thank you Leszek, you are right, i found the bug in my code, it was not exact translation now my javascript code can do the same numerical calculation, with same accuracy!
On Wednesday, June 26, 2019 at 3:50:56 PM UTC+8, Jerry Jin wrote: > > thank you, i'll come back later with sample code > > On Wednesday, June 26, 2019 at 1:56:34 PM UTC+8, Leszek Swirski wrote: >> >> To be clear: both C++ and JS use the exact same "double" representation, >> neither is more precise than the other. In fact we use the exact same >> machine instructions to feel with them. The significant digits/decimal >> digits mentioned are when printing, not when calculating. >> >> So, if it's not x87 extra precision, then it must either be a non exact >> translation from C++ to JavaScript, or a bug. If you have a small repro >> that demonstrates the difference, then please file a bug and we'll take a >> look. >> >> - Leszek >> >> >> On Wed, 26 Jun 2019, 06:00 Jerry Jin, <[email protected]> wrote: >> >>> thank you, i googled around and found msvc double is using 64-bit sse, >>> not x87 fpu >>> >>> not sure if it's because c++ double has 16 significant figures and js >>> has 15 >>> >>> On Tuesday, June 25, 2019 at 8:51:58 PM UTC+8, Jakob Kummerow wrote: >>>> >>>> If your original C++ code uses doubles, then porting it to >>>> TypeScript/JavaScript should be possible without loss of accuracy (and >>>> using plain built-in Numbers, no external libraries). If you have a >>>> reproducible counter-example, it would be interesting to hear about the >>>> details. >>>> >>>> The only exception I can think of is if your C++ compiler decided to >>>> emit x87 FPU code (on x86 hardware) which would use 80 bits (instead of 64 >>>> bits) for intermediate results. But I don't think modern compilers still >>>> do >>>> that; they should emit 64-bit SSE instructions. >>>> >>>> Side note: modern JavaScript has BigInts, which have arbitrary >>>> precision, but only integer values. So they're probably not useful for >>>> your >>>> case, but depending on what problem you're solving, you might be able to >>>> reformulate it to use integers. >>>> >>>> >>>> On Tue, Jun 25, 2019 at 2:19 PM Dan Elphick <[email protected]> >>>> wrote: >>>> >>>>> On Tuesday, June 25, 2019 at 10:24:56 AM UTC+1, Jerry Jin wrote: >>>>>> >>>>>> Hello, >>>>>> I'm new to v8-dev, sorry if this was already asked before. >>>>>> >>>>>> my problem is, i re-implemented a c++ numerical lib using typescript, >>>>>> and some test cases failed due to float number not having enough accuracy >>>>>> (i debugged js/c++ code side by side with chrome devtool&msvc) >>>>>> right now, my only option is to user third party lib like `decimal.js` >>>>>> >>>>>> i wonder is there any way to increase float number accuracy like c++ >>>>>> double type? >>>>>> >>>>> >>>>>> example, is there anyway to store more than 15 decimal places in >>>>>> number >>>>>> >>>>> >>>>> Javascript and V8 already use doubles to represent floating point >>>>> numbers. However double precision floating point numbers can have about >>>>> 16 >>>>> significant figures, so it's not possible to get any more accuracy using >>>>> C++ double. You may be able to print more than that (e.g. pi.ToFixed(30) >>>>> but they won't be accurate beyond 15 decimal places. >>>>> >>>>> >>>>>> ``` >>>>>> > let pi = 3.141592653589793238462643383280; >>>>>> < undefined >>>>>> > pi >>>>>> < 3.141592653589793 >>>>>> ``` >>>>>> >>>>>> not sure if this is possible in v8, if yes, where should i start? >>>>>> >>>>>> thanks! >>>>>> >>>>> -- >>>>> -- >>>>> 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/cfed0633-c6e7-45c4-a5c0-ca9f6cfaec9a%40googlegroups.com >>>>> >>>>> <https://groups.google.com/d/msgid/v8-dev/cfed0633-c6e7-45c4-a5c0-ca9f6cfaec9a%40googlegroups.com?utm_medium=email&utm_source=footer> >>>>> . >>>>> For more options, visit https://groups.google.com/d/optout. >>>>> >>>> -- >>> -- >>> 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/bc1e8872-a93a-405b-94b7-0a8ad01e24f2%40googlegroups.com >>> >>> <https://groups.google.com/d/msgid/v8-dev/bc1e8872-a93a-405b-94b7-0a8ad01e24f2%40googlegroups.com?utm_medium=email&utm_source=footer> >>> . >>> For more options, visit https://groups.google.com/d/optout. >>> >> -- -- 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/79e10225-6522-4524-810b-fa2db0f2e0e3%40googlegroups.com.
