gcc/0202-Backport-SME-Handle-epilogues-that-contain-jumps.patch

From 31433584b018cb2dc81e2366351a57bf5e1c4e44 Mon Sep 17 00:00:00 2001
From: Richard Sandiford <richard.sandiford@arm.com>
Date: Tue, 17 Oct 2023 23:45:43 +0100
Subject: [PATCH 103/157] [Backport][SME] Handle epilogues that contain jumps

Reference: https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=aeb3f0436f8ae84e593eda9641fe4e6fdf0afb3e

The prologue/epilogue pass allows the prologue sequence to contain
jumps.  The sequence is then partitioned into basic blocks using
find_many_sub_basic_blocks.

This patch treats epilogues in a similar way.  Since only one block
might need to be split, the patch (re)introduces a find_sub_basic_blocks
routine to handle a single block.

The new routine hard-codes the assumption that split_block will chain
the new block immediately after the original block.  The routine doesn't
try to replicate the fix for PR81030, since that was specific to
gimple->rtl expansion.

The patch is needed for follow-on aarch64 patches that add conditional
code to the epilogue.  The tests are part of those patches.

gcc/
	* cfgbuild.h (find_sub_basic_blocks): Declare.
	* cfgbuild.cc (update_profile_for_new_sub_basic_block): New function,
	split out from...
	(find_many_sub_basic_blocks): ...here.
	(find_sub_basic_blocks): New function.
	* function.cc (thread_prologue_and_epilogue_insns): Handle
	epilogues that contain jumps.
---
 gcc/cfgbuild.cc | 95 +++++++++++++++++++++++++++++++++----------------
 gcc/cfgbuild.h  |  1 +
 gcc/function.cc |  4 +++
 3 files changed, 70 insertions(+), 30 deletions(-)

diff --git a/gcc/cfgbuild.cc b/gcc/cfgbuild.cc
index 646a06614..58b865f29 100644
--- a/gcc/cfgbuild.cc
+++ b/gcc/cfgbuild.cc
@@ -693,6 +693,43 @@ compute_outgoing_frequencies (basic_block b)
     }
 }
 
+/* Update the profile information for BB, which was created by splitting
+   an RTL block that had a non-final jump.  */
+
+static void
+update_profile_for_new_sub_basic_block (basic_block bb)
+{
+  edge e;
+  edge_iterator ei;
+
+  bool initialized_src = false, uninitialized_src = false;
+  bb->count = profile_count::zero ();
+  FOR_EACH_EDGE (e, ei, bb->preds)
+    {
+      if (e->count ().initialized_p ())
+	{
+	  bb->count += e->count ();
+	  initialized_src = true;
+	}
+      else
+	uninitialized_src = true;
+    }
+  /* When some edges are missing with read profile, this is
+     most likely because RTL expansion introduced loop.
+     When profile is guessed we may have BB that is reachable
+     from unlikely path as well as from normal path.
+
+     TODO: We should handle loops created during BB expansion
+     correctly here.  For now we assume all those loop to cycle
+     precisely once.  */
+  if (!initialized_src
+      || (uninitialized_src
+	   && profile_status_for_fn (cfun) < PROFILE_GUESSED))
+    bb->count = profile_count::uninitialized ();
+
+  compute_outgoing_frequencies (bb);
+}
+
 /* Assume that some pass has inserted labels or control flow
    instructions within a basic block.  Split basic blocks as needed
    and create edges.  */
@@ -744,40 +781,15 @@ find_many_sub_basic_blocks (sbitmap blocks)
   if (profile_status_for_fn (cfun) != PROFILE_ABSENT)
     FOR_BB_BETWEEN (bb, min, max->next_bb, next_bb)
       {
-	edge e;
-	edge_iterator ei;
-
 	if (STATE (bb) == BLOCK_ORIGINAL)
 	  continue;
 	if (STATE (bb) == BLOCK_NEW)
 	  {
-	    bool initialized_src = false, uninitialized_src = false;
-	    bb->count = profile_count::zero ();
-	    FOR_EACH_EDGE (e, ei, bb->preds)
-	      {
-		if (e->count ().initialized_p ())
-		  {
-		    bb->count += e->count ();
-		    initialized_src = true;
-		  }
-		else
-		  uninitialized_src = true;
-	      }
-	    /* When some edges are missing with read profile, this is
-	       most likely because RTL expansion introduced loop.
-	       When profile is guessed we may have BB that is reachable
-	       from unlikely path as well as from normal path.
-
-	       TODO: We should handle loops created during BB expansion
-	       correctly here.  For now we assume all those loop to cycle
-	       precisely once.  */
-	    if (!initialized_src
-		|| (uninitialized_src
-		     && profile_status_for_fn (cfun) < PROFILE_GUESSED))
-	      bb->count = profile_count::uninitialized ();
+	    update_profile_for_new_sub_basic_block (bb);
+	    continue;
 	  }
- 	/* If nothing changed, there is no need to create new BBs.  */
-	else if (EDGE_COUNT (bb->succs) == n_succs[bb->index])
+	/* If nothing changed, there is no need to create new BBs.  */
+	if (EDGE_COUNT (bb->succs) == n_succs[bb->index])
 	  {
 	    /* In rare occassions RTL expansion might have mistakely assigned
 	       a probabilities different from what is in CFG.  This happens
@@ -788,10 +800,33 @@ find_many_sub_basic_blocks (sbitmap blocks)
 	      update_br_prob_note (bb);
 	    continue;
 	  }
-
 	compute_outgoing_frequencies (bb);
       }
 
