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

448 lines
14 KiB

// Fast decoder: ~3x the speed of decode.c, but x86-64 specific.
// Also the table size grows by 2x.
//
// Could potentially be ported to ARM64 or other 64-bit archs that pass at
// least six arguments in registers.
4 years ago
#include "upb/decode_fast.h"
#include "upb/decode.int.h"
4 years ago
/* Must be last. */
4 years ago
#include "upb/port_def.inc"
// The standard set of arguments passed to each parsing function.
// Thanks to x86-64 calling conventions, these will stay in registers.
#define UPB_PARSE_PARAMS \
upb_decstate *d, const char *ptr, upb_msg *msg, const upb_msglayout *table, \
4 years ago
uint64_t hasbits, uint64_t data
4 years ago
#define UPB_PARSE_ARGS d, ptr, msg, table, hasbits, data
#define RETURN_GENERIC(msg) \
/* fprintf(stderr, msg); */ \
return fastdecode_generic(UPB_PARSE_ARGS);
typedef enum {
CARD_s = 0, /* Singular (optional, non-repeated) */
CARD_o = 1, /* Oneof */
CARD_r = 2 /* Repeated */
} upb_card;
UPB_INLINE
upb_msg *decode_newmsg_ceil(upb_decstate *d, const upb_msglayout *l,
int msg_ceil_bytes) {
size_t size = l->size + sizeof(upb_msg_internal);
char *msg_data;
if (UPB_LIKELY(msg_ceil_bytes > 0 && _upb_arenahas(&d->arena, msg_ceil_bytes))) {
UPB_ASSERT(size <= (size_t)msg_ceil_bytes);
msg_data = d->arena.head.ptr;
d->arena.head.ptr += size;
UPB_UNPOISON_MEMORY_REGION(msg_data, msg_ceil_bytes);
memset(msg_data, 0, msg_ceil_bytes);
UPB_POISON_MEMORY_REGION(msg_data + size, msg_ceil_bytes - size);
} else {
msg_data = (char*)upb_arena_malloc(&d->arena, size);
memset(msg_data, 0, size);
}
return msg_data + sizeof(upb_msg_internal);
}
UPB_FORCEINLINE
static const char *fastdecode_tagdispatch(upb_decstate *d, const char *ptr,
upb_msg *msg,
const upb_msglayout *table,
uint64_t hasbits, uint32_t tag) {
// Get 5 bits of field number (we pretend the continuation bit is a data bit,
// speculating that the second byte, if any, will be 0x01).
size_t idx = (tag & 0xf8) >> 3;
// Xor the actual tag with the expected tag (in the low bytes of the table)
// so that the field parser can verify the tag by comparing with zero.
uint64_t data = table->fasttable[idx].field_data ^ tag;
// Jump to the specialized field parser function.
return table->fasttable[idx].field_parser(UPB_PARSE_ARGS);
}
UPB_FORCEINLINE
static uint32_t fastdecode_loadtag(const char *ptr) {
uint16_t tag;
memcpy(&tag, ptr, 2);
return tag;
}
UPB_NOINLINE
static const char *fastdecode_isdonefallback(upb_decstate *d, const char *ptr,
upb_msg *msg,
const upb_msglayout *table,
uint64_t hasbits, int overrun) {
ptr = decode_isdonefallback_inl(d, ptr, overrun);
if (ptr == NULL) {
return fastdecode_err(d);
}
uint16_t tag = fastdecode_loadtag(ptr);
return fastdecode_tagdispatch(d, ptr, msg, table, hasbits, tag);
}
4 years ago
UPB_FORCEINLINE
const char *fastdecode_dispatch(upb_decstate *d, const char *ptr, upb_msg *msg,
const upb_msglayout *table, uint64_t hasbits) {
if (UPB_UNLIKELY(ptr >= d->limit_ptr)) {
int overrun = ptr - d->end;
if (UPB_LIKELY(overrun == d->limit)) {
// Parse is finished.
