* [PATCH/RFC 2/2] WPD: Enable whole program devirtualization at LTRANS
@ 2021-09-16 9:26 Feng Xue OS
2021-09-30 3:59 ` PING: " Feng Xue OS
0 siblings, 1 reply; 3+ messages in thread
From: Feng Xue OS @ 2021-09-16 9:26 UTC (permalink / raw)
To: Jan Hubicka, mjambor, Richard Biener, gcc-patches
[-- Attachment #1: Type: text/plain, Size: 1865 bytes --]
This patch is to extend applicability of full devirtualization to LTRANS stage.
Normally, whole program assumption would not hold when WPA splits
whole compilation into more than one LTRANS partitions. To avoid information
lost for WPD at LTRANS, we will record all vtable nodes and related member
function references into each partition.
Bootstrapped/regtested on x86_64-linux and aarch64-linux.
Thanks,
Feng
----
2021-09-07 Feng Xue <fxue@os.amperecomputing.com>
gcc/
* tree.h (TYPE_CXX_LOCAL): New macro for type using
base.nothrow_flag.
* tree-core.h (tree_base): Update comment on using
base.nothrow_flag to represent TYPE_CXX_LOCAL.
* ipa-devirt.c (odr_type_d::whole_program_local): Removed.
(odr_type_d::whole_program_local_p): Check TYPE_CXX_LOCAL flag
on type, and enable WPD at LTRANS when flag_devirtualize_fully
is true.
(get_odr_type): Remove setting whole_program_local flag on type.
(identify_whole_program_local_types): Replace whole_program_local
in odr_type_d by TYPE_CXX_LOCAL on type.
(maybe_record_node): Enable WPD at LTRANS when
flag_devirtualize_fully is true.
* ipa.c (can_remove_vtable_if_no_refs_p): Retain vtables at LTRANS
stage under full devirtualization.
* lto-cgraph.c (compute_ltrans_boundary): Add all defined vtables
to boundary of each LTRANS partition.
* lto-streamer-out.c (get_symbol_initial_value): Streaming out
initial value of vtable even its class is optimized away.
* lto-lang.c (lto_post_options): Disable full devirtualization
if flag_ltrans_devirtualize is false.
* tree-streamer-in.c (unpack_ts_base_value_fields): unpack value
of TYPE_CXX_LOCAL for a type from streaming data.
* tree-streamer-out.c (pack_ts_base_value_fields): pack value
ofTYPE_CXX_LOCAL for a type into streaming data.
---
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0002-WPD-Enable-whole-program-devirtualization-at-LTRANS.patch --]
[-- Type: text/x-patch; name="0002-WPD-Enable-whole-program-devirtualization-at-LTRANS.patch", Size: 11190 bytes --]
From 3af32b9aadff23d339750ada4541386b3d358edc Mon Sep 17 00:00:00 2001
From: Feng Xue <fxue@os.amperecomputing.com>
Date: Mon, 6 Sep 2021 20:34:50 +0800
Subject: [PATCH 2/2] WPD: Enable whole program devirtualization at LTRANS
Whole program assumption would not hold when WPA splits whole compilation
into more than one LTRANS partitions. To avoid information lost for WPD
at LTRANS, we will record all vtable nodes and related member function
references into each partition.
2021-09-07 Feng Xue <fxue@os.amperecomputing.com>
gcc/
* tree.h (TYPE_CXX_LOCAL): New macro for type using
base.nothrow_flag.
* tree-core.h (tree_base): Update comment on using
base.nothrow_flag to represent TYPE_CXX_LOCAL.
* ipa-devirt.c (odr_type_d::whole_program_local): Removed.
(odr_type_d::whole_program_local_p): Check TYPE_CXX_LOCAL flag
on type, and enable WPD at LTRANS when flag_devirtualize_fully
is true.
(get_odr_type): Remove setting whole_program_local flag on type.
(identify_whole_program_local_types): Replace whole_program_local
in odr_type_d by TYPE_CXX_LOCAL on type.
(maybe_record_node): Enable WPD at LTRANS when
flag_devirtualize_fully is true.
* ipa.c (can_remove_vtable_if_no_refs_p): Retain vtables at LTRANS
stage under full devirtualization.
* lto-cgraph.c (compute_ltrans_boundary): Add all defined vtables
to boundary of each LTRANS partition.
* lto-streamer-out.c (get_symbol_initial_value): Streaming out
initial value of vtable even its class is optimized away.
* lto-lang.c (lto_post_options): Disable full devirtualization
if flag_ltrans_devirtualize is false.
* tree-streamer-in.c (unpack_ts_base_value_fields): unpack value
of TYPE_CXX_LOCAL for a type from streaming data.
* tree-streamer-out.c (pack_ts_base_value_fields): pack value
ofTYPE_CXX_LOCAL for a type into streaming data.
---
gcc/ipa-devirt.c | 29 ++++++++++++++++++-----------
gcc/ipa.c | 7 ++++++-
gcc/lto-cgraph.c | 18 ++++++++++++++++++
gcc/lto-streamer-out.c | 12 +++++++++++-
gcc/lto/lto-lang.c | 6 ++++++
gcc/tree-core.h | 3 +++
gcc/tree-streamer-in.c | 11 ++++++++---
gcc/tree-streamer-out.c | 11 ++++++++---
gcc/tree.h | 5 +++++
9 files changed, 83 insertions(+), 19 deletions(-)
diff --git a/gcc/ipa-devirt.c b/gcc/ipa-devirt.c
index fcb097d7156..65e9ebbfb59 100644
--- a/gcc/ipa-devirt.c
+++ b/gcc/ipa-devirt.c
@@ -216,8 +216,6 @@ struct GTY(()) odr_type_d
int id;
/* Is it in anonymous namespace? */
bool anonymous_namespace;
- /* Set when type is not used outside of program. */
- bool whole_program_local;
/* Did we report ODR violation here? */
bool odr_violated;
/* Set when virtual table without RTTI prevailed table with. */
@@ -290,10 +288,18 @@ get_type_vtable (tree type)
bool
odr_type_d::whole_program_local_p ()
{
- if (flag_ltrans)
+ if (flag_ltrans && !flag_devirtualize_fully)
return false;
- return whole_program_local;
+ if (in_lto_p)
+ return TYPE_CXX_LOCAL (type);
+
+ /* Although a local class is always considered as whole program local in
+ LGEN stage, but may not in LTO stage if multiple duplicated primary
+ vtables are attached to the class due to C++ privatizing via -fno-weak.
