Coalesced non-commutative reduction using shfl

[tkf]

No description provided.

[codecov]

Codecov Report

Merging #78 (3431a79) into master (a1faf25) will decrease coverage by 2.68%.
The diff coverage is 45.32%.

@@            Coverage Diff             @@
##           master      #78      +/-   ##
==========================================
- Coverage   51.20%   48.52%   -2.69%     
==========================================
  Files           2        3       +1     
  Lines         166      305     +139     
==========================================
+ Hits           85      148      +63     
- Misses         81      157      +76     

Flag	Coverage Δ
Pkg.test-cpu	0.00% <0.00%> (ø)
Run.test-cpu	0.00% <0.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/shfl.jl	44.52% <44.52%> (ø)
src/api.jl	90.00% <100.00%> (+2.50%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a1faf25...3431a79. Read the comment docs.

[tkf]

julia> xs = CUDA.ones(2^30);

julia> @btime sum(xs);
  7.060 ms (1659 allocations: 28.08 KiB)

julia> @btime Folds.sum(xs);
  159.392 ms (78965 allocations: 1.21 MiB)

julia> @btime Folds.sum(xs, CoalescedCUDAEx(); init = 0.0f0);
  8.117 ms (2697 allocations: 44.95 KiB)