func PairingAggregatePkInG1(ctx Pairing, PK *P1Affine, pkValidate bool,
                            sig *P2Affine, sigGroupcheck bool, msg []byte,
                            optional ...[]byte) int { // aug
    var aug []byte
    var uaug *C.byte
    if len(optional) > 0 {
        aug = optional[0]
        if len(aug) > 0 {
            uaug = (*C.byte)(&aug[0])
        }
    }
    var umsg *C.byte
    if len(msg) > 0 {
        umsg = (*C.byte)(&msg[0])
    }

    r := C.blst_pairing_chk_n_aggr_pk_in_g1(&ctx[0],
                                            PK, C.bool(pkValidate),
                                            sig, C.bool(sigGroupcheck),
                                            umsg, C.size_t(len(msg)),
                                            uaug, C.size_t(len(aug)))

    return int(r)
}

func PairingMulNAggregatePkInG1(ctx Pairing, PK *P1Affine, pkValidate bool,
                                sig *P2Affine, sigGroupcheck bool,
                                rand *Scalar, randBits int, msg []byte,
                                optional ...[]byte) int { // aug
    var aug []byte
    var uaug *C.byte
    if len(optional) > 0 {
        aug = optional[0]
        if len(aug) > 0 {
            uaug = (*C.byte)(&aug[0])
        }
    }
    var umsg *C.byte
    if len(msg) > 0 {
        umsg = (*C.byte)(&msg[0])
    }

    r := C.blst_pairing_chk_n_mul_n_aggr_pk_in_g1(&ctx[0],
                                                  PK, C.bool(pkValidate),
                                                  sig, C.bool(sigGroupcheck),
                                                  &rand.b[0], C.size_t(randBits),
                                                  umsg, C.size_t(len(msg)),
                                                  uaug, C.size_t(len(aug)))

    return int(r)
}

//
// Serialization/Deserialization.
//

// P1 Serdes
func (p1 *P1Affine) Serialize() []byte {
    var out [BLST_P1_SERIALIZE_BYTES]byte
    C.blst_p1_affine_serialize((*C.byte)(&out[0]), p1)
    return out[:]
}

func (p1 *P1Affine) Deserialize(in []byte) *P1Affine {
    if len(in) != BLST_P1_SERIALIZE_BYTES {
        return nil
    }
    if C.blst_p1_deserialize(p1, (*C.byte)(&in[0])) != C.BLST_SUCCESS {
        return nil
    }
    return p1
}
func (p1 *P1Affine) Compress() []byte {
    var out [BLST_P1_COMPRESS_BYTES]byte
    C.blst_p1_affine_compress((*C.byte)(&out[0]), p1)
    return out[:]
}

func (p1 *P1Affine) Uncompress(in []byte) *P1Affine {
    if len(in) != BLST_P1_COMPRESS_BYTES {
        return nil
    }
    if C.blst_p1_uncompress(p1, (*C.byte)(&in[0])) != C.BLST_SUCCESS {
        return nil
    }
    return p1
}

func (p1 *P1Affine) InG1() bool {
  return bool(C.blst_p1_affine_in_g1(p1))
}

func (dummy *P1Affine) BatchUncompress(in [][]byte) []*P1Affine {
    // Allocate space for all of the resulting points. Later we'll save pointers
    // and return those so that the result could be used in other functions,
    // such as MultipleAggregateVerify.
    n := len(in)
    points := make([]P1Affine, n)
    pointsPtrs := make([]*P1Affine, n)

    numCores := runtime.GOMAXPROCS(0)
    numThreads := maxProcs
    if numThreads > numCores {
        numThreads = numCores
    }
    if numThreads > n {
        numThreads = n
    }
    // Each thread will determine next message to process by atomically
    // incrementing curItem, process corresponding point, and
    // repeat until n is exceeded. Each thread will send a result (true for
    // success, false for failure) into the channel when complete.
    resCh := make(chan bool, numThreads)
    valid := int32(1)
    curItem := uint32(0)
    for tid := 0; tid < numThreads; tid++ {
        go func() {
            for atomic.LoadInt32(&valid) > 0 {
                // Get a work item
                work := atomic.AddUint32(&curItem, 1) - 1
                if work >= uint32(n) {
                    break
                }
                if points[work].Uncompress(in[work]) == nil {
                    atomic.StoreInt32(&valid, 0)
                    break
                }
                pointsPtrs[work] = &points[work]
            }
            if atomic.LoadInt32(&valid) > 0 {
                resCh <- true
            } else {
                resCh <- false
            }
        }()
    }

