src/mat_pixel_resize.cpp

// Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2018 THL A29 Limited, a Tencent company. 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 "mat.h"

#include <limits.h>
#include <math.h>
#if __ARM_NEON
#include <arm_neon.h>
#endif // __ARM_NEON
#include "platform.h"

namespace ncnn {

#if NCNN_PIXEL
void resize_bilinear_c1(const unsigned char* src, int srcw, int srch, unsigned char* dst, int w, int h)
{
    return resize_bilinear_c1(src, srcw, srch, srcw, dst, w, h, w);
}

void resize_bilinear_c2(const unsigned char* src, int srcw, int srch, unsigned char* dst, int w, int h)
{
    return resize_bilinear_c2(src, srcw, srch, srcw * 2, dst, w, h, w * 2);
}

void resize_bilinear_c3(const unsigned char* src, int srcw, int srch, unsigned char* dst, int w, int h)
{
    return resize_bilinear_c3(src, srcw, srch, srcw * 3, dst, w, h, w * 3);
}

void resize_bilinear_c4(const unsigned char* src, int srcw, int srch, unsigned char* dst, int w, int h)
{
    return resize_bilinear_c4(src, srcw, srch, srcw * 4, dst, w, h, w * 4);
}

void resize_bilinear_c1(const unsigned char* src, int srcw, int srch, int srcstride, unsigned char* dst, int w, int h, int stride)
{
    const int INTER_RESIZE_COEF_BITS = 11;
    const int INTER_RESIZE_COEF_SCALE = 1 << INTER_RESIZE_COEF_BITS;
    //     const int ONE=INTER_RESIZE_COEF_SCALE;

    double scale_x = (double)srcw / w;
    double scale_y = (double)srch / h;

    int* buf = new int[w + h + w + h];

    int* xofs = buf;     //new int[w];
    int* yofs = buf + w; //new int[h];

    short* ialpha = (short*)(buf + w + h);    //new short[w * 2];
    short* ibeta = (short*)(buf + w + h + w); //new short[h * 2];

    float fx;
    float fy;
    int sx;
    int sy;

#define SATURATE_CAST_SHORT(X) (short)::std::min(::std::max((int)(X + (X >= 0.f ? 0.5f : -0.5f)), SHRT_MIN), SHRT_MAX);

    for (int dx = 0; dx < w; dx++)
    {
        fx = (float)((dx + 0.5) * scale_x - 0.5);
        sx = static_cast<int>(floor(fx));
        fx -= sx;

        if (sx < 0)
        {
            sx = 0;
            fx = 0.f;
        }
        if (sx >= srcw - 1)
        {
            sx = srcw - 2;
            fx = 1.f;
        }

        xofs[dx] = sx;

        float a0 = (1.f - fx) * INTER_RESIZE_COEF_SCALE;
        float a1 = fx * INTER_RESIZE_COEF_SCALE;

        ialpha[dx * 2] = SATURATE_CAST_SHORT(a0);
        ialpha[dx * 2 + 1] = SATURATE_CAST_SHORT(a1);
    }

    for (int dy = 0; dy < h; dy++)
    {
        fy = (float)((dy + 0.5) * scale_y - 0.5);
        sy = static_cast<int>(floor(fy));
        fy -= sy;

        if (sy < 0)
        {
            sy = 0;
            fy = 0.f;
        }
        if (sy >= srch - 1)
        {
            sy = srch - 2;
            fy = 1.f;
        }

        yofs[dy] = sy;

        float b0 = (1.f - fy) * INTER_RESIZE_COEF_SCALE;
        float b1 = fy * INTER_RESIZE_COEF_SCALE;

        ibeta[dy * 2] = SATURATE_CAST_SHORT(b0);
        ibeta[dy * 2 + 1] = SATURATE_CAST_SHORT(b1);
    }

#undef SATURATE_CAST_SHORT

    // loop body
    Mat rowsbuf0(w, (size_t)2u);
    Mat rowsbuf1(w, (size_t)2u);
    short* rows0 = (short*)rowsbuf0.data;
    short* rows1 = (short*)rowsbuf1.data;

    int prev_sy1 = -2;

    for (int dy = 0; dy < h; dy++)
    {
        sy = yofs[dy];

        if (sy == prev_sy1)
        {
            // reuse all rows
        }
        else if (sy == prev_sy1 + 1)
        {
            // hresize one row
            short* rows0_old = rows0;
            rows0 = rows1;
            rows1 = rows0_old;
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                const unsigned char* S1p = S1 + sx;
                rows1p[dx] = (S1p[0] * a0 + S1p[1] * a1) >> 4;

                ialphap += 2;
            }
        }
        else
        {
            // hresize two rows
            const unsigned char* S0 = src + srcstride * (sy);
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows0p = rows0;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                const unsigned char* S0p = S0 + sx;
                const unsigned char* S1p = S1 + sx;
                rows0p[dx] = (S0p[0] * a0 + S0p[1] * a1) >> 4;
                rows1p[dx] = (S1p[0] * a0 + S1p[1] * a1) >> 4;

                ialphap += 2;
            }
        }

        prev_sy1 = sy;

        // vresize
        short b0 = ibeta[0];
        short b1 = ibeta[1];

        short* rows0p = rows0;
        short* rows1p = rows1;
        unsigned char* Dp = dst + stride * (dy);

#if __ARM_NEON
        int nn = w >> 3;
#else
        int nn = 0;
#endif
        int remain = w - (nn << 3);

