Compare commits

..

No commits in common. "a120a887ddf4e2d8b695b95cd04c1696943e8649" and "478230ea665f0fa78f9df3bdfea413ae32e204f2" have entirely different histories.

7 changed files with 2 additions and 1448 deletions

View File

@ -1,204 +0,0 @@
From a145d61f5ed0aba1aa0a4adc47afcedf8eaf7579 Mon Sep 17 00:00:00 2001
From: "Vladimir N. Makarov" <vmakarov@redhat.com>
Date: Fri, 8 May 2020 16:51:40 -0400
Subject: [PATCH] [Backport] Improve hard reg preference propapagation.
Reference: https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=74dc179a6da33cd00f6d4a93fbb97dc84f610126
2020-05-08 Vladimir Makarov <vmakarov@redhat.com>
* ira-color.c (update_costs_from_allocno): Remove
conflict_cost_update_p argument. Propagate costs only along
threads. Always do conflict cost update. Add printing debugging
info.
(update_costs_from_copies): Add printing debugging info.
(restore_costs_from_copies): Ditto.
(assign_hard_reg): Improve debug info.
(push_only_colorable): Ditto. Call update_costs_from_prefs.
(color_allocnos): Remove update_costs_from_prefs.
2020-05-08 Vladimir Makarov <vmakarov@redhat.com>
* gcc.target/i386/pr92807-1.c: Improve the regex.
---
gcc/ira-color.c | 63 +++++++++++++++--------
gcc/testsuite/gcc.target/i386/pr92807-1.c | 2 +-
2 files changed, 42 insertions(+), 23 deletions(-)
diff --git a/gcc/ira-color.c b/gcc/ira-color.c
index b0fc159a8..536ae8da2 100644
--- a/gcc/ira-color.c
+++ b/gcc/ira-color.c
@@ -1356,13 +1356,11 @@ allocnos_conflict_p (ira_allocno_t a1, ira_allocno_t a2)
/* Update (decrease if DECR_P) HARD_REGNO cost of allocnos connected
by copies to ALLOCNO to increase chances to remove some copies as
- the result of subsequent assignment. Update conflict costs only
- for true CONFLICT_COST_UPDATE_P. Record cost updates if RECORD_P is
- true. */
+ the result of subsequent assignment. Update conflict costs.
+ Record cost updates if RECORD_P is true. */
static void
update_costs_from_allocno (ira_allocno_t allocno, int hard_regno,
- int divisor, bool decr_p, bool record_p,
- bool conflict_cost_update_p)
+ int divisor, bool decr_p, bool record_p)
{
int cost, update_cost, update_conflict_cost;
machine_mode mode;
@@ -1391,7 +1389,9 @@ update_costs_from_allocno (ira_allocno_t allocno, int hard_regno,
gcc_unreachable ();
if (another_allocno == from
- || allocnos_conflict_p (another_allocno, start))
+ || (ALLOCNO_COLOR_DATA (another_allocno) != NULL
+ && (ALLOCNO_COLOR_DATA (allocno)->first_thread_allocno
+ != ALLOCNO_COLOR_DATA (another_allocno)->first_thread_allocno)))
continue;
aclass = ALLOCNO_CLASS (another_allocno);
@@ -1419,15 +1419,13 @@ update_costs_from_allocno (ira_allocno_t allocno, int hard_regno,
cost = -cost;
update_cost = cp->freq * cost / divisor;
- update_conflict_cost = conflict_cost_update_p ? update_cost : 0;
-
- if (ALLOCNO_COLOR_DATA (another_allocno) != NULL
- && (ALLOCNO_COLOR_DATA (allocno)->first_thread_allocno
- != ALLOCNO_COLOR_DATA (another_allocno)->first_thread_allocno))
- /* Decrease conflict cost of ANOTHER_ALLOCNO if it is not
- in the same allocation thread. */
- update_conflict_cost /= COST_HOP_DIVISOR;
+ update_conflict_cost = update_cost;
+ if (internal_flag_ira_verbose > 5 && ira_dump_file != NULL)
+ fprintf (ira_dump_file,
+ " a%dr%d (hr%d): update cost by %d, conflict cost by %d\n",
+ ALLOCNO_NUM (another_allocno), ALLOCNO_REGNO (another_allocno),
+ hard_regno, update_cost, update_conflict_cost);
if (update_cost == 0)
continue;
@@ -1455,8 +1453,13 @@ update_costs_from_prefs (ira_allocno_t allocno)
start_update_cost ();
for (pref = ALLOCNO_PREFS (allocno); pref != NULL; pref = pref->next_pref)
- update_costs_from_allocno (allocno, pref->hard_regno,
- COST_HOP_DIVISOR, true, true, false);
+ {
+ if (internal_flag_ira_verbose > 5 && ira_dump_file != NULL)
+ fprintf (ira_dump_file, " Start updating from pref of hr%d for a%dr%d:\n",
+ pref->hard_regno, ALLOCNO_NUM (allocno), ALLOCNO_REGNO (allocno));
+ update_costs_from_allocno (allocno, pref->hard_regno,
+ COST_HOP_DIVISOR, true, true);
+ }
}
/* Update (decrease if DECR_P) the cost of allocnos connected to
@@ -1471,7 +1474,10 @@ update_costs_from_copies (ira_allocno_t allocno, bool decr_p, bool record_p)
hard_regno = ALLOCNO_HARD_REGNO (allocno);
ira_assert (hard_regno >= 0 && ALLOCNO_CLASS (allocno) != NO_REGS);
start_update_cost ();
- update_costs_from_allocno (allocno, hard_regno, 1, decr_p, record_p, true);
+ if (internal_flag_ira_verbose > 5 && ira_dump_file != NULL)
+ fprintf (ira_dump_file, " Start updating from a%dr%d by copies:\n",
+ ALLOCNO_NUM (allocno), ALLOCNO_REGNO (allocno));
+ update_costs_from_allocno (allocno, hard_regno, 1, decr_p, record_p);
}
/* Update conflict_allocno_hard_prefs of allocnos conflicting with
@@ -1519,9 +1525,12 @@ restore_costs_from_copies (ira_allocno_t allocno)
return;
records = ALLOCNO_COLOR_DATA (allocno)->update_cost_records;
start_update_cost ();
+ if (internal_flag_ira_verbose > 5 && ira_dump_file != NULL)
+ fprintf (ira_dump_file, " Start restoring from a%dr%d:\n",
+ ALLOCNO_NUM (allocno), ALLOCNO_REGNO (allocno));
for (curr = records; curr != NULL; curr = curr->next)
update_costs_from_allocno (allocno, curr->hard_regno,
- curr->divisor, true, false, true);
+ curr->divisor, true, false);
free_update_cost_record_list (records);
ALLOCNO_COLOR_DATA (allocno)->update_cost_records = NULL;
}
@@ -1936,6 +1945,8 @@ assign_hard_reg (ira_allocno_t a, bool retry_p)
if (internal_flag_ira_verbose > 5 && ira_dump_file != NULL)
fprintf (ira_dump_file, "(%d=%d,%d) ", hard_regno, cost, full_cost);
}
+ if (internal_flag_ira_verbose > 5 && ira_dump_file != NULL)
+ fprintf (ira_dump_file, "\n");
if (min_full_cost > mem_cost
/* Do not spill static chain pointer pseudo when non-local goto
is used. */
@@ -2114,7 +2125,7 @@ form_threads_from_copies (int cp_num)
if (internal_flag_ira_verbose > 3 && ira_dump_file != NULL)
fprintf
(ira_dump_file,
- " Forming thread by copy %d:a%dr%d-a%dr%d (freq=%d):\n",
+ " Forming thread by copy %d:a%dr%d-a%dr%d (freq=%d):\n",
cp->num, ALLOCNO_NUM (cp->first), ALLOCNO_REGNO (cp->first),
ALLOCNO_NUM (cp->second), ALLOCNO_REGNO (cp->second),
cp->freq);
@@ -2122,7 +2133,7 @@ form_threads_from_copies (int cp_num)
if (internal_flag_ira_verbose > 3 && ira_dump_file != NULL)
{
thread1 = ALLOCNO_COLOR_DATA (thread1)->first_thread_allocno;
- fprintf (ira_dump_file, " Result (freq=%d): a%dr%d(%d)",
+ fprintf (ira_dump_file, " Result (freq=%d): a%dr%d(%d)",
ALLOCNO_COLOR_DATA (thread1)->thread_freq,
ALLOCNO_NUM (thread1), ALLOCNO_REGNO (thread1),
ALLOCNO_FREQ (thread1));
@@ -2186,6 +2197,9 @@ form_threads_from_colorable_allocno (ira_allocno_t a)
ira_copy_t cp, next_cp;
int cp_num = 0;
+ if (internal_flag_ira_verbose > 3 && ira_dump_file != NULL)
+ fprintf (ira_dump_file, " Forming thread from allocno a%dr%d:\n",
+ ALLOCNO_NUM (a), ALLOCNO_REGNO (a));
for (cp = ALLOCNO_COPIES (a); cp != NULL; cp = next_cp)
{
if (cp->first == a)
@@ -2507,7 +2521,13 @@ remove_allocno_from_bucket_and_push (ira_allocno_t allocno, bool colorable_p)
static void
push_only_colorable (void)
{
+ if (internal_flag_ira_verbose > 3 && ira_dump_file != NULL)
+ fprintf (ira_dump_file, " Forming thread from colorable bucket:\n");
form_threads_from_bucket (colorable_allocno_bucket);
+ for (ira_allocno_t a = colorable_allocno_bucket;
+ a != NULL;
+ a = ALLOCNO_COLOR_DATA (a)->next_bucket_allocno)
+ update_costs_from_prefs (a);
sort_bucket (&colorable_allocno_bucket, bucket_allocno_compare_func);
for (;colorable_allocno_bucket != NULL;)
remove_allocno_from_bucket_and_push (colorable_allocno_bucket, true);
@@ -2689,7 +2709,7 @@ pop_allocnos_from_stack (void)
else if (assign_hard_reg (allocno, false))
{
if (internal_flag_ira_verbose > 3 && ira_dump_file != NULL)
- fprintf (ira_dump_file, "assign reg %d\n",
+ fprintf (ira_dump_file, " assign reg %d\n",
ALLOCNO_HARD_REGNO (allocno));
}
else if (ALLOCNO_ASSIGNED_P (allocno))
@@ -3204,7 +3224,6 @@ color_allocnos (void)
if (ALLOCNO_CLASS (a) != NO_REGS && ! empty_profitable_hard_regs (a))
{
ALLOCNO_COLOR_DATA (a)->in_graph_p = true;
- update_costs_from_prefs (a);
update_conflict_allocno_hard_prefs (a);
}
else
diff --git a/gcc/testsuite/gcc.target/i386/pr92807-1.c b/gcc/testsuite/gcc.target/i386/pr92807-1.c
index 00f92930a..02a0654dd 100644
--- a/gcc/testsuite/gcc.target/i386/pr92807-1.c
+++ b/gcc/testsuite/gcc.target/i386/pr92807-1.c
@@ -8,4 +8,4 @@ abs2 (unsigned int a)
return (a+s)^s;
}
-/* { dg-final { scan-assembler-not "leal" } } */
+/* { dg-final { scan-assembler-not "leal\[\t \]*\[^(\]*\\((.*),\\1\\)" } } */
--
2.38.1.windows.1

