Looks like a case of one particular CPU architecture experiencing surprising slowdowns for some particular instruction or pattern. Depending on how widespread the situation is, and on availability+cost of software-side workarounds, it's not immediately clear whether we want to do anything about it on the V8 side.
To narrow it down further: - you could inspect the generated machine code (--print-wasm-code should do the trick) and/or use low-level profiling (such as Linux' perf tool) to figure out what exactly the trigger is. If the hypothesis is right that it's some particular pattern, then it should be possible to write a small standalone C/C++ program that exhibits the same behavior (godbolt.org is very helpful for crafting exactly the right code for such things). - if you can't do that work yourself, then you could significantly help anyone trying to help you by providing a complete example (i.e., complete program text plus instructions on how to compile/run it, detailed and specific enough that one can simply copy/paste them to reproduce what you're seeing). It might be possible for us to include a small compiler-side change to avoid some pattern on some hardware -- *if* there is a feasible alternative (e.g. we obviously can't just replace "< 0" comparisons with "< 1" comparisons). It might also be the case that the right fix is a microcode update, or to buy new hardware, or to hope that real applications won't experience the same impact as a microbenchmark. Only learning more about the issue will tell. Escalating this to AMD, once you have a small self-contained example, definitely makes sense. On Tue, Apr 7, 2020 at 12:31 PM Immanuel Haffner <haffner.imman...@gmail.com> wrote: > And I should add that i don't see that behaviour for i64, float, or double. > > -- > -- > v8-dev mailing list > v8-dev@googlegroups.com > http://groups.google.com/group/v8-dev > --- > You received this message because you are subscribed to the Google Groups > "v8-dev" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to v8-dev+unsubscr...@googlegroups.com. > To view this discussion on the web visit > https://groups.google.com/d/msgid/v8-dev/53ea45d3-0c46-4a02-8d72-6e0bb14cf9f4%40googlegroups.com > <https://groups.google.com/d/msgid/v8-dev/53ea45d3-0c46-4a02-8d72-6e0bb14cf9f4%40googlegroups.com?utm_medium=email&utm_source=footer> > . > -- -- v8-dev mailing list v8-dev@googlegroups.com http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to v8-dev+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/v8-dev/CAKSzg3TQdWEVz622HF0eTPJfyFiALUtvjrtnCQLz7wf6c49btw%40mail.gmail.com.
