omnizip 0.4.7-arm-linux

data/lib/omnizip/parity/reed_solomon_decoder.rb

@@ -0,0 +1,263 @@
+# frozen_string_literal: true
+
+module Omnizip
+  module Parity
+    # Pure Reed-Solomon decoder for recovering missing blocks
+    # This is algorithm-only code with no I/O dependencies
+    #
+    # Recovers missing blocks by solving a system of linear equations
+    # using Gaussian elimination over GF(2^16)
+    class ReedSolomonDecoder
+      # Recover missing input blocks using available inputs and recovery blocks
+      #
+      # @param present_blocks [Hash<Integer, String>] Map of index => block data for present inputs
+      # @param recovery_blocks [Array<Hash>] Array of recovery block info: {data: String, exponent: Integer}
+      # @param missing_indices [Array<Integer>] Indices of missing input blocks to recover
+      # @param block_size [Integer] Size of each block in bytes
+      # @param total_inputs [Integer] Total number of input blocks (present + missing)
+      # @return [Hash<Integer, String>] Map of recovered index => block data
+      def self.decode(present_blocks, recovery_blocks, missing_indices,
+block_size, total_inputs)
+        raise ArgumentError, "Block size must be even" if block_size.odd?
+
+        if missing_indices.empty?
+          raise ArgumentError,
+                "No missing blocks to recover"
+        end
+        if recovery_blocks.size < missing_indices.size
+          raise ArgumentError,
+                "Not enough recovery blocks"
+        end
+
+        # Select base values using par2cmdline algorithm (same as encoder)
+        bases = Par2cmdlineAlgorithm.compute_bases(total_inputs)
+
+        # Build and solve the matrix system
+        solved_matrix = build_and_solve_matrix(
+          present_blocks.keys.sort,
+          missing_indices.sort,
+          recovery_blocks.map { |r| r[:exponent] },
+          bases,
+        )
+
+        # Reconstruct missing blocks using solved matrix
+        reconstruct_missing_blocks(
+          present_blocks,
+          recovery_blocks,
+          missing_indices,
+          solved_matrix,
+          block_size,
+        )
+      end
+
+      # Build augmented matrix and solve using Gaussian elimination
+      #
+      # @param present_indices [Array<Integer>] Sorted indices of present blocks
+      # @param missing_indices [Array<Integer>] Sorted indices of missing blocks
+      # @param recovery_exponents [Array<Integer>] Exponents of recovery blocks
+      # @param bases [Array<Integer>] Base values for all inputs
+      # @return [Array<Array<Integer>>] Solved left matrix
+      def self.build_and_solve_matrix(present_indices, missing_indices,
+recovery_exponents, bases)
+        num_present = present_indices.size
+        num_missing = missing_indices.size
+        recovery_exponents.size
+
+        # We use first num_missing recovery blocks to solve for missing data
+        num_rows = num_missing
+        num_cols = num_present + num_missing
+
+        # Allocate matrices
+        left_matrix = Array.new(num_rows) { Array.new(num_cols, 0) }
+        right_matrix = Array.new(num_rows) { Array.new(num_rows, 0) }
+
+        # Fill matrices using first num_missing recovery blocks
+        recovery_exponents.first(num_missing).each_with_index do |exponent, row|
+          # Left matrix: base^exponent for present inputs, identity for missing
+          present_indices.each_with_index do |idx, col|
+            left_matrix[row][col] = Galois16.power(bases[idx], exponent)
+          end
+
+          # Identity block for missing data (helps in solving)
+          num_missing.times do |col|
+            left_matrix[row][num_present + col] = (row == col ? 1 : 0)
+          end
+
+          # Right matrix: base^exponent for missing inputs
+          missing_indices.each_with_index do |idx, col|
+            right_matrix[row][col] = Galois16.power(bases[idx], exponent)
+          end
+        end
+
+        # Solve using Gaussian elimination
+        gaussian_elimination(left_matrix, right_matrix)
+
+        left_matrix
+      end
+
+      # Perform Gaussian elimination to solve the matrix system
+      # Modifies matrices in place
+      #
+      # @param left_matrix [Array<Array<Integer>>] Left matrix (will be modified)
+      # @param right_matrix [Array<Array<Integer>>] Right matrix (will be modified)
+      def self.gaussian_elimination(left_matrix, right_matrix)
+        num_rows = left_matrix.size
+        num_left_cols = left_matrix[0].size
+        num_right_cols = right_matrix[0].size
+
+        # Forward elimination + back substitution
+        num_rows.times do |pivot_row|
+          # Get pivot value from right matrix diagonal
+          pivot = right_matrix[pivot_row][pivot_row]
+          raise Omnizip::Error, "Singular matrix at row #{pivot_row}" if pivot.zero?