View File

@ -1,421 +0,0 @@
From 46ee0d6975bbea6246b83db08aa105c93d5fe3ef Mon Sep 17 00:00:00 2001
From: rfwang07 <wangrufeng5@huawei.com>
Date: Sat, 15 Jun 2024 12:59:51 +0800
Subject: [PATCH] Support LTO in AutoBOLT mode
---
bolt-plugin/bolt-plugin.cc | 35 +++++--
gcc/common.opt | 4 +-
gcc/final.c | 191 ++++++++++++++++++++++++++++++++-----
gcc/opts.c | 7 +-
4 files changed, 196 insertions(+), 41 deletions(-)
diff --git a/bolt-plugin/bolt-plugin.cc b/bolt-plugin/bolt-plugin.cc
index f357b00dd..06cf1911a 100644
--- a/bolt-plugin/bolt-plugin.cc
+++ b/bolt-plugin/bolt-plugin.cc
@@ -269,6 +269,8 @@ static string tmp_out_file_name = "a.out";
/* Binary or dynamic file after BOLT. */
static string bolt_opt_target;
+static bool autobolt_with_lto = false;
+
/* Format of bolt_optimize_options should be "reorder-functions=hfsort+ ...",
command 'llvm-bolt' has been added here. */
static string bolt_optimize_options ("llvm-bolt ");
@@ -450,9 +452,9 @@ cleanup_handler ()
fclose (bolt_file_fd);
}
- if (file_exist (tmp_out_file_name.c_str ())
- && file_exist (bolt_profile_name.c_str ())
- && is_bolt_opt_target ())
+ if (is_bolt_opt_target ()
+ && file_exist (tmp_out_file_name.c_str ())
+ && file_exist (bolt_profile_name.c_str ()))
{
do_bolt_opt ();
}
@@ -552,6 +554,10 @@ dump_func_to_bolt_profile_file (const struct func_info &func)
static enum ld_plugin_status
all_symbols_read_handler ()
{
+ if (autobolt_with_lto)
+ {
+ return LDPS_OK;
+ }
for (const auto &functions: weak_functions)
{
/* More than one weak function. */
@@ -748,7 +754,10 @@ claim_file_handler (const struct ld_plugin_input_file *file, int *claimed)
}
/* BOLT plugin does not need claimd number, so set *claimed to 0. */
*claimed = 0;
-
+ if (autobolt_with_lto)
+ {
+ return LDPS_OK;
+ }
obj.file = file;
obj.objfile = simple_object_start_read (file->fd, file->offset, NULL,
&errmsg, &err);
@@ -823,9 +832,12 @@ generate_bolt_profile_name (string file_name)
{
if (!bolt_dir_path.empty ())
{
- file_name = concat (get_current_dir_name (),
- separator, file_name.c_str (), NULL);
- file_name = mangle_path (file_name);
+ if (!autobolt_with_lto)
+ {
+ file_name = concat (get_current_dir_name (),
+ separator, file_name.c_str (), NULL);
+ file_name = mangle_path (file_name);
+ }
}
else
{
@@ -1019,7 +1031,10 @@ process_output_option (const string &flag_o)
/* bolt_profile_name may be overridden in
function process_auto_bolt_option and
process_bolt_use_option. */
- bolt_profile_name = gcc_options[o_index + 1];
+ if (autobolt_with_lto && bolt_opt_target.empty ())
+ bolt_profile_name = "default";
+ else
+ bolt_profile_name = gcc_options[o_index + 1];
bolt_profile_name.append (DEFAULT_BOLT_OUT_NAME_SUFFIX);
}
else
@@ -1047,13 +1062,15 @@ process_gcc_option ()
char *collect_gcc_option = getenv ("COLLECT_GCC_OPTIONS");
get_options_from_collect_gcc_options (collect_gcc, collect_gcc_option);
+ autobolt_with_lto = (match_gcc_option (flag_auto_bolt.c_str ()) != -1
+ && match_gcc_option ("-flto") != -1);
+ process_bolt_target_option (flag_bolt_target);
/* Function process_output_option should be processed before
process_auto_bolt_option to obtain correct bolt_profile_name. */
process_output_option (flag_o);
process_auto_bolt_option (flag_auto_bolt);
process_bolt_use_option (flag_bolt_use);
- process_bolt_target_option (flag_bolt_target);
process_bolt_option (flag_bolt_optimize_options);
if (match_gcc_option (flag_profile_use.c_str ()) != -1)
diff --git a/gcc/common.opt b/gcc/common.opt
index ea55355be..aad6fb281 100644
--- a/gcc/common.opt
+++ b/gcc/common.opt
@@ -2482,7 +2482,7 @@ Common Report Var(flag_auto_bolt)
Generate profile from AutoFDO or PGO and do BOLT optimization after linkage.
fauto-bolt=
-Common Joined RejectNegative
+Common Joined RejectNegative Var(auto_bolt)
Specify the feedback data directory required by BOLT-plugin. The default is the current directory.
fbolt-use
@@ -2494,7 +2494,7 @@ Common Joined RejectNegative Var
Do BOLT optimization after linkage with BOLT profile read from this option.
fbolt-target=
-Common Joined RejectNegative Var
+Common Joined RejectNegative Var(bolt_target)
Specify the BOLT optimization target binary.
fbolt-option=
diff --git a/gcc/final.c b/gcc/final.c
index da8d20958..807384514 100644
--- a/gcc/final.c
+++ b/gcc/final.c
@@ -82,6 +82,7 @@ along with GCC; see the file COPYING3. If not see
#include "print-rtl.h"
#include "function-abi.h"
#include "insn-codes.h"
+#include <sys/file.h>
#ifdef XCOFF_DEBUGGING_INFO
#include "xcoffout.h" /* Needed for external data declarations. */
@@ -4650,6 +4651,10 @@ leaf_renumber_regs_insn (rtx in_rtx)
#define ASM_FDO_CALLEE_FLAG ".fdo.callee "
+static bool autobolt_with_lto = false;
+static char *autobolt_curr_func_name;
+static FILE *bolt_file_fd = NULL;
+
/* Return the relative offset address of the start instruction of BB,
return -1 if it is empty instruction. */
@@ -4785,6 +4790,16 @@ alias_local_functions (const char *fnname)
return concat (fnname, "/", lbasename (dump_base_name), NULL);
}
+static char*
+add_suffix (const char *str)
+{
+ if (strlen (str) == 0 || strstr (str, "/") == NULL)
+ {
+ return xstrdup (str);
+ }
+ return concat (str, "/1", NULL);
+}
+
/* Return function bind type string. */
static const char *
@@ -4830,19 +4845,33 @@ dump_direct_callee_info_to_asm (basic_block bb, gcov_type call_count)
if (callee)
{
- char *func_name =
- alias_local_functions (get_fnname_from_decl (callee));
- fprintf (asm_out_file, "\t.string \"%x\"\n",
- INSN_ADDRESSES (INSN_UID (insn)));
-
- fprintf (asm_out_file, "\t.string \"%s%s\"\n",
- ASM_FDO_CALLEE_FLAG,
- func_name);
-
- fprintf (asm_out_file,
- "\t.string \"" HOST_WIDE_INT_PRINT_DEC "\"\n",
- call_count);
-
+ char *func_name;
+ if (!autobolt_with_lto)
+ {
+ func_name = alias_local_functions (
+ get_fnname_from_decl (callee));
