2.6.0 & FPU

[russian]

I am migrating my project from 2.4.3 to 2.6.0 and it dies on startup, while printing the first line of the console greeting.

I was able to isolate the problem to CORTEX_USE_FPU in chconf.h - everything works without FPU option (#define CORTEX_USE_FPU FALSE) but breaks when I change this to TRUE. I am using stm32f4 discovery, optimization none -O0 and "-mcpu=cortex-m4 -mfloat-abi=softfp -mfpu=fpv4-sp-d16 -fsingle-precision-constant" - these flags used to work for me with 2.4.3

Please advice

[Tectu]

I don't know about the FPU thing but lately there were some issues with the -O0 thing. Since you posted on the ChibiOS/GFX sub forum, I assume that you're using that as well, right? Can you please try to compile the application with -O2? The talk is going on here.


~ Tectu

[russian]

I am reproducing this with my "hello Chibi" application - so, no GFX here, just a shell. Anyway, I have recompiled it with -O2 but it did not help

[Giovanni]

Could you try one of the standard demos and see if it fails for you when enabling the FPU? this would allow me to test this on my side.

Giovanni

[russian]

Standard demo compiled with standard Makefile (with the addition of 'USE_FPU = yes') does work.

I have adjusted the Makefile to compile my project and it also works fine now with FPU. Historically I was always using Eclipse to compile my stuff, I will now try to figure out how to bring Eclipse compiler/linker preferences in sync with the Makefile and report which option was causing the issue.

PS: there is a typo in the makefile:
# Enable this if you really want to use the STM FWLib.
ifeq ($(USE_FPU),)
USE_FPU = no
endif

[Giovanni]

I am glad you resolved the problem and thanks for finding the typo.

Giovanni

[russian]

In order to fix Eclipse compilation, I had to explicitly define a global symbol "-DCORTEX_USE_FPU=TRUE" - the '#define CORTEX_USE_FPU TRUE' line in chconf.h without the global symbol was causing the issue.
I believe this was not the case with 2.4.3.

Is it expected behavior that this setting is expected to be defined as a global symbol and not just as a line in the chconf.h now?

[Giovanni]

It should work even in chconf.h if accompanied by the proper compiler options, CORTEX_USE_FPU is not sufficient.

Giovanni

[russian]

It should work even in chconf.h if accompanied by the proper compiler options, CORTEX_USE_FPU is not sufficient.

Giovanni

I am not sure what you are saying Here are my results (all other options are the same across all configurations)
2.4.3: works with only "#define CORTEX_USE_FPU TRUE" without "-DCORTEX_USE_FPU=TRUE"

2.6.0: unhandedException indefinite loop in case of "#define CORTEX_USE_FPU TRUE" without "-DCORTEX_USE_FPU=TRUE" (change in behavior comparing to 2.4.3)
2.6.0: works with "#define CORTEX_USE_FPU TRUE" AND "-DCORTEX_USE_FPU=TRUE"