aboutsummaryrefslogtreecommitdiffstats
path: root/libraries/spongycastle/core/src/main/java/org/spongycastle/pqc/crypto/rainbow/util/ComputeInField.java
diff options
context:
space:
mode:
Diffstat (limited to 'libraries/spongycastle/core/src/main/java/org/spongycastle/pqc/crypto/rainbow/util/ComputeInField.java')
-rw-r--r--libraries/spongycastle/core/src/main/java/org/spongycastle/pqc/crypto/rainbow/util/ComputeInField.java490
1 files changed, 490 insertions, 0 deletions
diff --git a/libraries/spongycastle/core/src/main/java/org/spongycastle/pqc/crypto/rainbow/util/ComputeInField.java b/libraries/spongycastle/core/src/main/java/org/spongycastle/pqc/crypto/rainbow/util/ComputeInField.java
new file mode 100644
index 000000000..2094fb031
--- /dev/null
+++ b/libraries/spongycastle/core/src/main/java/org/spongycastle/pqc/crypto/rainbow/util/ComputeInField.java
@@ -0,0 +1,490 @@
+package org.spongycastle.pqc.crypto.rainbow.util;
+
+/**
+ * This class offers different operations on matrices in field GF2^8.
+ * <p/>
+ * Implemented are functions:
+ * - finding inverse of a matrix
+ * - solving linear equation systems using the Gauss-Elimination method
+ * - basic operations like matrix multiplication, addition and so on.
+ */
+
+public class ComputeInField
+{
+
+ private short[][] A; // used by solveEquation and inverse
+ short[] x;
+
+ /**
+ * Constructor with no parameters
+ */
+ public ComputeInField()
+ {
+ }
+
+
+ /**
+ * This function finds a solution of the equation Bx = b.
+ * Exception is thrown if the linear equation system has no solution
+ *
+ * @param B this matrix is the left part of the
+ * equation (B in the equation above)
+ * @param b the right part of the equation
+ * (b in the equation above)
+ * @return x the solution of the equation if it is solvable
+ * null otherwise
+ * @throws RuntimeException if LES is not solvable
+ */
+ public short[] solveEquation(short[][] B, short[] b)
+ {
+ try
+ {
+
+ if (B.length != b.length)
+ {
+ throw new RuntimeException(
+ "The equation system is not solvable");
+ }
+
+ /** initialize **/
+ // this matrix stores B and b from the equation B*x = b
+ // b is stored as the last column.
+ // B contains one column more than rows.
+ // In this column we store a free coefficient that should be later subtracted from b
+ A = new short[B.length][B.length + 1];
+ // stores the solution of the LES
+ x = new short[B.length];
+
+ /** copy B into the global matrix A **/
+ for (int i = 0; i < B.length; i++)
+ { // rows
+ for (int j = 0; j < B[0].length; j++)
+ { // cols
+ A[i][j] = B[i][j];
+ }
+ }
+
+ /** copy the vector b into the global A **/
+ //the free coefficient, stored in the last column of A( A[i][b.length]
+ // is to be subtracted from b
+ for (int i = 0; i < b.length; i++)
+ {
+ A[i][b.length] = GF2Field.addElem(b[i], A[i][b.length]);
+ }
+
+ /** call the methods for gauss elimination and backward substitution **/
+ computeZerosUnder(false); // obtain zeros under the diagonal
+ substitute();
+
+ return x;
+
+ }
+ catch (RuntimeException rte)
+ {
+ return null; // the LES is not solvable!
+ }
+ }
+
+ /**
+ * This function computes the inverse of a given matrix using the Gauss-
+ * Elimination method.
+ * <p/>
+ * An exception is thrown if the matrix has no inverse
+ *
+ * @param coef the matrix which inverse matrix is needed
+ * @return inverse matrix of the input matrix.
+ * If the matrix is singular, null is returned.
+ * @throws RuntimeException if the given matrix is not invertible
+ */
+ public short[][] inverse(short[][] coef)
+ {
+ try
+ {
+ /** Initialization: **/
+ short factor;
+ short[][] inverse;
+ A = new short[coef.length][2 * coef.length];
+ if (coef.length != coef[0].length)
+ {
+ throw new RuntimeException(
+ "The matrix is not invertible. Please choose another one!");
+ }
+
+ /** prepare: Copy coef and the identity matrix into the global A. **/
+ for (int i = 0; i < coef.length; i++)
+ {
+ for (int j = 0; j < coef.length; j++)
+ {
+ //copy the input matrix coef into A
+ A[i][j] = coef[i][j];
+ }
+ // copy the identity matrix into A.