+ fprintf (asm_out_file, "\t.string \"%x\"\n",
+ INSN_ADDRESSES (INSN_UID (insn)));
+
+ fprintf (asm_out_file, "\t.string \"%s%s\"\n",
+ ASM_FDO_CALLEE_FLAG,
+ func_name);
+
+ fprintf (asm_out_file,
+ "\t.string \"" HOST_WIDE_INT_PRINT_DEC "\"\n",
+ call_count);
+ }
+ else
+ {
+ func_name = xstrdup (
+ get_fnname_from_decl (callee));
+ fprintf (bolt_file_fd,
+ "1 %s %x 1 %s 0 0 " HOST_WIDE_INT_PRINT_DEC "\n",
+ autobolt_curr_func_name,
+ INSN_ADDRESSES (INSN_UID (insn)),
+ func_name,
+ call_count);
+ }
if (dump_file)
{
fprintf (dump_file, "call: %x --> %s\n",
@@ -4880,7 +4909,10 @@ dump_edge_jump_info_to_asm (basic_block bb, gcov_type bb_count)
/* This is a reserved assert for the original design. If this
assert is found, use the address of the previous instruction
as edge_start_addr. */
- gcc_assert (edge_start_addr != edge_end_addr);
+ if (edge_start_addr == edge_end_addr)
+ {
+ continue;
+ }
if (dump_file)
{
@@ -4890,10 +4922,24 @@ dump_edge_jump_info_to_asm (basic_block bb, gcov_type bb_count)
if (edge_count > 0)
{
- fprintf (asm_out_file, "\t.string \"%x\"\n", edge_start_addr);
- fprintf (asm_out_file, "\t.string \"%x\"\n", edge_end_addr);
- fprintf (asm_out_file, "\t.string \"" HOST_WIDE_INT_PRINT_DEC "\"\n",
- edge_count);
+ if (!autobolt_with_lto)
+ {
+ fprintf (asm_out_file, "\t.string \"%x\"\n", edge_start_addr);
+ fprintf (asm_out_file, "\t.string \"%x\"\n", edge_end_addr);
+ fprintf (asm_out_file,
+ "\t.string \"" HOST_WIDE_INT_PRINT_DEC "\"\n",
+ edge_count);
+ }
+ else
+ {
+ fprintf (bolt_file_fd,
+ "1 %s %x 1 %s %x 0 " HOST_WIDE_INT_PRINT_DEC "\n",
+ autobolt_curr_func_name,
+ edge_start_addr,
+ autobolt_curr_func_name,
+ edge_end_addr,
+ edge_count);
+ }
}
}
@@ -4924,12 +4970,17 @@ static void
dump_function_info_to_asm (const char *fnname)
{
char *func_name = alias_local_functions (fnname);
- fprintf (asm_out_file, "\t.string \"%s%s\"\n",
- ASM_FDO_CALLER_FLAG, func_name);
- fprintf (asm_out_file, "\t.string \"%s%d\"\n",
- ASM_FDO_CALLER_SIZE_FLAG, get_function_end_addr ());
- fprintf (asm_out_file, "\t.string \"%s%s\"\n",
- ASM_FDO_CALLER_BIND_FLAG, simple_get_function_bind ());
+ if (!autobolt_with_lto)
+ {
+ fprintf (asm_out_file, "\t.string \"%s%s\"\n",
+ ASM_FDO_CALLER_FLAG, func_name);
+ fprintf (asm_out_file, "\t.string \"%s%d\"\n",
+ ASM_FDO_CALLER_SIZE_FLAG,
+ get_function_end_addr ());
+ fprintf (asm_out_file, "\t.string \"%s%s\"\n",
+ ASM_FDO_CALLER_BIND_FLAG,
+ simple_get_function_bind ());
+ }
if (dump_file)
{
@@ -5041,6 +5092,90 @@ dump_profile_to_elf_sections ()
}
}
+#define DEFAULT_BOLT_OUT_NAME "default.fdata"
+static char *bolt_profile_name = DEFAULT_BOLT_OUT_NAME;
+static void
+open_bolt_profile_file ()
+{
+ char *file_path;
+
+ if (auto_bolt && strlen (auto_bolt) != 0)
+ file_path = lrealpath (auto_bolt);
+ else
+ file_path = lrealpath (dump_dir_name);
+ if (bolt_target && strlen (bolt_target) != 0)
+ bolt_profile_name = concat (
+ file_path, "/", lbasename (bolt_target), ".fdata", NULL);
+ else
+ bolt_profile_name = concat (
+ file_path, "/", DEFAULT_BOLT_OUT_NAME, NULL);
+ free (file_path);
+
+ if (bolt_file_fd == NULL)
+ {
+ bolt_file_fd = fopen (bolt_profile_name, "at");
+ if (!bolt_file_fd)
+ {
+ error ("Failed to open the file: %s."
+ " Please check whether the target path exists.",
+ bolt_profile_name);
+ }
+ if (flock (fileno (bolt_file_fd), LOCK_EX) == -1)
+ {
+ error ("Failed to lock the file: %s.",
+ bolt_profile_name);
+ }
+ }
+ free (bolt_profile_name);
+}
+
+inline static bool
+is_bolt_opt_target ()
+{
+ bool is_target = true;
+ if (bolt_target && strlen (bolt_target) != 0)
+ {
+ char *target_name = concat (lbasename (bolt_target),
+ ".ltrans", NULL);
+ if (strncmp (target_name,
+ lbasename (dump_base_name),
+ strlen (target_name)) != 0)
+ {
+ is_target = false;
+ }
+ free (target_name);
+ }
+ return is_target;
+}
+
+static void
+dump_profile_to_bolt_file ()
+{
+ /* Avoid empty functions. */
+ if (TREE_CODE (cfun->decl) != FUNCTION_DECL)
+ {
+ return;
+ }
+
+ if (!is_bolt_opt_target ())
+ {
+ return;
+ }
+
+ open_bolt_profile_file ();
+ const char *fnname = get_fnname_from_decl (current_function_decl);
+ autobolt_curr_func_name = add_suffix (fnname);
+ dump_fdo_info_to_asm (fnname);
+ free (autobolt_curr_func_name);
+ fflush (bolt_file_fd);
+ if (flock (fileno (bolt_file_fd), LOCK_UN) == -1)
+ {
+ error ("Failed to unlock the file: %s.", bolt_profile_name);
+ }
+ fclose (bolt_file_fd);
+ bolt_file_fd = NULL;
+}
+
/* Turn the RTL into assembly. */
static unsigned int
rest_of_handle_final (void)
@@ -5111,7 +5246,13 @@ rest_of_handle_final (void)
if (flag_auto_bolt)
{
- dump_profile_to_elf_sections ();
+ if (flag_ltrans)
+ {
+ autobolt_with_lto = true;
+ dump_profile_to_bolt_file ();
+ }
+ else
+ dump_profile_to_elf_sections ();
}
return 0;
diff --git a/gcc/opts.c b/gcc/opts.c
index 30ac57eec..c0ccd0853 100644
--- a/gcc/opts.c
+++ b/gcc/opts.c
@@ -1166,10 +1166,6 @@ finish_options (struct gcc_options *opts, struct gcc_options *opts_set,
if (opts->x_flag_vtable_verify && opts->x_flag_lto)
sorry ("vtable verification is not supported with LTO");
- /* Currently -fauto-bolt is not supported for LTO. */
- if (opts->x_flag_auto_bolt && opts->x_flag_lto)
- sorry ("%<-fauto-bolt%> is not supported with LTO");
-
/* Currently -fbolt-use is not supported for LTO. */
if (opts->x_flag_bolt_use && opts->x_flag_lto)
sorry ("-fbolt-use is not supported with LTO");
@@ -2970,6 +2966,7 @@ common_handle_option (struct gcc_options *opts,
case OPT_fauto_bolt_:
opts->x_flag_auto_bolt = true;
+ opts->x_auto_bolt = xstrdup (arg);
/* FALLTHRU */
case OPT_fauto_bolt:
if (opts->x_flag_bolt_use)
@@ -2985,7 +2982,7 @@ common_handle_option (struct gcc_options *opts,
break;
case OPT_fbolt_target_:
- /* Deferred. */
+ opts->x_bolt_target = xstrdup (arg);
break;
case OPT_fbolt_option_:
--
2.39.2 (Apple Git-143)