#if __ARM_NEON
#if __aarch64__
        int16x4_t _b0 = vdup_n_s16(b0);
        int16x4_t _b1 = vdup_n_s16(b1);
        int32x4_t _v2 = vdupq_n_s32(2);
        for (; nn > 0; nn--)
        {
            int16x4_t _rows0p_sr4 = vld1_s16(rows0p);
            int16x4_t _rows1p_sr4 = vld1_s16(rows1p);
            int16x4_t _rows0p_1_sr4 = vld1_s16(rows0p + 4);
            int16x4_t _rows1p_1_sr4 = vld1_s16(rows1p + 4);

            int32x4_t _rows0p_sr4_mb0 = vmull_s16(_rows0p_sr4, _b0);
            int32x4_t _rows1p_sr4_mb1 = vmull_s16(_rows1p_sr4, _b1);
            int32x4_t _rows0p_1_sr4_mb0 = vmull_s16(_rows0p_1_sr4, _b0);
            int32x4_t _rows1p_1_sr4_mb1 = vmull_s16(_rows1p_1_sr4, _b1);

            int32x4_t _acc = _v2;
            _acc = vsraq_n_s32(_acc, _rows0p_sr4_mb0, 16);
            _acc = vsraq_n_s32(_acc, _rows1p_sr4_mb1, 16);

            int32x4_t _acc_1 = _v2;
            _acc_1 = vsraq_n_s32(_acc_1, _rows0p_1_sr4_mb0, 16);
            _acc_1 = vsraq_n_s32(_acc_1, _rows1p_1_sr4_mb1, 16);

            int16x4_t _acc16 = vshrn_n_s32(_acc, 2);
            int16x4_t _acc16_1 = vshrn_n_s32(_acc_1, 2);

            uint8x8_t _Dp = vqmovun_s16(vcombine_s16(_acc16, _acc16_1));

            vst1_u8(Dp, _Dp);

            Dp += 8;
            rows0p += 8;
            rows1p += 8;
        }
#else
        if (nn > 0)
        {
            asm volatile(
                "vdup.s16   d16, %8         \n"
                "mov        r4, #2          \n"
                "vdup.s16   d17, %9         \n"
                "vdup.s32   q12, r4         \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "0:                         \n"
                "vmull.s16  q0, d2, d16     \n"
                "vmull.s16  q1, d3, d16     \n"
                "vorr.s32   q10, q12, q12   \n"
                "vorr.s32   q11, q12, q12   \n"
                "vmull.s16  q2, d6, d17     \n"
                "vmull.s16  q3, d7, d17     \n"
                "vsra.s32   q10, q0, #16    \n"
                "vsra.s32   q11, q1, #16    \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "vsra.s32   q10, q2, #16    \n"
                "vsra.s32   q11, q3, #16    \n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "vshrn.s32  d20, q10, #2    \n"
                "vshrn.s32  d21, q11, #2    \n"
                "vqmovun.s16 d20, q10        \n"
                "vst1.8     {d20}, [%2]!    \n"
                "subs       %3, #1          \n"
                "bne        0b              \n"
                "sub        %0, #16         \n"
                "sub        %1, #16         \n"
                : "=r"(rows0p), // %0
                "=r"(rows1p), // %1
                "=r"(Dp),     // %2
                "=r"(nn)      // %3
                : "0"(rows0p),
                "1"(rows1p),
                "2"(Dp),
                "3"(nn),
                "r"(b0), // %8
                "r"(b1)  // %9
                : "cc", "memory", "r4", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11", "q12");
        }
#endif // __aarch64__
#endif // __ARM_NEON
        for (; remain; --remain)
        {
            //             D[x] = (rows0[x]*b0 + rows1[x]*b1) >> INTER_RESIZE_COEF_BITS;
            *Dp++ = (unsigned char)(((short)((b0 * (short)(*rows0p++)) >> 16) + (short)((b1 * (short)(*rows1p++)) >> 16) + 2) >> 2);
        }

        ibeta += 2;
    }

    delete[] buf;
}

void resize_bilinear_c2(const unsigned char* src, int srcw, int srch, int srcstride, unsigned char* dst, int w, int h, int stride)
{
    const int INTER_RESIZE_COEF_BITS = 11;
    const int INTER_RESIZE_COEF_SCALE = 1 << INTER_RESIZE_COEF_BITS;
    //     const int ONE=INTER_RESIZE_COEF_SCALE;

    double scale_x = (double)srcw / w;
    double scale_y = (double)srch / h;

    int* buf = new int[w + h + w + h];

    int* xofs = buf;     //new int[w];
    int* yofs = buf + w; //new int[h];

    short* ialpha = (short*)(buf + w + h);    //new short[w * 2];
    short* ibeta = (short*)(buf + w + h + w); //new short[h * 2];

    float fx;
    float fy;
    int sx;
    int sy;

#define SATURATE_CAST_SHORT(X) (short)::std::min(::std::max((int)(X + (X >= 0.f ? 0.5f : -0.5f)), SHRT_MIN), SHRT_MAX);

    for (int dx = 0; dx < w; dx++)
    {
        fx = (float)((dx + 0.5) * scale_x - 0.5);
        sx = static_cast<int>(floor(fx));
        fx -= sx;

        if (sx < 0)
        {
            sx = 0;
            fx = 0.f;
        }
        if (sx >= srcw - 1)
        {
            sx = srcw - 2;
            fx = 1.f;
        }

        xofs[dx] = sx * 2;

        float a0 = (1.f - fx) * INTER_RESIZE_COEF_SCALE;
        float a1 = fx * INTER_RESIZE_COEF_SCALE;

        ialpha[dx * 2] = SATURATE_CAST_SHORT(a0);
        ialpha[dx * 2 + 1] = SATURATE_CAST_SHORT(a1);
    }

    for (int dy = 0; dy < h; dy++)
    {
        fy = (float)((dy + 0.5) * scale_y - 0.5);
        sy = static_cast<int>(floor(fy));
        fy -= sy;

        if (sy < 0)
        {
            sy = 0;
            fy = 0.f;
        }
        if (sy >= srch - 1)
        {
            sy = srch - 2;
            fy = 1.f;
        }

        yofs[dy] = sy;

        float b0 = (1.f - fy) * INTER_RESIZE_COEF_SCALE;
        float b1 = fy * INTER_RESIZE_COEF_SCALE;

        ibeta[dy * 2] = SATURATE_CAST_SHORT(b0);
        ibeta[dy * 2 + 1] = SATURATE_CAST_SHORT(b1);
    }

