Multiplication of two small numbers results in infinity

John Batch

I'm running floating point calculations in the M3 core of a Concerto F28M36, and certain calculations multiplying two small numbers together results in -Inf as a result.

I created a simple program that reproduces the issue I'm seeing. I've included the contents of main.c below.

float test_var;
int main(void) {

    test_var = 4.174816e-36;
    
    test_var *= 0.001;
    
    return 0;
}

The project is using the TI ARM compiler, version 5.2.4. Is the result -Infinity an expected result, and if so, are there any compiler options available to provide a result of zero instead? I would expect a result of 0 if this is considered an underflow.

over 10 years ago

0 Archaeologist over 10 years ago

TI__Guru* 84285 points

The correct value is a denormal number. I would expect denormals to be treated as zero, but the library doesn't always handle special values correctly when floating-point hardware isn't available. What command-line options are you using? Also see http://processors.wiki.ti.com/index.php/IEEE_754_Support_in_TI_Compilers

0 John Batch over 10 years ago in reply to Archaeologist

Prodigy 170 points

Here is the compiler invocation :
c:/ti/ccsv6/tools/compiler/ti-cgt-arm_5.2.4/bin/armcl" -mv7M3 --code_state=16 --abi=eabi -me -Ooff --opt_for_speed=0 --include_path="c:/ti/ccsv6/tools/compiler/ti-cgt-arm_5.2.4/include" -g --gcc --define=ccs --display_error_number --diag_warning=225 --diag_wrap=off --gen_func_subsections=on --ual --preproc_with_compile --preproc_dependency="main.pp" "../main.c"
'Finished building: ../main.c'

Here is the linker invocation:
"c:/ti/ccsv6/tools/compiler/ti-cgt-arm_5.2.4/bin/armcl" -mv7M3 --code_state=16 --abi=eabi -me -Ooff --opt_for_speed=0 -g --gcc --define=ccs --display_error_number --diag_warning=225 --diag_wrap=off --gen_func_subsections=on --ual -z -m"compiler_float_bug.map" --heap_size=0 --stack_size=256 -i"c:/ti/ccsv6/tools/compiler/ti-cgt-arm_5.2.4/lib" -i"c:/ti/ccsv6/tools/compiler/ti-cgt-arm_5.2.4/include" --reread_libs --warn_sections --display_error_number --diag_wrap=off --xml_link_info="compiler_float_bug_linkInfo.xml" --rom_model -o "compiler_float_bug.out" "./main.obj" "../F28M36x_generic_M3_RAM.cmd" -l"rtsv7M3_T_le_eabi.lib"

Is that the information you are looking for?

0 Archaeologist over 10 years ago in reply to John Batch

TI__Guru* 84285 points

Yes, thank you. However, I am unable to reproduce the issue. Where are you seeing the "infinity" value? Is it in the variable view in CCS? What is the actual bit content of variable test_var?

0 John Batch over 10 years ago in reply to Archaeologist

Prodigy 170 points

I am seeing the infinity value in the expressions window of the debugger. As I step through the code, I see that test_var is correctly assigned to 4.174816e-36, but when I step over the test_var *= 0.001; line, the number changes to -Infinity.

Variable contents before step:
0x04B193B8

Variable contents after step:
0xFF800000

Would it be helpful for me to send you the entire project?

0 Archaeologist over 10 years ago in reply to John Batch

TI__Guru* 84285 points

Yeah, that's -Inf alright. I can't see how this might be happening. Yes, I think I need to see the project

0 John Batch over 10 years ago in reply to Archaeologist

Prodigy 170 points

I've attached the project I created to reproduce this issue. Included in the debug directory is the .out file I was using to step through the code. Let me know if there is anything else you'd like me to provide.

compiler_float_bug.zip

0 Archaeologist over 10 years ago in reply to John Batch

TI__Guru* 84285 points

There's a bug in the conversion from double to float for this value. I'm still analyzing the issue, so there's no defect number yet. To work around the problem, change 0.001 to 0.001f to avoid double arithmetic

0 Archaeologist over 10 years ago in reply to Archaeologist

TI__Guru* 84285 points

I've submitted this bug as SDSCM00051870.

0 John Batch over 10 years ago in reply to Archaeologist

Prodigy 170 points

Thanks for looking into the issue. The workaround will will work great.

Code Composer Studio™︎

Code Composer Studio forum

Multiplication of two small numbers results in infinity