View File

@ -1,24 +0,0 @@
From 69cae9164ec85ef416c29baa3dd8b2b11edf8b71 Mon Sep 17 00:00:00 2001
From: yzyssdd <yuzeyang4@huawei.com>
Date: Tue, 18 Jun 2024 11:16:15 +0800
Subject: [PATCH] Fix bug in detecting abnormal revisit for bb
---
gcc/tree-ssa-llc-allocate.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/gcc/tree-ssa-llc-allocate.c b/gcc/tree-ssa-llc-allocate.c
index 3f6ff3623..a1e7b0eb6 100644
--- a/gcc/tree-ssa-llc-allocate.c
+++ b/gcc/tree-ssa-llc-allocate.c
@@ -2348,7 +2348,6 @@ revisit_bb_abnormal_p (basic_block bb, std::vector<int> &bb_visited,
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Already visited bb index %d. Abort.\n",
bb->index);
- unused_edges.insert (std::make_pair (src_bb_idx, bb->index));
return true;
}
--
2.33.0

View File

@ -1,29 +0,0 @@
From 50812a995ffda86acaab56f94ae99732b2a99c85 Mon Sep 17 00:00:00 2001
From: Chernonog Viacheslav <chernonog.vyacheslav@huawei.com>
Date: Wed, 8 May 2024 19:24:27 +0800
Subject: [PATCH] Change set_of to reg_set_p, and add check for global_regs
fix for I9JDHE
---
gcc/rtl-matcher.h | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/gcc/rtl-matcher.h b/gcc/rtl-matcher.h
index 6aed8d98d..5310f6266 100644
--- a/gcc/rtl-matcher.h
+++ b/gcc/rtl-matcher.h
@@ -56,8 +56,9 @@ check_def_chain_ref (df_ref ref, rtx reg)
if (!ref || !DF_REF_INSN_INFO (ref))
return false;
- return !global_regs[REGNO (reg)]
- || set_of (reg, DF_REF_INSN (ref));
+ return !(REGNO (reg) < FIRST_PSEUDO_REGISTER
+ && global_regs[REGNO (reg)])
+ || reg_set_p (reg, DF_REF_INSN (ref));
}
/* Get the single def instruction of the reg being used in the insn. */
--
2.33.0

View File

@ -1,54 +0,0 @@
From 0dcab0809ada30b09f553dbbe5b2df8d830e95f3 Mon Sep 17 00:00:00 2001
From: Pronin Alexander 00812787 <pronin.alexander@huawei.com>
Date: Thu, 13 Jun 2024 22:04:16 +0800
Subject: [PATCH] Remove erroneous pattern from gimple ifcvt
---
gcc/match.pd | 2 +-
gcc/testsuite/gcc.dg/ifcvt-gimple-1.c | 21 +++++++++++++++++++++
2 files changed, 22 insertions(+), 1 deletion(-)
create mode 100644 gcc/testsuite/gcc.dg/ifcvt-gimple-1.c
diff --git a/gcc/match.pd b/gcc/match.pd
index 1097cd926..efd2db209 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -3450,7 +3450,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
)
(if (flag_if_conversion_gimple)
- (for simple_op (plus minus bit_and bit_ior bit_xor)
+ (for simple_op (plus minus bit_ior bit_xor)
(simplify
(cond @0 (simple_op @1 INTEGER_CST@2) @1)
(switch
diff --git a/gcc/testsuite/gcc.dg/ifcvt-gimple-1.c b/gcc/testsuite/gcc.dg/ifcvt-gimple-1.c
new file mode 100644
index 000000000..381a4ad51
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/ifcvt-gimple-1.c
@@ -0,0 +1,21 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -fno-inline -fif-conversion-gimple" } */
+
+#include <stdlib.h>
+
+void foo(int a, int *p) {
+ *p = a;
+}
+
+void verify (int a) {
+ if (a != 3)
+ abort ();
+}
+
+int main() {
+ int a = 0;
+ foo (3, &a);
+ int tmp = (a > 7) ? a & 1 : a;
+ verify (tmp);
+ return 0;
+}
--
2.33.0