+ Thus, we can not set TYPE_CXX_LOCAL flag for local class at LGEN stage
+ when building ORD type. */
+ return anonymous_namespace || decl_function_context (TYPE_NAME (type));
}
/* Return TRUE if ODR type may have any instance. */
@@ -2007,11 +2013,6 @@ get_odr_type (tree type, bool insert)
else
val->anonymous_namespace = 0;
- if (!in_lto_p
- && (val->anonymous_namespace
- || decl_function_context (TYPE_NAME (type))))
- val->whole_program_local = true;
-
build_bases = COMPLETE_TYPE_P (val->type);
insert_to_odr_array = true;
*slot = val;
@@ -2602,9 +2603,15 @@ identify_whole_program_local_types (void)
if (multi_vtable_p)
continue;
+
+ /* Mark all equivalent types in the ODR type as whole program local,
+ because representative type of the ODR type at LTRANS might not
+ be the one at WPA. */
+ FOR_EACH_VEC_ELT (*(type->types), i, equiv_type)
+ TYPE_CXX_LOCAL (equiv_type) = 1;
}
- type->whole_program_local = true;
+ TYPE_CXX_LOCAL (type->type) = 1;
}
delete no_rtti_files;
@@ -2721,7 +2728,7 @@ maybe_record_node (vec <cgraph_node *> &nodes,
Currently we ignore these functions in speculative devirtualization.
??? Maybe it would make sense to be more aggressive for LTO even
elsewhere. */
- if (!flag_ltrans
+ if ((!flag_ltrans || flag_devirtualize_fully)
&& !pure_virtual
&& type_in_anonymous_namespace_p (DECL_CONTEXT (target))
&& (!target_node
diff --git a/gcc/ipa.c b/gcc/ipa.c
index 6e200a906b5..52d6a9e63be 100644
--- a/gcc/ipa.c
+++ b/gcc/ipa.c
@@ -268,7 +268,7 @@ can_remove_vtable_if_no_refs_p (varpool_node *vnode)
if (!flag_devirtualize_fully)
return true;
- if (DECL_EXTERNAL (vnode->decl))
+ if (DECL_EXTERNAL (vnode->decl) && !vnode->in_other_partition)
return true;
/* We will force generating vtables in LGEN stage even they are "unused",
@@ -276,6 +276,11 @@ can_remove_vtable_if_no_refs_p (varpool_node *vnode)
if (!in_lto_p && flag_generate_lto)
return false;
+ /* All vtables seen at LTRANS stage are result of dead class elimination
+ at WPA, no need to prune them further. */
+ if (flag_ltrans)
+ return false;
+
return true;
}
diff --git a/gcc/lto-cgraph.c b/gcc/lto-cgraph.c
index 7c3e276a8ea..12c82eebc3f 100644
--- a/gcc/lto-cgraph.c
+++ b/gcc/lto-cgraph.c
@@ -859,6 +859,24 @@ compute_ltrans_boundary (lto_symtab_encoder_t in_encoder)
lto_set_symtab_encoder_encode_initializer (encoder, vnode);
create_references (encoder, vnode);
}
+
+ /* Add all defined vtables to streaming encoder, so that we could
+ reconstruct whole type inheritance graph in LTRANS as what it was in
+ WPA to enable late full devirtualization.
+
+ TODO: For a partition, only need to encode a subset of vtables, not
+ all of them. */
+ if (flag_devirtualize_fully && flag_ltrans_devirtualize && flag_wpa)
+ {
+ varpool_node *vnode;
+
+ FOR_EACH_DEFINED_VARIABLE (vnode)
+ if (DECL_VIRTUAL_P (vnode->decl) && !DECL_EXTERNAL (vnode->decl)
+ && TYPE_CXX_LOCAL (DECL_CONTEXT (vnode->decl))
+ && lto_symtab_encoder_lookup (encoder, vnode) == LCC_NOT_FOUND)
+ lto_symtab_encoder_encode (encoder, vnode);
+ }
+
/* Pickle in also the initializer of all referenced readonly variables
to help folding. Constant pool variables are not shared, so we must
pickle those too. */
diff --git a/gcc/lto-streamer-out.c b/gcc/lto-streamer-out.c
index 643d9275a95..c8dd2aee7b4 100644
--- a/gcc/lto-streamer-out.c
+++ b/gcc/lto-streamer-out.c
@@ -437,7 +437,17 @@ get_symbol_initial_value (lto_symtab_encoder_t encoder, tree expr)
scalar values. */
if (!(vnode = varpool_node::get (expr))
|| !lto_symtab_encoder_encode_initializer_p (encoder, vnode))
- initial = error_mark_node;
+ {
+ /* Even when a class is optimized away, devirtualization at LTRANS
+ still need to extract member virtual functions from initial value
+ of vtable. */
+ if (flag_devirtualize_fully && flag_ltrans_devirtualize
+ && flag_wpa && DECL_VIRTUAL_P (expr) && !DECL_EXTERNAL (expr))
+ return initial;
+
+ initial = error_mark_node;
+ }
+
if (initial != error_mark_node)
{
long max_size = 30;
diff --git a/gcc/lto/lto-lang.c b/gcc/lto/lto-lang.c
index 92f499643b5..ab37a5e6c60 100644
--- a/gcc/lto/lto-lang.c
+++ b/gcc/lto/lto-lang.c
@@ -870,6 +870,12 @@ lto_post_options (const char **pfilename ATTRIBUTE_UNUSED)
/* During LTRANS, we are not looking at the whole program, only
a subset of the whole callgraph. */
flag_whole_program = 0;
+
+ /* Since full devirtualization at LTRANS requires streaming extra data
+ on vtables, it only takes effect when flag_ltrans_devirtualize is
+ also on. */
+ if (!flag_ltrans_devirtualize)
+ flag_devirtualize_fully = 0;
}
if (flag_wpa)
diff --git a/gcc/tree-core.h b/gcc/tree-core.h
index 192e1d51fc2..382cc097ba8 100644
--- a/gcc/tree-core.h
+++ b/gcc/tree-core.h
@@ -1358,6 +1358,9 @@ struct GTY(()) tree_base {
DECL_NONALIASED in
VAR_DECL
+ TYPE_CXX_LOCAL in
+ all types
+
deprecated_flag:
TREE_DEPRECATED in
diff --git a/gcc/tree-streamer-in.c b/gcc/tree-streamer-in.c
index 848981aa20b..42de93a3668 100644
--- a/gcc/tree-streamer-in.c
+++ b/gcc/tree-streamer-in.c
@@ -129,10 +129,15 @@ unpack_ts_base_value_fields (struct bitpack_d *bp, tree expr)
bp_unpack_value (bp, 1);
TREE_ASM_WRITTEN (expr) = (unsigned) bp_unpack_value (bp, 1);
if (TYPE_P (expr))
- TYPE_ARTIFICIAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ {
+ TYPE_ARTIFICIAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ TYPE_CXX_LOCAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ }
else
- TREE_NO_WARNING (expr) = (unsigned) bp_unpack_value (bp, 1);
- TREE_NOTHROW (expr) = (unsigned) bp_unpack_value (bp, 1);
+ {
+ TREE_NO_WARNING (expr) = (unsigned) bp_unpack_value (bp, 1);
+ TREE_NOTHROW (expr) = (unsigned) bp_unpack_value (bp, 1);
+ }
TREE_STATIC (expr) = (unsigned) bp_unpack_value (bp, 1);
if (TREE_CODE (expr) != TREE_BINFO)
TREE_PRIVATE (expr) = (unsigned) bp_unpack_value (bp, 1);
diff --git a/gcc/tree-streamer-out.c b/gcc/tree-streamer-out.c
index 1f9ce52d541..0ff03b7838c 100644
--- a/gcc/tree-streamer-out.c
+++ b/gcc/tree-streamer-out.c
@@ -102,10 +102,15 @@ pack_ts_base_value_fields (struct bitpack_d *bp, tree expr)
bp_pack_value (bp, (TREE_CODE (expr) != SSA_NAME
? 0 : TREE_ASM_WRITTEN (expr)), 1);
if (TYPE_P (expr))
- bp_pack_value (bp, TYPE_ARTIFICIAL (expr), 1);
+ {
+ bp_pack_value (bp, TYPE_ARTIFICIAL (expr), 1);
+ bp_pack_value (bp, TYPE_CXX_LOCAL (expr), 1);
+ }
else
- bp_pack_value (bp, TREE_NO_WARNING (expr), 1);
- bp_pack_value (bp, TREE_NOTHROW (expr), 1);
+ {
+ bp_pack_value (bp, TREE_NO_WARNING (expr), 1);
+ bp_pack_value (bp, TREE_NOTHROW (expr), 1);
+ }
bp_pack_value (bp, TREE_STATIC (expr), 1);
if (TREE_CODE (expr) != TREE_BINFO)
bp_pack_value (bp, TREE_PRIVATE (expr), 1);
diff --git a/gcc/tree.h b/gcc/tree.h
index 0ad5faeaf93..8fcc2fc3922 100644
--- a/gcc/tree.h
+++ b/gcc/tree.h
@@ -2303,6 +2303,11 @@ extern tree vector_element_bits_tree (const_tree);
#define TYPE_FINAL_P(NODE) \
(RECORD_OR_UNION_CHECK (NODE)->base.default_def_flag)
+/* If nonzero, it indicates that a C++ type (mainly polymorphic class) may
+ be referenced outside the program being compiled. This is used in C++
+ devirtualization. */
+#define TYPE_CXX_LOCAL(NODE) (TYPE_CHECK (NODE)->base.nothrow_flag)
+
/* The debug output functions use the symtab union field to store
information specific to the debugging format. The different debug
output hooks store different types in the union field. These three
--
2.17.1
^ permalink raw reply [flat|nested] 3+ messages in thread
* PING: [PATCH/RFC 2/2] WPD: Enable whole program devirtualization at LTRANS
2021-09-16 9:26 [PATCH/RFC 2/2] WPD: Enable whole program devirtualization at LTRANS Feng Xue OS
@ 2021-09-30 3:59 ` Feng Xue OS
2021-10-15 3:50 ` PING^2: " Feng Xue OS
0 siblings, 1 reply; 3+ messages in thread
From: Feng Xue OS @ 2021-09-30 3:59 UTC (permalink / raw)
To: Jan Hubicka, mjambor, Richard Biener, gcc-patches
[-- Attachment #1: Type: text/plain, Size: 2315 bytes --]
Made some minor changes.
Thanks,
Feng
________________________________________
From: Feng Xue OS
Sent: Thursday, September 16, 2021 5:26 PM
To: Jan Hubicka; mjambor@suse.cz; Richard Biener; gcc-patches@gcc.gnu.org
Cc: JiangNing OS
Subject: [PATCH/RFC 2/2] WPD: Enable whole program devirtualization at LTRANS
This patch is to extend applicability of full devirtualization to LTRANS stage.
Normally, whole program assumption would not hold when WPA splits
whole compilation into more than one LTRANS partitions. To avoid information
lost for WPD at LTRANS, we will record all vtable nodes and related member
function references into each partition.
Bootstrapped/regtested on x86_64-linux and aarch64-linux.
Thanks,
Feng
----
2021-09-07 Feng Xue <fxue@os.amperecomputing.com>
gcc/
* tree.h (TYPE_CXX_LOCAL): New macro for type using
base.nothrow_flag.
* tree-core.h (tree_base): Update comment on using
base.nothrow_flag to represent TYPE_CXX_LOCAL.
* ipa-devirt.c (odr_type_d::whole_program_local): Removed.
(odr_type_d::whole_program_local_p): Check TYPE_CXX_LOCAL flag
on type, and enable WPD at LTRANS when flag_devirtualize_fully
is true.
(get_odr_type): Remove setting whole_program_local flag on type.
(identify_whole_program_local_types): Replace whole_program_local
in odr_type_d by TYPE_CXX_LOCAL on type.
(maybe_record_node): Enable WPD at LTRANS when
flag_devirtualize_fully is true.
* ipa.c (can_remove_vtable_if_no_refs_p): Retain vtables at LTRANS
stage under full devirtualization.
* lto-cgraph.c (compute_ltrans_boundary): Add all defined vtables
to boundary of each LTRANS partition.
* lto-streamer-out.c (get_symbol_initial_value): Streaming out
initial value of vtable even its class is optimized away.
* lto-lang.c (lto_post_options): Disable full devirtualization
if flag_ltrans_devirtualize is false.
* tree-streamer-in.c (unpack_ts_base_value_fields): unpack value
of TYPE_CXX_LOCAL for a type from streaming data.
* tree-streamer-out.c (pack_ts_base_value_fields): pack value
ofTYPE_CXX_LOCAL for a type into streaming data.
---
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0002-WPD-Enable-whole-program-devirtualization-at-LTRANS.patch --]
[-- Type: text/x-patch; name="0002-WPD-Enable-whole-program-devirtualization-at-LTRANS.patch", Size: 11933 bytes --]
From 2c0d243b0c092585561c732bac490700f41001fb Mon Sep 17 00:00:00 2001
From: Feng Xue <fxue@os.amperecomputing.com>
Date: Mon, 6 Sep 2021 20:34:50 +0800
Subject: [PATCH 2/2] WPD: Enable whole program devirtualization at LTRANS
Whole program assumption would not hold when WPA splits whole compilation
into more than one LTRANS partitions. To avoid information lost for WPD
at LTRANS, we will record all vtable nodes and related member function
references into each partition.
