If it's an option for you to maintain an own fork of V8, then extending the header with the versions you have and putting them behind a compile-time macro might be a good choice. Otherwise I'd suggest trying to upstream your changes via submitting a pull request to https://chromium-review.googlesource.com/ (feel free to check out the contributing guide <https://v8.dev/docs/contribute>).
On Friday, October 28, 2022 at 3:03:50 PM UTC+2 [email protected] wrote: > I have written alternative implementations of the DoubleToInteger, > DoubleToWebIDLInt64, DoubleToInt32, FastD2UI, IsInt32Double, and > IsUint32Double functions in src/numbers/conversions-inl.h for x86_64 using > SSE2 intrinsics. > > I have done some benchmarking, and my alternative implementations > of DoubleToInteger, DoubleToWebIDLInt64, FastD2UI, IsInt32Double, and > IsUint32Double outperform the current implementations of DoubleToInteger, > DoubleToWebIDLInt64, FastD2UI, IsInt32Double, and IsUint32Double in the > src/numbers/conversions-inl.h on x86_64 platforms. > > My alternative implementation of DoubleToInt32 has similar performance to > the existing implementation of DoubleToInt32 in > src/numbers/conversions-inl.h for values that are within the range of a > 32-bit signed integer, and my alternative implementation of DoubleToInt32 > is much faster for values that are outside of the range of a 32-bit signed > integer. > > I also would like to add an option to enable compiling with the existing > implementations of the functions src/numbers/conversions-inl.h instead of > my alternative implementations of these functions on x86_64 platforms. > > My alternative implementation of the DoubleToWebIDLInt64, DoubleToInt32, > FastD2UI, IsInt32Double, and IsUint32Double methods has dependencies on > the _mm_cvttsd_si64 and _mm_cvtsi64_sd intrinsics that are not available on > 32-bit x86 targets (including 32-bit x86 targets that have SSE2 support). > > How do I add an option that would allow V8 to be compiled with my > alternative implementations of the conversion methods in > src/numbers/conversions-inl.h instead of the existing methods in > src/numbers/conversions-inl.h (and vice versa)? > > -- -- v8-dev mailing list [email protected] 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 [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/v8-dev/a9aa6c70-46b4-46bc-881c-0ddf970f8156n%40googlegroups.com.