*(uint32_t*)msg |= hasbits >> 16; // Sync hasbits.
return ptr;
} else {
return fastdecode_isdonefallback(d, ptr, msg, table, hasbits, overrun);
}
}
// Read two bytes of tag data (for a one-byte tag, the high byte is junk).
uint16_t tag = fastdecode_loadtag(ptr);
return fastdecode_tagdispatch(d, ptr, msg, table, hasbits, tag);
4 years ago
}
4 years ago
UPB_FORCEINLINE
static bool fastdecode_checktag(uint64_t data, int tagbytes) {
4 years ago
if (tagbytes == 1) {
return (data & 0xff) == 0;
} else {
return (data & 0xffff) == 0;
}
}
UPB_FORCEINLINE
static void *fastdecode_getfield_ofs(upb_decstate *d, const char *ptr,
upb_msg *msg, uint64_t *data,
uint64_t *hasbits, upb_array **outarr,
void **end, int valbytes,
upb_card card, bool hasbit_is_idx) {
size_t ofs = *data >> 48;
void *field = (char *)msg + ofs;
4 years ago
switch (card) {
case CARD_s:
// Set hasbit and return pointer to scalar field.
if (hasbit_is_idx) {
*hasbits |= 1ull << ((*data >> 32) & 63);
} else {
*hasbits |= *data;
}
return field;
case CARD_r: {
// Get pointer to upb_array and allocate/expand if necessary.
uint8_t elem_size_lg2 = __builtin_ctz(valbytes);
upb_array **arr_p = field;
upb_array *arr;
char *begin;
*hasbits >>= 16;
*(uint32_t*)msg |= *hasbits;
*hasbits = 0;
if (UPB_LIKELY(!*arr_p)) {
arr = _upb_array_new(&d->arena, 8, elem_size_lg2);
*arr_p = arr;
} else {
arr = *arr_p;
}
begin = _upb_array_ptr(arr);
field = begin + (arr->len * valbytes);
*end = begin + (arr->size * valbytes);
*data = fastdecode_loadtag(ptr);
*outarr = arr;
return field;
}
default:
UPB_UNREACHABLE();
4 years ago
}
}
UPB_FORCEINLINE
static void *fastdecode_getfield(upb_decstate *d, const char *ptr, upb_msg *msg,
uint64_t *data, uint64_t *hasbits,
int valbytes, upb_card card) {
return fastdecode_getfield_ofs(d, ptr, msg, data, hasbits, NULL, NULL,
valbytes, card, false);
}
/* varint fields **************************************************************/
4 years ago
UPB_FORCEINLINE
static uint64_t fastdecode_munge(uint64_t val, int valbytes, bool zigzag) {
if (valbytes == 1) {
return val != 0;
} else if (zigzag) {
if (valbytes == 4) {
uint32_t n = val;
return (n >> 1) ^ -(int32_t)(n & 1);
} else if (valbytes == 8) {
return (val >> 1) ^ -(int64_t)(val & 1);
}
UPB_UNREACHABLE();
}
return val;
}
4 years ago
UPB_FORCEINLINE
4 years ago
static const char *fastdecode_varint(UPB_PARSE_PARAMS, int tagbytes,
int valbytes, upb_card card, bool zigzag) {
uint64_t val;
void *dst;
if (UPB_UNLIKELY(!fastdecode_checktag(data, tagbytes))) {
RETURN_GENERIC("varint field tag mismatch\n");
4 years ago
}
dst = fastdecode_getfield(d, ptr, msg, &data, &hasbits, valbytes,
card);
ptr += tagbytes + 1;
val = (uint8_t)ptr[-1];
if (UPB_UNLIKELY(val & 0x80)) {
4 years ago
int i;
for (i = 0; i < 8; i++) {
ptr++;
uint64_t byte = (uint8_t)ptr[-1];
val += (byte - 1) << (7 + 7 * i);
if (UPB_LIKELY((byte & 0x80) == 0)) goto done;
}
ptr++;
uint64_t byte = (uint8_t)ptr[-1];
if (byte > 1) return fastdecode_err(d);
val += (byte - 1) << 63;
}
done:
val = fastdecode_munge(val, valbytes, zigzag);
memcpy(dst, &val, valbytes);
return fastdecode_dispatch(d, ptr, msg, table, hasbits);
4 years ago
}
4 years ago
#define z_ZZ true
#define b_ZZ false
#define v_ZZ false
4 years ago
/* Generate all varint functions.
