Document _Alignof(max_align_t) about riscv-elf-psabi-doc HOT 3 CLOSED

Which exact SysV ABIs are you referring to here? Give a URL please.

max_align_t greater than the stack alignment is problematic. Our stack alignment for ilp32{,f,d} and lp64{,f,d} is 16, and it is 4 for ilp32e. GCC reports 16 for _Alignof(max_align_t) for all of these ABIs, including ilp32e, but we already know that the ilp32e ABI is broken and are planning to fix it with a new EABI. Anyways, documenting it as 16 looks correct.

from riscv-elf-psabi-doc.

Which exact SysV ABIs are you referring to here?

The x86-64 psABI and the Intel386 psABI, I'll see if I can find a link.

GCC reports 16 for _Alignof(max_align_t) for all of these ABIs, including ilp32e,

Well, GCC has to do that for the Intel386 psABI. _Alignof(max_align_t) is defined as the largest alignment requirement of a "fundamental" type, and the Intel386 psABI supports the fundamental type long float which has an alignment requirement of 16. So AFAICT GCC hands are tied.

The only way to lower the value of _Alignof(max_align_t) for an ABI is, per definition, to lower the largest alignment requirement of the fundamental types in that ABI.

from riscv-elf-psabi-doc.

I would suggest
http://refspecs.linuxfoundation.org/elf/
which has multiple SVR4 ABIs including the i386 and x86_64 ones. I don't see a reference to max_align_t in the x86_64 one, but I need some time to look at this.

The i386 port does stack aligning when necessary, but the RISC-V port does not, which is why the ilp32e ABI is broken.

Anyways, this suggestion is uncontroversial. 16 is correct. Assuming we need to document this.

from riscv-elf-psabi-doc.