[fpc-devel] "Blank slate" next version of FPC

[Jonas Maebe]

On 23/02/2019 01:30, J. Gareth Moreton wrote:
> Given that the reference is an offset from %rbp, the compiler will know 
> for sure that the variable is local, so there won't be any problems with 
> multi-threading that prevents a lot of temporary register storage for 
> safety reasons.

Unpacking records could be done if their address is not taken anywhere. 
The compiler already supports keeping records in registers under the 
same conditions, but only up to a certain size (two native registers) 
and only using the same layout as they would have in memory (so no 
explicit splitting out fields into separate registers).

On x86, it does this less often than on other architectures, because 
that architecture does not have efficient instructions to extract/insert 
bits from/into registers. However, one improvement that could be made 
there in the compiler is to ignore this fact if every field of the 
record already corresponds to exactly one register.


Jonas