[lto] Merge streamer hooks from pph branch. (issue4568043)

Richard Guenther rguenther@suse.de
Wed Jun 1 12:08:00 GMT 2011


On Tue, 31 May 2011, Diego Novillo wrote:

> 
> This patch merges the LTO streamer hooks from the pph branch.
> 
> These hooks are meant to separate streaming work that is specific to
> GIMPLE from other modules that may want to stream their own data
> structures.  In the PPH branch, we are using the streamer to save
> front end data structures, so there was a bunch of work and checks
> that made no sense in the front end.
> 
> All this module-specific work can be moved to the hooks defined in
> struct lto_streamer_hooks.
> 
> One of the concerns I had with the hooks is potential overhead when
> streaming because of the new indirect calls added by the patch.  This
> does not seem to be the case.
> 
> I did several timing runs using insn-emit.o (the biggest object file
> we generate in a bootstrap) and using an LTO link of cc1.  The times
> are slightly in favour of the hooks, actually (the patch removes some
> checks from the core tree pickling routines), but the difference is
> insignificant.
> 
> Over 5 runs on an idle system: Without this patch, we compile
> insn-emit.o in 41.56 secs (wall time), with the patch the time is
> 41.34 secs (wall time).
> 
> The other change in the patch is to reserve enough tags in enum
> LTO_tags to cover all the possible tree codes.  This exposes the bug
> that I fixed yesterday in the pph branch (the sign of the shifts in
> lto_output_int_in_range).  This means that we need to use
> output_record_start (LTO_null) instead of output_zero to output
> 0-delimiters, which increases the size of the output by less than 1%
> on average.

Can you split this fix out from this merge?

> Tested with LTO profiledbootstrap on x86_64.  OK for trunk?

Comments inline

> 
> Diego.
> 
> 
> 	* Makefile.in (cgraphunit.o): Add dependency on LTO_STREAMER_H.
> 	* cgraphunit.c: Include lto-streamer.h
> 	(cgraph_finalize_compilation_unit): Call gimple_streamer_hooks_init
> 	if LTO is enabled.
> 	* lto-streamer-in.c (unpack_value_fields): Do not check
> 	for TS_SSA_NAME, TS_STATEMENT_LIST and TS_OMP_CLAUSE.
> 	Call lto_streamer_hooks.unpack_value_fields if set.
> 	(lto_materialize_tree): For unhandled nodes, first try to
> 	call lto_streamer_hooks.alloc_tree, if it exists.
> 	(lto_input_ts_decl_common_tree_pointers): Move reading of
> 	DECL_INITIAL to gimple_streamer_read_tree.
> 	(lto_input_tree_pointers): Do not handle TS_SSA_NAME,
> 	TS_STATEMENT_LIST, TS_OMP_CLAUSE and TS_OPTIMIZATION.
> 	(lto_read_tree): Call lto_streamer_hooks.read_tree if
> 	set.
> 	Only register symbols in symtab if
> 	lto_streamer_hooks.register_decls_in_symtab_p is set.
> 	(gimple_streamer_read_tree): New.
> 	(lto_reader_init): Rename from lto_init_reader.
> 	Move initialization code to gimple_streamer_reader_init.
> 	Call lto_streamer_hooks.reader_init if set.
> 	(gimple_streamer_reader_init): New.
> 	* lto-streamer-out.c (pack_value_fields): Do not handle
> 	TS_SSA_NAME, TS_STATEMENT_LIST and TS_OMP_CLAUSE.
> 	Call lto_streamer_hooks.pack_value_fields if set.
> 	(lto_output_tree_ref): For tree nodes that are not
> 	normally indexable, call
> 	lto_streamer_hooks.indexable_with_decls_p before giving
> 	up.
> 	(lto_output_ts_decl_common_tree_pointers): Move handling
> 	for FUNCTION_DECL and TRANSLATION_UNIT_DECL to
> 	gimple_streamer_write_tree.
> 	Move assertion for NULL DECL_SAVED_TREEs to
> 	gimple_streamer_write_tree.
> 	(lto_output_ts_decl_with_vis_tree_pointers): Call
> 	output_record_start with LTO_null instead of output_zero.
> 	(lto_output_ts_binfo_tree_pointers): Likewise.
> 	(lto_output_tree): Likewise.
> 	(output_eh_try_list): Likewise.
> 	(output_eh_region): Likewise.
> 	(output_eh_lp): Likewise.
> 	(output_eh_regions): Likewise.
> 	(output_bb): Likewise.
> 	(output_function): Likewise.
> 	(output_unreferenced_globals): Likewise.
> 	(lto_output_tree_pointers): Do not handle TS_SSA_NAME,
> 	TS_STATEMENT_LIST, TS_OMP_CLAUSE and TS_OPTIMIZATION.
> 	(lto_output_tree_header): Call
> 	lto_streamer_hooks.is_streamable instead of
> 	lto_is_streamable.
> 	Call lto_streamer_hooks.output_tree_header if set.
> 	(lto_write_tree): Call lto_streamer_hooks.write_tree if
> 	set.
> 	(gimple_streamer_write_tree): New.
> 	(lto_output_tree): If
> 	lto_streamer_hooks.has_unique_integer_csts_p is set,
> 	lookup the constant in the streamer cache first.
> 	* lto-streamer.c (lto_streamer_cache_create): Call
> 	lto_streamer_hooks.preload_common_nodes instead of
> 	lto_preload_common_nodes.
> 	(lto_is_streamable): Move from lto-streamer.h
> 	(gimple_streamer_hooks_init): New.
> 	(streamer_hooks): New.
> 	(streamer_hooks_): Declare.
> 	(streamer_hooks_init): New.
> 	* lto-streamer.h (struct output_block): Forward declare.
> 	(struct lto_input_block): Likewise.
> 	(struct data_in): Likewise.
> 	(struct bitpack_d): Likewise.
> 	(enum LTO_tags): Reserve MAX_TREE_CODES
> 	to cover all front end codes as well.
> 	(struct lto_streamer_hooks): Declare.
> 	(gimple_streamer_hooks_init): Declare.
> 	(gimple_streamer_write_tree): Declare.
> 	(gimple_streamer_read_tree): Declare.
> 	(streamer_hooks): Declare.
> 	(streamer_hooks_init): Declare.
> 	(lto_tag_is_tree_code_p): Check max value against MAX_TREE_CODES.
> 	(lto_is_streamable): Move to lto-streamer.c
> 	(lto_output_int_in_range): Change << to >> when shifting VAL.
> 
> lto/ChangeLog
> 2011-05-31  Diego Novillo  <dnovillo@google.com>
> 
> 	* lto.c (lto_main): Call gimple_streamer_hooks_init.
> 
> diff --git a/gcc/cgraphunit.c b/gcc/cgraphunit.c
> index 834acb1..22fde67 100644
> --- a/gcc/cgraphunit.c
> +++ b/gcc/cgraphunit.c
> @@ -140,6 +140,7 @@ along with GCC; see the file COPYING3.  If not see
>  #include "plugin.h"
>  #include "ipa-inline.h"
>  #include "ipa-utils.h"
> +#include "lto-streamer.h"
>  
>  static void cgraph_expand_all_functions (void);
>  static void cgraph_mark_functions_to_output (void);
> @@ -1092,6 +1093,10 @@ cgraph_finalize_compilation_unit (void)
>  {
>    timevar_push (TV_CGRAPH);
>  
> +  /* If LTO is enabled, initialize the streamer hooks needed by GIMPLE.  */
> +  if (flag_lto)
> +    gimple_streamer_hooks_init ();

