term_esc_rgb() dominates profiles

[dankamongmen]

Running a profile against the (optimized) build of notcurses, I get this output in notcurses-view:

-   27.05%     0.00%  notcurses-view  [unknown]                 [.] 0x000055cc0▒
   - 0x55cc08c4e040                                                            ◆
      + 22.02% fprintf                                                         ▒
        3.27% notcurses_render                                                 ▒
        0.54% _IO_default_xsputn                                               ▒
-   23.06%     1.01%  notcurses-view  libc-2.29.so              [.] fprintf    ▒
   - 22.04% fprintf                                                            ▒
        13.47% __vfprintf_internal                                             ▒
        3.73% _IO_default_xsputn                                               ▒
        2.40% _itoa_word                                                       ▒
        1.56% __strchrnul_avx2                                                 ▒
   - 0.93% 0x55cc08c4e040                                                      ▒
        0.82% fprintf                                                          ▒
-   15.03%    14.07%  notcurses-view  libc-2.29.so              [.] __vfprintf_▒
   + 13.09% 0x55cc08c4e040                                                     ▒
     0.97% __vfprintf_internal                                                 ▒
-   12.31%     0.01%  notcurses-view  [kernel.vmlinux]          [k] entry_SYSCA▒
   - 12.30% entry_SYSCALL_64_after_hwframe                                     ▒
      + 12.27% do_syscall_64                                                   ▒
-   12.27%     0.11%  notcurses-view  [kernel.vmlinux]          [k] do_syscall_▒
   - 12.16% do_syscall_64                                                      ▒
      + 11.86% __x64_sys_write                                                

absolutely dominated by fprintf(), which is entirely due to term_esc_rgb(). if we can get that sped up, we're going to recover a hell of a lot of cycles.

for notcurses-demo, it was even more pronounced.

[dankamongmen]

Yeah, dropping the fprintf() entirely boosts FPS by 36%! w0000000000000000t. alright let's do this.

[dankamongmen]

Reducing it to a lookup table and some fputs() boosted us from 2300 FPS to 2600, and combining all those fputs() boosted us further to 2750. That's almost a 20% increase, not bad at all for an hours' work :D :D :D. Just broke 2800! (xfce4-terminal 80x70)

[dankamongmen]

[schwarzgerat](0) $ ./notcurses-demo -p ../data/ -c
Term: xterm with direct-color indexing

122896 renders, 14.1s total (7.1e-05s min, 0.00478s max, 0.000115s avg 8725.1 fps)
171.27MiB total (0.00B min, 118.37KiB max, 1.43KiB avg)
0 failed renders
Emits/elides: def 72976/18641 fg 4264137/1974870 bg 4815725/1493123
 Elide rates: 20.35% 31.65% 23.67%
Cells emitted: 6422251 elided: 373710512 (98.31%)

        total│frames│output(B)│rendering│%r│    FPS║
══╤═╤════════╪══════╪═════════╪═════════╪══╪═══════╣
 0│i│   2.00s│   129│   1.77Mi│ 116.18ms│ 5│ 1110.3║
 1│x│   5.03s│   302│   9.05Mi│ 308.93ms│ 6│  977.6║
 2│e│   3.11s│   171│   8.91Mi│ 340.14ms│10│  502.7║
 3│t│  17.05s│   633│   1.70Mi│ 186.49ms│ 1│ 3394.1║
 4│c│   3.38s│    33│ 630.83Ki│  33.72ms│ 0│  978.4║
 5│g│   1.31s│   768│  74.35Mi│    1.20s│91│  638.7║
 6│s│   5.08s│   321│ 424.43Ki│  63.16ms│ 1│ 5082.0║
 7│w│  10.00s│119187│   7.28Mi│    9.23s│92│12909.5║
 8│u│  21.07s│    63│ 550.11Ki│  34.95ms│ 0│ 1802.3║
 9│b│   1.01s│    10│ 141.67Ki│   7.44ms│ 0│ 1342.5║
10│v│  19.60s│   753│  52.67Mi│    1.96s│ 9│  384.1║
11│l│   1.04s│    74│ 367.93Ki│  35.14ms│ 3│ 2105.3║
12│p│   5.00s│    18│ 337.71Ki│  16.25ms│ 0│ 1107.4║
13│o│  13.01s│   434│  13.09Mi│ 547.58ms│ 4│  792.6║
══╧═╧════════╪══════╪═════════╪═════════╪══╪═══════╝
      107.75s│122896│ 171.27Mi│   14.08s│13│
[schwarzgerat](0) $

alacritty 91x34, hot shit!