On Mon, Sep 7, 2015 at 11:42 AM, Caleb Spare wrote:
I'm probably missing something, but I don't see a way to do the sorts of
things we're discussing in this thread without being *very* familiar with
the runtime and compilers.
Ah, the unsafe thing was more relevant to the table proposal. For
intrinsics, I would imagine that this new tool would emit either asm
code (in "go tool asm" format), or emit binary code directly that
followed the Go calling convention.

It's a separate compiler, but not a separate Go compiler.

Having said that, that's assuming that you can factor out the
SIMD-rich code from the other code. For image processing, I would
generally assume that to be the case. Yes, that means that this new
tool has to write the whole loop, not just one line in the inner loop,
but if you really care about performance, I'd expect that you (via
this tool) want want to consider e.g. registerization and other state
over the whole of the loop.

You received this message because you are subscribed to the Google Groups "golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Search Discussions

Discussion Posts


Follow ups

Related Discussions

Discussion Navigation
viewthread | post
posts ‹ prev | 19 of 36 | next ›
Discussion Overview
groupgolang-nuts @
postedSep 4, '15 at 1:00p
activeSep 13, '15 at 6:15a



site design / logo © 2021 Grokbase