Performance of LLVM 13 vs LLVM 12 - laurynas-biveinis/unodb GitHub Wiki

commit

Filtered for unodb::db:

  • micro_benchmark_key_prefix: 2% slowdown (unpredictable_prepend_key_prefix) to 2% speedup (unpredictable_cut_key_prefix)
  • micro_benchmark_n4: 7% slowdown (n4_full_scan/4096) to 6% speedup (full_n4_sequential_delete/512)
  • micro_benchmark_n16: 4% slowdown (grow_n4_to_n16_randomly/20) to 13% speedup (shrink_n48_to_n16_randomly/64)
  • micro_benchmark_n48: 3% slowdown (minimal_n48_tree_full_scan/8) to 6% speedup (n48_sequential_add/2)
  • micro_benchmark_n256: 27% slowdown (grow_n48_to_n256_randomly/8) to 5% speedup (full_n256_tree_sequential_delete/192)

Filtered for unodb::olc_db:

  • micro_benchmark_key_prefix: 4% slowdown (unpredictable_get_shared_length) to 0% slowdown (unpredictable_cut_key_prefix)
  • micro_benchmark_n4: 4% slowdown (full_n4_sequential_delete/100) to 0% slowdown (full_n4_to_minimal_random_delete/32768)
  • micro_benchmark_n16: 5% slowdown (full_n16_tree_sequential_delete/512) to 1% speedup (shrink_n48_to_n16_randomly/64)
  • micro_benchmark_n48: 4% slowdown (n48_sequential_add/512) to 3% speedup (shrink_n256_to_n48_randomly/512)
  • micro_benchmark_n256: 14% slowdown (grow_n48_to_n256_randomly/8) to 1% speedup (grow_n48_to_n256_sequentially/2)