diff mbox series

[1/2] Fix PR 105532: match.pd patterns calling tree_nonzero_bits with vector types

Message ID 1667425595-2654-2-git-send-email-apinski@marvell.com
State Accepted
Headers
Series tree_nonzero_bits vs vector and complex types |

Checks

Context Check Description
snail/gcc-patch-check success Github commit url

Commit Message

Li, Pan2 via Gcc-patches Nov. 2, 2022, 9:46 p.m. UTC 
  From: Andrew Pinski <apinski@marvell.com>

Even though this PR was reported with an ubsan issue, the problem is
tree_nonzero_bits is being called with an expression which is a vector type.
This fixes three patterns I noticed which does that.
And adds a testcase for one of the patterns.

OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions

gcc/ChangeLog:

	PR tree-optimization/105532
	* match.pd (~(X >> Y) -> ~X >> Y): Check if it is an integral
	type before calling tree_nonzero_bits.
	(popcount(X) + popcount(Y)): Likewise.
	(popcount(X&C1)): Likewise.

gcc/testsuite/ChangeLog:

	* gcc.c-torture/compile/vector-shift-1.c: New test.
---
 gcc/match.pd                                  | 25 +++++++++++--------
 .../gcc.c-torture/compile/vector-shift-1.c    |  8 ++++++
 2 files changed, 22 insertions(+), 11 deletions(-)
 create mode 100644 gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c

Comments

Richard Biener Nov. 5, 2022, 11:44 a.m. UTC | #1 
On Wed, Nov 2, 2022 at 10:47 PM apinski--- via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> From: Andrew Pinski <apinski@marvell.com>
>
> Even though this PR was reported with an ubsan issue, the problem is
> tree_nonzero_bits is being called with an expression which is a vector type.

It seems to me the semantics
for vectors should be clear but the users didn't expect that result?

> This fixes three patterns I noticed which does that.
> And adds a testcase for one of the patterns.
>
> OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions

OK.

> gcc/ChangeLog:
>
>         PR tree-optimization/105532
>         * match.pd (~(X >> Y) -> ~X >> Y): Check if it is an integral
>         type before calling tree_nonzero_bits.
>         (popcount(X) + popcount(Y)): Likewise.
>         (popcount(X&C1)): Likewise.
>
> gcc/testsuite/ChangeLog:
>
>         * gcc.c-torture/compile/vector-shift-1.c: New test.
> ---
>  gcc/match.pd                                  | 25 +++++++++++--------
>  .../gcc.c-torture/compile/vector-shift-1.c    |  8 ++++++
>  2 files changed, 22 insertions(+), 11 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index 194ba8f5188..5833e05a926 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -1371,7 +1371,8 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
>     /* For logical right shifts, this is possible only if @0 doesn't
>        have MSB set and the logical right shift is changed into
>        arithmetic shift.  */
> -   (if (!wi::neg_p (tree_nonzero_bits (@0)))
> +   (if (INTEGRAL_TYPE_P (type)
> +        && !wi::neg_p (tree_nonzero_bits (@0)))
>      (with { tree stype = signed_type_for (TREE_TYPE (@0)); }
>       (convert (rshift (bit_not! (convert:stype @0)) @1))))))
>
> @@ -7518,7 +7519,8 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
>  /* popcount(X) + popcount(Y) is popcount(X|Y) when X&Y must be zero.  */
>  (simplify
>    (plus (POPCOUNT:s @0) (POPCOUNT:s @1))
> -  (if (wi::bit_and (tree_nonzero_bits (@0), tree_nonzero_bits (@1)) == 0)
> +  (if (INTEGRAL_TYPE_P (type)
> +       && wi::bit_and (tree_nonzero_bits (@0), tree_nonzero_bits (@1)) == 0)
>      (POPCOUNT (bit_ior @0 @1))))
>
>  /* popcount(X) == 0 is X == 0, and related (in)equalities.  */
> @@ -7550,15 +7552,16 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
>  (for pfun (POPCOUNT PARITY)
>    (simplify
>      (pfun @0)
> -    (with { wide_int nz = tree_nonzero_bits (@0); }
> -      (switch
> -       (if (nz == 1)
> -         (convert @0))
> -       (if (wi::popcount (nz) == 1)
> -         (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
> -           (convert (rshift:utype (convert:utype @0)
> -                                  { build_int_cst (integer_type_node,
> -                                                   wi::ctz (nz)); }))))))))
> +    (if (INTEGRAL_TYPE_P (type))
> +     (with { wide_int nz = tree_nonzero_bits (@0); }
> +       (switch
> +        (if (nz == 1)
> +          (convert @0))
> +        (if (wi::popcount (nz) == 1)
> +          (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
> +            (convert (rshift:utype (convert:utype @0)
> +                                   { build_int_cst (integer_type_node,
> +                                                    wi::ctz (nz)); })))))))))
>
>  #if GIMPLE
>  /* 64- and 32-bits branchless implementations of popcount are detected:
> diff --git a/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c b/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
> new file mode 100644
> index 00000000000..142ea56d5bb
> --- /dev/null
> +++ b/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
> @@ -0,0 +1,8 @@
> +typedef unsigned char __attribute__((__vector_size__ (1))) U;
> +
> +U
> +foo (U u)
> +{
> +  u = u == u;
> +  return (~(u >> 255));
> +}
> --
> 2.17.1
>
Richard Biener Dec. 20, 2022, 1:45 p.m. UTC | #2 
On Sat, Nov 5, 2022 at 12:44 PM Richard Biener
<richard.guenther@gmail.com> wrote:
>
> On Wed, Nov 2, 2022 at 10:47 PM apinski--- via Gcc-patches
> <gcc-patches@gcc.gnu.org> wrote:
> >
> > From: Andrew Pinski <apinski@marvell.com>
> >
> > Even though this PR was reported with an ubsan issue, the problem is
> > tree_nonzero_bits is being called with an expression which is a vector type.
>
> It seems to me the semantics
> for vectors should be clear but the users didn't expect that result?
>
> > This fixes three patterns I noticed which does that.
> > And adds a testcase for one of the patterns.
> >
> > OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions
>
> OK.

