|
|
|
/*
|
|
|
|
* Copyright (c) 2009-2021, Google LLC
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions are met:
|
|
|
|
* * Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* * Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
* * Neither the name of Google LLC nor the
|
|
|
|
* names of its contributors may be used to endorse or promote products
|
|
|
|
* derived from this software without specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
|
|
|
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL Google LLC BE LIABLE FOR ANY DIRECT,
|
|
|
|
* INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
|
|
|
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
|
|
|
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
|
|
|
|
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
|
|
|
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "upb/decode.h"
|
|
|
|
|
|
|
|
#include <string.h>
|
|
|
|
|
|
|
|
#include "upb/internal/array.h"
|
|
|
|
#include "upb/internal/decode.h"
|
|
|
|
#include "upb/msg_internal.h"
|
|
|
|
#include "upb/upb.h"
|
|
|
|
|
|
|
|
// Must be last.
|
|
|
|
#include "upb/port_def.inc"
|
|
|
|
|
|
|
|
/* Maps descriptor type -> elem_size_lg2. */
|
|
|
|
static const uint8_t desctype_to_elem_size_lg2[] = {
|
|
|
|
-1, /* invalid descriptor type */
|
|
|
|
3, /* DOUBLE */
|
|
|
|
2, /* FLOAT */
|
|
|
|
3, /* INT64 */
|
|
|
|
3, /* UINT64 */
|
|
|
|
2, /* INT32 */
|
|
|
|
3, /* FIXED64 */
|
|
|
|
2, /* FIXED32 */
|
|
|
|
0, /* BOOL */
|
|
|
|
UPB_SIZE(3, 4), /* STRING */
|
|
|
|
UPB_SIZE(2, 3), /* GROUP */
|
|
|
|
UPB_SIZE(2, 3), /* MESSAGE */
|
|
|
|
UPB_SIZE(3, 4), /* BYTES */
|
|
|
|
2, /* UINT32 */
|
|
|
|
2, /* ENUM */
|
|
|
|
2, /* SFIXED32 */
|
|
|
|
3, /* SFIXED64 */
|
|
|
|
2, /* SINT32 */
|
|
|
|
3, /* SINT64 */
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Maps descriptor type -> upb map size. */
|
|
|
|
static const uint8_t desctype_to_mapsize[] = {
|
|
|
|
-1, /* invalid descriptor type */
|
|
|
|
8, /* DOUBLE */
|
|
|
|
4, /* FLOAT */
|
|
|
|
8, /* INT64 */
|
|
|
|
8, /* UINT64 */
|
|
|
|
4, /* INT32 */
|
|
|
|
8, /* FIXED64 */
|
|
|
|
4, /* FIXED32 */
|
|
|
|
1, /* BOOL */
|
|
|
|
UPB_MAPTYPE_STRING, /* STRING */
|
|
|
|
sizeof(void*), /* GROUP */
|
|
|
|
sizeof(void*), /* MESSAGE */
|
|
|
|
UPB_MAPTYPE_STRING, /* BYTES */
|
|
|
|
4, /* UINT32 */
|
|
|
|
4, /* ENUM */
|
|
|
|
4, /* SFIXED32 */
|
|
|
|
8, /* SFIXED64 */
|
|
|
|
4, /* SINT32 */
|
|
|
|
8, /* SINT64 */
|
|
|
|
};
|
|
|
|
|
|
|
|
static const unsigned FIXED32_OK_MASK = (1 << kUpb_FieldType_Float) |
|
|
|
|
(1 << kUpb_FieldType_Fixed32) |
|
|
|
|
(1 << kUpb_FieldType_SFixed32);
|
|
|
|
|
|
|
|
static const unsigned FIXED64_OK_MASK = (1 << kUpb_FieldType_Double) |
|
|
|
|
(1 << kUpb_FieldType_Fixed64) |
|
|
|
|
(1 << kUpb_FieldType_SFixed64);
|
|
|
|
|
|
|
|
/* Three fake field types for MessageSet. */
|
|
|
|
#define TYPE_MSGSET_ITEM 19
|
|
|
|
#define TYPE_COUNT 19
|
|
|
|
|
|
|
|
/* Op: an action to be performed for a wire-type/field-type combination. */
|
|
|
|
#define OP_UNKNOWN -1 /* Unknown field. */
|
|
|
|
#define OP_MSGSET_ITEM -2
|
|
|
|
#define OP_SCALAR_LG2(n) (n) /* n in [0, 2, 3] => op in [0, 2, 3] */
|
|
|
|
#define OP_ENUM 1
|
|
|
|
#define OP_STRING 4
|
|
|
|
#define OP_BYTES 5
|
|
|
|
#define OP_SUBMSG 6
|
|
|
|
/* Scalar fields use only ops above. Repeated fields can use any op. */
|
|
|
|
#define OP_FIXPCK_LG2(n) (n + 5) /* n in [2, 3] => op in [7, 8] */
|
|
|
|
#define OP_VARPCK_LG2(n) (n + 9) /* n in [0, 2, 3] => op in [9, 11, 12] */
|
|
|
|
#define OP_PACKED_ENUM 13
|
|
|
|
|
|
|
|
static const int8_t varint_ops[] = {
|
|
|
|
OP_UNKNOWN, /* field not found */
|
|
|
|
OP_UNKNOWN, /* DOUBLE */
|
|
|
|
OP_UNKNOWN, /* FLOAT */
|
|
|
|
OP_SCALAR_LG2(3), /* INT64 */
|
|
|
|
OP_SCALAR_LG2(3), /* UINT64 */
|
|
|
|
OP_SCALAR_LG2(2), /* INT32 */
|
|
|
|
OP_UNKNOWN, /* FIXED64 */
|
|
|
|
OP_UNKNOWN, /* FIXED32 */
|
|
|
|
OP_SCALAR_LG2(0), /* BOOL */
|
|
|
|
OP_UNKNOWN, /* STRING */
|
|
|
|
OP_UNKNOWN, /* GROUP */
|
|
|
|
OP_UNKNOWN, /* MESSAGE */
|
|
|
|
OP_UNKNOWN, /* BYTES */
|
|
|
|
OP_SCALAR_LG2(2), /* UINT32 */
|
|
|
|
OP_ENUM, /* ENUM */
|
|
|
|
OP_UNKNOWN, /* SFIXED32 */
|
|
|
|
OP_UNKNOWN, /* SFIXED64 */
|
|
|
|
OP_SCALAR_LG2(2), /* SINT32 */
|
|
|
|
OP_SCALAR_LG2(3), /* SINT64 */
|
|
|
|
OP_UNKNOWN, /* MSGSET_ITEM */
|
|
|
|
};
|
|
|
|
|
|
|
|
static const int8_t delim_ops[] = {
|
|
|
|
/* For non-repeated field type. */
|
|
|
|
OP_UNKNOWN, /* field not found */
|
|
|
|
OP_UNKNOWN, /* DOUBLE */
|
|
|
|
OP_UNKNOWN, /* FLOAT */
|
|
|
|
OP_UNKNOWN, /* INT64 */
|
|
|
|
OP_UNKNOWN, /* UINT64 */
|
|
|
|
OP_UNKNOWN, /* INT32 */
|
|
|
|
OP_UNKNOWN, /* FIXED64 */
|
|
|
|
OP_UNKNOWN, /* FIXED32 */
|
|
|
|
OP_UNKNOWN, /* BOOL */
|
|
|
|
OP_STRING, /* STRING */
|
|
|
|
OP_UNKNOWN, /* GROUP */
|
|
|
|
OP_SUBMSG, /* MESSAGE */
|
|
|
|
OP_BYTES, /* BYTES */
|
|
|
|
OP_UNKNOWN, /* UINT32 */
|
|
|
|
OP_UNKNOWN, /* ENUM */
|
|
|
|
OP_UNKNOWN, /* SFIXED32 */
|
|
|
|
OP_UNKNOWN, /* SFIXED64 */
|
|
|
|
OP_UNKNOWN, /* SINT32 */
|
|
|
|
OP_UNKNOWN, /* SINT64 */
|
|
|
|
OP_UNKNOWN, /* MSGSET_ITEM */
|
|
|
|
/* For repeated field type. */
|
|
|
|
OP_FIXPCK_LG2(3), /* REPEATED DOUBLE */
|
|
|
|
OP_FIXPCK_LG2(2), /* REPEATED FLOAT */
|
|
|
|
OP_VARPCK_LG2(3), /* REPEATED INT64 */
|
|
|
|
OP_VARPCK_LG2(3), /* REPEATED UINT64 */
|
|
|
|
OP_VARPCK_LG2(2), /* REPEATED INT32 */
|
|
|
|
OP_FIXPCK_LG2(3), /* REPEATED FIXED64 */
|
|
|
|
OP_FIXPCK_LG2(2), /* REPEATED FIXED32 */
|
|
|
|
OP_VARPCK_LG2(0), /* REPEATED BOOL */
|
|
|
|
OP_STRING, /* REPEATED STRING */
|
|
|
|
OP_SUBMSG, /* REPEATED GROUP */
|
|
|
|
OP_SUBMSG, /* REPEATED MESSAGE */
|
|
|
|
OP_BYTES, /* REPEATED BYTES */
|
|
|
|
OP_VARPCK_LG2(2), /* REPEATED UINT32 */
|
|
|
|
OP_PACKED_ENUM, /* REPEATED ENUM */
|
|
|
|
OP_FIXPCK_LG2(2), /* REPEATED SFIXED32 */
|
|
|
|
OP_FIXPCK_LG2(3), /* REPEATED SFIXED64 */
|
|
|
|
OP_VARPCK_LG2(2), /* REPEATED SINT32 */
|
|
|
|
OP_VARPCK_LG2(3), /* REPEATED SINT64 */
|
|
|
|
/* Omitting MSGSET_*, because we never emit a repeated msgset type */
|
|
|
|
};
|
|
|
|
|
|
|
|
typedef union {
|
|
|
|
bool bool_val;
|
|
|
|
uint32_t uint32_val;
|
|
|
|
uint64_t uint64_val;
|
|
|
|
uint32_t size;
|
|
|
|
} wireval;
|
|
|
|
|
|
|
|
static const char* _upb_Decoder_DecodeMessage(upb_Decoder* d, const char* ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
const upb_MiniTable* layout);
|
|
|
|
|
|
|
|
UPB_NORETURN static void* _upb_Decoder_ErrorJmp(upb_Decoder* d,
|
|
|
|
upb_DecodeStatus status) {
|
|
|
|
assert(status != kUpb_DecodeStatus_Ok);
|
|
|
|
UPB_LONGJMP(d->err, status);
|
|
|
|
}
|
|
|
|
|
|
|
|
const char* _upb_FastDecoder_ErrorJmp(upb_Decoder* d, int status) {
|
|
|
|
assert(status != kUpb_DecodeStatus_Ok);
|
|
|
|
UPB_LONGJMP(d->err, status);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
static void _upb_Decoder_VerifyUtf8(upb_Decoder* d, const char* buf, int len) {
|
|
|
|
if (!_upb_Decoder_VerifyUtf8Inline(buf, len)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_BadUtf8);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool _upb_Decoder_Reserve(upb_Decoder* d, upb_Array* arr, size_t elem) {
|
|
|
|
bool need_realloc = arr->capacity - arr->size < elem;
|
|
|
|
if (need_realloc && !_upb_array_realloc(arr, arr->size + elem, &d->arena)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
|
|
|
return need_realloc;
|
|
|
|
}
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
const char* ptr;
|
|
|
|
uint64_t val;
|
|
|
|
} _upb_DecodeLongVarintReturn;
|
|
|
|
|
|
|
|
UPB_NOINLINE
|
|
|
|
static _upb_DecodeLongVarintReturn _upb_Decoder_DecodeLongVarint(
|
|
|
|
const char* ptr, uint64_t val) {
|
|
|
|
_upb_DecodeLongVarintReturn ret = {NULL, 0};
|
|
|
|
uint64_t byte;
|
|
|
|
int i;
|
|
|
|
for (i = 1; i < 10; i++) {
|
|
|
|
byte = (uint8_t)ptr[i];
|
|
|
|
val += (byte - 1) << (i * 7);
|
|
|
|
if (!(byte & 0x80)) {
|
|
|
|
ret.ptr = ptr + i + 1;
|
|
|
|
ret.val = val;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeVarint(upb_Decoder* d, const char* ptr,
|
|
|
|
uint64_t* val) {
|
|
|
|
uint64_t byte = (uint8_t)*ptr;
|
|
|
|
if (UPB_LIKELY((byte & 0x80) == 0)) {
|
|
|
|
*val = byte;
|
|
|
|
return ptr + 1;
|
|
|
|
} else {
|
|
|
|
_upb_DecodeLongVarintReturn res = _upb_Decoder_DecodeLongVarint(ptr, byte);
|
|
|
|
if (!res.ptr) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
*val = res.val;
|
|
|
|
return res.ptr;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeTag(upb_Decoder* d, const char* ptr,
|
|
|
|
uint32_t* val) {
|
|
|
|
uint64_t byte = (uint8_t)*ptr;
|
|
|
|
if (UPB_LIKELY((byte & 0x80) == 0)) {
|
|
|
|
*val = byte;
|
|
|
|
return ptr + 1;
|
|
|
|
} else {
|
|
|
|
const char* start = ptr;
|
|
|
|
_upb_DecodeLongVarintReturn res = _upb_Decoder_DecodeLongVarint(ptr, byte);
|
|
|
|
if (!res.ptr || res.ptr - start > 5 || res.val > UINT32_MAX) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
*val = res.val;
|
|
|
|
return res.ptr;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* upb_Decoder_DecodeSize(upb_Decoder* d, const char* ptr,
|
|
|
|
uint32_t* size) {
|
|
|
|
uint64_t size64;
|
|
|
|
ptr = _upb_Decoder_DecodeVarint(d, ptr, &size64);
|
|
|
|
if (size64 >= INT32_MAX || ptr - d->end + (int)size64 > d->limit) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
*size = size64;
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void _upb_Decoder_MungeInt32(wireval* val) {
|
|
|
|
if (!_upb_IsLittleEndian()) {
|
|
|
|
/* The next stage will memcpy(dst, &val, 4) */
|
|
|
|
val->uint32_val = val->uint64_val;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void _upb_Decoder_Munge(int type, wireval* val) {
|
|
|
|
switch (type) {
|
|
|
|
case kUpb_FieldType_Bool:
|
|
|
|
val->bool_val = val->uint64_val != 0;
|
|
|
|
break;
|
|
|
|
case kUpb_FieldType_SInt32: {
|
|
|
|
uint32_t n = val->uint64_val;
|
|
|
|
val->uint32_val = (n >> 1) ^ -(int32_t)(n & 1);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case kUpb_FieldType_SInt64: {
|
|
|
|
uint64_t n = val->uint64_val;
|
|
|
|
val->uint64_val = (n >> 1) ^ -(int64_t)(n & 1);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case kUpb_FieldType_Int32:
|
|
|
|
case kUpb_FieldType_UInt32:
|
|
|
|
case kUpb_FieldType_Enum:
|
|
|
|
_upb_Decoder_MungeInt32(val);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static upb_Message* _upb_Decoder_NewSubMessage(
|
|
|
|
upb_Decoder* d, const upb_MiniTable_Sub* subs,
|
|
|
|
const upb_MiniTable_Field* field) {
|
|
|
|
const upb_MiniTable* subl = subs[field->submsg_index].submsg;
|
|
|
|
upb_Message* msg = _upb_Message_New_inl(subl, &d->arena);
|
|
|
|
if (!msg) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
return msg;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_NOINLINE
|
|
|
|
const char* _upb_Decoder_IsDoneFallback(upb_Decoder* d, const char* ptr,
|
|
|
|
int overrun) {
|
|
|
|
int status;
|
|
|
|
ptr = _upb_Decoder_IsDoneFallbackInline(d, ptr, overrun, &status);
|
|
|
|
if (ptr == NULL) _upb_Decoder_ErrorJmp(d, status);
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* _upb_Decoder_ReadString(upb_Decoder* d, const char* ptr,
|
|
|
|
int size, upb_StringView* str) {
|
|
|
|
if (d->options & kUpb_DecodeOption_AliasString) {
|
|
|
|
str->data = ptr;
|
|
|
|
} else {
|
|
|
|
char* data = upb_Arena_Malloc(&d->arena, size);
|
|
|
|
if (!data) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
memcpy(data, ptr, size);
|
|
|
|
str->data = data;
|
|
|
|
}
|
|
|
|
str->size = size;
|
|
|
|
return ptr + size;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeSubMessage(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Message* submsg,
|
|
|
|
const upb_MiniTable_Sub* subs, const upb_MiniTable_Field* field, int size) {
|
|
|
|
int saved_delta = _upb_Decoder_PushLimit(d, ptr, size);