+ for (int j = coef.length; j < 2 * coef.length; j++)
+ {
+ A[i][j] = 0;
+ }
+ A[i][i + A.length] = 1;
+ }
+
+ /** Elimination operations to get the identity matrix from the left side of A. **/
+ // modify A to get 0s under the diagonal.
+ computeZerosUnder(true);
+
+ // modify A to get only 1s on the diagonal: A[i][j] =A[i][j]/A[i][i].
+ for (int i = 0; i < A.length; i++)
+ {
+ factor = GF2Field.invElem(A[i][i]);
+ for (int j = i; j < 2 * A.length; j++)
+ {
+ A[i][j] = GF2Field.multElem(A[i][j], factor);
+ }
+ }
+
+ //modify A to get only 0s above the diagonal.
+ computeZerosAbove();
+
+ // copy the result (the second half of A) in the matrix inverse.
+ inverse = new short[A.length][A.length];
+ for (int i = 0; i < A.length; i++)
+ {
+ for (int j = A.length; j < 2 * A.length; j++)
+ {
+ inverse[i][j - A.length] = A[i][j];
+ }
+ }
+ return inverse;
+
+ }
+ catch (RuntimeException rte)
+ {
+ // The matrix is not invertible! A new one should be generated!
+ return null;
+ }
+ }
+
+ /**
+ * Elimination under the diagonal.
+ * This function changes a matrix so that it contains only zeros under the
+ * diagonal(Ai,i) using only Gauss-Elimination operations.
+ * <p/>
+ * It is used in solveEquaton as well as in the function for
+ * finding an inverse of a matrix: {@link}inverse. Both of them use the
+ * Gauss-Elimination Method.
+ * <p/>
+ * The result is stored in the global matrix A
+ *
+ * @param usedForInverse This parameter shows if the function is used by the
+ * solveEquation-function or by the inverse-function and according
+ * to this creates matrices of different sizes.
+ * @throws RuntimeException in case a multiplicative inverse of 0 is needed
+ */
+ private void computeZerosUnder(boolean usedForInverse)
+ throws RuntimeException
+ {
+
+ //the number of columns in the global A where the tmp results are stored
+ int length;
+ short tmp = 0;
+
+ //the function is used in inverse() - A should have 2 times more columns than rows
+ if (usedForInverse)
+ {
+ length = 2 * A.length;
+ }
+ //the function is used in solveEquation - A has 1 column more than rows
+ else
+ {
+ length = A.length + 1;
+ }
+
+ //elimination operations to modify A so that that it contains only 0s under the diagonal
+ for (int k = 0; k < A.length - 1; k++)
+ { // the fixed row
+ for (int i = k + 1; i < A.length; i++)
+ { // rows
+ short factor1 = A[i][k];
+ short factor2 = GF2Field.invElem(A[k][k]);
+
+ //The element which multiplicative inverse is needed, is 0
+ //in this case is the input matrix not invertible
+ if (factor2 == 0)
+ {
+ throw new RuntimeException("Matrix not invertible! We have to choose another one!");
+ }
+
+ for (int j = k; j < length; j++)
+ {// columns
+ // tmp=A[k,j] / A[k,k]
+ tmp = GF2Field.multElem(A[k][j], factor2);
+ // tmp = A[i,k] * A[k,j] / A[k,k]
+ tmp = GF2Field.multElem(factor1, tmp);
+ // A[i,j]=A[i,j]-A[i,k]/A[k,k]*A[k,j];
+ A[i][j] = GF2Field.addElem(A[i][j], tmp);
+ }
+ }
+ }
+ }
+
+ /**
+ * Elimination above the diagonal.
+ * This function changes a matrix so that it contains only zeros above the
+ * diagonal(Ai,i) using only Gauss-Elimination operations.