You didn't push this yet?

> > gcc/ChangeLog:
> >
> >         PR tree-optimization/105532
> >         * match.pd (~(X >> Y) -> ~X >> Y): Check if it is an integral
> >         type before calling tree_nonzero_bits.
> >         (popcount(X) + popcount(Y)): Likewise.
> >         (popcount(X&C1)): Likewise.
> >
> > gcc/testsuite/ChangeLog:
> >
> >         * gcc.c-torture/compile/vector-shift-1.c: New test.
> > ---
> >  gcc/match.pd                                  | 25 +++++++++++--------
> >  .../gcc.c-torture/compile/vector-shift-1.c    |  8 ++++++
> >  2 files changed, 22 insertions(+), 11 deletions(-)
> >  create mode 100644 gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
> >
> > diff --git a/gcc/match.pd b/gcc/match.pd
> > index 194ba8f5188..5833e05a926 100644
> > --- a/gcc/match.pd
> > +++ b/gcc/match.pd
> > @@ -1371,7 +1371,8 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> >     /* For logical right shifts, this is possible only if @0 doesn't
> >        have MSB set and the logical right shift is changed into
> >        arithmetic shift.  */
> > -   (if (!wi::neg_p (tree_nonzero_bits (@0)))
> > +   (if (INTEGRAL_TYPE_P (type)
> > +        && !wi::neg_p (tree_nonzero_bits (@0)))
> >      (with { tree stype = signed_type_for (TREE_TYPE (@0)); }
> >       (convert (rshift (bit_not! (convert:stype @0)) @1))))))
> >
> > @@ -7518,7 +7519,8 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> >  /* popcount(X) + popcount(Y) is popcount(X|Y) when X&Y must be zero.  */
> >  (simplify
> >    (plus (POPCOUNT:s @0) (POPCOUNT:s @1))
> > -  (if (wi::bit_and (tree_nonzero_bits (@0), tree_nonzero_bits (@1)) == 0)
> > +  (if (INTEGRAL_TYPE_P (type)
> > +       && wi::bit_and (tree_nonzero_bits (@0), tree_nonzero_bits (@1)) == 0)
> >      (POPCOUNT (bit_ior @0 @1))))
> >
> >  /* popcount(X) == 0 is X == 0, and related (in)equalities.  */
> > @@ -7550,15 +7552,16 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> >  (for pfun (POPCOUNT PARITY)
> >    (simplify
> >      (pfun @0)
> > -    (with { wide_int nz = tree_nonzero_bits (@0); }
> > -      (switch
> > -       (if (nz == 1)
> > -         (convert @0))
> > -       (if (wi::popcount (nz) == 1)
> > -         (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
> > -           (convert (rshift:utype (convert:utype @0)
> > -                                  { build_int_cst (integer_type_node,
> > -                                                   wi::ctz (nz)); }))))))))
> > +    (if (INTEGRAL_TYPE_P (type))
> > +     (with { wide_int nz = tree_nonzero_bits (@0); }
> > +       (switch
> > +        (if (nz == 1)
> > +          (convert @0))
> > +        (if (wi::popcount (nz) == 1)
> > +          (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
> > +            (convert (rshift:utype (convert:utype @0)
> > +                                   { build_int_cst (integer_type_node,
> > +                                                    wi::ctz (nz)); })))))))))
> >
> >  #if GIMPLE
> >  /* 64- and 32-bits branchless implementations of popcount are detected:
> > diff --git a/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c b/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
> > new file mode 100644
> > index 00000000000..142ea56d5bb
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
> > @@ -0,0 +1,8 @@
> > +typedef unsigned char __attribute__((__vector_size__ (1))) U;
> > +
> > +U
> > +foo (U u)
> > +{
> > +  u = u == u;
> > +  return (~(u >> 255));
> > +}
> > --
> > 2.17.1
> >
diff mbox series