Ugh.  Isn't there a better entry for this?  Are you going to add

  if (flag_pph)
    init_hooks_some_other_way ();

here?  It looks it rather belongs to opts.c or toplev.c if the hooks
are really initialized dependent on compiler flags.

>    /* If we're here there's no current function anymore.  Some frontends
>       are lazy in clearing these.  */
>    current_function_decl = NULL;
> diff --git a/gcc/lto-streamer-in.c b/gcc/lto-streamer-in.c
> index 88966f2..801fe6f 100644
> --- a/gcc/lto-streamer-in.c
> +++ b/gcc/lto-streamer-in.c
> @@ -1833,6 +1833,7 @@ static void
>  unpack_value_fields (struct bitpack_d *bp, tree expr)
>  {
>    enum tree_code code;
> +  lto_streamer_hooks *h = streamer_hooks ();

A function to access a global ... we have lang_hooks and targetm,
so please simply use streamer_hooks as a variable.
streamer_hooks ()->preload_common_nodes (cache) looks super-ugly.

>    code = TREE_CODE (expr);
>  
> @@ -1864,26 +1865,11 @@ unpack_value_fields (struct bitpack_d *bp, tree expr)
>    if (CODE_CONTAINS_STRUCT (code, TS_BLOCK))
>      unpack_ts_block_value_fields (bp, expr);
>  
> -  if (CODE_CONTAINS_STRUCT (code, TS_SSA_NAME))
> -    {
> -      /* We only stream the version number of SSA names.  */
> -      gcc_unreachable ();
> -    }
> -
> -  if (CODE_CONTAINS_STRUCT (code, TS_STATEMENT_LIST))
> -    {
> -      /* This is only used by GENERIC.  */
> -      gcc_unreachable ();
> -    }
> -
> -  if (CODE_CONTAINS_STRUCT (code, TS_OMP_CLAUSE))
> -    {
> -      /* This is only used by High GIMPLE.  */
> -      gcc_unreachable ();
> -    }
> -
>    if (CODE_CONTAINS_STRUCT (code, TS_TRANSLATION_UNIT_DECL))
>      unpack_ts_translation_unit_decl_value_fields (bp, expr);
> +
> +  if (h->unpack_value_fields)
> +    h->unpack_value_fields (bp, expr);

I suppose the LTO implementation has a gcc_unreachable () for
the cases we do not handle here?

>  }
>  
>  
> @@ -1935,8 +1921,17 @@ lto_materialize_tree (struct lto_input_block *ib, struct data_in *data_in,
>      }
>    else
>      {
> -      /* All other nodes can be materialized with a raw make_node call.  */
> -      result = make_node (code);
> +      lto_streamer_hooks *h = streamer_hooks ();
> +
> +      /* For all other nodes, see if the streamer knows how to allocate
> +	 it.  */
> +      if (h->alloc_tree)
> +	result = h->alloc_tree (code, ib, data_in);
> +
> +      /* If the hook did not handle it, materialize the tree with a raw
> +	 make_node call.  */
> +      if (result == NULL_TREE)
> +	result = make_node (code);
>      }
>  
>  #ifdef LTO_STREAMER_DEBUG
> @@ -2031,12 +2026,8 @@ lto_input_ts_decl_common_tree_pointers (struct lto_input_block *ib,
>  {
>    DECL_SIZE (expr) = lto_input_tree (ib, data_in);
>    DECL_SIZE_UNIT (expr) = lto_input_tree (ib, data_in);
> -
> -  if (TREE_CODE (expr) != FUNCTION_DECL
> -      && TREE_CODE (expr) != TRANSLATION_UNIT_DECL)
> -    DECL_INITIAL (expr) = lto_input_tree (ib, data_in);
> -

Why move those?  DECL_INITIAL _is_ in decl_common.

>    DECL_ATTRIBUTES (expr) = lto_input_tree (ib, data_in);
> +
>    /* Do not stream DECL_ABSTRACT_ORIGIN.  We cannot handle debug information
>       for early inlining so drop it on the floor instead of ICEing in
>       dwarf2out.c.  */
> @@ -2433,38 +2424,15 @@ lto_input_tree_pointers (struct lto_input_block *ib, struct data_in *data_in,
>    if (CODE_CONTAINS_STRUCT (code, TS_EXP))
>      lto_input_ts_exp_tree_pointers (ib, data_in, expr);
>  
> -  if (CODE_CONTAINS_STRUCT (code, TS_SSA_NAME))
> -    {
> -      /* We only stream the version number of SSA names.  */
> -      gcc_unreachable ();
> -    }
> -
>    if (CODE_CONTAINS_STRUCT (code, TS_BLOCK))
>      lto_input_ts_block_tree_pointers (ib, data_in, expr);
>  
>    if (CODE_CONTAINS_STRUCT (code, TS_BINFO))
>      lto_input_ts_binfo_tree_pointers (ib, data_in, expr);
>  
> -  if (CODE_CONTAINS_STRUCT (code, TS_STATEMENT_LIST))
> -    {
> -      /* This should only appear in GENERIC.  */
> -      gcc_unreachable ();
> -    }
> -
>    if (CODE_CONTAINS_STRUCT (code, TS_CONSTRUCTOR))
>      lto_input_ts_constructor_tree_pointers (ib, data_in, expr);
>  
> -  if (CODE_CONTAINS_STRUCT (code, TS_OMP_CLAUSE))
> -    {
> -      /* This should only appear in High GIMPLE.  */
> -      gcc_unreachable ();
> -    }
> -
> -  if (CODE_CONTAINS_STRUCT (code, TS_OPTIMIZATION))
> -    {
> -      sorry ("optimization options not supported yet");
> -    }
> -

Where do those checks go?  Or do we simply lose them?

