Compiler/CC3200: bug in Ti compiler preprocessor - #define float constant is set to zero

it turns out it isn't the preprocessor.

it turns out that float arithmetic doesn't work correctly in function arguments. not even with literals.

Hi Tim,

Could you send a code snippet please?

~roger

that is a code snippet above.

what I'm seeing is that float arithmetic isn't evaluating properly, although in relaxed mode a statement like:

float a = 2.0 + 2.0;

does work somewhat.

Hi Tim,

That code snippet above, works ok for me, so I guess some other option is different.

Can you try this test to make sure we're using the same test-case?

#include "hw_types.h"
#include "hw_memmap.h"
#include "hw_gpio.h"
#include "hw_ints.h"
#include "pin.h"
#include "rom.h"
#include "rom_map.h"
#include "gpio.h"
#include "prcm.h"
#include "interrupt.h"

#include "uart_if.h"

extern void (* const g_pfnVectors[])(void);

#define API_POLL_QUEUE_INTERVAL 30.0f
const float _API_POLL_QUEUE_INTERVAL=30.0f;

void testFloatAddAndPrint(float a, float b)
{
    Report("adding #define float + const float: %f + %f = %f\n\r",a,b,a+b);
}

void testFloatPrint(void)
{
    Report("adding #define float + const float: %f + %f = %f\n\r",API_POLL_QUEUE_INTERVAL,_API_POLL_QUEUE_INTERVAL,API_POLL_QUEUE_INTERVAL+_API_POLL_QUEUE_INTERVAL);
}


void main()
{
    MAP_IntVTableBaseSet((unsigned long)&g_pfnVectors[0]);
    MAP_IntMasterEnable();
    MAP_IntEnable(FAULT_SYSTICK);
    PRCMCC3200MCUInit();

    MAP_PRCMPeripheralClkEnable(PRCM_GPIOA1, PRCM_RUN_MODE_CLK);
    MAP_PinTypeGPIO(PIN_01, PIN_MODE_0, false);
    MAP_GPIODirModeSet(GPIOA1_BASE, 0x4, GPIO_DIR_MODE_OUT);
    MAP_PinTypeGPIO(PIN_02, PIN_MODE_0, false);
    MAP_GPIODirModeSet(GPIOA1_BASE, 0x8, GPIO_DIR_MODE_OUT);

    InitTerm();

    testFloatAddAndPrint(API_POLL_QUEUE_INTERVAL+_API_POLL_QUEUE_INTERVAL,API_POLL_QUEUE_INTERVAL+_API_POLL_QUEUE_INTERVAL);
    testFloatAddAndPrint(API_POLL_QUEUE_INTERVAL,_API_POLL_QUEUE_INTERVAL);
    testFloatPrint();
}

Thanks,

~roger

Hi Roger thanks, this is a good approach.

I compiled your main.c with my existing compiler and settings.

I had to remove FREERTOS flags and add startup_ccs.c and uart_if.c from the existing project.

This is the output I got:

adding #define float + const float: 60.000000 + 60.000000 = 120.000000

adding #define float + const float: 30.000000 + 30.000000 = 60.000000

adding #define float + const float: 0.000000 + 30.000000 = 60.000000

which is different to what I was seeing, but still bat-*** crazy.

 

I switched from compiler 5.2.8 to 16.9.0 LTS and I got the same output:

adding #define float + const float: 60.000000 + 60.000000 = 120.000000

adding #define float + const float: 30.000000 + 30.000000 = 60.000000

adding #define float + const float: 0.000000 + 30.000000 = 60.000000

What output do you get?

Hmm, ok,

Could you add a couple of compiler switches, if you haven't got these already.

1.  -pdr (--issue_remarks) (to add extra compiler remarks)

This will generate compiler remarks and may give us some clues.

2.  -ppl (--preproc_with_line) -ppa (--preproc_with_compile) (to generate .pp pre-processor file and continue)

This should generate a .pp file after the pre-processor, and give you something like :-

const float _API_POLL_QUEUE_INTERVAL=30.0f;

void testFloatAddAndPrint(float a, float b)
{
    Report("adding #define float + const float: %f + %f = %f\n\r",a,b,a+b);
}

void testFloatPrint(void)
{
    Report("adding #define float + const float: %f + %f = %f\n\r",30.0f,_API_POLL_QUEUE_INTERVAL,30.0f+_API_POLL_QUEUE_INTERVAL);
}


void main()
{
    IntVTableBaseSet((unsigned long)&g_pfnVectors[0]);
    IntMasterEnable();
    IntEnable(15);
    PRCMCC3200MCUInit();

    PRCMPeripheralClkEnable(0x00000007, 0x00000001);
    PinTypeGPIO(0x00000000, 0x00000000, 0);
    GPIODirModeSet(0x40005000, 0x4, 0x00000001);
    PinTypeGPIO(0x00000001, 0x00000000, 0);
    GPIODirModeSet(0x40005000, 0x8, 0x00000001);

    InitTerm();

    testFloatAddAndPrint(30.0f+_API_POLL_QUEUE_INTERVAL,30.0f+_API_POLL_QUEUE_INTERVAL);
    testFloatAddAndPrint(30.0f,_API_POLL_QUEUE_INTERVAL);
    testFloatPrint();
}

and I get the output as expected :-

adding #define float + const float: 60.000000 + 60.000000 = 120.000000
adding #define float + const float: 30.000000 + 30.000000 = 60.000000
adding #define float + const float: 30.000000 + 30.000000 = 60.000000

Can you send your complete compile options?

~roger

I switched from strict to relaxed FP mode and I got the same answer.

I also turned off the option --fp_reassoc and I got the same answer:

adding #define float + const float: 60.000000 + 60.000000 = 120.000000

adding #define float + const float: 30.000000 + 30.000000 = 60.000000

adding #define float + const float: 0.000000 + 30.000000 = 60.000000

however, I changed the --float_support from fpalib to vfplib and the output changed:

adding #define float + const float: 60.000000 + 60.000000 = 120.000000
adding #define float + const float: 30.000000 + 30.000000 = 60.000000
adding #define float + const float: 30.000000 + 30.000000 = 60.000000

 

 

Hi Roger,

-pdr -ppl and --ppa don't work for me in CCS. would I have to set these on the command line?

here are the flags as reported by CCS:

-mv7M4 --code_state=16 --float_support=fpalib -me -Ooff --opt_for_speed=0 --fp_mode=relaxed --include_path="/Applications/ti/ccsv7/tools/compiler/ti-cgt-arm_16.9.0.LTS/include" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/simplelink_extlib/provisioninglib" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/oslib/" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/driverlib/" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/simplelink/" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/simplelink/source" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/simplelink/include" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/inc/" --include_path="/Users/tim/ti/CC3200SDK_1.2.0/cc3200-sdk/example/common" --define=ccs --define=cc3200 -g --c89 --c++03 --relaxed_ansi --diag_warning=225 --diag_wrap=off --display_error_number --issue_remarks --gen_func_subsections=on --abi=eabi --fp_reassoc=off

seems that the --float_support is the problem, and I can't recall when this was set in the project.

what are your reccomended options for FP support on the cc3200?

thanks Roger

I'm not sure when exactly this was set, but would Ti consider disabling this option in CCS until the bug is fixed?

Hi Tim,

Yes, I agree - we definitely need to look to see how we can avoid hitting this one again.

~roger