|
|
|
|
const upb_MiniTable* subl = subs[field->submsg_index].submsg;
|
|
|
|
if (--d->depth < 0) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_MaxDepthExceeded);
|
|
|
|
}
|
|
|
|
ptr = _upb_Decoder_DecodeMessage(d, ptr, submsg, subl);
|
|
|
|
if (d->end_group != DECODE_NOGROUP) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
_upb_Decoder_PopLimit(d, ptr, saved_delta);
|
|
|
|
d->depth++;
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DoDecodeGroup(upb_Decoder* d, const char* ptr,
|
|
|
|
upb_Message* submsg,
|
|
|
|
const upb_MiniTable* subl,
|
|
|
|
uint32_t number) {
|
|
|
|
if (--d->depth < 0) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_MaxDepthExceeded);
|
|
|
|
}
|
|
|
|
if (_upb_Decoder_IsDone(d, &ptr)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
ptr = _upb_Decoder_DecodeMessage(d, ptr, submsg, subl);
|
|
|
|
if (d->end_group != number) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
d->end_group = DECODE_NOGROUP;
|
|
|
|
d->depth++;
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeUnknownGroup(upb_Decoder* d,
|
|
|
|
const char* ptr,
|
|
|
|
uint32_t number) {
|
|
|
|
return _upb_Decoder_DoDecodeGroup(d, ptr, NULL, NULL, number);
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeKnownGroup(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Message* submsg,
|
|
|
|
const upb_MiniTable_Sub* subs, const upb_MiniTable_Field* field) {
|
|
|
|
const upb_MiniTable* subl = subs[field->submsg_index].submsg;
|
|
|
|
return _upb_Decoder_DoDecodeGroup(d, ptr, submsg, subl, field->number);
|
|
|
|
}
|
|
|
|
|
|
|
|
static char* upb_Decoder_EncodeVarint32(uint32_t val, char* ptr) {
|
|
|
|
do {
|
|
|
|
uint8_t byte = val & 0x7fU;
|
|
|
|
val >>= 7;
|
|
|
|
if (val) byte |= 0x80U;
|
|
|
|
*(ptr++) = byte;
|
|
|
|
} while (val);
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void _upb_Decoder_AddUnknownVarints(upb_Decoder* d, upb_Message* msg,
|
|
|
|
uint32_t val1, uint32_t val2) {
|
|
|
|
char buf[20];
|
|
|
|
char* end = buf;
|
|
|
|
end = upb_Decoder_EncodeVarint32(val1, end);
|
|
|
|
end = upb_Decoder_EncodeVarint32(val2, end);
|
|
|
|
|
|
|
|
if (!_upb_Message_AddUnknown(msg, buf, end - buf, &d->arena)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_NOINLINE
|
|
|
|
static bool _upb_Decoder_CheckEnumSlow(upb_Decoder* d, const char* ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
const upb_MiniTable_Enum* e,
|
|
|
|
const upb_MiniTable_Field* field,
|
|
|
|
uint32_t v) {
|
|
|
|
if (_upb_MiniTable_CheckEnumValueSlow(e, v)) return true;
|
|
|
|
|
|
|
|
// Unrecognized enum goes into unknown fields.
|
|
|
|
// For packed fields the tag could be arbitrarily far in the past, so we
|
|
|
|
// just re-encode the tag and value here.
|
|
|
|
uint32_t tag = ((uint32_t)field->number << 3) | kUpb_WireType_Varint;
|
Fixed bug when parsing an unknown value in a proto2 enum extension. #fuzzing
Proto2 enum parsing is the only case where we have to look at the wire value (not merely the tag) to decide whether the field is known or unknown. If the value is unknown, we need to put the value in the Unknown Fields, but for an extension we no longer have easy access to the message, because for extensions we replace the `msg` pointer with a pointer to the extension. The bug occurred when we were treating the fake `upb_Message*` (which was actually a pointer to an extension) as a real `upb_Message*` that can have unknown fields.
This CL fixes the problem by preserving the true message pointer in `d->unknown_msg` when we are parsing an extension.
This also required fixing a bug in MiniTable building when fasttables are enabled. We need to set the table_mask to `-1` to disable fasttable parsing, not `0`.
For unknown reasons, this CL appears to speed up parsing somewhat significantly. Ideally we should be tracking parsing performance better over time, as it is possible this is merely regaining performance that was lost at a different time:
```
benchy --reference=srcfs third_party/upb/benchmarks:benchmark
10 / 10 [=================================================================================================================] 100.00% 2m32s
(Generated by http://go/benchy. Settings: --runs 5 --reference "srcfs")
name old cpu/op new cpu/op delta
BM_ArenaOneAlloc 23.9ns ± 6% 23.7ns ± 4% ~ (p=0.180 n=53+51)
BM_ArenaInitialBlockOneAlloc 7.62ns ± 4% 7.70ns ± 5% +0.99% (p=0.024 n=59+60)
BM_LoadAdsDescriptor_Upb<NoLayout> 6.60ms ±10% 6.57ms ± 8% ~ (p=0.607 n=47+54)
BM_LoadAdsDescriptor_Upb<WithLayout> 6.92ms ± 5% 6.88ms ± 8% ~ (p=0.257 n=54+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 14.2ms ± 8% 14.0ms ± 7% -1.38% (p=0.025 n=58+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 14.3ms ± 8% 14.2ms ± 8% -1.16% (p=0.031 n=58+57)
BM_Parse_Upb_FileDesc<UseArena, Copy> 15.9µs ± 4% 14.6µs ± 4% -7.85% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 14.5µs ± 4% 13.3µs ± 5% -8.50% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 15.7µs ± 4% 14.4µs ± 5% -7.99% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 14.2µs ± 5% 13.0µs ± 4% -8.56% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 26.3µs ± 4% 26.2µs ± 4% ~ (p=0.195 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 13.3µs ± 5% 13.2µs ± 4% ~ (p=0.085 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 12.9µs ± 4% 12.8µs ± 3% -0.66% (p=0.023 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 10.9µs ± 6% 10.9µs ± 4% ~ (p=0.063 n=59+58)
BM_SerializeDescriptor_Proto2 7.57µs ± 6% 7.62µs ± 6% ~ (p=0.147 n=57+58)
BM_SerializeDescriptor_Upb 12.8µs ± 4% 12.8µs ± 4% ~ (p=0.163 n=59+56)
name old time/op new time/op delta
BM_ArenaOneAlloc 23.9ns ± 5% 23.7ns ± 4% ~ (p=0.172 n=53+51)
BM_ArenaInitialBlockOneAlloc 7.62ns ± 4% 7.70ns ± 5% +1.02% (p=0.017 n=59+60)
BM_LoadAdsDescriptor_Upb<NoLayout> 6.60ms ±10% 6.58ms ± 8% ~ (p=0.727 n=47+55)
BM_LoadAdsDescriptor_Upb<WithLayout> 6.92ms ± 5% 6.88ms ± 8% ~ (p=0.260 n=54+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 14.2ms ± 7% 14.0ms ± 7% -1.40% (p=0.019 n=58+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 14.3ms ± 8% 14.2ms ± 8% -1.13% (p=0.037 n=58+57)
BM_Parse_Upb_FileDesc<UseArena, Copy> 15.9µs ± 4% 14.6µs ± 3% -7.88% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 14.5µs ± 4% 13.3µs ± 5% -8.46% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 15.7µs ± 4% 14.4µs ± 5% -7.99% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 14.2µs ± 5% 13.0µs ± 4% -8.56% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 26.3µs ± 4% 26.2µs ± 4% ~ (p=0.224 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 13.3µs ± 5% 13.2µs ± 4% ~ (p=0.098 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 12.9µs ± 4% 12.8µs ± 3% -0.68% (p=0.015 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 10.9µs ± 6% 10.9µs ± 4% ~ (p=0.052 n=59+58)
BM_SerializeDescriptor_Proto2 7.56µs ± 6% 7.62µs ± 6% ~ (p=0.111 n=58+58)
BM_SerializeDescriptor_Upb 12.8µs ± 4% 12.8µs ± 4% ~ (p=0.241 n=56+56)
name old allocs/op new allocs/op delta
BM_ArenaOneAlloc 1.00 ± 0% 1.00 ± 0% ~ (all samples are equal)
BM_ArenaInitialBlockOneAlloc 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<NoLayout> 5.98k ± 0% 5.98k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<WithLayout> 5.98k ± 0% 5.98k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<NoLayout> 80.9k ± 0% 80.9k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<WithLayout> 82.1k ± 0% 82.1k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Copy> 7.00 ± 0% 7.00 ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Alias> 7.00 ± 0% 7.00 ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 765 ± 0% 765 ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 9.00 ± 0% 9.00 ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Proto2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Upb 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
name old peak-mem(Bytes)/op new peak-mem(Bytes)/op delta
BM_ArenaOneAlloc 344 ± 0% 344 ± 0% ~ (all samples are equal)
BM_ArenaInitialBlockOneAlloc 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<NoLayout> 9.60M ± 0% 9.60M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<WithLayout> 9.68M ± 0% 9.68M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<NoLayout> 6.41M ± 0% 6.41M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<WithLayout> 6.44M ± 0% 6.44M ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Copy> 36.5k ± 0% 36.5k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Alias> 36.5k ± 0% 36.5k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 35.8k ± 0% 35.8k ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 40.7k ± 0% 40.7k ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Proto2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Upb 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
name old speed new speed delta
BM_LoadAdsDescriptor_Upb<NoLayout> 113MB/s ± 9% 113MB/s ± 8% ~ (p=0.712 n=47+55)
BM_LoadAdsDescriptor_Upb<WithLayout> 107MB/s ± 8% 108MB/s ± 8% ~ (p=0.200 n=55+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 52.5MB/s ± 8% 53.3MB/s ± 7% +1.51% (p=0.018 n=59+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 51.9MB/s ± 7% 52.4MB/s ± 8% +1.01% (p=0.050 n=58+58)
BM_Parse_Upb_FileDesc<UseArena, Copy> 473MB/s ± 4% 514MB/s ± 4% +8.52% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 518MB/s ± 4% 566MB/s ± 5% +9.30% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 480MB/s ± 4% 521MB/s ± 5% +8.69% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 528MB/s ± 4% 578MB/s ± 4% +9.36% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 286MB/s ± 4% 287MB/s ± 4% ~ (p=0.195 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 566MB/s ± 5% 570MB/s ± 4% ~ (p=0.085 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 583MB/s ± 5% 587MB/s ± 3% +0.64% (p=0.023 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 688MB/s ± 6% 693MB/s ± 4% ~ (p=0.063 n=59+58)
BM_SerializeDescriptor_Proto2 995MB/s ± 6% 988MB/s ± 5% ~ (p=0.147 n=57+58)
BM_SerializeDescriptor_Upb 586MB/s ± 4% 589MB/s ± 4% ~ (p=0.163 n=59+56)
```
PiperOrigin-RevId: 462022073
2 years ago
|
|
|
upb_Message* unknown_msg =
|
|
|
|
field->mode & kUpb_LabelFlags_IsExtension ? d->unknown_msg : msg;
|
|
|
|
_upb_Decoder_AddUnknownVarints(d, unknown_msg, tag, v);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static bool _upb_Decoder_CheckEnum(upb_Decoder* d, const char* ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
const upb_MiniTable_Enum* e,
|
|
|
|
const upb_MiniTable_Field* field,
|
|
|
|
wireval* val) {
|
|
|
|
uint32_t v = val->uint32_val;
|
|
|
|
|
|
|
|
_kUpb_FastEnumCheck_Status status = _upb_MiniTable_CheckEnumValueFast(e, v);
|
|
|
|
if (UPB_LIKELY(status == _kUpb_FastEnumCheck_ValueIsInEnum)) return true;
|
|
|
|
return _upb_Decoder_CheckEnumSlow(d, ptr, msg, e, field, v);
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_NOINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeEnumArray(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Message* msg, upb_Array* arr,
|
|
|
|
const upb_MiniTable_Sub* subs, const upb_MiniTable_Field* field,
|
|
|
|
wireval* val) {
|
|
|
|
const upb_MiniTable_Enum* e = subs[field->submsg_index].subenum;
|
|
|
|
if (!_upb_Decoder_CheckEnum(d, ptr, msg, e, field, val)) return ptr;
|
|
|
|
void* mem = UPB_PTR_AT(_upb_array_ptr(arr), arr->size * 4, void);
|
|
|
|
arr->size++;
|
|
|
|
memcpy(mem, val, 4);
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeFixedPacked(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Array* arr, wireval* val,
|
|
|
|
const upb_MiniTable_Field* field, int lg2) {
|
|
|
|
int mask = (1 << lg2) - 1;
|
|
|
|
size_t count = val->size >> lg2;
|
|
|
|
if ((val->size & mask) != 0) {
|
|
|
|
// Length isn't a round multiple of elem size.
