Re: [patch, Fortran] Fix PR 118862, overwide shft

[Steve Kargl]

On Sun, Feb 16, 2025 at 09:36:20AM +0100, Thomas Koenig wrote:
> 
> this patch is a variation of Jakub's patch in the PR, which
> avoids overflow on the mask used for exponentiation and
> fixes unsigned HOST_WIDE_INT.  I tried testing this on
> a POWER machine, but --with-build-config=bootstrap-ubsan
> fails bootstrap there.
> 
> Regression-tested. OK for trunk?
> 

Yes.  Thanks for the patch.

-- 
Steve

[patch, fortran, committed] Remove duplicate output of save attribute

[Thomas Koenig]

Hello world,

I just committed the attached patch as obvious, simple and having
no user impact), as r15-7575.

Previously, a dump looked like

attributes: (VARIABLE IMPLICIT-SAVE BIND(C) SAVE-IMPLICIT)

and now it is

attributes: (VARIABLE IMPLICIT-SAVE BIND(C))

Best regards

Thomas

Remove double output of attr->save.

In the recent patch for dumping all attributes, there were
duplicates for attr->save, which is output via gfc_code2string
previously. This patch removes that double output.

gcc/fortran/ChangeLog:

* dump-parse-tree.cc (show_attr): Remove double output
of attr->save.
diff --git a/gcc/fortran/dump-parse-tree.cc b/gcc/fortran/dump-parse-tree.cc
index adc07da133e..7726b708ad8 100644
--- a/gcc/fortran/dump-parse-tree.cc
+++ b/gcc/fortran/dump-parse-tree.cc
@@ -1017,10 +1017,6 @@ show_attr (symbol_attribute *attr, const char * module)
 fputs (" AUTOMATIC", dumpfile);
   if (attr->class_pointer)
 fputs (" CLASS-POINTER", dumpfile);
-  if (attr->save == SAVE_EXPLICIT)
-fputs (" SAVE-EXPLICIT", dumpfile);
-  if (attr->save == SAVE_IMPLICIT)
-fputs (" SAVE-IMPLICIT", dumpfile);
   if (attr->used_in_submodule)
 fputs (" USED-IN-SUBMODULE", dumpfile);
   if (attr->use_only)

[patch, Fortran] Fix PR 118862, overwide shft

[Thomas Koenig]

Hello world,

this patch is a variation of Jakub's patch in the PR, which
avoids overflow on the mask used for exponentiation and
fixes unsigned HOST_WIDE_INT.  I tried testing this on
a POWER machine, but --with-build-config=bootstrap-ubsan
fails bootstrap there.

Regression-tested. OK for trunk?

Best regards

Thomas

Avoid shift wider than unsigned HOST_WIDE_INT on unsigned integer 
exponentiation.


gcc/fortran/ChangeLog:

PR fortran/118862
* trans-expr.cc (gfc_conv_cst_int_power): Use functions for
unsigned wide integer.
(gfc_conv_cst_uint_power): Avoid generating the mask if it would
overflow an unsigned HOST_WIDE_INT.  Format fixes.

diff --git a/gcc/fortran/trans-expr.cc b/gcc/fortran/trans-expr.cc
index 1329efcd6eb..14301265742 100644
--- a/gcc/fortran/trans-expr.cc
+++ b/gcc/fortran/trans-expr.cc
@@ -3604,7 +3604,7 @@ gfc_conv_cst_int_power (gfc_se * se, tree lhs, tree rhs)
if (bit_size(rhs) < bit_size(lhs))  ! Checked here.
 	 return lhs ** rhs;
 
-   mask = (1 < bit_size(a) - 1) / 2;
+   mask = 1 << (bit_size(a) - 1) / 2;
return lhs ** (n & rhs);
  }
if (rhs > bit_size(lhs))  ! Checked here.
@@ -3624,13 +3624,13 @@ gfc_conv_cst_uint_power (gfc_se * se, tree lhs, tree rhs)
   tree vartmp_odd[POWI_TABLE_SIZE], vartmp_even[POWI_TABLE_SIZE];
 
   /* Anything ** 0 is one.  */
-  if (tree_int_cst_sgn (rhs) == 0)
+  if (integer_zerop (rhs))
 {
   se->expr = build_int_cst (type, 1);
   return 1;
 }
 
-  if (!wi::fits_shwi_p (wrhs))
+  if (!wi::fits_uhwi_p (wrhs))
 return 0;
 
   n = wrhs.to_uhwi ();
@@ -3642,19 +3642,18 @@ gfc_conv_cst_uint_power (gfc_se * se, tree lhs, tree rhs)
 			tmp, build_int_cst (type, 1));
 
   lhs_prec = TYPE_PRECISION (type);
-  rhs_prec = TYPE_PRECISION (TREE_TYPE(rhs));
+  rhs_prec = TYPE_PRECISION (TREE_TYPE (rhs));
 
-  if (rhs_prec >= lhs_prec)
+  if (rhs_prec >= lhs_prec && lhs_prec <= HOST_BITS_PER_WIDE_INT)
 {
-  unsigned HOST_WIDE_INT mask;
-  mask = (((unsigned HOST_WIDE_INT) 1) << (lhs_prec - 1)) - 1;
+  unsigned HOST_WIDE_INT mask = (HOST_WIDE_INT_1U << (lhs_prec - 1)) - 1;
   n_odd = n & mask;
 }
   else
 n_odd = n;
 
   memset (vartmp_odd, 0, sizeof (vartmp_odd));
-  vartmp_odd[0] = build_int_cst(type, 1);
+  vartmp_odd[0] = build_int_cst (type, 1);
   vartmp_odd[1] = lhs;
   odd_branch = gfc_conv_powi (se, n_odd, vartmp_odd);
   even_branch = NULL_TREE;
@@ -3666,7 +3665,7 @@ gfc_conv_cst_uint_power (gfc_se * se, tree lhs, tree rhs)
   if (n_odd != n)
 	{
 	  memset (vartmp_even, 0, sizeof (vartmp_even));
-	  vartmp_even[0] = build_int_cst(type, 1);
+	  vartmp_even[0] = build_int_cst (type, 1);
 	  vartmp_even[1] = lhs;
 	  even_branch = gfc_conv_powi (se, n, vartmp_even);
 	}

Re: [patch, Fortran] Fix PR 118862, overwide shft

[Thomas Koenig]

Am 17.02.25 um 05:53 schrieb Steve Kargl:

On Sun, Feb 16, 2025 at 09:36:20AM +0100, Thomas Koenig wrote:


this patch is a variation of Jakub's patch in the PR, which
avoids overflow on the mask used for exponentiation and
fixes unsigned HOST_WIDE_INT.  I tried testing this on
a POWER machine, but --with-build-config=bootstrap-ubsan
fails bootstrap there.

Regression-tested. OK for trunk?



Yes.  Thanks for the patch.


Committed, thanks for the review!

Best regards

Thomas