Jon Masters

AppliedMicro launches first 64-bit ARM server chip

Eugene Crosser April 28, 2012 05:18

Did they change the command word size, or is it another hack a la x286? (shudder)

Michael K Johnson April 28, 2012 21:06

That wording seems to me to set up a false dichotomy.

There’s no need to double text size to no purpose. Also, given that it can run 32-bit programs as well as 64-bit programs, it would be even sillier to have 64-bit instruction words to waste 32 bits per word; do you really want to waste silicon in instruction decoding in order to double the size of text sections? Providing 64-bit addresses and 31 general-purpose 64-bit registers doesn’t sound like a hack to me. (Yes, LPAE is supported for backward compatibility. LPAE itself is the “hack” inasmuch as there is a hack; continuing support for it for compatibility is not per se a hack.)

I’m not an ARM expert by any means, but it doesn’t look like a hack to me.

Eugene Crosser April 29, 2012 08:17

It is just my understanding that quite often you need an offset encoded inside the command word. If the maximum value of this offset is much smaller than the addressable memory it quickly becomes a problem.

I may be very far off mark, my understanding of processor design is stuck in the ’80s.

Michael K Johnson April 29, 2012 08:23

Oh, as in global vs. local references. Given that even on x86-64 we are looking at an option for 32bit text and 64bit data, I have trouble worrying about the size of local references. The main reason you would want larger local references was if the instruction word length were longer… :-)

Eugene Crosser April 29, 2012 08:36

This is just my pet peeve because I have to deal with 24-bit-offset s360 code now… Not much, luckily.

Imported from Google+ — content and formatting may not be reliable