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
_upb_Decoder_Reserve(d, arr, count);
|
|
|
|
void* mem = UPB_PTR_AT(_upb_array_ptr(arr), arr->size << lg2, void);
|
|
|
|
arr->size += count;
|
|
|
|
// Note: if/when the decoder supports multi-buffer input, we will need to
|
|
|
|
// handle buffer seams here.
|
|
|
|
if (_upb_IsLittleEndian()) {
|
|
|
|
memcpy(mem, ptr, val->size);
|
|
|
|
ptr += val->size;
|
|
|
|
} else {
|
|
|
|
const char* end = ptr + val->size;
|
|
|
|
char* dst = mem;
|
|
|
|
while (ptr < end) {
|
|
|
|
if (lg2 == 2) {
|
|
|
|
uint32_t val;
|
|
|
|
memcpy(&val, ptr, sizeof(val));
|
|
|
|
val = _upb_BigEndian_Swap32(val);
|
|
|
|
memcpy(dst, &val, sizeof(val));
|
|
|
|
} else {
|
|
|
|
UPB_ASSERT(lg2 == 3);
|
|
|
|
uint64_t val;
|
|
|
|
memcpy(&val, ptr, sizeof(val));
|
|
|
|
val = _upb_BigEndian_Swap64(val);
|
|
|
|
memcpy(dst, &val, sizeof(val));
|
|
|
|
}
|
|
|
|
ptr += 1 << lg2;
|
|
|
|
dst += 1 << lg2;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeVarintPacked(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Array* arr, wireval* val,
|
|
|
|
const upb_MiniTable_Field* field, int lg2) {
|
|
|
|
int scale = 1 << lg2;
|
|
|
|
int saved_limit = _upb_Decoder_PushLimit(d, ptr, val->size);
|
|
|
|
char* out = UPB_PTR_AT(_upb_array_ptr(arr), arr->size << lg2, void);
|
|
|
|
while (!_upb_Decoder_IsDone(d, &ptr)) {
|
|
|
|
wireval elem;
|
|
|
|
ptr = _upb_Decoder_DecodeVarint(d, ptr, &elem.uint64_val);
|
|
|
|
_upb_Decoder_Munge(field->descriptortype, &elem);
|
|
|
|
if (_upb_Decoder_Reserve(d, arr, 1)) {
|
|
|
|
out = UPB_PTR_AT(_upb_array_ptr(arr), arr->size << lg2, void);
|
|
|
|
}
|
|
|
|
arr->size++;
|
|
|
|
memcpy(out, &elem, scale);
|
|
|
|
out += scale;
|
|
|
|
}
|
|
|
|
_upb_Decoder_PopLimit(d, ptr, saved_limit);
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_NOINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeEnumPacked(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Message* msg, upb_Array* arr,
|
|
|
|
const upb_MiniTable_Sub* subs, const upb_MiniTable_Field* field,
|
|
|
|
wireval* val) {
|
|
|
|
const upb_MiniTable_Enum* e = subs[field->submsg_index].subenum;
|
|
|
|
int saved_limit = _upb_Decoder_PushLimit(d, ptr, val->size);
|
|
|
|
char* out = UPB_PTR_AT(_upb_array_ptr(arr), arr->size * 4, void);
|
|
|
|
while (!_upb_Decoder_IsDone(d, &ptr)) {
|
|
|
|
wireval elem;
|
|
|
|
ptr = _upb_Decoder_DecodeVarint(d, ptr, &elem.uint64_val);
|
|
|
|
_upb_Decoder_MungeInt32(&elem);
|
|
|
|
if (!_upb_Decoder_CheckEnum(d, ptr, msg, e, field, &elem)) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (_upb_Decoder_Reserve(d, arr, 1)) {
|
|
|
|
out = UPB_PTR_AT(_upb_array_ptr(arr), arr->size * 4, void);
|
|
|
|
}
|
|
|
|
arr->size++;
|
|
|
|
memcpy(out, &elem, 4);
|
|
|
|
out += 4;
|
|
|
|
}
|
|
|
|
_upb_Decoder_PopLimit(d, ptr, saved_limit);
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* _upb_Decoder_DecodeToArray(upb_Decoder* d, const char* ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
const upb_MiniTable_Sub* subs,
|
|
|
|
const upb_MiniTable_Field* field,
|
|
|
|
wireval* val, int op) {
|
|
|
|
upb_Array** arrp = UPB_PTR_AT(msg, field->offset, void);
|
|
|
|
upb_Array* arr = *arrp;
|
|
|
|
void* mem;
|
|
|
|
|
|
|
|
if (arr) {
|
|
|
|
_upb_Decoder_Reserve(d, arr, 1);
|
|
|
|
} else {
|
|
|
|
size_t lg2 = desctype_to_elem_size_lg2[field->descriptortype];
|
|
|
|
arr = _upb_Array_New(&d->arena, 4, lg2);
|
|
|
|
if (!arr) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
*arrp = arr;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (op) {
|
|
|
|
case OP_SCALAR_LG2(0):
|
|
|
|
case OP_SCALAR_LG2(2):
|
|
|
|
case OP_SCALAR_LG2(3):
|
|
|
|
/* Append scalar value. */
|
|
|
|
mem = UPB_PTR_AT(_upb_array_ptr(arr), arr->size << op, void);
|
|
|
|
arr->size++;
|
Optimized decoder and paved the way for parsing extensions.
The primary motivation for this change is to avoid referring to the
`upb_msglayout` object when we are trying to fetch the `upb_msglayout`
object for a sub-message. This will help pave the way for parsing
extensions. We also implement several optimizations so that we can
make this change without regressing performance.
Normally we compute the layout for a sub-message field like so:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *layout,
const upb_msglayout_field *field) {
return layout->submsgs[field->submsg_index]
}
```
The reason for this indirection is to avoid storing a pointer directly
in `upb_msglayout_field`, as this would double its size (from 12 to 24
bytes on 64-bit architectures) which is wasteful as this pointer is
only needed for message typed fields.
However `get_submsg_layout` as written above does not work for
extensions, as they will not have entries in the message's
`layout->submsgs` array by nature, and we want to avoid creating
an entire fake `upb_msglayout` for each such extension since that
would also be wasteful.
This change removes the dependency on `upb_msglayout` by passing down
the `submsgs` array instead:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *const *submsgs,
const upb_msglayout_field *field) {
return submsgs[field->submsg_index]
}
```
This will pave the way for parsing extensions, as we can more easily
create an alternative `submsgs` array for extension fields without
extra overhead or waste.
Along the way several optimizations presented themselves that allow
a nice increase in performance:
1. Passing the parsed `wireval` by address instead of by value ended
up avoiding an expensive and useless stack copy (this is on Clang,
which was used for all measurements).
2. When field numbers are densely packed, we can find a field by number
with a single indexed lookup instead of linear search. At codegen
time we can compute the maximum field number that will allow such
an indexed lookup.
3. For fields that do require linear search, we can start the linear
search at the location where we found the previous field, taking
advantage of the fact that field numbers are generally increasing.
4. When the hasbit index is less than 32 (the common case) we can use
a less expensive code sequence to set it.
5. We check for the hasbit case before the oneof case, as optional
fields are more common than oneof fields.
Benchmark results indicate a 20% improvement in parse speed with a
small code size increase:
```
name old time/op new time/op delta
ArenaOneAlloc 21.3ns ± 0% 21.5ns ± 0% +0.96% (p=0.000 n=12+12)
ArenaInitialBlockOneAlloc 6.32ns ± 0% 6.32ns ± 0% +0.03% (p=0.000 n=12+10)
LoadDescriptor_Upb 53.5µs ± 1% 51.5µs ± 2% -3.70% (p=0.000 n=12+12)
LoadAdsDescriptor_Upb 2.78ms ± 2% 2.68ms ± 0% -3.57% (p=0.000 n=12+12)
LoadDescriptor_Proto2 240µs ± 0% 240µs ± 0% +0.12% (p=0.001 n=12+12)
LoadAdsDescriptor_Proto2 12.8ms ± 0% 12.7ms ± 0% -1.15% (p=0.000 n=12+10)
Parse_Upb_FileDesc<UseArena,Copy> 13.2µs ± 2% 10.7µs ± 0% -18.49% (p=0.000 n=10+12)
Parse_Upb_FileDesc<UseArena,Alias> 11.3µs ± 0% 9.6µs ± 0% -15.11% (p=0.000 n=12+11)
Parse_Upb_FileDesc<InitBlock,Copy> 12.7µs ± 0% 10.3µs ± 0% -19.00% (p=0.000 n=10+12)
Parse_Upb_FileDesc<InitBlock,Alias> 10.9µs ± 0% 9.2µs ± 0% -15.82% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,NoArena,Copy> 29.4µs ± 0% 29.5µs ± 0% +0.61% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,UseArena,Copy> 20.7µs ± 2% 20.6µs ± 2% ~ (p=0.260 n=12+11)
Parse_Proto2<FileDesc,InitBlock,Copy> 16.7µs ± 1% 16.7µs ± 0% -0.25% (p=0.036 n=12+10)
Parse_Proto2<FileDescSV,InitBlock,Alias> 16.5µs ± 0% 16.5µs ± 0% +0.20% (p=0.016 n=12+11)
SerializeDescriptor_Proto2 5.30µs ± 1% 5.36µs ± 1% +1.09% (p=0.000 n=12+11)
SerializeDescriptor_Upb 12.9µs ± 0% 13.0µs ± 0% +0.90% (p=0.000 n=12+11)
FILE SIZE VM SIZE
-------------- --------------
+1.5% +176 +1.6% +176 upb/decode.c
+1.8% +176 +1.9% +176 decode_msg
+0.4% +64 +0.4% +64 upb/def.c
+1.4% +64 +1.4% +64 _upb_symtab_addfile
+1.2% +48 +1.4% +48 upb/reflection.c
+15% +32 +18% +32 upb_msg_set
+2.9% +16 +3.1% +16 upb_msg_mutable
-9.3% -288 [ = ] 0 [Unmapped]
[ = ] 0 +0.2% +288 TOTAL
```
4 years ago
|
|
|
memcpy(mem, val, 1 << op);
|
|
|
|
return ptr;
|
|
|
|
case OP_STRING:
|
|
|
|
_upb_Decoder_VerifyUtf8(d, ptr, val->size);
|
|
|
|
/* Fallthrough. */
|
|
|
|
case OP_BYTES: {
|
|
|
|
/* Append bytes. */
|
|
|
|
upb_StringView* str = (upb_StringView*)_upb_array_ptr(arr) + arr->size;
|
|
|
|
arr->size++;
|
|
|
|
return _upb_Decoder_ReadString(d, ptr, val->size, str);
|
|
|
|
}
|
|
|
|
case OP_SUBMSG: {
|
|
|
|
/* Append submessage / group. */
|
|
|
|
upb_Message* submsg = _upb_Decoder_NewSubMessage(d, subs, field);
|
|
|
|
*UPB_PTR_AT(_upb_array_ptr(arr), arr->size * sizeof(void*),
|
|
|
|
upb_Message*) = submsg;
|
|
|
|
arr->size++;
|
|
|
|
if (UPB_UNLIKELY(field->descriptortype == kUpb_FieldType_Group)) {
|
|
|
|
return _upb_Decoder_DecodeKnownGroup(d, ptr, submsg, subs, field);
|
|
|
|
} else {
|
|
|
|
return _upb_Decoder_DecodeSubMessage(d, ptr, submsg, subs, field,
|
|
|
|
val->size);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
case OP_FIXPCK_LG2(2):
|
|
|
|
case OP_FIXPCK_LG2(3):
|
|
|
|
return _upb_Decoder_DecodeFixedPacked(d, ptr, arr, val, field,
|
|
|
|
op - OP_FIXPCK_LG2(0));
|
|
|
|
case OP_VARPCK_LG2(0):
|
|
|
|
case OP_VARPCK_LG2(2):
|
|
|
|
case OP_VARPCK_LG2(3):
|
|
|
|
return _upb_Decoder_DecodeVarintPacked(d, ptr, arr, val, field,
|
|
|
|
op - OP_VARPCK_LG2(0));
|
|
|
|
case OP_ENUM:
|
|
|
|
return _upb_Decoder_DecodeEnumArray(d, ptr, msg, arr, subs, field, val);
|
|
|
|
case OP_PACKED_ENUM:
|
|
|
|
return _upb_Decoder_DecodeEnumPacked(d, ptr, msg, arr, subs, field, val);
|
|
|
|
default:
|
|
|
|
UPB_UNREACHABLE();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* _upb_Decoder_DecodeToMap(upb_Decoder* d, const char* ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
const upb_MiniTable_Sub* subs,
|
|
|
|
const upb_MiniTable_Field* field,
|
|
|
|
wireval* val) {
|
|
|
|
upb_Map** map_p = UPB_PTR_AT(msg, field->offset, upb_Map*);
|
|
|
|
upb_Map* map = *map_p;
|
|
|
|
upb_MapEntry ent;
|
|
|
|
const upb_MiniTable* entry = subs[field->submsg_index].submsg;
|
|
|
|
|
|
|
|
if (!map) {
|
|
|
|
/* Lazily create map. */
|
|
|
|
const upb_MiniTable_Field* key_field = &entry->fields[0];
|
|
|
|
const upb_MiniTable_Field* val_field = &entry->fields[1];
|
|
|
|
char key_size = desctype_to_mapsize[key_field->descriptortype];
|
|
|
|
char val_size = desctype_to_mapsize[val_field->descriptortype];
|
|
|
|
UPB_ASSERT(key_field->offset == 0);
|
|
|
|
UPB_ASSERT(val_field->offset == sizeof(upb_StringView));
|
|
|
|
map = _upb_Map_New(&d->arena, key_size, val_size);
|
|
|
|
*map_p = map;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Parse map entry. */
|
|
|
|
memset(&ent, 0, sizeof(ent));
|
|
|
|
|
|
|
|
if (entry->fields[1].descriptortype == kUpb_FieldType_Message ||
|
|
|
|
entry->fields[1].descriptortype == kUpb_FieldType_Group) {
|
|
|
|
/* Create proactively to handle the case where it doesn't appear. */
|
|
|
|
ent.v.val =
|
|
|
|
upb_value_ptr(_upb_Message_New(entry->subs[0].submsg, &d->arena));
|
|
|
|
}
|
|
|
|
|
|
|
|
const char* start = ptr;
|
|
|
|
ptr = _upb_Decoder_DecodeSubMessage(d, ptr, &ent.k, subs, field, val->size);
|
|
|
|
// check if ent had any unknown fields
|
|
|
|
size_t size;
|
|
|
|
upb_Message_GetUnknown(&ent.k, &size);
|
|
|
|
if (size != 0) {
|
|
|
|
uint32_t tag = ((uint32_t)field->number << 3) | kUpb_WireType_Delimited;
|
|
|
|
_upb_Decoder_AddUnknownVarints(d, msg, tag, (uint32_t)(ptr - start));
|
|
|
|
if (!_upb_Message_AddUnknown(msg, start, ptr - start, &d->arena)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (_upb_Map_Insert(map, &ent.k, map->key_size, &ent.v, map->val_size,
|
|
|
|
&d->arena) == _kUpb_MapInsertStatus_OutOfMemory) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* _upb_Decoder_DecodeToSubMessage(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Message* msg,
|
|
|
|
const upb_MiniTable_Sub* subs, const upb_MiniTable_Field* field,
|
|
|
|
wireval* val, int op) {
|
|
|
|
void* mem = UPB_PTR_AT(msg, field->offset, void);
|
|
|
|
int type = field->descriptortype;
|
|
|
|
|
|
|
|
if (UPB_UNLIKELY(op == OP_ENUM) &&
|
|
|
|
!_upb_Decoder_CheckEnum(d, ptr, msg, subs[field->submsg_index].subenum,
|
|
|
|
field, val)) {
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Set presence if necessary. */
|
Optimized decoder and paved the way for parsing extensions.
