Re: Overview Of New Intel Core i7(Nehalem) Processor

What the compiler does under the covers isn't relevant to the language. You can also cast a pointer to double, but it won't mean anything.

Cheers

Phil Hobbs

:) Good one but not really fair. Don't insult the 4004. It was one of the first.

Yeah. At least it had effective i/d space separation.

John

I never said it did. I said that you can pass an array to "&", which most definitely produces a pointer. I only mentioned pointer arithmetic to make it clear what the type of the pointer is (it's a pointer to the array type, not the element type).

No, I'm elaborating upon my original statement, as you seem to be unclear on what it said.

It's not a meaningful classification. Arrays and structures are no more like each other than they are like integers.

Note that passing a structure to a function *does* cause the whole thing to get copied onto the stack. It's only arrays which are treated differently.

No, trying to assign a structure to a pointer isn't a pointer conversion error, it's an incompatible type error. You can suppress a pointer conversion error using an explicit cast. You can't suppress an incompatible type error as the compiler has no idea as to how it could perform the conversion. It isn't simply that it isn't letting you, it's that it *cannot* do what you ask.

If you used a counter instead of just a latch to hold the "code page" value, you could make the counter increment on the fetch of the last instruction of a page and thus make it so that the code could run several K without having to worry about the boundaries in the straight line code.

ERROR!! C itself does not have anything to say about intermixing them, that is an implementation thing. Not does C itself have anything to say about any memory management hardware. I do hope you are not willfully conflating of the problems of poor target architectures and marginal implementations for what the language actually is.

If you meant the switcher, it only needs access to some hardware pointers. The swapper needs access to the memory management hardware and the DMA hardware and disk I/O.

Sure, but those really wide 9 and 10 bit counters are slow.

John

But most everybody does it that way.

Not does C itself have

I hope you are not assuming that just because a sensible compiler is possible that you can actually buy one.

John

Using those registers for their intended purpose you mean. Or you pretending an obvious lack of understanding of the content of the 8086 databook?

John, your prejudices are beginning to demean you.

On Jun 14, 3:05 am, "JosephKK" wrote: [...]

I was lumping the memory swapping and the task switcher together as one hunk. They are the only part(s) that need access to the whole space. Everything else can be confined to its little area.

DI is the Destination Index register; you're supposed to be able to use (i.e. modify) it within the function.

A lot of performance-critical code even repurposes BP (which is intended to remain fixed throughout the function) as a general-purpose register, and just indexes off SP for parameters and locals.

The only register whose "intended purpose" is as a stack pointer is SP, and that's intended to be used for both return addresses and data.

Sorry, but this makes no sense at all. Having a flat address space doesn't imply sections can't be protected!

They could be, but usually aren't. Hence the hundreds (thousands?) of buffer overflow exploits that have been sprung on Windows.

If either memcpy was safe, or if programmers used it properly... oops, sorry, I was fantasizing.

Is there a C compiler that doesn't push everything on the stack when doing a call, or that keeps a memcpy from trashing code?

John

things a

the

design.

Interesting, how do you make the jump or branch instructions work correctly (hit the right page)?

I suggest you recheck the operation of multiply, divide, and many I/O instructions again. Not to mention iteration constructions and the use of CX. Original 8086/8088, things started changing with the

On the PDP-8, the last instruction on a page would fall through to the first one on the next page. Direct jumps were only to the current page.

The 8088 did something funny with paging that allowed the DOS HIGH trick to work, locating DOS above the 1M boundary.

John

Trying to build decent compiler implementations on a poor to marginal hardware platform to target the same poor hardware platform causes many ugly compromises.