For balanced and max partitioning this adds proper partitioning of asm
and related symbols.
The special symbols are partitioned with 1to1 and joined together if
there is no name conflict. All other symbols are partitioned with the
requested partitioning.
In typical usage with small number of toplevel assembly and no name
conflicts, all special symbols will be in the single first partition.
balanced partitioning will continue filling last asm partition.
gcc/lto/ChangeLog:
* lto-partition.cc (join_partitions): Declare.
(lto_1_to_1_map): Split out to..
(map_1_to_1): ..here.
(symbol_nonrenameable): New.
(create_asm_partition): Replaced by..
(create_asm_partitions): ..this.
(lto_max_map): Use new create_asm_partitions.
(lto_balanced_map): Use new create_asm_partitions.
gcc/testsuite/ChangeLog:
* gcc.dg/lto/toplevel-extended-asm-2_0.c: More partitionings.
* gcc.dg/lto/toplevel-extended-asm-2_1.c: Likewise.
---
gcc/lto/lto-partition.cc | 144 +++++++++++++++---
.../gcc.dg/lto/toplevel-extended-asm-2_0.c | 7 +-
.../gcc.dg/lto/toplevel-extended-asm-2_1.c | 5 +-
3 files changed, 130 insertions(+), 26 deletions(-)
diff --git a/gcc/lto/lto-partition.cc b/gcc/lto/lto-partition.cc
index 8c6ec6f8338..5a2bbea6cd6 100644
--- a/gcc/lto/lto-partition.cc
+++ b/gcc/lto/lto-partition.cc
@@ -45,6 +45,8 @@ vec<ltrans_partition> ltrans_partitions;
static void add_symbol_to_partition (ltrans_partition part,
toplevel_node *node);
+static ltrans_partition join_partitions (ltrans_partition into,
+ ltrans_partition from);
/* Helper for qsort; compare partitions and return one with smaller order. */
@@ -364,11 +366,10 @@ node_into_file_partition (toplevel_node* node,
add_symbol_to_partition (partition, node);
}
-/* Group cgrah nodes by input files. This is used mainly for testing
- right now. */
-
-void
-lto_1_to_1_map (void)
+/* Group cgrah nodes by input files. Used for symbols that must remain
+ together. */
+static void
+map_1_to_1 (bool forced_symbols_only)
{
symtab_node *node;
hash_map<lto_file_decl_data *, ltrans_partition> pmap;
@@ -379,6 +380,9 @@ lto_1_to_1_map (void)
|| symbol_partitioned_p (node))
continue;
+ if (forced_symbols_only && !node->must_remain_in_tu && !node->no_reorder)
+ continue;
+
node_into_file_partition (node, pmap);
}
@@ -386,29 +390,99 @@ lto_1_to_1_map (void)
for (anode = symtab->first_asm_symbol (); anode; anode = anode->next)
node_into_file_partition (anode, pmap);
- create_partition_if_empty ();
-
/* Order partitions by order of symbols because they are linked into binary
that way. */
ltrans_partitions.qsort (cmp_partitions_order);
}
-/* Creates partition with all toplevel assembly.
+/* Group cgrah nodes by input files. This is used mainly for testing
+ right now. */
+
+void
+lto_1_to_1_map (void)
+{
+ map_1_to_1 (false);
+ create_partition_if_empty ();
+}
+
+/* There may be some rare false positives since we do not have another
+ flag to track this nonrenameability directly. */
+static bool
+symbol_nonrenameable (symtab_node* node)
+{
+ return (node->must_remain_in_tu && node->ref_by_asm
+ && !TREE_PUBLIC (node->decl));
+}
+
+/* Toplevel assembly and symbols referenced by it can be required to remain
+ in the same partition and not be renamed.
+ noreroder symbols are also handled here to keep their order in respect to
+ these symbols.
- Before toplevel asm could be partitioned, all toplevel asm was inserted
- into first partition.
- This function achieves similar behavior for partitionings that cannot
- easily satisfy requirements of toplevel asm. */
+ This functions partitions these symbols with 1to1 partitioning and unites
+ translation units to target_size as long as there is no name conflict.