View File

@ -1,678 +0,0 @@
From 25014ea924bfe3659e88195636ec08f87dd72c07 Mon Sep 17 00:00:00 2001
From: Mingchuan Wu <wumingchuan1992@foxmail.com>
Date: Fri, 21 Jun 2024 20:26:04 +0800
Subject: [PATCH] Try to use AI model to guide optimization.
---
gcc/Makefile.in | 1 +
gcc/common.opt | 7 ++
gcc/config/aarch64/aarch64.c | 130 ++++++++++++++++++++
gcc/ipa-hardware-detection.c | 228 +++++++++++++++++++++++++++++++++++
gcc/opts-common.c | 154 +++++++++++++++++++++++
gcc/opts.c | 11 ++
gcc/passes.def | 1 +
gcc/timevar.def | 1 +
gcc/tree-pass.h | 2 +
9 files changed, 535 insertions(+)
create mode 100644 gcc/ipa-hardware-detection.c
diff --git a/gcc/Makefile.in b/gcc/Makefile.in
index aed321d27..f21bc5f9a 100644
--- a/gcc/Makefile.in
+++ b/gcc/Makefile.in
@@ -1398,6 +1398,7 @@ OBJS = \
inchash.o \
incpath.o \
init-regs.o \
+ ipa-hardware-detection.o \
internal-fn.o \
ipa-struct-reorg/ipa-struct-reorg.o \
ipa-cp.o \
diff --git a/gcc/common.opt b/gcc/common.opt
index aad6fb281..9b32ea50a 100644
--- a/gcc/common.opt
+++ b/gcc/common.opt
@@ -188,6 +188,9 @@ const char *main_input_basename
Variable
int main_input_baselength
+Variable
+bool optimize_machine
+
; Which options have been printed by --help.
Variable
char *help_printed
@@ -467,6 +470,10 @@ Ofast
Common Optimization
Optimize for speed disregarding exact standards compliance.
+Om
+Common Optimization
+Optimize for radical optimization for machines.
+
Og
Common Optimization
Optimize for debugging experience rather than speed or size.
diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index 2117326ba..e67e77e6a 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -14416,6 +14416,135 @@ aarch64_sve_adjust_stmt_cost (vect_cost_for_stmt kind,
return stmt_cost;
}
+/* Check whether in C language or LTO with only C language. */
+extern bool lang_c_p (void);
+
+static void
+override_C_optimize_options (struct gcc_options *opts)
+{
+ opts->x_flag_ipa_reorder_fields = 1;
+ opts->x_flag_ipa_struct_reorg = 6;
+ opts->x_struct_layout_optimize_level = 6;
+ opts->x_flag_gnu89_inline = 1;
+ opts->x_flag_ccmp2 = 1;
+ opts->x_flag_array_widen_compare = 1;
+ opts->x_flag_convert_minmax = 1;
+ opts->x_flag_tree_slp_transpose_vectorize = 1;
+ opts->x_param_max_inline_insns_auto = 64;
+ opts->x_param_inline_unit_growth = 96;
+ opts->x_flag_cmlt_arith = 1;
+}
+
+/* Check whether in CPP language or LTO with only CPP language. */
+static bool
+lang_cpp_p (void)
+{
+ const char *language_string = lang_hooks.name;
+ if (!language_string)
+ {
+ return false;
+ }
+ if (lang_GNU_CXX ())
+ {
+ return true;
+ }
+ else if (strcmp (language_string, "GNU GIMPLE") == 0) // for LTO check
+ {
+ unsigned i = 0;
+ tree t = NULL_TREE;
+ FOR_EACH_VEC_SAFE_ELT (all_translation_units, i, t)
+ {
+ language_string = TRANSLATION_UNIT_LANGUAGE (t);
+ if (language_string == NULL
+ || strncmp (lang_hooks.name, "GNU C++", 7))
+ {
+ return false;
+ }
+ }
+ return true;
+ }
+ return false;
+}
+
+static void
+override_CPP_optimize_options (struct gcc_options *opts)
+{
+ opts->x_flag_finite_loops = 1;
+ opts->x_flag_omit_frame_pointer = 1;
+ opts->x_flag_sized_deallocation = 0;
+ opts->x_flag_loop_elim = 1;
+ opts->x_flag_convert_minmax = 1;
+ opts->x_param_early_inlining_insns = 256;
+ opts->x_param_max_inline_insns_auto = 128;
+ opts->x_param_inline_unit_growth = 256;
+ opts->x_flag_cmlt_arith = 1;
+}
+
+static void
+override_optimize_options_1 (struct gcc_options *opts)
+{
+ opts->x_flag_split_ldp_stp = 1;
+ opts->x_flag_if_conversion_gimple = 1;
+ opts->x_param_tree_forwprop_perm = 1;
+ opts->x_flag_ifcvt_allow_complicated_cmps = 1;
+ opts->x_param_ifcvt_allow_register_renaming = 2;
+ opts->x_param_max_rtl_if_conversion_unpredictable_cost = 48;
+ opts->x_param_max_rtl_if_conversion_predictable_cost = 48;
+}
+
+static void
+override_Fortran_optimize_options (struct gcc_options *opts)
+{
+ opts->x_flag_unroll_loops = 1;
+ opts->x_flag_unconstrained_commons = 1;
+ opts->x_param_ipa_cp_eval_threshold = 1;
+ opts->x_param_ipa_cp_unit_growth = 80;
+ opts->x_param_ipa_cp_max_recursive_depth = 8;
+ opts->x_param_large_unit_insns = 30000;
+ opts->x_flag_ira_loop_pressure = 1;
+ opts->x_flag_inline_functions_called_once = 0;
+ opts->x_flag_ira_algorithm = IRA_ALGORITHM_PRIORITY;
+ opts->x_flag_delayed_branch = 1;
+ opts->x_flag_gcse_las = 1;
+ opts->x_flag_gcse_sm = 1;
+ opts->x_flag_ipa_pta = 1;
+ opts->x_flag_reorder_blocks_and_partition = 1;
+ opts->x_flag_reorder_blocks = 1;
+ opts->x_flag_crypto_accel_aes = 1;
+ opts->x_param_flexible_seg_len = 1;
+}
+
+/* Reset the optimize option.
+ After checking the model result, this function can
+ reset the more appropriate options. */
+static void
+reset_machine_option (struct gcc_options *opts)
+{
+ if (!(opts->x_optimize_machine)
+ || strstr (opts->x_aarch64_tune_string, "hip09") == NULL)
+ {
+ return;
+ }
+
+ const char *ai_infer_level = getenv ("AI_INFER_LEVEL");
+ if (ai_infer_level)
+ {
+ override_optimize_options_1 (opts);
+ if (lang_c_p ())
+ {
+ override_C_optimize_options (opts);
+ }
+ else if (lang_cpp_p ())
+ {
+ override_CPP_optimize_options (opts);
+ }
+ else if (lang_GNU_Fortran ())
+ {
+ override_Fortran_optimize_options (opts);
+ }
+ }
+}
+
/* Implement targetm.vectorize.add_stmt_cost. */
static unsigned
aarch64_add_stmt_cost (void *data, int count, enum vect_cost_for_stmt kind,
@@ -15060,6 +15189,7 @@ aarch64_override_options_internal (struct gcc_options *opts)
if (opts->x_aarch64_tune_string == NULL)
opts->x_aarch64_tune_string = selected_tune->name;
+ reset_machine_option (opts);
aarch64_override_options_after_change_1 (opts);
}
diff --git a/gcc/ipa-hardware-detection.c b/gcc/ipa-hardware-detection.c
new file mode 100644
index 000000000..f127ebe2c
--- /dev/null
+++ b/gcc/ipa-hardware-detection.c
@@ -0,0 +1,228 @@
+/* Hardware Detection.
+ Copyright (C) 2022-2022 Free Software Foundation, Inc.
+This file is part of GCC.
+GCC is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the
+Free Software Foundation; either version 3, or (at your option) any
+later version.
+GCC is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+for more details.
+You should have received a copy of the GNU General Public License
+along with GCC; see the file COPYING3.  If not see
+<http://www.gnu.org/licenses/>. */
+
+#include "config.h"
+#include "system.h"
+#include "coretypes.h"
+#include "backend.h"
+#include "target.h"
+#include "tree.h"
+#include "gimple.h"
+#include "tree-pass.h"
+#include "gimple-ssa.h"
+#include "tree-pretty-print.h"
+#include "fold-const.h"
+#include "gimplify.h"
+#include "gimple-iterator.h"
