Article: Why Apple Won’t ARM the MacBook
By: David Kanter (dkanter.delete@this.realworldtech.com), May 10, 2011 1:18 am
Room: Moderated Discussions
none (none@none.com) on 5/10/11 wrote:
---------------------------
>There are some mistakes about ARM:
>
>1. except for NEON, decoding is easy, even for detecting
>2-byte vs 4-byte instructions (which are the only sizes
>available in Thumb mode)
>
>2. Corte-A15 is Eagle and vice versa :-)
Thanks for that correction.
>3. the implicit barrel shifter indeed is an oddity, but it is
>no real issue: either split the instruction or add a pipe
>stage
Still ugly.
>4. in ARM mode (as opposed to Thumb mode) setting condition
>flags is an option
Yes, I got that.
>5. Thumb initially dropped the shifter and "predication" (I
>certainly don't like that term, it has become tainted, I
>prefer "conditional execution" :-) to fit in 16-bit opcodes,
>(32-bit Thumb came later) not to ease micro-architecture
>design.
Predication/conditional execution. It's all the same. You can argue predication is more for SIMD architectures, but I'm not sure it matters what you call it. The point is that you now have at least 3 inputs for most operations...
David
---------------------------
>There are some mistakes about ARM:
>
>1. except for NEON, decoding is easy, even for detecting
>2-byte vs 4-byte instructions (which are the only sizes
>available in Thumb mode)
>
>2. Corte-A15 is Eagle and vice versa :-)
Thanks for that correction.
>3. the implicit barrel shifter indeed is an oddity, but it is
>no real issue: either split the instruction or add a pipe
>stage
Still ugly.
>4. in ARM mode (as opposed to Thumb mode) setting condition
>flags is an option
Yes, I got that.
>5. Thumb initially dropped the shifter and "predication" (I
>certainly don't like that term, it has become tainted, I
>prefer "conditional execution" :-) to fit in 16-bit opcodes,
>(32-bit Thumb came later) not to ease micro-architecture
>design.
Predication/conditional execution. It's all the same. You can argue predication is more for SIMD architectures, but I'm not sure it matters what you call it. The point is that you now have at least 3 inputs for most operations...
David