Hi!
PCH doesn't work properly in --enable-host-pie configurations on
powerpc*-linux*.
The problem is that the rs6000_builtin_info and rs6000_instance_info
arrays mix pointers to .rodata/.data (bifname and attr_string point
to string literals in .rodata section, and the next member is either NULL
or &rs6000_instance_info[XXX]) and GC member (tree fntype).
Now, for normal GC this works just fine, we emit
{
&rs6000_instance_info[0].fntype,
1 * (RS6000_INST_MAX),
sizeof (rs6000_instance_info[0]),
>_ggc_mx_tree_node,
>_pch_nx_tree_node
},
{
&rs6000_builtin_info[0].fntype,
1 * (RS6000_BIF_MAX),
sizeof (rs6000_builtin_info[0]),
>_ggc_mx_tree_node,
>_pch_nx_tree_node
},
GC roots which are strided and thus cover only the fntype members of all
the elements of the two arrays.
For PCH though it actually results in saving those huge arrays (one is
130832 bytes, another 81568 bytes) into the .gch files and loading them back
in full. While the bifname and attr_string and next pointers are marked as
GTY((skip)), they are actually saved to point to the .rodata and .data
sections of the process which writes the PCH, but because cc1/cc1plus etc.
are position independent executables with --enable-host-pie, when it is
loaded from the PCH file, it can point in a completely different addresses
where nothing is mapped at all or some random different thing appears at.
While gengtype supports the callback option, that one is meant for
relocatable function pointers and doesn't work in the case of GTY arrays
inside of .data section anyway.
So, either we'd need to add some further GTY extensions, or the following
patch instead reworks it such that the fntype members which were the only
reason for PCH in those arrays are moved to separate arrays.
Size-wise in .data sections it is (in bytes):
vanilla patched
rs6000_builtin_info 130832 110704
rs6000_instance_info 81568 40784
rs6000_overload_info 7392 7392
rs6000_builtin_info_fntype 0 10064
rs6000_instance_info_fntype 0 20392
sum 219792 189336
where previously we saved/restored for PCH those 130832+81568 bytes, now we
save/restore just 10064+20392 bytes, so this change is beneficial for the
data section size.
Unfortunately, it grows the size of the rs6000_init_generated_builtins
function, vanilla had 218328 bytes, patched has 228668.
When I applied
void
rs6000_init_generated_builtins ()
{
+ bifdata *rs6000_builtin_info_p;
+ tree *rs6000_builtin_info_fntype_p;
+ ovlddata *rs6000_instance_info_p;
+ tree *rs6000_instance_info_fntype_p;
+ ovldrecord *rs6000_overload_info_p;
+ __asm ("" : "=r" (rs6000_builtin_info_p) : "0" (rs6000_builtin_info));
+ __asm ("" : "=r" (rs6000_builtin_info_fntype_p) : "0"
(rs6000_builtin_info_fntype));
+ __asm ("" : "=r" (rs6000_instance_info_p) : "0" (rs6000_instance_info));
+ __asm ("" : "=r" (rs6000_instance_info_fntype_p) : "0"
(rs6000_instance_info_fntype));
+ __asm ("" : "=r" (rs6000_overload_info_p) : "0" (rs6000_overload_info));
+ #define rs6000_builtin_info rs6000_builtin_info_p
+ #define rs6000_builtin_info_fntype rs6000_builtin_info_fntype_p
+ #define rs6000_instance_info rs6000_instance_info_p
+ #define rs6000_instance_info_fntype rs6000_instance_info_fntype_p
+ #define rs6000_overload_info rs6000_overload_info_p
+
hack by hand, the size of the function is 209700 though, so if really
wanted, we could add __attribute__((__noipa__)) to the function when
building with recent enough GCC and pass pointers to the first elements
of the 5 arrays to the function as arguments. If you want such a change,
could that be done incrementally?
Bootstrapped/regtested on powerpc64le-linux and powerpc64-linux (-m32/-m64
testing there), ok for trunk and after a while for release branches?
2024-06-03 Jakub Jelinek <ja...@redhat.com>
PR target/115324
* config/rs6000/rs6000-gen-builtins.cc (write_decls): Remove
GTY markup from struct bifdata and struct ovlddata and remove their
fntype members. Change next member in struct ovlddata and
first_instance member of struct ovldrecord to have int type rather
than struct ovlddata *. Remove GTY markup from rs6000_builtin_info
and rs6000_instance_info arrays, declare new
rs6000_builtin_info_fntype and rs6000_instance_info_fntype arrays,
which have GTY markup.