The primary motivation for this change is to avoid referring to the
`upb_msglayout` object when we are trying to fetch the `upb_msglayout`
object for a sub-message. This will help pave the way for parsing
extensions. We also implement several optimizations so that we can
make this change without regressing performance.
Normally we compute the layout for a sub-message field like so:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *layout,
const upb_msglayout_field *field) {
return layout->submsgs[field->submsg_index]
}
```
The reason for this indirection is to avoid storing a pointer directly
in `upb_msglayout_field`, as this would double its size (from 12 to 24
bytes on 64-bit architectures) which is wasteful as this pointer is
only needed for message typed fields.
However `get_submsg_layout` as written above does not work for
extensions, as they will not have entries in the message's
`layout->submsgs` array by nature, and we want to avoid creating
an entire fake `upb_msglayout` for each such extension since that
would also be wasteful.
This change removes the dependency on `upb_msglayout` by passing down
the `submsgs` array instead:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *const *submsgs,
const upb_msglayout_field *field) {
return submsgs[field->submsg_index]
}
```
This will pave the way for parsing extensions, as we can more easily
create an alternative `submsgs` array for extension fields without
extra overhead or waste.
Along the way several optimizations presented themselves that allow
a nice increase in performance:
1. Passing the parsed `wireval` by address instead of by value ended
up avoiding an expensive and useless stack copy (this is on Clang,
which was used for all measurements).
2. When field numbers are densely packed, we can find a field by number
with a single indexed lookup instead of linear search. At codegen
time we can compute the maximum field number that will allow such
an indexed lookup.
3. For fields that do require linear search, we can start the linear
search at the location where we found the previous field, taking
advantage of the fact that field numbers are generally increasing.
4. When the hasbit index is less than 32 (the common case) we can use
a less expensive code sequence to set it.
5. We check for the hasbit case before the oneof case, as optional
fields are more common than oneof fields.
Benchmark results indicate a 20% improvement in parse speed with a
small code size increase:
```
name old time/op new time/op delta
ArenaOneAlloc 21.3ns ± 0% 21.5ns ± 0% +0.96% (p=0.000 n=12+12)
ArenaInitialBlockOneAlloc 6.32ns ± 0% 6.32ns ± 0% +0.03% (p=0.000 n=12+10)
LoadDescriptor_Upb 53.5µs ± 1% 51.5µs ± 2% -3.70% (p=0.000 n=12+12)
LoadAdsDescriptor_Upb 2.78ms ± 2% 2.68ms ± 0% -3.57% (p=0.000 n=12+12)
LoadDescriptor_Proto2 240µs ± 0% 240µs ± 0% +0.12% (p=0.001 n=12+12)
LoadAdsDescriptor_Proto2 12.8ms ± 0% 12.7ms ± 0% -1.15% (p=0.000 n=12+10)
Parse_Upb_FileDesc<UseArena,Copy> 13.2µs ± 2% 10.7µs ± 0% -18.49% (p=0.000 n=10+12)
Parse_Upb_FileDesc<UseArena,Alias> 11.3µs ± 0% 9.6µs ± 0% -15.11% (p=0.000 n=12+11)
Parse_Upb_FileDesc<InitBlock,Copy> 12.7µs ± 0% 10.3µs ± 0% -19.00% (p=0.000 n=10+12)
Parse_Upb_FileDesc<InitBlock,Alias> 10.9µs ± 0% 9.2µs ± 0% -15.82% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,NoArena,Copy> 29.4µs ± 0% 29.5µs ± 0% +0.61% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,UseArena,Copy> 20.7µs ± 2% 20.6µs ± 2% ~ (p=0.260 n=12+11)
Parse_Proto2<FileDesc,InitBlock,Copy> 16.7µs ± 1% 16.7µs ± 0% -0.25% (p=0.036 n=12+10)
Parse_Proto2<FileDescSV,InitBlock,Alias> 16.5µs ± 0% 16.5µs ± 0% +0.20% (p=0.016 n=12+11)
SerializeDescriptor_Proto2 5.30µs ± 1% 5.36µs ± 1% +1.09% (p=0.000 n=12+11)
SerializeDescriptor_Upb 12.9µs ± 0% 13.0µs ± 0% +0.90% (p=0.000 n=12+11)
FILE SIZE VM SIZE
-------------- --------------
+1.5% +176 +1.6% +176 upb/decode.c
+1.8% +176 +1.9% +176 decode_msg
+0.4% +64 +0.4% +64 upb/def.c
+1.4% +64 +1.4% +64 _upb_symtab_addfile
+1.2% +48 +1.4% +48 upb/reflection.c
+15% +32 +18% +32 upb_msg_set
+2.9% +16 +3.1% +16 upb_msg_mutable
-9.3% -288 [ = ] 0 [Unmapped]
[ = ] 0 +0.2% +288 TOTAL
```
4 years ago
|
|
|
if (field->presence > 0) {
|
|
|
|
_upb_sethas_field(msg, field);
|
|
|
|
} else if (field->presence < 0) {
|
|
|
|
/* Oneof case */
|
|
|
|
uint32_t* oneof_case = _upb_oneofcase_field(msg, field);
|
|
|
|
if (op == OP_SUBMSG && *oneof_case != field->number) {
|
|
|
|
memset(mem, 0, sizeof(void*));
|
|
|
|
}
|
|
|
|
*oneof_case = field->number;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Store into message. */
|
|
|
|
switch (op) {
|
|
|
|
case OP_SUBMSG: {
|
|
|
|
upb_Message** submsgp = mem;
|
|
|
|
upb_Message* submsg = *submsgp;
|
|
|
|
if (!submsg) {
|
|
|
|
submsg = _upb_Decoder_NewSubMessage(d, subs, field);
|
|
|
|
*submsgp = submsg;
|
|
|
|
}
|
|
|
|
if (UPB_UNLIKELY(type == kUpb_FieldType_Group)) {
|
|
|
|
ptr = _upb_Decoder_DecodeKnownGroup(d, ptr, submsg, subs, field);
|
|
|
|
} else {
|
|
|
|
ptr = _upb_Decoder_DecodeSubMessage(d, ptr, submsg, subs, field,
|
|
|
|
val->size);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case OP_STRING:
|
|
|
|
_upb_Decoder_VerifyUtf8(d, ptr, val->size);
|
|
|
|
/* Fallthrough. */
|
|
|
|
case OP_BYTES:
|
|
|
|
return _upb_Decoder_ReadString(d, ptr, val->size, mem);
|
|
|
|
case OP_SCALAR_LG2(3):
|
Optimized decoder and paved the way for parsing extensions.
The primary motivation for this change is to avoid referring to the
`upb_msglayout` object when we are trying to fetch the `upb_msglayout`
object for a sub-message. This will help pave the way for parsing
extensions. We also implement several optimizations so that we can
make this change without regressing performance.
Normally we compute the layout for a sub-message field like so:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *layout,
const upb_msglayout_field *field) {
return layout->submsgs[field->submsg_index]
}
```
The reason for this indirection is to avoid storing a pointer directly
in `upb_msglayout_field`, as this would double its size (from 12 to 24
bytes on 64-bit architectures) which is wasteful as this pointer is
only needed for message typed fields.
However `get_submsg_layout` as written above does not work for
extensions, as they will not have entries in the message's
`layout->submsgs` array by nature, and we want to avoid creating
an entire fake `upb_msglayout` for each such extension since that
would also be wasteful.
This change removes the dependency on `upb_msglayout` by passing down
the `submsgs` array instead:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *const *submsgs,
const upb_msglayout_field *field) {
return submsgs[field->submsg_index]
}
```
This will pave the way for parsing extensions, as we can more easily
create an alternative `submsgs` array for extension fields without
extra overhead or waste.
Along the way several optimizations presented themselves that allow
a nice increase in performance:
1. Passing the parsed `wireval` by address instead of by value ended
up avoiding an expensive and useless stack copy (this is on Clang,
which was used for all measurements).
2. When field numbers are densely packed, we can find a field by number
with a single indexed lookup instead of linear search. At codegen
time we can compute the maximum field number that will allow such
an indexed lookup.
3. For fields that do require linear search, we can start the linear
search at the location where we found the previous field, taking
advantage of the fact that field numbers are generally increasing.
4. When the hasbit index is less than 32 (the common case) we can use
a less expensive code sequence to set it.
5. We check for the hasbit case before the oneof case, as optional
fields are more common than oneof fields.
Benchmark results indicate a 20% improvement in parse speed with a
small code size increase:
```
name old time/op new time/op delta
ArenaOneAlloc 21.3ns ± 0% 21.5ns ± 0% +0.96% (p=0.000 n=12+12)
ArenaInitialBlockOneAlloc 6.32ns ± 0% 6.32ns ± 0% +0.03% (p=0.000 n=12+10)
LoadDescriptor_Upb 53.5µs ± 1% 51.5µs ± 2% -3.70% (p=0.000 n=12+12)
LoadAdsDescriptor_Upb 2.78ms ± 2% 2.68ms ± 0% -3.57% (p=0.000 n=12+12)
LoadDescriptor_Proto2 240µs ± 0% 240µs ± 0% +0.12% (p=0.001 n=12+12)
LoadAdsDescriptor_Proto2 12.8ms ± 0% 12.7ms ± 0% -1.15% (p=0.000 n=12+10)
Parse_Upb_FileDesc<UseArena,Copy> 13.2µs ± 2% 10.7µs ± 0% -18.49% (p=0.000 n=10+12)
Parse_Upb_FileDesc<UseArena,Alias> 11.3µs ± 0% 9.6µs ± 0% -15.11% (p=0.000 n=12+11)
Parse_Upb_FileDesc<InitBlock,Copy> 12.7µs ± 0% 10.3µs ± 0% -19.00% (p=0.000 n=10+12)
Parse_Upb_FileDesc<InitBlock,Alias> 10.9µs ± 0% 9.2µs ± 0% -15.82% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,NoArena,Copy> 29.4µs ± 0% 29.5µs ± 0% +0.61% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,UseArena,Copy> 20.7µs ± 2% 20.6µs ± 2% ~ (p=0.260 n=12+11)
Parse_Proto2<FileDesc,InitBlock,Copy> 16.7µs ± 1% 16.7µs ± 0% -0.25% (p=0.036 n=12+10)
Parse_Proto2<FileDescSV,InitBlock,Alias> 16.5µs ± 0% 16.5µs ± 0% +0.20% (p=0.016 n=12+11)
SerializeDescriptor_Proto2 5.30µs ± 1% 5.36µs ± 1% +1.09% (p=0.000 n=12+11)
SerializeDescriptor_Upb 12.9µs ± 0% 13.0µs ± 0% +0.90% (p=0.000 n=12+11)
FILE SIZE VM SIZE
-------------- --------------
+1.5% +176 +1.6% +176 upb/decode.c
+1.8% +176 +1.9% +176 decode_msg
+0.4% +64 +0.4% +64 upb/def.c
+1.4% +64 +1.4% +64 _upb_symtab_addfile
+1.2% +48 +1.4% +48 upb/reflection.c
+15% +32 +18% +32 upb_msg_set
+2.9% +16 +3.1% +16 upb_msg_mutable
-9.3% -288 [ = ] 0 [Unmapped]
[ = ] 0 +0.2% +288 TOTAL
```
4 years ago
|
|
|
memcpy(mem, val, 8);
|
|
|
|
break;
|
|
|
|
case OP_ENUM:
|
|
|
|
case OP_SCALAR_LG2(2):
|
Optimized decoder and paved the way for parsing extensions.
The primary motivation for this change is to avoid referring to the
`upb_msglayout` object when we are trying to fetch the `upb_msglayout`
object for a sub-message. This will help pave the way for parsing
extensions. We also implement several optimizations so that we can
make this change without regressing performance.
