BCD is not relevant to integer values where all bits are used. GMP is slow. Very very slow.
Making this configurable means adding an additional, little used code path. One will end up being a second class citizen. Similarly for allowing configurable minimum units. You can look at vector2.h for many of the places but not all. Seth On Mon, Nov 25, 2024, 7:25 PM Liang Jia <[email protected]> wrote: > Hi Mark and Seth, > > Thanks for your reply. > > Yes, you're right, it will lose some precision when using double. > > Just curious, I did some search online. > *1. I found that the key players in the PCB market still have limits.* > OrCAD X: 200 in. x 200 in = 5 meter * 5 meter; > > https://www.cadence.com/en_US/home/tools/pcb-design-and-analysis/orcad.html#pcb-layout > Altium Designer: 200 in. x 200 in = 2.5 meter * 2.5 meter > > https://www.altium.com/documentation/knowledge-base/altium-designer/indicate-visual-cues-at-the-coordinate-limit-of-pcb-design-space-beyond-which-objects-should-not-be-placed > Allegro: I can't find any detail information, but I tried the software, it > seems still have limit. *Anyone know the board size limit for Allegro?* > > *2. If we really want to solve this problem, there are options below.* > 2.1 Using software integer library, such as GMP; > We can give an option to the user, let the user choose to enable it > or not. > If enabled, Kicad can support a bigger board size, but software will > slow; > If enabled, Kicad runs as usual. > https://gmplib.org/ > 2.2 Using Binary coded-decimal, > > https://stackoverflow.com/questions/2624973/why-doesnt-my-processor-have-built-in-bigint-support > https://en.wikipedia.org/wiki/Binary-coded_decimal > 2.3 *Give the control of precision and board size to the user.* > If a user wants to have a smaller board size, he/she will have more > precision location and something else; > If a user wants to have a bigger board size, he/she will have less > precision location and something else; > > *@Mark what's your opinion? * > > > *@Seth,* > >Addressing this means reworking our internal coordinate system > Could you please kindly give me some location for those codes? so I can > dig into it. > > > Sincerely > Liang > > On Tue, 26 Nov 2024 at 02:02, Mark Roszko <[email protected]> wrote: > >> Floats are not accurate beyond 6-7 digits and worse, floating point >> behavior is actually not fully well defined. You can actually get different >> quirks depending on platform, compiler and arch. Though generally the risk >> isn't in the arithmetic but supporting functions that are part of libc. >> >> >And why did Kicad choose integer as internal measurement resolution? >> >> Using integers is standard programming behavior for applications that >> want well defined and bounded mathematical behavior. >> >> When you do floating point math, even if we ignore the inaccurate power >> part of the number, that error still sits in the number. When you carry out >> sufficient and numerous operations using numbers that carry these not used >> digits, they can actually creep in and start affecting the digits you do >> care about and cause errors. >> >> >> >Last, so that means there is no way to handle this issue? >> >> >> We have come to the conclusion that if somebody needs PCB designs larger >> than 4 meters, which is already an ridiculous size. They need to discuss >> with us the use case which is already going to be ridiculously niche for >> that one person because there is no standard PCB manufacturing equipment >> for a board of that size. >> >> >> On Mon, Nov 25, 2024, 8:01 AM Liang Jia <[email protected]> >> wrote: >> >>> Hi Mark, >>> >>> Thanks for your email. >>> >>> Could you please explain more about the relation between "Change the >>> measurement store from 32-bit to *64-bit*" and "128-bit integer math >>> support in processors"? >>> And why did Kicad choose integer as internal measurement resolution? >>> >>> >mm is not adequate to express the resolution of position data required >>> for a board design. >>> How about this? use mm for internal measurement resolution, but use >>> nanometer or *double(instead of integer)* for board design. >>> >>> >>> Last, so that means there is no way to handle this issue? >>> >>> On Mon, 25 Nov 2024 at 19:21, Mark Roszko <[email protected]> wrote: >>> >>>> >This means it is possible to create boards up to approximately 4 >>>> meters by 4 meters >>>> >>>> Yes, KiCad is limited to 4x4 meter boards currently. >>>> >>>> > Change the measurement store from 32-bit to 64-bit, so Kicad can >>>> support a larger board. >>>> >>>> This is not an easy task in the slightest. Specifically because we >>>> can't get 128-bit integer math support in