commit 951929b96784d2ffbc88fe9cdbd49bb9a4096032
Author: bin.cheng <chengbin.cb@alibaba-inc.com>
Date:   Tue Mar 23 15:44:36 2021 +0800

    Use programing order preserved RPO in loop distribution.
    
    Tree loop distribution uses RPO to build reduced dependence graph,
    it's important that RPO preserves the original programing order.
    Though it usually does so, when distributing loop nest, exit BB can
    be placed before some loop BBs while after loop header.  This patch
    fixes the issue by calling rev_post_order_and_mark_dfs_back_seme.
    
    gcc/ChangeLog:
    
            PR tree-optimization/98736
            * tree-loop-distribution.c
            * (loop_distribution::bb_top_order_init):
            Compute RPO with programing order preserved by calling function
            rev_post_order_and_mark_dfs_back_seme.
    
    gcc/testsuite/ChangeLog:
    
            PR tree-optimization/98736
            * gcc.c-torture/execute/pr98736.c: New test.

diff --git a/gcc/testsuite/gcc.c-torture/execute/pr98736.c b/gcc/testsuite/gcc.c-torture/execute/pr98736.c
new file mode 100644
index 00000000000..c066abcd86a
--- /dev/null
+++ b/gcc/testsuite/gcc.c-torture/execute/pr98736.c
@@ -0,0 +1,14 @@
+/* PR tree-optimization/98736 */
+
+int a[6];
+char b, c;
+int main() {
+  int d[4] = {0, 0, 0, 0};
+  for (c = 0; c <= 5; c++) {
+    for (b = 2; b != 0; b++)
+      a[c] = 8;
+    a[c] = d[3];
+  }
+  if (a[0] != 0)
+    __builtin_abort();
+}
diff --git a/gcc/tree-loop-distribution.c b/gcc/tree-loop-distribution.c
index 7ee19fc8677..583bb062b76 100644
--- a/gcc/tree-loop-distribution.c
+++ b/gcc/tree-loop-distribution.c
@@ -3152,11 +3152,19 @@ loop_distribution::distribute_loop (class loop *loop, vec<gimple *> stmts,
 void loop_distribution::bb_top_order_init (void)
 {
   int rpo_num;
-  int *rpo = XNEWVEC (int, last_basic_block_for_fn (cfun));
+  int *rpo = XNEWVEC (int, n_basic_blocks_for_fn (cfun) - NUM_FIXED_BLOCKS);
+  edge entry = single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun));
+  bitmap exit_bbs = BITMAP_ALLOC (NULL);
 
   bb_top_order_index = XNEWVEC (int, last_basic_block_for_fn (cfun));
   bb_top_order_index_size = last_basic_block_for_fn (cfun);
-  rpo_num = pre_and_rev_post_order_compute_fn (cfun, NULL, rpo, true);
+
+  entry->flags &= ~EDGE_DFS_BACK;
+  bitmap_set_bit (exit_bbs, EXIT_BLOCK);
+  rpo_num = rev_post_order_and_mark_dfs_back_seme (cfun, entry, exit_bbs, true,
+						   rpo, NULL);
+  BITMAP_FREE (exit_bbs);
+
   for (int i = 0; i < rpo_num; i++)
     bb_top_order_index[rpo[i]] = i;