(write_bif_static_init): Adjust for the above changes.
(write_ovld_static_init): Likewise.
(write_init_bif_table): Likewise.
(write_init_ovld_table): Likewise.
* config/rs6000/rs6000-builtin.cc (rs6000_init_builtins): Likewise.
* config/rs6000/rs6000-c.cc (find_instance): Likewise. Make static.
(altivec_resolve_overloaded_builtin): Adjust for the above changes.
--- gcc/config/rs6000/rs6000-gen-builtins.cc.jj 2024-05-31 22:09:39.300059155
+0200
+++ gcc/config/rs6000/rs6000-gen-builtins.cc 2024-06-03 10:29:06.787214893
+0200
@@ -2261,20 +2261,19 @@ write_decls (void)
fprintf (header_file, "};\n\n");
fprintf (header_file, "#define PPC_MAXRESTROPNDS 3\n");
- fprintf (header_file, "struct GTY(()) bifdata\n");
+ fprintf (header_file, "struct bifdata\n");
fprintf (header_file, "{\n");
- fprintf (header_file, " const char *GTY((skip(\"\"))) bifname;\n");
- fprintf (header_file, " bif_enable GTY((skip(\"\"))) enable;\n");
- fprintf (header_file, " tree fntype;\n");
- fprintf (header_file, " insn_code GTY((skip(\"\"))) icode;\n");
- fprintf (header_file, " int nargs;\n");
- fprintf (header_file, " int bifattrs;\n");
- fprintf (header_file, " int restr_opnd[PPC_MAXRESTROPNDS];\n");
- fprintf (header_file, " restriction GTY((skip(\"\")))
restr[PPC_MAXRESTROPNDS];\n");
- fprintf (header_file, " int restr_val1[PPC_MAXRESTROPNDS];\n");
- fprintf (header_file, " int restr_val2[PPC_MAXRESTROPNDS];\n");
- fprintf (header_file, " const char *GTY((skip(\"\"))) attr_string;\n");
- fprintf (header_file, " rs6000_gen_builtins GTY((skip(\"\")))
assoc_bif;\n");
+ fprintf (header_file, " const char *bifname;\n");
+ fprintf (header_file, " bif_enable enable;\n");
+ fprintf (header_file, " insn_code icode;\n");
+ fprintf (header_file, " int nargs;\n");
+ fprintf (header_file, " int bifattrs;\n");
+ fprintf (header_file, " int restr_opnd[PPC_MAXRESTROPNDS];\n");
+ fprintf (header_file, " restriction restr[PPC_MAXRESTROPNDS];\n");
+ fprintf (header_file, " int restr_val1[PPC_MAXRESTROPNDS];\n");
+ fprintf (header_file, " int restr_val2[PPC_MAXRESTROPNDS];\n");
+ fprintf (header_file, " const char *attr_string;\n");
+ fprintf (header_file, " rs6000_gen_builtins assoc_bif;\n");
fprintf (header_file, "};\n\n");
fprintf (header_file, "#define bif_init_bit\t\t(0x00000001)\n");
@@ -2353,24 +2352,28 @@ write_decls (void)
fprintf (header_file, "\n");
fprintf (header_file,
- "extern GTY(()) bifdata rs6000_builtin_info[RS6000_BIF_MAX];\n\n");
+ "extern bifdata rs6000_builtin_info[RS6000_BIF_MAX];\n\n");
- fprintf (header_file, "struct GTY(()) ovlddata\n");
+ fprintf (header_file,
+ "extern GTY(()) tree
rs6000_builtin_info_fntype[RS6000_BIF_MAX];\n\n");
+
+ fprintf (header_file, "struct ovlddata\n");
fprintf (header_file, "{\n");
- fprintf (header_file, " const char *GTY((skip(\"\"))) bifname;\n");
- fprintf (header_file, " rs6000_gen_builtins GTY((skip(\"\"))) bifid;\n");
- fprintf (header_file, " tree fntype;\n");
- fprintf (header_file, " ovlddata *GTY((skip(\"\"))) next;\n");
+ fprintf (header_file, " const char *bifname;\n");
+ fprintf (header_file, " rs6000_gen_builtins bifid;\n");
+ fprintf (header_file, " int next;\n");
fprintf (header_file, "};\n\n");
fprintf (header_file, "struct ovldrecord\n");
fprintf (header_file, "{\n");
fprintf (header_file, " const char *ovld_name;\n");
