C++ Logo

SG12

Advanced search

Subject: Re: [ub] Justification for < not being a total order on pointers?
From: Roger Orr (rogero_at_[hidden])
Date: 2013-10-19 06:58:44


Nevin Liber wrote:

> But, even if segmented architectures, unlikely though it is, do come
> back

It might be worth remembering that the x86 and x64 architectures *are*
segmented.

However the operating system does a good job of hiding this from the
application programmer who (normally) treats the address space as flat - the
user-mode segment registers (CS, DS) are set up with a base address of zero.

However, even in application space, Windows for example uses the fs register
to point to the current thread environment block.
(It is a minor nit that you can't access this address using fs from C++.)

I would hesitate to make decisions now that would preclude future use cases
for segment registers.
[ For example, OTTOMH, mapping to GPU memory, or managing NUMA extensions ?
]

Regards,
Roger.


SG12 list run by herb.sutter at gmail.com