Normally we compute the layout for a sub-message field like so:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *layout,
const upb_msglayout_field *field) {
return layout->submsgs[field->submsg_index]
}
```
The reason for this indirection is to avoid storing a pointer directly
in `upb_msglayout_field`, as this would double its size (from 12 to 24
bytes on 64-bit architectures) which is wasteful as this pointer is
only needed for message typed fields.
However `get_submsg_layout` as written above does not work for
extensions, as they will not have entries in the message's
`layout->submsgs` array by nature, and we want to avoid creating
an entire fake `upb_msglayout` for each such extension since that
would also be wasteful.
This change removes the dependency on `upb_msglayout` by passing down
the `submsgs` array instead:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *const *submsgs,
const upb_msglayout_field *field) {
return submsgs[field->submsg_index]
}
```
This will pave the way for parsing extensions, as we can more easily
create an alternative `submsgs` array for extension fields without
extra overhead or waste.
Along the way several optimizations presented themselves that allow
a nice increase in performance:
1. Passing the parsed `wireval` by address instead of by value ended
up avoiding an expensive and useless stack copy (this is on Clang,
which was used for all measurements).
2. When field numbers are densely packed, we can find a field by number
with a single indexed lookup instead of linear search. At codegen
time we can compute the maximum field number that will allow such
an indexed lookup.
3. For fields that do require linear search, we can start the linear
search at the location where we found the previous field, taking
advantage of the fact that field numbers are generally increasing.
4. When the hasbit index is less than 32 (the common case) we can use
a less expensive code sequence to set it.
5. We check for the hasbit case before the oneof case, as optional
fields are more common than oneof fields.
Benchmark results indicate a 20% improvement in parse speed with a
small code size increase:
```
name old time/op new time/op delta
ArenaOneAlloc 21.3ns ± 0% 21.5ns ± 0% +0.96% (p=0.000 n=12+12)
ArenaInitialBlockOneAlloc 6.32ns ± 0% 6.32ns ± 0% +0.03% (p=0.000 n=12+10)
LoadDescriptor_Upb 53.5µs ± 1% 51.5µs ± 2% -3.70% (p=0.000 n=12+12)
LoadAdsDescriptor_Upb 2.78ms ± 2% 2.68ms ± 0% -3.57% (p=0.000 n=12+12)
LoadDescriptor_Proto2 240µs ± 0% 240µs ± 0% +0.12% (p=0.001 n=12+12)
LoadAdsDescriptor_Proto2 12.8ms ± 0% 12.7ms ± 0% -1.15% (p=0.000 n=12+10)
Parse_Upb_FileDesc<UseArena,Copy> 13.2µs ± 2% 10.7µs ± 0% -18.49% (p=0.000 n=10+12)
Parse_Upb_FileDesc<UseArena,Alias> 11.3µs ± 0% 9.6µs ± 0% -15.11% (p=0.000 n=12+11)
Parse_Upb_FileDesc<InitBlock,Copy> 12.7µs ± 0% 10.3µs ± 0% -19.00% (p=0.000 n=10+12)
Parse_Upb_FileDesc<InitBlock,Alias> 10.9µs ± 0% 9.2µs ± 0% -15.82% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,NoArena,Copy> 29.4µs ± 0% 29.5µs ± 0% +0.61% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,UseArena,Copy> 20.7µs ± 2% 20.6µs ± 2% ~ (p=0.260 n=12+11)
Parse_Proto2<FileDesc,InitBlock,Copy> 16.7µs ± 1% 16.7µs ± 0% -0.25% (p=0.036 n=12+10)
Parse_Proto2<FileDescSV,InitBlock,Alias> 16.5µs ± 0% 16.5µs ± 0% +0.20% (p=0.016 n=12+11)
SerializeDescriptor_Proto2 5.30µs ± 1% 5.36µs ± 1% +1.09% (p=0.000 n=12+11)
SerializeDescriptor_Upb 12.9µs ± 0% 13.0µs ± 0% +0.90% (p=0.000 n=12+11)
FILE SIZE VM SIZE
-------------- --------------
+1.5% +176 +1.6% +176 upb/decode.c
+1.8% +176 +1.9% +176 decode_msg
+0.4% +64 +0.4% +64 upb/def.c
+1.4% +64 +1.4% +64 _upb_symtab_addfile
+1.2% +48 +1.4% +48 upb/reflection.c
+15% +32 +18% +32 upb_msg_set
+2.9% +16 +3.1% +16 upb_msg_mutable
-9.3% -288 [ = ] 0 [Unmapped]
[ = ] 0 +0.2% +288 TOTAL
```
4 years ago
|
|
|
memcpy(mem, val, 4);
|
|
|
|
break;
|
|
|
|
case OP_SCALAR_LG2(0):
|
Optimized decoder and paved the way for parsing extensions.
The primary motivation for this change is to avoid referring to the
`upb_msglayout` object when we are trying to fetch the `upb_msglayout`
object for a sub-message. This will help pave the way for parsing
extensions. We also implement several optimizations so that we can
make this change without regressing performance.
Normally we compute the layout for a sub-message field like so:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *layout,
const upb_msglayout_field *field) {
return layout->submsgs[field->submsg_index]
}
```
The reason for this indirection is to avoid storing a pointer directly
in `upb_msglayout_field`, as this would double its size (from 12 to 24
bytes on 64-bit architectures) which is wasteful as this pointer is
only needed for message typed fields.
However `get_submsg_layout` as written above does not work for
extensions, as they will not have entries in the message's
`layout->submsgs` array by nature, and we want to avoid creating
an entire fake `upb_msglayout` for each such extension since that
would also be wasteful.
This change removes the dependency on `upb_msglayout` by passing down
the `submsgs` array instead:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *const *submsgs,
const upb_msglayout_field *field) {
return submsgs[field->submsg_index]
}
```
This will pave the way for parsing extensions, as we can more easily
create an alternative `submsgs` array for extension fields without
extra overhead or waste.
Along the way several optimizations presented themselves that allow
a nice increase in performance:
1. Passing the parsed `wireval` by address instead of by value ended
up avoiding an expensive and useless stack copy (this is on Clang,
which was used for all measurements).
2. When field numbers are densely packed, we can find a field by number
with a single indexed lookup instead of linear search. At codegen
time we can compute the maximum field number that will allow such
an indexed lookup.
3. For fields that do require linear search, we can start the linear
search at the location where we found the previous field, taking
advantage of the fact that field numbers are generally increasing.
4. When the hasbit index is less than 32 (the common case) we can use
a less expensive code sequence to set it.
5. We check for the hasbit case before the oneof case, as optional
fields are more common than oneof fields.
Benchmark results indicate a 20% improvement in parse speed with a
small code size increase:
```
name old time/op new time/op delta
ArenaOneAlloc 21.3ns ± 0% 21.5ns ± 0% +0.96% (p=0.000 n=12+12)
ArenaInitialBlockOneAlloc 6.32ns ± 0% 6.32ns ± 0% +0.03% (p=0.000 n=12+10)
LoadDescriptor_Upb 53.5µs ± 1% 51.5µs ± 2% -3.70% (p=0.000 n=12+12)
LoadAdsDescriptor_Upb 2.78ms ± 2% 2.68ms ± 0% -3.57% (p=0.000 n=12+12)
LoadDescriptor_Proto2 240µs ± 0% 240µs ± 0% +0.12% (p=0.001 n=12+12)
LoadAdsDescriptor_Proto2 12.8ms ± 0% 12.7ms ± 0% -1.15% (p=0.000 n=12+10)
Parse_Upb_FileDesc<UseArena,Copy> 13.2µs ± 2% 10.7µs ± 0% -18.49% (p=0.000 n=10+12)
Parse_Upb_FileDesc<UseArena,Alias> 11.3µs ± 0% 9.6µs ± 0% -15.11% (p=0.000 n=12+11)
Parse_Upb_FileDesc<InitBlock,Copy> 12.7µs ± 0% 10.3µs ± 0% -19.00% (p=0.000 n=10+12)
Parse_Upb_FileDesc<InitBlock,Alias> 10.9µs ± 0% 9.2µs ± 0% -15.82% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,NoArena,Copy> 29.4µs ± 0% 29.5µs ± 0% +0.61% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,UseArena,Copy> 20.7µs ± 2% 20.6µs ± 2% ~ (p=0.260 n=12+11)
Parse_Proto2<FileDesc,InitBlock,Copy> 16.7µs ± 1% 16.7µs ± 0% -0.25% (p=0.036 n=12+10)
Parse_Proto2<FileDescSV,InitBlock,Alias> 16.5µs ± 0% 16.5µs ± 0% +0.20% (p=0.016 n=12+11)
SerializeDescriptor_Proto2 5.30µs ± 1% 5.36µs ± 1% +1.09% (p=0.000 n=12+11)
SerializeDescriptor_Upb 12.9µs ± 0% 13.0µs ± 0% +0.90% (p=0.000 n=12+11)
FILE SIZE VM SIZE
-------------- --------------
+1.5% +176 +1.6% +176 upb/decode.c
+1.8% +176 +1.9% +176 decode_msg
+0.4% +64 +0.4% +64 upb/def.c
+1.4% +64 +1.4% +64 _upb_symtab_addfile
+1.2% +48 +1.4% +48 upb/reflection.c
+15% +32 +18% +32 upb_msg_set
+2.9% +16 +3.1% +16 upb_msg_mutable
-9.3% -288 [ = ] 0 [Unmapped]
[ = ] 0 +0.2% +288 TOTAL
```
4 years ago
|
|
|
memcpy(mem, val, 1);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
UPB_UNREACHABLE();
|
|
|
|
}
|
|
|
|
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_NOINLINE
|
|
|
|
const char* _upb_Decoder_CheckRequired(upb_Decoder* d, const char* ptr,
|
|
|
|
const upb_Message* msg,
|
|
|
|
const upb_MiniTable* l) {
|
|
|
|
assert(l->required_count);
|
|
|
|
if (UPB_LIKELY((d->options & kUpb_DecodeOption_CheckRequired) == 0)) {
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
uint64_t msg_head;
|
|
|
|
memcpy(&msg_head, msg, 8);
|
|
|
|
msg_head = _upb_BigEndian_Swap64(msg_head);
|
|
|
|
if (upb_MiniTable_requiredmask(l) & ~msg_head) {
|
|
|
|
d->missing_required = true;
|
|
|
|
}
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static bool _upb_Decoder_TryFastDispatch(upb_Decoder* d, const char** ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
const upb_MiniTable* layout) {
|
|
|
|
#if UPB_FASTTABLE
|
|
|
|
if (layout && layout->table_mask != (unsigned char)-1) {
|
|
|
|
uint16_t tag = _upb_FastDecoder_LoadTag(*ptr);
|
|
|
|
intptr_t table = decode_totable(layout);
|
|
|
|
*ptr = _upb_FastDecoder_TagDispatch(d, *ptr, msg, table, 0, tag);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* upb_Decoder_SkipField(upb_Decoder* d, const char* ptr,
|
|
|
|
uint32_t tag) {
|
|
|
|
int field_number = tag >> 3;
|
|
|
|
int wire_type = tag & 7;
|
|
|
|
switch (wire_type) {
|
|
|
|
case kUpb_WireType_Varint: {
|
|
|
|
uint64_t val;
|
|
|
|
return _upb_Decoder_DecodeVarint(d, ptr, &val);
|
|
|
|
}
|
|
|
|
case kUpb_WireType_64Bit:
|
|
|
|
return ptr + 8;
|
|
|
|
case kUpb_WireType_32Bit:
|
|
|
|
return ptr + 4;
|
|
|
|
case kUpb_WireType_Delimited: {
|
|
|
|
uint32_t size;
|
|
|
|
ptr = upb_Decoder_DecodeSize(d, ptr, &size);
|
|
|
|
return ptr + size;
|
|
|
|
}
|
|
|
|
case kUpb_WireType_StartGroup:
|
|
|
|
return _upb_Decoder_DecodeUnknownGroup(d, ptr, field_number);
|
|
|
|
default:
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
enum {
|
|
|
|
kStartItemTag = ((1 << 3) | kUpb_WireType_StartGroup),
|
|
|
|
kEndItemTag = ((1 << 3) | kUpb_WireType_EndGroup),
|
|
|
|
kTypeIdTag = ((2 << 3) | kUpb_WireType_Varint),
|
|
|
|
kMessageTag = ((3 << 3) | kUpb_WireType_Delimited),
|
|
|
|
};
|
|
|
|
|
|
|
|
static void upb_Decoder_AddKnownMessageSetItem(
|
|
|
|
upb_Decoder* d, upb_Message* msg, const upb_MiniTable_Extension* item_mt,
|
|
|
|
const char* data, uint32_t size) {
|
|
|
|
upb_Message_Extension* ext =
|
|
|
|
_upb_Message_GetOrCreateExtension(msg, item_mt, &d->arena);
|
|
|
|
if (UPB_UNLIKELY(!ext)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
|
|
|
upb_Message* submsg =
|
|
|
|
_upb_Decoder_NewSubMessage(d, &ext->ext->sub, &ext->ext->field);
|
|
|
|
upb_DecodeStatus status = upb_Decode(data, size, submsg, item_mt->sub.submsg,
|
|
|
|
d->extreg, d->options, &d->arena);
|
|
|
|
memcpy(&ext->data, &submsg, sizeof(submsg));
|
|
|
|
if (status != kUpb_DecodeStatus_Ok) _upb_Decoder_ErrorJmp(d, status);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void upb_Decoder_AddUnknownMessageSetItem(upb_Decoder* d,
|
|
|
|
upb_Message* msg,
|
|
|
|
uint32_t type_id,
|
|
|
|
const char* message_data,
|
|
|
|
uint32_t message_size) {
|
|
|
|
char buf[60];
|
|
|
|
char* ptr = buf;
|
|
|
|
ptr = upb_Decoder_EncodeVarint32(kStartItemTag, ptr);
|
|
|
|
ptr = upb_Decoder_EncodeVarint32(kTypeIdTag, ptr);
|
|
|
|
ptr = upb_Decoder_EncodeVarint32(type_id, ptr);
|
|
|
|
ptr = upb_Decoder_EncodeVarint32(kMessageTag, ptr);
|
|
|
|
ptr = upb_Decoder_EncodeVarint32(message_size, ptr);
|
|
|
|
char* split = ptr;
|
|
|
|
|
|
|
|
ptr = upb_Decoder_EncodeVarint32(kEndItemTag, ptr);
|
|
|
|
char* end = ptr;
|
|
|
|
|
|
|
|
if (!_upb_Message_AddUnknown(msg, buf, split - buf, &d->arena) ||
|
|
|
|
!_upb_Message_AddUnknown(msg, message_data, message_size, &d->arena) ||
|
|
|
|
!_upb_Message_AddUnknown(msg, split, end - split, &d->arena)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void upb_Decoder_AddMessageSetItem(upb_Decoder* d, upb_Message* msg,
|
|
|
|
const upb_MiniTable* layout,
|
|
|
|
uint32_t type_id, const char* data,
|
|
|
|
uint32_t size) {
|
|
|
|
const upb_MiniTable_Extension* item_mt =
|
|
|
|
_upb_extreg_get(d->extreg, layout, type_id);
|
|
|
|
if (item_mt) {
|
|
|
|
upb_Decoder_AddKnownMessageSetItem(d, msg, item_mt, data, size);
|
|
|
|
} else {
|
|
|
|
upb_Decoder_AddUnknownMessageSetItem(d, msg, type_id, data, size);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* upb_Decoder_DecodeMessageSetItem(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Message* msg,
|
|
|
|
const upb_MiniTable* layout) {
|
|
|
|
uint32_t type_id = 0;
|
|
|
|
upb_StringView preserved = {NULL, 0};
|
|
|
|
typedef enum {
|
|
|
|
kUpb_HaveId = 1 << 0,
|
|
|
|
kUpb_HavePayload = 1 << 1,
|
|
|
|
} StateMask;
|
|
|
|
StateMask state_mask = 0;
|
|
|
|
while (!_upb_Decoder_IsDone(d, &ptr)) {
|
|
|
|
uint32_t tag;
|
|
|
|
ptr = _upb_Decoder_DecodeTag(d, ptr, &tag);
|
|
|
|
switch (tag) {
|
|
|
|
case kEndItemTag:
|
|
|
|
return ptr;
|
|
|
|
case kTypeIdTag: {
|
|
|
|
uint64_t tmp;
|
|
|
|
ptr = _upb_Decoder_DecodeVarint(d, ptr, &tmp);
|
|
|
|
if (state_mask & kUpb_HaveId) break; // Ignore dup.