+ * <p/>
+ * It is used in the inverse-function
+ * The result is stored in the global matrix A
+ *
+ * @throws RuntimeException in case a multiplicative inverse of 0 is needed
+ */
+ private void computeZerosAbove()
+ throws RuntimeException
+ {
+ short tmp = 0;
+ for (int k = A.length - 1; k > 0; k--)
+ { // the fixed row
+ for (int i = k - 1; i >= 0; i--)
+ { // rows
+ short factor1 = A[i][k];
+ short factor2 = GF2Field.invElem(A[k][k]);
+ if (factor2 == 0)
+ {
+ throw new RuntimeException("The matrix is not invertible");
+ }
+ for (int j = k; j < 2 * A.length; j++)
+ { // columns
+ // tmp = A[k,j] / A[k,k]
+ tmp = GF2Field.multElem(A[k][j], factor2);
+ // tmp = A[i,k] * A[k,j] / A[k,k]
+ tmp = GF2Field.multElem(factor1, tmp);
+ // A[i,j] = A[i,j] - A[i,k] / A[k,k] * A[k,j];
+ A[i][j] = GF2Field.addElem(A[i][j], tmp);
+ }
+ }
+ }
+ }
+
+
+ /**
+ * This function uses backward substitution to find x
+ * of the linear equation system (LES) B*x = b,
+ * where A a triangle-matrix is (contains only zeros under the diagonal)
+ * and b is a vector
+ * <p/>
+ * If the multiplicative inverse of 0 is needed, an exception is thrown.
+ * In this case is the LES not solvable
+ *
+ * @throws RuntimeException in case a multiplicative inverse of 0 is needed
+ */
+ private void substitute()
+ throws RuntimeException
+ {
+
+ // for the temporary results of the operations in field
+ short tmp, temp;
+
+ temp = GF2Field.invElem(A[A.length - 1][A.length - 1]);
+ if (temp == 0)
+ {
+ throw new RuntimeException("The equation system is not solvable");
+ }
+
+ /** backward substitution **/
+ x[A.length - 1] = GF2Field.multElem(A[A.length - 1][A.length], temp);
+ for (int i = A.length - 2; i >= 0; i--)
+ {
+ tmp = A[i][A.length];
+ for (int j = A.length - 1; j > i; j--)
+ {
+ temp = GF2Field.multElem(A[i][j], x[j]);
+ tmp = GF2Field.addElem(tmp, temp);
+ }
+
+ temp = GF2Field.invElem(A[i][i]);
+ if (temp == 0)
+ {
+ throw new RuntimeException("Not solvable equation system");
+ }
+ x[i] = GF2Field.multElem(tmp, temp);
+ }
+ }
+
+
+ /**
+ * This function multiplies two given matrices.
+ * If the given matrices cannot be multiplied due
+ * to different sizes, an exception is thrown.
+ *
+ * @param M1 -the 1st matrix
+ * @param M2 -the 2nd matrix
+ * @return A = M1*M2
+ * @throws RuntimeException in case the given matrices cannot be multiplied
+ * due to different dimensions.
+ */
+ public short[][] multiplyMatrix(short[][] M1, short[][] M2)
+ throws RuntimeException
+ {
+
+ if (M1[0].length != M2.length)
+ {
+ throw new RuntimeException("Multiplication is not possible!");
+ }
+ short tmp = 0;
+ A = new short[M1.length][M2[0].length];
+ for (int i = 0; i < M1.length; i++)
+ {
+ for (int j = 0; j < M2.length; j++)
+ {
+ for (int k = 0; k < M2[0].length; k++)
+ {
+ tmp = GF2Field.multElem(M1[i][j], M2[j][k]);
+ A[i][k] = GF2Field.addElem(A[i][k], tmp);
+ }
+ }
+ }
+ return A;
+ }
+
+ /**
+ * This function multiplies a given matrix with a one-dimensional array.
+ * <p/>
+ * An exception is thrown, if the number of columns in the matrix and
+ * the number of rows in the one-dim. array differ.
