i386.c (ix86_output_main_function_alignment_hack): New function.

	* i386.c (ix86_output_main_function_alignment_hack): New function.
	(TARGET_ASM_FUNCTION_PROLOGUE): Default to it.

	* flow.c (mark_dfs_back_edges): Move from loop_p ; mark back
	edges by EDGE_DFS_BACK flag.
	(dump_edge_info): Add dfs_back flag.
	* basic-block.h (EDGE_DFS_BACK): New constant.
	(mark_dfs_back_edges): Declare.
	* alias.c (loop_p): Remove.
	(mark_constant_function): Use mark_dfs_back_edges.

	* reg-stack.c (block_info_def): Add predecesors counter and stack_out.
	(reg_to_stack): Call mark_dfs_back_edges; count the predecesors.
	(compensate_edge): Break out from ...
	(convert_regs_1): ... here; do smart choosing of stack_out to copy.
	(convert_regs_2): Set block_done once block is really done;
	Do updating of the predecesors counts.

	* toplev.c (rest_of_compilation): Recompute block_for_insn
	before post-reload cfg_cleanup.
	* function.c (thread_prologue_epilogue_insns):
	Call set_block_for_new_insns when emitting prologue directly.

From-SVN: r44486

gcc/ChangeLog

+Mon Jul 30 22:16:08 CEST 2001  Jan Hubicka  <jh@suse.cz>
+
+	* i386.c (ix86_output_main_function_alignment_hack): New function.
+	(TARGET_ASM_FUNCTION_PROLOGUE): Default to it.
+
+	* flow.c (mark_dfs_back_edges): Move from loop_p ; mark back
+	edges by EDGE_DFS_BACK flag.
+	(dump_edge_info): Add dfs_back flag.
+	* basic-block.h (EDGE_DFS_BACK): New constant.
+	(mark_dfs_back_edges): Declare.
+	* alias.c (loop_p): Remove.
+	(mark_constant_function): Use mark_dfs_back_edges.
+
+	* reg-stack.c (block_info_def): Add predecesors counter and stack_out.
+	(reg_to_stack): Call mark_dfs_back_edges; count the predecesors.
+	(compensate_edge): Break out from ...
+	(convert_regs_1): ... here; do smart choosing of stack_out to copy.
+	(convert_regs_2): Set block_done once block is really done;
+	Do updating of the predecesors counts.
+
+	* toplev.c (rest_of_compilation): Recompute block_for_insn
+	before post-reload cfg_cleanup.
+	* function.c (thread_prologue_epilogue_insns):
+	Call set_block_for_new_insns when emitting prologue directly.
+
 2001-07-30  Andreas Jaeger  <aj@suse.de>
 
         * jump.c: Add prototype for mark_modified_reg.

gcc/alias.c

@@ -107,8 +107,6 @@ static int aliases_everything_p         PARAMS ((rtx));
 static int write_dependence_p           PARAMS ((rtx, rtx, int));
 static int nonlocal_mentioned_p         PARAMS ((rtx));
 
-static int loop_p                       PARAMS ((void));
-
 /* Set up all info needed to perform alias analysis on memory references.  */
 
 /* Returns the size in bytes of the mode of X.  */
@@ -2051,96 +2049,6 @@ nonlocal_mentioned_p (x)
   return 0;
 }
 
-/* Return non-zero if a loop (natural or otherwise) is present.
-   Inspired by Depth_First_Search_PP described in:
-
-     Advanced Compiler Design and Implementation
-     Steven Muchnick
-     Morgan Kaufmann, 1997
-
-   and heavily borrowed from flow_depth_first_order_compute.  */
-
-static int
-loop_p ()
-{
-  edge *stack;
-  int *pre;
-  int *post;
-  int sp;
-  int prenum = 1;
-  int postnum = 1;
-  sbitmap visited;
-
-  /* Allocate the preorder and postorder number arrays.  */
-  pre = (int *) xcalloc (n_basic_blocks, sizeof (int));
-  post = (int *) xcalloc (n_basic_blocks, sizeof (int));
-
-  /* Allocate stack for back-tracking up CFG.  */
-  stack = (edge *) xmalloc ((n_basic_blocks + 1) * sizeof (edge));
-  sp = 0;
-
-  /* Allocate bitmap to track nodes that have been visited.  */
-  visited = sbitmap_alloc (n_basic_blocks);
-
-  /* None of the nodes in the CFG have been visited yet.  */
-  sbitmap_zero (visited);
-
-  /* Push the first edge on to the stack.  */
-  stack[sp++] = ENTRY_BLOCK_PTR->succ;
-
-  while (sp)
-    {
-      edge e;
-      basic_block src;
-      basic_block dest;
-
-      /* Look at the edge on the top of the stack.  */
-      e = stack[sp - 1];
-      src = e->src;
-      dest = e->dest;
-
-      /* Check if the edge destination has been visited yet.  */
-      if (dest != EXIT_BLOCK_PTR && ! TEST_BIT (visited, dest->index))
-	{
-	  /* Mark that we have visited the destination.  */
-	  SET_BIT (visited, dest->index);
-
-	  pre[dest->index] = prenum++;
-
-	  if (dest->succ)
-	    {
-	      /* Since the DEST node has been visited for the first
-		 time, check its successors.  */
-	      stack[sp++] = dest->succ;
-	    }
-	  else
-	    post[dest->index] = postnum++;
-	}
-      else
-	{
-	  if (dest != EXIT_BLOCK_PTR && src != ENTRY_BLOCK_PTR
-	      && pre[src->index] >= pre[dest->index]
-	      && post[dest->index] == 0)
-	    break;
-
-	  if (! e->succ_next && src != ENTRY_BLOCK_PTR)
-	    post[src->index] = postnum++;
-
-	  if (e->succ_next)
-	    stack[sp - 1] = e->succ_next;
-	  else
-	    sp--;
-	}
-    }
-
-  free (pre);
-  free (post);
-  free (stack);
-  sbitmap_free (visited);
-
-  return sp;
-}
-
 /* Mark the function if it is constant.  */
 