+
+          # Scale pivot row to make pivot = 1
+          unless pivot == 1
+            # Scale left matrix
+            num_left_cols.times do |col|
+              next if left_matrix[pivot_row][col].zero?
+
+              left_matrix[pivot_row][col] =
+                Galois16.divide(left_matrix[pivot_row][col], pivot)
+            end
+
+            # Scale right matrix
+            right_matrix[pivot_row][pivot_row] = 1
+            ((pivot_row + 1)...num_right_cols).each do |col|
+              next if right_matrix[pivot_row][col].zero?
+
+              right_matrix[pivot_row][col] =
+                Galois16.divide(right_matrix[pivot_row][col], pivot)
+            end
+          end
+
+          # Eliminate column in all other rows
+          num_rows.times do |row|
+            next if row == pivot_row
+
+            scale = right_matrix[row][pivot_row]
+            next if scale.zero?
+
+            if scale == 1
+              # Optimization: just XOR subtract rows
+              num_left_cols.times do |col|
+                next if left_matrix[pivot_row][col].zero?
+
+                left_matrix[row][col] = Galois16.add(
+                  left_matrix[row][col],
+                  left_matrix[pivot_row][col],
+                )
+              end
+
+              (pivot_row...num_right_cols).each do |col|
+                next if right_matrix[pivot_row][col].zero?
+
+                right_matrix[row][col] = Galois16.add(
+                  right_matrix[row][col],
+                  right_matrix[pivot_row][col],
+                )
+              end
+            else
+              # General case: row -= pivot_row * scale
+              num_left_cols.times do |col|
+                next if left_matrix[pivot_row][col].zero?
+
+                scaled = Galois16.multiply(left_matrix[pivot_row][col], scale)
+                left_matrix[row][col] =
+                  Galois16.add(left_matrix[row][col], scaled)
+              end
+
+              (pivot_row...num_right_cols).each do |col|
+                next if right_matrix[pivot_row][col].zero?
+
+                scaled = Galois16.multiply(right_matrix[pivot_row][col], scale)
+                right_matrix[row][col] =
+                  Galois16.add(right_matrix[row][col], scaled)
+              end
+            end
+          end
+        end
+      end
+
+      # Reconstruct missing blocks using solved matrix
+      #
+      # @param present_blocks [Hash<Integer, String>] Present input blocks
+      # @param recovery_blocks [Array<Hash>] Recovery blocks with :data and :exponent
+      # @param missing_indices [Array<Integer>] Indices to recover
+      # @param solved_matrix [Array<Array<Integer>>] Solved coefficient matrix
+      # @param block_size [Integer] Block size in bytes
+      # @return [Hash<Integer, String>] Recovered blocks
+      def self.reconstruct_missing_blocks(present_blocks, recovery_blocks,
+missing_indices, solved_matrix, block_size)
+        recovered = {}
+
+        missing_indices.each_with_index do |missing_idx, row|
+          # Initialize missing block to zeros
+          recovered_block = "\x00".b * block_size
+
+          # Add contribution from each present input block
+          present_blocks.each do |present_idx, present_data|
+            # Find column for this present block
+            col = present_blocks.keys.sort.index(present_idx)
+            factor = solved_matrix[row][col]
+
+            next if factor.zero?
+
+            # recovered_block ^= present_data * factor
+            process_block(factor, present_data, recovered_block, block_size)
+          end
+
+          # Add contribution from each recovery block we're using
+          num_present = present_blocks.size
+          recovery_blocks.first(missing_indices.size).each_with_index do |recovery_info, rec_idx|
+            col = num_present + rec_idx
+            factor = solved_matrix[row][col]
+
+            next if factor.zero?
