feat: bring back WebGPU

[crowlKats]

Fixes #21514

[sigmaSd]

does this not impact startup time anymore ?

[lino-levan]

Linking this PR for context: servo/core-foundation-rs#608

[hastebrot]

if it impacts startup time then it should be put behind a flag, if this helps. otherwise everything will slowdown including deployctl (deno deploy cli) which currently is 20 times faster than az (azure cli tool) just to output the help text on my machine.

[crowlKats]

We are working on reducing the startup time. A flag would not help. Also, the startup time originally was increased by 3ms, and now is at 1ms increase, and we are working on further reducing it.

[chances]

Requires a wgpu release...

@crowlKats What is missing from wgpu, exactly?

[crowlKats]

Requires a wgpu release...

@crowlKats What is missing from wgpu, exactly?

@chances due to our setup with the wgpu repo, we require the js & rust code to be pulled from a commit that is a release of wgpu.
The blockers for wgpu release from my end is a core-graphics-types release and then a metal-rs release.

[petamoriken]

Added many suggestions related to inspect 🙇

[KnorpelSenf]

This fixes #21514, feel free to update the description

[ry]

@crowlKats can you provide start up time benchmarks? Also max RSS for a blank file?

[crowlKats]

@ry

[crowlKats]

Memory:

[bartlomieju]

On my machine:

hyperfine --warmup 30 -S none './target/release/deno run empty.js' 'deno run empty.js'
Benchmark 1: ./target/release/deno run empty.js
  Time (mean ± σ):      16.3 ms ±   1.0 ms    [User: 10.3 ms, System: 3.9 ms]
  Range (min … max):    15.4 ms …  22.5 ms    189 runs

  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet PC without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.

Benchmark 2: deno run empty.js
  Time (mean ± σ):      14.3 ms ±   0.3 ms    [User: 10.1 ms, System: 4.2 ms]
  Range (min … max):    13.8 ms …  16.4 ms    205 runs

  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet PC without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.

Summary
  'deno run empty.js' ran
    1.14 ± 0.07 times faster than './target/release/deno run empty.js'

/usr/bin/time -l ./target/release/deno run empty.js
        0.03 real         0.01 user         0.00 sys
            38584320  maximum resident set size

/usr/bin/time -l deno run empty.js
        0.02 real         0.01 user         0.00 sys
            38273024  maximum resident set size

[mmastrac]

/bench startup_hello_world,startup_nop_empty,startup_nop_trivial

[denobot]

startup_hello_world

	time	user	system	relative
node	27.887 ms	21.973 ms	7.132 ms	+194.9%
bun	9.456 ms	4.116 ms	6.413 ms	best
deno	15.391 ms	9.995 ms	6.672 ms	+62.8%
deno-baseline	15.152 ms	10.052 ms	6.365 ms	+60.2%

startup_nop_empty

	time	user	system	relative
node	22.526 ms	16.879 ms	6.872 ms	+681.6%
bun	2.882 ms	0.918 ms	1.843 ms	best
deno	15.263 ms	10.067 ms	6.464 ms	+429.6%
deno-baseline	14.955 ms	9.424 ms	6.802 ms	+418.9%

startup_nop_trivial

	time	user	system	relative
node	21.986 ms	16.613 ms	6.58 ms	+92.4%
bun	11.426 ms	5.124 ms	7.47 ms	best
deno	15.137 ms	10.211 ms	6.192 ms	+32.5%
deno-baseline	14.952 ms	9.681 ms	6.539 ms	+30.9%

---

start: Fri, 08 Dec 2023 16:29:41 GMT

id: 1cab344b-ca4a-4422-a124-4a40b7f564e8

server: 093a6a04-1f84-4244-a37b-9a2e40cc5782

[mmastrac]

Comparing to deno 1.38.5 on my machine after a clean && build I see no impact:

09:52 $ hyperfine --warmup 30 -S none './target/release/deno run /tmp/empty.js' './target/release/deno-2 run /tmp/empty.js'
Benchmark 1: ./target/release/deno run /tmp/empty.js
  Time (mean ± σ):      13.3 ms ±   0.5 ms    [User: 9.9 ms, System: 3.4 ms]
  Range (min … max):    12.6 ms …  16.3 ms    212 runs
 
Benchmark 2: ./target/release/deno-2 run /tmp/empty.js
  Time (mean ± σ):      13.2 ms ±   0.4 ms    [User: 9.8 ms, System: 3.4 ms]
  Range (min … max):    12.4 ms …  15.3 ms    233 runs
 
Summary
  './target/release/deno-2 run /tmp/empty.js' ran
    1.01 ± 0.05 times faster than './target/release/deno run /tmp/empty.js'

[bartlomieju]

This is 4th place we define this env var :) we should move it to test_util.ts in a follow up. Can you handle that @crowlKats?

[crowlKats]

sure

[bartlomieju]

What about set()? I think on the web one can override this value

[crowlKats]

nope, not overwritable, however all the navigator getters should have no-op setters. I'll fix that up in a follow up with tests.

[lino-levan]

Nice work Leo (and the whole Deno team). Excited to see this brought back into Deno.