re PR target/89578 (5% runtime regression for 481.wrf at -Ofast -flto)

2019-03-08  Richard Biener  <rguenther@suse.de>

	PR middle-end/89578
	* cfgloop.h (struct loop): Add owned_clique field.
	* cfgloopmanip.c (copy_loop_info): Copy it.
	* tree-cfg.c (gimple_duplicate_bb): Do not remap owned_clique
	cliques.
	* tree-inline.c (copy_loops): Remap owned_clique.
	* lto-streamer-in.c (input_cfg): Stream owned_clique.
	* lto-streamer-out.c (output_cfg): Likewise.

From-SVN: r269484

7 files changed, 25 insertions, 1 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 72a1c3f..b46c5af 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,14 @@
+2019-03-08  Richard Biener  <rguenther@suse.de>
+
+	PR middle-end/89578
+	* cfgloop.h (struct loop): Add owned_clique field.
+	* cfgloopmanip.c (copy_loop_info): Copy it.
+	* tree-cfg.c (gimple_duplicate_bb): Do not remap owned_clique
+	cliques.
+	* tree-inline.c (copy_loops): Remap owned_clique.
+	* lto-streamer-in.c (input_cfg): Stream owned_clique.
+	* lto-streamer-out.c (output_cfg): Likewise.
+
 2019-03-08  Jakub Jelinek  <jakub@redhat.com>
 
 	PR target/80190
diff --git a/gcc/cfgloop.h b/gcc/cfgloop.h
index 2e93af3..e82cd7a 100644
--- a/gcc/cfgloop.h
+++ b/gcc/cfgloop.h
@@ -227,6 +227,10 @@ struct GTY ((chain_next ("%h.next"))) loop {
      Other values means unroll with the given unrolling factor.  */
   unsigned short unroll;
 
+  /* If this loop was inlined the main clique of the callee which does
+     not need remapping when copying the loop body.  */
+  unsigned short owned_clique;
+
   /* For SIMD loops, this is a unique identifier of the loop, referenced
      by IFN_GOMP_SIMD_VF, IFN_GOMP_SIMD_LANE and IFN_GOMP_SIMD_LAST_LANE
      builtins.  */
diff --git a/gcc/cfgloopmanip.c b/gcc/cfgloopmanip.c
index 7eb587a..bfee48e 100644
--- a/gcc/cfgloopmanip.c
+++ b/gcc/cfgloopmanip.c
@@ -1024,6 +1024,7 @@ copy_loop_info (struct loop *loop, struct loop *target)
   target->force_vectorize = loop->force_vectorize;
   target->in_oacc_kernels_region = loop->in_oacc_kernels_region;
   target->unroll = loop->unroll;
+  target->owned_clique = loop->owned_clique;
 }
 
 /* Copies copy of LOOP as subloop of TARGET loop, placing newly
diff --git a/gcc/lto-streamer-in.c b/gcc/lto-streamer-in.c
index 6d78e66..7727b9b 100644
--- a/gcc/lto-streamer-in.c
+++ b/gcc/lto-streamer-in.c
@@ -826,6 +826,7 @@ input_cfg (struct lto_input_block *ib, struct data_in *data_in,
       /* Read OMP SIMD related info.  */
       loop->safelen = streamer_read_hwi (ib);
       loop->unroll = streamer_read_hwi (ib);
+      loop->owned_clique = streamer_read_hwi (ib);
       loop->dont_vectorize = streamer_read_hwi (ib);
       loop->force_vectorize = streamer_read_hwi (ib);
       loop->simduid = stream_read_tree (ib, data_in);
diff --git a/gcc/lto-streamer-out.c b/gcc/lto-streamer-out.c
index a72016a..b6e395b 100644
--- a/gcc/lto-streamer-out.c
+++ b/gcc/lto-streamer-out.c
@@ -1938,6 +1938,7 @@ output_cfg (struct output_block *ob, struct function *fn)
       /* Write OMP SIMD related info.  */
       streamer_write_hwi (ob, loop->safelen);
       streamer_write_hwi (ob, loop->unroll);
+      streamer_write_hwi (ob, loop->owned_clique);
       streamer_write_hwi (ob, loop->dont_vectorize);
       streamer_write_hwi (ob, loop->force_vectorize);
       stream_write_tree (ob, loop->simduid, true);
diff --git a/gcc/tree-cfg.c b/gcc/tree-cfg.c
index 088fc7b..f433efc 100644
--- a/gcc/tree-cfg.c
+++ b/gcc/tree-cfg.c
@@ -6244,7 +6244,8 @@ gimple_duplicate_bb (basic_block bb, copy_bb_data *id)
 	      op = TREE_OPERAND (op, 0);
 	    if ((TREE_CODE (op) == MEM_REF
 		 || TREE_CODE (op) == TARGET_MEM_REF)
-		&& MR_DEPENDENCE_CLIQUE (op) > 1)
+		&& MR_DEPENDENCE_CLIQUE (op) > 1
+		&& MR_DEPENDENCE_CLIQUE (op) != bb->loop_father->owned_clique)
 	      {
 		if (!id->dependence_map)
 		  id->dependence_map = new hash_map<dependence_hash,
diff --git a/gcc/tree-inline.c b/gcc/tree-inline.c
index d3e53d2..956e39f 100644
--- a/gcc/tree-inline.c
+++ b/gcc/tree-inline.c
@@ -2670,6 +2670,11 @@ copy_loops (copy_body_data *id,
 	    cfun->has_unroll = true;
 	  if (dest_loop->force_vectorize)
 	    cfun->has_force_vectorize_loops = true;
+	  if (id->src_cfun->last_clique != 0)
+	    dest_loop->owned_clique
+	      = remap_dependence_clique (id,
+					 src_loop->owned_clique
+					 ? src_loop->owned_clique : 1);
 
 	  /* Finally place it into the loop array and the loop tree.  */
 	  place_new_loop (cfun, dest_loop);