    // Collect the threads
    result := true
    for i := 0; i < numThreads; i++ {
        if ! <-resCh {
            result = false
        }
    }
    if atomic.LoadInt32(&valid) == 0 || result == false {
        return nil
    }
    return pointsPtrs
}

func (p1 *P1) Serialize() []byte {
    var out [BLST_P1_SERIALIZE_BYTES]byte
    C.blst_p1_serialize((*C.byte)(&out[0]), p1)
    return out[:]
}
func (p1 *P1) Compress() []byte {
    var out [BLST_P1_COMPRESS_BYTES]byte
    C.blst_p1_compress((*C.byte)(&out[0]), p1)
    return out[:]
}

func (p1 *P1) Mult(scalar []byte, optional ...int) *P1 {
    var nbits int
    if len(optional) > 0 {
        nbits = optional[0]
    } else {
        nbits = len(scalar)*8
    }
    var ret P1
    C.blst_p1_mult(&ret, p1, (*C.byte)(&scalar[0]), C.size_t(nbits))
    return &ret
}

func P1Generator() *P1 {
    return C.blst_p1_generator()
}

//
// Affine
//

func (p *P1) ToAffine() *P1Affine {
    var pa P1Affine
    C.blst_p1_to_affine(&pa, p)
    return &pa
}

func (p *P1) FromAffine(pa *P1Affine) {
    C.blst_p1_from_affine(p, pa)
}

//
// Hash
//
func HashToG1(msg []byte, dst []byte,
        optional ...[]byte) *P1 { // aug
    var q P1

    // Handle zero length message
    var msgC *C.byte
    if len(msg) > 0 {
        msgC = (*C.byte)(&msg[0])
    }

    var dstC *C.byte
    if len(dst) > 0 {
        dstC = (*C.byte)(&dst[0])
    }

    var aug []byte
    var augC *C.byte
    if len(optional) > 0 {
        aug = optional[0]
        if len(aug) > 0 {
            augC = (*C.byte)(&aug[0])
        }
    }

    C.blst_hash_to_g1(&q, msgC, C.size_t(len(msg)),
                          dstC, C.size_t(len(dst)),
                          augC, C.size_t(len(aug)))
    return &q
}

func EncodeToG1(msg []byte, dst []byte,
        optional ...[]byte) *P1 { // aug
    var q P1

    // Handle zero length message
    var msgC *C.byte
    if len(msg) > 0 {
        msgC = (*C.byte)(&msg[0])
    }

    var dstC *C.byte
    if len(dst) > 0 {
        dstC = (*C.byte)(&dst[0])
    }

    var aug []byte
    var augC *C.byte
    if len(optional) > 0 {
        aug = optional[0]
        if len(aug) > 0 {
            augC = (*C.byte)(&aug[0])
        }
    }

    C.blst_encode_to_g1(&q, msgC, C.size_t(len(msg)),
                            dstC, C.size_t(len(dst)),
                            augC, C.size_t(len(aug)))
    return &q
}

//
// Multi-point/scalar operations
//

func P1sToAffine(points []*P1, optional ...int) P1Affines {
    var npoints int
    if len(optional) > 0 {
        npoints = optional[0]
    } else {
        npoints = len(points)
    }
    ret := make([]P1Affine, npoints)
    _cgoCheckPointer := func(...interface{}) {}
    C.blst_p1s_to_affine(&ret[0], &points[0], C.size_t(npoints))
    return ret
}

func (points P1s) ToAffine(optional ...P1Affines) P1Affines {
    npoints := len(points)
    var ret P1Affines

    if len(optional) > 0 {  // used in benchmark
        ret = optional[0]
        if len(ret) < npoints {
            panic("npoints mismatch")
        }
    } else {
        ret = make([]P1Affine, npoints)
    }

    if maxProcs < 2 || npoints < 768 {
        C.go_p1slice_to_affine(&ret[0], &points[0], C.size_t(npoints))
        return ret
    }

    nslices := (npoints + 511) / 512
    if nslices > maxProcs {
        nslices = maxProcs
    }
    delta, rem := npoints/nslices + 1, npoints%nslices

    var wg sync.WaitGroup
    wg.Add(nslices)
    for x := 0; x < npoints; x += delta {
        if rem == 0 {
            delta -= 1
        }
        rem -= 1
        go func(out *P1Affine, inp *P1, delta int) {
            C.go_p1slice_to_affine(out, inp, C.size_t(delta))
            wg.Done()
        }(&ret[x], &points[x], delta)
    }
    wg.Wait()

