S32K中的主题 Re: S32K144 AMMCLIB MLIB_Div_F32() Precision Problem

S32K144 AMMCLIB MLIB_Div_F32() Precision Problem

420113461 — Tue, 08 Sep 2020 03:05:38 GMT

When using ammclib tool library for division calculation, the last 15 bit(bit14 --- bit0) results of each calculation are 0, and the actual accuracy is only 16 bit(bit30 --- bit15).

Re: S32K144 AMMCLIB MLIB_Div_F32() Precision Problem

lukaszadrapa — Tue, 08 Sep 2020 05:50:25 GMT

Hi,

I can see it corresponds to the description. There's a note:

Regards,

Lukas

Re: S32K144 AMMCLIB MLIB_Div_F32() Precision Problem

420113461 — Tue, 08 Sep 2020 06:45:03 GMT

What's the difference between the three versions of ammclib library? Will it improve the calculation accuracy?

Re: S32K144 AMMCLIB MLIB_Div_F32() Precision Problem

lukaszadrapa — Tue, 08 Sep 2020 09:57:45 GMT

There's no difference. It just means if you have want to get object code for evaluation or if you want to buy a license for production (either object code only or full source code).

Regards,

Lukas

Re: S32K144 AMMCLIB MLIB_Div_F32() Precision Problem

petrz_ — Wed, 13 Oct 2021 12:11:44 GMT

Hi,

we added a new library function MLIB_DivSatAccurate_F32 in the AMMCLib release 1.1.26. This function calculates the division with full 32-bit accuracy of the result. The function is faster than 64-bit by 32-bit division (if you relied on compiler's optimization capabilities) but not as fast as MLIB_DivSat_F32.