TMS320C6748: Floating point representation

Cyril Lassouane

Part Number: TMS320C6748

Hello,

In the context of software integrations, I noticed different results (float data single precision) between algorithms running on the C6748 DSP and the intel x86 PC architecture.
On these 2 architectures, the source code is the same, and the floating point configuration is the same (fp:strict to respect IEEE 754 rules).

The different results are due to a different floating numbers representation, for instance:

Floating point value : 47.849457
Representation of this floating point value on DSP C6748 : 47.84946
Representation of this floating point value on PC Intel x86 : 47.8494568

Another example:

Floating point value : 46.6782722
Representation of this floating point value on DSP C6748 : 46.67827
Representation of this floating point value on PC Intel x86 : 46.6782722

Is there a way to avoid this limitation in order to have the same results as the x86 architecture ?

Thanks,

Cyril

over 2 years ago

0 Aravind Batni over 2 years ago

TI__Expert 8885 points

Hi Cyril,

Not aware of any compiler options to match to x86 representation. Looping in Compiler Team for any inputs.

Thanks

0 George Mock over 2 years ago

TI__Guru**** 232640 points

The C6000 compiler has no features that allow you to match the floating point computations of any other CPU. The article Floating Point Optimization may be helpful.

Thanks and regards,

-George

Processors

Processors forum

TMS320C6748: Floating point representation