 void
@@ -2157,7 +2065,7 @@ mark_constant_function ()
     return;
 
   /* A loop might not return which counts as a side effect.  */
-  if (loop_p ())
+  if (mark_dfs_back_edges ())
     return;
 
   nonlocal_mentioned = 0;

gcc/basic-block.h

@@ -145,6 +145,7 @@ typedef struct edge_def {
 #define EDGE_ABNORMAL_CALL	8
 #define EDGE_EH			16
 #define EDGE_FAKE		32
+#define EDGE_DFS_BACK		64
 
 #define EDGE_COMPLEX	(EDGE_ABNORMAL | EDGE_ABNORMAL_CALL | EDGE_EH)
 
@@ -650,6 +651,7 @@ extern void conflict_graph_print        PARAMS ((conflict_graph, FILE*));
 extern conflict_graph conflict_graph_compute 
                                         PARAMS ((regset,
 						 partition));
+extern bool mark_dfs_back_edges		PARAMS ((void));
 
 /* In dominance.c */
 

gcc/config/i386/i386.c

@@ -574,6 +574,7 @@ static HOST_WIDE_INT ix86_GOT_alias_set PARAMS ((void));
 static void ix86_adjust_counter PARAMS ((rtx, HOST_WIDE_INT));
 static rtx ix86_expand_aligntest PARAMS ((rtx, int));
 static void ix86_expand_strlensi_unroll_1 PARAMS ((rtx, rtx));
+static void ix86_output_main_function_alignment_hack PARAMS ((FILE *f, int));
 
 struct ix86_address
 {
@@ -634,6 +635,10 @@ static int ix86_comp_type_attributes PARAMS ((tree, tree));
 							  HOST_WIDE_INT));
 #  undef TARGET_ASM_FUNCTION_PROLOGUE
 #  define TARGET_ASM_FUNCTION_PROLOGUE ix86_osf_output_function_prologue
+#else
+#  undef TARGET_ASM_FUNCTION_PROLOGUE
+#  define TARGET_ASM_FUNCTION_PROLOGUE \
+   ix86_output_main_function_alignment_hack
 #endif
 
 #undef TARGET_ASM_OPEN_PAREN
@@ -10775,3 +10780,45 @@ ix86_memory_move_cost (mode, class, in)
 		* (int) GET_MODE_SIZE (mode) / 4);
     }
 }
+
+/* Most of current runtimes (Jul 2001) do not align stack properly when
+   entering main, so emit an wrapper to align stack before the real main
+   code is called.
+  
+   This can eventually go if we manage to fix the runtimes or teach gcc
+   to dynamically align stack in main automatically.
+
+   Adding check to configure is probably not good idea, as binarry can move
+   from one shared library to older.  */
+
+static void
+ix86_output_main_function_alignment_hack (file, size)
+     FILE *file;
+     int size ATTRIBUTE_UNUSED;
+{
+  rtx label;
+  char buf[256];
+  /* Check that we see main function with maximally 8 bytes of arguments.
+     if so, emit the hack to align stack for runtimes, where this constraint
+     is broken.  */
+  if (strcmp (cfun->name, "main"))
+    return;
+  if (cfun->pops_args || cfun->args_size > 12)
+    return;
+  if (PREFERRED_STACK_BOUNDARY <= 2)
+    return;
+  label = gen_label_rtx ();
+  fprintf (file, "\tpushl\t%%ebp\n");
+  fprintf (file, "\tmovl\t%%esp, %%ebp\n");
+  fprintf (file, "\tandl\t$0xfffffff0, %%esp\n");
+  fprintf (file, "\tpushl\t%%ebp\n");
+  fprintf (file, "\tpushl\t16(%%ebp)\n");
+  fprintf (file, "\tpushl\t12(%%ebp)\n");
+  fprintf (file, "\tpushl\t8(%%ebp)\n");
+  fprintf (file, "\tcall\t");
+  ASM_GENERATE_INTERNAL_LABEL (buf, "L", CODE_LABEL_NUMBER (label));
+  assemble_name (file, buf);
+  fprintf (file, "\n\tleave\n");
+  fprintf (file, "\tret\n");
+  ASM_OUTPUT_INTERNAL_LABEL (file, "L", CODE_LABEL_NUMBER (label));
+}