- fprintf (header_file, " ovlddata *first_instance;\n");
+ fprintf (header_file, " int first_instance;\n");
fprintf (header_file, "};\n\n");
fprintf (header_file,
- "extern GTY(()) ovlddata rs6000_instance_info[RS6000_INST_MAX];\n");
+ "extern ovlddata rs6000_instance_info[RS6000_INST_MAX];\n");
+ fprintf (header_file, "extern GTY(()) tree "
+ "rs6000_instance_info_fntype[RS6000_INST_MAX];\n");
fprintf (header_file, "extern ovldrecord rs6000_overload_info[];\n\n");
fprintf (header_file, "extern void rs6000_init_generated_builtins ();\n\n");
@@ -2481,7 +2484,7 @@ write_bif_static_init (void)
fprintf (init_file, "bifdata rs6000_builtin_info[RS6000_BIF_MAX] =\n");
fprintf (init_file, " {\n");
fprintf (init_file, " { /* RS6000_BIF_NONE: */\n");
- fprintf (init_file, " \"\", ENB_ALWAYS, 0, CODE_FOR_nothing, 0,\n");
+ fprintf (init_file, " \"\", ENB_ALWAYS, CODE_FOR_nothing, 0,\n");
fprintf (init_file, " 0, {0, 0, 0}, {RES_NONE, RES_NONE, RES_NONE},\n");
fprintf (init_file, " {0, 0, 0}, {0, 0, 0}, \"\", RS6000_BIF_NONE\n");
fprintf (init_file, " },\n");
@@ -2493,8 +2496,6 @@ write_bif_static_init (void)
bifp->proto.bifname);
fprintf (init_file, " /* enable*/\t%s,\n",
enable_string[bifp->stanza]);
- /* Type must be instantiated at run time. */
- fprintf (init_file, " /* fntype */\t0,\n");
fprintf (init_file, " /* icode */\tCODE_FOR_%s,\n",
bifp->patname);
fprintf (init_file, " /* nargs */\t%d,\n",
@@ -2586,6 +2587,8 @@ write_bif_static_init (void)
fprintf (init_file, " },\n");
}
fprintf (init_file, " };\n\n");
+
+ fprintf (init_file, "tree rs6000_builtin_info_fntype[RS6000_BIF_MAX];\n\n");
}
/* Write the decls and initializers for rs6000_overload_info[] and
@@ -2598,7 +2601,7 @@ write_ovld_static_init (void)
"- RS6000_OVLD_NONE] =\n");
fprintf (init_file, " {\n");
fprintf (init_file, " { /* RS6000_OVLD_NONE: */\n");
- fprintf (init_file, " \"\", NULL\n");
+ fprintf (init_file, " \"\", -1\n");
fprintf (init_file, " },\n");
for (int i = 0; i <= curr_ovld_stanza; i++)
{
@@ -2607,7 +2610,7 @@ write_ovld_static_init (void)
fprintf (init_file, " /* ovld_name */\t\"%s\",\n",
ovld_stanzas[i].intern_name);
/* First-instance must currently be instantiated at run time. */
- fprintf (init_file, " /* first_instance */\tNULL\n");
+ fprintf (init_file, " /* first_instance */\t-1\n");
fprintf (init_file, " },\n");
}
fprintf (init_file, " };\n\n");
@@ -2615,7 +2618,7 @@ write_ovld_static_init (void)
fprintf (init_file, "ovlddata rs6000_instance_info[RS6000_INST_MAX] =\n");
fprintf (init_file, " {\n");
fprintf (init_file, " { /* RS6000_INST_NONE: */\n");
- fprintf (init_file, " \"\", RS6000_BIF_NONE, NULL_TREE, NULL\n");
+ fprintf (init_file, " \"\", RS6000_BIF_NONE, -1\n");
fprintf (init_file, " },\n");
for (int i = 0; i <= curr_ovld; i++)
{
@@ -2625,19 +2628,20 @@ write_ovld_static_init (void)
ovlds[i].proto.bifname);
fprintf (init_file, " /* bifid */\tRS6000_BIF_%s,\n",
ovlds[i].bif_id_name);
- /* Type must be instantiated at run time. */
- fprintf (init_file, " /* fntype */\t0,\n");
fprintf (init_file, " /* next */\t");
if (i < curr_ovld
&& !strcmp (ovlds[i+1].proto.bifname, ovlds[i].proto.bifname))
fprintf (init_file,
- "&rs6000_instance_info[RS6000_INST_%s]\n",
+ "RS6000_INST_%s\n",
ovlds[i+1].ovld_id_name);
else
- fprintf (init_file, "NULL\n");
