/// @file rainbow_keypair_computation.c
/// @brief Implementations for functions in rainbow_keypair_computation.h
///

#include "rainbow_keypair_computation.h"
#include "blas.h"
#include "blas_comm.h"
#include "rainbow_blas.h"
#include "rainbow_keypair.h"
#include <stdint.h>
#include <stdlib.h>
#include <string.h>

void PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_extcpk_to_pk(pk_t *pk, const ext_cpk_t *cpk) {
    const unsigned char *idx_l1 = cpk->l1_Q1;
    const unsigned char *idx_l2 = cpk->l2_Q1;
    for (unsigned i = 0; i < _V1; i++) {
        for (unsigned j = i; j < _V1; j++) {
            unsigned pub_idx = PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_idx_of_trimat(i, j, _PUB_N);
            memcpy(&pk->pk[_PUB_M_BYTE * pub_idx], idx_l1, _O1_BYTE);
            memcpy((&pk->pk[_PUB_M_BYTE * pub_idx]) + _O1_BYTE, idx_l2, _O2_BYTE);
            idx_l1 += _O1_BYTE;
            idx_l2 += _O2_BYTE;
        }
    }
    idx_l1 = cpk->l1_Q2;
    idx_l2 = cpk->l2_Q2;
    for (unsigned i = 0; i < _V1; i++) {
        for (unsigned j = _V1; j < _V1 + _O1; j++) {
            unsigned pub_idx = PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_idx_of_trimat(i, j, _PUB_N);
            memcpy(&pk->pk[_PUB_M_BYTE * pub_idx], idx_l1, _O1_BYTE);
            memcpy((&pk->pk[_PUB_M_BYTE * pub_idx]) + _O1_BYTE, idx_l2, _O2_BYTE);
            idx_l1 += _O1_BYTE;
            idx_l2 += _O2_BYTE;
        }
    }
    idx_l1 = cpk->l1_Q3;
    idx_l2 = cpk->l2_Q3;
    for (unsigned i = 0; i < _V1; i++) {
        for (unsigned j = _V1 + _O1; j < _PUB_N; j++) {
            unsigned pub_idx = PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_idx_of_trimat(i, j, _PUB_N);
            memcpy(&pk->pk[_PUB_M_BYTE * pub_idx], idx_l1, _O1_BYTE);
            memcpy((&pk->pk[_PUB_M_BYTE * pub_idx]) + _O1_BYTE, idx_l2, _O2_BYTE);
            idx_l1 += _O1_BYTE;
            idx_l2 += _O2_BYTE;
        }
    }
    idx_l1 = cpk->l1_Q5;
    idx_l2 = cpk->l2_Q5;
    for (unsigned i = _V1; i < _V1 + _O1; i++) {
        for (unsigned j = i; j < _V1 + _O1; j++) {
            unsigned pub_idx = PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_idx_of_trimat(i, j, _PUB_N);
            memcpy(&pk->pk[_PUB_M_BYTE * pub_idx], idx_l1, _O1_BYTE);
            memcpy((&pk->pk[_PUB_M_BYTE * pub_idx]) + _O1_BYTE, idx_l2, _O2_BYTE);
            idx_l1 += _O1_BYTE;
            idx_l2 += _O2_BYTE;
        }
    }
    idx_l1 = cpk->l1_Q6;
    idx_l2 = cpk->l2_Q6;
    for (unsigned i = _V1; i < _V1 + _O1; i++) {
        for (unsigned j = _V1 + _O1; j < _PUB_N; j++) {
            unsigned pub_idx = PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_idx_of_trimat(i, j, _PUB_N);
            memcpy(&pk->pk[_PUB_M_BYTE * pub_idx], idx_l1, _O1_BYTE);
            memcpy((&pk->pk[_PUB_M_BYTE * pub_idx]) + _O1_BYTE, idx_l2, _O2_BYTE);
            idx_l1 += _O1_BYTE;
            idx_l2 += _O2_BYTE;
        }
    }
    idx_l1 = cpk->l1_Q9;
    idx_l2 = cpk->l2_Q9;
    for (unsigned i = _V1 + _O1; i < _PUB_N; i++) {
        for (unsigned j = i; j < _PUB_N; j++) {
            unsigned pub_idx = PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_idx_of_trimat(i, j, _PUB_N);
            memcpy(&pk->pk[_PUB_M_BYTE * pub_idx], idx_l1, _O1_BYTE);
            memcpy((&pk->pk[_PUB_M_BYTE * pub_idx]) + _O1_BYTE, idx_l2, _O2_BYTE);
            idx_l1 += _O1_BYTE;
            idx_l2 += _O2_BYTE;
        }
    }
}

