/* * File: dclxvi-20130329/fp6e.c * Author: Ruben Niederhagen, Peter Schwabe * Public Domain */ #include #include //AVANT //#include "fp6e.h" //APRES #include "mul.h" extern "C" { #include "fpe.h" #include "fp2e.h" #include "fp6e.h" } extern const fp2e_t bn_ypminus1; extern const fp2e_t bn_ypminus1_squ; extern const fpe_t bn_zeta; extern const fpe_t bn_zeta2; void fp6e_short_coeffred(fp6e_t rop) { fp2e_short_coeffred(rop->m_a); fp2e_short_coeffred(rop->m_b); fp2e_short_coeffred(rop->m_c); } // Set fp6e_t rop to given value: void fp6e_set(fp6e_t rop, const fp6e_t op) { fp2e_set(rop->m_a, op->m_a); fp2e_set(rop->m_b, op->m_b); fp2e_set(rop->m_c, op->m_c); } // Initialize an fp6e, set to value given in three fp2es void fp6e_set_fp2e(fp6e_t rop, const fp2e_t a, const fp2e_t b, const fp2e_t c) { fp2e_set(rop->m_a, a); fp2e_set(rop->m_b, b); fp2e_set(rop->m_c, c); } // Set rop to one: void fp6e_setone(fp6e_t rop) { fp2e_setzero(rop->m_a); fp2e_setzero(rop->m_b); fp2e_setone(rop->m_c); } // Set rop to zero: void fp6e_setzero(fp6e_t rop) { fp2e_setzero(rop->m_a); fp2e_setzero(rop->m_b); fp2e_setzero(rop->m_c); } // Compare for equality: int fp6e_iseq(const fp6e_t op1, const fp6e_t op2) { int ret = fp2e_iseq(op1->m_a, op2->m_a); ret = ret && fp2e_iseq(op1->m_b, op2->m_b); ret = ret && fp2e_iseq(op1->m_c, op2->m_c); return ret; } int fp6e_isone(const fp6e_t op) { int ret = fp2e_iszero(op->m_a); ret = ret && fp2e_iszero(op->m_b); ret = ret && fp2e_isone(op->m_c); return ret; } int fp6e_iszero(const fp6e_t op) { int ret = fp2e_iszero(op->m_a); ret = ret && fp2e_iszero(op->m_b); ret = ret && fp2e_iszero(op->m_c); return ret; } void fp6e_cmov(fp6e_t rop, const fp6e_t op, int c) { fp2e_cmov(rop->m_a, op->m_a, c); fp2e_cmov(rop->m_b, op->m_b, c); fp2e_cmov(rop->m_c, op->m_c, c); } // Add two fp6e, store result in rop: void fp6e_add(fp6e_t rop, const fp6e_t op1, const fp6e_t op2) { fp2e_add(rop->m_a, op1->m_a, op2->m_a); fp2e_add(rop->m_b, op1->m_b, op2->m_b); fp2e_add(rop->m_c, op1->m_c, op2->m_c); } // Subtract op2 from op1, store result in rop: void fp6e_sub(fp6e_t rop, const fp6e_t op1, const fp6e_t op2) { fp2e_sub(rop->m_a, op1->m_a, op2->m_a); fp2e_sub(rop->m_b, op1->m_b, op2->m_b); fp2e_sub(rop->m_c, op1->m_c, op2->m_c); } // Subtract op2 from op1, store result in rop: void fp6e_neg(fp6e_t rop, const fp6e_t op) { fp2e_neg(rop->m_a, op->m_a); fp2e_neg(rop->m_b, op->m_b); fp2e_neg(rop->m_c, op->m_c); } // Multiply two fp6e, store result in rop: void fp6e_mul(fp6e_t rop, const fp6e_t op1, const fp6e_t op2) { fp2e_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6; // Needed for intermediary values // See "Multiplication and