2021-09-07 Feng Xue <fxue@os.amperecomputing.com>
gcc/
* tree.h (TYPE_CXX_LOCAL): New macro for type using
base.nothrow_flag.
* tree-core.h (tree_base): Update comment on using
base.nothrow_flag to represent TYPE_CXX_LOCAL.
* ipa-devirt.c (odr_type_d::whole_program_local): Removed.
(odr_type_d::whole_program_local_p): Check TYPE_CXX_LOCAL flag
on type, and enable WPD at LTRANS when flag_devirtualize_fully
is true.
(get_odr_type): Remove setting whole_program_local flag on type.
(identify_whole_program_local_types): Replace whole_program_local
in odr_type_d by TYPE_CXX_LOCAL on type.
(maybe_record_node): Enable WPD at LTRANS when
flag_devirtualize_fully is true.
* ipa.c (can_remove_vtable_if_no_refs_p): Retain vtables at LTRANS
stage under full devirtualization.
* lto-cgraph.c (compute_ltrans_boundary): Add all defined vtables
to boundary of each LTRANS partition.
* lto-streamer-out.c (get_symbol_initial_value): Streaming out
initial value of vtable even its class is optimized away.
* lto-streamer-in.c (lto_input_tree): There might be more than
one decls in dref_queue, register debuginfo for all of them.
* lto-lang.c (lto_post_options): Disable full devirtualization
if flag_ltrans_devirtualize is false.
* tree-streamer-in.c (unpack_ts_base_value_fields): unpack value
of TYPE_CXX_LOCAL for a type from streaming data.
* tree-streamer-out.c (pack_ts_base_value_fields): pack value
ofTYPE_CXX_LOCAL for a type into streaming data.
temp
---
gcc/ipa-devirt.c | 29 ++++++++++++++++++-----------
gcc/ipa.c | 7 ++++++-
gcc/lto-cgraph.c | 18 ++++++++++++++++++
gcc/lto-streamer-in.c | 3 +--
gcc/lto-streamer-out.c | 12 +++++++++++-
gcc/lto/lto-lang.c | 6 ++++++
gcc/tree-core.h | 3 +++
gcc/tree-streamer-in.c | 11 ++++++++---
gcc/tree-streamer-out.c | 11 ++++++++---
gcc/tree.h | 5 +++++
10 files changed, 84 insertions(+), 21 deletions(-)
diff --git a/gcc/ipa-devirt.c b/gcc/ipa-devirt.c
index a7d04388dab..4ff551bace8 100644
--- a/gcc/ipa-devirt.c
+++ b/gcc/ipa-devirt.c
@@ -216,8 +216,6 @@ struct GTY(()) odr_type_d
int id;
/* Is it in anonymous namespace? */
bool anonymous_namespace;
- /* Set when type is not used outside of program. */
- bool whole_program_local;
/* Did we report ODR violation here? */
bool odr_violated;
/* Set when virtual table without RTTI prevailed table with. */
@@ -290,10 +288,18 @@ get_type_vtable (tree type)
bool
odr_type_d::whole_program_local_p ()
{
- if (flag_ltrans)
+ if (flag_ltrans && !flag_devirtualize_fully)
return false;
- return whole_program_local;
+ if (in_lto_p)
+ return TYPE_CXX_LOCAL (type);
+
+ /* Although a local class is always considered as whole program local in
+ LGEN stage, but may not in LTO stage if multiple duplicated primary
+ vtables are attached to the class due to C++ privatizing via -fno-weak.
+ Thus, we can not set TYPE_CXX_LOCAL flag for local class at LGEN stage
+ when building ORD type. */
+ return anonymous_namespace || decl_function_context (TYPE_NAME (type));
}
/* Return TRUE if ODR type may have any instance. */
@@ -2007,11 +2013,6 @@ get_odr_type (tree type, bool insert)
else
val->anonymous_namespace = 0;
- if (!in_lto_p
- && (val->anonymous_namespace
- || decl_function_context (TYPE_NAME (type))))
- val->whole_program_local = true;
-
build_bases = COMPLETE_TYPE_P (val->type);
insert_to_odr_array = true;
*slot = val;
@@ -2607,9 +2608,15 @@ identify_whole_program_local_types (void)
if (multi_vtable_p)
continue;
+
+ /* Mark all equivalent types in the ODR type as whole program local,
+ because representative type of the ODR type at LTRANS might not
+ be the one at WPA. */
+ FOR_EACH_VEC_ELT (*(type->types), i, equiv_type)
+ TYPE_CXX_LOCAL (equiv_type) = 1;
}
- type->whole_program_local = true;
+ TYPE_CXX_LOCAL (type->type) = 1;
}
delete no_rtti_files;
@@ -2726,7 +2733,7 @@ maybe_record_node (vec <cgraph_node *> &nodes,
Currently we ignore these functions in speculative devirtualization.
??? Maybe it would make sense to be more aggressive for LTO even
elsewhere. */
- if (!flag_ltrans
+ if ((!flag_ltrans || flag_devirtualize_fully)
&& !pure_virtual
&& type_in_anonymous_namespace_p (DECL_CONTEXT (target))
&& (!target_node
diff --git a/gcc/ipa.c b/gcc/ipa.c
index 6e200a906b5..52d6a9e63be 100644
--- a/gcc/ipa.c
+++ b/gcc/ipa.c
@@ -268,7 +268,7 @@ can_remove_vtable_if_no_refs_p (varpool_node *vnode)
if (!flag_devirtualize_fully)
return true;
- if (DECL_EXTERNAL (vnode->decl))
+ if (DECL_EXTERNAL (vnode->decl) && !vnode->in_other_partition)
return true;
/* We will force generating vtables in LGEN stage even they are "unused",
@@ -276,6 +276,11 @@ can_remove_vtable_if_no_refs_p (varpool_node *vnode)
if (!in_lto_p && flag_generate_lto)
return false;
+ /* All vtables seen at LTRANS stage are result of dead class elimination
+ at WPA, no need to prune them further. */
+ if (flag_ltrans)
+ return false;
+
return true;
}
diff --git a/gcc/lto-cgraph.c b/gcc/lto-cgraph.c
index 7c3e276a8ea..12c82eebc3f 100644
--- a/gcc/lto-cgraph.c
+++ b/gcc/lto-cgraph.c
@@ -859,6 +859,24 @@ compute_ltrans_boundary (lto_symtab_encoder_t in_encoder)
lto_set_symtab_encoder_encode_initializer (encoder, vnode);
create_references (encoder, vnode);
}
+
+ /* Add all defined vtables to streaming encoder, so that we could
+ reconstruct whole type inheritance graph in LTRANS as what it was in
+ WPA to enable late full devirtualization.
