Update to 0.7 #61
Update to 0.7 #61
Conversation
test/scratch.jl
Outdated
| @@ -0,0 +1,189 @@ | |||
| NearestNeighbors.knn_point!(tree, [.1,.2,.3], false, Vector{Float64}(5), Vector{Int}(5), NearestNeighbors.always_false) | |||
There was a problem hiding this comment.
Accidentally committed this file?
There was a problem hiding this comment.
Oops, yes! Removing...
Codecov Report
@@ Coverage Diff @@
## master #61 +/- ##
==========================================
+ Coverage 93.89% 99.35% +5.45%
==========================================
Files 14 13 -1
Lines 508 311 -197
==========================================
- Hits 477 309 -168
+ Misses 31 2 -29
Continue to review full report at Codecov.
|
|
A question: should I revert the |
|
Nah, just keep it in the same :) |
|
Currently blocked on StaticArrays 0.7 support. |
KristofferC
force-pushed
the
no-reinterpret
branch
from
bfbe593
to
a0a77af
Compare
KristofferC
changed the title
reinterpret, make copies of data instead
|
Updated this. I couldn't get the |
KristofferC
force-pushed
the
no-reinterpret
branch
from
a0a77af
to
900dc50
Compare
|
No worries, fine with me!
… Am 27.06.2018 um 14:22 schrieb Kristoffer Carlsson ***@***.***>:
Updated this.
I couldn't get the reorder thing to work with the data free trees so, for now, I removed it. @rened Is this feature important to you?
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or mute the thread.
|
Fix for #59, alternative approach to #60
Since
ReinterpretArrayseems to have an unacceptable indexing overhead here (see #60 and https://discourse.julialang.org/t/big-overhead-with-the-new-lazy-reshape-reinterpret/7635/1), @KristofferC advices tocopyinstead ofreshape(reinterpret(data...),...).This PR implements two parts required for this approach:
1- A
@noinline reinterpret_or_copyfunction that usesreinterpretfor julia versions prior to the introduction ofReinterpretArraylazy wrappers, and makes a reinterpreted copy otherwise.2- Change from the deprecated uninitialized contructor
Array{T}(::Int...)to the new formArray{T}(uninitialized, ::Int...). Not doing this introduced a very large performance regression under julia master. To remain backward compatible I used@compat, which adds a dependency to Compat.jl.With this, the timings for a few tests I tried were identical to the ones in the master branch under v0.6.