The "llvm.mem.parallel_loop_access" is an annotation on loads and stores that indicate they do not depend on other iterations. @simd causes them to be sprinkled throughout the loop when the LLVM IR is generated. The lack of "load <*n* x float>" indicates that the LLVM vectorizer gave up. I'm not sure what spooked it. The code looks like it should have vectorized. I'll investigate.
- [julia-users] Article on `@simd` Arch Robison
- Re: [julia-users] Article on `@simd` Elliot Saba
- Re: [julia-users] Article on `@simd` Elliot Saba
- Re: [julia-users] Article on `@simd` Patrick O'Leary
- Re: [julia-users] Article on `@simd` Jacob Quinn
- Re: [julia-users] Article on `@sim... Arch Robison
- Re: [julia-users] Article on `... David Smith
- Re: [julia-users] Article on `@simd` Stefan Karpinski
- Re: [julia-users] Article on `@simd` Arch Robison
- Re: [julia-users] Article on `@simd` Jacob Quinn
- Re: [julia-users] Article on `@simd` Arch Robison
- Re: [julia-users] Article on `@sim... Jacob Quinn
- Re: [julia-users] Article on `... Arch Robison
- Re: [julia-users] Article on `... Job van der Zwan
- Re: [julia-users] Article ... Stefan Karpinski
- [julia-users] Re: Article on `@simd` Gunnar Farnebäck
- Re: [julia-users] Re: Article on `@simd` Arch Robison
- Re: [julia-users] Re: Article on `@simd` Arch Robison
- Re: [julia-users] Re: Article on `@simd... Uwe Fechner
- Re: [julia-users] Re: Article on `... Arch Robison
- Re: [julia-users] Re: Article ... gael . mcdon