avoid buffering the output twice when writing to lepton file

[mcroomp]

Track the carry overflow to we can immediately write out bytes to output stream and not have to worry about adjusting for carry.

[Melirius]

Idea is very nice, but performance is bad:

main 82d6547

2024-11-15T21:03:34.699Z INFO  [lepton_jpeg::structs::lepton_file_writer] compressing to Lepton format
2024-11-15T21:03:35.173Z INFO  [lepton_jpeg::structs::lepton_file_writer] Number of threads: 8
2024-11-15T21:03:36.610Z INFO  [lepton_jpeg::structs::lepton_file_writer] worker threads 9632ms of CPU time in 1436ms of wall time
2024-11-15T21:03:36.610Z INFO  [lepton_jpeg::structs::lepton_file_writer] decompressing to verify contents
2024-11-15T21:03:38.216Z INFO  [lepton_jpeg_util] compressed input 22171278, output 17324076 bytes (compression = 28.0%)
2024-11-15T21:03:38.216Z INFO  [lepton_jpeg_util] Main thread CPU: 3517ms, Worker thread CPU: 20710 ms, walltime: 3517 ms

 Performance counter stats for 'taskset -c 10 nice -n -20 target/release/lepton_jpeg_util images/img_52MP_7k.jpg images/img_52MP_7k2.lep':

       871 003 064      cache-references                                                        (41,51%)
        86 439 870      cache-misses                     #    9,92% of all cache refs           (41,52%)
    15 795 148 123      cycles                                                                  (41,72%)
       834 608 725      ic_fetch_stall.ic_stall_back_pressure                                        (41,87%)
     1 014 898 692      stalled-cycles-frontend          #    6,43% frontend cycles idle        (42,14%)
    39 107 914 014      instructions                     #    2,48  insn per cycle            
                                                  #    0,03  stalled cycles per insn     (42,43%)
     4 396 207 542      branch-instructions                                                     (42,50%)
       162 547 993      branch-misses                    #    3,70% of all branches             (42,32%)
     5 434 637 214      ic_fetch_stall.ic_stall_any                                             (42,14%)
        45 374 631      ic_fetch_stall.ic_stall_dq_empty                                        (41,99%)
        72 770 962      l2_cache_misses_from_ic_miss                                            (41,75%)
     2 149 017 778      l2_latency.l2_cycles_waiting_on_fills                                        (41,47%)
           184 005      faults                                                                
                 1      migrations                                                            

       3,551824055 seconds time elapsed

       3,224073000 seconds user
       0,324906000 seconds sys

this PR

2024-11-16T11:07:51.300Z INFO  [lepton_jpeg::structs::lepton_file_writer] compressing to Lepton format
2024-11-16T11:07:51.773Z INFO  [lepton_jpeg::structs::lepton_file_writer] Number of threads: 8
2024-11-16T11:07:53.262Z INFO  [lepton_jpeg::structs::lepton_file_writer] worker threads 9925ms of CPU time in 1488ms of wall time
2024-11-16T11:07:53.262Z INFO  [lepton_jpeg::structs::lepton_file_writer] decompressing to verify contents
2024-11-16T11:07:54.920Z INFO  [lepton_jpeg_util] compressed input 22171278, output 17324076 bytes (compression = 28.0%)
2024-11-16T11:07:54.920Z INFO  [lepton_jpeg_util] Main thread CPU: 3620ms, Worker thread CPU: 21327 ms, walltime: 3620 ms

 Performance counter stats for 'taskset -c 10 nice -n -20 target/release/lepton_jpeg_util images/img_52MP_7k.jpg images/img_52MP_7k2.lep':

       865 085 762      cache-references                                                        (41,79%)
        81 144 742      cache-misses                     #    9,38% of all cache refs           (42,09%)
    16 326 000 065      cycles                                                                  (42,24%)
       968 571 728      ic_fetch_stall.ic_stall_back_pressure                                        (42,49%)
     1 047 639 608      stalled-cycles-frontend          #    6,42% frontend cycles idle        (42,35%)
    39 674 257 535      instructions                     #    2,43  insn per cycle            
                                                  #    0,03  stalled cycles per insn     (42,22%)
     4 364 055 943      branch-instructions                                                     (42,12%)
       165 140 691      branch-misses                    #    3,78% of all branches             (41,80%)
     5 544 312 995      ic_fetch_stall.ic_stall_any                                             (41,69%)
        39 725 098      ic_fetch_stall.ic_stall_dq_empty                                        (41,43%)
        66 843 172      l2_cache_misses_from_ic_miss                                            (41,31%)
     2 165 667 711      l2_latency.l2_cycles_waiting_on_fills                                        (41,54%)
           183 994      faults                                                                
                 1      migrations                                                            

       3,654640632 seconds time elapsed

       3,323893000 seconds user
       0,328890000 seconds sys

[Melirius]

c5b6938 helps, but still much slower than main

2024-11-20T20:56:19.974Z INFO  [lepton_jpeg::structs::lepton_file_writer] compressing to Lepton format
2024-11-20T20:56:20.453Z INFO  [lepton_jpeg::structs::lepton_file_writer] Number of threads: 8
2024-11-20T20:56:21.944Z INFO  [lepton_jpeg::structs::lepton_file_writer] worker threads 9985ms of CPU time in 1490ms of wall time
2024-11-20T20:56:21.944Z INFO  [lepton_jpeg::structs::lepton_file_writer] decompressing to verify contents
2024-11-20T20:56:23.567Z INFO  [lepton_jpeg_util] compressed input 22171278, output 17324076 bytes (compression = 28.0%)
2024-11-20T20:56:23.567Z INFO  [lepton_jpeg_util] Main thread CPU: 3593ms, Worker thread CPU: 21155 ms, walltime: 3593 ms

 Performance counter stats for 'taskset -c 10 nice -n -20 target/release/lepton_jpeg_util images/img_52MP_7k.jpg images/img_52MP_7k2.lep':

       836 854 716      cache-references                                                        (41,97%)
        75 722 326      cache-misses                     #    9,05% of all cache refs           (41,79%)
    16 120 935 933      cycles                                                                  (41,65%)
       956 121 210      ic_fetch_stall.ic_stall_back_pressure                                        (41,42%)
     1 106 567 900      stalled-cycles-frontend          #    6,86% frontend cycles idle        (41,53%)
    39 302 320 657      instructions                     #    2,44  insn per cycle            
                                                  #    0,03  stalled cycles per insn     (41,52%)
     4 328 449 995      branch-instructions                                                     (41,74%)
       166 308 719      branch-misses                    #    3,84% of all branches             (42,04%)
     5 614 369 841      ic_fetch_stall.ic_stall_any                                             (42,26%)
        42 227 531      ic_fetch_stall.ic_stall_dq_empty                                        (42,33%)
        62 449 857      l2_cache_misses_from_ic_miss                                            (42,46%)
     2 284 954 295      l2_latency.l2_cycles_waiting_on_fills                                        (42,25%)
           183 975      faults                                                                
                 1      migrations                                                            

       3,628224562 seconds time elapsed

       3,283399000 seconds user
       0,342937000 seconds sys