#undef SATURATE_CAST_SHORT

    // loop body
    Mat rowsbuf0(w * 2 + 2, (size_t)2u);
    Mat rowsbuf1(w * 2 + 2, (size_t)2u);
    short* rows0 = (short*)rowsbuf0.data;
    short* rows1 = (short*)rowsbuf1.data;

    int prev_sy1 = -2;

    for (int dy = 0; dy < h; dy++)
    {
        sy = yofs[dy];

        if (sy == prev_sy1)
        {
            // reuse all rows
        }
        else if (sy == prev_sy1 + 1)
        {
            // hresize one row
            short* rows0_old = rows0;
            rows0 = rows1;
            rows1 = rows0_old;
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];

                const unsigned char* S1p = S1 + sx;
#if __ARM_NEON
                int16x4_t _a0a1XX = vld1_s16(ialphap);
                int16x4_t _a0a0a1a1 = vzip_s16(_a0a1XX, _a0a1XX).val[0];
                uint8x8_t _S1 = uint8x8_t();

                _S1 = vld1_lane_u8(S1p, _S1, 0);
                _S1 = vld1_lane_u8(S1p + 1, _S1, 1);
                _S1 = vld1_lane_u8(S1p + 2, _S1, 2);
                _S1 = vld1_lane_u8(S1p + 3, _S1, 3);

                int16x8_t _S116 = vreinterpretq_s16_u16(vmovl_u8(_S1));
                int16x4_t _S1lowhigh = vget_low_s16(_S116);
                int32x4_t _S1ma0a1 = vmull_s16(_S1lowhigh, _a0a0a1a1);
                int32x2_t _rows1low = vadd_s32(vget_low_s32(_S1ma0a1), vget_high_s32(_S1ma0a1));
                int32x4_t _rows1 = vcombine_s32(_rows1low, vget_high_s32(_S1ma0a1));
                int16x4_t _rows1_sr4 = vshrn_n_s32(_rows1, 4);
                vst1_s16(rows1p, _rows1_sr4);
#else
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                rows1p[0] = (S1p[0] * a0 + S1p[2] * a1) >> 4;
                rows1p[1] = (S1p[1] * a0 + S1p[3] * a1) >> 4;
#endif // __ARM_NEON

                ialphap += 2;
                rows1p += 2;
            }
        }
        else
        {
            // hresize two rows
            const unsigned char* S0 = src + srcstride * (sy);
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows0p = rows0;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                const unsigned char* S0p = S0 + sx;
                const unsigned char* S1p = S1 + sx;
#if __ARM_NEON
                int16x4_t _a0 = vdup_n_s16(a0);
                int16x4_t _a1 = vdup_n_s16(a1);
                uint8x8_t _S0 = uint8x8_t();
                uint8x8_t _S1 = uint8x8_t();

                _S0 = vld1_lane_u8(S0p, _S0, 0);
                _S0 = vld1_lane_u8(S0p + 1, _S0, 1);
                _S0 = vld1_lane_u8(S0p + 2, _S0, 2);
                _S0 = vld1_lane_u8(S0p + 3, _S0, 3);

                _S1 = vld1_lane_u8(S1p, _S1, 0);
                _S1 = vld1_lane_u8(S1p + 1, _S1, 1);
                _S1 = vld1_lane_u8(S1p + 2, _S1, 2);
                _S1 = vld1_lane_u8(S1p + 3, _S1, 3);

                int16x8_t _S016 = vreinterpretq_s16_u16(vmovl_u8(_S0));
                int16x8_t _S116 = vreinterpretq_s16_u16(vmovl_u8(_S1));
                int16x4_t _S0lowhigh = vget_low_s16(_S016);
                int16x4_t _S1lowhigh = vget_low_s16(_S116);
                int32x2x2_t _S0S1low_S0S1high = vtrn_s32(vreinterpret_s32_s16(_S0lowhigh), vreinterpret_s32_s16(_S1lowhigh));
                int32x4_t _rows01 = vmull_s16(vreinterpret_s16_s32(_S0S1low_S0S1high.val[0]), _a0);
                _rows01 = vmlal_s16(_rows01, vreinterpret_s16_s32(_S0S1low_S0S1high.val[1]), _a1);
                int16x4_t _rows01_sr4 = vshrn_n_s32(_rows01, 4);
                int16x4_t _rows1_sr4 = vext_s16(_rows01_sr4, _rows01_sr4, 2);
                vst1_s16(rows0p, _rows01_sr4);
                vst1_s16(rows1p, _rows1_sr4);
#else
                rows0p[0] = (S0p[0] * a0 + S0p[2] * a1) >> 4;
                rows0p[1] = (S0p[1] * a0 + S0p[3] * a1) >> 4;
                rows1p[0] = (S1p[0] * a0 + S1p[2] * a1) >> 4;
                rows1p[1] = (S1p[1] * a0 + S1p[3] * a1) >> 4;
#endif // __ARM_NEON

                ialphap += 2;
                rows0p += 2;
                rows1p += 2;
            }
        }

        prev_sy1 = sy;

        // vresize
        short b0 = ibeta[0];
        short b1 = ibeta[1];

        short* rows0p = rows0;
        short* rows1p = rows1;
        unsigned char* Dp = dst + stride * (dy);

#if __ARM_NEON
        int nn = (w * 2) >> 3;
#else
        int nn = 0;
#endif
        int remain = (w * 2) - (nn << 3);

