Yagarto GCC and Cortex M3

Tony Whitehall wrote:
> I am writing a program that need floating point maths. So I have
> included libgcc in my build.
I suggest to "include" libm too (option -lm)

>I keep  on receiving the following Error
> message from the linker,"libgcc.a uses VFP whereas xxx.elf does not".
Verify that the target-options are also given for linking (-mthumb 
-mcpu-cortex-m3), don't call the linker (*-ld) directly but use 
compiler-frontend (*-gcc).

> Clifford Slocombe said that in default GCC compile the code with
> -mhard-float. Yet when I compile the code without -msoft-float , I still
> receive the above message. If I compile with -mHard-float, I get
> hard-float conflict with specified fpu.
Should not be needed for your target.

> My question is do I have to rebuild libgcc.a ? Clifford Slocombe seem to
> suggest it is not necessary.
I don't think so.

> If I need to build libgcc.a, how do I do it ? I have installed minGW and
> MSYS but I don't know how to use it with Yagarto.
AFAIK Michael Fischer provides buildscripts. Cygwin might be a better 
environment to build a toolchain. With the -nocygwin compiler-option 
binaries can be generated which can work without cygwin-DLLs on any 
MS-Windows system. But I expect building/updating the toolchain is not 
needed.

Hi Martin,

Thank you for your response.

>> I am writing a program that need floating point maths. So I have
>> included libgcc in my build.
>I suggest to "include" libm too (option -lm)
I am Sorry I didn't mention that I also include libc and Libm.

>>I keep  on receiving the following Error
>> message from the linker,"libgcc.a uses VFP whereas xxx.elf does not".
>Verify that the target-options are also given for linking (-mthumb
>-mcpu-cortex-m3), don't call the linker (*-ld) directly but use
>compiler-frontend (*-gcc).

I am using  LD = $(GNUTOOLS)/arm-elf-gcc, I believe that is using the 
compiler front.

After your response I include -mthumb and mcpu-cortex-m3 as part of the 
linking option. Unfortunately, it make no difference.

It seem my best solution is to avoid using int64_t variables or make use 
my own int64_t divide and mod , as the compiler uses link in libgcc for 
these functions. Is this correct ?

>> Clifford Slocombe said that in default GCC compile the code with
>> -mhard-float. Yet when I compile the code without -msoft-float , I still
>> receive the above message. If I compile with -mHard-float, I get
>> hard-float conflict with specified fpu.
>Should not be needed for your target.

>> My question is do I have to rebuild libgcc.a ? Clifford Slocombe seem to
>> suggest it is not necessary.
>I don't think so.

>> If I need to build libgcc.a, how do I do it ? I have installed minGW and
>> MSYS but I don't know how to use it with Yagarto.
>AFAIK Michael Fischer provides buildscripts. Cygwin might be a better
>environment to build a toolchain. With the -nocygwin compiler-option
>binaries can be generated which can work without cygwin-DLLs on any
>MS-Windows system. But I expect building/updating the toolchain is not
>needed.

I have no experience in using cygwin. I am chicken out, so for the 
moment I will try to avoid using libgcc or implement 64 bits divide and 
mod.


Best Regards
Tony

Hi Tom,

You response hit the nail right on the head. The only option I didn't 
try was -mfpu=vfp. When I compiled with -mfpu=vfp, I can now link with 
libgcc.a

Thank you, Tom and Martin for your help.

Best Regard,
Tony

Hallo,

I'm having the same problem. Up to now I used a custom build gcc 
toolchain, but maintaining it is too much an efford, so I decided to 
switch to Yagarto.

My Makefile looks like this:

1

A_FLAGS = -mcpu=cortex-m3 -mthumb -Wa,-gdwarf-2

2

C_FLAGS = -mcpu=cortex-m3 -mthumb -O0 -gdwarf-2 \

3

          -mfpu=vfp -msoft-float \

4

          $(patsubst %,-I%,$(INC_DIRS)) -I. \

5

          -std=gnu99 \

6

          -ffunction-sections -fdata-sections \

7

          -Wall -Wextra -Wimplicit -Wcast-align -Wpointer-arith \

8

          -Wredundant-decls -Wshadow -Wcast-qual -Wcast-align -Wnested-externs \

9

          -Wno-cast-qual -Wno-missing-prototypes  -Wno-strict-prototypes -Wno-missing-declarations

10

L_FLAGS = -T stm32f10x_md.ld \

11

          -nostartfiles -Wl,--gc-sections \

12

          -lc -lm -lgcc \

13

          -mfpu=vfp -msoft-float

As you can see, I tried -mfpu=vfp -msoft-float, but it did not work for 
me. Maybe there is still something wrong.

Current Yagartos arm-elf-gcc was build without the configure option 
"-with-float=soft", maybe that has something to do with it.

Is there anybody, who uses Yagarto for Cortex-M3 an can provide me his 
Makefile?

Any other suggestions?

Kind regards, Oliver.