>    if (CODE_CONTAINS_STRUCT (code, TS_TARGET_OPTION))
>      lto_input_ts_target_option (ib, expr);
>  
> @@ -2655,20 +2623,29 @@ lto_read_tree (struct lto_input_block *ib, struct data_in *data_in,
>  	       enum LTO_tags tag)
>  {
>    tree result;
> +  lto_streamer_hooks *h = streamer_hooks ();
>  
>    result = lto_materialize_tree (ib, data_in, tag);
>  
>    /* Read all the pointer fields in RESULT.  */
>    lto_input_tree_pointers (ib, data_in, result);
>  
> +  /* Call back into the streaming module to read anything else it
> +     may need.  */
> +  if (h->read_tree)
> +    h->read_tree (ib, data_in, result);
> +
>    /* We should never try to instantiate an MD or NORMAL builtin here.  */
>    if (TREE_CODE (result) == FUNCTION_DECL)
>      gcc_assert (!lto_stream_as_builtin_p (result));
>  
> -  if (TREE_CODE (result) == VAR_DECL)
> -    lto_register_var_decl_in_symtab (data_in, result);
> -  else if (TREE_CODE (result) == FUNCTION_DECL && !DECL_BUILT_IN (result))
> -    lto_register_function_decl_in_symtab (data_in, result);
> +  if (h->register_decls_in_symtab_p)
> +    {
> +      if (TREE_CODE (result) == VAR_DECL)
> +	lto_register_var_decl_in_symtab (data_in, result);
> +      else if (TREE_CODE (result) == FUNCTION_DECL && !DECL_BUILT_IN (result))
> +	lto_register_function_decl_in_symtab (data_in, result);
> +    }

I would say we should defer symtab registering to uniquify_nodes time,
when we are sure we completely streamed in the tree we want to register
(then we don't have to deal with partially read trees in those functions).

Can you work on a preparatory patch for this please?  That would get
rid for the need of this hook and clean up things at the same time.

>  
>    /* end_marker = */ lto_input_1_unsigned (ib);
>  
> @@ -2682,6 +2659,22 @@ lto_read_tree (struct lto_input_block *ib, struct data_in *data_in,
>  }
>  
>  
> +/* LTO streamer hook for reading GIMPLE trees.  IB and DATA_IN are as in
> +   lto_read_tree.  EXPR is the tree was materialized by lto_read_tree and
> +   needs GIMPLE specific data to be filled in.  */
> +
> +void
> +gimple_streamer_read_tree (struct lto_input_block *ib,
> +			   struct data_in *data_in,
> +			   tree expr)
> +{
> +  if (DECL_P (expr)
> +      && TREE_CODE (expr) != FUNCTION_DECL
> +      && TREE_CODE (expr) != TRANSLATION_UNIT_DECL)
> +    DECL_INITIAL (expr) = lto_input_tree (ib, data_in);

What's wrong with doing this in the decl-common path?

> +}
> +
> +
>  /* Read and INTEGER_CST node from input block IB using the per-file
>     context in DATA_IN.  */
>  
> @@ -2763,16 +2756,23 @@ lto_input_tree (struct lto_input_block *ib, struct data_in *data_in)
>  /* Initialization for the LTO reader.  */
>  
>  void
> -lto_init_reader (void)
> +lto_reader_init (void)
>  {
>    lto_streamer_init ();
> -
> -  memset (&lto_stats, 0, sizeof (lto_stats));
> -  bitmap_obstack_initialize (NULL);
> -
>    file_name_hash_table = htab_create (37, hash_string_slot_node,
>  				      eq_string_slot_node, free);
> +  if (streamer_hooks ()->reader_init)
> +    streamer_hooks ()->reader_init ();
> +}
> +
>  
> +/* GIMPLE streamer hook for initializing the LTO reader.  */
> +
> +void
> +gimple_streamer_reader_init (void)
> +{
> +  memset (&lto_stats, 0, sizeof (lto_stats));
> +  bitmap_obstack_initialize (NULL);
>    gimple_register_cfg_hooks ();
>  }
>  
> diff --git a/gcc/lto-streamer-out.c b/gcc/lto-streamer-out.c
> index b3b81bd..90e3252 100644
> --- a/gcc/lto-streamer-out.c
> +++ b/gcc/lto-streamer-out.c
> @@ -558,6 +558,7 @@ static void
>  pack_value_fields (struct bitpack_d *bp, tree expr)
>  {
>    enum tree_code code;
> +  lto_streamer_hooks *h = streamer_hooks ();
>  
>    code = TREE_CODE (expr);
>  
> @@ -589,26 +590,11 @@ pack_value_fields (struct bitpack_d *bp, tree expr)
>    if (CODE_CONTAINS_STRUCT (code, TS_BLOCK))
>      pack_ts_block_value_fields (bp, expr);
>  
> -  if (CODE_CONTAINS_STRUCT (code, TS_SSA_NAME))
> -    {
> -      /* We only stream the version number of SSA names.  */
> -      gcc_unreachable ();
> -    }
> -
> -  if (CODE_CONTAINS_STRUCT (code, TS_STATEMENT_LIST))
> -    {
> -      /* This is only used by GENERIC.  */
> -      gcc_unreachable ();
> -    }
> -
> -  if (CODE_CONTAINS_STRUCT (code, TS_OMP_CLAUSE))
> -    {
> -      /* This is only used by High GIMPLE.  */
> -      gcc_unreachable ();
> -    }
> -
>    if (CODE_CONTAINS_STRUCT (code, TS_TRANSLATION_UNIT_DECL))
>      pack_ts_translation_unit_decl_value_fields (bp, expr);
> +
> +  if (h->pack_value_fields)
> +    h->pack_value_fields (bp, expr);
>  }
>  
>  
> @@ -772,9 +758,23 @@ lto_output_tree_ref (struct output_block *ob, tree expr)
>        break;
>  
>      default:
> -      /* No other node is indexable, so it should have been handled
> -	 by lto_output_tree.  */
> -      gcc_unreachable ();
> +      {
> +	lto_streamer_hooks *h = streamer_hooks ();
> +
> +	/* See if streamer hooks allows this node to be indexable with
> +	   VAR_DECLs.  */