#if __ARM_NEON
#if __aarch64__
        int16x4_t _b0 = vdup_n_s16(b0);
        int16x4_t _b1 = vdup_n_s16(b1);
        int32x4_t _v2 = vdupq_n_s32(2);
        for (; nn > 0; nn--)
        {
            int16x4_t _rows0p_sr4 = vld1_s16(rows0p);
            int16x4_t _rows1p_sr4 = vld1_s16(rows1p);
            int16x4_t _rows0p_1_sr4 = vld1_s16(rows0p + 4);
            int16x4_t _rows1p_1_sr4 = vld1_s16(rows1p + 4);

            int32x4_t _rows0p_sr4_mb0 = vmull_s16(_rows0p_sr4, _b0);
            int32x4_t _rows1p_sr4_mb1 = vmull_s16(_rows1p_sr4, _b1);
            int32x4_t _rows0p_1_sr4_mb0 = vmull_s16(_rows0p_1_sr4, _b0);
            int32x4_t _rows1p_1_sr4_mb1 = vmull_s16(_rows1p_1_sr4, _b1);

            int32x4_t _acc = _v2;
            _acc = vsraq_n_s32(_acc, _rows0p_sr4_mb0, 16);
            _acc = vsraq_n_s32(_acc, _rows1p_sr4_mb1, 16);

            int32x4_t _acc_1 = _v2;
            _acc_1 = vsraq_n_s32(_acc_1, _rows0p_1_sr4_mb0, 16);
            _acc_1 = vsraq_n_s32(_acc_1, _rows1p_1_sr4_mb1, 16);

            int16x4_t _acc16 = vshrn_n_s32(_acc, 2);
            int16x4_t _acc16_1 = vshrn_n_s32(_acc_1, 2);

            uint8x8_t _Dp = vqmovun_s16(vcombine_s16(_acc16, _acc16_1));

            vst1_u8(Dp, _Dp);

            Dp += 8;
            rows0p += 8;
            rows1p += 8;
        }
#else
        if (nn > 0)
        {
            asm volatile(
                "vdup.s16   d16, %8         \n"
                "mov        r4, #2          \n"
                "vdup.s16   d17, %9         \n"
                "vdup.s32   q12, r4         \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "0:                         \n"
                "vmull.s16  q0, d2, d16     \n"
                "vmull.s16  q1, d3, d16     \n"
                "vorr.s32   q10, q12, q12   \n"
                "vorr.s32   q11, q12, q12   \n"
                "vmull.s16  q2, d6, d17     \n"
                "vmull.s16  q3, d7, d17     \n"
                "vsra.s32   q10, q0, #16    \n"
                "vsra.s32   q11, q1, #16    \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "vsra.s32   q10, q2, #16    \n"
                "vsra.s32   q11, q3, #16    \n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "vshrn.s32  d20, q10, #2    \n"
                "vshrn.s32  d21, q11, #2    \n"
                "vqmovun.s16 d20, q10        \n"
                "vst1.8     {d20}, [%2]!    \n"
                "subs       %3, #1          \n"
                "bne        0b              \n"
                "sub        %0, #16         \n"
                "sub        %1, #16         \n"
                : "=r"(rows0p), // %0
                "=r"(rows1p), // %1
                "=r"(Dp),     // %2
                "=r"(nn)      // %3
                : "0"(rows0p),
                "1"(rows1p),
                "2"(Dp),
                "3"(nn),
                "r"(b0), // %8
                "r"(b1)  // %9
                : "cc", "memory", "r4", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11", "q12");
        }
#endif // __aarch64__
#endif // __ARM_NEON
        for (; remain; --remain)
        {
            //             D[x] = (rows0[x]*b0 + rows1[x]*b1) >> INTER_RESIZE_COEF_BITS;
            *Dp++ = (unsigned char)(((short)((b0 * (short)(*rows0p++)) >> 16) + (short)((b1 * (short)(*rows1p++)) >> 16) + 2) >> 2);
        }

        ibeta += 2;
    }

    delete[] buf;
}

void resize_bilinear_c3(const unsigned char* src, int srcw, int srch, int srcstride, unsigned char* dst, int w, int h, int stride)
{
    const int INTER_RESIZE_COEF_BITS = 11;
    const int INTER_RESIZE_COEF_SCALE = 1 << INTER_RESIZE_COEF_BITS;
    //     const int ONE=INTER_RESIZE_COEF_SCALE;

    double scale_x = (double)srcw / w;
    double scale_y = (double)srch / h;

    int* buf = new int[w + h + w + h];

    int* xofs = buf;     //new int[w];
    int* yofs = buf + w; //new int[h];

    short* ialpha = (short*)(buf + w + h);    //new short[w * 2];
    short* ibeta = (short*)(buf + w + h + w); //new short[h * 2];

    float fx;
    float fy;
    int sx;
    int sy;

#define SATURATE_CAST_SHORT(X) (short)::std::min(::std::max((int)(X + (X >= 0.f ? 0.5f : -0.5f)), SHRT_MIN), SHRT_MAX);

    for (int dx = 0; dx < w; dx++)
    {
        fx = (float)((dx + 0.5) * scale_x - 0.5);
        sx = static_cast<int>(floor(fx));
        fx -= sx;

        if (sx < 0)
        {
            sx = 0;
            fx = 0.f;
        }
        if (sx >= srcw - 1)
        {
            sx = srcw - 2;
            fx = 1.f;
        }

        xofs[dx] = sx * 3;

        float a0 = (1.f - fx) * INTER_RESIZE_COEF_SCALE;
        float a1 = fx * INTER_RESIZE_COEF_SCALE;

        ialpha[dx * 2] = SATURATE_CAST_SHORT(a0);
        ialpha[dx * 2 + 1] = SATURATE_CAST_SHORT(a1);
    }

    for (int dy = 0; dy < h; dy++)
    {
        fy = (float)((dy + 0.5) * scale_y - 0.5);
        sy = static_cast<int>(floor(fy));
        fy -= sy;

        if (sy < 0)
        {
            sy = 0;
            fy = 0.f;
        }
        if (sy >= srch - 1)
        {
            sy = srch - 2;
            fy = 1.f;
        }

        yofs[dy] = sy;

        float b0 = (1.f - fy) * INTER_RESIZE_COEF_SCALE;
        float b1 = fy * INTER_RESIZE_COEF_SCALE;

        ibeta[dy * 2] = SATURATE_CAST_SHORT(b0);
        ibeta[dy * 2 + 1] = SATURATE_CAST_SHORT(b1);
    }

