public inbox for
 help / color / mirror / Atom feed
* [COMMITTED] Clear NAN when reading back a global range if necessary.
@ 2022-11-09 17:21 Aldy Hernandez
  0 siblings, 0 replies; only message in thread
From: Aldy Hernandez @ 2022-11-09 17:21 UTC (permalink / raw)
  To: GCC patches; +Cc: Jakub Jelinek, Andrew MacLeod, Aldy Hernandez

When reading back from the global store, we must clear the NAN bit if
necessary.  The reason it's not happening is because the constructor
sets a NAN by default (when HONOR_NANS).  We must be careful to clear
the NAN bit if the original range didn't have a NAN.

I have commented the reason we use the constructor instead of filling
out the fields by hand, because it wasn't clear at re-reading this

	PR 107569/tree-optimization


	* (frange_storage_slot::get_frange): Clear
	NAN if appropriate.
	* (range_tests_floats): New test.
 gcc/ | 9 ++++++++-
 gcc/         | 9 +++++++++
 2 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/gcc/ b/gcc/
index 462447ba250..b57701f0ea1 100644
--- a/gcc/
+++ b/gcc/
@@ -276,13 +276,20 @@ frange_storage_slot::get_frange (frange &r, tree type) const
-  // Use the constructor because it will canonicalize the range.
+  // We use the constructor to create the new range instead of writing
+  // out the bits into the frange directly, because the global range
+  // being read may be being inlined into a function with different
+  // restrictions as when it was originally written.  We want to make
+  // sure the resulting range is canonicalized correctly for the new
+  // consumer.
   r = frange (type, m_min, m_max, m_kind);
   // The constructor will set the NAN bits for HONOR_NANS, but we must
   // make sure to set the NAN sign if known.
   if (HONOR_NANS (type) && (m_pos_nan ^ m_neg_nan) == 1)
     r.update_nan (m_neg_nan);
+  else if (!m_pos_nan && !m_neg_nan)
+    r.clear_nan ();
diff --git a/gcc/ b/gcc/
index 859c7fb4af9..852ac09f2c4 100644
--- a/gcc/
+++ b/gcc/
@@ -4051,6 +4051,15 @@ range_tests_floats ()
       ASSERT_TRUE (real_isinf (&r0.lower_bound (), true));
       ASSERT_TRUE (real_isinf (&r0.upper_bound (), true));
+  // Test that reading back a global range yields the same result as
+  // what we wrote into it.
+  tree ssa = make_temp_ssa_name (float_type_node, NULL, "blah");
+  r0.set_varying (float_type_node);
+  r0.clear_nan ();
+  set_range_info (ssa, r0);
+  get_global_range_query ()->range_of_expr (r1, ssa);
+  ASSERT_EQ (r0, r1);
 // Run floating range tests for various combinations of NAN and INF

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2022-11-09 17:22 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-09 17:21 [COMMITTED] Clear NAN when reading back a global range if necessary Aldy Hernandez

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).