with VAR_DECLs?  More like "similar to global decls."?

> +	if (h->indexable_with_decls_p && h->indexable_with_decls_p (expr))
> +	  {
> +	    output_record_start (ob, LTO_global_decl_ref);
> +	    lto_output_var_decl_index (ob->decl_state, ob->main_stream, expr);

Why hook it this way and not

             if (h->output_tree_ref
                 && h->output_tree_ref (...))
               break;
             gcc_unreachable (); 

I find the flag vs. function hook stuff somewhat odd.

> +	  }
> +	else
> +	  {
> +	    /* No other node is indexable, so it should have been
> +	      handled by lto_output_tree.  */
> +	    gcc_unreachable ();
> +	  }
> +      }
>      }
>  }
>  
> @@ -883,27 +883,10 @@ lto_output_ts_decl_common_tree_pointers (struct output_block *ob, tree expr,
>    lto_output_tree_or_ref (ob, DECL_SIZE (expr), ref_p);
>    lto_output_tree_or_ref (ob, DECL_SIZE_UNIT (expr), ref_p);
>  
> -  if (TREE_CODE (expr) != FUNCTION_DECL
> -      && TREE_CODE (expr) != TRANSLATION_UNIT_DECL)
> -    {
> -      tree initial = DECL_INITIAL (expr);
> -      if (TREE_CODE (expr) == VAR_DECL
> -	  && (TREE_STATIC (expr) || DECL_EXTERNAL (expr))
> -	  && initial)
> -	{
> -	  lto_varpool_encoder_t varpool_encoder = ob->decl_state->varpool_node_encoder;
> -	  struct varpool_node *vnode = varpool_get_node (expr);
> -	  if (!vnode)
> -	    initial = error_mark_node;
> -	  else if (!lto_varpool_encoder_encode_initializer_p (varpool_encoder,
> -							      vnode))
> -	    initial = NULL;
> -	}
> -    
> -      lto_output_tree_or_ref (ob, initial, ref_p);
> -    }
> +  /* Do not stream DECL_INITIAL.  */

That's a gross comment ;)  We _do_ stream it, but now in a hook.
I suppose all the streamer functions should lose their lto_ prefix -
reading them with such comments will get confusing ...

>  
>    lto_output_tree_or_ref (ob, DECL_ATTRIBUTES (expr), ref_p);
> +
>    /* Do not stream DECL_ABSTRACT_ORIGIN.  We cannot handle debug information
>       for early inlining so drop it on the floor instead of ICEing in
>       dwarf2out.c.  */
> @@ -931,11 +914,6 @@ lto_output_ts_decl_non_common_tree_pointers (struct output_block *ob,
>  {
>    if (TREE_CODE (expr) == FUNCTION_DECL)
>      {
> -      /* DECL_SAVED_TREE holds the GENERIC representation for DECL.
> -	 At this point, it should not exist.  Either because it was
> -	 converted to gimple or because DECL didn't have a GENERIC
> -	 representation in this TU.  */
> -      gcc_assert (DECL_SAVED_TREE (expr) == NULL_TREE);
>        lto_output_tree_or_ref (ob, DECL_ARGUMENTS (expr), ref_p);
>        lto_output_tree_or_ref (ob, DECL_RESULT (expr), ref_p);
>      }
> @@ -955,7 +933,7 @@ lto_output_ts_decl_with_vis_tree_pointers (struct output_block *ob, tree expr,
>    if (DECL_ASSEMBLER_NAME_SET_P (expr))
>      lto_output_tree_or_ref (ob, DECL_ASSEMBLER_NAME (expr), ref_p);
>    else
> -    output_zero (ob);
> +    output_record_start (ob, LTO_null);
>  
>    lto_output_tree_or_ref (ob, DECL_SECTION_NAME (expr), ref_p);
>    lto_output_tree_or_ref (ob, DECL_COMDAT_GROUP (expr), ref_p);
> @@ -1136,7 +1114,7 @@ lto_output_ts_binfo_tree_pointers (struct output_block *ob, tree expr,
>       is needed to build the empty BINFO node on the reader side.  */
>    FOR_EACH_VEC_ELT (tree, BINFO_BASE_BINFOS (expr), i, t)
>      lto_output_tree_or_ref (ob, t, ref_p);
> -  output_zero (ob);
> +  output_record_start (ob, LTO_null);
>  
>    lto_output_tree_or_ref (ob, BINFO_OFFSET (expr), ref_p);
>    lto_output_tree_or_ref (ob, BINFO_VTABLE (expr), ref_p);
> @@ -1256,12 +1234,6 @@ lto_output_tree_pointers (struct output_block *ob, tree expr, bool ref_p)
>    if (CODE_CONTAINS_STRUCT (code, TS_EXP))
>      lto_output_ts_exp_tree_pointers (ob, expr, ref_p);
>  
> -  if (CODE_CONTAINS_STRUCT (code, TS_SSA_NAME))
> -    {
> -      /* We only stream the version number of SSA names.  */
> -      gcc_unreachable ();
> -    }
> -
>    if (CODE_CONTAINS_STRUCT (code, TS_BLOCK))
>      lto_output_ts_block_tree_pointers (ob, expr, ref_p);
>  
> @@ -1271,21 +1243,6 @@ lto_output_tree_pointers (struct output_block *ob, tree expr, bool ref_p)
>    if (CODE_CONTAINS_STRUCT (code, TS_CONSTRUCTOR))
>      lto_output_ts_constructor_tree_pointers (ob, expr, ref_p);
>  
> -  if (CODE_CONTAINS_STRUCT (code, TS_STATEMENT_LIST))
> -    {
> -      /* This should only appear in GENERIC.  */
> -      gcc_unreachable ();
> -    }
> -
> -  if (CODE_CONTAINS_STRUCT (code, TS_OMP_CLAUSE))
> -    {
> -      /* This should only appear in High GIMPLE.  */
> -      gcc_unreachable ();
> -    }
> -
> -  if (CODE_CONTAINS_STRUCT (code, TS_OPTIMIZATION))
> -    sorry ("gimple bytecode streams do not support the optimization attribute");
> -
>    if (CODE_CONTAINS_STRUCT (code, TS_TARGET_OPTION))
>      lto_output_ts_target_option (ob, expr);
>  
> @@ -1304,18 +1261,20 @@ lto_output_tree_header (struct output_block *ob, tree expr)
>  {
>    enum LTO_tags tag;
>    enum tree_code code;
> +  lto_streamer_hooks *h = streamer_hooks ();
>  
>    /* We should not see any non-GIMPLE tree nodes here.  */
>    code = TREE_CODE (expr);
> -  if (!lto_is_streamable (expr))
> -    internal_error ("tree code %qs is not supported in gimple streams",
> -		    tree_code_name[code]);
> +  if (h->is_streamable && !h->is_streamable (expr))
> +    internal_error ("tree code %qs is not supported in %s streams",
> +		    h->name, tree_code_name[code]);

