This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[new-regalloc-branch] bugfix for `get_free_reg'
- To: gcc-patches at gcc dot gnu dot org
- Subject: [new-regalloc-branch] bugfix for `get_free_reg'
- From: Denis Chertykov <denisc at overta dot ru>
- Date: Mon, 30 Jul 2001 20:46:11 +0400
- Cc: denisc at overta dot ru, Daniel Berlin <dan at cgsoftware dot com>, Michael Matz <matzmich at cs dot tu-berlin dot de>
Few times ago I'm already fix similar bug.
It was:
http://gcc.gnu.org/ml/gcc-patches/2001-02/msg00063.html
If we test hard reg `c' for mode `mode' in:
if (!TEST_HARD_REG_BIT (dont_begin_colors, c)
&& TEST_HARD_REG_BIT (free_colors, c)
&& HARD_REGNO_MODE_OK (c, mode))
then we needn't test `c' for mode `mode' anymore.
- if (i != size)
- {
- c += i;
- continue;
- }
- for (i = 1; i < size && HARD_REGNO_MODE_OK (c + i, mode); i++);
--------------------------------^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
We alredy have tested `c' for mode `mode'.
Test a hardreg `c+1' for mode `mode' is wrong.
Mon Jul 30 20:21:07 2001 Denis Chertykov <denisc@overta.ru>
* ra.c (get_free_reg): Wrong test removed.
(reg_alloc): Fix formatting.
Index: ra.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/Attic/ra.c,v
retrieving revision 1.1.2.25
diff -c -3 -p -r1.1.2.25 ra.c
*** ra.c 2001/07/29 18:35:42 1.1.2.25
--- ra.c 2001/07/30 16:20:41
*************** get_free_reg (dont_begin_colors, free_co
*** 3124,3135 ****
int i, size;
size = HARD_REGNO_NREGS (c, mode);
for (i = 1; i < size && TEST_HARD_REG_BIT (free_colors, c + i); i++);
- if (i != size)
- {
- c += i;
- continue;
- }
- for (i = 1; i < size && HARD_REGNO_MODE_OK (c + i, mode); i++);
if (i == size)
{
if (size < 2 || (c & 1) == 0)
--- 3124,3129 ----
*************** reg_alloc (void)
*** 4135,4141 ****
else
{
allocate_reg_info (max_reg_num (), FALSE, TRUE);
! reg_scan_update (get_insns(), BLOCK_END (n_basic_blocks - 1), max_regno);
regclass (get_insns (), max_reg_num (), rtl_dump_file);
}
dump_ra (df);
--- 4129,4136 ----
else
{
allocate_reg_info (max_reg_num (), FALSE, TRUE);
! reg_scan_update (get_insns(), BLOCK_END (n_basic_blocks - 1),
! max_regno);
regclass (get_insns (), max_reg_num (), rtl_dump_file);
}
dump_ra (df);