Assembler component doesn't recognize rsqrtps

[Borogove]

Don't know if this should be in bugs or feature-requests, but the Assembler seems to parse sqrtps (square root) and rcpps (reciprocal approximation) just fine, but not rsqrtps (reciprocal square root approximation). sqrtps takes up to 134 cycles to complete, but with rsqrtps and a little bit of arithmetic to refine the approximation, you can get a good audio-quality square root in only about 20 cycles.

[infuzion]

Interesting... could you roll your own rsqrtps, or does it just have to be opend in SM's Assembly?

[Borogove]

Interesting... could you roll your own rsqrtps, or does it just have to be opend in SM's Assembly?

As noted, I'm after the performance that I can only get from rsqrtps (a single machine instruction that there's currently just no way to get in SM -- unless the assembler module has an undocumented "emit" pseudo-op). While I'm waiting, I can use sqrtps instead, but as noted it's substantially slower.

[rl]

This needs to be bumped. Almost 6 years gone, still not implemented!

Stupid restriction all over the place, maybe it's time for a hexcode assembly module?

[infuzion]

Yep a request, I just added it here for you: viewtopic.php?f=5&t=10775