|
|
|
|
state_mask |= kUpb_HaveId;
|
|
|
|
type_id = tmp;
|
|
|
|
if (state_mask & kUpb_HavePayload) {
|
|
|
|
upb_Decoder_AddMessageSetItem(d, msg, layout, type_id, preserved.data,
|
|
|
|
preserved.size);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case kMessageTag: {
|
|
|
|
uint32_t size;
|
|
|
|
ptr = upb_Decoder_DecodeSize(d, ptr, &size);
|
|
|
|
const char* data = ptr;
|
|
|
|
ptr += size;
|
|
|
|
if (state_mask & kUpb_HavePayload) break; // Ignore dup.
|
|
|
|
state_mask |= kUpb_HavePayload;
|
|
|
|
if (state_mask & kUpb_HaveId) {
|
|
|
|
upb_Decoder_AddMessageSetItem(d, msg, layout, type_id, data, size);
|
|
|
|
} else {
|
|
|
|
// Out of order, we must preserve the payload.
|
|
|
|
preserved.data = data;
|
|
|
|
preserved.size = size;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
// We do not preserve unexpected fields inside a message set item.
|
|
|
|
ptr = upb_Decoder_SkipField(d, ptr, tag);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
|
|
|
|
static const upb_MiniTable_Field* _upb_Decoder_FindField(
|
|
|
|
upb_Decoder* d, const upb_MiniTable* l, uint32_t field_number,
|
|
|
|
int* last_field_index) {
|
|
|
|
static upb_MiniTable_Field none = {0, 0, 0, 0, 0, 0};
|
|
|
|
if (l == NULL) return &none;
|
|
|
|
|
|
|
|
size_t idx = ((size_t)field_number) - 1; // 0 wraps to SIZE_MAX
|
|
|
|
if (idx < l->dense_below) {
|
|
|
|
/* Fastest case: index into dense fields. */
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (l->dense_below < l->field_count) {
|
|
|
|
/* Linear search non-dense fields. Resume scanning from last_field_index
|
|
|
|
* since fields are usually in order. */
|
|
|
|
int last = *last_field_index;
|
|
|
|
for (idx = last; idx < l->field_count; idx++) {
|
|
|
|
if (l->fields[idx].number == field_number) {
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (idx = l->dense_below; idx < last; idx++) {
|
|
|
|
if (l->fields[idx].number == field_number) {
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (d->extreg) {
|
|
|
|
switch (l->ext) {
|
|
|
|
case kUpb_ExtMode_Extendable: {
|
|
|
|
const upb_MiniTable_Extension* ext =
|
|
|
|
_upb_extreg_get(d->extreg, l, field_number);
|
|
|
|
if (ext) return &ext->field;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case kUpb_ExtMode_IsMessageSet:
|
|
|
|
if (field_number == _UPB_MSGSET_ITEM) {
|
|
|
|
static upb_MiniTable_Field item = {0, 0, 0, 0, TYPE_MSGSET_ITEM, 0};
|
|
|
|
return &item;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return &none; /* Unknown field. */
|
|
|
|
|
|
|
|
found:
|
|
|
|
UPB_ASSERT(l->fields[idx].number == field_number);
|
|
|
|
*last_field_index = idx;
|
|
|
|
return &l->fields[idx];
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeWireValue(
|
|
|
|
upb_Decoder* d, const char* ptr, const upb_MiniTable_Field* field,
|
|
|
|
int wire_type, wireval* val, int* op) {
|
|
|
|
switch (wire_type) {
|
|
|
|
case kUpb_WireType_Varint:
|
|
|
|
ptr = _upb_Decoder_DecodeVarint(d, ptr, &val->uint64_val);
|
|
|
|
*op = varint_ops[field->descriptortype];
|
|
|
|
_upb_Decoder_Munge(field->descriptortype, val);
|
|
|
|
return ptr;
|
|
|
|
case kUpb_WireType_32Bit:
|
|
|
|
memcpy(&val->uint32_val, ptr, 4);
|
|
|
|
val->uint32_val = _upb_BigEndian_Swap32(val->uint32_val);
|
|
|
|
*op = OP_SCALAR_LG2(2);
|
|
|
|
if (((1 << field->descriptortype) & FIXED32_OK_MASK) == 0) {
|
|
|
|
*op = OP_UNKNOWN;
|
|
|
|
}
|
|
|
|
return ptr + 4;
|
|
|
|
case kUpb_WireType_64Bit:
|
|
|
|
memcpy(&val->uint64_val, ptr, 8);
|
|
|
|
val->uint64_val = _upb_BigEndian_Swap64(val->uint64_val);
|
|
|
|
*op = OP_SCALAR_LG2(3);
|
|
|
|
if (((1 << field->descriptortype) & FIXED64_OK_MASK) == 0) {
|
|
|
|
*op = OP_UNKNOWN;
|
|
|
|
}
|
|
|
|
return ptr + 8;
|
|
|
|
case kUpb_WireType_Delimited: {
|
|
|
|
int ndx = field->descriptortype;
|
|
|
|
if (upb_FieldMode_Get(field) == kUpb_FieldMode_Array) ndx += TYPE_COUNT;
|
|
|
|
ptr = upb_Decoder_DecodeSize(d, ptr, &val->size);
|
|
|
|
*op = delim_ops[ndx];
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
case kUpb_WireType_StartGroup:
|
|
|
|
val->uint32_val = field->number;
|
|
|
|
if (field->descriptortype == kUpb_FieldType_Group) {
|
|
|
|
*op = OP_SUBMSG;
|
|
|
|
} else if (field->descriptortype == TYPE_MSGSET_ITEM) {
|
|
|
|
*op = OP_MSGSET_ITEM;
|
|
|
|
} else {
|
|
|
|
*op = OP_UNKNOWN;
|
|
|
|
}
|
|
|
|
return ptr;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_FORCEINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeKnownField(
|
|
|
|
upb_Decoder* d, const char* ptr, upb_Message* msg,
|
|
|
|
const upb_MiniTable* layout, const upb_MiniTable_Field* field, int op,
|
|
|
|
wireval* val) {
|
|
|
|
const upb_MiniTable_Sub* subs = layout->subs;
|
|
|
|
uint8_t mode = field->mode;
|
|
|
|
|
|
|
|
if (UPB_UNLIKELY(mode & kUpb_LabelFlags_IsExtension)) {
|
|
|
|
const upb_MiniTable_Extension* ext_layout =
|
|
|
|
(const upb_MiniTable_Extension*)field;
|
|
|
|
upb_Message_Extension* ext =
|
|
|
|
_upb_Message_GetOrCreateExtension(msg, ext_layout, &d->arena);
|
|
|
|
if (UPB_UNLIKELY(!ext)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
Fixed bug when parsing an unknown value in a proto2 enum extension. #fuzzing
Proto2 enum parsing is the only case where we have to look at the wire value (not merely the tag) to decide whether the field is known or unknown. If the value is unknown, we need to put the value in the Unknown Fields, but for an extension we no longer have easy access to the message, because for extensions we replace the `msg` pointer with a pointer to the extension. The bug occurred when we were treating the fake `upb_Message*` (which was actually a pointer to an extension) as a real `upb_Message*` that can have unknown fields.
This CL fixes the problem by preserving the true message pointer in `d->unknown_msg` when we are parsing an extension.
This also required fixing a bug in MiniTable building when fasttables are enabled. We need to set the table_mask to `-1` to disable fasttable parsing, not `0`.
For unknown reasons, this CL appears to speed up parsing somewhat significantly. Ideally we should be tracking parsing performance better over time, as it is possible this is merely regaining performance that was lost at a different time:
```
benchy --reference=srcfs third_party/upb/benchmarks:benchmark
10 / 10 [=================================================================================================================] 100.00% 2m32s
(Generated by http://go/benchy. Settings: --runs 5 --reference "srcfs")
name old cpu/op new cpu/op delta
BM_ArenaOneAlloc 23.9ns ± 6% 23.7ns ± 4% ~ (p=0.180 n=53+51)
BM_ArenaInitialBlockOneAlloc 7.62ns ± 4% 7.70ns ± 5% +0.99% (p=0.024 n=59+60)
BM_LoadAdsDescriptor_Upb<NoLayout> 6.60ms ±10% 6.57ms ± 8% ~ (p=0.607 n=47+54)
BM_LoadAdsDescriptor_Upb<WithLayout> 6.92ms ± 5% 6.88ms ± 8% ~ (p=0.257 n=54+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 14.2ms ± 8% 14.0ms ± 7% -1.38% (p=0.025 n=58+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 14.3ms ± 8% 14.2ms ± 8% -1.16% (p=0.031 n=58+57)
BM_Parse_Upb_FileDesc<UseArena, Copy> 15.9µs ± 4% 14.6µs ± 4% -7.85% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 14.5µs ± 4% 13.3µs ± 5% -8.50% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 15.7µs ± 4% 14.4µs ± 5% -7.99% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 14.2µs ± 5% 13.0µs ± 4% -8.56% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 26.3µs ± 4% 26.2µs ± 4% ~ (p=0.195 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 13.3µs ± 5% 13.2µs ± 4% ~ (p=0.085 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 12.9µs ± 4% 12.8µs ± 3% -0.66% (p=0.023 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 10.9µs ± 6% 10.9µs ± 4% ~ (p=0.063 n=59+58)
BM_SerializeDescriptor_Proto2 7.57µs ± 6% 7.62µs ± 6% ~ (p=0.147 n=57+58)
BM_SerializeDescriptor_Upb 12.8µs ± 4% 12.8µs ± 4% ~ (p=0.163 n=59+56)
name old time/op new time/op delta
BM_ArenaOneAlloc 23.9ns ± 5% 23.7ns ± 4% ~ (p=0.172 n=53+51)
BM_ArenaInitialBlockOneAlloc 7.62ns ± 4% 7.70ns ± 5% +1.02% (p=0.017 n=59+60)
BM_LoadAdsDescriptor_Upb<NoLayout> 6.60ms ±10% 6.58ms ± 8% ~ (p=0.727 n=47+55)
BM_LoadAdsDescriptor_Upb<WithLayout> 6.92ms ± 5% 6.88ms ± 8% ~ (p=0.260 n=54+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 14.2ms ± 7% 14.0ms ± 7% -1.40% (p=0.019 n=58+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 14.3ms ± 8% 14.2ms ± 8% -1.13% (p=0.037 n=58+57)
BM_Parse_Upb_FileDesc<UseArena, Copy> 15.9µs ± 4% 14.6µs ± 3% -7.88% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 14.5µs ± 4% 13.3µs ± 5% -8.46% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 15.7µs ± 4% 14.4µs ± 5% -7.99% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 14.2µs ± 5% 13.0µs ± 4% -8.56% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 26.3µs ± 4% 26.2µs ± 4% ~ (p=0.224 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 13.3µs ± 5% 13.2µs ± 4% ~ (p=0.098 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 12.9µs ± 4% 12.8µs ± 3% -0.68% (p=0.015 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 10.9µs ± 6% 10.9µs ± 4% ~ (p=0.052 n=59+58)
BM_SerializeDescriptor_Proto2 7.56µs ± 6% 7.62µs ± 6% ~ (p=0.111 n=58+58)
BM_SerializeDescriptor_Upb 12.8µs ± 4% 12.8µs ± 4% ~ (p=0.241 n=56+56)
name old allocs/op new allocs/op delta
BM_ArenaOneAlloc 1.00 ± 0% 1.00 ± 0% ~ (all samples are equal)
BM_ArenaInitialBlockOneAlloc 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<NoLayout> 5.98k ± 0% 5.98k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<WithLayout> 5.98k ± 0% 5.98k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<NoLayout> 80.9k ± 0% 80.9k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<WithLayout> 82.1k ± 0% 82.1k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Copy> 7.00 ± 0% 7.00 ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Alias> 7.00 ± 0% 7.00 ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 765 ± 0% 765 ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 9.00 ± 0% 9.00 ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Proto2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Upb 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
name old peak-mem(Bytes)/op new peak-mem(Bytes)/op delta
BM_ArenaOneAlloc 344 ± 0% 344 ± 0% ~ (all samples are equal)
BM_ArenaInitialBlockOneAlloc 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<NoLayout> 9.60M ± 0% 9.60M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<WithLayout> 9.68M ± 0% 9.68M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<NoLayout> 6.41M ± 0% 6.41M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<WithLayout> 6.44M ± 0% 6.44M ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Copy> 36.5k ± 0% 36.5k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Alias> 36.5k ± 0% 36.5k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 35.8k ± 0% 35.8k ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 40.7k ± 0% 40.7k ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Proto2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Upb 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
name old speed new speed delta
BM_LoadAdsDescriptor_Upb<NoLayout> 113MB/s ± 9% 113MB/s ± 8% ~ (p=0.712 n=47+55)
BM_LoadAdsDescriptor_Upb<WithLayout> 107MB/s ± 8% 108MB/s ± 8% ~ (p=0.200 n=55+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 52.5MB/s ± 8% 53.3MB/s ± 7% +1.51% (p=0.018 n=59+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 51.9MB/s ± 7% 52.4MB/s ± 8% +1.01% (p=0.050 n=58+58)
BM_Parse_Upb_FileDesc<UseArena, Copy> 473MB/s ± 4% 514MB/s ± 4% +8.52% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 518MB/s ± 4% 566MB/s ± 5% +9.30% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 480MB/s ± 4% 521MB/s ± 5% +8.69% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 528MB/s ± 4% 578MB/s ± 4% +9.36% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 286MB/s ± 4% 287MB/s ± 4% ~ (p=0.195 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 566MB/s ± 5% 570MB/s ± 4% ~ (p=0.085 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 583MB/s ± 5% 587MB/s ± 3% +0.64% (p=0.023 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 688MB/s ± 6% 693MB/s ± 4% ~ (p=0.063 n=59+58)
BM_SerializeDescriptor_Proto2 995MB/s ± 6% 988MB/s ± 5% ~ (p=0.147 n=57+58)
BM_SerializeDescriptor_Upb 586MB/s ± 4% 589MB/s ± 4% ~ (p=0.163 n=59+56)
```
PiperOrigin-RevId: 462022073
2 years ago
|
|
|
d->unknown_msg = msg;
|
|
|
|
msg = &ext->data;
|
|
|
|
subs = &ext->ext->sub;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (mode & kUpb_FieldMode_Mask) {
|
|
|
|
case kUpb_FieldMode_Array:
|
|
|
|
return _upb_Decoder_DecodeToArray(d, ptr, msg, subs, field, val, op);
|
|
|
|
case kUpb_FieldMode_Map:
|
|
|
|
return _upb_Decoder_DecodeToMap(d, ptr, msg, subs, field, val);
|
|
|
|
case kUpb_FieldMode_Scalar:
|
|
|
|
return _upb_Decoder_DecodeToSubMessage(d, ptr, msg, subs, field, val, op);
|
|
|
|
default:
|
|
|
|
UPB_UNREACHABLE();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* _upb_Decoder_ReverseSkipVarint(const char* ptr,
|
|
|
|
uint32_t val) {
|
|
|
|
uint32_t seen = 0;
|
|
|
|
do {
|
|
|
|
ptr--;
|
|
|
|
seen <<= 7;
|
|
|
|
seen |= *ptr & 0x7f;
|
|
|
|
} while (seen != val);
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const char* _upb_Decoder_DecodeUnknownField(upb_Decoder* d,
|
|
|
|
const char* ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
int field_number,
|
|
|
|
int wire_type, wireval val) {
|
|
|
|
if (field_number == 0) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed);
|
|
|
|
|
|
|
|
// Since unknown fields are the uncommon case, we do a little extra work here
|
|
|
|
// to walk backwards through the buffer to find the field start. This frees
|
|
|
|
// up a register in the fast paths (when the field is known), which leads to
|
|
|
|
// significant speedups in benchmarks.
