By: Gabriele Svelto (gabriele.svelto.delete@this.gmail.com), January 23, 2017 12:53 am
Room: Moderated Discussions
anon (spam.delete.delete@this.this.spam.com) on January 22, 2017 2:24 am wrote:
> Also maybe the CPU designers don't like register windows.
I'm sure architectural considerations might have had a part in that. Register windows are probably bad enough on their own, but Fujitsu also had to use other kludges like kinda-variable length encoding of instructions to make room for their extensions in the opcode space.
That being said they already claimed in a number of their papers that they re-use most of the backend for both their SPARC and mainframe CPUs, so they probably figured out that an AArch64 front-end was a technically better option than the legacy SPARC since they didn't need binary compatibility.
> Also maybe the CPU designers don't like register windows.
I'm sure architectural considerations might have had a part in that. Register windows are probably bad enough on their own, but Fujitsu also had to use other kludges like kinda-variable length encoding of instructions to make room for their extensions in the opcode space.
That being said they already claimed in a number of their papers that they re-use most of the backend for both their SPARC and mainframe CPUs, so they probably figured out that an AArch64 front-end was a technically better option than the legacy SPARC since they didn't need binary compatibility.