Raghuveer Devulapalli

> is this sort implementation possible on AVX2 era instructions or is that forever stuck on scalar implementation? https://github.com/intel/x86-simd-sort/pull/119 adds AVX2 versions but the performance gains won't match AVX-512. >...

Also, what is the array size?

Hmm not sure if I am measuring it wrong but I am seeing results differently. PR #121 compares copying the original array and sort indices using `keyvalue_qsort` against `ippsSortRadixIndexAscend_32s` and...

sure. I can take a look.

apologies, I haven't had time yet. I will try to take a look at it this week.

HI @KungFuJesus yeah looks like for the data distribution you have, I am also measuring ippindexsort to be faster than avx-512 argsort. I wonder if the new pivot selection might...

Very likely that your binutils/assembler is old. `C++ linker for the host machine: c++ ld.bfd 2.27-44`. Could you update that and re-build?

Hi, If you pass in the array and array of indices initialized to [0, 1, 2, 3, .. n] as arguments for key-value pair, PR #2 can be adapted to...

> avx512_qsort_kv(double *keys, uint64_t *indexes, int64_t arrsize) is a little worse than my simple bubble sorting. @xiangyunzhou do you mind sharing your bubble sort so we can compare and add...

> I want to get the top 32 index of 400 sorting by a float pfWeight. Have you tried benchmarking for a larger array? The code size of AVX512 qsort...