GitLab

This avoids some (mostly false) warnings about missing initializers.

Nearly all structs derived from amd64_attr_t has a size member anyway.
Simplify the code and move it to the base amd64_attr_t.

iropt: Implement 'computed_value_Proj_Builtin' for 'ir_bk_clz', 'ir_bk_ctz', 'ir_bk_ffs', 'ir_bk_parity' and 'ir_bk_popcount'.

Either the conv is used as the new result or it has further users.
In either case it is used afterwards and cannot be removed.

The mode of both nodes is the mode for the GP register class already.

'match_immediate_32()' in the meantime got a parameter 'must_match_ip_relative'.

The documented variable was removed in 2011.

This meant initializing the address mode variant all over the ia32
backend, on the other hand this will be useful if we ever switch the
ia32 backend to a style without NoReg nodes (like amd64).

We only use the immediate and scale yet on ia32.

After a SubSP the top end of the stack is used and not just unused
alignment padding anymore.

Only the backends know how memperm are implementd and whether they use
sp- or fp-relative addressing. This should fix sparc memperm problems.

Long-term I'd like to remove the memperm node and use a callback that
creates a sequence of target specific nodes instead (similar to how
spills and reloads are created).