New Features/Functions
-
fp16
- half precision floating point support has been added - #221 -
Confidence Connected Components
confidenceCC
- #221 -
Deconvolutions - #221
-
Reduction using keys - #221
-
Neural network based convolution and gradient functions - #221
-
Support for uniform ranges in approx1 and approx2 functions - #234
-
Array class methods - #233
-
New Examples
Breaking APIs
Fixes
- Fixed wrapper validations in
create_sparse_from_host
- #198 - Added a workaround for
bench_cg
example on less capable GPUs - #200 - Fixed missing info in
Array.device_ptr
function documentation - #210 - Corrected invert operation to use non-in-place bit wise inversion - #228