Apologies for jumping in. I tried compiling my home computer emulator code with -mtail-call to see if it makes a difference, but I have a hard time starting Chrome (or Chrome Dev) into a mode where it accepts the new instructions. On the console I get:
~~~ Uncaught (in promise) RuntimeError: Aborted(CompileError: WebAssembly.instantiate(): Compiling function #221 failed: Invalid opcode 0x12 (enable with --experimental-wasm-return_call) @+126726). Build with -s ASSERTIONS=1 for more info. ~~~ I tried (on Mac): - enabling "Experiments" in Chrome Dev and then: - starting the Chrome Dev executable with --experimental-wasm-return_call fom the command line option, no effect - after googling (and finding https://leaningtech.com/extreme-webassembly-2-the-sad-state-of-webassembly-tail-calls/), starting the Chrome Dev exe with --js-flags=”--experimental-wasm-return-call", also no effect Any ideas (or links) grealy appreciated :) On Thursday, 25 November 2021 at 02:05:15 UTC+1 Soeren Balko wrote: > I found some faint hints that Clang has -mdisable-tail-calls (see > disable-tail-calls.c > | searchcode > <https://searchcode.com/file/307703938/src/llvm-project/clang/test/CodeGen/disable-tail-calls.c/> > ) > , so I figured I may try -menable-tail-calls. Not sure if that did > anything, but I ended up with a very small difference (6 bytes) in file > sizes for my FFmpeg build (42789880 vs 42789886 bytes). Not sure where that > comes from. That being said, the aforementioned ~1% performance difference > could indeed be noise. I'll do some more tests to see what's going on. I > may have to disassemble the wasm binary to figure out if there really is > anything as a result of the -menable-tail-calls flag. > > > > On Wednesday, November 24, 2021 at 4:49:26 PM UTC+10 [email protected] > wrote: > >> Hmm, I tried finding documentation on -menable-tail-call or an >> implementation of it but couldn't, so then I tried using it to see if it >> did anything. I built the in-tree sqlite benchmark at -O2 with no extra >> flags, with -mtail-call, and with -menable-tail-call. The version built >> with -mtail-call contained 53 instances of `return_call` or >> `return_call_indirect` instructions, but the version with no flags and the >> version with -menable-tail-call were identical. I also confirmed that >> Emscripten does not warn or error out on unknown flags passed at link time, >> so I can only conclude that -menable-tail-call is not a real flag. The >> performance difference you saw must have been noise. >> >> On Wed, Nov 24, 2021 at 12:00 AM Soeren Balko <[email protected]> >> wrote: >> >>> Thanks Thomas - haven't run into any issues. There wasn't any measurable >>> performance uplift when using -mtail-call though, which is probably >>> because my workload (FFmpeg) may not use it much. >>> >>> That being said, I previously played around with the -menable-tail-call >>> linker flag. And unlike -mtail-call, it does two things: firstly, it >>> does produce a small (~1%), but consistent performance uplift. At the same >>> time, it doesn't seem to make use of the new WebAssembly opcodes (didn't >>> require me to enable those in V8). Not sure what -menable-tail-call >>> actually does, but it seems different from -mtail-call. Also, it cannot >>> be used as a compiler flag, ie. only the linker will accept it. >>> >>> Any ideas? >>> >>> Soeren >>> >>> On Wednesday, November 24, 2021 at 10:57:46 AM UTC+10 [email protected] >>> wrote: >>> >>>> Hi Soeren, >>>> >>>> The compiler flag -mtail-call is what you want. This support has been >>>> in upstream LLVM for a while, but it hasn't been used very much, so if you >>>> run into bugs with it, please let us know! >>>> >>>> Best, >>>> >>>> Thomas >>>> >>>> On Tue, Nov 23, 2021 at 6:14 PM Soeren Balko <[email protected]> >>>> wrote: >>>> >>>>> I was wondering about the state of WebAssembly tail call support in >>>>> Emscripten. It's available as an experimental V8 feature in Chrome and >>>>> will >>>>> hopefully be mainstreamed in the not-too distant future. >>>>> >>>>> Couldn't find much information on how to build with tail call support >>>>> in Emscripten, but I suspect that adding the -menable-tail-call >>>>> linker flag does the trick? Or is there more to it? >>>>> >>>>> @Alon - can you please confirm? >>>>> >>>>> Thank you! >>>>> Soeren >>>>> >>>>> -- >>>>> 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 [email protected]. >>>>> To view this discussion on the web visit >>>>> https://groups.google.com/d/msgid/emscripten-discuss/bdcc8072-b7a0-454c-bd05-68b0ae780c77n%40googlegroups.com >>>>> >>>>> <https://groups.google.com/d/msgid/emscripten-discuss/bdcc8072-b7a0-454c-bd05-68b0ae780c77n%40googlegroups.com?utm_medium=email&utm_source=footer> >>>>> . >>>>> >>>> -- >>> 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 [email protected]. >>> >> To view this discussion on the web visit >>> https://groups.google.com/d/msgid/emscripten-discuss/f697836d-10ef-41a0-85bb-565698302427n%40googlegroups.com >>> >>> <https://groups.google.com/d/msgid/emscripten-discuss/f697836d-10ef-41a0-85bb-565698302427n%40googlegroups.com?utm_medium=email&utm_source=footer> >>> . >>> >> -- 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 [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/emscripten-discuss/236b54bc-4e50-498c-a6df-5656c34eacb9n%40googlegroups.com.
