On Thu, Jan 9, 2025, at 21:10, Maciej W. Rozycki wrote:
> On Thu, 9 Jan 2025, Eric W. Biederman wrote:
>
>> > So, this would be the 100% correct for alpha then which would not loose
>> > any functionality even for 32-bit binaries?
>>
>> I don't think it is correct to think about 32-bit binaries on alpha.
>>
>> Alpha never had a 32bit instruction set. But at some point it looks
>> like binaries that could not handle more than 31 bits of address
>> space got ported and someone implemented a work-around. I guess this
>> is the --taso option that Arnd mentioned.
>
> This also saves some code space in non-PIE and plain static executables
> as it takes fewer machine instructions to load a 64-bit address that is
> known beforehand to be a sign-extended 32-bit value.
>
> This is similar to the MIPS n32 ABI, which also implies a 32-bit address
> space while still using 64-bit registers for everything, starting from
> stack slots (it's also ILP32 with the `long long' C data type only making
> proper use of the full width of the CPU registers, while Alpha's --taso
> ABI is I believe IP32 (?) with the plain `long' C data type still 64-bit,
> just as with the regular LP64 ABI).
I'm pretty sure it's still LP64 on Alpha Linux with gcc. There is an
-mpointer-size=32 option in gcc for VMS, but I don't see anything like
that in Linux. The only thing that is implemented here is the option
for the linker that sets the EF_ALPHA_32BIT bit, but none of the
code generation takes advantage of the upper bits being zero.
Arnd
