Re: OFFTOPIC: (Re: 3.0 wishlist....)-->64bit

Noah Beck (
Fri, 23 Jan 1998 00:51:50 -0500

Adam McKee wrote:
> What about gcc? I'm not totally informed in these matters, but from what
> I've seen, there are many hoops a compiler must jump through to compile
> for the IA-64, especially with decent optimizations.

There was a Merced thread on this list a couple of months ago; I haven't
heard anything since.

> IA-64 takes the attitude of "let's offload lots of work to the compiler,
> and get rid of as much hairy logic on the chip as possible so we can
> devote logic to actually doing work..." which I really like as you might
> tell.

Compilers must become more sophisticated to deal with more advanced
architectures. You can extract much more instruction-level parallelism
that way.

> I'm sure IA-64 isn't nearly as full of original design ideas as Intel
> would have us believe -- they say it's not RISC, for example, but it sure
> looks like RISC.

I have to agree with them, it's not RISC. It's VLIW, but they gave it a
marketing name, EPIC. There's been considerable VLIW research from HP,
Intel's design partner in the project. As for originality, well, it
seems many architectural tricks used in modern processors were first
used in the 60's or 70's. IMO, Merced has more originality than I've
seen in a new processor architecture recently.

> Comments, corrections, flames?

Based on my experience, Microprocessor Report has amazingly accurate
descriptions of Intel projects. These two links provide very good
information, IMO:

(Oct. 97)

(Dec. 96)

I'm sure Linux developers' experience with 64-bit Sparc and Alpha will
make the 64-bitness of Merced a very minor porting issue. As long as
somebody (HP, Intel) is nice enough to port gcc to it, I don't see how
any other operating system has much of an advantage over Linux :)