaf_flags_t> GTY((skip)) arg_flags;
>>> +  eaf_flags_t retslot_flags;
>>>  bool writes_errno;
>>> 
>>>  modref_summary ();
>>> diff --git a/gcc/tree-ssa-structalias.c b/gcc/tree-ssa-structalias.c
>>> index 35971a54e02..99072df0768 100644
>>> --- a/gcc/tree-ssa-structalias.c
>>> +++ b/gcc/tree-ssa-structalias.c
>>> @@ -4254,17 +4254,28 @@ handle_rhs_call (gcall *stmt, vec<ce_s> *results,
>>>      && gimple_call_lhs (stmt) != NULL_TREE
>>>      && TREE_ADDRESSABLE (TREE_TYPE (gimple_call_lhs (stmt))))
>>>    {
>>> -      auto_vec<ce_s> tmpc;
>>> -      struct constraint_expr *c;
>>> -      unsigned i;
>>> +      int flags = gimple_call_retslot_flags (stmt);
>>> +      if ((flags & (EAF_NOESCAPE | EAF_NOT_RETURNED))
>>> +	  != (EAF_NOESCAPE | EAF_NOT_RETURNED))
>>> +	{
>>> +	  auto_vec<ce_s> tmpc;
>>> 
>>> -      get_constraint_for_address_of (gimple_call_lhs (stmt), &tmpc);
>>> +	  get_constraint_for_address_of (gimple_call_lhs (stmt), &tmpc);
>>> 
>>> -      make_constraints_to (callescape->id, tmpc);
>>> -      if (writes_global_memory)
>>> -	make_constraints_to (escaped_id, tmpc);
>>> -      FOR_EACH_VEC_ELT (tmpc, i, c)
>>> -	results->safe_push (*c);
>>> +	  if (!(flags & (EAF_NOESCAPE | EAF_NODIRECTESCAPE)))
>>> +	    {
>>> +	      make_constraints_to (callescape->id, tmpc);
>>> +	      if (writes_global_memory)
>>> +		make_constraints_to (escaped_id, tmpc);
>>> +	    }
>>> +	  if (!(flags & EAF_NOT_RETURNED))
>>> +	    {
>>> +	      struct constraint_expr *c;
>>> +	      unsigned i;
>>> +	      FOR_EACH_VEC_ELT (tmpc, i, c)
>>> +		results->safe_push (*c);
>>> +	    }
>>> +	}
>>>    }
>>> }
>>> 
>>> </cut>
>>