// Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2023 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.

#version 450

#if NCNN_fp16_storage
#extension GL_EXT_shader_16bit_storage: require
struct sfpvec8 { f16vec4 abcd; f16vec4 efgh; };
#endif
#if NCNN_fp16_arithmetic
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
#endif

layout (constant_id = 0) const int op_type = 0;

#define shape_constant_id_offset 1
layout (constant_id = shape_constant_id_offset + 0) const int adims = 0;
layout (constant_id = shape_constant_id_offset + 1) const int aw = 0;
layout (constant_id = shape_constant_id_offset + 2) const int ah = 0;
layout (constant_id = shape_constant_id_offset + 3) const int ad = 0;
layout (constant_id = shape_constant_id_offset + 4) const int ac = 0;
layout (constant_id = shape_constant_id_offset + 5) const int acstep = 0;

layout (constant_id = shape_constant_id_offset + 6) const int bdims = 0;
layout (constant_id = shape_constant_id_offset + 7) const int bw = 0;
layout (constant_id = shape_constant_id_offset + 8) const int bh = 0;
layout (constant_id = shape_constant_id_offset + 9) const int bd = 0;
layout (constant_id = shape_constant_id_offset + 10) const int bc = 0;
layout (constant_id = shape_constant_id_offset + 11) const int bcstep = 0;

layout (constant_id = shape_constant_id_offset + 12) const int outdims = 0;
layout (constant_id = shape_constant_id_offset + 13) const int outw = 0;
layout (constant_id = shape_constant_id_offset + 14) const int outh = 0;
layout (constant_id = shape_constant_id_offset + 15) const int outd = 0;
layout (constant_id = shape_constant_id_offset + 16) const int outc = 0;
layout (constant_id = shape_constant_id_offset + 17) const int outcstep = 0;

#if NCNN_image_shader
layout (binding = 0) uniform unfp sampler3D a_blob_3d;
layout (binding = 1) uniform unfp sampler3D b_blob_3d;
layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
#else
layout (binding = 0) readonly buffer a_blob { sfpvec8 a_blob_data[]; };
layout (binding = 1) readonly buffer b_blob { sfp b_blob_data[]; };
layout (binding = 2) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
#endif

layout (push_constant) uniform parameter
{
    int adims;
    int aw;
    int ah;
    int ad;
    int ac;
    int acstep;

    int bdims;
    int bw;
    int bh;
    int bd;
    int bc;
    int bcstep;

    int outdims;
    int outw;
    int outh;
    int outd;
    int outc;
    int outcstep;
} p;

void main()
{
    int gx = int(gl_GlobalInvocationID.x);
    int gy = int(gl_GlobalInvocationID.y);
    int gz = int(gl_GlobalInvocationID.z);

    if (gx >= psc(outw) || gy >= psc(outh) * psc(outd) || gz >= psc(outc))
        return;

    int yd = gy / psc(outh);
    int yh = gy % psc(outh);

    // explicit broadcast
    int ax = min(gx, psc(aw) - 1);
    int ay = min(yd, psc(ad) - 1) * psc(ah) + min(yh, psc(ah) - 1);
    int az = min(gz, psc(ac) - 1);

    int bx = min(gx, psc(bw) - 1);
    int by = min(yd, psc(bd) - 1) * psc(bh) + min(yh, psc(bh) - 1);
    int bz = min(gz, psc(bc) - 1);

#if NCNN_image_shader
    afpvec8 v1 = image3d_ld8(a_blob_3d, ivec3(ax, ay, az));
    afp b = image3d_ld1(b_blob_3d, ivec3(bx, by, bz));
#else
    int ai = az * psc(acstep) + ay * psc(aw) + ax;
    int bi = bz * psc(bcstep) + by * psc(bw) + bx;

    afpvec8 v1 = buffer_ld8(a_blob_data, ai);
    afp b = buffer_ld1(b_blob_data, bi);
#endif
    afpvec8 v2;
    v2[0] = afpvec4(b);
    v2[1] = afpvec4(b);

    afpvec8 res;

    if (op_type == 0)
    {
        res[0] = v1[0] + v2[0];
        res[1] = v1[1] + v2[1];
    }
    if (op_type == 1)
    {
        res[0] = v1[0] - v2[0];
        res[1] = v1[1] - v2[1];
    }
    if (op_type == 2)
    {
        res[0] = v1[0] * v2[0];
        res[1] = v1[1] * v2[1];
    }
    if (op_type == 3)
    {
        res[0] = v1[0] / v2[0];
        res[1] = v1[1] / v2[1];
    }
    if (op_type == 4)
    {
        res[0] = max(v1[0], v2[0]);
        res[1] = max(v1[1], v2[1]);
    }
    if (op_type == 5)
    {
        res[0] = min(v1[0], v2[0]);
        res[1] = min(v1[1], v2[1]);
    }
    if (op_type == 6)
    {
        res[0] = pow(v1[0], v2[0]);
        res[1] = pow(v1[1], v2[1]);
    }
    if (op_type == 7)
    {
        res[0] = v2[0] - v1[0];
        res[1] = v2[1] - v1[1];
    }
    if (op_type == 8)
    {
        res[0] = v2[0] / v1[0];
        res[1] = v2[1] / v1[1];
    }
    if (op_type == 9)
    {
        res[0] = pow(v2[0], v1[0]);
        res[1] = pow(v2[1], v1[1]);
    }
    if (op_type == 10)
    {
#if NCNN_moltenvk
        res[0] = afpvec4(atan(vec4(v1[0]), vec4(v2[0])));
        res[1] = afpvec4(atan(vec4(v1[1]), vec4(v2[1])));
#else
        res[0] = atan(v1[0], v2[0]);
        res[1] = atan(v1[1], v2[1]);
#endif
    }
    if (op_type == 11)
    {
#if NCNN_moltenvk
        res[0] = afpvec4(atan(vec4(v2[0]), vec4(v1[0])));
        res[1] = afpvec4(atan(vec4(v2[1]), vec4(v1[1])));
#else
        res[0] = atan(v2[0], v1[0]);
        res[1] = atan(v2[1], v1[1]);
#endif
    }

#if NCNN_image_shader
    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
#else
    int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
    buffer_st8(top_blob_data, gi, res);
#endif
}