I'd say this hook should always exist.

>    /* The header of a tree node consists of its tag, the size of
>       the node, and any other information needed to instantiate
>       EXPR on the reading side (such as the number of slots in
>       variable sized nodes).  */
>    tag = lto_tree_code_to_tag (code);
> +  gcc_assert ((unsigned) tag < (unsigned) LTO_NUM_TAGS);

Doesn't Honzas streaming this as enum already assert this?

>    output_record_start (ob, tag);
>  
>    /* The following will cause bootstrap miscomparisons.  Enable with care.  */
> @@ -1338,6 +1297,11 @@ lto_output_tree_header (struct output_block *ob, tree expr)
>      output_sleb128 (ob, TREE_VEC_LENGTH (expr));
>    else if (CODE_CONTAINS_STRUCT (code, TS_BINFO))
>      output_uleb128 (ob, BINFO_N_BASE_BINFOS (expr));
> +
> +  /* Allow the streamer to write any streamer-specific information
> +     needed to instantiate the node when reading.  */
> +  if (h->output_tree_header)
> +    h->output_tree_header (ob, expr);
>  }
>  
>  
> @@ -1385,6 +1349,7 @@ static void
>  lto_write_tree (struct output_block *ob, tree expr, bool ref_p)
>  {
>    struct bitpack_d bp;
> +  lto_streamer_hooks *h = streamer_hooks ();
>  
>    /* Write the header, containing everything needed to materialize
>       EXPR on the reading side.  */
> @@ -1399,11 +1364,58 @@ lto_write_tree (struct output_block *ob, tree expr, bool ref_p)
>    /* Write all the pointer fields in EXPR.  */
>    lto_output_tree_pointers (ob, expr, ref_p);
>  
> +  /* Call back into the streaming module to see if it needs to write
> +     anything that was not written by the common streamer.  */
> +  if (h->write_tree)
> +    h->write_tree (ob, expr, ref_p);
> +
>    /* Mark the end of EXPR.  */
>    output_zero (ob);
>  }
>  
>  
> +/* GIMPLE hook for writing GIMPLE-specific parts of trees.  OB, EXPR
> +   and REF_P are as in lto_write_tree.  */
> +
> +void
> +gimple_streamer_write_tree (struct output_block *ob, tree expr, bool ref_p)
> +{
> +  if (TREE_CODE (expr) == FUNCTION_DECL)
> +    {
> +      /* DECL_SAVED_TREE holds the GENERIC representation for DECL.
> +	 At this point, it should not exist.  Either because it was
> +	 converted to gimple or because DECL didn't have a GENERIC
> +	 representation in this TU.  */
> +      gcc_assert (DECL_SAVED_TREE (expr) == NULL_TREE);
> +    }

I think we can drop this check.

> +  if (DECL_P (expr)
> +      && TREE_CODE (expr) != FUNCTION_DECL
> +      && TREE_CODE (expr) != TRANSLATION_UNIT_DECL)
> +    {
> +      /* Handle DECL_INITIAL for symbols.  */
> +      tree initial = DECL_INITIAL (expr);
> +      if (TREE_CODE (expr) == VAR_DECL
> +	  && (TREE_STATIC (expr) || DECL_EXTERNAL (expr))
> +	  && initial)
> +	{
> +	  lto_varpool_encoder_t varpool_encoder;
> +	  struct varpool_node *vnode;
> +
> +	  varpool_encoder = ob->decl_state->varpool_node_encoder;
> +	  vnode = varpool_get_node (expr);
> +	  if (!vnode)
> +	    initial = error_mark_node;
> +	  else if (!lto_varpool_encoder_encode_initializer_p (varpool_encoder,
> +							      vnode))
> +	    initial = NULL;
> +	}
> +
> +      lto_output_tree_or_ref (ob, initial, ref_p);
> +    }
> +}
> +
> +
>  /* Emit the integer constant CST to output block OB.  If REF_P is true,
>     CST's type will be emitted as a reference.  */
>  
> @@ -1430,7 +1442,7 @@ lto_output_tree (struct output_block *ob, tree expr, bool ref_p)
>  
>    if (expr == NULL_TREE)
>      {
> -      output_zero (ob);
> +      output_record_start (ob, LTO_null);
>        return;
>      }
>  
> @@ -1438,8 +1450,27 @@ lto_output_tree (struct output_block *ob, tree expr, bool ref_p)
>       to be materialized by the reader (to implement TYPE_CACHED_VALUES).  */
>    if (TREE_CODE (expr) == INTEGER_CST)
>      {
> -      lto_output_integer_cst (ob, expr, ref_p);
> -      return;
> +      bool is_special;
> +
> +     /* There are some constants that are special to the streamer
> +	(e.g., void_zero_node, truthvalue_false_node).
> +	These constants cannot be rematerialized with
> +	build_int_cst_wide because they may actually lack a type (like
> +	void_zero_node) and they need to be pointer-identical to trees
> +	materialized by the compiler tables like global_trees or
> +	c_global_trees.
> +
> +	If the streamer told us that it has special constants, they
> +	will be preloaded in the streamer cache.  If we find a match,
> +	then stream the constant as a reference so the reader can
> +	re-materialize it from the cache.  */
> +      is_special = streamer_hooks ()->has_unique_integer_csts_p
> +		   && lto_streamer_cache_lookup (ob->writer_cache, expr, NULL);
> +      if (!is_special)
> +	{
> +	  lto_output_integer_cst (ob, expr, ref_p);
> +	  return;
> +	}