static void calculate_F_from_Q_ref(sk_t *Fs, const sk_t *Qs, sk_t *Ts) {
    // Layer 1
    // F_sk.l1_F1s[i] = Q_pk.l1_F1s[i]
    memcpy(Fs->l1_F1, Qs->l1_F1, _O1_BYTE * N_TRIANGLE_TERMS(_V1));

    // F_sk.l1_F2s[i] = ( Q_pk.l1_F1s[i] + Q_pk.l1_F1s[i].transpose() ) * T_sk.t1 + Q_pk.l1_F2s[i]
    memcpy(Fs->l1_F2, Qs->l1_F2, _O1_BYTE * _V1 * _O1);
    batch_2trimat_madd(Fs->l1_F2, Qs->l1_F1, Ts->t1, _V1, _V1_BYTE, _O1, _O1_BYTE);

    /*
          Layer 2
          computations:

          F_sk.l2_F1s[i] = Q_pk.l2_F1s[i]

          Q1_T1 = Q_pk.l2_F1s[i]*T_sk.t1
          F_sk.l2_F2s[i] =              Q1_T1 + Q_pk.l2_F2s[i]     + Q_pk.l2_F1s[i].transpose() * T_sk.t1
          F_sk.l2_F5s[i] = UT( t1_tr* ( Q1_T1 + Q_pk.l2_F2s[i] ) ) + Q_pk.l2_F5s[i]

          Q1_Q1T_T4 =  (Q_pk.l2_F1s[i] + Q_pk.l2_F1s[i].transpose()) * t4
          #Q1_Q1T_T4 =  Q1_Q1T * t4
          Q2_T3 = Q_pk.l2_F2s[i]*T_sk.t3
          F_sk.l2_F3s[i] =           Q1_Q1T_T4 + Q2_T3 + Q_pk.l2_F3s[i]
          F_sk.l2_F6s[i] = t1_tr * ( Q1_Q1T_T4 + Q2_T3 + Q_pk.l2_F3s[i] )
                          +  Q_pk.l2_F2s[i].transpose() * t4
                          + (Q_pk.l2_F5s[i] + Q_pk.l2_F5s[i].transpose())*T_sk.t3   + Q_pk.l2_F6s[i]

      */
    memcpy(Fs->l2_F1, Qs->l2_F1, _O2_BYTE * N_TRIANGLE_TERMS(_V1)); // F_sk.l2_F1s[i] = Q_pk.l2_F1s[i]

    // F_sk.l2_F2s[i] =              Q1_T1 + Q_pk.l2_F2s[i]     + Q_pk.l2_F1s[i].transpose() * T_sk.t1
    // F_sk.l2_F5s[i] = UT( t1_tr* ( Q1_T1 + Q_pk.l2_F2s[i] ) ) + Q_pk.l2_F5s[i]
    memcpy(Fs->l2_F2, Qs->l2_F2, _O2_BYTE * _V1 * _O1);
    batch_trimat_madd(Fs->l2_F2, Qs->l2_F1, Ts->t1, _V1, _V1_BYTE, _O1, _O2_BYTE); // Q1_T1+ Q2

    unsigned char tempQ[_O1 * _O1 * _O2_BYTE + 32];
    memset(tempQ, 0, _O1 * _O1 * _O2_BYTE);
    batch_matTr_madd(tempQ, Ts->t1, _V1, _V1_BYTE, _O1, Fs->l2_F2, _O1, _O2_BYTE); // t1_tr*(Q1_T1+Q2)
    memcpy(Fs->l2_F5, Qs->l2_F5, _O2_BYTE * N_TRIANGLE_TERMS(_O1));                // F5
    PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_UpperTrianglize(Fs->l2_F5, tempQ, _O1, _O2_BYTE);            // UT( ... )

    batch_trimatTr_madd(Fs->l2_F2, Qs->l2_F1, Ts->t1, _V1, _V1_BYTE, _O1, _O2_BYTE); // F2 = Q1_T1 + Q2 + Q1^tr*t1

