Optimization takes a long time!!

[themixup]

During the Reelase build process, compilation is quick but optimization takes a long time. But  Debug type is fast.

How can I identify which parts are taking the most time during optimization?

* emcmake cmake -S ../ -DCMAKE_BUILD_TYPE=Release

* build time : 30minutes more...

* cmakelist build link option

"SHELL:-s EXPORTED_FUNCTIONS='[\"_malloc\", \"_free\"]'"
"SHELL:-s EXPORTED_RUNTIME_METHODS='[\"ccall\", \"cwrap\", \"FS\", \"stringToUTF16\", \"UTF8ToString\", \"lengthBytesUTF8\", \"allocateUTF8\"]'"
"SHELL:-s NO_EXIT_RUNTIME=1"
"SHELL:-s ELIMINATE_DUPLICATE_FUNCTIONS=1"
"SHELL:-s ERROR_ON_UNDEFINED_SYMBOLS=1"
"SHELL:-s ASSERTIONS=2"
"SHELL:-s FORCE_FILESYSTEM=1"
"SHELL:-s MAX_WEBGL_VERSION=2"
"SHELL:-s WASM=1"
"SHELL:-v"
"SHELL:--bind"
"SHELL:-s ALLOW_MEMORY_GROWTH=1"
"SHELL:-s ALLOW_TABLE_GROWTH=1"
"SHELL:-s TOTAL_MEMORY=1024MB"
"SHELL:-s STACK_SIZE=5MB"

*  my setting

cpu :i7-13700
ram :32GB
ssd :500GB

emsdk : 3.1.53

[Sam Clegg]

You can run your link command with `EMPROFILE=2` set in the environment and emscripten will print a breakdown of the various link steps.

If you want even more details you can use `EMPROFILE=1` to produce a trace that can be rendered as html.

See https://emscripten.org/docs/optimizing/Profiling-Toolchain.html

cheers,

sam

--
You received this message because you are subscribed to the Google Groups "emscripten-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to emscripten-disc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/emscripten-discuss/1b28b3d5-fdc3-432d-bf07-0010570cb2acn%40googlegroups.com.

[themixup]

Thank you. sam.

I changed the options as instructed and did the profiling. After checking the results, I identified sections where time was being consumed. 

Could *._emval_await be affecting this? When compared to other projects, there is a difference in *._emval_await.

  

profile.html

{
    "pid": 110007,
    "subprocessPid": 110007,
    "op": "spawn",
    "targetPid": 110114,
    "time": 1720413197.1,
    "cmdLine": [
      "/home/test/dev/emsdk/upstream/bin/wasm-opt",
      "--strip-target-features",
      "--post-emscripten",
      "-O3",
      "--low-memory-unused",
      "--asyncify",
      "--pass-arg=asyncify-asserts",
      "--pass-arg=asyncify...@env.invoke_*,env.__asyncjs__*,*.fd_sync,*.emscripten_promise_await,*.emscripten_idb_load,*.emscripten_idb_store,*.emscripten_idb_delete,*.emscripten_idb_exists,*.emscripten_idb_clear,*.emscripten_idb_load_blob,*.emscripten_idb_store_blob,*.emscripten_sleep,*.emscripten_wget_data,*.emscripten_scan_registers,*.emscripten_lazy_load_code,*._load_secondary_module,*.emscripten_fiber_swap,*.SDL_Delay,*._emval_await",
      "--zero-filled-memory",
      "--pass-arg=directize-initial-contents-immutable",
      "test.wasm",
      "-o",
      "test.wasm",
      "--mvp-features",
      "--enable-mutable-globals",
      "--enable-sign-ext"
    ]
  },
  {
    "pid": 110007,
    "subprocessPid": 110007,
    "op": "wait",
    "targetPid": 110114,
    "time": 1720413197.1
  },
  {
    "pid": 110007,
    "subprocessPid": 110007,
    "op": "finish",
    "targetPid": 110114,
    "time": 1720416042.543,
    "returncode": 0
  },

2024년 7월 6일 토요일 오전 1시 36분 27초 UTC+9에 s...@google.com님이 작성:

[Sam Clegg]

It seems that what is slow here is the `--asyncify` pass in wasm-opt.   IIUC that pass will be slower or faster depending on how much of your program need to be asyncified. 

If you are using embind with asyncify I don't think you can avoid the need to asyncify calls to _emval_await so I'm not sure what you can do to try to speed up that pass.  Others may have some tricks they know about?

As an alternative to asyncify we also have `-sJSPI` which doesn't require this pass, but is also not yet available in stable browsers.

cheers,

sam

To view this discussion on the web visit https://groups.google.com/d/msgid/emscripten-discuss/f6b39d3a-6d4c-425d-83f7-561f891dcf3dn%40googlegroups.com.

[Alon Zakai]

Asyncify can definitely be very slow in some cases, yeah, in large programs.

If you can't use JSPI in production yet you can at least use it in local testing at least, for much faster iteration.

To view this discussion on the web visit https://groups.google.com/d/msgid/emscripten-discuss/CAL_va293xnbBEz%2BZ_Dx5%2BS0K%2BuNK9Ky508AMzr%3D9orQH1FzNFw%40mail.gmail.com.

[themixup]

Thank you!!
After removing the Asyncify option, the build time and size have been significantly reduced!!

2024년 7월 9일 화요일 오전 4시 40분 59초 UTC+9에 s...@google.com님이 작성: