Floating point instructions

[Vivek Sankaranarayanan]

Posted on August 17, 2017 at 15:16

I am using STM32F303VC with System Workbench IDE.

I can't for the life of me get the hard floating point option to work. The compiler never generates floating point instructions. It' only generating soft code with library calls. Anyone knows what the problem could be?

[Tesla DeLorean]

Posted on August 17, 2017 at 17:01

>>Anyone knows what the problem could be?

Settings and defines sent to the compiler? Not hard to get GNU/GCC to compile FPU instructions.

The FPU only supports 32-bit floats, not doubles.

1.23 is a double, 1.23f is a float

printf/scanf use doubles

The FPU only supports simple arithmetic functions, anything complex or transcendental will use mix of software/hardware.

Most stuff tends to use the Soft ABI, though that might be changing now the M7 and M4F are getting more prevalent.

[Vivek Sankaranarayanan]

Posted on August 17, 2017 at 19:46

The FPU only supports 32-bit floats, not doubles.

1.23 is a double, 1.23f is a float

printf/scanf use doubles

The FPU only supports simple arithmetic functions, anything complex or transcendental will use mix of software/hardware.

Most stuff tends to use the Soft ABI, though that might be changing now the M7 and M4F are getting more prevalent.

So i have a simple statement saying:

float x; x = 1.25; x = x*x;

x = 1.25; x = x*x;

x = x*x;

Going through the dissasembly I don't see a single floating point instruction. Even if it were double, the compiler must be making use of 32bit floating point hardware to do double right?

[Tesla DeLorean]

Posted on August 17, 2017 at 21:32

The optimizer should flatten that to constants.

This was one of my test cases, built with GNU/GCC 4.9.3

float test1(float x, float y)

{

  return(x + y + sqrtf((x * x) + (y * y)));

}

demo.lss listing

...

08002578 <test1>:

//******************************************************************************

♯ include <math.h>

float test1(float x, float y)

{

 8002578:    b508          push    {r3, lr}

 800257a:    ed2d 8b02     vpush    {d8}

 800257e:    eeb0 8a40     vmov.f32    s16, s0

  return(x + y + sqrtf((x * x) + (y * y)));

 8002582:    ee20 0aa0     vmul.f32    s0, s1, s1

//******************************************************************************

♯ include <math.h>

float test1(float x, float y)

{

 8002586:    eef0 8a60     vmov.f32    s17, s1

  return(x + y + sqrtf((x * x) + (y * y)));

 800258a:    eea8 0a08     vfma.f32    s0, s16, s16

 800258e:    f001 fcd9     bl    8003f44 <sqrtf>

 8002592:    ee38 8a28     vadd.f32    s16, s16, s17

}

 8002596:    ee38 0a00     vadd.f32    s0, s16, s0

 800259a:    ecbd 8b02     vpop    {d8}

 800259e:    bd08          pop    {r3, pc}

Compiling file: main.c

arm-none-eabi-gcc -c -mcpu=cortex-m4 -mthumb -mfloat-abi=hard -mfpu=fpv4-sp-d16

-Os -ffunction-sections -fdata-sections -Wall -Wstrict-prototypes -Wextra -std=gnu89

-g -ggdb3 -fverbose-asm -Wa,-ahlms=out/main.lst  -DUSE_STDPERIPH_DRIVER 

-DSTM32F4XX  -DUSE_STM32F4_DISCOVERY  -DUSE_USB_OTG_FS  -DHSE_VALUE=8000000 

-D__FPU_PRESENT=1 -MD -MP -MF out/main.d -I.  -Iinc main.c -o out/main.o

Linking target: out/demo.elf

arm-none-eabi-g++ -mcpu=cortex-m4 -mthumb -mfloat-abi=hard -mfpu=fpv4-sp-d16

-Tstm32f4xx.ld -g -Wl,-Map=out/demo.map,--cref,--no-warn-mismatch -Wl,--gc-sections

-nostartfiles  out/startup_stm32f4xx.o out/diskio.o out/ff.o out/main.o out/misc.o

out/newlib_stubs.o out/stm32f4_discovery.o out/stm32f4_discovery_sdio_sd.o

out/stm32f4xx_dma.o out/stm32f4xx_gpio.o out/stm32f4xx_rcc.o out/stm32f4xx_sdio.o

out/stm32fxxx_it.o out/system_stm32f4xx.o    -o out/demo.elf

[RomainR.]

Posted on August 17, 2017 at 17:12

Hi,

Can you check the configuration of your project this way:

Into 

System Workbench Project \ Properties \ C/C++ Build \ Settings \ Tool Settings \ Preprocessor

Check if __FPU_PRESENT and __FPU_USED symbols are defined

And then into MCU Settings Floating Point ABI is hard

Best regards

[Vivek Sankaranarayanan]

Posted on August 17, 2017 at 19:42

I have both those symbols defined in the preprocessor.

But my problem could be this. I cannot select anything under Float Point Hardware. I have only No unit listed.

Could this be the problem?

[RomainR.]

Posted on August 17, 2017 at 20:36

What is your SW4STM32 version ?

Maybe upgrade to latest version should be fix your issue.

Best regards