    // Q1_Q1T_T4 =  (Q_pk.l2_F1s[i] + Q_pk.l2_F1s[i].transpose()) * t4
    // Q2_T3 = Q_pk.l2_F2s[i]*T_sk.t3
    // F_sk.l2_F3s[i] =           Q1_Q1T_T4 + Q2_T3 + Q_pk.l2_F3s[i]
    memcpy(Fs->l2_F3, Qs->l2_F3, _V1 * _O2 * _O2_BYTE);
    batch_2trimat_madd(Fs->l2_F3, Qs->l2_F1, Ts->t4, _V1, _V1_BYTE, _O2, _O2_BYTE);  // Q1_Q1T_T4
    batch_mat_madd(Fs->l2_F3, Qs->l2_F2, _V1, Ts->t3, _O1, _O1_BYTE, _O2, _O2_BYTE); // Q2_T3

    // F_sk.l2_F6s[i] = t1_tr * ( Q1_Q1T_T4 + Q2_T3 + Q_pk.l2_F3s[i] )
    //                +  Q_pk.l2_F2s[i].transpose() * t4
    //                + (Q_pk.l2_F5s[i] + Q_pk.l2_F5s[i].transpose())*T_sk.t3   + Q_pk.l2_F6s[i]
    memcpy(Fs->l2_F6, Qs->l2_F6, _O1 * _O2 * _O2_BYTE);
    batch_matTr_madd(Fs->l2_F6, Ts->t1, _V1, _V1_BYTE, _O1, Fs->l2_F3, _O2, _O2_BYTE); // t1_tr * ( Q1_Q1T_T4 + Q2_T3 + Q_pk.l2_F3s[i] )
    batch_2trimat_madd(Fs->l2_F6, Qs->l2_F5, Ts->t3, _O1, _O1_BYTE, _O2, _O2_BYTE);    // (Q_pk.l2_F5s[i] + Q_pk.l2_F5s[i].transpose())*T_sk.t3
    batch_bmatTr_madd(Fs->l2_F6, Qs->l2_F2, _O1, Ts->t4, _V1, _V1_BYTE, _O2, _O2_BYTE);
}

static void calculate_Q_from_F_cyclic_ref(cpk_t *Qs, const sk_t *Fs, const sk_t *Ts) {
    // Layer 1: Computing Q5, Q3, Q6, Q9

    //  Q_pk.l1_F5s[i] = UT( T1tr* (F1 * T1 + F2) )
    const unsigned char *t2 = Ts->t4;
    sk_t tempQ;
    memcpy(tempQ.l1_F2, Fs->l1_F2, _O1_BYTE * _V1 * _O1);
    batch_trimat_madd(tempQ.l1_F2, Fs->l1_F1, Ts->t1, _V1, _V1_BYTE, _O1, _O1_BYTE); // F1*T1 + F2
    memset(tempQ.l2_F1, 0, sizeof(tempQ.l2_F1));
    memset(tempQ.l2_F2, 0, sizeof(tempQ.l2_F2));
    batch_matTr_madd(tempQ.l2_F1, Ts->t1, _V1, _V1_BYTE, _O1, tempQ.l1_F2, _O1, _O1_BYTE); // T1tr*(F1*T1 + F2)
    memset(Qs->l1_Q5, 0, _O1_BYTE * N_TRIANGLE_TERMS(_O1));
    PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_UpperTrianglize(Qs->l1_Q5, tempQ.l2_F1, _O1, _O1_BYTE); // UT( ... )   // Q5

    /*
          F1_T2     = F1 * t2
          F2_T3     = F2 * t3
          F1_F1T_T2 + F2_T3 = F1_T2 + F2_T3 + F1tr * t2
          Q_pk.l1_F3s[i] =         F1_F1T_T2 + F2_T3
          Q_pk.l1_F6s[i] = T1tr* ( F1_F1T_T2 + F2_T3 ) + F2tr * t2
          Q_pk.l1_F9s[i] = UT( T2tr* ( F1_T2 + F2_T3 ) )
      */
    memset(Qs->l1_Q3, 0, _O1_BYTE * _V1 * _O2);
    memset(Qs->l1_Q6, 0, _O1_BYTE * _O1 * _O2);
    memset(Qs->l1_Q9, 0, _O1_BYTE * N_TRIANGLE_TERMS(_O2));