#undef SATURATE_CAST_SHORT

    // loop body
    Mat rowsbuf0(w * 3 + 1, (size_t)2u);
    Mat rowsbuf1(w * 3 + 1, (size_t)2u);
    short* rows0 = (short*)rowsbuf0.data;
    short* rows1 = (short*)rowsbuf1.data;

    int prev_sy1 = -2;

    for (int dy = 0; dy < h; dy++)
    {
        sy = yofs[dy];

        if (sy == prev_sy1)
        {
            // reuse all rows
        }
        else if (sy == prev_sy1 + 1)
        {
            // hresize one row
            short* rows0_old = rows0;
            rows0 = rows1;
            rows1 = rows0_old;
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                const unsigned char* S1p = S1 + sx;
#if __ARM_NEON
                int16x4_t _a0 = vdup_n_s16(a0);
                int16x4_t _a1 = vdup_n_s16(a1);
                uint8x8_t _S1 = uint8x8_t();

                _S1 = vld1_lane_u8(S1p, _S1, 0);
                _S1 = vld1_lane_u8(S1p + 1, _S1, 1);
                _S1 = vld1_lane_u8(S1p + 2, _S1, 2);
                _S1 = vld1_lane_u8(S1p + 3, _S1, 3);
                _S1 = vld1_lane_u8(S1p + 4, _S1, 4);
                _S1 = vld1_lane_u8(S1p + 5, _S1, 5);

                int16x8_t _S116 = vreinterpretq_s16_u16(vmovl_u8(_S1));
                int16x4_t _S1low = vget_low_s16(_S116);
                int16x4_t _S1high = vext_s16(_S1low, vget_high_s16(_S116), 3);
                int32x4_t _rows1 = vmull_s16(_S1low, _a0);
                _rows1 = vmlal_s16(_rows1, _S1high, _a1);
                int16x4_t _rows1_sr4 = vshrn_n_s32(_rows1, 4);
                vst1_s16(rows1p, _rows1_sr4);
#else
                rows1p[0] = (S1p[0] * a0 + S1p[3] * a1) >> 4;
                rows1p[1] = (S1p[1] * a0 + S1p[4] * a1) >> 4;
                rows1p[2] = (S1p[2] * a0 + S1p[5] * a1) >> 4;
#endif // __ARM_NEON

                ialphap += 2;
                rows1p += 3;
            }
        }
        else
        {
            // hresize two rows
            const unsigned char* S0 = src + srcstride * (sy);
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows0p = rows0;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                const unsigned char* S0p = S0 + sx;
                const unsigned char* S1p = S1 + sx;
#if __ARM_NEON
                int16x4_t _a0 = vdup_n_s16(a0);
                int16x4_t _a1 = vdup_n_s16(a1);
                uint8x8_t _S0 = uint8x8_t();
                uint8x8_t _S1 = uint8x8_t();

                _S0 = vld1_lane_u8(S0p, _S0, 0);
                _S0 = vld1_lane_u8(S0p + 1, _S0, 1);
                _S0 = vld1_lane_u8(S0p + 2, _S0, 2);
                _S0 = vld1_lane_u8(S0p + 3, _S0, 3);
                _S0 = vld1_lane_u8(S0p + 4, _S0, 4);
                _S0 = vld1_lane_u8(S0p + 5, _S0, 5);

                _S1 = vld1_lane_u8(S1p, _S1, 0);
                _S1 = vld1_lane_u8(S1p + 1, _S1, 1);
                _S1 = vld1_lane_u8(S1p + 2, _S1, 2);
                _S1 = vld1_lane_u8(S1p + 3, _S1, 3);
                _S1 = vld1_lane_u8(S1p + 4, _S1, 4);
                _S1 = vld1_lane_u8(S1p + 5, _S1, 5);

                int16x8_t _S016 = vreinterpretq_s16_u16(vmovl_u8(_S0));
                int16x8_t _S116 = vreinterpretq_s16_u16(vmovl_u8(_S1));
                int16x4_t _S0low = vget_low_s16(_S016);
                int16x4_t _S1low = vget_low_s16(_S116);
                int16x4_t _S0high = vext_s16(_S0low, vget_high_s16(_S016), 3);
                int16x4_t _S1high = vext_s16(_S1low, vget_high_s16(_S116), 3);
                int32x4_t _rows0 = vmull_s16(_S0low, _a0);
                int32x4_t _rows1 = vmull_s16(_S1low, _a0);
                _rows0 = vmlal_s16(_rows0, _S0high, _a1);
                _rows1 = vmlal_s16(_rows1, _S1high, _a1);
                int16x4_t _rows0_sr4 = vshrn_n_s32(_rows0, 4);
                int16x4_t _rows1_sr4 = vshrn_n_s32(_rows1, 4);
                vst1_s16(rows0p, _rows0_sr4);
                vst1_s16(rows1p, _rows1_sr4);
#else
                rows0p[0] = (S0p[0] * a0 + S0p[3] * a1) >> 4;
                rows0p[1] = (S0p[1] * a0 + S0p[4] * a1) >> 4;
                rows0p[2] = (S0p[2] * a0 + S0p[5] * a1) >> 4;
                rows1p[0] = (S1p[0] * a0 + S1p[3] * a1) >> 4;
                rows1p[1] = (S1p[1] * a0 + S1p[4] * a1) >> 4;
                rows1p[2] = (S1p[2] * a0 + S1p[5] * a1) >> 4;
#endif // __ARM_NEON

                ialphap += 2;
                rows0p += 3;
                rows1p += 3;
            }
        }

        prev_sy1 = sy;

