public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] ipa/109607 - properly gimplify conversions introduced by IPA param manipulation
@ 2023-04-27  7:24 Richard Biener
  0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2023-04-27  7:24 UTC (permalink / raw)
  To: gcc-patches; +Cc: Martin Jambor

The following addresses IPA param manipulation (through IPA SRA)
replacing

  BIT_FIELD_REF <*this_8(D), 8, 56>

with

  BIT_FIELD_REF <VIEW_CONVERT_EXPR<const struct profile_count>(ISRA.814), 8, 56>

which is supposed to be invalid GIMPLE (ISRA.814 is a register).
There's currently insufficient checking in place to catch this in the
IL verifier but I am working on that as part of fixing PR109594.

The solution for the particular testcase I am running into this is
to split the conversion to a separate stmt.  Generally the modification
phase is set up for this but the extra_stmts sequence isn't passed
around everywhere.  The following passes it to modify_expression
from modify_assignment when rewriting the RHS.

Bootstrapped and tested on x86_64-unknown-linux-gnu.

OK for trunk?

Thanks,
Richard.

	PR ipa/109607
	* ipa-param-manipulation.h
	(ipa_param_body_adjustments::modify_expression): Add extra_stmts
	argument.
	* ipa-param-manipulation.cc
	(ipa_param_body_adjustments::modify_expression): Likewise.
	When we need a conversion and the replacement is a register
	split the conversion out.
	(ipa_param_body_adjustments::modify_assignment): Pass
	extra_stmts to RHS modify_expression.

	* g++.dg/torture/pr109607.C: New testcase.
---
 gcc/ipa-param-manipulation.cc           | 10 ++++++++--
 gcc/ipa-param-manipulation.h            |  2 +-
 gcc/testsuite/g++.dg/torture/pr109607.C | 13 +++++++++++++
 3 files changed, 22 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/torture/pr109607.C

diff --git a/gcc/ipa-param-manipulation.cc b/gcc/ipa-param-manipulation.cc
index 42488ee09c3..33dcab9c33c 100644
--- a/gcc/ipa-param-manipulation.cc
+++ b/gcc/ipa-param-manipulation.cc
@@ -1825,7 +1825,8 @@ ipa_param_body_adjustments::replace_removed_params_ssa_names (tree old_name,
    necessary conversions.  */
 
 bool
-ipa_param_body_adjustments::modify_expression (tree *expr_p, bool convert)
+ipa_param_body_adjustments::modify_expression (tree *expr_p, bool convert,
+					       gimple_seq *extra_stmts)
 {
   tree expr = *expr_p;
 
@@ -1860,6 +1861,11 @@ ipa_param_body_adjustments::modify_expression (tree *expr_p, bool convert)
       gcc_checking_assert (tree_to_shwi (TYPE_SIZE (TREE_TYPE (expr)))
 			   == tree_to_shwi (TYPE_SIZE (TREE_TYPE (repl))));
       tree vce = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (expr), repl);
+      if (is_gimple_reg (repl))
+	{
+	  gcc_assert (extra_stmts);
+	  vce = force_gimple_operand (vce, extra_stmts, true, NULL_TREE);
+	}
       *expr_p = vce;
     }
   else
@@ -1887,7 +1893,7 @@ ipa_param_body_adjustments::modify_assignment (gimple *stmt,
   lhs_p = gimple_assign_lhs_ptr (stmt);
 
   any = modify_expression (lhs_p, false);
-  any |= modify_expression (rhs_p, false);
+  any |= modify_expression (rhs_p, false, extra_stmts);
   if (any
       && !useless_type_conversion_p (TREE_TYPE (*lhs_p), TREE_TYPE (*rhs_p)))
     {
diff --git a/gcc/ipa-param-manipulation.h b/gcc/ipa-param-manipulation.h
index 9cc957ae7bb..9798eedf0b6 100644
--- a/gcc/ipa-param-manipulation.h
+++ b/gcc/ipa-param-manipulation.h
@@ -373,7 +373,7 @@ private:
 						    unsigned unit_offset);
   ipa_param_body_replacement *lookup_first_base_replacement (tree base);
   tree replace_removed_params_ssa_names (tree old_name, gimple *stmt);
-  bool modify_expression (tree *expr_p, bool convert);
+  bool modify_expression (tree *expr_p, bool convert, gimple_seq * = nullptr);
   bool modify_assignment (gimple *stmt, gimple_seq *extra_stmts);
   bool modify_call_stmt (gcall **stmt_p, gimple *orig_stmt);
   bool modify_cfun_body ();
diff --git a/gcc/testsuite/g++.dg/torture/pr109607.C b/gcc/testsuite/g++.dg/torture/pr109607.C
new file mode 100644
index 00000000000..f7e2a7457af
--- /dev/null
+++ b/gcc/testsuite/g++.dg/torture/pr109607.C
@@ -0,0 +1,13 @@
+// { dg-do compile }
+// { dg-require-effective-target lp64 }
+
+enum profile_quality { GUESSED_LOCAL };
+struct profile_count {
+  long m_val : 61;
+  profile_quality m_quality : 3;
+  bool verify() {
+    m_quality ? __builtin_abort (), 0 : 0;
+    return m_val == GUESSED_LOCAL;
+  }
+} count;
+void verify_flow_info() { count.verify(); }
-- 
2.35.3

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] ipa/109607 - properly gimplify conversions introduced by IPA param manipulation
       [not found] <notmuch-sha1-303b2e15d6e6379f358f76d79baab4438f9eddca>
@ 2023-04-27  9:49 ` Martin Jambor
  0 siblings, 0 replies; 2+ messages in thread
From: Martin Jambor @ 2023-04-27  9:49 UTC (permalink / raw)
  To: Richard Biener; +Cc: GCC Patches

Hi,

On Thu, Apr 27 2023, Richard Biener wrote:
> The following addresses IPA param manipulation (through IPA SRA)
> replacing
>
>   BIT_FIELD_REF <*this_8(D), 8, 56>
>
> with
>
>   BIT_FIELD_REF <VIEW_CONVERT_EXPR<const struct profile_count>(ISRA.814), 8, 56>
>
> which is supposed to be invalid GIMPLE (ISRA.814 is a register).
> There's currently insufficient checking in place to catch this in the
> IL verifier but I am working on that as part of fixing PR109594.
>
> The solution for the particular testcase I am running into this is
> to split the conversion to a separate stmt.  Generally the modification
> phase is set up for this but the extra_stmts sequence isn't passed
> around everywhere.  The following passes it to modify_expression
> from modify_assignment when rewriting the RHS.
>
> Bootstrapped and tested on x86_64-unknown-linux-gnu.
>
> OK for trunk?

Yes, thank you!

Martin


>
> Thanks,
> Richard.
>
> 	PR ipa/109607
> 	* ipa-param-manipulation.h
> 	(ipa_param_body_adjustments::modify_expression): Add extra_stmts
> 	argument.
> 	* ipa-param-manipulation.cc
> 	(ipa_param_body_adjustments::modify_expression): Likewise.
> 	When we need a conversion and the replacement is a register
> 	split the conversion out.
> 	(ipa_param_body_adjustments::modify_assignment): Pass
> 	extra_stmts to RHS modify_expression.
>
> 	* g++.dg/torture/pr109607.C: New testcase.

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2023-04-27  9:49 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-04-27  7:24 [PATCH] ipa/109607 - properly gimplify conversions introduced by IPA param manipulation Richard Biener
     [not found] <notmuch-sha1-303b2e15d6e6379f358f76d79baab4438f9eddca>
2023-04-27  9:49 ` Martin Jambor

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).