Floating Point Rounding Compiler Error ??

smh123

Other Parts Discussed in Thread: CCSTUDIO

I am seeing a floating point rounding condition (error??) that I do not expect. It causes a problem when two floats are added together and the resultant value is assigned to an integer. Here's the code snippet from my pwm configuration routine...

#define CLK_RATE_US 0.010    // 10 ns
float totalPrdTm_us;                     // pwm period
float  t1Tm_us;                               // transistor 1 active time
float  t2Tm_us;                                // transistor 2 active time
Uint16 totalPrdClks;                     // clocks in pwm period

t1Tm_us = .20;
t2Tm_us = .40;

totalPrdTm_us = t1Tm_us + t2Tm_us; // total period = 0.60 in watch window (float format)
// totalPrdTm_us as displayed in watch window in hex format is 0x3F199999
totalPrdClks = totalPrdTm_us / (float)CLK_RATE_US ; // result = 59, expect 60

totalPrdTm_us = 0.60; // total period
// totalPrdTm_us as displayed in watch window in hex format is 0x3F19999A
totalPrdClks = totalPrdTm_us / (float)CLK_RATE_US ; // result = 60, as expected

I can work around this by adding some very small value as follows:

totalPrdClks = totalPrdTm_us / (float)CLK_RATE_US + 0.0001;

This is annoying. I have checked this on another processor / compiler and the result of the addition results in the correct hex value of 0x3F19999A, and hence the correct final value of 60.

Is this the expected behavior of the CCS compiler?

This is a DSP/BIOS project running on the F28335 experimenter's kit.

CCS 3.3.82.13

Integrated Develpment 5.98.0.34

BIOS 5.33.04

Codgen 5.14

Silicon Rev 00.00.02

Regards,

Scott

over 16 years ago

0 Lori Heustess over 16 years ago

TI__Guru* 91145 points

Hello, Scott,

I've tried to reproduce this on a 2833x both with the normal RTS FPU library and the fast RTS FPU library. So far I have not had any luck. If you can provide a small test case to support@ti.com it would help.

0 smh123 over 16 years ago in reply to Lori Heustess

Intellectual 935 points

In the hopes that this may help others, I got the following reply in response to providing the following additional info to tech support:

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

I did some more testing using the TI examples and also my own code. I only get this problem when running in a DSP/BIOS project. To reproduce most simply I used the Hello example at:
        C:\CCStudio_v3.3\bios_5_33_04\packages\ti\bios\examples\basic\hello\ezdsp28335\hello.pjt

I added a few lines to the Hello.c file to reproduce the problem on an F28335 Experimenter's Kit ...

Void main()
{

////////// START OF MY CODE/////////
#define CLOCK_RATE_US    0.010
Uint16    u1;
float        f1, f2, f3;
    f1 = 0.2;
    f2 = 0.4;
    f3 = f1 + f2;

    u1 = (f1 + f2) / (float)CLOCK_RATE_US ;     // u1 == 59
    u1 = f3 / (float)CLOCK_RATE_US ;                 // u1 == 59
    u1 = f3 / 0.01;                                                    // u1 == 59

// This is my temporary work-around
    f3 += 0.0001;
    u1 = f3 / 0.01;    // u1 == 60 CORRECT
/////////// END OF MY CODE///////

    LOG_printf(&trace, "hello world!");

    /* fall into DSP/BIOS idle loop */
    return;
}

The actual c file is attached. I have also attached a pdf with relevant info. If you need more info, let me know. I am at a loss at this point but my guess would be that somehow I'm not linking in the correct fp library (????).

Regards,
    Scott

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

The tools team has accepted the defect (tracking # SDOCM00059580) and committed to fix for the next update of BIOS, v5.33.06. At this time I do not have a release date for this version, but when it is released it will be posted on the Update Advisor web site.

Texas Instruments

Software Technical Support

0 smh123 over 16 years ago in reply to smh123

Intellectual 935 points

Here's the workaround received from TI support:

The workaround is to add the following line as the first line in your 'main()' function:

asm(" SETFLG RNDF32=1 ; Enable rounding in FPU32 mode.");

Scott

C2000™︎ microcontrollers

C2000 microcontrollers forum

Floating Point Rounding Compiler Error ??