tfjs 3.4.0 regression: wasm causes browser hang #4932
Comments
|
#4796 this might be related , will try to test in my local as well and update , thank you |
|
I test in my local, the problem is gone after turning off the SIMD flag in chrome. We are working on a fix. Thank you for reporting the bug! |
|
@mattsoulanille This maybe related to the recent changes to bazel wasm build, #4769 |
|
any updates on this one? as much as i'd love to use tfjs 3.4.0 due to support of tf2, this is a blocker for me. |
|
No updates yet, but I've been able to reproduce it. I agree with Ping that this is likely related to changes made in #4769, and I'm taking a look. |
|
@mattsoulanille - thanks - 'taking a look' is an update, just wanted to know what's going on as it's a blocker for me. |
Without the linker option `PTHREAD_POOL_SIZE` set, the wasm backend does not create any webworkers and spins forever waiting for them to be created (crashing the browser). This PR sets `PTHREAD_POOL_SIZE` to `Math.min(4, Math.max(1, (navigator.hardwareConcurrency || 1) / 2))`, meaning it will use half of the logical cores available up to a max of 4 (or only 1 if `navigator.hardwareConcurrency` is not available). See [this comment](emscripten-core/emscripten#10263 (comment)) for details on why this works. This option apparently supersedes the value passed to [`pthreadpool_create` in `backend.cc`](https://github.com/tensorflow/tfjs/blob/c9dfebddfa34f531d2f0c363b6ea574a9fe9745d/tfjs-backend-wasm/src/cc/backend.cc#L65-L66), but its current setting should match the value computed in the `.cc` file. Fixes #4932 This PR also fixes a bug where webworkers created by the wasm backend were not removed when `dispose` is called. Fixes #4796. Fixes #4934 as well.
|
@mattsoulanille thanks! given it's a regression that causes hangs, does this warrant spinning up a tfjs service build (e.g., 3.4.1)? |
|
Hi @vladmandic , we are releasing 3.5.0 with this fix today. |
tfjs 3.4.0 was just released and unfortunately wasm module is somewhat broken
to be precise,
tfjs-backend-wasm.wasmandtfjs-backend-wasm-simd.wasmseem to work (although haven't tested in depth)what's definitely broken is
tfjs-backend-wasm-threaded-simd.wasmand symptoms are simple - complete browser hang,
up to a point where browser window is nonresponsive to close requests and requires a hard kill
to reproduce:
WebAssembly SIMD supportin chrome://flagsWebAssembly threads supportin chrome://flagswasmthis used to work in all recent versions of tfjs
environment: tfjs 3.4.0 on windows 10 with chrome 89 or edge 89
The text was updated successfully, but these errors were encountered: