// Xavier Hsinyuan is pleased to support the open source community by making
// ncnn available.
//
// Copyright (C) 2021 Xavier Hsinyuan <me@lstlx.com>. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this
// file except in compliance with the License. You may obtain a copy of the
// License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
// WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
// License for the specific language governing permissions and limitations under
// the License.

#include "hardswish_riscv.h"

#if __riscv_vector
#include <riscv_vector.h>
#endif // __riscv_vector

namespace ncnn {

HardSwish_riscv::HardSwish_riscv()
{
#if __riscv_vector
    support_packing = true;
#if __riscv_zfh
    support_fp16_storage = true;
#endif
#endif // __riscv_vector
}

int HardSwish_riscv::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
{
#if __riscv_vector && __riscv_zfh
    int elembits = bottom_top_blob.elembits();

    if (opt.use_fp16_storage && elembits == 16)
    {
        return forward_inplace_fp16s(bottom_top_blob, opt);
    }
#endif

    int w = bottom_top_blob.w;
    int h = bottom_top_blob.h;
    int d = bottom_top_blob.d;
    int channels = bottom_top_blob.c;
    int elempack = bottom_top_blob.elempack;
    int size = w * h * d * elempack;

    #pragma omp parallel for num_threads(opt.num_threads)
    for (int q = 0; q < channels; q++)
    {
        float* ptr = bottom_top_blob.channel(q);

#if __riscv_vector
        int n = size;
        while (n > 0)
        {
            size_t vl = vsetvl_e32m8(n);
            vfloat32m8_t _p = vle32_v_f32m8(ptr, vl);

            vbool4_t _lower = vmflt_vf_f32m8_b4(_p, lower, vl);
            vbool4_t _higher = vmfgt_vf_f32m8_b4(_p, upper, vl);
            vbool4_t _apply = vmnor_mm_b4(_lower, _higher, vl);
            _p = vfmerge_vfm_f32m8(_lower, _p, .0f, vl);

            vfloat32m8_t _p0 = vfadd_vf_f32m8_m(
                                   _apply, _p, /*op1*/ vfmul_vf_f32m8_m(_apply, _p, _p, alpha, vl), beta,
                                   vl);
            _p = vfmul_vv_f32m8_m(_apply, _p, /*op1*/ _p, _p0, vl);

            vse32_v_f32m8(ptr, _p, vl);
            ptr += vl;
            n -= vl;
        }
#else  // __riscv_vector
        for (int i = 0; i < size; i++)
        {
            if (ptr[i] < lower)
                ptr[i] = 0.f;
            else if (ptr[i] > upper)
                ;
            else
                ptr[i] = ptr[i] * (ptr[i] * alpha + beta);
        }
#endif // __riscv_vector
    }

    return 0;
}

#if __riscv_vector && __riscv_zfh
int HardSwish_riscv::forward_inplace_fp16s(Mat& bottom_top_blob, const Option& opt) const
{
    int w = bottom_top_blob.w;
    int h = bottom_top_blob.h;
    int d = bottom_top_blob.d;
    int channels = bottom_top_blob.c;
    int elempack = bottom_top_blob.elempack;
    int size = w * h * d * elempack;

    #pragma omp parallel for num_threads(opt.num_threads)
    for (int q = 0; q < channels; q++)
    {
        __fp16* ptr = bottom_top_blob.channel(q);

        int n = size;
        while (n > 0)
        {
            size_t vl = vsetvl_e16m8(n);
            vfloat16m8_t _p = vle16_v_f16m8(ptr, vl);

            vbool2_t _lower = vmflt_vf_f16m8_b2(_p, lower, vl);
            vbool2_t _higher = vmfgt_vf_f16m8_b2(_p, upper, vl);
            vbool2_t _apply = vmnor_mm_b2(_lower, _higher, vl);
            _p = vfmerge_vfm_f16m8(_lower, _p, .0f, vl);

            vfloat16m8_t _p0 = vfadd_vf_f16m8_m(
                                   _apply, _p, /*op1*/ vfmul_vf_f16m8_m(_apply, _p, _p, alpha, vl), beta,
                                   vl);
            _p = vfmul_vv_f16m8_m(_apply, _p, /*op1*/ _p, _p0, vl);

            vse16_v_f16m8(ptr, _p, vl);
            ptr += vl;
            n -= vl;
        }
    }

    return 0;
}
#endif

} // namespace ncnn