From 5c1cd7b026d56fb3469da6d48a6ad6a42e140252 Mon Sep 17 00:00:00 2001
From: liyancheng <412998149@qq.com>
Date: Thu, 8 Jun 2023 16:08:32 +0800
Subject: [PATCH] [PGO] Bugfix for ICE with -freorder-blocks-and-partition and
 inline-asm goto

If asm goto has a crossing_edge, redirect it to a new bb and
insert an unconditional jump to orig dest from the new bb.
Avoid optimizing indirect jumps across different partitions in
pass ira.
---
 gcc/bb-reorder.c                              | 40 +++++++++++++++++++
 gcc/ira.c                                     | 20 ++++++++++
 .../rtl/aarch64/bb-reorder-and-inline-asm.c   | 17 ++++++++
 3 files changed, 77 insertions(+)
 create mode 100644 gcc/testsuite/gcc.dg/rtl/aarch64/bb-reorder-and-inline-asm.c

diff --git a/gcc/bb-reorder.c b/gcc/bb-reorder.c
index c635010c69f..6fccadb7b47 100644
--- a/gcc/bb-reorder.c
+++ b/gcc/bb-reorder.c
@@ -2116,6 +2116,46 @@ fix_crossing_conditional_branches (void)
 	{
 	  rtx_insn *old_jump = BB_END (cur_bb);
 
+	  /* If asm goto has a crossing_edge, redirect it to a new bb and
+	     insert an unconditional jump to orig dest from the new bb.
+	     In some architecture that conditional branches can not span
+	     all of memory, the unconditional jump will transfer into
+	     indirect jump in fix_crossing_unconditional_branches ().  */
+	  rtx asm_op = extract_asm_operands (PATTERN (old_jump));
+	  if (asm_op)
+	    {
+	      rtx_code_label *new_label = gen_label_rtx ();
+	      basic_block old_dest = crossing_edge->dest;
+
+	      /* Put the new label and a jump in the new basic block.  */
+	      rtx_insn *label = emit_label (new_label);
+	      rtx_code_label *old_label = block_label (old_dest);
+	      rtx_insn *jump = emit_jump_insn (targetm.gen_jump (old_label));
+	      JUMP_LABEL (jump) = old_label;
+
+	      /* Create the new basic block and put it in last position.  */
+	      basic_block last_bb = EXIT_BLOCK_PTR_FOR_FN (cfun)->prev_bb;
+	      basic_block new_bb = create_basic_block (label, jump, last_bb);
+	      new_bb->aux = last_bb->aux;
+	      last_bb->aux = new_bb;
+	      /* Make sure the new bb count is the same as crossing edge.  */
+	      new_bb->count = crossing_edge->count ();
+
+	      emit_barrier_after_bb (new_bb);
+
+	      new_edge = make_single_succ_edge (new_bb, old_dest, 0);
+	      new_edge->flags |= EDGE_CROSSING;
+	      /* Make sure the new basic block is in the same partition.  */
+	      BB_SET_PARTITION (new_bb, BB_PARTITION (old_dest));
+	      /* Make orignal edge jump to new dest.  */
+	      redirect_edge_and_branch (crossing_edge, new_bb);
+	      crossing_edge->flags &= ~EDGE_CROSSING;
+	      if (dump_file)
+		fprintf (dump_file, "\nRedirect asm goto from bb %d"
+				    " to bb %d.\n\n",
+				    old_dest->index, new_bb->index);
+	    }
+
 	  /* Check to make sure the jump instruction is a
 	     conditional jump.  */
 
diff --git a/gcc/ira.c b/gcc/ira.c
index 681ec2f46f9..24afaf8dc41 100644
--- a/gcc/ira.c
+++ b/gcc/ira.c
@@ -3918,6 +3918,7 @@ static void
 indirect_jump_optimize (void)
 {
   basic_block bb;
+  basic_block dest_bb = NULL;
   bool rebuild_p = false;
 
   FOR_EACH_BB_REVERSE_FN (bb, cfun)
@@ -3927,6 +3928,25 @@ indirect_jump_optimize (void)
 	  || find_reg_note (insn, REG_NON_LOCAL_GOTO, NULL_RTX))
 	continue;
 
+      /* In pass reorder_blocks_and_partition, if the architecture does
+	 not have unconditional branches that can span all of memory,
+	 convert crossing unconditional branches into indirect jumps.
+	 We should avoid optimizing indirect jumps across different
+	 partitions here.  */
+      if (flag_reorder_blocks_and_partition && !HAS_LONG_UNCOND_BRANCH)
+	{
+	  if (JUMP_LABEL (insn) && BLOCK_FOR_INSN (JUMP_LABEL (insn)))
+	    dest_bb = BLOCK_FOR_INSN (JUMP_LABEL (insn));
+
+	  if (dest_bb && BB_PARTITION (bb) != BB_PARTITION (dest_bb))
+	    {
+	      if (dump_file)
+		fprintf (dump_file, "\nSkip indirect_jump_optimize due to"
+				    " edge crossing different partition.\n");
+	      continue;
+	    }
+	}
+
       rtx x = pc_set (insn);
       if (!x || !REG_P (SET_SRC (x)))
 	continue;
diff --git a/gcc/testsuite/gcc.dg/rtl/aarch64/bb-reorder-and-inline-asm.c b/gcc/testsuite/gcc.dg/rtl/aarch64/bb-reorder-and-inline-asm.c
new file mode 100644
index 00000000000..fe960746d63
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/rtl/aarch64/bb-reorder-and-inline-asm.c
@@ -0,0 +1,17 @@
+/* { dg-do compile { target aarch64-*-* } } */
+/* { dg-options "-O2 -freorder-blocks-and-partition -save-temps -fdump-rtl-bbpart" } */
+
+int a;
+__attribute__((__cold__)) int b(char *);
+void d(void) {
+e:
+  for (; a;)
+    ;
+  b("");
+  asm goto("" : : : : c);
+  asm("");
+c:
+  goto e;
+}
+
+/* { dg-final { scan-assembler-times {d.cold:} 1 } }  */
\ No newline at end of file
-- 
Gitee