Optimize unchecked indexing into chunks and chunks_mut by the8472 · Pull Request...

the8472 commented 21 days ago

# BEFORE

$ rustc +nightly -Copt-level=3 -Ccodegen-units=1 -Clto=fat chunks.rs
$ perf stat ./chunks

 Performance counter stats for './chunks':

          3,177.03 msec task-clock                #    1.000 CPUs utilized
                 4      context-switches          #    0.001 K/sec
                 0      cpu-migrations            #    0.000 K/sec
           984,006      page-faults               #    0.310 M/sec
    13,092,199,322      cycles                    #    4.121 GHz                      (83.29%)
       384,543,475      stalled-cycles-frontend   #    2.94% frontend cycles idle     (83.35%)
     7,414,280,722      stalled-cycles-backend    #   56.63% backend cycles idle      (83.38%)
    50,493,980,662      instructions              #    3.86  insn per cycle
                                                  #    0.15  stalled cycles per insn  (83.29%)
     6,625,375,297      branches                  # 2085.396 M/sec                    (83.38%)
         3,087,652      branch-misses             #    0.05% of all branches          (83.31%)

       3.178079469 seconds time elapsed

       2.327156000 seconds user
       0.762041000 seconds sys

# AFTER

$ ./build/x86_64-unknown-linux-gnu/stage1/bin/rustc -Copt-level=3 -Ccodegen-units=1 -Clto=fat chunks.rs
$ perf stat ./chunks

 Performance counter stats for './chunks':

          2,705.76 msec task-clock                #    1.000 CPUs utilized
                 4      context-switches          #    0.001 K/sec
                 0      cpu-migrations            #    0.000 K/sec
           984,005      page-faults               #    0.364 M/sec
    11,156,763,039      cycles                    #    4.123 GHz                      (83.26%)
       342,198,882      stalled-cycles-frontend   #    3.07% frontend cycles idle     (83.37%)
     6,486,263,637      stalled-cycles-backend    #   58.14% backend cycles idle      (83.37%)
    40,553,476,617      instructions              #    3.63  insn per cycle
                                                  #    0.16  stalled cycles per insn  (83.37%)
     6,668,429,113      branches                  # 2464.532 M/sec                    (83.37%)
         3,099,636      branch-misses             #    0.05% of all branches          (83.26%)

       2.706725288 seconds time elapsed

       1.782083000 seconds user
       0.848424000 seconds sys