ggml : always define ggml_fp16_t as uint16_t

[ggerganov]

ref ggml-org/ggml#720

[slaren]

I don't see any obvious issues, but I also don't know much about the ARM code.

[slaren]

A few more suspicious casts to float in ggml-quants.c:

5632:        const float d = y[i].d * (float)x[i].d;
5633:        const float dmin = -y[i].d * (float)x[i].dmin;
5782:        const float d = y[i].d * (float)x[i].d;
5783:        const float dmin = -y[i].d * (float)x[i].dmin;
6436:        const float d = y[i].d * (float)x[i].d;
6638:        const float d = y[i].d * (float)x[i].d;
7141:        sum_mins += y[i].d * (float)x[i].d[1] * summi;
7143:        const float d = y[i].d * (float)x[i].d[0];
7801:        const float d = y[i].d * (float)x[i].d;
7943:        const float d = y[i].d * (float)x[i].d;
8682:        const float d_all = (float)x[i].d;