Is this because of the affordances of the language?
-
-
-
It’s because GLSL puts a lot of effort into ergonomics of vector code. The syntactic sugar runs in the right direction: you want to write vector code because it’s easy, not just because it’s fast.
- 2 more replies
New conversation -
-
-
isn't the pain semi-unavoidable without hardware support to hide the lane width?
-
I actually mostly think that 128 bits is enough for anyone.
- 3 more replies
New conversation -
-
-
I've spent the past 3 weeks or so SIMD-ifying my path tracer, and I feel this. Though I think myself and
@Lokathor are helping this situation in Rust somewhat with https://docs.rs/wide and https://docs.rs/ultraviolet ;D -
I’d like to plug pathfinder_simd too :)
- 1 more reply
New conversation -
-
-
It's not. Use ISPC.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Typically, the inherent issue is that language does not map well to what the hardware actually capable of doing well. Modern C++. Hoping that a compiler that is capable optimising your code will appear in the future is not a paradigm. ISPC is a good example. And its story.
- 1 more reply
New conversation -
-
-
But GLSL is usually scalarized and then vectorized over the problem space, so the elements of the vector are really four different iterations of the loop over the input vertices or output fragments. Using vector types for SIMD would just be manual loop unrolling.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.