[PATCH] Use size_type_node instead of sizetype for stride overflow checking (PR fortran/46945)
Jakub Jelinek
jakub@redhat.com
Wed Dec 15 18:46:00 GMT 2010
Hi!
Using sizetype for testing whether stride * element_size will overflow
is wrong, because unlike all sane unsigned types, sizetype doesn't have
modulo semantics, but instead has undefined overflow.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for
trunk?
2010-12-15 Jakub Jelinek <jakub@redhat.com>
PR fortran/46945
* trans-array.c (gfc_array_init_size): Perform stride overflow
checking and multiplication by element_size in size_type_node instead
of sizetype, return value with size_type_node type instead of
sometimes with sizetype and sometimes with gfc_array_index_type.
* gfortran.dg/pr46945.f90: New test.
--- gcc/fortran/trans-array.c.jj 2010-12-14 08:11:31.000000000 +0100
+++ gcc/fortran/trans-array.c 2010-12-15 12:16:49.000000000 +0100
@@ -4154,15 +4154,15 @@ gfc_array_init_size (tree descriptor, in
size of an element to get the total size. */
tmp = TYPE_SIZE_UNIT (gfc_get_element_type (type));
/* Convert to size_t. */
- element_size = fold_convert (sizetype, tmp);
- stride = fold_convert (sizetype, stride);
+ element_size = fold_convert (size_type_node, tmp);
+ stride = fold_convert (size_type_node, stride);
/* First check for overflow. Since an array of type character can
have zero element_size, we must check for that before
dividing. */
tmp = fold_build2_loc (input_location, TRUNC_DIV_EXPR,
- sizetype,
- TYPE_MAX_VALUE (sizetype), element_size);
+ size_type_node,
+ TYPE_MAX_VALUE (size_type_node), element_size);
tmp = fold_build3_loc (input_location, COND_EXPR, integer_type_node,
gfc_unlikely (fold_build2_loc (input_location, LT_EXPR,
boolean_type_node, tmp,
@@ -4178,7 +4178,7 @@ gfc_array_init_size (tree descriptor, in
*overflow, tmp);
*overflow = gfc_evaluate_now (tmp, pblock);
- size = fold_build2_loc (input_location, MULT_EXPR, sizetype,
+ size = fold_build2_loc (input_location, MULT_EXPR, size_type_node,
stride, element_size);
if (poffset != NULL)
@@ -4190,11 +4190,11 @@ gfc_array_init_size (tree descriptor, in
if (integer_zerop (or_expr))
return size;
if (integer_onep (or_expr))
- return gfc_index_zero_node;
+ return build_int_cst (size_type_node, 0);
var = gfc_create_var (TREE_TYPE (size), "size");
gfc_start_block (&thenblock);
- gfc_add_modify (&thenblock, var, size_zero_node);
+ gfc_add_modify (&thenblock, var, build_int_cst (size_type_node, 0));
thencase = gfc_finish_block (&thenblock);
gfc_start_block (&elseblock);
--- gcc/testsuite/gfortran.dg/pr46945.f90.jj 2010-12-15 12:22:11.000000000 +0100
+++ gcc/testsuite/gfortran.dg/pr46945.f90 2010-12-15 12:21:46.000000000 +0100
@@ -0,0 +1,10 @@
+! PR fortran/46945
+! { dg-do run }
+! { dg-options "-O -ftree-vrp -fno-tree-ccp -fno-tree-fre" }
+
+program pr46945
+ real, allocatable :: a(:,:,:)
+ integer :: n
+ n = 0
+ allocate (a(n,n,n))
+end program pr46945
Jakub
More information about the Gcc-patches
mailing list