* {s,o,r} x {b1,v4,z4,v8,z8} x {1bt,2bt} */
4 years ago
#define F(card, type, valbytes, tagbytes) \
const char *upb_p##card##type##valbytes##_##tagbytes##bt(UPB_PARSE_PARAMS) { \
return fastdecode_varint(UPB_PARSE_ARGS, tagbytes, valbytes, CARD_##card, \
type##_ZZ); \
4 years ago
}
4 years ago
4 years ago
#define TYPES(card, tagbytes) \
F(card, b, 1, tagbytes) \
F(card, v, 4, tagbytes) \
F(card, v, 8, tagbytes) \
F(card, z, 4, tagbytes) \
F(card, z, 8, tagbytes)
#define TAGBYTES(card) \
TYPES(card, 1) \
TYPES(card, 2)
TAGBYTES(s)
TAGBYTES(o)
/* TAGBYTES(r) */
4 years ago
#undef z_ZZ
#undef b_ZZ
#undef v_ZZ
#undef o_ONEOF
#undef s_ONEOF
#undef r_ONEOF
#undef F
#undef TYPES
#undef TAGBYTES
/* string fields **************************************************************/
UPB_FORCEINLINE
static bool fastdecode_boundscheck(const char *ptr, size_t len,
const char *end) {
uintptr_t uptr = (uintptr_t)ptr;
uintptr_t uend = (uintptr_t)end + 16;
uintptr_t res = uptr + len;
return res < uptr || res > uend;
}
UPB_NOINLINE
static const char *fastdecode_copystring(struct upb_decstate *d,
size_t len, upb_msg *msg,
const upb_msglayout *table,
uint64_t hasbits, upb_strview *str) {
char *ptr = (char*)str->data;
char *data = upb_arena_malloc(&d->arena, len);
if (!data) {
return fastdecode_err(d);
}
memcpy(data, ptr, len);
str->data = data;
return fastdecode_dispatch(d, ptr + len, msg, table, hasbits);
}
UPB_FORCEINLINE
static const char *fastdecode_string(UPB_PARSE_PARAMS, int tagbytes,
upb_card card) {
upb_strview *dst;
const char *str;
int64_t len;
if (UPB_UNLIKELY(!fastdecode_checktag(data, tagbytes))) {
RETURN_GENERIC("string field tag mismatch\n");
}
dst = fastdecode_getfield(d, ptr, msg, &data, &hasbits,
sizeof(upb_strview), card);
len = (int8_t)ptr[tagbytes];
str = ptr + tagbytes + 1;
dst->data = str;
dst->size = len;
if (UPB_UNLIKELY(fastdecode_boundscheck(str, len, d->limit_ptr))) {
dst->size = 0;
RETURN_GENERIC("string field len >1 byte\n");
}
if (!d->alias) {
return fastdecode_copystring(d, len, msg, table, hasbits, dst);
}
return fastdecode_dispatch(d, str + len, msg, table, hasbits);
}
const char *upb_pss_1bt(UPB_PARSE_PARAMS) {
return fastdecode_string(UPB_PARSE_ARGS, 1, CARD_s);
}
const char *upb_pos_1bt(UPB_PARSE_PARAMS) {
return fastdecode_string(UPB_PARSE_ARGS, 1, CARD_o);
}
const char *upb_pss_2bt(UPB_PARSE_PARAMS) {
return fastdecode_string(UPB_PARSE_ARGS, 2, CARD_s);
}
const char *upb_pos_2bt(UPB_PARSE_PARAMS) {
return fastdecode_string(UPB_PARSE_ARGS, 2, CARD_o);
}
/* message fields *************************************************************/
UPB_FORCEINLINE
static const char *fastdecode_submsg(UPB_PARSE_PARAMS, int tagbytes,
I think this may have reached the optimization limit. ------------------------------------------------------------------------- Benchmark Time CPU Iterations ------------------------------------------------------------------------- BM_ArenaOneAlloc 21 ns 21 ns 32994231 BM_ArenaInitialBlockOneAlloc 6 ns 6 ns 116318005 BM_ParseDescriptorNoHeap 3028 ns 3028 ns 231138 2.34354GB/s BM_ParseDescriptor 3557 ns 3557 ns 196583 1.99498GB/s BM_ParseDescriptorProto2NoArena 33228 ns 33226 ns 21196 218.688MB/s BM_ParseDescriptorProto2WithArena 22863 ns 22861 ns 30666 317.831MB/s BM_SerializeDescriptorProto2 5444 ns 5444 ns 127368 1.30348GB/s