[PATCH] fold-const: Fix ICE in fold_read_from_constant_string on invalid code [PR99204]

Jakub Jelinek jakub@redhat.com
Tue Feb 23 08:46:16 GMT 2021


Hi!

fold_read_from_constant_string and expand_expr_real_1 have code to optimize
constant reads from string (tree vs. rtl).
If the STRING_CST array type has zero low bound, index is fold converted to
sizetype and so the compare_tree_int works fine, but if it has some other
low bound, it calls size_diffop_loc and that function from 2 sizetype
operands creates a ssizetype difference.  expand_expr_real_1 then uses
tree_fits_uhwi_p + compare_tree_int and so works fine, but fold-const.c
only checked if index is INTEGER_CST and calls compare_tree_int, which means
for negative index it will succeed and result in UB in the compiler.

This patch just follows what expand_expr_real_1 is doing.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2021-02-22  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/99204
	* fold-const.c (fold_read_from_constant_string): Check that
	tree_fits_uhwi_p (index) rather than just that index is INTEGER_CST.

	* gfortran.dg/pr99204.f90: New test.

--- gcc/fold-const.c.jj	2021-01-04 10:25:39.031231582 +0100
+++ gcc/fold-const.c	2021-02-22 19:46:57.167342075 +0100
@@ -15433,7 +15433,7 @@ fold_read_from_constant_string (tree exp
       if (string
 	  && TYPE_MODE (TREE_TYPE (exp)) == TYPE_MODE (TREE_TYPE (TREE_TYPE (string)))
 	  && TREE_CODE (string) == STRING_CST
-	  && TREE_CODE (index) == INTEGER_CST
+	  && tree_fits_uhwi_p (index)
 	  && compare_tree_int (index, TREE_STRING_LENGTH (string)) < 0
 	  && is_int_mode (TYPE_MODE (TREE_TYPE (TREE_TYPE (string))),
 			  &char_mode)
--- gcc/testsuite/gfortran.dg/pr99204.f90.jj	2021-02-22 19:52:55.749431186 +0100
+++ gcc/testsuite/gfortran.dg/pr99204.f90	2021-02-22 19:51:07.410612678 +0100
@@ -0,0 +1,10 @@
+! PR tree-optimization/99204
+! { dg-do compile }
+! { dg-options "-O2 -w" }
+
+program pr99204
+  character :: c
+  integer :: i = -12345678
+  c = 'abc'(i:i)
+  print *, c
+end

	Jakub



More information about the Gcc-patches mailing list