???  We should not arrive here for such global trees.  Please do not
merge this part of the patch as part of the hook introducing (keep
patches simple, make them do a single thing ...)

>      }
>  
>    existed_p = lto_streamer_cache_insert (ob->writer_cache, expr, &ix);
> @@ -1486,7 +1517,7 @@ output_eh_try_list (struct output_block *ob, eh_catch first)
>        lto_output_tree_ref (ob, n->label);
>      }
>  
> -  output_zero (ob);
> +  output_record_start (ob, LTO_null);
>  }
>  
>  
> @@ -1501,7 +1532,7 @@ output_eh_region (struct output_block *ob, eh_region r)
>  
>    if (r == NULL)
>      {
> -      output_zero (ob);
> +      output_record_start (ob, LTO_null);
>        return;
>      }
>  
> @@ -1564,7 +1595,7 @@ output_eh_lp (struct output_block *ob, eh_landing_pad lp)
>  {
>    if (lp == NULL)
>      {
> -      output_zero (ob);
> +      output_record_start (ob, LTO_null);
>        return;
>      }
>  
> @@ -1633,9 +1664,9 @@ output_eh_regions (struct output_block *ob, struct function *fn)
>  	}
>      }
>  
> -  /* The 0 either terminates the record or indicates that there are no
> -     eh_records at all.  */
> -  output_zero (ob);
> +  /* The LTO_null either terminates the record or indicates that there
> +     are no eh_records at all.  */
> +  output_record_start (ob, LTO_null);
>  }
>  
>  
> @@ -1880,10 +1911,10 @@ output_bb (struct output_block *ob, basic_block bb, struct function *fn)
>  	      output_sleb128 (ob, region);
>  	    }
>  	  else
> -	    output_zero (ob);
> +	    output_record_start (ob, LTO_null);
>  	}
>  
> -      output_zero (ob);
> +      output_record_start (ob, LTO_null);
>  
>        for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
>  	{
> @@ -1896,7 +1927,7 @@ output_bb (struct output_block *ob, basic_block bb, struct function *fn)
>  	    output_phi (ob, phi);
>  	}
>  
> -      output_zero (ob);
> +      output_record_start (ob, LTO_null);
>      }
>  }
>  
> @@ -2053,7 +2084,7 @@ output_function (struct cgraph_node *node)
>      output_bb (ob, bb, fn);
>  
>    /* The terminator for this function.  */
> -  output_zero (ob);
> +  output_record_start (ob, LTO_null);
>  
>    output_cfg (ob, fn);
>  
> @@ -2167,7 +2198,7 @@ output_unreferenced_globals (cgraph_node_set set, varpool_node_set vset)
>        }
>    symbol_alias_set_destroy (defined);
>  
> -  output_zero (ob);
> +  output_record_start (ob, LTO_null);
>  
>    produce_asm (ob, NULL);
>    destroy_output_block (ob);
> @@ -2238,6 +2269,8 @@ static void
>  lto_writer_init (void)
>  {
>    lto_streamer_init ();
> +  if (streamer_hooks ()->writer_init)
> +    streamer_hooks ()->writer_init ();

This hook should always exist.  Why is this called in a context with
lto_*?

>  }
>  
>  
> diff --git a/gcc/lto-streamer.c b/gcc/lto-streamer.c
> index abccd10..4281bd9 100644
> --- a/gcc/lto-streamer.c
> +++ b/gcc/lto-streamer.c
> @@ -579,7 +579,7 @@ lto_streamer_cache_create (void)
>    /* Load all the well-known tree nodes that are always created by
>       the compiler on startup.  This prevents writing them out
>       unnecessarily.  */
> -  lto_preload_common_nodes (cache);
> +  streamer_hooks ()->preload_common_nodes (cache);
>  
>    return cache;
>  }
> @@ -724,3 +724,70 @@ lto_check_version (int major, int minor)
>  		 major, minor,
>  		 LTO_major_version, LTO_minor_version);
>  }
> +
> +
> +/* Return true if EXPR is a tree node that can be written to disk.  */
> +
> +static inline bool
> +lto_is_streamable (tree expr)
> +{
> +  enum tree_code code = TREE_CODE (expr);
> +
> +  /* Notice that we reject SSA_NAMEs as well.  We only emit the SSA
> +     name version in lto_output_tree_ref (see output_ssa_names).  */
> +  return !is_lang_specific (expr)
> +	 && code != SSA_NAME
> +	 && code != CALL_EXPR
> +	 && code != LANG_TYPE
> +	 && code != MODIFY_EXPR
> +	 && code != INIT_EXPR
> +	 && code != TARGET_EXPR
> +	 && code != BIND_EXPR
> +	 && code != WITH_CLEANUP_EXPR
> +	 && code != STATEMENT_LIST
> +	 && code != OMP_CLAUSE
> +	 && code != OPTIMIZATION_NODE
> +	 && (code == CASE_LABEL_EXPR
> +	     || code == DECL_EXPR
> +	     || TREE_CODE_CLASS (code) != tcc_statement);
> +}

This change (with the removal of the cases with the unreachable()s)
could be made separately (directly calling it instead of hooking it).

