S32DS / e200z4 hard floats working? (MPC5746R)

jamesmurray — Fri, 13 Oct 2017 12:15:25 GMT

I might be missing something here, but I don't seem to be able to get S32DS for Power v1.2 to actually use hardware floating point maths.

I created the most basic project and made sure that "Use Hardware Floating Point" is ticked, but there are no recognisable floating point instructions in the disassembly.

Unticking, making clean and rebuilding generates output that is exactly the same.

My source:

--------------------

#include "derivative.h" /* include peripheral declarations */

extern void xcptn_xmpl(void);

volatile float test;

int main(void)

{
volatile int counter = 0;
xcptn_xmpl (); /* Configure and Enable Interrupts */

/* Loop forever */

   for(;;) {
          counter++;
          test = counter * 1.2345;
   }
}

--------------------

Disassembly snippet 1:

--------------------

           test = counter * 1.2345;
1003826:   c2 7f          se_lwz r7,8(r31)
1003828:   01 73          se_mr   r3,r7
100382a:   79 ff fa a3    e_bl    10032cc <__floatsidf>
100382e:   02 40          se_mtar r8,r4
1003830:   01 37          se_mr   r7,r3
1003832:   01 73          se_mr   r3,r7
1003834:   03 04          se_mfar r4,r8
1003836:   70 a7 e7 f3    e_lis   r5,16371
100383a:   70 b8 c0 83    e_or2i r5,49283
100383e:   70 c2 e2 6e    e_lis   r6,4718
1003842:   70 d2 c7 8d    e_or2i r6,38797
1003846:   79 ff f8 23    e_bl    1003068 <__muldf3>
100384a:   02 40          se_mtar r8,r4
100384c:   01 37          se_mr   r7,r3
100384e:   01 73          se_mr   r3,r7
1003850:   03 04          se_mfar r4,r8
1003852:   79 ff fa fb    e_bl    100334c <__truncdfsf2>
1003856:   01 36          se_mr   r6,r3
1003858:   70 e8 e0 00    e_lis   r7,16384
100385c:   54 c7 00 20    e_stw   r6,32(r7)

--------------------

Disassembly snippet 2:

--------------------

010032cc <__floatsidf>:
__floatsidf():
10032cc:   18 21 06 d8    e_stwu r1,-40(r1)
10032d0:   00 80          se_mflr r0
10032d2:   7c 67 fc 70    e_srwi r7,r3,31
10032d6:   48 36          se_li   r6,3
10032d8:   db 01          se_stw r0,44(r1)
10032da:   d2 61          se_stw r6,8(r1)
10032dc:   d3 71          se_stw r7,12(r1)
10032de:   2a 03          se_cmpi r3,0
10032e0:   7a 02 00 1a    e_bne   10032fa <__floatsidf+0x2e>
10032e4:   48 27          se_li   r7,2
10032e6:   d2 71          se_stw r7,8(r1)
10032e8:   18 61 80 08    e_addi r3,r1,8
10032ec:   78 00 01 11    e_bl    10033fc <__pack_d>
10032f0:   cb 01          se_lwz r0,44(r1)
10032f2:   00 90          se_mtlr r0
10032f4:   18 21 80 28    e_addi r1,r1,40
10032f8:   00 04          se_blr
10032fa:   2a 07          se_cmpi r7,0
10032fc:   7a 12 00 0e    e_beq   100330a <__floatsidf+0x3e>
1003300:   18 03 ab 80    e_cmpi cr0,r3,-2147483648
1003304:   7a 12 00 3e    e_beq   1003342 <__floatsidf+0x76>
1003308:   00 33          se_neg r3
100330a:   7c 67 00 34    cntlzw r7,r3
100330e:   7c 65 0c 70    e_srwi r5,r3,1
1003312:   18 c7 84 fd    e_addi r6,r7,-3
1003316:   18 87 80 1d    e_addi r4,r7,29
100331a:   2a 06          se_cmpi r6,0
100331c:   7c 60 30 30    slw     r0,r3,r6
1003320:   18 e7 b0 1f    e_subfic r7,r7,31
1003324:   18 c4 b0 1f    e_subfic r6,r4,31
1003328:   42 43          se_slw r3,r4
100332a:   7c a6 34 30    srw     r6,r5,r6
100332e:   d4 71          se_stw r7,16(r1)
1003330:   48 05          se_li   r5,0
1003332:   7c c6 00 1e    isellt r6,r6,r0
1003336:   7c 63 28 1e    isellt r3,r3,r5
100333a:   d6 61          se_stw r6,24(r1)
100333c:   d7 31          se_stw r3,28(r1)
100333e:   79 ff ff aa    e_b     10032e8 <__floatsidf+0x1c>
1003342:   70 78 e1 e0    e_lis   r3,49632
1003346:   48 04          se_li   r4,0
1003348:   79 ff ff a8    e_b     10032f0 <__floatsidf+0x24>

--------------------

According to VLEPEM.pdf, I was expecting to see opcodes:

efsXXXX or efdXXXX for hardware floating point instructions.

I also disassembled many of the libraries that are installed (libgcc.a) and wasn't able to find one that included any of those either - they all appeared to be using that slow software implementation shown above.

Also,

- is there a core reference manual for e200z425n3?

- is there an accurate VLE reference manual that includes all opcodes that S32DS is using? (isellt above is missing from VLEPEM.pdf)

James

Re: S32DS / e200z4 hard floats working? (MPC5746R)

jamesmurray — Fri, 13 Oct 2017 20:10:47 GMT

I can answer my own question here.

I hadn't used floats on an embedded platform before and didn't realise that the compiler is automatically treating 1.2345 in my example as 'double' not 'float'. Using 1.2345f instead lets the compiler use the hardware single precision floating point instructions.

Adding -Wdouble-promotion to the compiler flags warns of the original coding error.

James

Re: S32DS / e200z4 hard floats working? (MPC5746R)

stanish — Mon, 16 Oct 2017 12:10:26 GMT

Hi James,

You are right. Thanks for sharing the solution!

BTW: if you are interested in the optimized production grade single precision math routines you can possibly use AMMCLIBs (Automotive Math and Motor Control Library Set|NXP )

I'd recommend you to install the latest S32DS Power update which includes ammclib v1.1.9 libs plus new project wizard integration.

Hope it helps.

Stan

S32 Design StudioのトピックRe: S32DS / e200z4 hard floats working? (MPC5746R)

S32DS / e200z4 hard floats working? (MPC5746R)

Re: S32DS / e200z4 hard floats working? (MPC5746R)

Re: S32DS / e200z4 hard floats working? (MPC5746R)