On 09/15/14 09:57, Ilya Enkovich wrote:
On 15 Sep 09:20, Jeff Law wrote:
On 09/15/14 01:20, Ilya Enkovich wrote:
2014-06-05 Ilya Enkovich <ilya.enkov...@intel.com>
* tree-ssa-ccp.c: Include tree-chkp.h.
(insert_clobber_before_stack_restore): Handle
BUILT_IN_CHKP_BNDRET calls.
diff --git a/gcc/tree-ssa-ccp.c b/gcc/tree-ssa-ccp.c
index eeefeaf..e99bcf5 100644
--- a/gcc/tree-ssa-ccp.c
+++ b/gcc/tree-ssa-ccp.c
@@ -144,6 +144,7 @@ along with GCC; see the file COPYING3. If not see
#include "diagnostic-core.h"
#include "dbgcnt.h"
#include "params.h"
+#include "tree-chkp.h"
/* Possible lattice values. */
@@ -1880,6 +1881,9 @@ insert_clobber_before_stack_restore (tree saved_val, tree
var,
else if (gimple_assign_ssa_name_copy_p (stmt))
insert_clobber_before_stack_restore (gimple_assign_lhs (stmt), var,
visited);
+ else if (gimple_code (stmt) == GIMPLE_CALL
+ && chkp_gimple_call_builtin_p (stmt, BUILT_IN_CHKP_BNDRET))
+ continue;
else
gcc_assert (is_gimple_debug (stmt));
I think Richi also wanted the
gimple_code (stmt) == GIMPLE_CALL
Turned into:
is_gimple_call (stmt)
Presumably using chkp_gimple_call_builtin_p rather than
gimple_call_builtin_p is related to using different codes for the
instrumented builtin calls? Otherwise why use
chkp_gimple_call_builtin_p rather than gimple_call_builtin_p?
jeff
The reason is that target may provide own version of builtin (and currently
does it) for instrumentation. Therefore we should obtain fndecl from target
and then compare. That is what chkp_gimple_call_builtin_p does. I just moved
gimple code check into chkp_gimple_call_builtin_p to make similar to
gimple_call_builtin_p.
Thanks,
Ilya
--
2014-09-15 Ilya Enkovich <ilya.enkov...@intel.com>
* tree-ssa-ccp.c: Include tree-chkp.h.
(insert_clobber_before_stack_restore): Handle
BUILT_IN_CHKP_BNDRET calls.
OK.
Jeff
