Jun Wu 40a88364be indexedlog: replace div with shr to make checksum faster ... Summary: Spot `div` slowness using Linux's `perf` tool. | Disassembly of section .text: | | 0000000000018990 <indexedlog::checksum_table::ChecksumTable::check_range>: | _ZN10indexedlog14checksum_table13ChecksumTable11check_range17h2303c96b1e035e20E(): 1.36 | push %rax 0.18 | mov %rdx,%r8 0.54 | mov $0x1,%cl | test %r8,%r8 | je 60 0.54 | add %rsi,%r8 0.72 | cmp 0x30(%rdi),%r8 | ja 64 0.27 | mov 0x28(%rdi),%r9 0.27 | test %r9,%r9 | je 6a 0.36 | add $0xffffffffffffffff,%r8 0.18 | xor %edx,%edx 0.45 | mov %rsi,%rax 0.36 | div %r9 43.72 | mov %rax,%rsi | xor %edx,%edx | mov %r8,%rax 0.18 | div %r9 42.82 | add $0x1,%rax 0.09 | cmp %rax,%rsi | jae 60 2.17 | cmpq $0x0,0x60(%rdi) | je 78 | mov 0x50(%rdi),%rcx | cmpb $0x0,(%rcx) 1.63 | sete %cl 0.18 | xchg %ax,%ax |50: test $0x1,%cl | je 64 0.45 | add $0x1,%rsi 0.81 | mov $0x1,%cl 0.09 | cmp %rax,%rsi | jb 50 |60: mov %ecx,%eax | pop %rcx 2.62 | retq |64: xor %ecx,%ecx | mov %ecx,%eax | pop %rcx | retq |6a: lea panic_loc.a.llvm.9800112514578621117,%rdi | callq core::panicking::panic | ud2 |78: lea panic_bounds_check_loc.7.llvm.9800112514578621117,%rdi | xor %esi,%esi | xor %edx,%edx | callq core::panicking::panic_bounds_check | ud2 Change `chunk_size` to `chunk_size_log`. Replace `div` with `shr` to make it significantly faster: Before: index lookup (memory) 1.118 ms index lookup (disk, no verify) 2.078 ms index lookup (disk, verified) 7.687 ms After: index lookup (memory) 1.066 ms index lookup (disk, no verify) 1.992 ms index lookup (disk, verified) 3.591 ms Reviewed By: DurhamG, markbt Differential Revision: D7554992 fbshipit-source-id: c24189ced722d880af6ca0d64967eb762363d9e3		2018-04-17 18:54:39 -07:00
..
benches	indexedlog: use OpenOptions	2018-04-13 21:51:46 -07:00
src	indexedlog: replace div with shr to make checksum faster	2018-04-17 18:54:39 -07:00
Cargo.toml	indexedlog: separate benchmarks	2018-04-13 21:51:42 -07:00