+
+ TODO: For a partition, only need to encode a subset of vtables, not
+ all of them. */
+ if (flag_devirtualize_fully && flag_ltrans_devirtualize && flag_wpa)
+ {
+ varpool_node *vnode;
+
+ FOR_EACH_DEFINED_VARIABLE (vnode)
+ if (DECL_VIRTUAL_P (vnode->decl) && !DECL_EXTERNAL (vnode->decl)
+ && TYPE_CXX_LOCAL (DECL_CONTEXT (vnode->decl))
+ && lto_symtab_encoder_lookup (encoder, vnode) == LCC_NOT_FOUND)
+ lto_symtab_encoder_encode (encoder, vnode);
+ }
+
/* Pickle in also the initializer of all referenced readonly variables
to help folding. Constant pool variables are not shared, so we must
pickle those too. */
diff --git a/gcc/lto-streamer-in.c b/gcc/lto-streamer-in.c
index eb8a7dc57b0..b0902a28d59 100644
--- a/gcc/lto-streamer-in.c
+++ b/gcc/lto-streamer-in.c
@@ -1905,11 +1905,10 @@ lto_input_tree (class lto_input_block *ib, class data_in *data_in)
}
tree t = lto_input_tree_1 (ib, data_in, tag, 0);
- if (!dref_queue.is_empty ())
+ while (!dref_queue.is_empty ())
{
dref_entry e = dref_queue.pop ();
debug_hooks->register_external_die (e.decl, e.sym, e.off);
- gcc_checking_assert (dref_queue.is_empty ());
}
return t;
}
diff --git a/gcc/lto-streamer-out.c b/gcc/lto-streamer-out.c
index a26d4885800..c36efb2e25c 100644
--- a/gcc/lto-streamer-out.c
+++ b/gcc/lto-streamer-out.c
@@ -437,7 +437,17 @@ get_symbol_initial_value (lto_symtab_encoder_t encoder, tree expr)
scalar values. */
if (!(vnode = varpool_node::get (expr))
|| !lto_symtab_encoder_encode_initializer_p (encoder, vnode))
- initial = error_mark_node;
+ {
+ /* Even when a class is optimized away, devirtualization at LTRANS
+ still needs to extract addresses of member virtual functions from
+ initial value of vtable. */
+ if (flag_devirtualize_fully && flag_ltrans_devirtualize
+ && flag_wpa && DECL_VIRTUAL_P (expr) && !DECL_EXTERNAL (expr))
+ return initial;
+
+ initial = error_mark_node;
+ }
+
if (initial != error_mark_node)
{
long max_size = 30;
diff --git a/gcc/lto/lto-lang.c b/gcc/lto/lto-lang.c
index c13c7e45ac1..a48b55f7ada 100644
--- a/gcc/lto/lto-lang.c
+++ b/gcc/lto/lto-lang.c
@@ -870,6 +870,12 @@ lto_post_options (const char **pfilename ATTRIBUTE_UNUSED)
/* During LTRANS, we are not looking at the whole program, only
a subset of the whole callgraph. */
flag_whole_program = 0;
+
+ /* Since full devirtualization at LTRANS requires streaming extra data
+ on vtables, it only takes effect when flag_ltrans_devirtualize is
+ also on. */
+ if (!flag_ltrans_devirtualize)
+ flag_devirtualize_fully = 0;
}
if (flag_wpa)
diff --git a/gcc/tree-core.h b/gcc/tree-core.h
index 4498c3ad127..08d59af022b 100644
--- a/gcc/tree-core.h
+++ b/gcc/tree-core.h
@@ -1328,6 +1328,9 @@ struct GTY(()) tree_base {
DECL_NONALIASED in
VAR_DECL
+ TYPE_CXX_LOCAL in
+ all types
+
deprecated_flag:
TREE_DEPRECATED in
diff --git a/gcc/tree-streamer-in.c b/gcc/tree-streamer-in.c
index 984b1e269cf..1597fc102b0 100644
--- a/gcc/tree-streamer-in.c
+++ b/gcc/tree-streamer-in.c
@@ -129,10 +129,15 @@ unpack_ts_base_value_fields (struct bitpack_d *bp, tree expr)
bp_unpack_value (bp, 1);
TREE_ASM_WRITTEN (expr) = (unsigned) bp_unpack_value (bp, 1);
if (TYPE_P (expr))
- TYPE_ARTIFICIAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ {
+ TYPE_ARTIFICIAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ TYPE_CXX_LOCAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ }
else
- TREE_NO_WARNING (expr) = (unsigned) bp_unpack_value (bp, 1);
- TREE_NOTHROW (expr) = (unsigned) bp_unpack_value (bp, 1);
+ {
+ TREE_NO_WARNING (expr) = (unsigned) bp_unpack_value (bp, 1);
+ TREE_NOTHROW (expr) = (unsigned) bp_unpack_value (bp, 1);
+ }
TREE_STATIC (expr) = (unsigned) bp_unpack_value (bp, 1);
if (TREE_CODE (expr) != TREE_BINFO)
TREE_PRIVATE (expr) = (unsigned) bp_unpack_value (bp, 1);
diff --git a/gcc/tree-streamer-out.c b/gcc/tree-streamer-out.c
index 1a43534d117..7ec719ed59b 100644
--- a/gcc/tree-streamer-out.c
+++ b/gcc/tree-streamer-out.c
@@ -102,10 +102,15 @@ pack_ts_base_value_fields (struct bitpack_d *bp, tree expr)
bp_pack_value (bp, (TREE_CODE (expr) != SSA_NAME
? 0 : TREE_ASM_WRITTEN (expr)), 1);
if (TYPE_P (expr))
- bp_pack_value (bp, TYPE_ARTIFICIAL (expr), 1);
+ {
+ bp_pack_value (bp, TYPE_ARTIFICIAL (expr), 1);
+ bp_pack_value (bp, TYPE_CXX_LOCAL (expr), 1);
+ }
else
- bp_pack_value (bp, TREE_NO_WARNING (expr), 1);
- bp_pack_value (bp, TREE_NOTHROW (expr), 1);
+ {
+ bp_pack_value (bp, TREE_NO_WARNING (expr), 1);
+ bp_pack_value (bp, TREE_NOTHROW (expr), 1);
+ }
bp_pack_value (bp, TREE_STATIC (expr), 1);
if (TREE_CODE (expr) != TREE_BINFO)
bp_pack_value (bp, TREE_PRIVATE (expr), 1);
diff --git a/gcc/tree.h b/gcc/tree.h
index e5547e8491c..6842de7a24b 100644
--- a/gcc/tree.h
+++ b/gcc/tree.h
@@ -2254,6 +2254,11 @@ extern tree vector_element_bits_tree (const_tree);
#define TYPE_FINAL_P(NODE) \
(RECORD_OR_UNION_CHECK (NODE)->base.default_def_flag)
+/* If nonzero, it indicates that a C++ type (mainly polymorphic class) may
