LoongArch64: Update README.md

[XiWeiGu]

@azuresky01 @martin-frbg @yinshiyou , Could you help me review this PR? Thanks!

[martin-frbg]

Looks good to me (though of course I am not that familiar with the Loongson hardware range - I assume a user would know whether their hardware is LSX or LASX ?) Incidentally, is there a known performance or correctness problem that (still) makes it necessary to use the "simple" level3 threading code on LoongArch64 ?

[XiWeiGu]

Using the "simple" level 3 threading code is mainly for performance reasons, and performance data will be provided later.

[XiWeiGu]

Test data on the 3A5000 2.5GHz with a peak double-precision floating-point performance of 160 GFlops is as follows:

When using USE_SIMPLE_THREADED_LEVEL3=1 on the LoongArch64 platform, there are fewer L3 cache conflicts, resulting in better performance. However, I have also compared this with the X86_64 platform and found the opposite phenomenon, where using USE_SIMPLE_THREADED_LEVEL3=1 often leads to the lowest performance. This may be related to cache design (just a guess).
So I still recommend enabling USE_SIMPLE_THREADED_LEVEL3 on the LoongArch64 platform for now.

[martin-frbg]

Thank you very much for the detailed explanation and data.