On Thu, Jul 30, 2015 at 03:30:12PM -0500, Lynn A. Boger wrote:
> PR66870
> * gcc/config/rs6000/rs6000.c: Add check for no_split_stack
> function attribute along with flag_split_stack check to
> determine when to generate split stack prologue for
> ppc64 and ppc64le.
Looks good to me, except that the changelog entry should mention the
modified functions, for example:
PR target/66870
* gcc/config/rs6000/rs6000.c (rs6000_emit_prologue): Check for
no_split_stack function attribute along with flag_split_stack.
(rs6000_expand_split_stack_prologue): Likewise.
Also, formatting rules for gcc say to not split a line after an
operator.
> + int using_split_stack = flag_split_stack &&
> + (lookup_attribute ("no_split_stack", DECL_ATTRIBUTES (cfun->decl))
> + == NULL);
The "&&" belongs on the next line, with parentheses added so that emacs
and indent will line up the continuation nicely.
int using_split_stack = (flag_split_stack
&& (lookup_attribute ("no_split_stack",
DECL_ATTRIBUTES (cfun->decl))
== NULL));
David, the following is another piece of the PR66870 fixes. This
stops shrink-wrap from moving insns around in the first few blocks of
a function, in a way that is incorrect given that r12 is live.
Bootstrapped and regression tested powerpc64le-linux (and
powerpc64-linux by Lynn).
PR target/66870
* config/rs6000/rs6000.c (machine_function): Add split_stack_argp_used.
(rs6000_emit_prologue): Set it.
(rs6000_set_up_by_prologue): Specify r12 when split_stack_argp_used.
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 5d9ff88..dc2e20c 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -166,6 +166,7 @@ typedef struct GTY(()) machine_function
rtx sdmode_stack_slot;
/* Alternative internal arg pointer for -fsplit-stack. */
rtx split_stack_arg_pointer;
+ bool split_stack_argp_used;
/* Flag if r2 setup is needed with ELFv2 ABI. */
bool r2_setup_needed;
} machine_function;
@@ -24458,6 +24459,7 @@ rs6000_emit_prologue (void)
__morestack was called, it left the arg pointer to the old
stack in r29. Otherwise, the arg pointer is the top of the
current frame. */
+ cfun->machine->split_stack_argp_used = true;
if (sp_adjust)
{
rtx r12 = gen_rtx_REG (Pmode, 12);
@@ -33711,6 +33713,8 @@ rs6000_set_up_by_prologue (struct
hard_reg_set_container *set)
&& TARGET_MINIMAL_TOC
&& get_pool_size () != 0)
add_to_hard_reg_set (&set->set, Pmode, RS6000_PIC_OFFSET_TABLE_REGNUM);
+ if (cfun->machine->split_stack_argp_used)
+ add_to_hard_reg_set (&set->set, Pmode, 12);
}
--
Alan Modra
Australia Development Lab, IBM