gcc/flow.c

@@ -1528,6 +1528,99 @@ mark_critical_edges ()
     }
 }
 
+/* Mark the back edges in DFS traversal.
+   Return non-zero if a loop (natural or otherwise) is present.
+   Inspired by Depth_First_Search_PP described in:
+
+     Advanced Compiler Design and Implementation
+     Steven Muchnick
+     Morgan Kaufmann, 1997
+
+   and heavily borrowed from flow_depth_first_order_compute.  */
+
+bool
+mark_dfs_back_edges ()
+{
+  edge *stack;
+  int *pre;
+  int *post;
+  int sp;
+  int prenum = 1;
+  int postnum = 1;
+  sbitmap visited;
+  bool found = false;
+
+  /* Allocate the preorder and postorder number arrays.  */
+  pre = (int *) xcalloc (n_basic_blocks, sizeof (int));
+  post = (int *) xcalloc (n_basic_blocks, sizeof (int));
+
+  /* Allocate stack for back-tracking up CFG.  */
+  stack = (edge *) xmalloc ((n_basic_blocks + 1) * sizeof (edge));
+  sp = 0;
+
+  /* Allocate bitmap to track nodes that have been visited.  */
+  visited = sbitmap_alloc (n_basic_blocks);
+
+  /* None of the nodes in the CFG have been visited yet.  */
+  sbitmap_zero (visited);
+
+  /* Push the first edge on to the stack.  */
+  stack[sp++] = ENTRY_BLOCK_PTR->succ;
+
+  while (sp)
+    {
+      edge e;
+      basic_block src;
+      basic_block dest;
+
+      /* Look at the edge on the top of the stack.  */
+      e = stack[sp - 1];
+      src = e->src;
+      dest = e->dest;
+      e->flags &= ~EDGE_DFS_BACK;
+
+      /* Check if the edge destination has been visited yet.  */
+      if (dest != EXIT_BLOCK_PTR && ! TEST_BIT (visited, dest->index))
+	{
+	  /* Mark that we have visited the destination.  */
+	  SET_BIT (visited, dest->index);
+
+	  pre[dest->index] = prenum++;
+
+	  if (dest->succ)
+	    {
+	      /* Since the DEST node has been visited for the first
+		 time, check its successors.  */
+	      stack[sp++] = dest->succ;
+	    }
+	  else
+	    post[dest->index] = postnum++;
+	}
+      else
+	{
+	  if (dest != EXIT_BLOCK_PTR && src != ENTRY_BLOCK_PTR
+	      && pre[src->index] >= pre[dest->index]
+	      && post[dest->index] == 0)
+	    e->flags |= EDGE_DFS_BACK, found = true;
+
+	  if (! e->succ_next && src != ENTRY_BLOCK_PTR)
+	    post[src->index] = postnum++;
+
+	  if (e->succ_next)
+	    stack[sp - 1] = e->succ_next;
+	  else
+	    sp--;
+	}
+    }
+
+  free (pre);
+  free (post);
+  free (stack);
+  sbitmap_free (visited);
+
+  return found;
+}
+
 /* Split a block BB after insn INSN creating a new fallthru edge.
    Return the new edge.  Note that to keep other parts of the compiler happy,
    this function renumbers all the basic blocks so that the new
@@ -7779,7 +7872,7 @@ dump_edge_info (file, e, do_succ)
   if (e->flags)
     {
       static const char * const bitnames[] = {
-	"fallthru", "crit", "ab", "abcall", "eh", "fake"
+	"fallthru", "crit", "ab", "abcall", "eh", "fake", "dfs_back"
       };
       int comma = 0;
       int i, flags = e->flags;

gcc/function.c

@@ -7226,7 +7226,8 @@ thread_prologue_and_epilogue_insns (f)
 	  inserted = 1;
 	}
       else
-	emit_insn_after (seq, f);
+	set_block_for_new_insns (emit_insn_after (seq, f),
+		       		 ENTRY_BLOCK_PTR->succ);
     }
 #endif
 

gcc/toplev.c

@@ -3473,11 +3473,13 @@ rest_of_compilation (decl)
   verify_flow_info ();
 #endif
 
+  compute_bb_for_insn (get_max_uid ());
+
   /* If optimizing, then go ahead and split insns now.  */
   if (optimize > 0)
     split_all_insns (0);
 
-  cleanup_cfg (0);
+  cleanup_cfg (optimize ? CLEANUP_EXPENSIVE : 0);
 
   /* On some machines, the prologue and epilogue code, or parts thereof,
      can be represented as RTL.  Doing so lets us schedule insns between
@@ -3485,8 +3487,6 @@ rest_of_compilation (decl)
      scheduling to operate in the epilogue.  */
   thread_prologue_and_epilogue_insns (insns);
 
-  compute_bb_for_insn (get_max_uid ());
-
   if (optimize)
     {
       cleanup_cfg (CLEANUP_EXPENSIVE | CLEANUP_CROSSJUMP);