BM_SerializeDescriptor 12509 ns 12508 ns 55816 580.914MB/s $ perf stat bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap 2020-10-08 14:07:06 Running bazel-bin/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_ParseDescriptorNoHeap 3071 ns 3071 ns 227743 2.31094GB/s Performance counter stats for 'bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap': 1,050.22 msec task-clock # 0.978 CPUs utilized 4 context-switches # 0.004 K/sec 0 cpu-migrations # 0.000 K/sec 179 page-faults # 0.170 K/sec 3,875,796,334 cycles # 3.690 GHz 13,282,835,967 instructions # 3.43 insn per cycle 2,887,725,848 branches # 2749.627 M/sec 8,324,912 branch-misses # 0.29% of all branches 1.073924364 seconds time elapsed 1.042806000 seconds user 0.008021000 seconds sys Profile: 23.96% benchmark benchmark [.] upb_prm_1bt_max192b 22.44% benchmark benchmark [.] fastdecode_dispatch 18.96% benchmark benchmark [.] upb_pss_1bt 14.20% benchmark benchmark [.] upb_psv4_1bt 8.33% benchmark benchmark [.] upb_prm_1bt_max64b 6.66% benchmark benchmark [.] upb_prm_1bt_max128b 1.29% benchmark benchmark [.] upb_psm_1bt_max64b 0.77% benchmark benchmark [.] fastdecode_generic 0.55% benchmark [kernel.kallsyms] [k] smp_call_function_single 0.42% benchmark [kernel.kallsyms] [k] _raw_spin_lock_irqsave 0.42% benchmark benchmark [.] upb_psm_1bt_max256b 0.31% benchmark benchmark [.] upb_psb1_1bt 0.21% benchmark benchmark [.] upb_plv4_5bv 0.14% benchmark benchmark [.] upb_psb1_2bt 0.12% benchmark benchmark [.] decode_longvarint64 0.08% benchmark [kernel.kallsyms] [k] vsnprintf 0.07% benchmark [kernel.kallsyms] [k] _raw_spin_lock 0.07% benchmark benchmark [.] _upb_msg_new 0.06% benchmark ld-2.31.so [.] check_match
4 years ago
int msg_ceil_bytes, upb_card card) {
if (UPB_UNLIKELY(!fastdecode_checktag(data, tagbytes))) {
RETURN_GENERIC("submessage field tag mismatch\n");
}
if (--d->depth == 0) return fastdecode_err(d);
upb_msg **submsg;
upb_array *arr;
void *end;
uint32_t submsg_idx = data;
4 years ago
submsg_idx >>= 16;
const upb_msglayout *subl = table->submsgs[submsg_idx];
submsg = fastdecode_getfield_ofs(d, ptr, msg, &data, &hasbits, &arr, &end,
sizeof(upb_msg *), card, true);
if (card == CARD_s) {
*(uint32_t*)msg |= hasbits >> 16;
hasbits = 0;
}
again:
if (card == CARD_r) {
if (UPB_UNLIKELY(submsg == end)) {
4 years ago
size_t old_size = arr->size;
size_t old_bytes = old_size * sizeof(upb_msg*);
size_t new_size = old_size * 2;
size_t new_bytes = new_size * sizeof(upb_msg*);
char *old_ptr = _upb_array_ptr(arr);
char *new_ptr = upb_arena_realloc(&d->arena, old_ptr, old_bytes, new_bytes);
4 years ago
arr->size = new_size;
arr->data = _upb_array_tagptr(new_ptr, 3);
submsg = (void*)(new_ptr + (old_size * sizeof(upb_msg*)));
end = (void*)(new_ptr + (new_size * sizeof(upb_msg*)));
}
}
upb_msg* child = *submsg;
if (card == CARD_r || UPB_LIKELY(!child)) {
*submsg = child = decode_newmsg_ceil(d, subl, msg_ceil_bytes);
}
ptr += tagbytes + 1;
size_t len = (uint8_t)ptr[-1];
int saved_delta;
if (UPB_UNLIKELY(len & 0x80)) {
int i;
for (i = 0; i < 3; i++) {
ptr++;
size_t byte = (uint8_t)ptr[-1];
len += (byte - 1) << (7 + 7 * i);
if (UPB_LIKELY((byte & 0x80) == 0)) goto done;
}
ptr++;
size_t byte = (uint8_t)ptr[-1];
// len is limited by 2gb not 4gb, hence 8 and not 16 as normally expected
// for a 32 bit varint.
