MAINT: Optimize numpy.count_nonzero for int types using SIMD operations#18183

Merged

mattip merged 14 commits intonumpy:masterfrom

touqir14:master

Feb 15, 2021

Commits on Jan 18, 2021

Added support for SIMD operations for int types in numpy.count_nonzero function
touqir14
committed
d2e7768
Copy full SHA for d2e7768
Added tests for i1,i2,i4,i8 types for numpy.count_nonzero function
touqir14
committed
c716a12
Copy full SHA for c716a12

Commits on Jan 19, 2021

Merged count_nonzero_int16/int32/int64 into count_nonzero_int and added benchmarks
touqir14
committed
15cf37d
Copy full SHA for 15cf37d
Removed commented out code from PyArray_CountNonzero
touqir14
committed
2b41cbf
Copy full SHA for 2b41cbf

Commits on Feb 5, 2021

Merge remote-tracking branch 'upstream/master'
touqir14
committed
ed3d080
Copy full SHA for ed3d080
Replaced manual sums with horizontal simd sums for count_nonzero_16/64
touqir14
committed
87c5d51
Copy full SHA for 87c5d51

Commits on Feb 7, 2021

fixed CI errors and optimized further simd_16 and simd_32
touqir14
committed
65892ef
Copy full SHA for 65892ef
some fixes for the build problems
touqir14
committed
022cc66
Copy full SHA for 022cc66
another attempt to fix build issues
touqir14
committed
6895bab
Copy full SHA for 6895bab
removed the target variable and changed the loop as suggested by Sayed Adel
touqir14
committed
89d6e55
Copy full SHA for 89d6e55
Modified PyArray_CountNonzero to discriminate between types based on elsize
touqir14
committed
534132e
Copy full SHA for 534132e
Ensured overflow does not happen for 16 and 32 bit ints
touqir14
committed
1eb91a3
Copy full SHA for 1eb91a3

Commits on Feb 13, 2021

cleanup
seiko2plus
authored
d208702
Copy full SHA for d208702
fix up
seiko2plus
authored
85e2ce9
Copy full SHA for 85e2ce9