Effect of removing redundant nullptr checks in deleters - laurynas-biveinis/unodb GitHub Wiki
baseline commit, patch commit, filtered with --benchmark_filter="(shrink|delete).*unodb::db"
-
micro_benchmark_node4: 2% slowdown (full_node4_to_minimal_sequential_delete<unodb::db>/65532) to 11% speedup (shrink_node16_to_node4_randomly<unodb::db>/64) -
micro_benchmark_node16: 3% slowdown (full_node16_tree_random_delete<unodb::db>/32768) to 7% speedup (shrink_node48_to_node16_randomly<unodb::db>/16383) -
micro_benchmark_node48: 3% slowdown (full_node48_tree_sequential_delete<unodb::db>/192) to 12% speedup (shrink_node256_to_node48_randomly<unodb::db>/64) -
micro_benchmark_node256: 1% slowdown
perf stat for shrink_node256_to_node48_randomly<unodb::db>/64:
baseline:
$ perf stat ./micro_benchmark_node48 --benchmark_filter="shrink_node256_to_node48_randomly<unodb::db>/64"
2021-05-06T05:51:49+02:00
Running ./micro_benchmark_node48
Run on (8 X 3800 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x4)
L1 Instruction 32 KiB (x4)
L2 Unified 256 KiB (x4)
L3 Unified 8192 KiB (x1)
Load Average: 0.25, 0.49, 0.51
----------------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
----------------------------------------------------------------------------------------------------------
shrink_node256_to_node48_randomly<unodb::db>/64 14.2 us 14.0 us 49952 items_per_second=4.56595M/s size=475.906k
Performance counter stats for './micro_benchmark_node48 --benchmark_filter=shrink_node256_to_node48_randomly<unodb::db>/64':
18,349.19 msec task-clock # 0.999 CPUs utilized
170 context-switches # 0.009 K/sec
0 cpu-migrations # 0.000 K/sec
344 page-faults # 0.019 K/sec
69,793,580,619 cycles # 3.804 GHz (83.33%)
12,757,622,496 stalled-cycles-frontend # 18.28% frontend cycles idle (83.32%)
3,666,650,080 stalled-cycles-backend # 5.25% backend cycles idle (66.68%)
182,994,791,626 instructions # 2.62 insn per cycle
# 0.07 stalled cycles per insn (83.35%)
32,695,626,969 branches # 1781.856 M/sec (83.35%)
71,736,753 branch-misses # 0.22% of all branches (83.32%)
18.368685049 seconds time elapsed
18.254159000 seconds user
0.096011000 seconds sys
patch:
$ perf stat ./micro_benchmark_node48 --benchmark_filter="shrink_node256_to_node48_randomly<unodb::db>/64"
2021-05-06T05:52:20+02:00
Running ./micro_benchmark_node48
Run on (8 X 3800 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x4)
L1 Instruction 32 KiB (x4)
L2 Unified 256 KiB (x4)
L3 Unified 8192 KiB (x1)
Load Average: 0.39, 0.50, 0.51
----------------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
----------------------------------------------------------------------------------------------------------
shrink_node256_to_node48_randomly<unodb::db>/64 12.6 us 12.4 us 56269 items_per_second=5.14701M/s size=475.906k
Performance counter stats for './micro_benchmark_node48 --benchmark_filter=shrink_node256_to_node48_randomly<unodb::db>/64':
20,263.39 msec task-clock # 0.999 CPUs utilized
42 context-switches # 0.002 K/sec
0 cpu-migrations # 0.000 K/sec
344 page-faults # 0.017 K/sec
77,066,687,284 cycles # 3.803 GHz (83.32%)
13,001,846,782 stalled-cycles-frontend # 16.87% frontend cycles idle (83.34%)
3,967,510,685 stalled-cycles-backend # 5.15% backend cycles idle (66.68%)
201,972,140,130 instructions # 2.62 insn per cycle
# 0.06 stalled cycles per insn (83.34%)
36,100,564,556 branches # 1781.566 M/sec (83.34%)
64,862,577 branch-misses # 0.18% of all branches (83.32%)
20.281543370 seconds time elapsed
20.187859000 seconds user
0.076014000 seconds sys