Add flag -ftree-loop-if-convert-stores.

This patch adds a flag that controls the replacement of the memory
writes that are in predicated basic blocks with a full write:

for (...)
  if (cond)
    A[i] = foo

is replaced with:

for (...)
  A[i] = cond ? foo : A[i]

In order to do this, we have to call gimple_could_trap_p instead of
gimple_assign_rhs_could_trap_p, as we have to also check that the LHS
of assign stmts does not trap.

	* common.opt (ftree-loop-if-convert-stores): New flag.
	* doc/invoke.texi (ftree-loop-if-convert-stores): Documented.
	* tree-if-conv.c (ifc_temp_var): Pass an extra parameter GSI.  Insert
	the created statement before GSI.
	(if_convertible_phi_p): Allow virtual phi nodes when
	flag_loop_if_convert_stores is set.
	(if_convertible_gimple_assign_stmt_p): Allow memory reads and writes
	Do not handle types that do not match is_gimple_reg_type.
	Remove loop and bb parameters.  Call gimple_could_trap_p instead of
	when flag_loop_if_convert_stores is set, as LHS can contain
	memory refs.
	(if_convertible_stmt_p): Remove loop and bb parameters.  Update calls
	to if_convertible_gimple_assign_stmt_p.
	(if_convertible_loop_p): Update call to if_convertible_stmt_p.
	(replace_phi_with_cond_gimple_assign_stmt): Renamed
	predicate_scalar_phi.  Do not handle virtual phi nodes.
	(ifconvert_phi_nodes): Renamed predicate_all_scalar_phis.
	Call predicate_scalar_phi.
	(insert_gimplified_predicates): Insert the gimplified predicate of a BB
	just after the labels for flag_loop_if_convert_stores, otherwise
	insert the predicate in the end of the BB.
	(predicate_mem_writes): New.
	(combine_blocks): Call predicate_all_scalar_phis.  When
	flag_loop_if_convert_stores is set, call predicate_mem_writes.
	(tree_if_conversion): Call mark_sym_for_renaming when
	flag_loop_if_convert_stores is set.
	(main_tree_if_conversion): Return TODO_update_ssa_only_virtuals when
	flag_loop_if_convert_stores is set.

	* gcc.dg/tree-ssa/ifc-4.c: New.
	* gcc.dg/tree-ssa/ifc-7.c: New.

From-SVN: r163530

1 files changed, 19 insertions, 1 deletions

diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index 148117d..7e069f0 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -383,7 +383,8 @@ Objective-C and Objective-C++ Dialects}.
 -fstrict-aliasing -fstrict-overflow -fthread-jumps -ftracer -ftree-bit-ccp @gol
 -ftree-builtin-call-dce -ftree-ccp -ftree-ch -ftree-copy-prop @gol
 -ftree-copyrename -ftree-dce -ftree-dominator-opts -ftree-dse @gol
--ftree-forwprop -ftree-fre -ftree-loop-if-convert -ftree-loop-im @gol
+-ftree-forwprop -ftree-fre -ftree-loop-if-convert @gol
+-ftree-loop-if-convert-memory-writes -ftree-loop-im @gol
 -ftree-phiprop -ftree-loop-distribution -ftree-loop-distribute-patterns @gol
 -ftree-loop-ivcanon -ftree-loop-linear -ftree-loop-optimize @gol
 -ftree-parallelize-loops=@var{n} -ftree-pre -ftree-pta -ftree-reassoc @gol
@@ -6931,6 +6932,23 @@ the innermost loops in order to improve the ability of the
 vectorization pass to handle these loops.  This is enabled by default
 if vectorization is enabled.
 
+@item -ftree-loop-if-convert-stores
+Attempt to also if-convert conditional jumps containing memory writes.
+This transformation can be unsafe for multi-threaded programs as it
+transforms conditional memory writes into unconditional memory writes.
+For example,
+@smallexample
+for (i = 0; i < N; i++)
+  if (cond)
+    A[i] = expr;
+@end smallexample
+would be transformed to
+@smallexample
+for (i = 0; i < N; i++)
+  A[i] = cond ? expr : A[i];
+@end smallexample
+potentially producing data races.
+
 @item -ftree-loop-distribution
 Perform loop distribution.  This flag can improve cache performance on
 big loop bodies and allow further loop optimizations, like