Cortex-M4 & GCC - float behavior

5
votes

If I compile GCC project for Cortex-M4 (LPC4357) and use -mcpu=cortex-m4 switch floats aren't working (calls blx __addsf3, which eventually branches to stmia command, which results in an error interrupt, probably due to bad values in registers). If I compile with -mcpu=cortex-m3 it calls bl __addsf3, which has a different source and works.

I think I might have to fix other switches in order to make floats work with -mcpu=cortex-m4... I tried several things, but this is really strange.

Relevant compiler and linker switches:

Compiler: -mthumb -mcpu=cortex-m4 -std=c99 -c -g -D DEBUG -D gcc

Linker: -nostartfiles -nostdlib -nodefaultlibs -fno-exceptions -mthumb -mcpu=cortex-m4 -mfloat-abi=hard -mfpu=fpv4-sp-d16 -O0 -lgcc -lc

Can you tell if I'm missing something obvious here?

1
gccfloating-pointembeddedcortex-m
gcc takes a while to catch up to newer chips, even if the knowledge of those chips has been around for a long while. So use the cutting edge gcc, and if it doesnt do what you want then either contribute to gcc or wait or check llvm/clang to see how caught up they are (same problem, delay between hardware and compiler). – old_timer
Did you try to use the CodeRed compiler? It is Gcc base too. – Felipe Lavratti
No, I have to prepare the sample with this compiler (although good idea, I should try newer versions, there have been some since). – Anja V.

1 Answers

14
votes

With the GNU Tools for ARM Embedded Processors toolchain provided by ARM, you need to compile with:

 -mthumb -mcpu=cortex-m4 -mfloat-abi=hard -mfpu=fpv4-sp-d16

to use hardware floating point. See the following table from the readme.txt file:

 Architecture options usage
--------------------------------------------------------------------
| ARM Core | Command Line Options                       | multilib |
|----------|--------------------------------------------|----------|
|Cortex-M0+| -mthumb -mcpu=cortex-m0plus                | armv6-m  |
|Cortex-M0 | -mthumb -mcpu=cortex-m0                    |          |
|Cortex-M1 | -mthumb -mcpu=cortex-m1                    |          |
|          |--------------------------------------------|          |
|          | -mthumb -march=armv6-m                     |          |
|----------|--------------------------------------------|----------|
|Cortex-M3 | -mthumb -mcpu=cortex-m3                    | armv7-m  |
|          |--------------------------------------------|          |
|          | -mthumb -march=armv7-m                     |          |
|----------|--------------------------------------------|----------|
|Cortex-M4 | -mthumb -mcpu=cortex-m4                    | armv7e-m |
|(No FP)   |--------------------------------------------|          |
|          | -mthumb -march=armv7e-m                    |          |
|----------|--------------------------------------------|----------|
|Cortex-M4 | -mthumb -mcpu=cortex-m4 -mfloat-abi=softfp | armv7e-m |
|(Soft FP) | -mfpu=fpv4-sp-d16                          | /softfp  |
|          |--------------------------------------------|          |
|          | -mthumb -march=armv7e-m -mfloat-abi=softfp |          |
|          | -mfpu=fpv4-sp-d16                          |          |
|----------|--------------------------------------------|----------|
|Cortex-M4 | -mthumb -mcpu=cortex-m4 -mfloat-abi=hard   | armv7e-m |
|(Hard FP) | -mfpu=fpv4-sp-d16                          | /fpu     |
|          |--------------------------------------------|          |
|          | -mthumb -march=armv7e-m -mfloat-abi=hard   |          |
|          | -mfpu=fpv4-sp-d16                          |          |
|----------|--------------------------------------------|----------|
|Cortex-R4 | [-mthumb] -march=armv7-r                   | armv7-ar |
|Cortex-R5 |                                            | /thumb   |
|Cortex-R7 |                                            |      |
|(No FP)   |                                            |          |
|----------|--------------------------------------------|----------|
|Cortex-R4 | [-mthumb] -march=armv7-r -mfloat-abi=softfp| armv7-ar |
|Cortex-R5 | -mfpu=vfpv3-d16                            | /thumb   |
|Cortex-R7 |                                            | /softfp  |
|(Soft FP) |                                            |          |
|----------|--------------------------------------------|----------|
|Cortex-R4 | [-mthumb] -march=armv7-r -mfloat-abi=hard  | armv7-ar |
|Cortex-R5 | -mfpu=vfpv3-d16                            | /thumb   |
|Cortex-R7 |                                            | /fpu     |
|(Hard FP) |                                            |          |
|----------|--------------------------------------------|----------|
|Cortex-A* | [-mthumb] -march=armv7-a                   | armv7-ar |
|(No FP)   |                                            | /thumb   |
|----------|--------------------------------------------|----------|
|Cortex-A* | [-mthumb] -march=armv7-a -mfloat-abi=softfp| armv7-ar |
|(Soft FP) | -mfpu=vfpv3-d16                            | /thumb   |
|          |                                            | /softfp  |
|----------|--------------------------------------------|----------|
|Cortex-A* | [-mthumb] -march=armv7-a -mfloat-abi=hard  | armv7-ar |
|(Hard FP) | -mfpu=vfpv3-d16                            | /thumb   |
|          |                                            | /fpu     |
--------------------------------------------------------------------

You should really read the README file that come with gcc toolchains no matter where you get them from.