+#include "tree-ssa-loop-manip.h"
+#include "tree-ssa-loop.h"
+#include "ssa.h"
+#include "tree-into-ssa.h"
+#include "cfganal.h"
+#include "cfgloop.h"
+#include "gimple-pretty-print.h"
+#include "tree-cfg.h"
+#include "cgraph.h"
+#include "print-tree.h"
+#include "cfghooks.h"
+#include "gimple-fold.h"
+
+namespace {
+
+static basic_block
+create_abort_bb (basic_block last_bb)
+{
+ basic_block bb = create_empty_bb (last_bb);
+ if (last_bb->loop_father != NULL)
+ {
+ add_bb_to_loop (bb, last_bb->loop_father);
+ loops_state_set (LOOPS_NEED_FIXUP);
+ }
+ gimple_stmt_iterator gsi = gsi_last_bb (bb);
+ tree fn = builtin_decl_implicit (BUILT_IN_ABORT);
+ gimple *g = gimple_build_call (fn, 0);
+ gsi_insert_after (&gsi, g, GSI_NEW_STMT);
+ return bb;
+}
+
+static basic_block
+create_part_bb (basic_block last_bb, tree part_base)
+{
+ basic_block bb = create_empty_bb (last_bb);
+ if (last_bb->loop_father != NULL)
+ {
+ add_bb_to_loop (bb, last_bb->loop_father);
+ loops_state_set (LOOPS_NEED_FIXUP);
+ }
+ gimple_stmt_iterator gsi = gsi_last_bb (bb);
+ gsi_insert_after (&gsi, gimple_build_nop (), GSI_NEW_STMT);
+ /* This number is used to efficiently identify the supported part range. */
+ tree part_cond = gimplify_build2 (
+ &gsi, PLUS_EXPR, unsigned_type_node, part_base,
+ build_int_cst (unsigned_type_node, 4294963967));
+ gcond *cond = gimple_build_cond (LE_EXPR, part_cond,
+ build_int_cst (unsigned_type_node, 2),
+ NULL_TREE, NULL_TREE);
+ gimple_set_location (cond, input_location);
+ gsi_insert_before (&gsi, cond, GSI_SAME_STMT);
+ gsi_remove (&gsi, true);
+ return bb;
+}
+
+static void
+create_detection_bb ()
+{
+ edge old_e = single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun));
+ basic_block ret_bb = old_e->dest;
+
+ basic_block detection_bb = create_empty_bb (ENTRY_BLOCK_PTR_FOR_FN (cfun));
+ if (ENTRY_BLOCK_PTR_FOR_FN (cfun)->loop_father != NULL)
+ {
+ add_bb_to_loop (detection_bb, ENTRY_BLOCK_PTR_FOR_FN (cfun)->loop_father);
+ loops_state_set (LOOPS_NEED_FIXUP);
+ }
+ tree cpuid_decl = build_decl (input_location, VAR_DECL,
+ get_identifier ("cpuid"), unsigned_type_node);
+ add_local_decl (cfun, cpuid_decl);
+
+ gimple_stmt_iterator gsi = gsi_last_bb (detection_bb);
+ vec<tree, va_gc> *outputs = NULL;
+ tree purpose = build_string (strlen ("=r"), "=r");
+ tree output = build_tree_list (
+ build_tree_list (NULL_TREE, purpose), cpuid_decl);
+ vec_safe_push (outputs, output);
+ gasm *asm_stmt = gimple_build_asm_vec (
+ "mrs %0, MIDR_EL1", NULL, outputs, NULL, NULL);
+ gsi_insert_after (&gsi, asm_stmt, GSI_NEW_STMT);
+ gsi_insert_after (&gsi, gimple_build_nop (), GSI_NEW_STMT);
+
+ tree implementer = gimplify_build2 (
+ &gsi, RSHIFT_EXPR, unsigned_type_node, cpuid_decl,
+ build_int_cst (unsigned_type_node, 24));
+ tree part_base = gimplify_build2 (
+ &gsi, RSHIFT_EXPR, unsigned_type_node, cpuid_decl,
+ build_int_cst (unsigned_type_node, 4));
+ tree part = gimplify_build2 (
+ &gsi, BIT_AND_EXPR, unsigned_type_node, part_base,
+ build_int_cst (unsigned_type_node, 4095));
+ gcond *implementer_cond = gimple_build_cond (
+ EQ_EXPR, implementer,
+ build_int_cst (unsigned_type_node, 72),
+ NULL_TREE, NULL_TREE);
+ gimple_set_location (implementer_cond, input_location);
+ gsi_insert_before (&gsi, implementer_cond, GSI_SAME_STMT);
+ gsi_remove (&gsi, true);
+
+ basic_block part_bb = create_part_bb (detection_bb, part);
+ basic_block abort_bb = create_abort_bb (part_bb);
+
+ remove_edge_raw (old_e);
+ make_single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun),
+ detection_bb, EDGE_FALLTHRU);
+ edge etrue = make_edge (detection_bb, part_bb, EDGE_TRUE_VALUE);
+ etrue->probability = profile_probability::likely ();
+ edge efalse = make_edge (detection_bb, abort_bb, EDGE_FALSE_VALUE);
+ efalse->probability = profile_probability::unlikely ();
+ edge part_true = make_edge (part_bb, ret_bb, EDGE_TRUE_VALUE);
+ part_true->probability = profile_probability::likely ();
+ edge part_false = make_edge (part_bb, abort_bb, EDGE_FALSE_VALUE);
+ part_false->probability = profile_probability::unlikely ();
+ make_single_succ_edge (abort_bb, ret_bb, EDGE_FALLTHRU);
+ if (dom_info_available_p (CDI_DOMINATORS))
+ {
+ set_immediate_dominator (CDI_DOMINATORS, part_bb, detection_bb);
+ set_immediate_dominator (CDI_DOMINATORS, ret_bb, detection_bb);
+ set_immediate_dominator (CDI_DOMINATORS, abort_bb, detection_bb);
+ }
+}
+
+const pass_data pass_data_ipa_hardware_detection =
+{
+ SIMPLE_IPA_PASS,
+ "hardware_detection",
+ OPTGROUP_NONE,
+ TV_IPA_HARDWARE_DETECTION,
+ (PROP_cfg | PROP_ssa),
+ 0,
+ 0,
+ 0,
+ (TODO_update_ssa | TODO_verify_all)
+};
+
+class pass_ipa_hardware_detection : public simple_ipa_opt_pass
+{
+public:
+ pass_ipa_hardware_detection (gcc::context *ctxt)
+ : simple_ipa_opt_pass (pass_data_ipa_hardware_detection, ctxt)
+ {}
+
+ virtual bool gate (function *);
+ virtual unsigned int execute (function *);
+}; // class pass_ipa_hardware_detection
+
+bool
+pass_ipa_hardware_detection::gate (function *)
+{
+ const char *ai_infer_level = getenv ("AI_INFER_LEVEL");
+ return (ai_infer_level
+ && optimize_machine > 0
+ /* Only enable in lto or whole_program.  */
+ && (in_lto_p || flag_whole_program));
+}
+
+unsigned int
+pass_ipa_hardware_detection::execute (function *)
+{
+ unsigned int ret = 0;
+ cgraph_node *cnode;
+ FOR_EACH_FUNCTION (cnode)
+ {
+ if (!cnode->real_symbol_p ())
+ {
+ continue;
+ }
+ if (cnode->definition)
+ {
+ if (!cnode->has_gimple_body_p () || cnode->inlined_to)
+ continue;
+
+ cnode->get_body ();
+ function *fn = DECL_STRUCT_FUNCTION (cnode->decl);
+ if (!fn)
+ continue;
+
+ if (DECL_NAME (cnode->decl)
+ && MAIN_NAME_P (DECL_NAME (cnode->decl)))
+ {
+ push_cfun (fn);
+ calculate_dominance_info (CDI_DOMINATORS);
+
+ create_detection_bb ();
+
+ cgraph_edge::rebuild_edges ();
+ free_dominance_info (CDI_DOMINATORS);
+ pop_cfun ();
+ }
+ }
+ }
+ return ret;
+}
+} // anon namespace
+
+simple_ipa_opt_pass *
+make_pass_ipa_hardware_detection (gcc::context *ctxt)
+{
+ return new pass_ipa_hardware_detection (ctxt);
+}
diff --git a/gcc/opts-common.c b/gcc/opts-common.c
index bf82b05c8..52e28e2dc 100644
--- a/gcc/opts-common.c
+++ b/gcc/opts-common.c
@@ -926,6 +926,158 @@ opts_concat (const char *first, ...)
return newstr;
}
+typedef int64_t (*run_ai_model_func)(int, const char **,
+ const char *, int, int64_t *);
+#define PTR_UNION_TYPE(TOTYPE) union { void *_q; TOTYPE _nq; }
+#define PTR_UNION_AS_VOID_PTR(NAME) (NAME._q)
+#define PTR_UNION_AS_CAST_PTR(NAME) (NAME._nq)
+
+static int64_t
+ai_infer_optimization (int argc, const char **argv,
