topic Re: Hsfpe source files in Classic/Legacy CodeWarrior

Hsfpe source files

lukekim — Wed, 11 Jan 2017 06:45:32 GMT

Hi,

where can i find source files of SFPE libraries?

Thanks.

Re: Hsfpe source files

stanish — Wed, 18 Jan 2017 12:37:39 GMT

Hi luke,

I guess you are looking math SPFP library source codes.

Could you be little more specific about the source files you are missing?

Which CodeWarrior do you use? CW for MPC55xx/56xx v2.x?

Thanks!

Stan

Re: Hsfpe source files

lukekim — Mon, 23 Jan 2017 00:51:45 GMT

Hi, Stanislav.

Thanks for your reply.

Your guess is right.

We use two versions of CW.

- CW MCU V10. 7

- CW for MPC55XX and MPC56XX 2.10

We are using a MPC5674F.

thanks.

Re: Hsfpe source files

stanish — Mon, 23 Jan 2017 16:27:23 GMT

Hi Luke,

There is a complete .mcp project of FP library in CodeWarrior Classic:

<CW for MPC55xx and MPC56xx 2.10>\PowerPC_EABI_Support\MSL\MSL_C\PPC_EABI\Project\fdlibm.*.mcp

sources are available here:

<CW for MPC55xx and MPC56xx 2.10>\PowerPC_EABI_Support\MSL\MSL_C\PPC_EABI\SRC\

In CodeWarrior Eclipse (V10.7) SP floating point routines are located here:

<CW MCU v10.7>\MCU\PA_Support\ewl\EWL_C\src\sun_math\Single_precision\

Hope it helps.

Stan

Re: Hsfpe source files

lukekim — Tue, 31 Jan 2017 06:01:40 GMT

Hi, Stanislav.

I made a test function like this.

void floatTestFunc()

{

int a;

float x=1.0f,y=1.0f,z;

z = x* y;

a = (int) z;

}

If I build upper codes with below libraries in Eclipse Codewarrior, there is no error.

- libc_E200z760_VLE. a

- librt_E200z760_VLE. a

But, if I build the codes without libraries, there are some errors like following.

- undefined : '_f_ftoi' Referenced 'from floatTestFunc' in......

- undefined : '_f_mul' Referenced 'from floatTestFunc' in......

But, I can't find related source files in upper folder.

I would like to build with source files regarding floating points.

Thanks.

Re: Hsfpe source files

stanish — Tue, 31 Jan 2017 12:41:14 GMT

Luke,

The functions you are referring to are software emulated versions of float multiplication/conversion float to int.

These functions are being used by the compiler in case floating point flag is set to software (-fp soft).

Unfortunately there is no source code available for these functions, they are available only in object form

<CW MCU v10.x>\MCU\PA_Support\ewl\EWL_Runtime\Runtime_PA\Source\sfpe.*.a"

Please make sure your compiler has -fp SPFP or -fp SPFP_Only flag since it seems you are using a MCU with SPFP unit (e200z7) so you don't have to emulate single precision floating point operations using a library routines.

Hope it helps.

Stan

Re: Hsfpe source files

lukekim — Tue, 07 Feb 2017 01:42:33 GMT

Hi Stanislav,

To clarify my understanding is correct, if I change the floating point option to SFPE, the errors will be cleared. Is it correct?

Actually, I don't know the difference between Software and SFPE.

Thanks.

Your response helped me a lot.

Re: Hsfpe source files

stanish — Tue, 07 Feb 2017 17:19:38 GMT

Hi Luke,

Yes your assumption is correct - errors should disappear if you switch to SPFP or SPFP_Only.

* SPFP (Single Precision Floating Point) is in fact kind of Hardware floating point support. This option causes that single precision - 32 bit float number calculations are compiled into a single hardware instruction (or sequence of these instructions)

For more details please see EFPU2 unit - http://www.nxp.com/assets/documents/data/en/reference-manuals/e200z760RM.pdf

Double precision (64-bit ) operations are emulated by SW since there is no hw support for this on chip.

*Software - this option must be used on the devices without any floating point unit and both single/float and double precision calculation are emulated using software routines with sequence of an integer instructions. These routines significantly decreases CPU speed performance since instead of a single instruction there is a complex routine.

Hope it helps.

Stan,