+
+ Remaining symbols can be partitioned with any strategy. */
static void
-create_asm_partition (void)
+create_asm_partitions (int64_t target_size)
{
- struct asm_node *anode = symtab->first_asm_symbol ();
- if (anode)
+ if (!symtab->first_asm_symbol ())
+ return;
+ map_1_to_1 (true);
+
+ size_t join_into = 0;
+ size_t join_from = 0;
+ hash_set<const char*> nonrenameable_symbols;
+ lto_symtab_encoder_iterator lsei;
+
+ for (; join_from < ltrans_partitions.length (); join_into++)
{
- ltrans_partition partition = new_partition ("asm_nodes");
- for (; anode; anode = anode->next)
- add_symbol_to_partition (partition, anode);
+ ltrans_partitions[join_into] = ltrans_partitions[join_from];
+ ltrans_partition p_into = ltrans_partitions[join_into];
+ nonrenameable_symbols.empty ();
+
+ bool first_partition = true;
+ for (; join_from < ltrans_partitions.length (); join_from++)
+ {
+ ltrans_partition p_from = ltrans_partitions[join_from];
+ if (p_into->insns > target_size)
+ break;
+
+ lto_symtab_encoder_t encoder = p_from->encoder;
+ /* All symbols that cannot be renamed and might collide. */
+ for (lsei = lsei_start (encoder); !lsei_end_p (lsei);
+ lsei_next (&lsei))
+ {
+ toplevel_node* tnode = lsei_node (lsei);
+ if (symtab_node *snode = dyn_cast <symtab_node*> (tnode))
+ {
+ if (symbol_nonrenameable (snode))
+ if (nonrenameable_symbols.add (snode->asm_name ()))
+ goto finish_partition;
+ }
+ else if (asm_node *anode = dyn_cast <asm_node*> (tnode))
+ {
+ symtab_node* snode;
+ unsigned i = 0;
+ /* This covers symbols defined by extended assembly. */
+ for (; anode->symbols_referenced.iterate (i, &snode); i++)
+ {
+ if (symbol_nonrenameable (snode))
+ if (nonrenameable_symbols.add (snode->asm_name ()))
+ goto finish_partition;
+ }
+ }
+ }
+ first_partition = false;
+
+ if (p_into != p_from)
+ join_partitions (p_into, p_from);
+ }
+finish_partition: {}
+ if (first_partition)
+ join_from++;
}
+ ltrans_partitions.truncate (join_into);
}
/* Maximal partitioning. Put every new symbol into new partition if possible.
*/
@@ -419,6 +493,9 @@ lto_max_map (void)
symtab_node *node;
ltrans_partition partition;
+ /* Needed for toplevel assembly. */
+ map_1_to_1 (true);
+
FOR_EACH_SYMBOL (node)
{
if (node->get_partitioning_class () != SYMBOL_PARTITION
@@ -428,7 +505,6 @@ lto_max_map (void)
add_symbol_to_partition (partition, node);
}
- create_asm_partition ();
create_partition_if_empty ();
}
@@ -1169,11 +1245,34 @@ lto_balanced_map (int n_lto_partitions, int
max_partition_size)
if (partition_size < param_min_partition_size)
partition_size = param_min_partition_size;
npartitions = 1;
- partition = new_partition ("");
if (dump_file)
fprintf (dump_file, "Total unit size: %" PRId64 ", partition size: %"
PRId64 "\n",
total_size, partition_size);
+ create_asm_partitions (partition_size);
+ if (ltrans_partitions.length ())
+ {
+ int64_t insns = 0;
+ unsigned partitions = ltrans_partitions.length ();
+ for (unsigned i = 0; i + 1 < partitions ; i++)
+ insns += ltrans_partitions[i]->insns;
+
+ total_size -= insns;
+ if (partition_size)
+ n_lto_partitions -= insns / partition_size;
+ if (n_lto_partitions < 1)
+ n_lto_partitions = 1;
+
+ partition_size = total_size / n_lto_partitions;
+ if (partition_size < param_min_partition_size)
+ partition_size = param_min_partition_size;
+
+ partition = ltrans_partitions[partitions - 1];
+ }
+ else
+ partition = new_partition ("");
+
+
auto_vec<symtab_node *> next_nodes;
for (unsigned i = 0; i < order.length (); i++)
@@ -1434,11 +1533,10 @@ lto_balanced_map (int n_lto_partitions, int
max_partition_size)
next_nodes.safe_push (noreorder[noreorder_pos++]);
/* For one partition the cost of boundary should be 0 unless we added final
symbols here (these are not accounted) or we have accounting bug. */
- gcc_assert (next_nodes.length () || npartitions != 1 || !best_cost ||
best_cost == -1);
+ gcc_assert (next_nodes.length () || ltrans_partitions.length () != 1
+ || !best_cost || best_cost == -1);
add_sorted_nodes (next_nodes, partition);
- create_asm_partition ();
-
if (dump_file)
{
fprintf (dump_file, "\nPartition sizes:\n");
diff --git a/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_0.c
b/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_0.c
index 19ac7c21b2d..84a7284d669 100644
--- a/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_0.c
+++ b/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_0.c
@@ -1,10 +1,13 @@
/* { dg-lto-do link } */
-/* { dg-lto-options {{-O2 -flto -flto-partition=1to1} } } */
+/* { dg-lto-options {{-O2 -flto -flto-partition=1to1} {-O2 -flto
-flto-partition=max} {-O2 -flto -flto-partition=cache}} } */
extern int use_statics ();
extern int asm_var;
+static int a;
+asm (".local %cc0\n %cc0:" :: ":"(&a));
+
int main() {
- return use_statics ();
+ return a + use_statics ();
}
diff --git a/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_1.c
b/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_1.c
index a115f139eaa..2b6c5539177 100644
--- a/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_1.c
+++ b/gcc/testsuite/gcc.dg/lto/toplevel-extended-asm-2_1.c
@@ -6,7 +6,10 @@ asm("%cc0:" :: ":" (&static_asm_var));
extern int asm_var;
asm("%cc0:" :: ":" (&asm_var));
+static int a;
+asm (".local %cc0\n %cc0:" :: ":"(&a));
+
int use_statics () {
static_asm_fn ();
- return static_asm_var;
+ return static_asm_var + a;
}
--
2.51.1
