Pearu Peterson

While using asin from asinh for complex inputs is the right way to go (as implemented in this PR), asin itself on complex inputs has very poor accuracy. I am...

As it turns out, asin/asinh improvements in `xla/client/lib/math.cc` will not be available for JAX. Therefore, I am currently working on submitting the modified Hull algorithm to StableHLO so that asin/asinh...

https://github.com/openxla/stablehlo/pull/2357 implements the modified Hull et al algorithm for StableHLO. The asin implementations for both XLA and StableHLO are now generated using https://github.com/pearu/functional_algorithms tool that includes the validation of the...

Copying a relevant note from #48 to here: The composition of slices is a slice. See https://github.com/Quansight-Labs/rfcs/tree/pearu/rfc0005/RFC0003-sparse-roadmap#slicing for the mathematical background.

Data points: - running all rbc tests using CPU-only omniscidb server takes about 12 min. - running all rbs tests using CUDA-enabled omniscidb server takes about 20 min

Keeping the issue open as there are other tests that can be speed up.

``` for fn in rbc/tests/test_*.py; do echo $fn; (pytest $fn | tail -1) ; done ``` leads to (using CUDA enabled omniscidb) ``` rbc/tests/test_ci.py ======================== 3 skipped, 2 warnings in...

See Slack discussions for ideas: 1. https://quansight.slack.com/archives/C01CAU0U1FH/p1605526178096400

Related PRs: - https://github.com/omnisci/omniscidb-internal/pull/5492 - https://github.com/xnd-project/rbc/pull/322

Tests in rbc/tests/heavyai/test_text_encoding.py cover the cases where TextEncodingDict variables are passed around but we don't have tests for accessing the content of TextEncodingDict variables. In short, we should be able...