Why would Apple Not tweak the A4's instruction set?
Although Apple's A4 processor in the iPad/iPhone 4 is based on the ARM RISC-type processor, it appears it's instruction set is extensible. With Apple's purchase of PA Semi and Intrinsity, I figured it makes sense that Apple would take advantage of the golden opportunity to tweak their custom silicon for improved performance of Objective-C and the many software layers that ride on it to really "enhance the user experience." But I've seen no mention of the possibility anywhere. Does anyone have any insights into what's synergistically possible between the hardware and software?
5 comments
Adding coprocessors on the same silicon is one thing, but changing the instruction set is a little extreme. They license a core from ARM and changes to that core would be more costly then adding some custom silicon to the chip.
Let's take an open source software analogy. I got this wonderful code from a group that is really picky about patches but has a good plug-in architecture.
I can change the core code. When new software shows up, I have to try and re-patch the software to add back in my core changes. I might try to submit my changes to the group, but if they are not broadly useful, they won't get adopted.
I can add my stuff through the plug-in architecture. It is a known expansion path, cheaper, and has less of a chance of breakage due to it being the preferred, documented method. Sure, when new software shows up, I will probably need to make some changes, but it will be easier.
Let's take an open source software analogy. I got this wonderful code from a group that is really picky about patches but has a good plug-in architecture.
I can change the core code. When new software shows up, I have to try and re-patch the software to add back in my core changes. I might try to submit my changes to the group, but if they are not broadly useful, they won't get adopted.
I can add my stuff through the plug-in architecture. It is a known expansion path, cheaper, and has less of a chance of breakage due to it being the preferred, documented method. Sure, when new software shows up, I will probably need to make some changes, but it will be easier.
I believe it is common for ARM licensees to add accelerators for multimedia, etc.
However, extending the processor itself goes against the entire RISC philosophy. Making the core more complicated makes speeding it more difficult and will not produce significant gain unless the additional instructions are very common.
A more complicated and less orthogonal instruction set also makes the compiler's job much more difficult.
However, extending the processor itself goes against the entire RISC philosophy. Making the core more complicated makes speeding it more difficult and will not produce significant gain unless the additional instructions are very common.
A more complicated and less orthogonal instruction set also makes the compiler's job much more difficult.
I'm wondering then what are the benefits of Apple doing it's own silicon? Perhaps it would be the obvious and what they've actually told us... making the circuitry massively smaller so they can fit in a bigger battery. Hmmmm.
This is very interesting. It is probably easy to find out by running a decompiler on code taken from the iPhone 4 or iPad.
I think Apple's ultimate goal is to dump an outside CPU altogether and create its own.