> +
> +/* Initialize all the streamer hooks used for streaming GIMPLE.  */
> +
> +void
> +gimple_streamer_hooks_init (void)

It's more like lto_streamer_hooks_init, no?  You are basically
turning lto_streamer_* to tree_streamer_* and make lto_streamer_*
tree_streamer_* + lto_streamer_hooks, no?  It would be really
nice to also reflect this in the function naming (and eventually
moving the tree_streamer_* stuff to separate files).  It really
confuses me ;)

> +{
> +  lto_streamer_hooks *h = streamer_hooks_init ();

Heh, and the type is called lto_streamer_hooks ...

> +  h->name = "gimple";
> +  h->reader_init = gimple_streamer_reader_init;
> +  h->writer_init = NULL;
> +  h->preload_common_nodes = lto_preload_common_nodes;
> +  h->is_streamable = lto_is_streamable;
> +  h->write_tree = gimple_streamer_write_tree;
> +  h->read_tree = gimple_streamer_read_tree;
> +  h->register_decls_in_symtab_p = true;
> +}
> +
> +
> +/* Return the current set of streamer hooks.  Note that only one set
> +   of streamer hooks can be active at a time.  */
> +
> +static lto_streamer_hooks streamer_hooks_;
> +
> +lto_streamer_hooks *
> +streamer_hooks (void)
> +{
> +  return &streamer_hooks_;
> +}
> +
> +
> +/* Initialize and return the current set of streamer hooks.  */
> +
> +lto_streamer_hooks *
> +streamer_hooks_init (void)
> +{
> +  lto_streamer_hooks *h = streamer_hooks ();
> +  memset (h, 0, sizeof (lto_streamer_hooks));
> +  return h;
> +}
> diff --git a/gcc/lto-streamer.h b/gcc/lto-streamer.h
> index e8410d4..035aa5e 100644
> --- a/gcc/lto-streamer.h
> +++ b/gcc/lto-streamer.h
> @@ -33,6 +33,12 @@ along with GCC; see the file COPYING3.  If not see
>  #include "alloc-pool.h"
>  #include "gcov-io.h"
>  
> +/* Forward declarations to avoid including unnecessary headers.  */
> +struct output_block;
> +struct lto_input_block;
> +struct data_in;
> +struct bitpack_d;
> +
>  /* Define when debugging the LTO streamer.  This causes the writer
>     to output the numeric value for the memory address of the tree node
>     being emitted.  When debugging a problem in the reader, check the
> @@ -186,7 +192,7 @@ enum LTO_tags
>  
>       Conversely, to map between LTO tags and tree/gimple codes, the
>       reverse operation must be applied.  */
> -  LTO_bb0 = 1 + NUM_TREE_CODES + LAST_AND_UNUSED_GIMPLE_CODE,
> +  LTO_bb0 = 1 + MAX_TREE_CODES + LAST_AND_UNUSED_GIMPLE_CODE,
>    LTO_bb1,

This (and related) change can be merged separately.

>    /* EH region holding the previous statement.  */
> @@ -741,6 +747,98 @@ struct data_in
>  };
>  
>  
> +/* Streamer hooks.  These functions do additional processing as
> +   needed by the module.  There are two types of callbacks, those that
> +   replace the default behavior and those that supplement it.  Any
> +   or all of these hooks can be NULL.  */
> +typedef struct lto_streamer_hooks {
> +  /* A string identifying this streamer.  */
> +  const char *name;
> +
> +  /* Called by lto_streamer_cache_create to instantiate a cache of
> +     well-known nodes.  These are tree nodes that are always
> +     instantiated by the compiler on startup.  Additionally, these
> +     nodes need to be shared.  This function should call
> +     lto_streamer_cache_append on every tree node that it wishes to
> +     preload in the streamer cache.  This way, the writer will only
> +     write out a reference to the tree and the reader will instantiate
> +     the tree out of this pre-populated cache.  */
> +  void (*preload_common_nodes) (struct lto_streamer_cache_d *);
> +
> +  /* Called by lto_reader_init after it does basic initialization.  */
> +  void (*reader_init) (void);
> +
> +  /* Called by lto_writer_init after it does basic initalization.  */
> +  void (*writer_init) (void);
> +
> +  /* Return true if the given tree is supported by this streamer.  */
> +  bool (*is_streamable) (tree);
> +
> +  /* Called by lto_write_tree after writing all the common parts of
> +     a tree.  If defined, the callback is in charge of writing all
> +     the fields that lto_write_tree did not write out.  Arguments
> +     are as in lto_write_tree.
> +
> +     The following tree fields are not handled by common code:
> +
> +	DECL_ABSTRACT_ORIGIN
> +	DECL_INITIAL
> +	DECL_SAVED_TREE
> +
> +     Callbacks may choose to ignore or handle them.  If handled,
> +     the reader should read them in the exact same sequence written
> +     by the writer.  */
> +  void (*write_tree) (struct output_block *, tree, bool);
> +
> +  /* Called by lto_read_tree after reading all the common parts of
> +     a tree.  If defined, the callback is in charge of reading all
> +     the fields that lto_read_tree did not read in.  Arguments
> +     are as in lto_read_tree.  */
> +  void (*read_tree) (struct lto_input_block *, struct data_in *, tree);
> +
> +  /* Called by lto_output_tree_ref to determine if the given tree node
> +     should be emitted as a reference to the table of declarations
> +     (the same table that holds VAR_DECLs).  */
> +  bool (*indexable_with_decls_p) (tree);
> +
> +  /* Called by pack_value_fields to store any non-pointer fields
> +     in the tree structure.  The arguments are as in pack_value_fields.  */
> +  void (*pack_value_fields) (struct bitpack_d *, tree);
> +
> +  /* Called by unpack_value_fields to retrieve any non-pointer fields
> +     in the tree structure.  The arguments are as in unpack_value_fields.  */
> +  void (*unpack_value_fields) (struct bitpack_d *, tree);
> +
> +  /* Non-zero if the streamer should register decls in the LTO
> +     global symbol tables.  */
> +  unsigned register_decls_in_symtab_p : 1;
> +
> +  /* Non-zero if the streamer has special constants that cannot be
> +     shared and are used in pointer-equality tests (e.g., void_zero_node,
> +     truthvalue_false_node, etc).  These constants will be present in
> +     the streamer cache and should be streamed as references.  */
> +  unsigned has_unique_integer_csts_p : 1;
> +
> +  /* Called by lto_materialize_tree for tree nodes that it does not
> +     know how to allocate memory for.  If defined, this hook should
> +     return a new tree node of the given code.  The data_in and
> +     input_block arguments are passed in case the hook needs to
> +     read more data from the stream to allocate the node.
> +     If this hook returns NULL, then lto_materialize_tree will attempt
> +     to allocate the tree by calling make_node directly.  */
> +  tree (*alloc_tree) (enum tree_code, struct lto_input_block *,
> +                      struct data_in *);
> +
> +  /* Called by lto_output_tree_header to write any streamer-specific
> +     information needed to allocate the tree.  This hook may assume
> +     that the basic header data (tree code, etc) has already been
> +     written.  It should only write any extra data needed to allocate
> +     the node (e.g., in the case of CALL_EXPR, this hook would write
> +     the number of arguments to the CALL_EXPR).  */
> +  void (*output_tree_header) (struct output_block *, tree);
> +} lto_streamer_hooks;

