Re: [RFC PATCH 14/43] target/loongarch: Implement vmax/vmin

[Richard Henderson]

On 12/24/22 00:16, Song Gao wrote:
> +INSN_LSX(vmax_b,           vvv)
> +INSN_LSX(vmax_h,           vvv)
> +INSN_LSX(vmax_w,           vvv)
> +INSN_LSX(vmax_d,           vvv)
> +INSN_LSX(vmin_b,           vvv)
> +INSN_LSX(vmin_h,           vvv)
> +INSN_LSX(vmin_w,           vvv)
> +INSN_LSX(vmin_d,           vvv)
> +INSN_LSX(vmax_bu,          vvv)
> +INSN_LSX(vmax_hu,          vvv)
> +INSN_LSX(vmax_wu,          vvv)
> +INSN_LSX(vmax_du,          vvv)
> +INSN_LSX(vmin_bu,          vvv)
> +INSN_LSX(vmin_hu,          vvv)
> +INSN_LSX(vmin_wu,          vvv)
> +INSN_LSX(vmin_du,          vvv)

These can be implemented with tcg_gen_gvec_{smin,umin,smax,umax}.

> +INSN_LSX(vmaxi_b,          vv_i)
> +INSN_LSX(vmaxi_h,          vv_i)
> +INSN_LSX(vmaxi_w,          vv_i)
> +INSN_LSX(vmaxi_d,          vv_i)
> +INSN_LSX(vmini_b,          vv_i)
> +INSN_LSX(vmini_h,          vv_i)
> +INSN_LSX(vmini_w,          vv_i)
> +INSN_LSX(vmini_d,          vv_i)
> +INSN_LSX(vmaxi_bu,         vv_i)
> +INSN_LSX(vmaxi_hu,         vv_i)
> +INSN_LSX(vmaxi_wu,         vv_i)
> +INSN_LSX(vmaxi_du,         vv_i)
> +INSN_LSX(vmini_bu,         vv_i)
> +INSN_LSX(vmini_hu,         vv_i)
> +INSN_LSX(vmini_wu,         vv_i)
> +INSN_LSX(vmini_du,         vv_i)

These have no direct immediate variant.  Use a combination pattern with dup + 
minmax.


r~