Hi, Before replacing with completly new code of LPCNet/mozilla, I tried to use new vector code.
Does anyone try it? https://github.com/jg1uaa/LPCNet.drowe67/tree/replace_vec_avx The differences from master: vec.h, vec_avx.h, vec_neon.h, tansig_table.h: replaced with new mozilla's code nneth, nnet.c: simply include new vector header, new accum8x4() is not supported yet test_vec.c: fix compile error with new header On my A10-7860K(OpenBSD/amd64) machine, SSE works faster than previous vec_sse.h so I removed it. How about on other machines/CPUs? If there is good enough performance, I will try replace nnet.c. -- SASANO Takayoshi (JG1UAA) <[email protected]> _______________________________________________ Freetel-codec2 mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freetel-codec2
