Protocol Buffers - Google's data interchange format (grpc依赖) https://developers.google.com/protocol-buffers/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

291 lines
9.0 KiB

upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
/*
* upb - a minimalist implementation of protocol buffers.
*
* Copyright (c) 2011 Google Inc. See LICENSE for details.
* Author: Josh Haberman <jhaberman@gmail.com>
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
*/
#include <stdlib.h>
#include "upb/handlers.h"
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
/* upb_mhandlers **************************************************************/
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
static upb_mhandlers *upb_mhandlers_new() {
upb_mhandlers *m = malloc(sizeof(*m));
upb_inttable_init(&m->fieldtab, 8, sizeof(upb_itofhandlers_ent));
m->startmsg = NULL;
m->endmsg = NULL;
m->is_group = false;
#ifdef UPB_USE_JIT_X64
m->tablearray = NULL;
#endif
return m;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
static upb_fhandlers *_upb_mhandlers_newfhandlers(upb_mhandlers *m, uint32_t n,
upb_fieldtype_t type,
bool repeated) {
upb_itofhandlers_ent *e = upb_inttable_lookup(&m->fieldtab, n);
// TODO: design/refine the API for changing the set of fields or modifying
// existing handlers.
if (e) return NULL;
upb_fhandlers new_f = {type, repeated, UPB_ATOMIC_INIT(0),
n, -1, m, NULL, UPB_NO_VALUE, NULL, NULL, NULL, NULL, NULL,
#ifdef UPB_USE_JIT_X64
0, 0, 0,
#endif
NULL};
upb_fhandlers *ptr = malloc(sizeof(*ptr));
memcpy(ptr, &new_f, sizeof(upb_fhandlers));
upb_itofhandlers_ent ent = {false, ptr};
upb_inttable_insert(&m->fieldtab, n, &ent);
return ptr;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
upb_fhandlers *upb_mhandlers_newfhandlers(upb_mhandlers *m, uint32_t n,
upb_fieldtype_t type, bool repeated) {
assert(type != UPB_TYPE(MESSAGE));
assert(type != UPB_TYPE(GROUP));
return _upb_mhandlers_newfhandlers(m, n, type, repeated);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
upb_fhandlers *upb_mhandlers_newfhandlers_subm(upb_mhandlers *m, uint32_t n,
upb_fieldtype_t type,
bool repeated,
upb_mhandlers *subm) {
assert(type == UPB_TYPE(MESSAGE) || type == UPB_TYPE(GROUP));
assert(subm);
upb_fhandlers *f = _upb_mhandlers_newfhandlers(m, n, type, repeated);
if (!f) return NULL;
f->submsg = subm;
if (type == UPB_TYPE(GROUP))
_upb_mhandlers_newfhandlers(subm, n, UPB_TYPE_ENDGROUP, false);
return f;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
/* upb_handlers ***************************************************************/
upb_handlers *upb_handlers_new() {
upb_handlers *h = malloc(sizeof(*h));
upb_atomic_init(&h->refcount, 1);
h->msgs_len = 0;
h->msgs_size = 4;
h->msgs = malloc(h->msgs_size * sizeof(*h->msgs));
h->should_jit = true;
return h;
}
void upb_handlers_ref(upb_handlers *h) { upb_atomic_ref(&h->refcount); }
void upb_handlers_unref(upb_handlers *h) {
if (upb_atomic_unref(&h->refcount)) {
for (int i = 0; i < h->msgs_len; i++) {
upb_mhandlers *mh = h->msgs[i];
for(upb_inttable_iter j = upb_inttable_begin(&mh->fieldtab);
!upb_inttable_done(j);
j = upb_inttable_next(&mh->fieldtab, j)) {
upb_itofhandlers_ent *e = upb_inttable_iter_value(j);
free(e->f);
}
upb_inttable_free(&mh->fieldtab);
#ifdef UPB_USE_JIT_X64
free(mh->tablearray);
#endif
free(mh);
}
free(h->msgs);
free(h);
}
}
upb_mhandlers *upb_handlers_newmhandlers(upb_handlers *h) {
if (h->msgs_len == h->msgs_size) {
h->msgs_size *= 2;
h->msgs = realloc(h->msgs, h->msgs_size * sizeof(*h->msgs));
}
upb_mhandlers *mh = upb_mhandlers_new();
h->msgs[h->msgs_len++] = mh;
return mh;
}
typedef struct {
upb_mhandlers *mh;
} upb_mtab_ent;
static upb_mhandlers *upb_regmsg_dfs(upb_handlers *h, const upb_msgdef *m,
upb_onmsgreg *msgreg_cb,
upb_onfieldreg *fieldreg_cb,
void *closure, upb_strtable *mtab) {
upb_mhandlers *mh = upb_handlers_newmhandlers(h);
upb_mtab_ent e = {mh};
upb_strtable_insert(mtab, m->base.fqname, &e);
if (msgreg_cb) msgreg_cb(closure, mh, m);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
upb_msg_iter i;
for(i = upb_msg_begin(m); !upb_msg_done(i); i = upb_msg_next(m, i)) {
upb_fielddef *f = upb_msg_iter_field(i);
upb_fhandlers *fh;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
if (upb_issubmsg(f)) {
upb_mhandlers *sub_mh;
upb_mtab_ent *subm_ent;
// The table lookup is necessary to break the DFS for type cycles.