+ const char *mcpu_option,
+ int argc_hw, int64_t *argv_hw)
+{
+ /* Load dependent AI-framework libraries. */
+ void *onnxruntime_lib_handle = NULL;
+ const char *onnxruntime_lib_path = "libonnxruntime.so";
+
+ onnxruntime_lib_handle = dlopen (onnxruntime_lib_path,
+ RTLD_LAZY | RTLD_GLOBAL);
+ if (!onnxruntime_lib_handle)
+ {
+ return -1;
+ }
+
+ void *ai4c_lib_handle = NULL;
+ const char *ai4c_lib_path = "libONNXRunner.so";
+
+ ai4c_lib_handle = dlopen (ai4c_lib_path, RTLD_LAZY | RTLD_GLOBAL);
+ if (!ai4c_lib_handle)
+ {
+ return -1;
+ }
+
+ /* Clear any existing error. */
+ dlerror ();
+
+ /* Run AI4Compiler model. */
+ if (ai4c_lib_handle == NULL || onnxruntime_lib_handle == NULL)
+ {
+ return -1;
+ }
+
+ run_ai_model_func run_ai_model;
+ PTR_UNION_TYPE (run_ai_model_func) run_ai_model_func_union;
+ PTR_UNION_AS_VOID_PTR (run_ai_model_func_union)
+ = dlsym (ai4c_lib_handle, "runONNXModelOptimizer");
+ run_ai_model = PTR_UNION_AS_CAST_PTR (run_ai_model_func_union);
+ if (!run_ai_model)
+ {
+ dlclose (ai4c_lib_handle);
+ dlclose (onnxruntime_lib_handle);
+ return -1;
+ }
+ int64_t model_pred = (*run_ai_model) (argc, argv,
+ mcpu_option, argc_hw, argv_hw);
+
+ if (ai4c_lib_handle)
+ dlclose (ai4c_lib_handle);
+
+ if (onnxruntime_lib_handle)
+ dlclose (onnxruntime_lib_handle);
+
+ if (model_pred == 1)
+ putenv ("AI_INFER_LEVEL=1");
+ return model_pred;
+}
+
+static int
+handle_lto_option (unsigned int lang_mask,
+ unsigned int num_decoded_options,
+ unsigned int argc,
+ const char **argv,
+ struct cl_decoded_option *&opt_array)
+{
+ int ret = 0;
+ char *lan = "";
+ char *compiler = xstrdup (argv[0]);
+ lan = strrchr (compiler, '/');
+ if (lan != NULL)
+ lan ++;
+ else
+ lan = compiler;
+ if (strstr (lan, "gcc") != NULL)
+ {
+ opt_array = XRESIZEVEC (struct cl_decoded_option, opt_array, argc + 2);
+ const char* lto_flag = "-flto=8";
+ decode_cmdline_option (&lto_flag, lang_mask,
+ &opt_array[num_decoded_options]);
+ ret++;
+ const char* ltopartition_flag = "-flto-partition=one";
+ decode_cmdline_option (&ltopartition_flag, lang_mask,
+ &opt_array[num_decoded_options + 1]);
+ ret++;
+ }
+ else if (strstr (lan, "g++") != NULL
+ || strstr (lan, "gfortran") != NULL)
+ {
+ opt_array = XRESIZEVEC (struct cl_decoded_option, opt_array, argc + 1);
+ const char* lto_flag = "-flto=8";
+ decode_cmdline_option (&lto_flag, lang_mask,
+ &opt_array[num_decoded_options]);
+ ret++;
+ }
+ if (compiler)
+ free (compiler);
+ return ret;
+}
+
+static int
+handle_machine_option (unsigned int lang_mask,
+ unsigned int num_decoded_options,
+ unsigned int argc,
+ const char **argv,
+ struct cl_decoded_option *&opt_array)
+{
+ int ret = 0;
+ bool flag_Om = false;
+ bool flag_hip09 = false;
+ for (unsigned i = 1; i < argc; i ++)
+ {
+ if (strcmp (argv[i], "-Om") == 0)
+ flag_Om = true;
+ if (strstr (argv[i], "mcpu=hip09") != NULL)
+ flag_hip09 = true;
+ }
+ if (!flag_hip09 || !flag_Om)
+ {
+ return ret;
+ }
+
+ const char *ai_infer_level = getenv ("AI_INFER_LEVEL");
+ if (ai_infer_level)
+ {
+ return ret;
+ }
+ int argc_hw = 6;
+ int64_t argv_hw[argc_hw] = {
+ global_options.x_param_simultaneous_prefetches,
+ global_options.x_param_l1_cache_size,
+ global_options.x_param_l1_cache_line_size,
+ global_options.x_param_l2_cache_size,
+ global_options.x_param_llc_capacity_per_core,
+ global_options.x_param_ipa_prefetch_distance_factor};
+ int64_t output_pred = ai_infer_optimization (
+ argc, argv, "hip09", argc_hw, argv_hw);
+ if (output_pred != 1)
+ {
+ return ret;
+ }
+
+ return handle_lto_option (lang_mask, num_decoded_options,
+ argc, argv, opt_array);
+}
+
/* Decode command-line options (ARGC and ARGV being the arguments of
main) into an array, setting *DECODED_OPTIONS to a pointer to that
array and *DECODED_OPTIONS_COUNT to the number of entries in the
@@ -987,6 +1139,8 @@ decode_cmdline_options_to_array (unsigned int argc, const char **argv,
num_decoded_options++;
}
+ num_decoded_options += handle_machine_option (lang_mask, num_decoded_options,
+ argc, argv, opt_array);
*decoded_options = opt_array;
*decoded_options_count = num_decoded_options;
prune_options (decoded_options, decoded_options_count, lang_mask);
diff --git a/gcc/opts.c b/gcc/opts.c
index c0ccd0853..dc61216c0 100644
--- a/gcc/opts.c
+++ b/gcc/opts.c
@@ -632,6 +632,15 @@ default_options_optimization (struct gcc_options *opts,
opts->x_optimize_debug = 1;
break;
+ case OPT_Om:
+ /* -Om adds flags to -O3 & -Ofast. */
+ opts->x_optimize_size = 0;
+ opts->x_optimize = 3;
+ opts->x_optimize_fast = 1;
+ opts->x_optimize_machine = true;
+ opts->x_optimize_debug = 0;
+ break;
+
case OPT_fopenacc:
if (opt->value)
openacc_mode = true;
@@ -2378,6 +2387,8 @@ common_handle_option (struct gcc_options *opts,
opts->x_flag_sanitize_coverage, value, true);
break;
+ case OPT_Om:
+ break;
case OPT_O:
case OPT_Os:
case OPT_Ofast:
diff --git a/gcc/passes.def b/gcc/passes.def
index b6006de22..8898b72fc 100644
--- a/gcc/passes.def
+++ b/gcc/passes.def
@@ -177,6 +177,7 @@ along with GCC; see the file COPYING3. If not see
compiled unit. */
INSERT_PASSES_AFTER (all_late_ipa_passes)
NEXT_PASS (pass_materialize_all_clones);
+ NEXT_PASS (pass_ipa_hardware_detection);
NEXT_PASS (pass_ipa_pta);
/* FIXME: this should a normal IP pass */
NEXT_PASS (pass_ipa_struct_reorg);
diff --git a/gcc/timevar.def b/gcc/timevar.def
index 929e9e1d3..66b21f166 100644
--- a/gcc/timevar.def
+++ b/gcc/timevar.def
@@ -81,6 +81,7 @@ DEFTIMEVAR (TV_IPA_CONSTANT_PROP , "ipa cp")
DEFTIMEVAR (TV_IPA_INLINING , "ipa inlining heuristics")
DEFTIMEVAR (TV_IPA_FNSPLIT , "ipa function splitting")
DEFTIMEVAR (TV_IPA_COMDATS , "ipa comdats")
+DEFTIMEVAR (TV_IPA_HARDWARE_DETECTION, "ipa detection")
DEFTIMEVAR (TV_IPA_PREFETCH , "ipa prefetch")
DEFTIMEVAR (TV_IPA_STRUCT_REORG , "ipa struct reorg optimization")
DEFTIMEVAR (TV_IPA_EXTEND_AUTO_PROFILE, "ipa extend auto profile")
diff --git a/gcc/tree-pass.h b/gcc/tree-pass.h
index 4d952884d..d3a41d0d5 100644
--- a/gcc/tree-pass.h
+++ b/gcc/tree-pass.h
@@ -513,6 +513,8 @@ extern ipa_opt_pass_d *make_pass_ipa_odr (gcc::context *ctxt);
extern ipa_opt_pass_d *make_pass_ipa_reference (gcc::context *ctxt);
extern ipa_opt_pass_d *make_pass_ipa_hsa (gcc::context *ctxt);
extern ipa_opt_pass_d *make_pass_ipa_pure_const (gcc::context *ctxt);
+extern simple_ipa_opt_pass *make_pass_ipa_hardware_detection (gcc::context *
+ ctxt);
extern simple_ipa_opt_pass *make_pass_ipa_prefetch (gcc::context *ctxt);
extern simple_ipa_opt_pass *make_pass_ipa_struct_reorg (gcc::context *ctxt);
extern simple_ipa_opt_pass *make_pass_ipa_extend_auto_profile (gcc::context
--
2.33.0