+ be referenced outside the program being compiled. This is used in C++
+ devirtualization. */
+#define TYPE_CXX_LOCAL(NODE) (TYPE_CHECK (NODE)->base.nothrow_flag)
+
/* The debug output functions use the symtab union field to store
information specific to the debugging format. The different debug
output hooks store different types in the union field. These three
--
2.17.1
^ permalink raw reply [flat|nested] 3+ messages in thread
* PING^2: [PATCH/RFC 2/2] WPD: Enable whole program devirtualization at LTRANS
2021-09-30 3:59 ` PING: " Feng Xue OS
@ 2021-10-15 3:50 ` Feng Xue OS
0 siblings, 0 replies; 3+ messages in thread
From: Feng Xue OS @ 2021-10-15 3:50 UTC (permalink / raw)
To: Jan Hubicka, mjambor, Richard Biener, gcc-patches
[-- Attachment #1: Type: text/plain, Size: 2287 bytes --]
Thanks,
Feng
________________________________________
From: Feng Xue OS
Sent: Thursday, September 16, 2021 5:26 PM
To: Jan Hubicka; mjambor@suse.cz; Richard Biener; gcc-patches@gcc.gnu.org
Cc: JiangNing OS
Subject: [PATCH/RFC 2/2] WPD: Enable whole program devirtualization at LTRANS
This patch is to extend applicability of full devirtualization to LTRANS stage.
Normally, whole program assumption would not hold when WPA splits
whole compilation into more than one LTRANS partitions. To avoid information
lost for WPD at LTRANS, we will record all vtable nodes and related member
function references into each partition.
Bootstrapped/regtested on x86_64-linux and aarch64-linux.
Thanks,
Feng
----
2021-09-07 Feng Xue <fxue@os.amperecomputing.com>
gcc/
* tree.h (TYPE_CXX_LOCAL): New macro for type using
base.nothrow_flag.
* tree-core.h (tree_base): Update comment on using
base.nothrow_flag to represent TYPE_CXX_LOCAL.
* ipa-devirt.c (odr_type_d::whole_program_local): Removed.
(odr_type_d::whole_program_local_p): Check TYPE_CXX_LOCAL flag
on type, and enable WPD at LTRANS when flag_devirtualize_fully
is true.
(get_odr_type): Remove setting whole_program_local flag on type.
(identify_whole_program_local_types): Replace whole_program_local
in odr_type_d by TYPE_CXX_LOCAL on type.
(maybe_record_node): Enable WPD at LTRANS when
flag_devirtualize_fully is true.
* ipa.c (can_remove_vtable_if_no_refs_p): Retain vtables at LTRANS
stage under full devirtualization.
* lto-cgraph.c (compute_ltrans_boundary): Add all defined vtables
to boundary of each LTRANS partition.
* lto-streamer-out.c (get_symbol_initial_value): Streaming out
initial value of vtable even its class is optimized away.
* lto-lang.c (lto_post_options): Disable full devirtualization
if flag_ltrans_devirtualize is false.
* tree-streamer-in.c (unpack_ts_base_value_fields): unpack value
of TYPE_CXX_LOCAL for a type from streaming data.
* tree-streamer-out.c (pack_ts_base_value_fields): pack value
ofTYPE_CXX_LOCAL for a type into streaming data.
---
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0002-WPD-Enable-whole-program-devirtualization-at-LTRANS.patch --]
[-- Type: text/x-patch; name="0002-WPD-Enable-whole-program-devirtualization-at-LTRANS.patch", Size: 12582 bytes --]
From 624aef44d72799ae488a431b4dce730f4b0fc28e Mon Sep 17 00:00:00 2001
From: Feng Xue <fxue@os.amperecomputing.com>
Date: Mon, 6 Sep 2021 20:34:50 +0800
Subject: [PATCH 2/2] WPD: Enable whole program devirtualization at LTRANS
Whole program assumption would not hold when WPA splits whole compilation
into more than one LTRANS partitions. To avoid information lost for WPD
at LTRANS, we will record all vtable nodes and related member function
references into each partition.
2021-09-07 Feng Xue <fxue@os.amperecomputing.com>
gcc/
* tree.h (TYPE_CXX_LOCAL): New macro for type using
base.nothrow_flag.
* tree-core.h (tree_base): Update comment on using
base.nothrow_flag to represent TYPE_CXX_LOCAL.
* ipa-devirt.c (odr_type_d::whole_program_local): Removed.
(odr_type_d::whole_program_local_p): Check TYPE_CXX_LOCAL flag
on type, and enable WPD at LTRANS when flag_devirtualize_fully
is true.
(get_odr_type): Remove setting whole_program_local flag on type.
(identify_whole_program_local_types): Replace whole_program_local
in odr_type_d by TYPE_CXX_LOCAL on type.
(maybe_record_node): Enable WPD at LTRANS when
flag_devirtualize_fully is true.
* ipa.c (can_remove_vtable_if_no_refs_p): Retain vtables at LTRANS
stage under full devirtualization.
* lto-cgraph.c (compute_ltrans_boundary): Add all defined vtables
to boundary of each LTRANS partition.
* lto-streamer-out.c (get_symbol_initial_value): Streaming out
initial value of vtable even its class is optimized away.
* lto-streamer-in.c (lto_input_tree): There might be more than
one decls in dref_queue, register debuginfo for all of them.
* lto-lang.c (lto_post_options): Disable full devirtualization
if flag_ltrans_devirtualize is false.
* tree-streamer-in.c (unpack_ts_base_value_fields): unpack value
of TYPE_CXX_LOCAL for a type from streaming data.
* tree-streamer-out.c (pack_ts_base_value_fields): pack value
ofTYPE_CXX_LOCAL for a type into streaming data.
