pytroll / pyresample / 35048142149
94%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 99
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 93.915%. Remained the same
35048142149

push

github

djhoese 
Optimize the EWA fornav inner loops without changing results

Restructure compute_ewa() and compute_ewa_single() so the per-grid-cell loop
only does the weight lookup and the two grid updates. The ellipse parameters
and weight table fields are now copied into locals and the array arguments are
declared restrict, so the compiler no longer has to assume every grid store may
have modified them and reload them each iteration. The input pixel's
fill/NaN check is hoisted out of the bounding box loops (once per swath pixel,
with a compact list of valid channels in the multi-channel path),
maximum_weight_mode is a template parameter so each mode gets its own
specialised loop, the weight table is over-allocated by one entry so the index
clamp can be dropped, and the bounding box test is evaluated on the float
bounds before any int cast so pixels far outside the grid on either side (the
common case for dask chunks) are rejected with four comparisons and NaN
coordinates fall out without isnan(). -fno-math-errno is added to the compile
flags so sqrt()/exp() can inline. No floating point operation changes order, so
outputs are bit-for-bit identical to the previous build; this was verified with
np.array_equal across both wrappers, all coordinate/image dtypes and both weight
modes. Synthetic timings show 1.1-1.3x on the single-channel path, most for
large ellipses and for chunks where most source pixels lie outside the grid.

13149 of 14001 relevant lines covered (93.91%)

2.82 hits per line

Jobs
ID Job ID Ran Files Coverage
1 run- - 35048142149.1 99
93.91
Source Files on build 35048142149