+ *
+ * @param M1 the matrix to be multiplied
+ * @param m the one-dimensional array to be multiplied
+ * @return M1*m
+ * @throws RuntimeException in case of dimension inconsistency
+ */
+ public short[] multiplyMatrix(short[][] M1, short[] m)
+ throws RuntimeException
+ {
+ if (M1[0].length != m.length)
+ {
+ throw new RuntimeException("Multiplication is not possible!");
+ }
+ short tmp = 0;
+ short[] B = new short[M1.length];
+ for (int i = 0; i < M1.length; i++)
+ {
+ for (int j = 0; j < m.length; j++)
+ {
+ tmp = GF2Field.multElem(M1[i][j], m[j]);
+ B[i] = GF2Field.addElem(B[i], tmp);
+ }
+ }
+ return B;
+ }
+
+ /**
+ * Addition of two vectors
+ *
+ * @param vector1 first summand, always of dim n
+ * @param vector2 second summand, always of dim n
+ * @return addition of vector1 and vector2
+ * @throws RuntimeException in case the addition is impossible
+ * due to inconsistency in the dimensions
+ */
+ public short[] addVect(short[] vector1, short[] vector2)
+ {
+ if (vector1.length != vector2.length)
+ {
+ throw new RuntimeException("Multiplication is not possible!");
+ }
+ short rslt[] = new short[vector1.length];
+ for (int n = 0; n < rslt.length; n++)
+ {
+ rslt[n] = GF2Field.addElem(vector1[n], vector2[n]);
+ }
+ return rslt;
+ }
+
+ /**
+ * Multiplication of column vector with row vector
+ *
+ * @param vector1 column vector, always n x 1
+ * @param vector2 row vector, always 1 x n
+ * @return resulting n x n matrix of multiplication
+ * @throws RuntimeException in case the multiplication is impossible due to
+ * inconsistency in the dimensions
+ */
+ public short[][] multVects(short[] vector1, short[] vector2)
+ {
+ if (vector1.length != vector2.length)
+ {
+ throw new RuntimeException("Multiplication is not possible!");
+ }
+ short rslt[][] = new short[vector1.length][vector2.length];
+ for (int i = 0; i < vector1.length; i++)
+ {
+ for (int j = 0; j < vector2.length; j++)
+ {
+ rslt[i][j] = GF2Field.multElem(vector1[i], vector2[j]);
+ }
+ }
+ return rslt;
+ }
+
+ /**
+ * Multiplies vector with scalar
+ *
+ * @param scalar galois element to multiply vector with
+ * @param vector vector to be multiplied
+ * @return vector multiplied with scalar
+ */
+ public short[] multVect(short scalar, short[] vector)
+ {
+ short rslt[] = new short[vector.length];
+ for (int n = 0; n < rslt.length; n++)
+ {
+ rslt[n] = GF2Field.multElem(scalar, vector[n]);
+ }
+ return rslt;
+ }
+
+ /**
+ * Multiplies matrix with scalar
+ *
+ * @param scalar galois element to multiply matrix with
+ * @param matrix 2-dim n x n matrix to be multiplied
+ * @return matrix multiplied with scalar
+ */
+ public short[][] multMatrix(short scalar, short[][] matrix)
+ {
+ short[][] rslt = new short[matrix.length][matrix[0].length];
+ for (int i = 0; i < matrix.length; i++)
+ {
+ for (int j = 0; j < matrix[0].length; j++)
+ {
+ rslt[i][j] = GF2Field.multElem(scalar, matrix[i][j]);
+ }
+ }
+ return rslt;
+ }
+
+ /**
+ * Adds the n x n matrices matrix1 and matrix2
+ *
+ * @param matrix1 first summand
+ * @param matrix2 second summand
+ * @return addition of matrix1 and matrix2; both having the dimensions n x n
+ * @throws RuntimeException in case the addition is not possible because of
+ * different dimensions of the matrices
+ */
+ public short[][] addSquareMatrix(short[][] matrix1, short[][] matrix2)
+ {
+ if (matrix1.length != matrix2.length || matrix1[0].length != matrix2[0].length)
+ {
+ throw new RuntimeException("Addition is not possible!");
+ }
+
+ short[][] rslt = new short[matrix1.length][matrix1.length];//
+ for (int i = 0; i < matrix1.length; i++)
+ {
+ for (int j = 0; j < matrix2.length; j++)
+ {
+ rslt[i][j] = GF2Field.addElem(matrix1[i][j], matrix2[i][j]);
+ }
+ }
+ return rslt;
+ }
+
+}