Since you have the code handy, can you provide a working example that includes a benchmark?

When n is negated, if that's the style chosen to re-write the loop, then look out for n starting off as the smallest possible value as two's complement representation is asymmetrical.

Isn't the decrement of i followed by a jz (or jnz) another way to write this.
for i := n; i != 0; i-- { }

Writing assembly for various architectures, a pre-decrement test against zero is the normal way I'm used to seeing, yes. But I wanted to warn against a flaw in the OP's suggestion.

Something else to ponder in implementing this... I hear some folk use a debugger; might they try and inspect the loop's variable and be confused that it's changed sign, or running in the opposite direction? Or perhaps DWARF can represent the transformation for the debugger to reverse? If other compilers have implemented this for other languages then they might give ideas.

OTOH, stepping through C code that's been built withgcc -O3 is pretty painful. It's fine as long as you're reading the assembly and not trying to match it up to the C code that produced it.

Ok, here is a simplified and extracted example. I called it "loop_test.go"

https://play.golang.org/p/uRdymTPtTCi

Alas, i don't see such a big difference in this case. The code is not the same, not so much register pressure.

Summary: maybe this is not so important.

I think it's still worth trying to measure gains across more tests, especially as some platforms have more register pressure than others. But I've also a niggling recollection that this has been suggested before and turned down because of debugger confusion, probably by @rsc. :-)