Patch

diff --git a/gcc/match.pd b/gcc/match.pd
index 194ba8f5188..5833e05a926 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -1371,7 +1371,8 @@  DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
    /* For logical right shifts, this is possible only if @0 doesn't
       have MSB set and the logical right shift is changed into
       arithmetic shift.  */
-   (if (!wi::neg_p (tree_nonzero_bits (@0)))
+   (if (INTEGRAL_TYPE_P (type)
+        && !wi::neg_p (tree_nonzero_bits (@0)))
     (with { tree stype = signed_type_for (TREE_TYPE (@0)); }
      (convert (rshift (bit_not! (convert:stype @0)) @1))))))
 
@@ -7518,7 +7519,8 @@  DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
 /* popcount(X) + popcount(Y) is popcount(X|Y) when X&Y must be zero.  */
 (simplify
   (plus (POPCOUNT:s @0) (POPCOUNT:s @1))
-  (if (wi::bit_and (tree_nonzero_bits (@0), tree_nonzero_bits (@1)) == 0)
+  (if (INTEGRAL_TYPE_P (type)
+       && wi::bit_and (tree_nonzero_bits (@0), tree_nonzero_bits (@1)) == 0)
     (POPCOUNT (bit_ior @0 @1))))
 
 /* popcount(X) == 0 is X == 0, and related (in)equalities.  */
@@ -7550,15 +7552,16 @@  DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
 (for pfun (POPCOUNT PARITY)
   (simplify
     (pfun @0)
-    (with { wide_int nz = tree_nonzero_bits (@0); }
-      (switch
-	(if (nz == 1)
-	  (convert @0))
-	(if (wi::popcount (nz) == 1)
-	  (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
-	    (convert (rshift:utype (convert:utype @0)
-				   { build_int_cst (integer_type_node,
-						    wi::ctz (nz)); }))))))))
+    (if (INTEGRAL_TYPE_P (type))
+     (with { wide_int nz = tree_nonzero_bits (@0); }
+       (switch
+	 (if (nz == 1)
+	   (convert @0))
+	 (if (wi::popcount (nz) == 1)
+	   (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
+	     (convert (rshift:utype (convert:utype @0)
+				    { build_int_cst (integer_type_node,
+						     wi::ctz (nz)); })))))))))
 
 #if GIMPLE
 /* 64- and 32-bits branchless implementations of popcount are detected:
diff --git a/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c b/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
new file mode 100644
index 00000000000..142ea56d5bb
--- /dev/null
+++ b/gcc/testsuite/gcc.c-torture/compile/vector-shift-1.c
@@ -0,0 +1,8 @@ 
+typedef unsigned char __attribute__((__vector_size__ (1))) U;
+
+U
+foo (U u)
+{
+  u = u == u;
+  return (~(u >> 255));
+}