Performance of GCC 11 vs GCC 10 - laurynas-biveinis/unodb GitHub Wiki
commit, GCC 10 vs 11.
-
micro_benchmark_key_prefix: 3% slowdown (unpredictable_prepend_key_prefix<unodb::db>) to 2% speedup (unpredictable_cut_key_prefix<unodb::db>)
--benchmark_filter=unodb::db:
-
micro_benchmark_n4: 3% slowdown (minimal_n4_sequential_insert<unodb::db>/16) to 3% speedup (full_n4_sequential_insert<unodb::db>/4096) -
micro_benchmark_n16: 23% slowdown (n16_sequential_add<unodb::db>/64) to 3% speedup (full_n16_tree_full_scan<unodb::db>/64) -
micro_benchmark_n48: 7% slowdown (full_n48_tree_random_delete<unodb::db>/192) to 3% speedup (full_n48_tree_full_scan<unodb::db>/4096) -
micro_benchmark_n256: 2% slowdown (full_n256_tree_sequential_delete<unodb::db>/512) to 5% speedup (full_n256_tree_full_scan<unodb::db>/32768)
micro_benchmark_n16 24 slowdown, GCC 10:
Performance counter stats for './micro_benchmark_n16 --benchmark_filter=n16_sequential_add<unodb::db>/64 --benchmark_repetitions=9':
13,337.62 msec task-clock # 0.999 CPUs utilized
20 context-switches # 0.001 K/sec
0 cpu-migrations # 0.000 K/sec
7,424 page-faults # 0.557 K/sec
33,382,560,983 cycles # 2.503 GHz (83.34%)
7,738,722,831 stalled-cycles-frontend # 23.18% frontend cycles idle (83.33%)
2,726,154,605 stalled-cycles-backend # 8.17% backend cycles idle (66.65%)
80,896,266,696 instructions # 2.42 insn per cycle
# 0.10 stalled cycles per insn (83.33%)
13,921,969,896 branches # 1043.813 M/sec (83.35%)
26,729,009 branch-misses # 0.19% of all branches (83.34%)
GCC 11:
13,439.69 msec task-clock # 0.999 CPUs utilized
29 context-switches # 0.002 K/sec
0 cpu-migrations # 0.000 K/sec
7,334 page-faults # 0.546 K/sec
33,638,383,411 cycles # 2.503 GHz (83.34%)
7,446,332,576 stalled-cycles-frontend # 22.14% frontend cycles idle (83.33%)
2,691,713,094 stalled-cycles-backend # 8.00% backend cycles idle (66.67%)
80,470,158,602 instructions # 2.39 insn per cycle
# 0.09 stalled cycles per insn (83.33%)
13,736,250,817 branches # 1022.066 M/sec (83.33%)
32,313,073 branch-misses # 0.24% of all branches (83.32%)
--benchmark_filter=unodb::olc_db:
-
micro_benchmark_n4: 3% slowdown (shrink_node16_to_n4_sequentially<unodb::olc_db>/25) to 4% speedup (full_n4_sequential_insert<unodb::olc_db>/100) -
micro_benchmark_n16: 2% slowdown (minimal_n16_tree_full_scan<unodb::olc_db>/512) to 9% speedup (n16_sequential_add<unodb::olc_db>/64) -
micro_benchmark_n48: 12% slowdown (n48_random_add<unodb::olc_db>/8) to 1% speedup (grow_n16_to_n48_sequentially<unodb::olc_db>/8) -
micro_benchmark_n256: 3% slowdown (grow_n48_to_n256_randomly<unodb::olc_db>/8) to 6% (full_n256_tree_random_delete<unodb::olc_db>/32768)