   FOR_EACH_BB_FN (bb, cfun)
     SET_STATE (bb, 0);
 }
+
+/* Like find_many_sub_basic_blocks, but look only within BB.  */
+
+void
+find_sub_basic_blocks (basic_block bb)
+{
+  basic_block end_bb = bb->next_bb;
+  find_bb_boundaries (bb);
+  if (bb->next_bb == end_bb)
+    return;
+
+  /* Re-scan and wire in all edges.  This expects simple (conditional)
+     jumps at the end of each new basic blocks.  */
+  make_edges (bb, end_bb->prev_bb, 1);
+
+  /* Update branch probabilities.  Expect only (un)conditional jumps
+     to be created with only the forward edges.  */
+  if (profile_status_for_fn (cfun) != PROFILE_ABSENT)
+    {
+      compute_outgoing_frequencies (bb);
+      for (bb = bb->next_bb; bb != end_bb; bb = bb->next_bb)
+	update_profile_for_new_sub_basic_block (bb);
+    }
+}
diff --git a/gcc/cfgbuild.h b/gcc/cfgbuild.h
index 85145da7f..53543bb75 100644
--- a/gcc/cfgbuild.h
+++ b/gcc/cfgbuild.h
@@ -24,5 +24,6 @@ extern bool inside_basic_block_p (const rtx_insn *);
 extern bool control_flow_insn_p (const rtx_insn *);
 extern void rtl_make_eh_edge (sbitmap, basic_block, rtx);
 extern void find_many_sub_basic_blocks (sbitmap);
+extern void find_sub_basic_blocks (basic_block);
 
 #endif /* GCC_CFGBUILD_H */
diff --git a/gcc/function.cc b/gcc/function.cc
index ddab43ca4..f4fc211a0 100644
--- a/gcc/function.cc
+++ b/gcc/function.cc
@@ -6126,6 +6126,8 @@ thread_prologue_and_epilogue_insns (void)
 		  && returnjump_p (BB_END (e->src)))
 		e->flags &= ~EDGE_FALLTHRU;
 	    }
+
+	  find_sub_basic_blocks (BLOCK_FOR_INSN (epilogue_seq));
 	}
       else if (next_active_insn (BB_END (exit_fallthru_edge->src)))
 	{
@@ -6234,6 +6236,8 @@ thread_prologue_and_epilogue_insns (void)
 	  set_insn_locations (seq, epilogue_location);
 
 	  emit_insn_before (seq, insn);
+
+	  find_sub_basic_blocks (BLOCK_FOR_INSN (insn));
 	}
     }
 
-- 
2.33.0