C++ temporary object storage

I'm using 4.92014Q4 compiler.

Jonathan

Hi Jonathan,

I tried your example with the 4.9 2014Q4 toolchain release and see the same sub sp, #52 at the beginning no matter how many call to A when compiling with -O3 or even -O2. To compile I had to only declare A via a prototype otherwise GCC would optimize it to 0. Is there something I missed? Can you share with us the command line you used to compile as well as the assembly output you get?

Best regards.

hi Thomas,

Here the information requested­ .

Here the full code source. I have changed it to simplify my test.

#include <cstdio>
#include <cstring>
#include <cstdint>
void C();
void D();
void E();

class myObject
{
private:
 char buf[8];
public:
 myObject(const char *p)
 {
  strcpy(buf, p);
 }

 const char *c_str() const
 {
  return buf;
 }

};

uint32_t * A(const myObject &a, const myObject &b);
uint32_t * B(const char *a, const char *b);

int main(void)
{
   C();
   D();
   while(1);
}

uint32_t * A(const myObject &a, const myObject &b)
{
 printf("%s%s", a.c_str(), b.c_str());
 return 0;
}

uint32_t * B(const char *a, const char *b)
{
 printf("%s%s", a, b);
 return 0;
}

void C()
{
 myObject t("t"), tt("tt");
 A(t, tt);
 A(t, tt);
}

void D()
{
 B("t", "tt");
 B("t", "tt");
}

void E()
{
 A("t", "tt");
 A("t", "tt");
 A("t", "tt");
 A("t", "tt");
}

The compiling line.

arm-none-eabi-g++ -mcpu=cortex-m3 -mthumb -Og -fsigned-char -ffunction-sections -fdata-sections -fno-inline-functions -fstack-usage -Wunused -Wuninitialized -Wall -Wextra -Wmissing-declarations -Wconversion -Wpointer-arith -Wpadded -Wshadow -Wlogical-op -Waggregate-return -Wfloat-equal -g3 -std=gnu++11 -fabi-version=0 -fno-exceptions -fno-rtti -fno-use-cxa-atexit -fno-threadsafe-statics -Wabi -Wctor-dtor-privacy -Wnoexcept -Wnon-virtual-dtor -Wstrict-null-sentinel -Wsign-promo -Wa,-adhlns="src/main.o.lst" -MMD -MP -MF"src/main.d" -MT"src/main.o" -c -o "src/main.o" "../src/main.cpp"

I have done some more test and its look like the reuse of the stack is only done when the optimisation is in -O2. and -O3.

I try all the flag that are activated when O2 is used and no one is produce the correct stack reuse. It's probably an internal optimisation that is use.

Do you think that this can be activated with a specific flag ?

Regards

Jonathan

Hi Jonathan,

You can compile with -Q -O2 --help=optimizers and then the same with -Og and diff the result. This will tell you all the difference between Og and O2 (the manual might not be up-to-date). My guess is that this optimization is due to -foptimize-sibling-calls.

Best regards.

Hi,

I try the optimisation flag and this change nothing on the stack usage. In the curent code, the SP is reserving #64 bytes on E() method.

I have already did this for the diff with the -Q command. And no flags seam to reproduce the optimisation level -O2. Is it possible that -O2 add some optimizer that cannot be enabled with flags ?

Regards

Jonathan

Indeed, I tried myself and activating all pass from -Og doesn't change a thing. Yes some optimization are not part of a pass and can thus not be enabled manually. In this case it might also be that an optimization is *disabled* by -Og rather than *enabled* by -O2. It seems that your only choice here is to use -O2.

Best regards,

Thomas

Hi,

I think for an embedded compiler this should be always enabled. Event with no optimisation at all. I checked and in C, the same problem is visible for small scope variable. The stack pointer is initialised to the total amount of variable this is used by a function, event if the variable is within a small scope.

Do you think that this can be enabled by default on this toolchain ?

Regards

Jonathan

Hi Jonathan,

Running a compiler without any optimization enabled is useful to have a fast development and debug cycle. In both these cases you will want the compilation to be as fast as possible as you will compile very often. There is also the case of running the compiler as a commit hook for instance. Therefore I don't think any optimization should be enabled by default (-O0).

Best regards.

Hi,

That look logical to me that with -O0 should not do any optimisation. But if we can enable this feature with a flag will be very usefull.

In my situation, we use an RTOS and when we compile in debug mode, we need to have 3 times the stack defined. This is anoying for us.

First I think that using -fstack-reuse=reuse-level https://gcc.gnu.org/onlinedocs/gcc/Code-Gen-Options.html should do this, but it doesn't look so.

Is this a bug ?

Regards

Jonathan