if ((subm_ent = upb_strtable_lookup(mtab, f->def->fqname)) != NULL) {
sub_mh = subm_ent->mh;
} else {
sub_mh = upb_regmsg_dfs(h, upb_downcast_msgdef(f->def), msgreg_cb,
fieldreg_cb, closure, mtab);
}
fh = upb_mhandlers_newfhandlers_subm(
mh, f->number, f->type, upb_isseq(f), sub_mh);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
} else {
fh = upb_mhandlers_newfhandlers(mh, f->number, f->type, upb_isseq(f));
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
if (fieldreg_cb) fieldreg_cb(closure, fh, f);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
return mh;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
upb_mhandlers *upb_handlers_regmsgdef(upb_handlers *h, const upb_msgdef *m,
upb_onmsgreg *msgreg_cb,
upb_onfieldreg *fieldreg_cb,
void *closure) {
upb_strtable mtab;
upb_strtable_init(&mtab, 8, sizeof(upb_mtab_ent));
upb_mhandlers *ret =
upb_regmsg_dfs(h, m, msgreg_cb, fieldreg_cb, closure, &mtab);
upb_strtable_free(&mtab);
return ret;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
/* upb_dispatcher *************************************************************/
void upb_dispatcher_init(upb_dispatcher *d, upb_status *status,
upb_exit_handler UPB_NORETURN *exit,
void *srcclosure) {
d->stack[0].f = NULL; // Should never be read.
d->limit = &d->stack[UPB_MAX_NESTING];
d->exitjmp = exit;
d->srcclosure = srcclosure;
d->top_is_implicit = false;
d->msgent = NULL;
d->top = NULL;
d->toplevel_msgent = NULL;
d->status = status;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
upb_dispatcher_frame *upb_dispatcher_reset(upb_dispatcher *d, void *closure,
upb_mhandlers *top) {
d->msgent = top;
d->toplevel_msgent = top;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
d->top = d->stack;
d->top->closure = closure;
d->top->is_sequence = false;
d->top->is_packed = false;
return d->top;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
void upb_dispatcher_uninit(upb_dispatcher *d) {
(void)d;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
void upb_dispatch_startmsg(upb_dispatcher *d) {
upb_flow_t flow = UPB_CONTINUE;
if (d->msgent->startmsg) d->msgent->startmsg(d->top->closure);
if (flow != UPB_CONTINUE) _upb_dispatcher_abortjmp(d);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
void upb_dispatch_endmsg(upb_dispatcher *d, upb_status *status) {
assert(d->top == d->stack);
if (d->msgent->endmsg) d->msgent->endmsg(d->top->closure, d->status);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
// TODO: should we avoid this copy by passing client's status obj to cbs?
upb_status_copy(status, d->status);
}
upb_dispatcher_frame *upb_dispatch_startseq(upb_dispatcher *d,
upb_fhandlers *f) {
if (d->top + 1 >= d->limit) {
upb_status_seterrliteral(d->status, "Nesting too deep.");
_upb_dispatcher_abortjmp(d);
}
upb_sflow_t sflow = UPB_CONTINUE_WITH(d->top->closure);
if (f->startseq) sflow = f->startseq(d->top->closure, f->fval);
if (sflow.flow != UPB_CONTINUE) {
_upb_dispatcher_abortjmp(d);
}
++d->top;
d->top->f = f;
d->top->is_sequence = true;
d->top->is_packed = false;
d->top->closure = sflow.closure;
return d->top;
}
upb_dispatcher_frame *upb_dispatch_endseq(upb_dispatcher *d) {
assert(d->top > d->stack);
assert(d->top->is_sequence);
upb_fhandlers *f = d->top->f;
--d->top;
upb_flow_t flow = UPB_CONTINUE;
if (f->endseq) flow = f->endseq(d->top->closure, f->fval);
if (flow != UPB_CONTINUE) {
_upb_dispatcher_abortjmp(d);
}
d->msgent = d->top->f ? d->top->f->submsg : d->toplevel_msgent;
return d->top;
}
upb_dispatcher_frame *upb_dispatch_startsubmsg(upb_dispatcher *d,
upb_fhandlers *f) {
if (d->top + 1 >= d->limit) {
upb_status_seterrliteral(d->status, "Nesting too deep.");
_upb_dispatcher_abortjmp(d);
}
upb_sflow_t sflow = UPB_CONTINUE_WITH(d->top->closure);
if (f->startsubmsg) sflow = f->startsubmsg(d->top->closure, f->fval);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
if (sflow.flow != UPB_CONTINUE) {
_upb_dispatcher_abortjmp(d);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
++d->top;
d->top->f = f;
d->top->is_sequence = false;
d->top->is_packed = false;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
d->top->closure = sflow.closure;
d->msgent = f->submsg;
upb_dispatch_startmsg(d);
return d->top;
}
upb_dispatcher_frame *upb_dispatch_endsubmsg(upb_dispatcher *d) {
assert(d->top > d->stack);
assert(!d->top->is_sequence);
upb_fhandlers *f = d->top->f;
if (d->msgent->endmsg) d->msgent->endmsg(d->top->closure, d->status);
d->msgent = d->top->f->msg;
--d->top;
upb_flow_t flow = UPB_CONTINUE;
if (f->endsubmsg) f->endsubmsg(d->top->closure, f->fval);
if (flow != UPB_CONTINUE) _upb_dispatcher_abortjmp(d);
return d->top;
}
bool upb_dispatcher_stackempty(upb_dispatcher *d) {
return d->top == d->stack;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
bool upb_dispatcher_islegalend(upb_dispatcher *d) {
if (d->top == d->stack) return true;
if (d->top - 1 == d->stack &&
d->top->is_sequence && !d->top->is_packed) return true;
return false;
}
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
void _upb_dispatcher_abortjmp(upb_dispatcher *d) {
d->exitjmp(d->srcclosure);
assert(false); // Never returns.
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}