// vim: ft=arm

{% include "arm64fp16_mmm_8h_scalar.tmpliq" label:"scalar_min", op:"fmin", from:from, to:to %}
{% include "arm64fp16_mmm_8h_scalar.tmpliq" label:"scalar_max", op:"fmax", from:from, to:to %}
{% include "arm64fp16_mmm_8h_scalar.tmpliq" label:"scalar_mul", op:"fmul", from:from, to:to %}
{% include "arm64fp16_mmm_8h_scalar.tmpliq" label:"scalar_add", op:"fadd", from:from, to:to %}
{% include "arm64fp16_mmm_8h_scalar.tmpliq" label:"scalar_sub", op:"fsub", from:from, to:to %}
{% include "arm64fp16_mmm_8h_scalar.tmpliq" label:"scalar_sub_flipped", op:"fsub", from:from, to:to, flipped:true %}

.clear:
{% for r in (from..to) %}
    eor         v{{r}}.8b, v{{r}}.8b, v{{r}}.8b
{% endfor %}
    b .non_linear_loop

.leaky_relu:
    add         x2, x0, #8
    ld1         {v4.s}[0], [ x2 ]
    dup         v4.8h, v4.h[0]

    // bsl cond/dst, then, else
    // fcmge dst, src, #0.0
    {% for r in (from..to) %}
        fmul  v0.8h, v{{r}}.8h, v4.8h
        fcmge v1.8h, v{{r}}.8h, #0.0
        bsl   v1.16b, v{{r}}.16b, v0.16b
        and   v{{r}}.16b, v1.16b, v1.16b
    {% endfor %}

    b .non_linear_loop


.q_scale:
.q_shl:
.q_shr:
    b .unsupported