Squaring in Pairing-Friendly Fields", section 4, Karatsuba method fp2e_mul(tmp3, op1->m_a, op2->m_a); fp2e_mul(tmp2, op1->m_b, op2->m_b); fp2e_mul(tmp1, op1->m_c, op2->m_c); fp2e_add(tmp4, op1->m_a, op1->m_b); //fp2e_short_coeffred(tmp4); fp2e_add(tmp5, op2->m_a, op2->m_b); //fp2e_short_coeffred(tmp5); fp2e_mul(tmp6, tmp4, tmp5); fp2e_sub2(tmp6, tmp2); //fp2e_short_coeffred(tmp6); fp2e_sub2(tmp6, tmp3); //fp2e_short_coeffred(tmp6); fp2e_mulxi(tmp6, tmp6); fp2e_add2(tmp6, tmp1); fp2e_add(tmp4, op1->m_b, op1->m_c); //fp2e_short_coeffred(tmp4); fp2e_add(tmp5, op2->m_b, op2->m_c); //fp2e_short_coeffred(tmp5); fp2e_mul(rop->m_b, tmp4, tmp5); fp2e_sub2(rop->m_b, tmp1); fp2e_sub2(rop->m_b, tmp2); //fp2e_short_coeffred(rop->m_b); fp2e_mulxi(tmp4, tmp3); fp2e_add2(rop->m_b, tmp4); fp2e_short_coeffred(rop->m_b); fp2e_add(tmp4, op1->m_a, op1->m_c); //fp2e_short_coeffred(tmp4); fp2e_add(tmp5, op2->m_a, op2->m_c); //fp2e_short_coeffred(tmp5); fp2e_set(rop->m_c, tmp6); fp2e_short_coeffred(rop->m_c); fp2e_mul(rop->m_a, tmp4, tmp5); fp2e_sub2(rop->m_a, tmp1); fp2e_add2(rop->m_a, tmp2); fp2e_sub2(rop->m_a, tmp3); fp2e_short_coeffred(rop->m_a); } // Compute the double of the Square of an fp6e, store result in rop, uses Chung-Hasan (CH-SQR3x in pairing-friendly fields) void fp6e_squaredouble(fp6e_t rop, const fp6e_t op) { //fp6e_mul(rop, op, op); //XXX make faster! fp2e_t s0, s1, s2, s3, s4, t; fp2e_square(s0,op->m_c); fp2e_add(t, op->m_a, op->m_c); fp2e_add(s1, t, op->m_b); fp2e_short_coeffred(s1); fp2e_square(s1, s1); fp2e_sub(s2, t, op->m_b); fp2e_short_coeffred(s2); fp2e_square(s2, s2); fp2e_mul(s3, op->m_a, op->m_b); fp2e_double(s3, s3); fp2e_square(s4, op->m_a); fp2e_mulxi(rop->m_c, s3); fp2e_add(rop->m_c, rop->m_c, s0); fp2e_double(rop->m_c, rop->m_c); fp2e_short_coeffred(rop->m_c); fp2e_mulxi(rop->m_b, s4); fp2e_sub(rop->m_b, s3, rop->m_b); fp2e_double(rop->m_b, rop->m_b); fp2e_sub(rop->m_b, s1, rop->m_b); fp2e_sub(rop->m_b, rop->m_b, s2); fp2e_short_coeffred(rop->m_b); fp2e_add(rop->m_a, s0, s4); fp2e_double(rop->m_a, rop->m_a); fp2e_sub(rop->m_a, s1, rop->m_a); fp2e_add(rop->m_a, rop->m_a, s2); fp2e_short_coeffred(rop->m_a); } // Multiply with tau: void fp6e_multau(fp6e_t rop, const fp6e_t op) { fp2e_t tmp1; fp2e_set(tmp1, op->m_b); fp2e_set(rop->m_b, op->m_c); fp2e_mulxi(rop->m_c, op->m_a); fp2e_set(rop->m_a, tmp1); } void fp6e_mul_fpe(fp6e_t rop, const fp6e_t op1, const fpe_t op2) { fp2e_mul_fpe(rop->m_a, op1->m_a, op2); fp2e_mul_fpe(rop->m_b, op1->m_b, op2); fp2e_mul_fpe(rop->m_c, op1->m_c, op2); } void fp6e_mul_fp2e(fp6e_t rop, const fp6e_t op1, const fp2e_t op2) { fp2e_mul(rop->m_a, op1->m_a, op2); fp2e_mul(rop->m_b, op1->m_b, op2); fp2e_mul(rop->m_c, op1->m_c, op2); } // Multiply