{{L}}return:
    ldmxcsr     [rsp + 4]
    add         rsp, 8

    pop r15
    pop r14
    pop r13
    pop r12
    pop rbx

{% if family == "windows" %}
    pop rsi
    pop rdi

    vmovaps xmm15, [rsp+16*9]
    vmovaps xmm14, [rsp+16*8]
    vmovaps xmm13, [rsp+16*7]
    vmovaps xmm12, [rsp+16*6]
    vmovaps xmm11, [rsp+16*5]
    vmovaps xmm10, [rsp+16*4]
    vmovaps xmm9, [rsp+16*3]
    vmovaps xmm8, [rsp+16*2]
    vmovaps xmm7, [rsp+16*1]
    vmovaps xmm6, [rsp]
{% endif %}

    mov rsp, rbp
    pop rbp
    ret

{% if msvc %}
{{arch}}_mmm_f32_{{size}}_{{suffix}} endp
_text ends
end

{% else %}
.cfi_endproc
{% endif %}