+
+            # recovered_block ^= recovery_data * factor
+            process_block(factor, recovery_info[:data], recovered_block,
+                          block_size)
+          end
+
+          recovered[missing_idx] = recovered_block
+        end
+
+        recovered
+      end
+
+      # Core operation: output_block ^= input_block * factor (in GF(2^16))
+      # Same as encoder's process_block
+      #
+      # @param factor [Integer] Galois field multiplier
+      # @param input_block [String] Input data
+      # @param output_block [String] Output data (modified in place)
+      # @param block_size [Integer] Block size in bytes
+      def self.process_block(factor, input_block, output_block, block_size)
+        num_words = block_size / 2
+
+        num_words.times do |i|
+          offset = i * 2
+
+          # Read 16-bit words (little-endian)
+          input_word = input_block.getbyte(offset) |
+            (input_block.getbyte(offset + 1) << 8)
+
+          output_word = output_block.getbyte(offset) |
+            (output_block.getbyte(offset + 1) << 8)
+
+          # Galois multiplication and addition (XOR)
+          result = Galois16.add(output_word,
+                                Galois16.multiply(input_word, factor))
+
+          # Write back as bytes (little-endian)
+          output_block.setbyte(offset, result & 0xFF)
+          output_block.setbyte(offset + 1, (result >> 8) & 0xFF)
+        end
+      end
+
+      private_class_method :build_and_solve_matrix, :gaussian_elimination,
+                           :reconstruct_missing_blocks, :process_block
+    end
+  end
+end

data/lib/omnizip/parity/reed_solomon_encoder.rb

@@ -0,0 +1,108 @@
+# frozen_string_literal: true
+
+module Omnizip
+  module Parity
+    # Pure Reed-Solomon encoder for creating recovery blocks
+    # This is algorithm-only code with no I/O dependencies
+    #
+    # Creates recovery blocks using Vandermonde matrix over GF(2^16):
+    #   Recovery[i] = sum(Input[j] * Base[j]^Exponent[i] for all j)
+    class ReedSolomonEncoder
+      # Create recovery blocks from input blocks
+      #
+      # @param input_blocks [Array<String>] Array of input block data (binary strings)
+      # @param block_size [Integer] Size of each block in bytes (must be even for 16-bit processing)
+      # @param exponents [Array<Integer>] Exponent for each recovery block (0-65535)
+      # @return [Array<String>] Array of recovery block data (binary strings)
+      def self.encode(input_blocks, block_size, exponents)
+        if block_size.odd?
+          raise ArgumentError,
+                "Block size must be even for 16-bit processing"
+        end
+        raise ArgumentError, "No input blocks provided" if input_blocks.empty?
+        raise ArgumentError, "No exponents provided" if exponents.empty?
+
+        # Validate all input blocks have correct size
+        input_blocks.each_with_index do |block, idx|
+          unless block.bytesize == block_size
+            raise ArgumentError,
+                  "Input block #{idx} has size #{block.bytesize}, expected #{block_size}"
+          end
+        end
+
+        # Select base values using par2cmdline algorithm
+        num_inputs = input_blocks.length
+        bases = Par2cmdlineAlgorithm.compute_bases(num_inputs)
+
+        # Create recovery blocks
+        recovery_blocks = []
+        exponents.each do |exponent|
+          recovery_block = create_recovery_block(input_blocks, bases, exponent,
+                                                 block_size)
+          recovery_blocks << recovery_block
+        end
+
+        recovery_blocks
+      end
+
+      # Create a single recovery block
+      #
+      # @param input_blocks [Array<String>] Input block data
+      # @param bases [Array<Integer>] Base values for each input
+      # @param exponent [Integer] Exponent for this recovery block
+      # @param block_size [Integer] Block size in bytes
+      # @return [String] Recovery block data
+      def self.create_recovery_block(input_blocks, bases, exponent, block_size)
+        # Initialize recovery block to zeros
+        recovery_data = "\x00".b * block_size
+
+        # Process each input block
+        input_blocks.each_with_index do |input_block, idx|
+          # Compute factor = base[idx]^exponent
+          factor = Galois16.power(bases[idx], exponent)
+
+          # Skip if factor is zero (optimization)
+          next if factor.zero?
+
+          # Add input_block * factor to recovery_block
+          process_block(factor, input_block, recovery_data, block_size)
+        end
+
+        recovery_data
+      end
+
+      # Core operation: output_block ^= input_block * factor (in GF(2^16))
+      # Processes data as 16-bit words in little-endian format
+      #
+      # @param factor [Integer] Galois field multiplier
+      # @param input_block [String] Input data
+      # @param output_block [String] Output data (modified in place)
+      # @param block_size [Integer] Block size in bytes
+      def self.process_block(factor, input_block, output_block, block_size)
+        # Process as 16-bit words (little-endian)
+        num_words = block_size / 2
+
+        num_words.times do |i|
+          offset = i * 2
+
+          # Read 16-bit words (little-endian)
+          input_word = input_block.getbyte(offset) |
+            (input_block.getbyte(offset + 1) << 8)
+
+          output_word = output_block.getbyte(offset) |
+            (output_block.getbyte(offset + 1) << 8)
+
+          # Galois multiplication and addition (XOR)
+          result = Galois16.add(output_word,
+                                Galois16.multiply(input_word, factor))
+
+          # Write back as bytes (little-endian)
+          output_block.setbyte(offset, result & 0xFF)
+          output_block.setbyte(offset + 1, (result >> 8) & 0xFF)
+        end
+      end
+
+      private_class_method :create_recovery_block, :process_block
+    end
+  end
+end