I committed this patch. It fixes a thinko I had in the builtin support that I
committed on November 29th. I had used the wrong field for figuring out the
attributes, and it meant that gcc.target/powerpc/recip-5.c started failing
since __builtin_recipdiv and __builtin_recipdivf were not considered const
functions if -ffast-math was used.
2011-12-02 Michael Meissner <meiss...@the-meissners.org>
PR target/51390
* config/rs6000/rs6000.c (def_builtin): Use the correct field to
figure out the function's attributes.
(struct builtin_description): Mark mask field as const since we no
longer need to modify it for SPE and PAIRED builtins.
Index: gcc/config/rs6000/rs6000.c
===================================================================
--- gcc/config/rs6000/rs6000.c (revision 181929)
+++ gcc/config/rs6000/rs6000.c (working copy)
@@ -215,10 +215,7 @@ static GTY(()) section *toc_section;
struct builtin_description
{
- /* mask is not const because we're going to alter it below. This
- nonsense will go away when we rewrite the -march infrastructure
- to give us more target flag bits. */
- unsigned int mask;
+ const unsigned int mask;
const enum insn_code icode;
const char *const name;
const enum rs6000_builtins code;
@@ -9394,7 +9391,7 @@ static void
def_builtin (const char *name, tree type, enum rs6000_builtins code)
{
tree t;
- unsigned classify = rs6000_builtin_info[(int)code].mask;
+ unsigned classify = rs6000_builtin_info[(int)code].attr;
const char *attr_string = "";
gcc_assert (name != NULL);
--
Michael Meissner, IBM
5 Technology Place Drive, M/S 2757, Westford, MA 01886-3141, USA
meiss...@linux.vnet.ibm.com fax +1 (978) 399-6899
