64-bit floating point

[FA]

Hi,

I have a few questions about using floating point in nRF5 chips. As discussed here , nRF51 has no FPU but we can simply use "float" in Keil and have the operations in software.

1- Besides power and speed, what is the advantage of using FPU? In terms of accuracy is there any difference between a 32-bit float in FPU and in software?

2- How can we enable/disable FPU in nRF52? any example projects?

3- Is nRF52 capable of doing IEEE-754 64-bit operations? For our case, software or FPU doesn't matter, just going to 64-bit is important. If yes, how can I define a 64-bit float?

I'm using nRF52-DK, SDK11, SD132 in Keil.

Thanks, FA

[Håkon Alseth]

Hi,

The difference between doing float manipulation with and without FPU enabled is quite huge. I did a test with CMSIS-DPS fft example (\CMSIS\DSP_Lib\Examples\arm_fft_bin_example\ARM) a while back, and the difference is huge. Also note that there's a fft example now in SDK v12 (I have not benchmarked that one).

Compiled in Keil, all with -O3 optimization.

nRF51x22 : 198 ms 
nRF52832 w/o FPU :40.8 ms 
nRF52832 w/FPU : 2.84 ms

You will definitely benefit from tuning on the FPU if you're doing double/float operations.

To answer your questions:

1. Besides power/speed when performing a single-precision operation, no change.

2. most examples have FPU enabled by default on the nRF52832. In keil, you set this in your project settings under "Target -> Floating point hardware" drop-down menu. Similar setting in IAR settings (Under the Target section). With GCC, the FPU is set under CFLAGS and LDFLAGS with "-mfloat-abi=hard -mfpu=fpv4-sp-d16"

3. The FPU in the nRF52 is a component that is "included" in the ARM M4F core that we have licensed from ARM. According to ARM, it is not fully compliant with IEEE 754 (only for Single Precision): infocenter.arm.com/.../index.jsp infocenter.arm.com/.../index.jsp

The FPU provides floating-point computation functionality that is compliant with the ANSI/IEEE Std 754-2008, IEEE Standard for Binary Floating-Point Arithmetic, referred to as the IEEE 754 standard. The FPU contains 32 single-precision extension registers, which you can also access as 16 doubleword registers for load, store, and move operations.

edit: Oh, almost forgot. a 64-bit float is a "double", and is of type "float64_t" (which is just typedef'ed from a double).

Cheers, Håkon

[paul_tanner]

Thx Hakon, that does fix the issue. I tried both sprintf and SEGGER_RTT_printf. The latter did not work so I use sprintf and feed the output of that to the RTT call so all is good.