diff options
Diffstat (limited to 'crypto/src/math/raw/Mod.cs')
-rw-r--r-- | crypto/src/math/raw/Mod.cs | 133 |
1 files changed, 65 insertions, 68 deletions
diff --git a/crypto/src/math/raw/Mod.cs b/crypto/src/math/raw/Mod.cs index a3435ea80..1dcfcb6b3 100644 --- a/crypto/src/math/raw/Mod.cs +++ b/crypto/src/math/raw/Mod.cs @@ -54,7 +54,6 @@ namespace Org.BouncyCastle.Math.Raw int bits = (len32 << 5) - Integers.NumberOfLeadingZeros((int)m[len32 - 1]); int len30 = (bits + 29) / 30; - int m0Inv30x4 = -(int)Inverse32(m[0]) << 2; int[] t = new int[4]; int[] D = new int[len30]; @@ -69,32 +68,33 @@ namespace Org.BouncyCastle.Math.Raw Array.Copy(M, 0, F, 0, len30); int eta = -1; + int m0Inv32 = (int)Inverse32((uint)M[0]); int maxDivsteps = GetMaximumDivsteps(bits); for (int divSteps = 0; divSteps < maxDivsteps; divSteps += 30) { eta = Divsteps30(eta, F[0], G[0], t); - UpdateDE30(len30, D, E, t, m0Inv30x4, M); + UpdateDE30(len30, D, E, t, m0Inv32, M); UpdateFG30(len30, F, G, t); } int signF = F[len30 - 1] >> 31; - Debug.Assert(-1 == signF | 0 == signF); - CNegate30(len30, signF, F); - - int signD = CNegate30(len30, signF, D); - Debug.Assert(-1 == signD | 0 == signD); - - // TODO 'D' should already be in [P, -P), but absent a proof we support [-2P, 2P) - signD = CSub30(len30, ~signD, D, M); - signD = CAdd30(len30, signD, D, M); - signD = CAdd30(len30, signD, D, M); - Debug.Assert(0 == signD); + /* + * D is in the range (-2.M, M). First, conditionally add M if D is negative, to bring it + * into the range (-M, M). Then normalize by conditionally negating (according to signF) + * and/or then adding M, to bring it into the range [0, M). + */ + int signD = D[len30 - 1] >> 31; + signD = CAdd30(len30, signD, D, M); + CNormalize30(len30, signF, D, M); Decode30(bits, D, 0, z, 0); Debug.Assert(0 != Nat.LessThan(len32, z, m)); + signF = CNegate30(len30, signF, F); + Debug.Assert(0 == signF); + return (uint)(EqualTo(len30, F, 1) & EqualToZero(len30, G)); } @@ -107,7 +107,6 @@ namespace Org.BouncyCastle.Math.Raw int bits = (len32 << 5) - Integers.NumberOfLeadingZeros((int)m[len32 - 1]); int len30 = (bits + 29) / 30; - int m0Inv30x4 = -(int)Inverse32(m[0]) << 2; int[] t = new int[4]; int[] D = new int[len30]; @@ -124,6 +123,7 @@ namespace Org.BouncyCastle.Math.Raw int clzG = Integers.NumberOfLeadingZeros(G[len30 - 1] | 1) - (len30 * 30 + 2 - bits); int eta = -1 - clzG; int lenDE = len30, lenFG = len30; + int m0Inv32 = (int)Inverse32((uint)M[0]); int maxDivsteps = GetMaximumDivsteps(bits); int divsteps = 0; @@ -135,7 +135,7 @@ namespace Org.BouncyCastle.Math.Raw divsteps += 30; eta = Divsteps30Var(eta, F[0], G[0], t); - UpdateDE30(lenDE, D, E, t, m0Inv30x4, M); + UpdateDE30(lenDE, D, E, t, m0Inv32, M); UpdateFG30(lenFG, F, G, t); int fn = F[lenFG - 1]; @@ -154,32 +154,30 @@ namespace Org.BouncyCastle.Math.Raw } int signF = F[lenFG - 1] >> 31; - Debug.Assert(-1 == signF | 0 == signF); - - if (0 != signF) - { - Negate30(lenFG, F); - Negate30(lenDE, D); - } - - if (!IsOne(lenFG, F)) - return false; + /* + * D is in the range (-2.M, M). First, conditionally add M if D is negative, to bring it + * into the range (-M, M). Then normalize by conditionally negating (according to signF) + * and/or then adding M, to bring it into the range [0, M). + */ int signD = D[lenDE - 1] >> 31; - Debug.Assert(-1 == signD | 0 == signD); - - // TODO 'D' should already be in [P, -P), but absent a proof we support [-2P, 2P) if (signD < 0) { - signD = Add30(len30, D, M); + signD = Add30(lenDE, D, M); } - else + if (signF < 0) { - signD = Sub30(len30, D, M); + signD = Negate30(lenDE, D); + signF = Negate30(lenFG, F); } + Debug.Assert(0 == signF); + + if (!IsOne(lenFG, F)) + return false; + if (signD < 0) { - signD = Add30(len30, D, M); + signD = Add30(lenDE, D, M); } Debug.Assert(0 == signD); @@ -263,21 +261,22 @@ namespace Org.BouncyCastle.Math.Raw return c; } - private static int CSub30(int len30, int cond, int[] D, int[] M) + private static void CNormalize30(int len30, int condNegate, int[] D, int[] M) { Debug.Assert(len30 > 0); Debug.Assert(D.Length >= len30); Debug.Assert(M.Length >= len30); int c = 0, last = len30 - 1; + int condAdd = (D[last] >> 31) ^ condNegate; for (int i = 0; i < last; ++i) { - c += D[i] - (M[i] & cond); + c += (D[i] ^ condNegate) - condNegate + (M[i] & condAdd); D[i] = c & M30; c >>= 30; } - c += D[last] - (M[last] & cond); - D[last] = c; c >>= 30; - return c; + c += (D[last] ^ condNegate) - condNegate + (M[last] & condAdd); + D[last] = c; + Debug.Assert(c >> 30 == 0); } private static void Decode30(int bits, int[] x, int xOff, uint[] z, int zOff) @@ -349,7 +348,7 @@ namespace Org.BouncyCastle.Math.Raw int f = f0, g = g0, m, w, x, y, z; int i = 30, limit, zeros; - for (;;) + for (; ; ) { // Use a sentinel bit to count zeros only up to i. zeros = Integers.NumberOfTrailingZeros(g | (-1 << i)); @@ -502,46 +501,46 @@ namespace Org.BouncyCastle.Math.Raw return c; } - private static int Sub30(int len30, int[] D, int[] M) - { - Debug.Assert(len30 > 0); - Debug.Assert(D.Length >= len30); - Debug.Assert(M.Length >= len30); - - int c = 0, last = len30 - 1; - for (int i = 0; i < last; ++i) - { - c += D[i] - M[i]; - D[i] = c & M30; c >>= 30; - } - c += D[last] - M[last]; - D[last] = c; c >>= 30; - return c; - } - - private static void UpdateDE30(int len30, int[] D, int[] E, int[] t, int m0Inv30x4, int[] M) + private static void UpdateDE30(int len30, int[] D, int[] E, int[] t, int m0Inv32, int[] M) { Debug.Assert(len30 > 0); Debug.Assert(D.Length >= len30); Debug.Assert(E.Length >= len30); Debug.Assert(M.Length >= len30); - Debug.Assert(m0Inv30x4 * M[0] == -1 << 2); + Debug.Assert(m0Inv32 * M[0] == 1); int u = t[0], v = t[1], q = t[2], r = t[3]; - int di, ei, i, md, me; + int di, ei, i, md, me, mi, sd, se; long cd, ce; + /* + * We accept D (E) in the range (-2.M, M) and conceptually add the modulus to the input + * value if it is initially negative. Instead of adding it explicitly, we add u and/or v (q + * and/or r) to md (me). + */ + sd = D[len30 - 1] >> 31; + se = E[len30 - 1] >> 31; + + md = (u & sd) + (v & se); + me = (q & sd) + (r & se); + + mi = M[0]; di = D[0]; ei = E[0]; cd = (long)u * di + (long)v * ei; ce = (long)q * di + (long)r * ei; - md = (m0Inv30x4 * (int)cd) >> 2; - me = (m0Inv30x4 * (int)ce) >> 2; + /* + * Subtract from md/me an extra term in the range [0, 2^30) such that the low 30 bits of the + * intermediate D/E values will be 0, allowing clean division by 2^30. The final D/E are + * thus in the range (-2.M, M), consistent with the input constraint. + */ + md -= (m0Inv32 * (int)cd + md) & M30; + me -= (m0Inv32 * (int)ce + me) & M30; - cd += (long)M[0] * md; - ce += (long)M[0] * me; + cd += (long)mi * md; + ce += (long)mi * me; Debug.Assert(((int)cd & M30) == 0); Debug.Assert(((int)ce & M30) == 0); @@ -551,14 +550,12 @@ namespace Org.BouncyCastle.Math.Raw for (i = 1; i < len30; ++i) { + mi = M[i]; di = D[i]; ei = E[i]; - cd += (long)u * di + (long)v * ei; - ce += (long)q * di + (long)r * ei; - - cd += (long)M[i] * md; - ce += (long)M[i] * me; + cd += (long)u * di + (long)v * ei + (long)mi * md; + ce += (long)q * di + (long)r * ei + (long)mi * me; D[i - 1] = (int)cd & M30; cd >>= 30; E[i - 1] = (int)ce & M30; ce >>= 30; @@ -606,4 +603,4 @@ namespace Org.BouncyCastle.Math.Raw G[len30 - 1] = (int)cg; } } -} +} \ No newline at end of file |