if (UPB_UNLIKELY(byte >= 8)) return fastdecode_err(d);
len += (byte - 1) << 28;
}
done:
if (ptr - d->end + (int)len > d->limit) {
return fastdecode_err(d);
}
saved_delta = decode_pushlimit(d, ptr, len);
ptr = fastdecode_dispatch(d, ptr, child, subl, 0);
decode_poplimit(d, saved_delta);
if (UPB_UNLIKELY(d->end_group != 0)) {
return fastdecode_err(d);
}
if (card == CARD_r) {
submsg++;
if (UPB_LIKELY(!decode_isdone(d, &ptr))) {
uint32_t tag = fastdecode_loadtag(ptr);
if (tagbytes == 1) {
if ((uint8_t)tag == (uint8_t)data) goto again;
} else {
if ((uint16_t)tag == (uint16_t)data) goto again;
}
arr->len = submsg - (upb_msg**)_upb_array_ptr(arr);
d->depth++;
return fastdecode_tagdispatch(d, ptr, msg, table, hasbits, tag);
} else {
arr->len = submsg - (upb_msg**)_upb_array_ptr(arr);
d->depth++;
return ptr;
}
}
d->depth++;
return fastdecode_dispatch(d, ptr, msg, table, hasbits);
}
I think this may have reached the optimization limit. ------------------------------------------------------------------------- Benchmark Time CPU Iterations ------------------------------------------------------------------------- BM_ArenaOneAlloc 21 ns 21 ns 32994231 BM_ArenaInitialBlockOneAlloc 6 ns 6 ns 116318005 BM_ParseDescriptorNoHeap 3028 ns 3028 ns 231138 2.34354GB/s BM_ParseDescriptor 3557 ns 3557 ns 196583 1.99498GB/s BM_ParseDescriptorProto2NoArena 33228 ns 33226 ns 21196 218.688MB/s BM_ParseDescriptorProto2WithArena 22863 ns 22861 ns 30666 317.831MB/s BM_SerializeDescriptorProto2 5444 ns 5444 ns 127368 1.30348GB/s BM_SerializeDescriptor 12509 ns 12508 ns 55816 580.914MB/s $ perf stat bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap 2020-10-08 14:07:06 Running bazel-bin/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_ParseDescriptorNoHeap 3071 ns 3071 ns 227743 2.31094GB/s Performance counter stats for 'bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap': 1,050.22 msec task-clock # 0.978 CPUs utilized 4 context-switches # 0.004 K/sec 0 cpu-migrations # 0.000 K/sec 179 page-faults # 0.170 K/sec 3,875,796,334 cycles # 3.690 GHz 13,282,835,967 instructions # 3.43 insn per cycle 2,887,725,848 branches # 2749.627 M/sec 8,324,912 branch-misses # 0.29% of all branches 1.073924364 seconds time elapsed 1.042806000 seconds user 0.008021000 seconds sys Profile: 23.96% benchmark benchmark [.] upb_prm_1bt_max192b 22.44% benchmark benchmark [.] fastdecode_dispatch 18.96% benchmark benchmark [.] upb_pss_1bt 14.20% benchmark benchmark [.] upb_psv4_1bt 8.33% benchmark benchmark [.] upb_prm_1bt_max64b 6.66% benchmark benchmark [.] upb_prm_1bt_max128b 1.29% benchmark benchmark [.] upb_psm_1bt_max64b 0.77% benchmark benchmark [.] fastdecode_generic 0.55% benchmark [kernel.kallsyms] [k] smp_call_function_single 0.42% benchmark [kernel.kallsyms] [k] _raw_spin_lock_irqsave 0.42% benchmark benchmark [.] upb_psm_1bt_max256b 0.31% benchmark benchmark [.] upb_psb1_1bt 0.21% benchmark benchmark [.] upb_plv4_5bv 0.14% benchmark benchmark [.] upb_psb1_2bt 0.12% benchmark benchmark [.] decode_longvarint64 0.08% benchmark [kernel.kallsyms] [k] vsnprintf 0.07% benchmark [kernel.kallsyms] [k] _raw_spin_lock 0.07% benchmark benchmark [.] _upb_msg_new 0.06% benchmark ld-2.31.so [.] check_match
4 years ago
#define F(card, tagbytes, size_ceil, ceil_arg) \
