Strange SIZE output for -march=rv32ec_zicsr

[sadkotheguest]

I'm trying to compile my project with last x-pack GCC for CH32V003 MCU.

set(CPU_FLAGS "-march=rv32ec_zicsr -mabi=ilp32e -fmessage-length=0 -msmall-data-limit=0 -msave-restore -fsigned-char -fno-common")
and
set(LD_FLAGS "-march=rv32ec -mabi=ilp32e -Wl,--gc-sections -Wl,--print-memory-usage -nostartfiles")

And always got 100% RAM usage.

When I'am switching to old one GCC 8.2 (from Mounriver IDE) with

set(CPU_FLAGS "-march=rv32ecxw -mabi=ilp32e -fmessage-length=0 -msmall-data-limit=0 -msave-restore -fsigned-char -fno-common")
and
set(LD_FLAGS "-march=rv32ecxw -mabi=ilp32e -Wl,--gc-sections -Wl,--print-memory-usage -nostartfiles")

everything looks ok.

---

Could you tell where should I dig to make the SIZE (%-age of RAM) work well?

[ilg-ul]

I suggest you configure the linker to produce the memory map with both compilers and compare the two files, perhaps you can identify some differences.

Anyway, if you use the regular newlib, 2K of RAM might not be enough for the most recent releases. One possible workaround is to try newlib-nano.

[ilg-ul]

I think you can update the .ld file to define the correct stack symbols without actually using that top memory, so size should report only up to the heap.

[sadkotheguest]

Ok, thanks for your advice.

[sadkotheguest]

@ilg-ul
When I am using CH32V307 with the same .ld script (-march=rv32imafc -mabi=ilp32f) & the last one GCC 12.2 - SIZE utility works well.

[ilg-ul]

I have no idea why.

Try to run manually size from 8.x on both binaries (8.x and 12.x), and similarly size from 12.x on both binaries, and compare the results. If the same binary results in different values, it is a problem with size, which you should report to GNU.

[sadkotheguest]

Great idea, thanks.

[sadkotheguest]

I found that the subj is the output of the LD with --print-memory-usage directive.

.stack ORIGIN(RAM) + LENGTH(RAM) - __stack_size :
    {
        . = ALIGN(4);
        PROVIDE(_susrstack = . );
        . = . + __stack_size;
        PROVIDE( _eusrstack = .);
    } >RAM 

and

.stack :
    {
        . = ALIGN(4);
        PROVIDE(_susrstack = . );
        . = . + __stack_size;
        PROVIDE( _eusrstack = .);
    } >RAM 

outputs are different :(

[ilg-ul]

Please note that removing the expression moves the stack right after bss, leaving no space for the heap.

I do not recommend such a change.

[ilg-ul]

I also suggest to use the gnu utility size, instead of the linker with that option.

[sadkotheguest]

Thanks for your advices.