View File

@ -61,7 +61,7 @@
Summary: Various compilers (C, C++, Objective-C, ...)
Name: gcc
Version: %{gcc_version}
Release: 62
Release: 58
License: GPLv3+ and GPLv3+ with exceptions and GPLv2+ with exceptions and LGPLv2+ and BSD
URL: https://gcc.gnu.org
@ -306,12 +306,6 @@ Patch195: 0195-add-whitelist-feature-for-OneProfile.patch
Patch196: 0196-fix-bugs-in-loop-detections-add-filter-to-SSA-statem.patch
Patch197: 0197-Add-hip09-machine-discribtion.patch
Patch198: 0198-bugfix-Modify-the-hip09-CPU-information.patch
Patch199: 0199-Backport-Improve-hard-reg-preference-propapagation.patch
Patch200: 0200-Support-LTO-in-AutoBOLT-mode.patch
Patch201: 0201-Fix-bug-in-detecting-abnormal-revisit-for-bb.patch
Patch202: 0202-Change-set_of-to-reg_set_p-and-add-check-for-global_.patch
Patch203: 0203-Remove-erroneous-pattern-from-gimple-ifcvt.patch
Patch204: 0204-Try-to-use-AI-model-to-guide-optimization.patch
%global gcc_target_platform %{_arch}-linux-gnu
%if %{build_go}
@ -962,12 +956,6 @@ not stable, so plugins must be rebuilt any time GCC is updated.
%patch196 -p1
%patch197 -p1
%patch198 -p1
%patch199 -p1
%patch200 -p1
%patch201 -p1
%patch202 -p1
%patch203 -p1
%patch204 -p1
%build
@ -1058,7 +1046,7 @@ CC="$CC" CFLAGS="$OPT_FLAGS" \
%ifarch ppc64le
--disable-multilib \
--enable-targets=powerpcle-linux \
--with-cpu-32=power8 --with-tune-32=power8 --with-cpu-64=power8 --with-tune-64=power8 \
--withc-cpu-32=power8 --with-tune-32=power8 --with-cpu-64=power8 --with-tune-64=power8 \
%endif
%ifarch sparc sparcv9 sparc64
@ -3002,30 +2990,6 @@ end
%doc rpm.doc/changelogs/libcc1/ChangeLog*
%changelog
* Fri Jun 21 2024 Mingchuan Wu <wumingchuan1992@foxmail.com> - 10.3.1-62
- Type:Sync
- ID:NA
- SUG:NA
- DESC: Sync patch from openeuler/gcc
* Tue Jun 18 2024 chenyuanfeng <yuanfeng.chen@shingroup.cn> - 10.3.1-61
- Type: Fix
- ID:NA
- SUG:NA
- DESC: fix typo
* Sat Jun 15 2024 rfwang07 <wangrufeng5@huawei.com> - 10.3.1-60
- Type:Sync
- ID:NA
- SUG:NA
- DESC: Sync patch from openeuler/gcc
* Sat Jun 15 2024 eastb233 <xiezhiheng@huawei.com> - 10.3.1-59
- Type:Sync
- ID:NA
- SUG:NA
- DESC: Sync patch from openeuler/gcc
* Fri Jun 14 2024 zhenyu zhao <zhaozhenyu17@huawei.com> - 10.3.1-58
- Type:Sync
- ID:NA