    return ret
}

//
// Batch addition
//

func P1AffinesAdd(points []*P1Affine, optional ...int) *P1 {
    var npoints int
    if len(optional) > 0 {
        npoints = optional[0]
    } else {
        npoints = len(points)
    }
    var ret P1
    _cgoCheckPointer := func(...interface{}) {}
    C.blst_p1s_add(&ret, &points[0], C.size_t(npoints))
    return &ret
}

func (points P1Affines) Add() *P1 {
    npoints := len(points)
    if maxProcs < 2 || npoints < 768 {
        var ret P1
        C.go_p1slice_add(&ret, &points[0], C.size_t(npoints))
        return &ret
    }

    nslices := (npoints + 511) / 512
    if nslices > maxProcs {
        nslices = maxProcs
    }
    delta, rem := npoints/nslices + 1, npoints%nslices

    msgs := make(chan P1, nslices)
    for x := 0; x < npoints; x += delta {
        if rem == 0 {
            delta -= 1
        }
        rem -= 1
        go func(points *P1Affine, delta int) {
            var ret P1
            C.go_p1slice_add(&ret, points, C.size_t(delta))
            msgs <- ret
        }(&points[x], delta)
    }

    ret := <- msgs
    for i := 1; i < nslices; i++ {
        msg := <- msgs
        C.blst_p1_add_or_double(&ret, &ret, &msg)
    }
    return &ret
}

func (points P1s) Add() *P1 {
    return points.ToAffine().Add()
}

//
// Multi-scalar multiplication
//

func P1AffinesMult(pointsIf interface{}, scalarsIf interface{}, nbits int) *P1 {
    var npoints int
    switch val := pointsIf.(type) {
    case []*P1Affine:
        npoints = len(val)
    case []P1Affine:
        npoints = len(val)
    case P1Affines:
        npoints = len(val)
    default:
        panic(fmt.Sprintf("unsupported type %T", val))
    }

    nbytes := (nbits+7)/8
    var scalars []*C.byte
    switch val := scalarsIf.(type) {
    case []byte:
        if len(val) < npoints*nbytes {
            return nil
        }
    case [][]byte:
        if len(val) < npoints {
            return nil
        }
        scalars = make([]*C.byte, npoints)
        for i := range scalars {
            scalars[i] = (*C.byte)(&val[i][0])
        }
    case []Scalar:
        if len(val) < npoints {
            return nil
        }
        if nbits <= 248 {
            scalars = make([]*C.byte, npoints)
            for i := range scalars {
                scalars[i] = &val[i].b[0]
            }
        }
    case []*Scalar:
        if len(val) < npoints {
            return nil
        }
        scalars = make([]*C.byte, npoints)
        for i := range scalars {
            scalars[i] = &val[i].b[0]
        }
    default:
        panic(fmt.Sprintf("unsupported type %T",val))
    }

    numThreads := maxProcs
    numCores := runtime.GOMAXPROCS(0)
    if numCores < maxProcs {
        numThreads = numCores
    }

    if numThreads < 2 || npoints < 32 {
        sz := int(C.blst_p1s_mult_pippenger_scratch_sizeof(C.size_t(npoints)))/8
        scratch := make([]uint64, sz)

        pointsBySlice := [2]*P1Affine{nil, nil}
        var p_points **P1Affine
        switch val := pointsIf.(type) {
        case []*P1Affine:
            p_points = &val[0]
        case []P1Affine:
            pointsBySlice[0] = &val[0]
            p_points = &pointsBySlice[0]
        case P1Affines:
            pointsBySlice[0] = &val[0]
            p_points = &pointsBySlice[0]
        }

        scalarsBySlice := [2]*C.byte{nil, nil}
        var p_scalars **C.byte
        switch val := scalarsIf.(type) {
        case []byte:
            scalarsBySlice[0] = (*C.byte)(&val[0])
            p_scalars = &scalarsBySlice[0]
        case [][]byte:
            p_scalars = &scalars[0]
        case []Scalar:
            if nbits > 248 {
                scalarsBySlice[0] = (*C.byte)(&val[0].b[0])
                p_scalars = &scalarsBySlice[0]
            } else {
                p_scalars = &scalars[0]
            }
        case []*Scalar:
            p_scalars = &scalars[0]
        }

        var ret P1
        _cgoCheckPointer := func(...interface{}) {}
        C.blst_p1s_mult_pippenger(&ret, p_points, C.size_t(npoints),
                                        p_scalars, C.size_t(nbits),
                                        (*C.limb_t)(&scratch[0]))

        for i := range(scalars) {
            scalars[i] = nil
        }

        return &ret
    }

    // this is sizeof(scratch[0])
    sz := int(C.blst_p1s_mult_pippenger_scratch_sizeof(0))/8

    nx, ny, window := breakdown(nbits, pippenger_window_size(npoints),
                                numThreads)

