Unexpected result using division with signed integers

Ian, I couldn't reproduce it. This is such a simply case that I doubt
compiler can do it wrong.

Can you please post the disassembled function for me to diagnose?

Joey
On Sep 6, 2013 7:53 PM, "Ian Duffy" <email address hidden>
wrote:

> New question #235262 on GCC ARM Embedded:
> https://answers.launchpad.net/gcc-arm-embedded/+question/235262
>
> The following code produces surprising results
> {
> signed int a = -9;
> unsigned int b = 3;
> signed int c = a / b;
> signed short d = a / b;
> }
>
> Expected c = -3, d = -3
> Get c = 1431655762, d = 21842
>
> Clearly the compiler is casting a to an unsigned int before doing the
> division and then casting the result back.
>
> However, regardless of the value of a or b, the result can always fit in a
> signed int with no overflow so there should be no problem always producing
> the correct answer.
>
> In pseudo code I'd expect the following to happen:
>
> signed int divide(signed int a, unsigned int b)
> {
> if b > abs(a) return 0;
> return a / (signed int)b;
> }
>
> I don't know if this is a bug or a subtle trap in the C spec ready to
> catch the unwary.
>
> For the record I'm using
> arm-none-eabi-gcc -mcpu=cortex-m4 -mthumb -g ...
>
>
>
>
>
>
>
> --
> You received this question notification because you are an answer
> contact for GCC ARM Embedded.
>

My test function looks like this:

signed int test(void)
{
 signed int a = -9;
 unsigned int b = 3;
 return a / b;
}

The assembler generated using -S switch looks lke this:

test:
.LFB110:
 .loc 1 46 0
 .cfi_startproc
 @ args = 0, pretend = 0, frame = 8
 @ frame_needed = 1, uses_anonymous_args = 0
 @ link register save eliminated.
 push {r7}
.LCFI0:
 .cfi_def_cfa_offset 4
 .cfi_offset 7, -4
 sub sp, sp, #12
.LCFI1:
 .cfi_def_cfa_offset 16
 add r7, sp, #0
.LCFI2:
 .cfi_def_cfa_register 7
 .loc 1 47 0
 mvn r3, #8
 str r3, [r7, #4]
 .loc 1 48 0
 mov r3, #3
 str r3, [r7, #0]
 .loc 1 49 0
 ldr r2, [r7, #4]
 ldr r3, [r7, #0]
 udiv r3, r2, r3
 .loc 1 50 0
 mov r0, r3
 add r7, r7, #12
 mov sp, r7
 pop {r7}
 bx lr
 .cfi_endproc

A bit more context ..
I'm compiling using CooCox IDE on Windows 8.
Tool chain path is "C:\Program Files (x86)\GNU Tools ARM Embedded\4.7 2013q1\bin"
My target is a STM32F407VG

I didn't include the full command in my earlier post as I thought some of the options were irrelevant, but just in case I'm mistaken here it is (apart from the include paths and filenames)
 arm-none-eabi-gcc -mcpu=cortex-m4 -mfpu=fpv4-sp-d16 -mfloat-abi=hard -mthumb -Wall -ffunction-sections -g -O0 -std=c99 -S -c -DSTM32F407VG -DSTM32F4XX -DUSE_STDPERIPH_DRIVER -DUSE_USB_OTG_FS -D__FPU_USED

The project I'm currently working on is quite large with lots of libraries. I'll try to create a bare-bones project with as little functionality as possible to reproduce the same behaviour.

Just to confirm that it also happens in a brand new project with just start up code and this function.

Thanks. I thought it would be surprising for the compiler to be wrong too :-)