pnumpy

pnumpy copied to clipboard

No central location for AVX math

6

In numpy, I notice numpy intrinsics being checking by xiegengxin [xiegengxin](https://github.com/numpy/numpy/commits?author=xiegengxin ) However, it is difficult to tell which routines are best because they are also checked in tis project....

tdimitri

Hyperthreading, L2/L3 Cache sizes, CPU C States, NUMA

2

At work, in BIOS we disable hyperthreading and turn off the power saving CPU C states. If we have multiple NUMA nodes, we sometimes run one process on NUMA node...

tdimitri

ufunc signature for true_divide and sqrt seem a little odd

1

I tried to make a signature of two inputs:int32, int32 returning output; float64 and this failed. I notice some inputs like int16 return float32. On sqrt, some inputs, like int8,...

tdimitri

need to work on compiler flags for vectorized code vs

2

Need to experiment more with pragma and compiler targets. We want the loader code to be compiled normally. We want the avx2 (256bit instructions) code to be compiled with -avx2...

tdimitri

need to route ufunc signatures we do not have hook for into threader

1

we can thread ufuncs we do not understand. For a binary_reduce, on a large array, we can divide the work up assigning each work chunk to a thread. each work...

tdimitri

need to see if we can tell numpy loop matching engine that we can directly compare int64 to uint64. (maybe it already works, just have not tested it yet)

tdimitri