+ fprintf (init_file, "-1\n");
fprintf (init_file, " },\n");
}
fprintf (init_file, " };\n\n");
+
+ fprintf (init_file,
+ "tree rs6000_instance_info_fntype[RS6000_INST_MAX];\n\n");
}
/* Write code to initialize the built-in function table. */
@@ -2647,7 +2651,7 @@ write_init_bif_table (void)
for (int i = 0; i <= curr_bif; i++)
{
fprintf (init_file,
- " rs6000_builtin_info[RS6000_BIF_%s].fntype"
+ " rs6000_builtin_info_fntype[RS6000_BIF_%s]"
"\n = %s;\n",
bifs[i].idname, bifs[i].fndecl);
@@ -2736,7 +2740,7 @@ write_init_ovld_table (void)
for (int i = 0; i <= curr_ovld; i++)
{
fprintf (init_file,
- " rs6000_instance_info[RS6000_INST_%s].fntype"
+ " rs6000_instance_info_fntype[RS6000_INST_%s]"
"\n = %s;\n",
ovlds[i].ovld_id_name, ovlds[i].fndecl);
@@ -2793,7 +2797,7 @@ write_init_ovld_table (void)
".first_instance\n",
stanza->stanza_id);
fprintf (init_file,
- " = &rs6000_instance_info[RS6000_INST_%s];\n\n",
+ " = RS6000_INST_%s;\n\n",
ovlds[i].ovld_id_name);
}
}
--- gcc/config/rs6000/rs6000-builtin.cc.jj 2024-04-15 10:16:58.646244857
+0200
+++ gcc/config/rs6000/rs6000-builtin.cc 2024-06-03 10:23:32.684662263 +0200
@@ -845,7 +845,7 @@ rs6000_init_builtins (void)
enum rs6000_gen_builtins fn_code = (enum rs6000_gen_builtins) i;
if (!rs6000_builtin_is_supported (fn_code))
continue;
- tree fntype = rs6000_builtin_info[i].fntype;
+ tree fntype = rs6000_builtin_info_fntype[i];
tree t = TREE_TYPE (fntype);
fprintf (stderr, "%s %s (", rs6000_type_string (t),
rs6000_builtin_info[i].bifname);
--- gcc/config/rs6000/rs6000-c.cc.jj 2024-04-15 10:16:58.653244762 +0200
+++ gcc/config/rs6000/rs6000-c.cc 2024-06-03 10:18:24.996754126 +0200
@@ -1664,22 +1664,25 @@ resolve_vec_step (resolution *res, vec<t
true. If we don't match, return error_mark_node and leave
UNSUPPORTED_BUILTIN alone. */
-tree
-find_instance (bool *unsupported_builtin, ovlddata **instance,
+static tree
+find_instance (bool *unsupported_builtin, int *instance,
rs6000_gen_builtins instance_code,
rs6000_gen_builtins fcode,
tree *types, tree *args, int nargs)
{
- while (*instance && (*instance)->bifid != instance_code)
- *instance = (*instance)->next;
+ while (*instance != -1
+ && rs6000_instance_info[*instance].bifid != instance_code)
+ *instance = rs6000_instance_info[*instance].next;
- ovlddata *inst = *instance;
- gcc_assert (inst != NULL);
+ int inst = *instance;
+ gcc_assert (inst != -1);
/* It is possible for an instance to require a data type that isn't
- defined on this target, in which case inst->fntype will be NULL. */
- if (!inst->fntype)
+ defined on this target, in which case rs6000_instance_info_fntype[inst]
+ will be NULL. */
+ if (!rs6000_instance_info_fntype[inst])
return error_mark_node;
- tree fntype = rs6000_builtin_info[inst->bifid].fntype;
+ rs6000_gen_builtins bifid = rs6000_instance_info[inst].bifid;
+ tree fntype = rs6000_builtin_info_fntype[bifid];
tree argtype = TYPE_ARG_TYPES (fntype);
bool args_compatible = true;
@@ -1696,12 +1699,12 @@ find_instance (bool *unsupported_builtin
if (args_compatible)
{
- if (rs6000_builtin_decl (inst->bifid, false) != error_mark_node
- && rs6000_builtin_is_supported (inst->bifid))
+ if (rs6000_builtin_decl (bifid, false) != error_mark_node
+ && rs6000_builtin_is_supported (bifid))
{
- tree ret_type = TREE_TYPE (inst->fntype);