const char *upb_p##card##m_##tagbytes##bt_max##size_ceil##b( \
UPB_PARSE_PARAMS) { \
return fastdecode_submsg(UPB_PARSE_ARGS, tagbytes, ceil_arg, CARD_##card); \
}
I think this may have reached the optimization limit. ------------------------------------------------------------------------- Benchmark Time CPU Iterations ------------------------------------------------------------------------- BM_ArenaOneAlloc 21 ns 21 ns 32994231 BM_ArenaInitialBlockOneAlloc 6 ns 6 ns 116318005 BM_ParseDescriptorNoHeap 3028 ns 3028 ns 231138 2.34354GB/s BM_ParseDescriptor 3557 ns 3557 ns 196583 1.99498GB/s BM_ParseDescriptorProto2NoArena 33228 ns 33226 ns 21196 218.688MB/s BM_ParseDescriptorProto2WithArena 22863 ns 22861 ns 30666 317.831MB/s BM_SerializeDescriptorProto2 5444 ns 5444 ns 127368 1.30348GB/s BM_SerializeDescriptor 12509 ns 12508 ns 55816 580.914MB/s $ perf stat bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap 2020-10-08 14:07:06 Running bazel-bin/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_ParseDescriptorNoHeap 3071 ns 3071 ns 227743 2.31094GB/s Performance counter stats for 'bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap': 1,050.22 msec task-clock # 0.978 CPUs utilized 4 context-switches # 0.004 K/sec 0 cpu-migrations # 0.000 K/sec 179 page-faults # 0.170 K/sec 3,875,796,334 cycles # 3.690 GHz 13,282,835,967 instructions # 3.43 insn per cycle 2,887,725,848 branches # 2749.627 M/sec 8,324,912 branch-misses # 0.29% of all branches 1.073924364 seconds time elapsed 1.042806000 seconds user 0.008021000 seconds sys Profile: 23.96% benchmark benchmark [.] upb_prm_1bt_max192b 22.44% benchmark benchmark [.] fastdecode_dispatch 18.96% benchmark benchmark [.] upb_pss_1bt 14.20% benchmark benchmark [.] upb_psv4_1bt 8.33% benchmark benchmark [.] upb_prm_1bt_max64b 6.66% benchmark benchmark [.] upb_prm_1bt_max128b 1.29% benchmark benchmark [.] upb_psm_1bt_max64b 0.77% benchmark benchmark [.] fastdecode_generic 0.55% benchmark [kernel.kallsyms] [k] smp_call_function_single 0.42% benchmark [kernel.kallsyms] [k] _raw_spin_lock_irqsave 0.42% benchmark benchmark [.] upb_psm_1bt_max256b 0.31% benchmark benchmark [.] upb_psb1_1bt 0.21% benchmark benchmark [.] upb_plv4_5bv 0.14% benchmark benchmark [.] upb_psb1_2bt 0.12% benchmark benchmark [.] decode_longvarint64 0.08% benchmark [kernel.kallsyms] [k] vsnprintf 0.07% benchmark [kernel.kallsyms] [k] _raw_spin_lock 0.07% benchmark benchmark [.] _upb_msg_new 0.06% benchmark ld-2.31.so [.] check_match
4 years ago
#define SIZES(card, tagbytes) \
F(card, tagbytes, 64, 64) \
F(card, tagbytes, 128, 128) \
F(card, tagbytes, 192, 192) \
F(card, tagbytes, 256, 256) \
F(card, tagbytes, max, -1)
I think this may have reached the optimization limit. ------------------------------------------------------------------------- Benchmark Time CPU Iterations ------------------------------------------------------------------------- BM_ArenaOneAlloc 21 ns 21 ns 32994231 BM_ArenaInitialBlockOneAlloc 6 ns 6 ns 116318005 BM_ParseDescriptorNoHeap 3028 ns 3028 ns 231138 2.34354GB/s BM_ParseDescriptor 3557 ns 3557 ns 196583 1.99498GB/s BM_ParseDescriptorProto2NoArena 33228 ns 33226 ns 21196 218.688MB/s BM_ParseDescriptorProto2WithArena 22863 ns 22861 ns 30666 317.831MB/s BM_SerializeDescriptorProto2 