processors, and many compilers do >>>> not support 128-bit integers. I think gcc has some experimental support. >>>> >>>> > Change the resolution of all objects to mm >>>> >>>> mm is not adequate to express the resolution of position data required >>>> for a board design. >>>> >>>> >>>> >>>> Basically the bug here is we simply do not tell the users the DXF is >>>> beyond our board support limit. >>>> >>>> >>>> On Mon, Nov 25, 2024 at 5:07 AM Liang Jia <[email protected]> >>>> wrote: >>>> >>>>> Hi All, >>>>> >>>>> I am writing to inquire about the challenges we are facing when >>>>> importing DXF files which contain large numbers into our system. >>>>> >>>>> I have noticed that when the number exceeds a certain threshold(such >>>>> as 4437 mm), the import process results in an int overflow error. >>>>> >>>>> I did the search below: >>>>> 1. Found that there was a ticket to track it, but it seems *it still >>>>> opens*. >>>>> https://gitlab.com/kicad/code/kicad/-/issues/12392 >>>>> >>>>> 2. From the Kicad document: >>>>> The internal measurement resolution of all objects in KiCad is *1 >>>>> nanometer*, and measurements are stored as *32-bit integers*. This >>>>> means it is possible to create boards up to approximately 4 meters by 4 >>>>> meters >>>>> I think the *root cause* is here: Kicad tried to convert the DXF >>>>> number into nanometer, but those numbers exceeded the limit of integer. >>>>> >>>>> Questions: >>>>> 1. Is there any workaround for this case, and let Kicad import those >>>>> files successfully? >>>>> 2. If I want to fix ticket 12392, what should I do? >>>>> Change the measurement store from 32-bit to 64-bit, so Kicad can >>>>> support a larger board. >>>>> Change the resolution of all objects to mm >>>>> >>>>> Looking forward to any comment or workaround. >>>>> >>>>> Sincerely >>>>> Liang >>>>> >>>>> -- >>>>> You received this message because you are subscribed to the Google >>>>> Groups "KiCad Developers" group. >>>>> To unsubscribe from this group and stop receiving emails from it, send >>>>> an email to [email protected]. >>>>> To view this discussion visit >>>>> https://groups.google.com/a/kicad.org/d/msgid/devlist/CAE0Ak8bd%3DgDwW%3DDWVXH-TKj%3Dr4nsgW370aNqB6PbP7T8b2QU6A%40mail.gmail.com >>>>> <https://groups.google.com/a/kicad.org/d/msgid/devlist/CAE0Ak8bd%3DgDwW%3DDWVXH-TKj%3Dr4nsgW370aNqB6PbP7T8b2QU6A%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>>> . >>>>> >>>> -- >>>> You received this message because you are subscribed to the Google >>>> Groups "KiCad Developers" group. >>>> To unsubscribe from this group and stop receiving emails from it, send >>>> an email to [email protected]. >>>> To view this discussion visit >>>> https://groups.google.com/a/kicad.org/d/msgid/devlist/CAJjB1qKoN689BoyB1uWpWYmDCpbaCRKCyW2qmnBOoG5Au_Nnfg%40mail.gmail.com >>>> <https://groups.google.com/a/kicad.org/d/msgid/devlist/CAJjB1qKoN689BoyB1uWpWYmDCpbaCRKCyW2qmnBOoG5Au_Nnfg%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>> . >>>> >>> -- >>> You received this message because you are subscribed to the Google >>> Groups "KiCad Developers" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to [email protected]. >>> To view this discussion visit >>> https://groups.google.com/a/kicad.org/d/msgid/devlist/CAE0Ak8bmAyB7ciCe6nrps8gw0meXgaj25r5Zq-o2Z9x_-8CFDA%40mail.gmail.com >>> <https://groups.google.com/a/kicad.org/d/msgid/devlist/CAE0Ak8bmAyB7ciCe6nrps8gw0meXgaj25r5Zq-o2Z9x_-8CFDA%40mail.gmail.com?utm_medium=email&utm_source=footer> >>> . >>> >> -- >> You received this message because you are subscribed to the Google Groups >> "KiCad Developers" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> To view this discussion visit >> https://groups.google.com/a/kicad.org/d/msgid/devlist/CAJjB1qLDDND2JZpaBCOHZpoz9HvsVF%2Brn%3D5nSGAQVDzyrFFRBA%40mail.gmail.com >> <https://groups.google.com/a/kicad.org/d/msgid/devlist/CAJjB1qLDDND2JZpaBCOHZpoz9HvsVF%2Brn%3D5nSGAQVDzyrFFRBA%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> > -- > You received this message because you are subscribed to the Google Groups > "KiCad Developers" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion visit > https://groups.google.com/a/kicad.org/d/msgid/devlist/CAE0Ak8awnYeEqbpz0SwNc7wAf_hmS1ybhDWNbRuyr7efEPUf-w%40mail.gmail.com > <https://groups.google.com/a/kicad.org/d/msgid/devlist/CAE0Ak8awnYeEqbpz0SwNc7wAf_hmS1ybhDWNbRuyr7efEPUf-w%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > -- You received this message because you are subscribed to the Google Groups "KiCad Developers" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/a/kicad.org/d/msgid/devlist/CAFdeG-rsD3kZTHX8w9xpxr-UXKj_rhRS%2B6t8UwfJcrQSDXGquQ%40mail.gmail.com.