---
gcc/ipa-devirt.c | 33 ++++++++++++++++++---------------
gcc/ipa.c | 7 ++++++-
gcc/lto-cgraph.c | 19 +++++++++++++++++++
gcc/lto-streamer-in.c | 3 +--
gcc/lto-streamer-out.c | 12 +++++++++++-
gcc/lto/lto-lang.c | 6 ++++++
gcc/tree-core.h | 3 +++
gcc/tree-streamer-in.c | 11 ++++++++---
gcc/tree-streamer-out.c | 16 +++++++++++++---
gcc/tree.h | 5 +++++
10 files changed, 90 insertions(+), 25 deletions(-)
diff --git a/gcc/ipa-devirt.c b/gcc/ipa-devirt.c
index 284c449c6c1..bb929f016f8 100644
--- a/gcc/ipa-devirt.c
+++ b/gcc/ipa-devirt.c
@@ -216,8 +216,6 @@ struct GTY(()) odr_type_d
int id;
/* Is it in anonymous namespace? */
bool anonymous_namespace;
- /* Set when type is not used outside of program. */
- bool whole_program_local;
/* Did we report ODR violation here? */
bool odr_violated;
/* Set when virtual table without RTTI prevailed table with. */
@@ -290,10 +288,15 @@ get_type_vtable (tree type)
bool
odr_type_d::whole_program_local_p ()
{
- if (flag_ltrans)
- return false;
-
- return whole_program_local;
+ if (in_lto_p)
+ return TYPE_CXX_LOCAL (type);
+
+ /* Although a local class is always considered as whole program local in
+ LGEN stage, but may not in LTO stage if multiple duplicated primary
+ vtables are attached to the class due to C++ privatizing via -fno-weak.
+ Thus, we can not set TYPE_CXX_LOCAL flag for local class at LGEN stage
+ when building ORD type. */
+ return anonymous_namespace || decl_function_context (TYPE_NAME (type));
}
/* Return TRUE if ODR type may have any instance. */
@@ -2007,11 +2010,6 @@ get_odr_type (tree type, bool insert)
else
val->anonymous_namespace = 0;
- if (!in_lto_p
- && (val->anonymous_namespace
- || decl_function_context (TYPE_NAME (type))))
- val->whole_program_local = true;
-
build_bases = COMPLETE_TYPE_P (val->type);
insert_to_odr_array = true;
*slot = val;
@@ -2630,17 +2628,22 @@ identify_whole_program_local_types (void)
if (multi_vtable_p)
continue;
- if (is_final)
+ /* Mark all equivalent types in the ODR type as whole program local,
+ because representative type of the ODR type at LTRANS might not
+ be the one at WPA. */
+ FOR_EACH_VEC_ELT (*(type->types), i, equiv_type)
{
+ TYPE_CXX_LOCAL (equiv_type) = 1;
+
/* If no derivation, whole program local type is marked as
FINAL, which could help devirtualization at LTRANS even
there will be more than one partitions. */
- FOR_EACH_VEC_ELT (*(type->types), i, equiv_type)
+ if (is_final)
TYPE_FINAL_P (equiv_type) = 1;
}
}
- type->whole_program_local = true;
+ TYPE_CXX_LOCAL (type->type) = 1;
if (is_final)
TYPE_FINAL_P (type->type) = 1;
}
@@ -2759,7 +2762,7 @@ maybe_record_node (vec <cgraph_node *> &nodes,
Currently we ignore these functions in speculative devirtualization.
??? Maybe it would make sense to be more aggressive for LTO even
elsewhere. */
- if (!flag_ltrans
+ if ((!flag_ltrans || flag_devirtualize_fully)
&& !pure_virtual
&& type_in_anonymous_namespace_p (DECL_CONTEXT (target))
&& (!target_node
diff --git a/gcc/ipa.c b/gcc/ipa.c
index 6e200a906b5..52d6a9e63be 100644
--- a/gcc/ipa.c
+++ b/gcc/ipa.c
@@ -268,7 +268,7 @@ can_remove_vtable_if_no_refs_p (varpool_node *vnode)
if (!flag_devirtualize_fully)
return true;
- if (DECL_EXTERNAL (vnode->decl))
+ if (DECL_EXTERNAL (vnode->decl) && !vnode->in_other_partition)
return true;
/* We will force generating vtables in LGEN stage even they are "unused",
@@ -276,6 +276,11 @@ can_remove_vtable_if_no_refs_p (varpool_node *vnode)
if (!in_lto_p && flag_generate_lto)
return false;
+ /* All vtables seen at LTRANS stage are result of dead class elimination
+ at WPA, no need to prune them further. */
+ if (flag_ltrans)
+ return false;
+
return true;
}
diff --git a/gcc/lto-cgraph.c b/gcc/lto-cgraph.c
index 7c3e276a8ea..fd272c4ffdf 100644
--- a/gcc/lto-cgraph.c
+++ b/gcc/lto-cgraph.c
@@ -859,6 +859,25 @@ compute_ltrans_boundary (lto_symtab_encoder_t in_encoder)
lto_set_symtab_encoder_encode_initializer (encoder, vnode);
create_references (encoder, vnode);
}
+
+ /* Add all defined vtables to streaming encoder, so that we could
+ reconstruct complete class hierarchy at LTRANS for any whole-program
+ local type as what it was at WPA to enable late full devirtualization.