    // |grid[]| holds "coordinates" and place for result
    grid := make([]struct { x, dx, y, dy int
                            point P1 }, nx*ny)

    dx := npoints/nx
    y := window*(ny-1)
    total := 0
    for ; total < nx; total++ {
        grid[total].x = total*dx
        grid[total].dx = dx
        grid[total].y = y
        grid[total].dy = nbits - y
    }
    grid[total-1].dx = npoints - grid[total-1].x

    for y > 0 {
        y -= window
        for i := 0; i < nx; i++ {
            grid[total].x = grid[i].x
            grid[total].dx = grid[i].dx
            grid[total].y = y
            grid[total].dy = window
            total++
        }
    }

    if numThreads > total {
        numThreads = total
    }

    msgsCh := make(chan int, ny)
    rowSync := make([]int32, ny)    // count up to |nx|
    curItem := int32(0)
    for tid := 0; tid < numThreads; tid++ {
        go func() {
            scratch := make([]uint64, sz << uint(window-1))
            pointsBySlice := [2]*P1Affine{nil, nil}
            scalarsBySlice := [2]*C.byte{nil, nil}
            _cgoCheckPointer := func(...interface{}) {}

            for {
                workItem := atomic.AddInt32(&curItem, 1) - 1
                if int(workItem) >= total {
                    break
                }

                x := grid[workItem].x
                y := grid[workItem].y

                var p_points **P1Affine
                switch val := pointsIf.(type) {
                case []*P1Affine:
                    p_points = &val[x]
                case []P1Affine:
                    pointsBySlice[0] = &val[x]
                    p_points = &pointsBySlice[0]
                case P1Affines:
                    pointsBySlice[0] = &val[x]
                    p_points = &pointsBySlice[0]
                }

                var p_scalars **C.byte
                switch val := scalarsIf.(type) {
                case []byte:
                    scalarsBySlice[0] = (*C.byte)(&val[x*nbytes])
                    p_scalars = &scalarsBySlice[0]
                case [][]byte:
                    p_scalars = &scalars[x]
                case []Scalar:
                    if nbits > 248 {
                        scalarsBySlice[0] = (*C.byte)(&val[x].b[0])
                        p_scalars = &scalarsBySlice[0]
                    } else {
                        p_scalars = &scalars[x]
                    }
                case []*Scalar:
                    p_scalars = &scalars[x]
                }

                C.blst_p1s_tile_pippenger(&grid[workItem].point,
                                          p_points, C.size_t(grid[workItem].dx),
                                          p_scalars, C.size_t(nbits),
                                          (*C.limb_t)(&scratch[0]),
                                          C.size_t(y), C.size_t(window));

                if atomic.AddInt32(&rowSync[y/window], 1) == int32(nx) {
                    msgsCh <- y     // "row" is done
                } else {
                    runtime.Gosched()   // be nice to the application
                }
            }

            pointsBySlice[0] = nil
            scalarsBySlice[0] = nil
        }()
    }

    var ret P1
    rows := make([]bool, ny)
    row := 0                        // actually index in |grid[]|
    for i := 0; i < ny; i++ {       // we expect |ny| messages, one per "row"
        y := <- msgsCh
        rows[y/window] = true       // mark the "row"
        for grid[row].y == y {      // if it's current "row", process it
            for row < total && grid[row].y == y {
                C.blst_p1_add_or_double(&ret, &ret, &grid[row].point)
                row++
            }
            if y == 0 {
                break               // one can as well 'return &ret' here
            }
            for j := 0; j < window; j++ {
                C.blst_p1_double(&ret, &ret)
            }
            y -= window
            if !rows[y/window] {    // see if next "row" was marked already
                break
            }
        }
    }

    for i := range(scalars) {
        scalars[i] = nil
    }

    return &ret
}

func (points P1Affines) Mult(scalarsIf interface{}, nbits int) *P1 {
    return P1AffinesMult(points, scalarsIf, nbits)
}

func (points P1s) Mult(scalarsIf interface{}, nbits int) *P1 {
    return points.ToAffine().Mult(scalarsIf, nbits)
}