Slow performance in Macenko normalization

[haoranch3n]

Hello,

I'm facing performance issues with Macenko normalization. Here are the timing results for each step on a typical 512x512 patch:

__convert_rgb2od took 0.14093 seconds
linalg_eigh took 0.00055 seconds
__find_HE took 0.28750 seconds
__find_concentration took 0.54008 seconds
__convert_rgb2od took 0.00351 seconds
linalg_eigh took 0.00012 seconds
__find_HE took 0.00606 seconds
__find_concentration took 0.01264 seconds

The bottleneck seems to be in the __convert_rgb2od, __find_HE, and __find_concentration steps, which are significantly slower than others. I've tried both CPU and GPU, but performance is still slow.

Does anyone have insights into why these specific steps might be taking so long or know of potential optimizations that could speed them up? Any insights or suggestions for optimization would be greatly appreciated!

Thanks in advance for your help!