Performance of LLVM 12 vs LLVM 11, take 2

commit

Filtered for unodb::db:

• micro_benchmark_key_prefix: 2% slowdown (unpredictable_cut_key_prefix) to 3% speedup (unpredictable_prepend_key_prefix)
• micro_benchmark_n4: 1% slowdown (full_n4_to_minimal_sequential_delete/32768) to 7% speedup (n4_full_scan/100)
• micro_benchmark_n16: 11% slowdown (shrink_n48_to_n16_randomly/64) to 14% speedup (full_n16_tree_random_delete/64)
• micro_benchmark_n48: 3% slowdown (grow_n16_to_n48_sequentially/64) to 4% speedup (full_n48_tree_sequential_delete/192)
• micro_benchmark_n256: 5% slowdown (full_n256_tree_random_delete/192) to 6% speedup (full_n256_tree_random_delete/196608)

Filtered for unodb::olc_db:

• micro_benchmark_key_prefix: 3% slowdown (unpredictable_leaf_key_prefix_split) to 3% speedup (unpredictable_prepend_key_prefix)
• micro_benchmark_n4: 7% slowdown (full_n4_sequential_delete/32768) to 0% slowdown (full_n4_random_insert/32768)
• micro_benchmark_n16: 5% slowdown (full_n16_tree_random_delete/246000) to 3% speedup (shrink_n48_to_n16_randomly/64)
• micro_benchmark_n48: 4% slowdown (shrink_n256_to_n48_randomly/512) to 3% speedup (full_n48_tree_sequential_delete/4096)
• micro_benchmark_n256: 0% slowdown (n256_sequential_add/64) to 12% speedup (grow_n48_to_n256_randomly/8)