        // vresize
        short b0 = ibeta[0];
        short b1 = ibeta[1];

        short* rows0p = rows0;
        short* rows1p = rows1;
        unsigned char* Dp = dst + stride * (dy);

#if __ARM_NEON
        int nn = (w * 3) >> 3;
#else
        int nn = 0;
#endif
        int remain = (w * 3) - (nn << 3);

#if __ARM_NEON
#if __aarch64__
        int16x4_t _b0 = vdup_n_s16(b0);
        int16x4_t _b1 = vdup_n_s16(b1);
        int32x4_t _v2 = vdupq_n_s32(2);
        for (; nn > 0; nn--)
        {
            int16x4_t _rows0p_sr4 = vld1_s16(rows0p);
            int16x4_t _rows1p_sr4 = vld1_s16(rows1p);
            int16x4_t _rows0p_1_sr4 = vld1_s16(rows0p + 4);
            int16x4_t _rows1p_1_sr4 = vld1_s16(rows1p + 4);

            int32x4_t _rows0p_sr4_mb0 = vmull_s16(_rows0p_sr4, _b0);
            int32x4_t _rows1p_sr4_mb1 = vmull_s16(_rows1p_sr4, _b1);
            int32x4_t _rows0p_1_sr4_mb0 = vmull_s16(_rows0p_1_sr4, _b0);
            int32x4_t _rows1p_1_sr4_mb1 = vmull_s16(_rows1p_1_sr4, _b1);

            int32x4_t _acc = _v2;
            _acc = vsraq_n_s32(_acc, _rows0p_sr4_mb0, 16);
            _acc = vsraq_n_s32(_acc, _rows1p_sr4_mb1, 16);

            int32x4_t _acc_1 = _v2;
            _acc_1 = vsraq_n_s32(_acc_1, _rows0p_1_sr4_mb0, 16);
            _acc_1 = vsraq_n_s32(_acc_1, _rows1p_1_sr4_mb1, 16);

            int16x4_t _acc16 = vshrn_n_s32(_acc, 2);
            int16x4_t _acc16_1 = vshrn_n_s32(_acc_1, 2);

            uint8x8_t _Dp = vqmovun_s16(vcombine_s16(_acc16, _acc16_1));

            vst1_u8(Dp, _Dp);

            Dp += 8;
            rows0p += 8;
            rows1p += 8;
        }
#else
        if (nn > 0)
        {
            asm volatile(
                "vdup.s16   d16, %8         \n"
                "mov        r4, #2          \n"
                "vdup.s16   d17, %9         \n"
                "vdup.s32   q12, r4         \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "0:                         \n"
                "vmull.s16  q0, d2, d16     \n"
                "vmull.s16  q1, d3, d16     \n"
                "vorr.s32   q10, q12, q12   \n"
                "vorr.s32   q11, q12, q12   \n"
                "vmull.s16  q2, d6, d17     \n"
                "vmull.s16  q3, d7, d17     \n"
                "vsra.s32   q10, q0, #16    \n"
                "vsra.s32   q11, q1, #16    \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "vsra.s32   q10, q2, #16    \n"
                "vsra.s32   q11, q3, #16    \n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "vshrn.s32  d20, q10, #2    \n"
                "vshrn.s32  d21, q11, #2    \n"
                "vqmovun.s16 d20, q10        \n"
                "vst1.8     {d20}, [%2]!    \n"
                "subs       %3, #1          \n"
                "bne        0b              \n"
                "sub        %0, #16         \n"
                "sub        %1, #16         \n"
                : "=r"(rows0p), // %0
                "=r"(rows1p), // %1
                "=r"(Dp),     // %2
                "=r"(nn)      // %3
                : "0"(rows0p),
                "1"(rows1p),
                "2"(Dp),
                "3"(nn),
                "r"(b0), // %8
                "r"(b1)  // %9
                : "cc", "memory", "r4", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11", "q12");
        }
#endif // __aarch64__
#endif // __ARM_NEON
        for (; remain; --remain)
        {
            //             D[x] = (rows0[x]*b0 + rows1[x]*b1) >> INTER_RESIZE_COEF_BITS;
            *Dp++ = (unsigned char)(((short)((b0 * (short)(*rows0p++)) >> 16) + (short)((b1 * (short)(*rows1p++)) >> 16) + 2) >> 2);
        }

        ibeta += 2;
    }

    delete[] buf;
}

void resize_bilinear_c4(const unsigned char* src, int srcw, int srch, int srcstride, unsigned char* dst, int w, int h, int stride)
{
    const int INTER_RESIZE_COEF_BITS = 11;
    const int INTER_RESIZE_COEF_SCALE = 1 << INTER_RESIZE_COEF_BITS;
    //     const int ONE=INTER_RESIZE_COEF_SCALE;

    double scale_x = (double)srcw / w;
    double scale_y = (double)srch / h;

    int* buf = new int[w + h + w + h];

    int* xofs = buf;     //new int[w];
    int* yofs = buf + w; //new int[h];

    short* ialpha = (short*)(buf + w + h);    //new short[w * 2];
    short* ibeta = (short*)(buf + w + h + w); //new short[h * 2];

    float fx;
    float fy;
    int sx;
    int sy;

