-fp

Controls floating-point code generation.

Syntax

  -fp 
  keyword

The choices for keyword are:

none | off

No floating point code generation.

soft[ware]

Use software libraries to perform floating-point operations. This is the default.

hard[ware]

Use the processor's built-in floating-point capabilities to perform floating-point operations.

dpfp

Use the processor's double-precision floating-point capabilities on the e500v2 processor.

spfp

Use software libraries for floating-point operations that use the double data type and use the e500 SPE-EFPU floating-point capabilities for other floating-point operations.

spfp_only

Use to have the compiler consider double and long double data types as floating point. This option is only supported for e200 (Zen or VLE) and e500v1 processors that support SPFP APU.

Note: When you downgrade from double data type to a floating point you will lose precision and range. If your expected numbers are within the range supported by a floating point data type, then this option might dramatically speed up and shrink your code. Do not use this option if you have instances in your project that depend on the size of a double data type.

fmadd

Equivalent to -fp hard -fp_contract.

Remarks

When using the -fp spfp_only option, the size of a double data type changes to a floating point data type, if you have existing code that is expecting to find certain bits at certain locations of the exponent or significand, then you will have to change that code to expect 4 byte doubles. Your code can make a test as shown in Example Test Code .

Listing: Example Test Code


  if (sizeof(double) == 4) {
     ...

  } else {

     ... }

The e500 and VLE library project files have targets and pre-built libraries (with SP in the name) that support this feature. Ensure you pick the right libraries to include in a project that supports this feature else you may call a function with a 8 byte double parameter and only pass a 4 byte double argument. The linker will report with a warning if you mix up the libraries - make sure you have linker warnings enabled.

If you have a library that doesn't use floating point, try setting it to none for the floating point model by using the -fp none option. Libaries with none floating point do not cause a warning when added to projects using another floating point model.

The sample code in Sample Code assumes that you are using the -fp spfp_only option and have included SP libraries. Your existing code makes a call to a MSL math function and a user defined function that takes a double argument and returns a double data type.

Listing: Sample Code


  #include <math.h>
  extern double my_func(double);

  extern double d1, d2;

  void main()

  {

        d1 = pow(d2, 2.0);

        d2 = my_func(d1);

  }

Following can be observed while executing the sample code in Sample Code :

2.0 will be treated as a 4 byte double constant (exactly like 2.0f).
Storage for d1 and d2 will be 4 bytes each (exactly like floats).
MSL will either inline or call a stub function for pow which will call powf.
my_func will receive and return a 4 byte double. As long as my_func doesn't do bit twiddling or require numbers not representable in a float, it will do its job correctly.
Note: If you are using a Zen processor and are using the -fp spfp_only option, ensure passing -spe_addl_vector instead of -spe_vector in order to have the compiler generate Multiply-Add instructions.