[PATCH] Decimal floats can never be an frange::singleton_p.

[PATCH] Decimal floats can never be an frange::singleton_p.

[Aldy Hernandez]

As Jakub mentioned in the PR, because many numbers have multiple
possible representations, we can't reliably return true here.

I'll commit this if tests pass.

I wonder if its worth even handling decimal floats in frange, since
there's a lot of things we can't represent.  I suppose even though we
could never propagate an actual value with VRP, we could fold
conditionals (symbolic and stuff outside ranges, etc) ??.  

Thoughts?

	PR middle-end/106831

gcc/ChangeLog:

	* value-range.cc (frange::singleton_p): Return false for
	DECIMAL_FLOAT_TYPE_P.
---
 gcc/value-range.cc | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/gcc/value-range.cc b/gcc/value-range.cc
index c3f668a811a..12a3750d078 100644
--- a/gcc/value-range.cc
+++ b/gcc/value-range.cc
@@ -647,6 +647,14 @@ frange::singleton_p (tree *result) const
       if (HONOR_NANS (m_type) && !get_nan ().no_p ())
 	return false;
 
+      // Because the significand is not normalized (there is no
+      // implicit leading "1").  Most values with less than 7
+      // significant digits have multiple possible representations.
+      // Zero has 192 possible representations (or twice that for
+      // signed zeros).
+      if (DECIMAL_FLOAT_TYPE_P (m_type))
+	return false;
+
       // Return the appropriate zero if known.
       if (HONOR_SIGNED_ZEROS (m_type) && zero_p ())
 	{
-- 
2.37.1

Re: [PATCH] Decimal floats can never be an frange::singleton_p.

[Jakub Jelinek]

On Mon, Sep 05, 2022 at 03:47:01PM +0200, Aldy Hernandez wrote:
> As Jakub mentioned in the PR, because many numbers have multiple
> possible representations, we can't reliably return true here.
> 
> I'll commit this if tests pass.
> 
> I wonder if its worth even handling decimal floats in frange, since
> there's a lot of things we can't represent.  I suppose even though we
> could never propagate an actual value with VRP, we could fold
> conditionals (symbolic and stuff outside ranges, etc) ??.  
> 
> Thoughts?
> 
> 	PR middle-end/106831
> 
> gcc/ChangeLog:
> 
> 	* value-range.cc (frange::singleton_p): Return false for
> 	DECIMAL_FLOAT_TYPE_P.
> ---
>  gcc/value-range.cc | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/gcc/value-range.cc b/gcc/value-range.cc
> index c3f668a811a..12a3750d078 100644
> --- a/gcc/value-range.cc
> +++ b/gcc/value-range.cc
> @@ -647,6 +647,14 @@ frange::singleton_p (tree *result) const
>        if (HONOR_NANS (m_type) && !get_nan ().no_p ())
>  	return false;
>  
> +      // Because the significand is not normalized (there is no
> +      // implicit leading "1").  Most values with less than 7
> +      // significant digits have multiple possible representations.
> +      // Zero has 192 possible representations (or twice that for
> +      // signed zeros).

The exact details (7, 192) are dependent on the format, for
decimal64 it is (16, 768) and for decimal128 (34, 12288).

	Jakub

Re: [PATCH] Decimal floats can never be an frange::singleton_p.

[Aldy Hernandez]

[-- Attachment #1: Type: text/plain, Size: 1697 bytes --]

Final version of the patch.  Disable frange for DECIMAL_FLOAT_MODE_P altogether.

Tested on x86-64 Linux with regstrap as well as mpfr tests.

