TiKV rust upgrade to nightly-2012-12-06 benchmarks

benchcmp-misc.txt

+ cargo benchcmp misc-before.txt misc-after.txt
 name                                                              misc-before.txt ns/iter  misc-after.txt ns/iter  diff ns/iter   diff %  speedup
 _bench_check_requirement                                          0                        0                                  0     NaN%    x NaN
 channel::bench_channel::bench_crossbeam_channel                   39                       39                                 0    0.00%   x 1.00
 channel::bench_channel::bench_mio_channel                         155                      148                               -7   -4.52%   x 1.05
 channel::bench_channel::bench_receiver_stream                     1,858,467                1,869,328                     10,861    0.58%   x 0.99
 channel::bench_channel::bench_receiver_stream_batch               1,103,490                1,103,264                       -226   -0.02%   x 1.00
 channel::bench_channel::bench_thread_channel                      28                       27                                -1   -3.57%   x 1.04
 channel::bench_channel::bench_util_channel                        40                       39                                -1   -2.50%   x 1.03
 channel::bench_channel::bench_util_loose                          85                       89                                 4    4.71%   x 0.96
 coprocessor::codec::bench_record_prefix_bigendian_check           940                      941                                1    0.11%   x 1.00
 coprocessor::codec::bench_record_prefix_equal_check               1,100                    1,101                              1    0.09%   x 1.00
 coprocessor::codec::bench_record_prefix_littleendian_check        817                      818                                1    0.12%   x 1.00
 coprocessor::codec::bench_record_prefix_start_with                817                      826                                9    1.10%   x 0.99
 coprocessor::codec::bench_table_prefix_check                      816                      818                                2    0.25%   x 1.00
 coprocessor::codec::bench_table_prefix_start_with                 816                      819                                3    0.37%   x 1.00
 coprocessor::codec::chunk::bench_chunk_build_offical              63,145                   62,897                          -248   -0.39%   x 1.00
 coprocessor::codec::chunk::bench_chunk_build_tidb                 29,878                   28,900                          -978   -3.27%   x 1.03
 coprocessor::codec::chunk::bench_chunk_iter_offical               33,841                   37,339                         3,498   10.34%   x 0.91
 coprocessor::codec::chunk::bench_chunk_iter_tidb                  38,088                   38,639                           551    1.45%   x 0.99
 coprocessor::codec::chunk::bench_encode_chunk                     18,284                   21,912                         3,628   19.84%   x 0.83
 coprocessor::dag::expr::scalar::bench_get_scalar_args_with_map    3,387                    3,361                            -26   -0.77%   x 1.01
 coprocessor::dag::expr::scalar::bench_get_scalar_args_with_match  1,137                    1,133                             -4   -0.35%   x 1.00
 raftkv::bench_async_snapshot                                      1,945                    1,888                            -57   -2.93%   x 1.03
 raftkv::bench_async_snapshots_noop                                164                      177                               13    7.93%   x 0.93
 raftkv::bench_async_write                                         1,257                    1,228                            -29   -2.31%   x 1.02
 serialization::bench_serialization::bench_decode_one              494                      474                              -20   -4.05%   x 1.04
 serialization::bench_serialization::bench_decode_two              879                      864                              -15   -1.71%   x 1.02
 serialization::bench_serialization::bench_encode_one              583                      580                               -3   -0.51%   x 1.01
 serialization::bench_serialization::bench_encode_two              970                      1,010                             40    4.12%   x 0.96
 writebatch::bench_writebatch::bench_writebatch_1                  39,876,213               36,526,210                -3,350,003   -8.40%   x 1.09
 writebatch::bench_writebatch::bench_writebatch_1024               14,261,664               13,616,758                  -644,906   -4.52%   x 1.05
 writebatch::bench_writebatch::bench_writebatch_128                14,308,605               13,507,428                  -801,177   -5.60%   x 1.06
 writebatch::bench_writebatch::bench_writebatch_16                 16,245,834               15,099,810                -1,146,024   -7.05%   x 1.08
 writebatch::bench_writebatch::bench_writebatch_2                  27,658,050               25,670,234                -1,987,816   -7.19%   x 1.08
 writebatch::bench_writebatch::bench_writebatch_256                13,912,289               12,997,538                  -914,751   -6.58%   x 1.07
 writebatch::bench_writebatch::bench_writebatch_32                 14,924,121               13,702,075                -1,222,046   -8.19%   x 1.09
 writebatch::bench_writebatch::bench_writebatch_4                  21,888,783               20,360,655                -1,528,128   -6.98%   x 1.08
 writebatch::bench_writebatch::bench_writebatch_512                13,762,431               12,134,391                -1,628,040  -11.83%   x 1.13
 writebatch::bench_writebatch::bench_writebatch_64                 14,229,843               12,149,712                -2,080,131  -14.62%   x 1.17
 writebatch::bench_writebatch::bench_writebatch_8                  18,194,036               15,890,786                -2,303,250  -12.66%   x 1.14
 writebatch::bench_writebatch::bench_writebatch_with_capacity      10,942                   5,934                         -5,008  -45.77%   x 1.84
 writebatch::bench_writebatch::bench_writebatch_without_capacity   10,317                   6,282                         -4,035  -39.11%   x 1.64