5444 ns 5444 ns 127368 1.30348GB/s BM_SerializeDescriptor 12509 ns 12508 ns 55816 580.914MB/s $ perf stat bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap 2020-10-08 14:07:06 Running bazel-bin/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_ParseDescriptorNoHeap 3071 ns 3071 ns 227743 2.31094GB/s Performance counter stats for 'bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap': 1,050.22 msec task-clock # 0.978 CPUs utilized 4 context-switches # 0.004 K/sec 0 cpu-migrations # 0.000 K/sec 179 page-faults # 0.170 K/sec 3,875,796,334 cycles # 3.690 GHz 13,282,835,967 instructions # 3.43 insn per cycle 2,887,725,848 branches # 2749.627 M/sec 8,324,912 branch-misses # 0.29% of all branches 1.073924364 seconds time elapsed 1.042806000 seconds user 0.008021000 seconds sys Profile: 23.96% benchmark benchmark [.] upb_prm_1bt_max192b 22.44% benchmark benchmark [.] fastdecode_dispatch 18.96% benchmark benchmark [.] upb_pss_1bt 14.20% benchmark benchmark [.] upb_psv4_1bt 8.33% benchmark benchmark [.] upb_prm_1bt_max64b 6.66% benchmark benchmark [.] upb_prm_1bt_max128b 1.29% benchmark benchmark [.] upb_psm_1bt_max64b 0.77% benchmark benchmark [.] fastdecode_generic 0.55% benchmark [kernel.kallsyms] [k] smp_call_function_single 0.42% benchmark [kernel.kallsyms] [k] _raw_spin_lock_irqsave 0.42% benchmark benchmark [.] upb_psm_1bt_max256b 0.31% benchmark benchmark [.] upb_psb1_1bt 0.21% benchmark benchmark [.] upb_plv4_5bv 0.14% benchmark benchmark [.] upb_psb1_2bt 0.12% benchmark benchmark [.] decode_longvarint64 0.08% benchmark [kernel.kallsyms] [k] vsnprintf 0.07% benchmark [kernel.kallsyms] [k] _raw_spin_lock 0.07% benchmark benchmark [.] _upb_msg_new 0.06% benchmark ld-2.31.so [.] check_match
4 years ago
#define TAGBYTES(card) \
SIZES(card, 1) \
SIZES(card, 2)
I think this may have reached the optimization limit. ------------------------------------------------------------------------- Benchmark Time CPU Iterations ------------------------------------------------------------------------- BM_ArenaOneAlloc 21 ns 21 ns 32994231 BM_ArenaInitialBlockOneAlloc 6 ns 6 ns 116318005 BM_ParseDescriptorNoHeap 3028 ns 3028 ns 231138 2.34354GB/s BM_ParseDescriptor 3557 ns 3557 ns 196583 1.99498GB/s BM_ParseDescriptorProto2NoArena 33228 ns 33226 ns 21196 218.688MB/s BM_ParseDescriptorProto2WithArena 22863 ns 22861 ns 30666 317.831MB/s BM_SerializeDescriptorProto2 5444 ns 5444 ns 127368 1.30348GB/s BM_SerializeDescriptor 12509 ns 12508 ns 55816 580.914MB/s $ perf stat bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap 2020-10-08 14:07:06 Running bazel-bin/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_ParseDescriptorNoHeap 3071 ns 3071 ns 227743 2.31094GB/s Performance counter stats for 'bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap': 1,050.22 msec task-clock # 0.978 CPUs utilized 4 context-switches # 0.004 K/sec 0 cpu-migrations # 0.000 K/sec 179 page-faults # 0.170 K/sec 3,875,796,334 cycles # 3.690 GHz 13,282,835,967 instructions # 3.43 insn per cycle 2,887,725,848 branches # 2749.627 M/sec 8,324,912 branch-misses # 0.29% of all branches 1.073924364 seconds time elapsed 1.042806000 seconds user 0.008021000 seconds sys Profile: 23.96% benchmark benchmark [.] upb_prm_1bt_max192b 22.44% benchmark benchmark [.] fastdecode_dispatch 18.96% benchmark benchmark [.] upb_pss_1bt 14.20% benchmark benchmark [.] upb_psv4_1bt 8.33% benchmark benchmark [.] upb_prm_1bt_max64b 6.66% benchmark benchmark [.] upb_prm_1bt_max128b 1.29% benchmark benchmark [.] upb_psm_1bt_max64b 0.77% benchmark benchmark [.] fastdecode_generic 0.55% benchmark [kernel.kallsyms] [k] smp_call_function_single 0.42% benchmark [kernel.kallsyms] [k] _raw_spin_lock_irqsave 0.42% benchmark benchmark [.] upb_psm_1bt_max256b 0.31% benchmark benchmark [.] upb_psb1_1bt 0.21% benchmark benchmark [.] upb_plv4_5bv 0.14% benchmark benchmark [.] upb_psb1_2bt 0.12% benchmark benchmark [.] decode_longvarint64 0.08% benchmark [kernel.kallsyms] [k] vsnprintf 0.07% benchmark [kernel.kallsyms] [k] _raw_spin_lock 0.07% benchmark benchmark [.] _upb_msg_new 0.06% benchmark ld-2.31.so [.] check_match
4 years ago
TAGBYTES(s)
TAGBYTES(o)
TAGBYTES(r)
I think this may have reached the optimization limit. ------------------------------------------------------------------------- Benchmark Time CPU Iterations ------------------------------------------------------------------------- BM_ArenaOneAlloc 21 ns 21 ns 32994231 BM_ArenaInitialBlockOneAlloc 6 ns 6 ns 116318005 BM_ParseDescriptorNoHeap 3028 ns 3028 ns 231138 2.34354GB/s BM_ParseDescriptor 3557 ns 3557 ns 196583 1.99498GB/s BM_ParseDescriptorProto2NoArena 33228 ns 33226 ns 21196 218.688MB/s BM_ParseDescriptorProto2WithArena 22863 ns 22861 ns 30666 317.831MB/s BM_SerializeDescriptorProto2 5444 ns 5444 ns 127368 1.30348GB/s BM_SerializeDescriptor 12509 ns 12508 ns 55816 580.914MB/s $ perf stat bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap 2020-10-08 14:07:06 Running bazel-bin/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_ParseDescriptorNoHeap 3071 ns 3071 ns 227743 2.31094GB/s Performance counter stats for 'bazel-bin/benchmark --benchmark_filter=BM_ParseDescriptorNoHeap': 1,050.22 msec task-clock # 0.978 CPUs utilized 4 context-switches # 0.004 K/sec 0 cpu-migrations # 0.000 K/sec 179 page-faults # 0.170 K/sec 3,875,796,334 cycles # 3.690 GHz 13,282,835,967 instructions # 3.43 insn per cycle 2,887,725,848 branches # 2749.627 M/sec 8,324,912 branch-misses # 0.29% of all branches 1.073924364 seconds time elapsed 1.042806000 seconds user 0.008021000 seconds sys Profile: 23.96% benchmark benchmark [.] upb_prm_1bt_max192b 22.44% benchmark benchmark [.] fastdecode_dispatch 18.96% benchmark benchmark [.] upb_pss_1bt 14.20% benchmark benchmark [.] upb_psv4_1bt 8.33% benchmark benchmark [.] upb_prm_1bt_max64b 6.66% benchmark benchmark [.] upb_prm_1bt_max128b 1.29% benchmark benchmark [.] upb_psm_1bt_max64b 0.77% benchmark benchmark [.] fastdecode_generic 0.55% benchmark [kernel.kallsyms] [k] smp_call_function_single 0.42% benchmark [kernel.kallsyms] [k] _raw_spin_lock_irqsave 0.42% benchmark benchmark [.] upb_psm_1bt_max256b 0.31% benchmark benchmark [.] upb_psb1_1bt 0.21% benchmark benchmark [.] upb_plv4_5bv 0.14% benchmark benchmark [.] upb_psb1_2bt 0.12% benchmark benchmark [.] decode_longvarint64 0.08% benchmark [kernel.kallsyms] [k] vsnprintf 0.07% benchmark [kernel.kallsyms] [k] _raw_spin_lock 0.07% benchmark benchmark [.] _upb_msg_new 0.06% benchmark ld-2.31.so [.] check_match
4 years ago
#undef TAGBYTES
#undef SIZES
#undef F