|
|
|
|
const char* start = ptr;
|
|
|
|
|
|
|
|
if (wire_type == kUpb_WireType_Delimited) ptr += val.size;
|
|
|
|
if (msg) {
|
|
|
|
switch (wire_type) {
|
|
|
|
case kUpb_WireType_Varint:
|
|
|
|
case kUpb_WireType_Delimited:
|
|
|
|
start--;
|
|
|
|
while (start[-1] & 0x80) start--;
|
|
|
|
break;
|
|
|
|
case kUpb_WireType_32Bit:
|
|
|
|
start -= 4;
|
|
|
|
break;
|
|
|
|
case kUpb_WireType_64Bit:
|
|
|
|
start -= 8;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(start == d->debug_valstart);
|
|
|
|
uint32_t tag = ((uint32_t)field_number << 3) | wire_type;
|
|
|
|
start = _upb_Decoder_ReverseSkipVarint(start, tag);
|
|
|
|
assert(start == d->debug_tagstart);
|
|
|
|
|
|
|
|
if (wire_type == kUpb_WireType_StartGroup) {
|
|
|
|
d->unknown = start;
|
|
|
|
d->unknown_msg = msg;
|
|
|
|
ptr = _upb_Decoder_DecodeUnknownGroup(d, ptr, field_number);
|
|
|
|
start = d->unknown;
|
|
|
|
d->unknown = NULL;
|
|
|
|
}
|
|
|
|
if (!_upb_Message_AddUnknown(msg, start, ptr - start, &d->arena)) {
|
|
|
|
_upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory);
|
|
|
|
}
|
|
|
|
} else if (wire_type == kUpb_WireType_StartGroup) {
|
|
|
|
ptr = _upb_Decoder_DecodeUnknownGroup(d, ptr, field_number);
|
|
|
|
}
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
UPB_NOINLINE
|
|
|
|
static const char* _upb_Decoder_DecodeMessage(upb_Decoder* d, const char* ptr,
|
|
|
|
upb_Message* msg,
|
|
|
|
const upb_MiniTable* layout) {
|
Optimized decoder and paved the way for parsing extensions.
The primary motivation for this change is to avoid referring to the
`upb_msglayout` object when we are trying to fetch the `upb_msglayout`
object for a sub-message. This will help pave the way for parsing
extensions. We also implement several optimizations so that we can
make this change without regressing performance.
Normally we compute the layout for a sub-message field like so:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *layout,
const upb_msglayout_field *field) {
return layout->submsgs[field->submsg_index]
}
```
The reason for this indirection is to avoid storing a pointer directly
in `upb_msglayout_field`, as this would double its size (from 12 to 24
bytes on 64-bit architectures) which is wasteful as this pointer is
only needed for message typed fields.
However `get_submsg_layout` as written above does not work for
extensions, as they will not have entries in the message's
`layout->submsgs` array by nature, and we want to avoid creating
an entire fake `upb_msglayout` for each such extension since that
would also be wasteful.
This change removes the dependency on `upb_msglayout` by passing down
the `submsgs` array instead:
```
const upb_msglayout *get_submsg_layout(
const upb_msglayout *const *submsgs,
const upb_msglayout_field *field) {
return submsgs[field->submsg_index]
}
```
This will pave the way for parsing extensions, as we can more easily
create an alternative `submsgs` array for extension fields without
extra overhead or waste.
Along the way several optimizations presented themselves that allow
a nice increase in performance:
1. Passing the parsed `wireval` by address instead of by value ended
up avoiding an expensive and useless stack copy (this is on Clang,
which was used for all measurements).
2. When field numbers are densely packed, we can find a field by number
with a single indexed lookup instead of linear search. At codegen
time we can compute the maximum field number that will allow such
an indexed lookup.
3. For fields that do require linear search, we can start the linear
search at the location where we found the previous field, taking
advantage of the fact that field numbers are generally increasing.
4. When the hasbit index is less than 32 (the common case) we can use
a less expensive code sequence to set it.
5. We check for the hasbit case before the oneof case, as optional
fields are more common than oneof fields.
Benchmark results indicate a 20% improvement in parse speed with a
small code size increase:
```
name old time/op new time/op delta
ArenaOneAlloc 21.3ns ± 0% 21.5ns ± 0% +0.96% (p=0.000 n=12+12)
ArenaInitialBlockOneAlloc 6.32ns ± 0% 6.32ns ± 0% +0.03% (p=0.000 n=12+10)
LoadDescriptor_Upb 53.5µs ± 1% 51.5µs ± 2% -3.70% (p=0.000 n=12+12)
LoadAdsDescriptor_Upb 2.78ms ± 2% 2.68ms ± 0% -3.57% (p=0.000 n=12+12)
LoadDescriptor_Proto2 240µs ± 0% 240µs ± 0% +0.12% (p=0.001 n=12+12)
LoadAdsDescriptor_Proto2 12.8ms ± 0% 12.7ms ± 0% -1.15% (p=0.000 n=12+10)
Parse_Upb_FileDesc<UseArena,Copy> 13.2µs ± 2% 10.7µs ± 0% -18.49% (p=0.000 n=10+12)
Parse_Upb_FileDesc<UseArena,Alias> 11.3µs ± 0% 9.6µs ± 0% -15.11% (p=0.000 n=12+11)
Parse_Upb_FileDesc<InitBlock,Copy> 12.7µs ± 0% 10.3µs ± 0% -19.00% (p=0.000 n=10+12)
Parse_Upb_FileDesc<InitBlock,Alias> 10.9µs ± 0% 9.2µs ± 0% -15.82% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,NoArena,Copy> 29.4µs ± 0% 29.5µs ± 0% +0.61% (p=0.000 n=12+12)
Parse_Proto2<FileDesc,UseArena,Copy> 20.7µs ± 2% 20.6µs ± 2% ~ (p=0.260 n=12+11)
Parse_Proto2<FileDesc,InitBlock,Copy> 16.7µs ± 1% 16.7µs ± 0% -0.25% (p=0.036 n=12+10)
Parse_Proto2<FileDescSV,InitBlock,Alias> 16.5µs ± 0% 16.5µs ± 0% +0.20% (p=0.016 n=12+11)
SerializeDescriptor_Proto2 5.30µs ± 1% 5.36µs ± 1% +1.09% (p=0.000 n=12+11)
SerializeDescriptor_Upb 12.9µs ± 0% 13.0µs ± 0% +0.90% (p=0.000 n=12+11)
FILE SIZE VM SIZE
-------------- --------------
+1.5% +176 +1.6% +176 upb/decode.c
+1.8% +176 +1.9% +176 decode_msg
+0.4% +64 +0.4% +64 upb/def.c
+1.4% +64 +1.4% +64 _upb_symtab_addfile
+1.2% +48 +1.4% +48 upb/reflection.c
+15% +32 +18% +32 upb_msg_set
+2.9% +16 +3.1% +16 upb_msg_mutable
-9.3% -288 [ = ] 0 [Unmapped]
[ = ] 0 +0.2% +288 TOTAL
```
4 years ago
|
|
|
int last_field_index = 0;
|
|
|
|
|
|
|
|
#if UPB_FASTTABLE
|
|
|
|
// The first time we want to skip fast dispatch, because we may have just been
|
|
|
|
// invoked by the fast parser to handle a case that it bailed on.
|
|
|
|
if (!_upb_Decoder_IsDone(d, &ptr)) goto nofast;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
while (!_upb_Decoder_IsDone(d, &ptr)) {
|
|
|
|
uint32_t tag;
|
|
|
|
const upb_MiniTable_Field* field;
|
|
|
|
int field_number;
|
|
|
|
int wire_type;
|
|
|
|
wireval val;
|
|
|
|
int op;
|
|
|
|
|
|
|
|
if (_upb_Decoder_TryFastDispatch(d, &ptr, msg, layout)) break;
|
|
|
|
|
|
|
|
#if UPB_FASTTABLE
|
|
|
|
nofast:
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifndef NDEBUG
|
|
|
|
d->debug_tagstart = ptr;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
UPB_ASSERT(ptr < d->limit_ptr);
|
|
|
|
ptr = _upb_Decoder_DecodeTag(d, ptr, &tag);
|
|
|
|
field_number = tag >> 3;
|
|
|
|
wire_type = tag & 7;
|
|
|
|
|
|
|
|
#ifndef NDEBUG
|
|
|
|
d->debug_valstart = ptr;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
if (wire_type == kUpb_WireType_EndGroup) {
|
|
|
|
d->end_group = field_number;
|
|
|
|
return ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
field = _upb_Decoder_FindField(d, layout, field_number, &last_field_index);
|
|
|
|
ptr = _upb_Decoder_DecodeWireValue(d, ptr, field, wire_type, &val, &op);
|
|
|
|
|
|
|
|
if (op >= 0) {
|
|
|
|
ptr = _upb_Decoder_DecodeKnownField(d, ptr, msg, layout, field, op, &val);
|
|
|
|
} else {
|
|
|
|
switch (op) {
|
|
|
|
case OP_UNKNOWN:
|
|
|
|
ptr = _upb_Decoder_DecodeUnknownField(d, ptr, msg, field_number,
|
|
|
|
wire_type, val);
|
|
|
|
break;
|
|
|
|
case OP_MSGSET_ITEM:
|
|
|
|
ptr = upb_Decoder_DecodeMessageSetItem(d, ptr, msg, layout);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return UPB_UNLIKELY(layout && layout->required_count)
|
|
|
|
? _upb_Decoder_CheckRequired(d, ptr, msg, layout)
|
|
|
|
: ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
const char* _upb_FastDecoder_DecodeGeneric(struct upb_Decoder* d,
|
|
|
|
const char* ptr, upb_Message* msg,
|
|
|
|
intptr_t table, uint64_t hasbits,
|
|
|
|
uint64_t data) {
|
|
|
|
(void)data;
|
|
|
|
*(uint32_t*)msg |= hasbits;
|
|
|
|
return _upb_Decoder_DecodeMessage(d, ptr, msg, decode_totablep(table));
|
|
|
|
}
|
|
|
|
|
|
|
|
static upb_DecodeStatus _upb_Decoder_DecodeTop(struct upb_Decoder* d,
|
|
|
|
const char* buf, void* msg,
|
|
|
|
const upb_MiniTable* l) {
|
|
|
|
if (!_upb_Decoder_TryFastDispatch(d, &buf, msg, l)) {
|
|
|
|
_upb_Decoder_DecodeMessage(d, buf, msg, l);
|
|
|
|
}
|
|
|
|
if (d->end_group != DECODE_NOGROUP) return kUpb_DecodeStatus_Malformed;
|
|
|
|
if (d->missing_required) return kUpb_DecodeStatus_MissingRequired;
|
|
|
|
return kUpb_DecodeStatus_Ok;
|
|
|
|
}
|
|
|
|
|
|
|
|
upb_DecodeStatus upb_Decode(const char* buf, size_t size, void* msg,
|
|
|
|
const upb_MiniTable* l,
|
|
|
|
const upb_ExtensionRegistry* extreg, int options,
|
|
|
|
upb_Arena* arena) {
|
|
|
|
upb_Decoder state;
|
Added API for copy vs. alias and added benchmarks to test both.