benchcmp-tikv

+ cargo benchcmp tikv-before.txt tikv-after.txt
 name                                                                                        tikv-before.txt ns/iter  tikv-after.txt ns/iter  diff ns/iter   diff %  speedup
 coprocessor::codec::batch::column::benches::bench_batch_decode                              27,960                   28,260                           300    1.07%   x 0.99
 coprocessor::codec::batch::column::benches::bench_push_datum_int                            20,659                   20,143                          -516   -2.50%   x 1.03
 coprocessor::codec::batch::column::benches::bench_retain                                    2,638                    2,388                           -250   -9.48%   x 1.10
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_by_vec_clone              46,848                   47,163                           315    0.67%   x 0.99
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_by_vec_push_raw_10bytes   53,429                   53,680                           251    0.47%   x 1.00
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_clone                     16,516                   16,154                          -362   -2.19%   x 1.02
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_clone_10bytes             53,795                   54,003                           208    0.39%   x 1.00
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_clone_and_decode          43,104                   42,998                          -106   -0.25%   x 1.00
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_clone_and_decode_decoded  321                      411                               90   28.04%   x 0.78
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_clone_decoded             312                      411                               99   31.73%   x 0.76
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_clone_naive               81,452                   79,672                        -1,780   -2.19%   x 1.02
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_push_raw_10bytes          55,041                   55,829                           788    1.43%   x 0.99
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_push_raw_4bytes           17,837                   18,226                           389    2.18%   x 0.98
 coprocessor::codec::batch::rows::benches::bench_lazy_batch_column_push_raw_9bytes           17,837                   18,202                           365    2.05%   x 0.98
 raftstore::store::fsm::router::tests::bench_send                                            153                      162                                9    5.88%   x 0.94
 storage::txn::store::benches::bench_fixture_batch_get                                       11,884                   9,438                         -2,446  -20.58%   x 1.26
 storage::txn::store::benches::bench_fixture_get                                             92                       104                               12   13.04%   x 0.88
 storage::txn::store::benches::bench_fixture_scanner                                         448,679                  479,242                       30,563    6.81%   x 0.94
 storage::txn::store::benches::bench_fixture_scanner_next                                    435,325                  433,957                       -1,368   -0.31%   x 1.00
 storage::txn::store::benches::bench_fixture_scanner_scan                                    569,422                  531,086                      -38,336   -6.73%   x 1.07
 util::codec::bytes::tests::bench_decode                                                     2,396                    3,096                            700   29.22%   x 0.77
 util::codec::bytes::tests::bench_decode_inplace                                             1,151                    1,920                            769   66.81%   x 0.60
 util::codec::bytes::tests::bench_decode_inplace_small                                       36                       40                                 4   11.11%   x 0.90
 util::codec::bytes::tests::bench_decode_small                                               70                       69                                -1   -1.43%   x 1.01
 util::codec::bytes::tests::bench_encode                                                     47                       46                                -1   -2.13%   x 1.02
 util::rocksdb::properties::tests::bench_mvcc_properties                                     2,189                    1,999                           -190   -8.68%   x 1.10

Bench result over efficient codec PR:

$ cargo benchcmp base.txt new_version.txt 
 name                                                                              base.txt ns/iter  new_version.txt ns/iter  diff ns/iter   diff %  speedup 
 byte::benches::bench_memcmp_decode_first_asc_large                                90                88                                 -2   -2.22%   x 1.02 
 byte::benches::bench_memcmp_decode_first_asc_large_original                       250               316                                66   26.40%   x 0.79 
 byte::benches::bench_memcmp_decode_first_desc_large                               114               110                                -4   -3.51%   x 1.04 
 byte::benches::bench_memcmp_decode_first_desc_large_original                      258               343                                85   32.95%   x 0.75 
 byte::benches::bench_memcmp_decode_first_in_place_asc_large                       121               168                                47   38.84%   x 0.72 
 byte::benches::bench_memcmp_decode_first_in_place_asc_large_original              166               234                                68   40.96%   x 0.71 
 byte::benches::bench_memcmp_decode_first_in_place_desc_large                      148               185                                37   25.00%   x 0.80 
 byte::benches::bench_memcmp_decode_first_in_place_desc_large_original             188               252                                64   34.04%   x 0.75 
 byte::benches::bench_memcmp_encode_all_asc_large                                  94                89                                 -5   -5.32%   x 1.06 
 byte::benches::bench_memcmp_encode_all_asc_large_naive                            179               160                               -19  -10.61%   x 1.12 
 byte::benches::bench_memcmp_encode_all_asc_large_original                         513               512                                -1   -0.19%   x 1.00 
 byte::benches::bench_memcmp_encode_all_asc_small                                  17                18                                  1    5.88%   x 0.94 
 byte::benches::bench_memcmp_encode_all_asc_small_naive                            26                23                                 -3  -11.54%   x 1.13 
 byte::benches::bench_memcmp_encode_all_desc_large                                 130               134                                 4    3.08%   x 0.97 
 byte::benches::bench_memcmp_encode_all_desc_large_original                        1,275             1,267                              -8   -0.63%   x 1.01 
 byte::benches::bench_memcmp_encode_all_desc_small                                 27                27                                  0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_buffer_decoder                               1                 1                                   0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_byteorder                                    2                 2                                   0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_bytes_buf                                    1                 1                                   0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_number_codec                                 1                 1                                   0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_number_codec_over_slice                      2                 2                                   0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_original                                     2                 2                                   0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_number_codec_large_buffer             5                 6                                   1   20.00%   x 0.83 
 number::benches::bench_decode_varint_normal_number_codec_small_buffer             8                 8                                   0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_number_codec_small_buffer_with_extra  8                 8                                   0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_original_large_buffer                 8                 8                                   0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_original_small_buffer                 8                 8                                   0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_original_small_buffer_with_extra      8                 8                                   0    0.00%   x 1.00 
 number::benches::bench_decode_varint_small_number_codec_large_buffer              2                 3                                   1   50.00%   x 0.67 
 number::benches::bench_decode_varint_small_number_codec_small_buffer              3                 3                                   0    0.00%   x 1.00 
 number::benches::bench_decode_varint_small_original_small_buffer                  2                 2                                   0    0.00%   x 1.00 
 number::benches::bench_encode_u64_le_buffer_encoder_slice                         1                 1                                   0    0.00%   x 1.00 
 number::benches::bench_encode_u64_le_buffer_encoder_vec                           1                 2                                   1  100.00%   x 0.50 
 number::benches::bench_encode_u64_le_byteorder                                    8                 8                                   0    0.00%   x 1.00 
 number::benches::bench_encode_u64_le_number_codec                                 2                 2                                   0    0.00%   x 1.00 
 number::benches::bench_encode_varint_naive                                        9                 10                                  1   11.11%   x 0.90 
 number::benches::bench_encode_varint_number_codec                                 4                 5                                   1   25.00%   x 0.80 
 number::benches::bench_encode_varint_original                                     9                 10                                  1   11.11%   x 0.90 
 number::benches::bench_encode_varint_protobuf                                     46                88                                 42   91.30%   x 0.52 

With 2019-01-01:

$ cargo benchcmp base.txt new_version_2.txt
 name                                                                              base.txt ns/iter  new_version_2.txt ns/iter  diff ns/iter   diff %  speedup 
 byte::benches::bench_memcmp_decode_first_asc_large                                90                90                                    0    0.00%   x 1.00 
 byte::benches::bench_memcmp_decode_first_asc_large_original                       250               327                                  77   30.80%   x 0.76 
 byte::benches::bench_memcmp_decode_first_desc_large                               114               113                                  -1   -0.88%   x 1.01 
 byte::benches::bench_memcmp_decode_first_desc_large_original                      258               334                                  76   29.46%   x 0.77 
 byte::benches::bench_memcmp_decode_first_in_place_asc_large                       121               159                                  38   31.40%   x 0.76 
 byte::benches::bench_memcmp_decode_first_in_place_asc_large_original              166               225                                  59   35.54%   x 0.74 
 byte::benches::bench_memcmp_decode_first_in_place_desc_large                      148               186                                  38   25.68%   x 0.80 
 byte::benches::bench_memcmp_decode_first_in_place_desc_large_original             188               244                                  56   29.79%   x 0.77 
 byte::benches::bench_memcmp_encode_all_asc_large                                  94                89                                   -5   -5.32%   x 1.06 
 byte::benches::bench_memcmp_encode_all_asc_large_naive                            179               168                                 -11   -6.15%   x 1.07 
 byte::benches::bench_memcmp_encode_all_asc_large_original                         513               523                                  10    1.95%   x 0.98 
 byte::benches::bench_memcmp_encode_all_asc_small                                  17                18                                    1    5.88%   x 0.94 
 byte::benches::bench_memcmp_encode_all_asc_small_naive                            26                23                                   -3  -11.54%   x 1.13 
 byte::benches::bench_memcmp_encode_all_desc_large                                 130               120                                 -10   -7.69%   x 1.08 
 byte::benches::bench_memcmp_encode_all_desc_large_original                        1,275             1,267                                -8   -0.63%   x 1.01 
 byte::benches::bench_memcmp_encode_all_desc_small                                 27                28                                    1    3.70%   x 0.96 
 number::benches::bench_decode_u64_le_buffer_decoder                               1                 1                                     0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_byteorder                                    2                 2                                     0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_bytes_buf                                    1                 1                                     0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_number_codec                                 1                 1                                     0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_number_codec_over_slice                      2                 2                                     0    0.00%   x 1.00 
 number::benches::bench_decode_u64_le_original                                     2                 2                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_number_codec_large_buffer             5                 5                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_number_codec_small_buffer             8                 8                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_number_codec_small_buffer_with_extra  8                 8                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_original_large_buffer                 8                 8                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_original_small_buffer                 8                 8                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_normal_original_small_buffer_with_extra      8                 8                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_small_number_codec_large_buffer              2                 3                                     1   50.00%   x 0.67 
 number::benches::bench_decode_varint_small_number_codec_small_buffer              3                 3                                     0    0.00%   x 1.00 
 number::benches::bench_decode_varint_small_original_small_buffer                  2                 2                                     0    0.00%   x 1.00 
 number::benches::bench_encode_u64_le_buffer_encoder_slice                         1                 1                                     0    0.00%   x 1.00 
 number::benches::bench_encode_u64_le_buffer_encoder_vec                           1                 2                                     1  100.00%   x 0.50 
 number::benches::bench_encode_u64_le_byteorder                                    8                 8                                     0    0.00%   x 1.00 
 number::benches::bench_encode_u64_le_number_codec                                 2                 2                                     0    0.00%   x 1.00 
 number::benches::bench_encode_varint_naive                                        9                 9                                     0    0.00%   x 1.00 
 number::benches::bench_encode_varint_number_codec                                 4                 5                                     1   25.00%   x 0.80 
 number::benches::bench_encode_varint_original                                     9                 10                                    1   11.11%   x 0.90 
 number::benches::bench_encode_varint_protobuf                                     46                84                                   38   82.61%   x 0.55 

I investigated why bench_memcmp_decode_first_in_place_asc_large becomes slower, turns out the cause is that Vec clone become slower:

#[bench]
fn bench(b: &mut test::Bencher) {
    let raw = vec![0u8; 1000];
    b.iter(|| {
        test::black_box(test::black_box(&raw).clone());
    });
}

In rustc 1.29 it is 34 ns/iter (+/- 6), in rustc 1.33 it is 121 ns/iter (+/- 21).

I checked the assembly generated by try_decode_first_in_place and bench_memcmp_decode_first_in_place_asc_large, they are mostly the same (only some register changes and some instruction reordering). So the slowness comes from the Rust core.

I opened an issue in rust-lang/rust: rust-lang/rust#57437

Thank you so much for your help @breeswich 🙏