    batch_trimat_madd(Qs->l1_Q3, Fs->l1_F1, t2, _V1, _V1_BYTE, _O2, _O1_BYTE);       // F1*T2
    batch_mat_madd(Qs->l1_Q3, Fs->l1_F2, _V1, Ts->t3, _O1, _O1_BYTE, _O2, _O1_BYTE); // F1_T2 + F2_T3

    memset(tempQ.l1_F2, 0, _O1_BYTE * _V1 * _O2);                                    // should be F3. assuming: _O1 >= _O2
    batch_matTr_madd(tempQ.l1_F2, t2, _V1, _V1_BYTE, _O2, Qs->l1_Q3, _O2, _O1_BYTE); // T2tr * ( F1_T2 + F2_T3 )
    PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_UpperTrianglize(Qs->l1_Q9, tempQ.l1_F2, _O2, _O1_BYTE);        // Q9

    batch_trimatTr_madd(Qs->l1_Q3, Fs->l1_F1, t2, _V1, _V1_BYTE, _O2, _O1_BYTE); // F1_F1T_T2 + F2_T3  // Q3

    batch_bmatTr_madd(Qs->l1_Q6, Fs->l1_F2, _O1, t2, _V1, _V1_BYTE, _O2, _O1_BYTE);    // F2tr*T2
    batch_matTr_madd(Qs->l1_Q6, Ts->t1, _V1, _V1_BYTE, _O1, Qs->l1_Q3, _O2, _O1_BYTE); // Q6
    /*
          Layer 2
          Computing Q9:

          F1_T2     = F1 * t2
          F2_T3     = F2 * t3
          Q9 = UT( T2tr*( F1*T2 + F2*T3 + F3 )  +  T3tr*( F5*T3 + F6 ) )
      */
    sk_t tempQ2;
    memcpy(tempQ2.l2_F3, Fs->l2_F3, _O2_BYTE * _V1 * _O2);                              /// F3 actually.
    batch_trimat_madd(tempQ2.l2_F3, Fs->l2_F1, t2, _V1, _V1_BYTE, _O2, _O2_BYTE);       // F1*T2 + F3
    batch_mat_madd(tempQ2.l2_F3, Fs->l2_F2, _V1, Ts->t3, _O1, _O1_BYTE, _O2, _O2_BYTE); // F1_T2 + F2_T3 + F3

    memset(tempQ.l2_F3, 0, _O2_BYTE * _V1 * _O2);
    batch_matTr_madd(tempQ.l2_F3, t2, _V1, _V1_BYTE, _O2, tempQ2.l2_F3, _O2, _O2_BYTE); // T2tr * ( ..... )

    memcpy(tempQ.l2_F6, Fs->l2_F6, _O2_BYTE * _O1 * _O2);
    batch_trimat_madd(tempQ.l2_F6, Fs->l2_F5, Ts->t3, _O1, _O1_BYTE, _O2, _O2_BYTE); // F5*T3 + F6

    batch_matTr_madd(tempQ.l2_F3, Ts->t3, _O1, _O1_BYTE, _O2, tempQ.l2_F6, _O2, _O2_BYTE); // T2tr*( ..... ) + T3tr*( ..... )
    memset(Qs->l2_Q9, 0, _O2_BYTE * N_TRIANGLE_TERMS(_O2));
    PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_UpperTrianglize(Qs->l2_Q9, tempQ.l2_F3, _O2, _O2_BYTE); // Q9
}

// Choosing implementations depends on the macros: _BLAS_SSE_ and _BLAS_AVX2_
#define calculate_F_from_Q_impl calculate_F_from_Q_ref
#define calculate_Q_from_F_cyclic_impl calculate_Q_from_F_cyclic_ref

void PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_calculate_F_from_Q(sk_t *Fs, const sk_t *Qs, sk_t *Ts) {
    calculate_F_from_Q_impl(Fs, Qs, Ts);
}

void PQCLEAN_RAINBOWIIICCYCLICCOMPRESSED_CLEAN_calculate_Q_from_F_cyclic(cpk_t *Qs, const sk_t *Fs, const sk_t *Ts) {
    calculate_Q_from_F_cyclic_impl(Qs, Fs, Ts);
}