Repository navigation
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #117 +/- ##
==========================================
+ Coverage 96.63% 96.96% +0.33%
==========================================
Files 17 17
Lines 2052 1881 -171
==========================================
- Hits 1983 1824 -159
+ Misses 69 57 -12 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
d49077f to
0555eba
Compare
|
Ok, the latest time benchmarks (under "bench summary") look pretty good, although the GitHub runner used for benchmarking is likely very noisy. |
4419b1f to
69d233e
Compare
f633eed to
2135e25
Compare
|
Benchmarks on this PR now almost universally beat master. |
(i)modwt_step!, improve modwt fix gpu+cpu `modwt_step` tests
* use isnothing, new julia 1.10 base functions * format benchmarks.jl * format FILTERS and SCHEMES
- delete failed tests because of a bug in Base on 1.10, coz no point testing a base function in a package... - readability, `count`, fix undefined variable error - up/downsample bool - improve denoise, nspin2circ, threshold - bitshifts where appropriate, rounding_div4 - avoid mutating matchingpursuit input testfunction sinpi
shr for detailn trailing_zeros for `maxtransformlevels` ndigits
* steptype Union (predOrUp) * indicate `filtermainloop...!`s are mutating
make lrange type stable use more UnitRanges
move some inbounds
`@Const(bases)`, argument formatting divrem move if into loop and L >= 0 don't set workgroup size gpu.jl testset for-loop bitvector
reduce usage of `unsafe_vectorslice` in filter dwt code modify `unsafe_dwt1level!` reduce allocations from `dwt_transform_cols!` reduce allocations from `dwt_transform_strided!` checkbounds in stridedcopy!
* annotate y with AbstractVector
* oopv::Union{Vector{T},Nothing}
* half in normalize
* split! and merge! checks
* stepj
* internal docs
* inline dwt1
mostly Integer -> Int
2135e25 to
d17763d
Compare
|
Ok benchmark looking good. Is this ready for review? Quite difficult to review this mixed bag of random changes. |
|
Admittedly touches many files in the same PR. Would you prefer if I split this up into 2 / 3 parts? I did make an effort to ensure that tests pass after each commit though, so it would be easy enough to cut off later commits. The first 5 commits could be part 1, they're the more minor changes. |
lift_inbounds!