Rename to tree_streamer_hooks, start a tree-streamer.h file.

> +
>  /* In lto-section-in.c  */
>  extern struct lto_input_block * lto_create_simple_input_block (
>  			       struct lto_file_decl_data *,
> @@ -851,17 +949,21 @@ extern intptr_t lto_orig_address_get (tree);
>  extern void lto_orig_address_remove (tree);
>  #endif
>  extern void lto_check_version (int, int);
> -
> +extern void gimple_streamer_hooks_init (void);
> +extern void gimple_streamer_write_tree (struct output_block *, tree, bool);
> +extern void gimple_streamer_read_tree (struct lto_input_block *,
> +				       struct data_in *, tree);
> +extern lto_streamer_hooks *streamer_hooks (void);
> +extern lto_streamer_hooks *streamer_hooks_init (void);
>  
>  /* In lto-streamer-in.c */
>  extern void lto_input_cgraph (struct lto_file_decl_data *, const char *);
> -extern void lto_init_reader (void);
> +extern void lto_reader_init (void);
>  extern tree lto_input_tree (struct lto_input_block *, struct data_in *);
>  extern void lto_input_function_body (struct lto_file_decl_data *, tree,
>  				     const char *);
>  extern void lto_input_constructors_and_inits (struct lto_file_decl_data *,
>  					      const char *);
> -extern void lto_init_reader (void);
>  extern struct data_in *lto_data_in_create (struct lto_file_decl_data *,
>  				    const char *, unsigned,
>  				    VEC(ld_plugin_symbol_resolution_t,heap) *);
> @@ -869,6 +971,7 @@ extern void lto_data_in_delete (struct data_in *);
>  extern const char *lto_input_string (struct data_in *,
>  				     struct lto_input_block *);
>  extern void lto_input_data_block (struct lto_input_block *, void *, size_t);
> +extern void gimple_streamer_reader_init (void);
>  
>  
>  /* In lto-streamer-out.c  */
> @@ -957,7 +1060,7 @@ extern VEC(lto_out_decl_state_ptr, heap) *lto_function_decl_states;
>  static inline bool
>  lto_tag_is_tree_code_p (enum LTO_tags tag)
>  {
> -  return tag > LTO_null && (unsigned) tag <= NUM_TREE_CODES;
> +  return tag > LTO_null && (unsigned) tag <= MAX_TREE_CODES;
>  }
>  
>  
> @@ -1063,29 +1166,6 @@ lto_stream_as_builtin_p (tree expr)
>  	      || DECL_BUILT_IN_CLASS (expr) == BUILT_IN_MD));
>  }
>  
> -/* Return true if EXPR is a tree node that can be written to disk.  */
> -static inline bool
> -lto_is_streamable (tree expr)
> -{
> -  enum tree_code code = TREE_CODE (expr);
> -
> -  /* Notice that we reject SSA_NAMEs as well.  We only emit the SSA
> -     name version in lto_output_tree_ref (see output_ssa_names).  */
> -  return !is_lang_specific (expr)
> -	 && code != SSA_NAME
> -	 && code != CALL_EXPR
> -	 && code != LANG_TYPE
> -	 && code != MODIFY_EXPR
> -	 && code != INIT_EXPR
> -	 && code != TARGET_EXPR
> -	 && code != BIND_EXPR
> -	 && code != WITH_CLEANUP_EXPR
> -	 && code != STATEMENT_LIST
> -	 && (code == CASE_LABEL_EXPR
> -	     || code == DECL_EXPR
> -	     || TREE_CODE_CLASS (code) != tcc_statement);
> -}
> -
>  DEFINE_DECL_STREAM_FUNCS (TYPE, type)
>  DEFINE_DECL_STREAM_FUNCS (FIELD_DECL, field_decl)
>  DEFINE_DECL_STREAM_FUNCS (FN_DECL, fn_decl)
> @@ -1226,11 +1306,11 @@ lto_output_int_in_range (struct lto_output_stream *obs,
>    val -= min;
>    lto_output_1_stream (obs, val & 255);
>    if (range >= 0xff)
> -    lto_output_1_stream (obs, (val << 8) & 255);
> +    lto_output_1_stream (obs, (val >> 8) & 255);
>    if (range >= 0xffff)
> -    lto_output_1_stream (obs, (val << 16) & 255);
> +    lto_output_1_stream (obs, (val >> 16) & 255);
>    if (range >= 0xffffff)
> -    lto_output_1_stream (obs, (val << 24) & 255);
> +    lto_output_1_stream (obs, (val >> 24) & 255);
>  }
>  
>  /* Input VAL into OBS and verify it is in range MIN...MAX that is supposed
> diff --git a/gcc/lto/lto.c b/gcc/lto/lto.c
> index 9d4e2ed..657d667 100644
> --- a/gcc/lto/lto.c
> +++ b/gcc/lto/lto.c
> @@ -2615,8 +2615,8 @@ void
>  lto_main (void)
>  {
>    lto_process_name ();
> -
> -  lto_init_reader ();
> +  gimple_streamer_hooks_init ();
> +  lto_reader_init ();

Thanks,
Richard.



More information about the Gcc-patches mailing list