This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tree-ssa] Gimplifying Java


On Thu, 2003-06-12 at 17:21, Andrew MacLeod wrote:
> On Thu, 2003-06-12 at 17:14, Jeff Sturm wrote:

> 
> I guess the general rule would probably be that an edge needs to be
> split if the src block has more than 1 *normal* edge, or the dest block
> has more than 1 *normal* predecessor.
> 
> Right now it just counts edges. The edge from 9 to 10 is marked abnormal
> right? so I can simple insert right at the end of block 9.
> 
> Let me try to handle that right now.. hold on..


Try this. Im running tests right now, so I dont know if it introduces
new failures or not, but I did notice it fixed one C++ testcase :-)
(FAIL: g++.dg/eh/cleanup1.C works with this patch :-)

TRy it an see if it helps.

Andrew

	* tree-cfg.c (bsi_commit_first_edge_insert): Only consider non-abnormal
	edges when determining whether an edge needs to be split.


Index: tree-cfg.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/Attic/tree-cfg.c,v
retrieving revision 1.1.4.105
diff -c -p -r1.1.4.105 tree-cfg.c
*** tree-cfg.c	12 Jun 2003 04:36:20 -0000	1.1.4.105
--- tree-cfg.c	12 Jun 2003 21:27:00 -0000
*************** bsi_commit_first_edge_insert (edge e, tr
*** 3824,3845 ****
    basic_block src, dest, new_bb;
    block_stmt_iterator bsi, tmp;
    tree_stmt_iterator tsi;
!   int single_exit, single_entry;
    enum find_location_action location;
    tree first, last, inserted_stmt, parent;
    bb_ann_t ann;
  
    first = last = NULL_TREE;
    src = e->src;
    dest = e->dest;
  
!   single_exit = (src->succ->succ_next == NULL);
!   single_entry = (dest->pred->pred_next == NULL);
  
    /* If it is a single exit block, and it isn't the entry block, and the edge 
       is not abnormal, then insert at the end of the block, if we can.  */
       
!   if (single_exit 
        && src != ENTRY_BLOCK_PTR 
        && ((e->flags & EDGE_ABNORMAL)) == 0)
      {
--- 3824,3852 ----
    basic_block src, dest, new_bb;
    block_stmt_iterator bsi, tmp;
    tree_stmt_iterator tsi;
!   int num_exit, num_entry;
    enum find_location_action location;
    tree first, last, inserted_stmt, parent;
    bb_ann_t ann;
+   edge e2;
  
    first = last = NULL_TREE;
    src = e->src;
    dest = e->dest;
  
!   num_exit = num_entry = 0;
!   for (e2 = src->succ; e2; e2 = e2->succ_next)
!     if (!(e->flags & EDGE_ABNORMAL))
!       num_exit++;
! 
!   for (e2 = src->pred; e2; e2 = e2->pred_next)
!     if (!(e->flags & EDGE_ABNORMAL))
!       num_entry++;
  
    /* If it is a single exit block, and it isn't the entry block, and the edge 
       is not abnormal, then insert at the end of the block, if we can.  */
       
!   if (num_exit == 1
        && src != ENTRY_BLOCK_PTR 
        && ((e->flags & EDGE_ABNORMAL)) == 0)
      {
*************** bsi_commit_first_edge_insert (edge e, tr
*** 3875,3881 ****
    /* If it is a single entry destination, and it isn't the exit block, the new
       stmt can be inserted at the beginning of the destination block.  */
  
!   if (single_entry && dest != EXIT_BLOCK_PTR)
      {
        bsi = bsi_start (dest);
        /* If it is an empty block, simply insert after this bsi, and the new stmt
--- 3882,3888 ----
    /* If it is a single entry destination, and it isn't the exit block, the new
       stmt can be inserted at the beginning of the destination block.  */
  
!   if (num_entry == 1 && dest != EXIT_BLOCK_PTR)
      {
        bsi = bsi_start (dest);
        /* If it is an empty block, simply insert after this bsi, and the new stmt


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]