On Mon, Sep 5, 2022 at 3:47 PM Aldy Hernandez <aldyh@redhat.com> wrote:
>
> As Jakub mentioned in the PR, because many numbers have multiple
> possible representations, we can't reliably return true here.
>
> I'll commit this if tests pass.
>
> I wonder if its worth even handling decimal floats in frange, since
> there's a lot of things we can't represent.  I suppose even though we
> could never propagate an actual value with VRP, we could fold
> conditionals (symbolic and stuff outside ranges, etc) ??.
>
> Thoughts?
>
>         PR middle-end/106831
>
> gcc/ChangeLog:
>
>         * value-range.cc (frange::singleton_p): Return false for
>         DECIMAL_FLOAT_TYPE_P.
> ---
>  gcc/value-range.cc | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/gcc/value-range.cc b/gcc/value-range.cc
> index c3f668a811a..12a3750d078 100644
> --- a/gcc/value-range.cc
> +++ b/gcc/value-range.cc
> @@ -647,6 +647,14 @@ frange::singleton_p (tree *result) const
>        if (HONOR_NANS (m_type) && !get_nan ().no_p ())
>         return false;
>
> +      // Because the significand is not normalized (there is no
> +      // implicit leading "1").  Most values with less than 7
> +      // significant digits have multiple possible representations.
> +      // Zero has 192 possible representations (or twice that for
> +      // signed zeros).
> +      if (DECIMAL_FLOAT_TYPE_P (m_type))
> +       return false;
> +
>        // Return the appropriate zero if known.
>        if (HONOR_SIGNED_ZEROS (m_type) && zero_p ())
>         {
> --
> 2.37.1
>

[-- Attachment #2: 0002-Disable-decimal-floating-point-in-frange.patch --]
[-- Type: text/x-patch, Size: 2458 bytes --]

From 4f32e08f7317c52ed72ee48447cb4dd5d68a5c28 Mon Sep 17 00:00:00 2001
From: Aldy Hernandez <aldyh@redhat.com>
Date: Mon, 5 Sep 2022 15:41:39 +0200
Subject: [PATCH] Disable decimal floating point in frange.

As Jakub mentioned in the PR, because many numbers have multiple
possible representations, we can't reliably return true for singleton_p.
For that matter, we may not be capable of modeling them just yet.
Disabling them until someone with DFP knowledge can opine or extend
frange.

	PR middle-end/106831

gcc/ChangeLog:

	* value-range.h (frange::supports_p): Disable decimal floats.
	* range-op-float.cc (frange_drop_inf): Remove DECIMAL_FLOAT_MODE_P
	check.
	(frange_drop_ninf): Same.
---
 gcc/range-op-float.cc | 10 ----------
 gcc/value-range.h     |  5 ++++-
 2 files changed, 4 insertions(+), 11 deletions(-)

diff --git a/gcc/range-op-float.cc b/gcc/range-op-float.cc
index 7301e5a060b..050f07a9867 100644
--- a/gcc/range-op-float.cc
+++ b/gcc/range-op-float.cc
@@ -204,11 +204,6 @@ frelop_early_resolve (irange &r, tree type,
 static inline void
 frange_drop_inf (frange &r, tree type)
 {
-  // FIXME: build_real() bails on decimal float modes when called with
-  // a max representable endpoint.
-  if (DECIMAL_FLOAT_MODE_P (TYPE_MODE (type)))
-    return;
-
   REAL_VALUE_TYPE max;
   real_max_representable (&max, type);
   frange tmp (type, r.lower_bound (), max);
@@ -221,11 +216,6 @@ frange_drop_inf (frange &r, tree type)
 static inline void
 frange_drop_ninf (frange &r, tree type)
 {
-  // FIXME: build_real() bails on decimal float modes when called with
-  // a max representable endpoint.
-  if (DECIMAL_FLOAT_MODE_P (TYPE_MODE (type)))
-    return;
-
   REAL_VALUE_TYPE min;
   real_min_representable (&min, type);
   frange tmp (type, min, r.upper_bound ());
diff --git a/gcc/value-range.h b/gcc/value-range.h
index bc00f3d5b08..645dc76c33a 100644
--- a/gcc/value-range.h
+++ b/gcc/value-range.h
@@ -338,7 +338,10 @@ public:
 	  value_range_kind = VR_RANGE);
   static bool supports_p (const_tree type)
   {
-    return SCALAR_FLOAT_TYPE_P (type);
+    // ?? Decimal floats can have multiple representations for the
+    // same number.  Supporting them may be as simple as just
+    // disabling them in singleton_p.  No clue.
+    return SCALAR_FLOAT_TYPE_P (type) && !DECIMAL_FLOAT_TYPE_P (type);
   }
   virtual tree type () const override;
   virtual void set (tree, tree, value_range_kind = VR_RANGE) override;
-- 
2.37.1