RzBuffer is inefficient (only on hot paths)

[Rot127]

Is your feature request related to a problem? Please describe.

All RzBuffer related functions which use the internal callbacks are inefficient on hot paths.
This makes RzBuffer unusable if many reads happen on it.

The reason is that the internal buffer uses different callbacks for different buffer types (read(), write(), get_size(), get_whole_buffer() etc.). This makes sense, but prevents the compiler from inlining functions.

Even the simplest functions which only return a pointer, consumes enormous CPU time (in my case 20-30%), because the function pointers are de-referenced every single time.
The compiler can't optimize them away.

Describe the solution you'd like

Move away from function pointers and replace them with switch cases directly calling the functions.
Or implementing "performance" versions of them. But this means double amount of maintenance.

Describe alternatives you've considered

Using raw ut8* + size. Which we want (?) to get away from.

Additional context
None

[XVilka]

RzBuffer is quite nice, I suggest adding something like:

if (buf->type == RZ_BUFFER_BYTES || buf->type == RZ_BUFFER_MMIO) then {
     void *ptr = buf->get_raw_ptr()
} else {
     ... slow path
}

[wargio]

that should be done only internally

[tushar3q34]

RzBuffer is quite nice, I suggest adding something like:

if (buf->kind == MEMORY) then {
     void *ptr = buf->get_raw_ptr()
} else {
     ... slow path
}

Can you please elaborate this? What exactly is buf->kind == MEMORY, is it a new type in RzBufferType?
Thank you.

[XVilka]

RzBuffer is quite nice, I suggest adding something like:

if (buf->kind == MEMORY) then {
     void *ptr = buf->get_raw_ptr()
} else {
     ... slow path
}

Can you please elaborate this? What exactly is buf->kind == MEMORY, is it a new type in RzBufferType? Thank you.

Yes, I updated my pseudocode example to be more precise.