Benchmark output:
$ bazel-bin/benchmarks/benchmark '--benchmark_filter=BM_Parse'
2020-11-11 15:39:04
Running bazel-bin/benchmarks/benchmark
Run on (72 X 3700 MHz CPU s)
CPU Caches:
L1 Data 32K (x36)
L1 Instruction 32K (x36)
L2 Unified 1024K (x36)
L3 Unified 25344K (x2)
-------------------------------------------------------------------------------------
Benchmark Time CPU Iterations
-------------------------------------------------------------------------------------
BM_Parse_Upb_FileDesc<UseArena, Copy> 4134 ns 4134 ns 168714 1.69152GB/s
BM_Parse_Upb_FileDesc<UseArena, Alias> 3487 ns 3487 ns 199509 2.00526GB/s
BM_Parse_Upb_FileDesc<InitBlock, Copy> 3727 ns 3726 ns 187581 1.87643GB/s
BM_Parse_Upb_FileDesc<InitBlock, Alias> 3110 ns 3110 ns 224970 2.24866GB/s
BM_Parse_Proto2<FileDesc, NoArena, Copy> 31132 ns 31132 ns 22437 229.995MB/s
BM_Parse_Proto2<FileDesc, UseArena, Copy> 21011 ns 21009 ns 33922 340.812MB/s
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 17976 ns 17975 ns 38808 398.337MB/s
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 17357 ns 17356 ns 40244 412.539MB/s
4 years ago
|
|
|
unsigned depth = (unsigned)options >> 16;
|
|
|
|
|
|
|
|
if (size <= 16) {
|
|
|
|
memset(&state.patch, 0, 32);
|
|
|
|
if (size) memcpy(&state.patch, buf, size);
|
|
|
|
buf = state.patch;
|
|
|
|
state.end = buf + size;
|
|
|
|
state.limit = 0;
|
|
|
|
options &= ~kUpb_DecodeOption_AliasString; // Can't alias patch buf.
|
|
|
|
} else {
|
|
|
|
state.end = buf + size - 16;
|
|
|
|
state.limit = 16;
|
|
|
|
}
|
|
|
|
|
|
|
|
state.extreg = extreg;
|
|
|
|
state.limit_ptr = state.end;
|
Fixed bug when parsing an unknown value in a proto2 enum extension. #fuzzing
Proto2 enum parsing is the only case where we have to look at the wire value (not merely the tag) to decide whether the field is known or unknown. If the value is unknown, we need to put the value in the Unknown Fields, but for an extension we no longer have easy access to the message, because for extensions we replace the `msg` pointer with a pointer to the extension. The bug occurred when we were treating the fake `upb_Message*` (which was actually a pointer to an extension) as a real `upb_Message*` that can have unknown fields.
This CL fixes the problem by preserving the true message pointer in `d->unknown_msg` when we are parsing an extension.
This also required fixing a bug in MiniTable building when fasttables are enabled. We need to set the table_mask to `-1` to disable fasttable parsing, not `0`.
For unknown reasons, this CL appears to speed up parsing somewhat significantly. Ideally we should be tracking parsing performance better over time, as it is possible this is merely regaining performance that was lost at a different time:
```
benchy --reference=srcfs third_party/upb/benchmarks:benchmark
10 / 10 [=================================================================================================================] 100.00% 2m32s
(Generated by http://go/benchy. Settings: --runs 5 --reference "srcfs")
name old cpu/op new cpu/op delta
BM_ArenaOneAlloc 23.9ns ± 6% 23.7ns ± 4% ~ (p=0.180 n=53+51)
BM_ArenaInitialBlockOneAlloc 7.62ns ± 4% 7.70ns ± 5% +0.99% (p=0.024 n=59+60)
BM_LoadAdsDescriptor_Upb<NoLayout> 6.60ms ±10% 6.57ms ± 8% ~ (p=0.607 n=47+54)
BM_LoadAdsDescriptor_Upb<WithLayout> 6.92ms ± 5% 6.88ms ± 8% ~ (p=0.257 n=54+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 14.2ms ± 8% 14.0ms ± 7% -1.38% (p=0.025 n=58+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 14.3ms ± 8% 14.2ms ± 8% -1.16% (p=0.031 n=58+57)
BM_Parse_Upb_FileDesc<UseArena, Copy> 15.9µs ± 4% 14.6µs ± 4% -7.85% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 14.5µs ± 4% 13.3µs ± 5% -8.50% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 15.7µs ± 4% 14.4µs ± 5% -7.99% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 14.2µs ± 5% 13.0µs ± 4% -8.56% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 26.3µs ± 4% 26.2µs ± 4% ~ (p=0.195 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 13.3µs ± 5% 13.2µs ± 4% ~ (p=0.085 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 12.9µs ± 4% 12.8µs ± 3% -0.66% (p=0.023 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 10.9µs ± 6% 10.9µs ± 4% ~ (p=0.063 n=59+58)
BM_SerializeDescriptor_Proto2 7.57µs ± 6% 7.62µs ± 6% ~ (p=0.147 n=57+58)
BM_SerializeDescriptor_Upb 12.8µs ± 4% 12.8µs ± 4% ~ (p=0.163 n=59+56)
name old time/op new time/op delta
BM_ArenaOneAlloc 23.9ns ± 5% 23.7ns ± 4% ~ (p=0.172 n=53+51)
BM_ArenaInitialBlockOneAlloc 7.62ns ± 4% 7.70ns ± 5% +1.02% (p=0.017 n=59+60)
BM_LoadAdsDescriptor_Upb<NoLayout> 6.60ms ±10% 6.58ms ± 8% ~ (p=0.727 n=47+55)
BM_LoadAdsDescriptor_Upb<WithLayout> 6.92ms ± 5% 6.88ms ± 8% ~ (p=0.260 n=54+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 14.2ms ± 7% 14.0ms ± 7% -1.40% (p=0.019 n=58+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 14.3ms ± 8% 14.2ms ± 8% -1.13% (p=0.037 n=58+57)
BM_Parse_Upb_FileDesc<UseArena, Copy> 15.9µs ± 4% 14.6µs ± 3% -7.88% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 14.5µs ± 4% 13.3µs ± 5% -8.46% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 15.7µs ± 4% 14.4µs ± 5% -7.99% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 14.2µs ± 5% 13.0µs ± 4% -8.56% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 26.3µs ± 4% 26.2µs ± 4% ~ (p=0.224 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 13.3µs ± 5% 13.2µs ± 4% ~ (p=0.098 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 12.9µs ± 4% 12.8µs ± 3% -0.68% (p=0.015 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 10.9µs ± 6% 10.9µs ± 4% ~ (p=0.052 n=59+58)
BM_SerializeDescriptor_Proto2 7.56µs ± 6% 7.62µs ± 6% ~ (p=0.111 n=58+58)
BM_SerializeDescriptor_Upb 12.8µs ± 4% 12.8µs ± 4% ~ (p=0.241 n=56+56)
name old allocs/op new allocs/op delta
BM_ArenaOneAlloc 1.00 ± 0% 1.00 ± 0% ~ (all samples are equal)
BM_ArenaInitialBlockOneAlloc 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<NoLayout> 5.98k ± 0% 5.98k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<WithLayout> 5.98k ± 0% 5.98k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<NoLayout> 80.9k ± 0% 80.9k ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<WithLayout> 82.1k ± 0% 82.1k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Copy> 7.00 ± 0% 7.00 ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Alias> 7.00 ± 0% 7.00 ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 765 ± 0% 765 ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 9.00 ± 0% 9.00 ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Proto2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Upb 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
name old peak-mem(Bytes)/op new peak-mem(Bytes)/op delta
BM_ArenaOneAlloc 344 ± 0% 344 ± 0% ~ (all samples are equal)
BM_ArenaInitialBlockOneAlloc 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<NoLayout> 9.60M ± 0% 9.60M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Upb<WithLayout> 9.68M ± 0% 9.68M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<NoLayout> 6.41M ± 0% 6.41M ± 0% ~ (all samples are equal)
BM_LoadAdsDescriptor_Proto2<WithLayout> 6.44M ± 0% 6.44M ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Copy> 36.5k ± 0% 36.5k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<UseArena, Alias> 36.5k ± 0% 36.5k ± 0% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 35.8k ± 0% 35.8k ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 40.7k ± 0% 40.7k ± 0% ~ (all samples are equal)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Proto2 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
BM_SerializeDescriptor_Upb 0.00 ±NaN% 0.00 ±NaN% ~ (all samples are equal)
name old speed new speed delta
BM_LoadAdsDescriptor_Upb<NoLayout> 113MB/s ± 9% 113MB/s ± 8% ~ (p=0.712 n=47+55)
BM_LoadAdsDescriptor_Upb<WithLayout> 107MB/s ± 8% 108MB/s ± 8% ~ (p=0.200 n=55+54)
BM_LoadAdsDescriptor_Proto2<NoLayout> 52.5MB/s ± 8% 53.3MB/s ± 7% +1.51% (p=0.018 n=59+59)
BM_LoadAdsDescriptor_Proto2<WithLayout> 51.9MB/s ± 7% 52.4MB/s ± 8% +1.01% (p=0.050 n=58+58)
BM_Parse_Upb_FileDesc<UseArena, Copy> 473MB/s ± 4% 514MB/s ± 4% +8.52% (p=0.000 n=57+59)
BM_Parse_Upb_FileDesc<UseArena, Alias> 518MB/s ± 4% 566MB/s ± 5% +9.30% (p=0.000 n=57+60)
BM_Parse_Upb_FileDesc<InitBlock, Copy> 480MB/s ± 4% 521MB/s ± 5% +8.69% (p=0.000 n=59+60)
BM_Parse_Upb_FileDesc<InitBlock, Alias> 528MB/s ± 4% 578MB/s ± 4% +9.36% (p=0.000 n=57+58)
BM_Parse_Proto2<FileDesc, NoArena, Copy> 286MB/s ± 4% 287MB/s ± 4% ~ (p=0.195 n=55+53)
BM_Parse_Proto2<FileDesc, UseArena, Copy> 566MB/s ± 5% 570MB/s ± 4% ~ (p=0.085 n=59+59)
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 583MB/s ± 5% 587MB/s ± 3% +0.64% (p=0.023 n=60+58)
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 688MB/s ± 6% 693MB/s ± 4% ~ (p=0.063 n=59+58)
BM_SerializeDescriptor_Proto2 995MB/s ± 6% 988MB/s ± 5% ~ (p=0.147 n=57+58)
BM_SerializeDescriptor_Upb 586MB/s ± 4% 589MB/s ± 4% ~ (p=0.163 n=59+56)
```
PiperOrigin-RevId: 462022073
2 years ago
|
|
|
state.unknown = NULL;
|
Added API for copy vs. alias and added benchmarks to test both.
Benchmark output:
$ bazel-bin/benchmarks/benchmark '--benchmark_filter=BM_Parse'
2020-11-11 15:39:04
Running bazel-bin/benchmarks/benchmark
Run on (72 X 3700 MHz CPU s)
CPU Caches:
L1 Data 32K (x36)
L1 Instruction 32K (x36)
L2 Unified 1024K (x36)
L3 Unified 25344K (x2)
-------------------------------------------------------------------------------------
Benchmark Time CPU Iterations
-------------------------------------------------------------------------------------
BM_Parse_Upb_FileDesc<UseArena, Copy> 4134 ns 4134 ns 168714 1.69152GB/s
BM_Parse_Upb_FileDesc<UseArena, Alias> 3487 ns 3487 ns 199509 2.00526GB/s
BM_Parse_Upb_FileDesc<InitBlock, Copy> 3727 ns 3726 ns 187581 1.87643GB/s
BM_Parse_Upb_FileDesc<InitBlock, Alias> 3110 ns 3110 ns 224970 2.24866GB/s
BM_Parse_Proto2<FileDesc, NoArena, Copy> 31132 ns 31132 ns 22437 229.995MB/s
BM_Parse_Proto2<FileDesc, UseArena, Copy> 21011 ns 21009 ns 33922 340.812MB/s
BM_Parse_Proto2<FileDesc, InitBlock, Copy> 17976 ns 17975 ns 38808 398.337MB/s
BM_Parse_Proto2<FileDescSV, InitBlock, Alias> 17357 ns 17356 ns 40244 412.539MB/s
4 years ago
|
|
|
state.depth = depth ? depth : 64;
|
|
|
|
state.end_group = DECODE_NOGROUP;
|
|
|
|
state.options = (uint16_t)options;
|
|
|
|
state.missing_required = false;
|
|
|
|
state.arena.head = arena->head;
|
|
|
|
state.arena.last_size = arena->last_size;
|
|
|
|
state.arena.cleanup_metadata = arena->cleanup_metadata;
|
|
|
|
state.arena.parent = arena;
|
|
|
|
|
|
|
|
upb_DecodeStatus status = UPB_SETJMP(state.err);
|
|
|
|
if (UPB_LIKELY(status == kUpb_DecodeStatus_Ok)) {
|
|
|
|
status = _upb_Decoder_DecodeTop(&state, buf, msg, l);
|
|
|
|
}
|
|
|
|
|
|
|
|
arena->head.ptr = state.arena.head.ptr;
|
|
|
|
arena->head.end = state.arena.head.end;
|
|
|
|
arena->cleanup_metadata = state.arena.cleanup_metadata;
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
|
|
|
#undef OP_UNKNOWN
|
|
|
|
#undef OP_SKIP
|
|
|
|
#undef OP_SCALAR_LG2
|
|
|
|
#undef OP_FIXPCK_LG2
|
|
|
|
#undef OP_VARPCK_LG2
|
|
|
|
#undef OP_STRING
|
|
|
|
#undef OP_BYTES
|
|
|
|
#undef OP_SUBMSG
|