#define SATURATE_CAST_SHORT(X) (short)::std::min(::std::max((int)(X + (X >= 0.f ? 0.5f : -0.5f)), SHRT_MIN), SHRT_MAX);

    for (int dx = 0; dx < w; dx++)
    {
        fx = (float)((dx + 0.5) * scale_x - 0.5);
        sx = static_cast<int>(floor(fx));
        fx -= sx;

        if (sx < 0)
        {
            sx = 0;
            fx = 0.f;
        }
        if (sx >= srcw - 1)
        {
            sx = srcw - 2;
            fx = 1.f;
        }

        xofs[dx] = sx * 4;

        float a0 = (1.f - fx) * INTER_RESIZE_COEF_SCALE;
        float a1 = fx * INTER_RESIZE_COEF_SCALE;

        ialpha[dx * 2] = SATURATE_CAST_SHORT(a0);
        ialpha[dx * 2 + 1] = SATURATE_CAST_SHORT(a1);
    }

    for (int dy = 0; dy < h; dy++)
    {
        fy = (float)((dy + 0.5) * scale_y - 0.5);
        sy = static_cast<int>(floor(fy));
        fy -= sy;

        if (sy < 0)
        {
            sy = 0;
            fy = 0.f;
        }
        if (sy >= srch - 1)
        {
            sy = srch - 2;
            fy = 1.f;
        }

        yofs[dy] = sy;

        float b0 = (1.f - fy) * INTER_RESIZE_COEF_SCALE;
        float b1 = fy * INTER_RESIZE_COEF_SCALE;

        ibeta[dy * 2] = SATURATE_CAST_SHORT(b0);
        ibeta[dy * 2 + 1] = SATURATE_CAST_SHORT(b1);
    }

#undef SATURATE_CAST_SHORT

    // loop body
    Mat rowsbuf0(w * 4, (size_t)2u);
    Mat rowsbuf1(w * 4, (size_t)2u);
    short* rows0 = (short*)rowsbuf0.data;
    short* rows1 = (short*)rowsbuf1.data;

    int prev_sy1 = -2;

    for (int dy = 0; dy < h; dy++)
    {
        sy = yofs[dy];

        if (sy == prev_sy1)
        {
            // reuse all rows
        }
        else if (sy == prev_sy1 + 1)
        {
            // hresize one row
            short* rows0_old = rows0;
            rows0 = rows1;
            rows1 = rows0_old;
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                const unsigned char* S1p = S1 + sx;
#if __ARM_NEON
                int16x4_t _a0 = vdup_n_s16(a0);
                int16x4_t _a1 = vdup_n_s16(a1);
                uint8x8_t _S1 = vld1_u8(S1p);
                int16x8_t _S116 = vreinterpretq_s16_u16(vmovl_u8(_S1));
                int16x4_t _S1low = vget_low_s16(_S116);
                int16x4_t _S1high = vget_high_s16(_S116);
                int32x4_t _rows1 = vmull_s16(_S1low, _a0);
                _rows1 = vmlal_s16(_rows1, _S1high, _a1);
                int16x4_t _rows1_sr4 = vshrn_n_s32(_rows1, 4);
                vst1_s16(rows1p, _rows1_sr4);
#else
                rows1p[0] = (S1p[0] * a0 + S1p[4] * a1) >> 4;
                rows1p[1] = (S1p[1] * a0 + S1p[5] * a1) >> 4;
                rows1p[2] = (S1p[2] * a0 + S1p[6] * a1) >> 4;
                rows1p[3] = (S1p[3] * a0 + S1p[7] * a1) >> 4;
#endif // __ARM_NEON

                ialphap += 2;
                rows1p += 4;
            }
        }
        else
        {
            // hresize two rows
            const unsigned char* S0 = src + srcstride * (sy);
            const unsigned char* S1 = src + srcstride * (sy + 1);

            const short* ialphap = ialpha;
            short* rows0p = rows0;
            short* rows1p = rows1;
            for (int dx = 0; dx < w; dx++)
            {
                sx = xofs[dx];
                short a0 = ialphap[0];
                short a1 = ialphap[1];

                const unsigned char* S0p = S0 + sx;
                const unsigned char* S1p = S1 + sx;
#if __ARM_NEON
                int16x4_t _a0 = vdup_n_s16(a0);
                int16x4_t _a1 = vdup_n_s16(a1);
                uint8x8_t _S0 = vld1_u8(S0p);
                uint8x8_t _S1 = vld1_u8(S1p);
                int16x8_t _S016 = vreinterpretq_s16_u16(vmovl_u8(_S0));
                int16x8_t _S116 = vreinterpretq_s16_u16(vmovl_u8(_S1));
                int16x4_t _S0low = vget_low_s16(_S016);
                int16x4_t _S1low = vget_low_s16(_S116);
                int16x4_t _S0high = vget_high_s16(_S016);
                int16x4_t _S1high = vget_high_s16(_S116);
                int32x4_t _rows0 = vmull_s16(_S0low, _a0);
                int32x4_t _rows1 = vmull_s16(_S1low, _a0);
                _rows0 = vmlal_s16(_rows0, _S0high, _a1);
                _rows1 = vmlal_s16(_rows1, _S1high, _a1);
                int16x4_t _rows0_sr4 = vshrn_n_s32(_rows0, 4);
                int16x4_t _rows1_sr4 = vshrn_n_s32(_rows1, 4);
                vst1_s16(rows0p, _rows0_sr4);
                vst1_s16(rows1p, _rows1_sr4);
#else
                rows0p[0] = (S0p[0] * a0 + S0p[4] * a1) >> 4;
                rows0p[1] = (S0p[1] * a0 + S0p[5] * a1) >> 4;
                rows0p[2] = (S0p[2] * a0 + S0p[6] * a1) >> 4;
                rows0p[3] = (S0p[3] * a0 + S0p[7] * a1) >> 4;
                rows1p[0] = (S1p[0] * a0 + S1p[4] * a1) >> 4;
                rows1p[1] = (S1p[1] * a0 + S1p[5] * a1) >> 4;
                rows1p[2] = (S1p[2] * a0 + S1p[6] * a1) >> 4;
                rows1p[3] = (S1p[3] * a0 + S1p[7] * a1) >> 4;
#endif // __ARM_NEON

                ialphap += 2;
                rows0p += 4;
                rows1p += 4;
            }
        }

        prev_sy1 = sy;

