How do I know at preprocessing time which compiler option was chosen between FPv5-D16 and FPv5-SP-D16?

[Nikita91]

The __SIZEOF_FLOAT__ and __SIZEOF_DOUBLE__ macros give the same result in both cases.

[waclawek.jan]

Try to output all preprocessor symbols (macros), see e.g. https://stackoverflow.com/questions/2224334/gcc-dump-preprocessor-defines (which refers to gcc documentation).

Maybe there is some specific __TARGET_FPU_xx or alike.

JW

[Nikita91]

Nice idea. Thank you.

I found a define quite useful:

__SOFTFP__ : is defined if soft FP is selected in configuration

__ARM_FP : is defined and set if hardware FP is selected in configuration

__ARM_FP Set if hardware floating-point is available.

Bits 1-3 indicate the supported floating-point precision levels. The other bits are reserved.

• Bit 1 - half precision (16-bit).
• Bit 2 - single precision (32-bit).
• Bit 3 - double precision (64-bit).

These bits can be bitwise or-ed together. Permitted values include:

• 0x04  for single-support.
• 0x0C  for single- and double-support.
• 0x0E  for half-, single-, and double-support.

On STM32H743:

with FPv5_SP_D16 I get 0x04

with FPv5_D16 I get 0x0E, but the same thing with none...

with softfp __ARM_FP is not defined

Same thing for __GCC_IEC_559 :

0 for FPv5_SP_D16 hard or soft, and soft+none

2 for FPv5_D16 or none hard or soft

But this only means that there is an implementation difference (with regard to the standard) between SP and DP.

[waclawek.jan]

Well, then the answer to your original question is, "you don't". It *is* an unpleasant but valid answer.

Another, extreme maybe, answer is: gcc is open source, you can make whatever change you wish to it.

That much for academic and impractical answers. I'm sure you know the practical ones already.

JW