+ tree ret_type = TREE_TYPE (rs6000_instance_info_fntype[inst]);
return altivec_build_resolved_builtin (args, nargs, fntype, ret_type,
- inst->bifid, fcode);
+ bifid, fcode);
}
else
*unsupported_builtin = true;
@@ -1884,11 +1887,11 @@ altivec_resolve_overloaded_builtin (loca
bool unsupported_builtin = false;
rs6000_gen_builtins instance_code;
bool supported = false;
- ovlddata *instance = rs6000_overload_info[adj_fcode].first_instance;
- gcc_assert (instance != NULL);
+ int instance = rs6000_overload_info[adj_fcode].first_instance;
+ gcc_assert (instance != -1);
/* Functions with no arguments can have only one overloaded instance. */
- gcc_assert (nargs > 0 || !instance->next);
+ gcc_assert (nargs > 0 || rs6000_instance_info[instance].next == -1);
/* Standard overload processing involves determining whether an instance
exists that is type-compatible with the overloaded function call. In
@@ -1989,16 +1992,18 @@ altivec_resolve_overloaded_builtin (loca
/* Standard overload processing. Look for an instance with compatible
parameter types. If it is supported in the current context, resolve
the overloaded call to that instance. */
- for (; instance != NULL; instance = instance->next)
+ for (; instance != -1; instance = rs6000_instance_info[instance].next)
{
+ tree fntype = rs6000_instance_info_fntype[instance];
+ rs6000_gen_builtins bifid = rs6000_instance_info[instance].bifid;
/* It is possible for an instance to require a data type that isn't
- defined on this target, in which case instance->fntype will be
+ defined on this target, in which case fntype will be
NULL. */
- if (!instance->fntype)
+ if (!fntype)
continue;
bool mismatch = false;
- tree nextparm = TYPE_ARG_TYPES (instance->fntype);
+ tree nextparm = TYPE_ARG_TYPES (fntype);
for (unsigned int arg_i = 0;
arg_i < nargs && nextparm != NULL;
@@ -2016,15 +2021,14 @@ altivec_resolve_overloaded_builtin (loca
if (mismatch)
continue;
- supported = rs6000_builtin_is_supported (instance->bifid);
- if (rs6000_builtin_decl (instance->bifid, false) != error_mark_node
+ supported = rs6000_builtin_is_supported (bifid);
+ if (rs6000_builtin_decl (bifid, false) != error_mark_node
&& supported)
{
- tree fntype = rs6000_builtin_info[instance->bifid].fntype;
- tree ret_type = TREE_TYPE (instance->fntype);
+ tree ret_type = TREE_TYPE (fntype);
+ fntype = rs6000_builtin_info_fntype[bifid];
return altivec_build_resolved_builtin (args, nargs, fntype,
- ret_type, instance->bifid,
- fcode);
+ ret_type, bifid, fcode);
}
else
{
@@ -2041,12 +2045,12 @@ altivec_resolve_overloaded_builtin (loca
{
/* Indicate that the instantiation of the overloaded builtin
name is not available with the target flags in effect. */
- rs6000_gen_builtins fcode = (rs6000_gen_builtins) instance->bifid;
+ rs6000_gen_builtins bifid = rs6000_instance_info[instance].bifid;
+ rs6000_gen_builtins fcode = (rs6000_gen_builtins) bifid;
rs6000_invalid_builtin (fcode);
/* Provide clarity of the relationship between the overload
and the instantiation. */
- const char *internal_name
- = rs6000_builtin_info[instance->bifid].bifname;
+ const char *internal_name = rs6000_builtin_info[bifid].bifname;
rich_location richloc (line_table, input_location);
inform (&richloc,
"overloaded builtin %qs is implemented by builtin %qs",
Jakub