        // vresize
        short b0 = ibeta[0];
        short b1 = ibeta[1];

        short* rows0p = rows0;
        short* rows1p = rows1;
        unsigned char* Dp = dst + stride * (dy);

#if __ARM_NEON
        int nn = (w * 4) >> 3;
#else
        int nn = 0;
#endif
        int remain = (w * 4) - (nn << 3);

#if __ARM_NEON
#if __aarch64__
        int16x4_t _b0 = vdup_n_s16(b0);
        int16x4_t _b1 = vdup_n_s16(b1);
        int32x4_t _v2 = vdupq_n_s32(2);
        for (; nn > 0; nn--)
        {
            int16x4_t _rows0p_sr4 = vld1_s16(rows0p);
            int16x4_t _rows1p_sr4 = vld1_s16(rows1p);
            int16x4_t _rows0p_1_sr4 = vld1_s16(rows0p + 4);
            int16x4_t _rows1p_1_sr4 = vld1_s16(rows1p + 4);

            int32x4_t _rows0p_sr4_mb0 = vmull_s16(_rows0p_sr4, _b0);
            int32x4_t _rows1p_sr4_mb1 = vmull_s16(_rows1p_sr4, _b1);
            int32x4_t _rows0p_1_sr4_mb0 = vmull_s16(_rows0p_1_sr4, _b0);
            int32x4_t _rows1p_1_sr4_mb1 = vmull_s16(_rows1p_1_sr4, _b1);

            int32x4_t _acc = _v2;
            _acc = vsraq_n_s32(_acc, _rows0p_sr4_mb0, 16);
            _acc = vsraq_n_s32(_acc, _rows1p_sr4_mb1, 16);

            int32x4_t _acc_1 = _v2;
            _acc_1 = vsraq_n_s32(_acc_1, _rows0p_1_sr4_mb0, 16);
            _acc_1 = vsraq_n_s32(_acc_1, _rows1p_1_sr4_mb1, 16);

            int16x4_t _acc16 = vshrn_n_s32(_acc, 2);
            int16x4_t _acc16_1 = vshrn_n_s32(_acc_1, 2);

            uint8x8_t _Dp = vqmovun_s16(vcombine_s16(_acc16, _acc16_1));

            vst1_u8(Dp, _Dp);

            Dp += 8;
            rows0p += 8;
            rows1p += 8;
        }
#else
        if (nn > 0)
        {
            asm volatile(
                "vdup.s16   d16, %8         \n"
                "mov        r4, #2          \n"
                "vdup.s16   d17, %9         \n"
                "vdup.s32   q12, r4         \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "0:                         \n"
                "vmull.s16  q0, d2, d16     \n"
                "vmull.s16  q1, d3, d16     \n"
                "vorr.s32   q10, q12, q12   \n"
                "vorr.s32   q11, q12, q12   \n"
                "vmull.s16  q2, d6, d17     \n"
                "vmull.s16  q3, d7, d17     \n"
                "vsra.s32   q10, q0, #16    \n"
                "vsra.s32   q11, q1, #16    \n"
                "pld        [%0, #128]      \n"
                "vld1.s16   {d2-d3}, [%0 :128]!\n"
                "vsra.s32   q10, q2, #16    \n"
                "vsra.s32   q11, q3, #16    \n"
                "pld        [%1, #128]      \n"
                "vld1.s16   {d6-d7}, [%1 :128]!\n"
                "vshrn.s32  d20, q10, #2    \n"
                "vshrn.s32  d21, q11, #2    \n"
                "vqmovun.s16 d20, q10        \n"
                "vst1.8     {d20}, [%2]!    \n"
                "subs       %3, #1          \n"
                "bne        0b              \n"
                "sub        %0, #16         \n"
                "sub        %1, #16         \n"
                : "=r"(rows0p), // %0
                "=r"(rows1p), // %1
                "=r"(Dp),     // %2
                "=r"(nn)      // %3
                : "0"(rows0p),
                "1"(rows1p),
                "2"(Dp),
                "3"(nn),
                "r"(b0), // %8
                "r"(b1)  // %9
                : "cc", "memory", "r4", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11", "q12");
        }
#endif // __aarch64__
#endif // __ARM_NEON
        for (; remain; --remain)
        {
            //             D[x] = (rows0[x]*b0 + rows1[x]*b1) >> INTER_RESIZE_COEF_BITS;
            *Dp++ = (unsigned char)(((short)((b0 * (short)(*rows0p++)) >> 16) + (short)((b1 * (short)(*rows1p++)) >> 16) + 2) >> 2);
        }

        ibeta += 2;
    }

    delete[] buf;
}

void resize_bilinear_yuv420sp(const unsigned char* src, int srcw, int srch, unsigned char* dst, int w, int h)
{
    // assert srcw % 2 == 0
    // assert srch % 2 == 0
    // assert w % 2 == 0
    // assert h % 2 == 0

    const unsigned char* srcY = src;
    unsigned char* dstY = dst;
    resize_bilinear_c1(srcY, srcw, srch, dstY, w, h);

    const unsigned char* srcUV = src + srcw * srch;
    unsigned char* dstUV = dst + w * h;
    resize_bilinear_c2(srcUV, srcw / 2, srch / 2, dstUV, w / 2, h / 2);
}
#endif // NCNN_PIXEL

} // namespace ncnn