an fp6e by a short fp6e, store result in rop: // the short fp6e op2 has a2 = 0, i.e. op2 = b2*tau + c2. void fp6e_mul_shortfp6e(fp6e_t rop, const fp6e_t op1, const fp6e_t op2) { fp2e_t tmp1, tmp2, tmp3, tmp4, tmp5; // Needed for intermediary values fp2e_mul(tmp2, op1->m_b, op2->m_b); // tmp2 = b1*b2 fp2e_mul(tmp1, op1->m_c, op2->m_c); // tmp1 = c1*c2 fp2e_mul(tmp3, op1->m_a, op2->m_b); // tmp3 = a1*b2 fp2e_mulxi(tmp3, tmp3); // tmp3 = a1*b2*xi fp2e_add(tmp5, tmp3, tmp1); // tmp5 = c1*c2 + a1*b2*xi fp2e_add(tmp4, op1->m_b, op1->m_c); // tmp4 = b1+c1 //fp2e_short_coeffred(tmp4); fp2e_add(tmp3, op2->m_b, op2->m_c); // tmp3 = b2+c2 //fp2e_short_coeffred(tmp3); fp2e_mul(rop->m_b, tmp4, tmp3); // b3 = (b1+c1)*(b2+c2) fp2e_sub2(rop->m_b, tmp1); fp2e_sub2(rop->m_b, tmp2); // b3 = b1*c2 + b2*c1 fp2e_short_coeffred(rop->m_b); fp2e_mul(rop->m_a, op1->m_a, op2->m_c); // a3 = a1*c2 fp2e_add2(rop->m_a, tmp2); // a3 = a1*c2 + b1*b2 fp2e_set(rop->m_c, tmp5); // c3 = c1*c2 + a1*b2*xi } void fp6e_invert(fp6e_t rop, const fp6e_t op) { fp2e_t tmp1, tmp2, tmp3, tmp4, tmp5; // Needed to store intermediary results // See "Implementing cryptographic pairings" fp2e_square(tmp1, op->m_c); fp2e_mul(tmp5, op->m_a, op->m_b); fp2e_mulxi(tmp5, tmp5); fp2e_sub2(tmp1, tmp5); // A fp2e_short_coeffred(tmp1); fp2e_square(tmp2, op->m_a); fp2e_mulxi(tmp2, tmp2); fp2e_mul(tmp5, op->m_b, op->m_c); fp2e_sub2(tmp2, tmp5); // B fp2e_short_coeffred(tmp2); fp2e_square(tmp3, op->m_b); fp2e_mul(tmp5, op->m_a, op->m_c); fp2e_sub2(tmp3, tmp5); // C //fp2e_short_coeffred(tmp3); fp2e_mul(tmp4, tmp3, op->m_b); fp2e_mulxi(tmp4, tmp4); fp2e_mul(tmp5, tmp1, op->m_c); fp2e_add2(tmp4, tmp5); fp2e_mul(tmp5, tmp2, op->m_a); fp2e_mulxi(tmp5, tmp5); fp2e_add2(tmp4, tmp5); // F fp2e_short_coeffred(tmp4); fp2e_invert(tmp4, tmp4); fp2e_mul(rop->m_a, tmp3, tmp4); fp2e_mul(rop->m_b, tmp2, tmp4); fp2e_mul(rop->m_c, tmp1, tmp4); } void fp6e_frobenius_p(fp6e_t rop, const fp6e_t op) { fp6e_set(rop, op); fp2e_conjugate(rop->m_a, rop->m_a); fp2e_conjugate(rop->m_b, rop->m_b); fp2e_conjugate(rop->m_c, rop->m_c); fp2e_mul(rop->m_b, rop->m_b, bn_ypminus1); fp2e_mul(rop->m_a, rop->m_a, bn_ypminus1_squ); } void fp6e_frobenius_p2(fp6e_t rop, const fp6e_t op) { fp2e_set(rop->m_c, op->m_c); fp2e_mul_fpe(rop->m_b, op->m_b, bn_zeta2); fp2e_mul_fpe(rop->m_a, op->m_a, bn_zeta); } // Print the fp6e: void fp6e_print(FILE * outfile, const fp6e_t op) { fprintf(outfile, "["); fp2e_print(outfile, op->m_a); fprintf(outfile, " * Y^2\n + "); fp2e_print(outfile, op->m_b); fprintf(outfile, " * Y\n + "); fp2e_print(outfile, op->m_c); fprintf(outfile, "]"); }