+
+ TODO: For a partition, only need to encode a subset of vtables, not all
+ of them. Some class hierarchy that is not referenced in the partition
+ could be excluded. */
+ if (flag_devirtualize_fully && flag_ltrans_devirtualize && flag_wpa)
+ {
+ varpool_node *vnode;
+
+ FOR_EACH_DEFINED_VARIABLE (vnode)
+ if (DECL_VIRTUAL_P (vnode->decl) && !DECL_EXTERNAL (vnode->decl)
+ && TYPE_CXX_LOCAL (DECL_CONTEXT (vnode->decl))
+ && lto_symtab_encoder_lookup (encoder, vnode) == LCC_NOT_FOUND)
+ lto_symtab_encoder_encode (encoder, vnode);
+ }
+
/* Pickle in also the initializer of all referenced readonly variables
to help folding. Constant pool variables are not shared, so we must
pickle those too. */
diff --git a/gcc/lto-streamer-in.c b/gcc/lto-streamer-in.c
index eb8a7dc57b0..b0902a28d59 100644
--- a/gcc/lto-streamer-in.c
+++ b/gcc/lto-streamer-in.c
@@ -1905,11 +1905,10 @@ lto_input_tree (class lto_input_block *ib, class data_in *data_in)
}
tree t = lto_input_tree_1 (ib, data_in, tag, 0);
- if (!dref_queue.is_empty ())
+ while (!dref_queue.is_empty ())
{
dref_entry e = dref_queue.pop ();
debug_hooks->register_external_die (e.decl, e.sym, e.off);
- gcc_checking_assert (dref_queue.is_empty ());
}
return t;
}
diff --git a/gcc/lto-streamer-out.c b/gcc/lto-streamer-out.c
index a26d4885800..c36efb2e25c 100644
--- a/gcc/lto-streamer-out.c
+++ b/gcc/lto-streamer-out.c
@@ -437,7 +437,17 @@ get_symbol_initial_value (lto_symtab_encoder_t encoder, tree expr)
scalar values. */
if (!(vnode = varpool_node::get (expr))
|| !lto_symtab_encoder_encode_initializer_p (encoder, vnode))
- initial = error_mark_node;
+ {
+ /* Even when a class is optimized away, devirtualization at LTRANS
+ still needs to extract addresses of member virtual functions from
+ initial value of vtable. */
+ if (flag_devirtualize_fully && flag_ltrans_devirtualize
+ && flag_wpa && DECL_VIRTUAL_P (expr) && !DECL_EXTERNAL (expr))
+ return initial;
+
+ initial = error_mark_node;
+ }
+
if (initial != error_mark_node)
{
long max_size = 30;
diff --git a/gcc/lto/lto-lang.c b/gcc/lto/lto-lang.c
index c13c7e45ac1..a48b55f7ada 100644
--- a/gcc/lto/lto-lang.c
+++ b/gcc/lto/lto-lang.c
@@ -870,6 +870,12 @@ lto_post_options (const char **pfilename ATTRIBUTE_UNUSED)
/* During LTRANS, we are not looking at the whole program, only
a subset of the whole callgraph. */
flag_whole_program = 0;
+
+ /* Since full devirtualization at LTRANS requires streaming extra data
+ on vtables, it only takes effect when flag_ltrans_devirtualize is
+ also on. */
+ if (!flag_ltrans_devirtualize)
+ flag_devirtualize_fully = 0;
}
if (flag_wpa)
diff --git a/gcc/tree-core.h b/gcc/tree-core.h
index 4498c3ad127..08d59af022b 100644
--- a/gcc/tree-core.h
+++ b/gcc/tree-core.h
@@ -1328,6 +1328,9 @@ struct GTY(()) tree_base {
DECL_NONALIASED in
VAR_DECL
+ TYPE_CXX_LOCAL in
+ all types
+
deprecated_flag:
TREE_DEPRECATED in
diff --git a/gcc/tree-streamer-in.c b/gcc/tree-streamer-in.c
index 984b1e269cf..1597fc102b0 100644
--- a/gcc/tree-streamer-in.c
+++ b/gcc/tree-streamer-in.c
@@ -129,10 +129,15 @@ unpack_ts_base_value_fields (struct bitpack_d *bp, tree expr)
bp_unpack_value (bp, 1);
TREE_ASM_WRITTEN (expr) = (unsigned) bp_unpack_value (bp, 1);
if (TYPE_P (expr))
- TYPE_ARTIFICIAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ {
+ TYPE_ARTIFICIAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ TYPE_CXX_LOCAL (expr) = (unsigned) bp_unpack_value (bp, 1);
+ }
else
- TREE_NO_WARNING (expr) = (unsigned) bp_unpack_value (bp, 1);
- TREE_NOTHROW (expr) = (unsigned) bp_unpack_value (bp, 1);
+ {
+ TREE_NO_WARNING (expr) = (unsigned) bp_unpack_value (bp, 1);
+ TREE_NOTHROW (expr) = (unsigned) bp_unpack_value (bp, 1);
+ }
TREE_STATIC (expr) = (unsigned) bp_unpack_value (bp, 1);
if (TREE_CODE (expr) != TREE_BINFO)
TREE_PRIVATE (expr) = (unsigned) bp_unpack_value (bp, 1);
diff --git a/gcc/tree-streamer-out.c b/gcc/tree-streamer-out.c
index 1a43534d117..151df839886 100644
--- a/gcc/tree-streamer-out.c
+++ b/gcc/tree-streamer-out.c
@@ -102,10 +102,20 @@ pack_ts_base_value_fields (struct bitpack_d *bp, tree expr)
bp_pack_value (bp, (TREE_CODE (expr) != SSA_NAME
? 0 : TREE_ASM_WRITTEN (expr)), 1);
if (TYPE_P (expr))
- bp_pack_value (bp, TYPE_ARTIFICIAL (expr), 1);
+ {
+ bp_pack_value (bp, TYPE_ARTIFICIAL (expr), 1);
+ /* Clear whole-program local flag of type if full devirtualization is
+ disabled at LTRANS stage. */
+ if (flag_wpa && (!flag_devirtualize_fully || !flag_ltrans_devirtualize))
+ bp_pack_value (bp, 0, 1);
+ else
+ bp_pack_value (bp, TYPE_CXX_LOCAL (expr), 1);
+ }
else
- bp_pack_value (bp, TREE_NO_WARNING (expr), 1);
- bp_pack_value (bp, TREE_NOTHROW (expr), 1);
+ {
+ bp_pack_value (bp, TREE_NO_WARNING (expr), 1);
+ bp_pack_value (bp, TREE_NOTHROW (expr), 1);
+ }
bp_pack_value (bp, TREE_STATIC (expr), 1);
if (TREE_CODE (expr) != TREE_BINFO)
bp_pack_value (bp, TREE_PRIVATE (expr), 1);
diff --git a/gcc/tree.h b/gcc/tree.h
index e5547e8491c..6842de7a24b 100644
--- a/gcc/tree.h
+++ b/gcc/tree.h
@@ -2254,6 +2254,11 @@ extern tree vector_element_bits_tree (const_tree);
#define TYPE_FINAL_P(NODE) \
(RECORD_OR_UNION_CHECK (NODE)->base.default_def_flag)
+/* If nonzero, it indicates that a C++ type (mainly polymorphic class) may
+ be referenced outside the program being compiled. This is used in C++
+ devirtualization. */
+#define TYPE_CXX_LOCAL(NODE) (TYPE_CHECK (NODE)->base.nothrow_flag)
+
/* The debug output functions use the symtab union field to store
information specific to the debugging format. The different debug
output hooks store different types in the union field. These three
--
2.17.1
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2021-10-15 3:50 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-09-16 9:26 [PATCH/RFC 2/2] WPD: Enable whole program devirtualization at LTRANS Feng Xue OS
2021-09-30 3:59 ` PING: " Feng Xue OS
2021-10-15 3:50 ` PING^2: " Feng Xue OS
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).