/* * Licensed to the Apache Software Foundation (ASF) under one * or more contributor license agreements. See the NOTICE file * distributed with this work for additional information * regarding copyright ownership. The ASF licenses this file * to you under the Apache License, Version 2.0 (the * "License"); you may not use this file except in compliance * with the License. You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, * software distributed under the License is distributed on an * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY * KIND, either express or implied. See the License for the * specific language governing permissions and limitations * under the License. */ package org.apache.sysml.runtime.matrix.data; /** * SparseBlock implementation that realizes a 'modified compressed sparse row' * representation, where each compressed row is stored as a separate SparseRow * object which provides flexibility for unsorted row appends without the need * for global reshifting of values/indexes but it incurs additional memory * overhead per row for object/array headers per row which also slows down * memory-bound operations due to higher memory bandwidth requirements. * */ public class SparseBlockMCSR extends SparseBlock { private static final long serialVersionUID = -4743624499258436199L; private SparseRow[] _rows = null; /** * Copy constructor sparse block abstraction. * * @param sblock sparse block to copy */ public SparseBlockMCSR(SparseBlock sblock) { //special case SparseBlockMCSR if( sblock instanceof SparseBlockMCSR ) { SparseRow[] orows = ((SparseBlockMCSR)sblock)._rows; _rows = new SparseRow[orows.length]; for( int i=0; i<_rows.length; i++ ) _rows[i] = new SparseRowVector(orows[i]); } //general case SparseBlock else { _rows = new SparseRow[sblock.numRows()]; for( int i=0; i<_rows.length; i++ ) { if( !sblock.isEmpty(i) ) { int apos = sblock.pos(i); int alen = sblock.size(i); _rows[i] = new SparseRowVector(alen); ((SparseRowVector)_rows[i]).setSize(alen); System.arraycopy(sblock.indexes(i), apos, _rows[i].indexes(), 0, alen); System.arraycopy(sblock.values(i), apos, _rows[i].values(), 0, alen); } } } } /** * Copy constructor old sparse row representation. * * @param rows array of sparse rows * @param deep if true, deep copy */ public SparseBlockMCSR(SparseRow[] rows, boolean deep) { if( deep ) { _rows = new SparseRow[rows.length]; for( int i=0; i<_rows.length; i++ ) { _rows[i] = (rows[i].size()==1) ? new SparseRowScalar( rows[i].indexes()[0], rows[i].values()[0]) : new SparseRowVector(rows[i]); } } else { _rows = rows; } } public SparseBlockMCSR(int rlen, int clen) { _rows = new SparseRow[rlen]; } /** * Get the estimated in-memory size of the sparse block in MCSR * with the given dimensions w/o accounting for overallocation. * * @param nrows number of rows * @param ncols number of columns * @param sparsity sparsity ratio * @return memory estimate */ public static long estimateMemory(long nrows, long ncols, double sparsity) { double cnnz = Math.max(SparseRowVector.initialCapacity, Math.ceil(sparsity*ncols)); double rlen = Math.min(nrows, Math.ceil(sparsity*nrows*ncols)); //Each sparse row has a fixed overhead of 8B (reference) + 32B (object) + //12B (3 int members), 32B (overhead int array), 32B (overhead double array), //Each non-zero value requires 12B for the column-index/value pair. //Overheads for arrays, objects, and references refer to 64bit JVMs //If nnz < than rows we have only also empty rows. double size = 16; //object size += rlen * (116 + cnnz * 12); //sparse rows size += 32 + nrows * 8d; //references // robustness for long overflows return (long) Math.min(size, Long.MAX_VALUE); } /////////////////// //SparseBlock implementation @Override public void allocate(int r) { if( _rows[r] == null ) _rows[r] = new SparseRowVector(); } @Override public void allocate(int r, int nnz) { if( _rows[r] == null ) { _rows[r] = (nnz == 1) ? new SparseRowScalar() : new SparseRowVector(nnz); } } @Override public void allocate(int r, int ennz, int maxnnz) { if( _rows[r] == null ) { _rows[r] = (ennz == 1) ? new SparseRowScalar() : new SparseRowVector(ennz, maxnnz); } } @Override public int numRows() { return _rows.length; } @Override public boolean isThreadSafe() { return true; } @Override public boolean isContiguous() { return false; } @Override public void reset() { for( SparseRow row : _rows ) if( row != null ) row.reset(row.size(), Integer.MAX_VALUE); } @Override public void reset(int ennz, int maxnnz) { for( SparseRow row : _rows ) if( row != null ) row.reset(ennz, maxnnz); } @Override public void reset(int r, int ennz, int maxnnz) { if( _rows[r] != null ) _rows[r].reset(ennz, maxnnz); } @Override public long size() { //recompute non-zeros to avoid redundant maintenance long nnz = 0; for( SparseRow row : _rows ) if( row != null ) nnz += row.size(); return nnz; } @Override public int size(int r) { //prior check with isEmpty(r) expected return (_rows[r]!=null) ? _rows[r].size() : 0; } @Override public long size(int rl, int ru) { int ret = 0; for( int i=rl; i<ru; i++ ) ret += (_rows[i]!=null) ? _rows[i].size() : 0; return ret; } @Override public long size(int rl, int ru, int cl, int cu) { long nnz = 0; for(int i=rl; i<ru; i++) if( !isEmpty(i) ) { int start = posFIndexGTE(i, cl); int end = posFIndexGTE(i, cu); nnz += (start!=-1) ? (end-start) : 0; } return nnz; } @Override public boolean isEmpty(int r) { return (_rows[r]==null || _rows[r].isEmpty()); } @Override public int[] indexes(int r) { //prior check with isEmpty(r) expected return _rows[r].indexes(); } @Override public double[] values(int r) { //prior check with isEmpty(r) expected return _rows[r].values(); } @Override public int pos(int r) { //arrays per row (always start 0) return 0; } @Override public boolean set(int r, int c, double v) { if( _rows[r] == null ) _rows[r] = new SparseRowScalar(); else if( _rows[r] instanceof SparseRowScalar && !_rows[r].isEmpty()) _rows[r] = new SparseRowVector(_rows[r]); return _rows[r].set(c, v); } @Override public void set(int r, SparseRow row, boolean deep) { //copy values into existing row to avoid allocation if( _rows[r] != null && _rows[r] instanceof SparseRowVector && ((SparseRowVector)_rows[r]).capacity() >= row.size() && deep ) ((SparseRowVector)_rows[r]).copy(row); //set new sparse row (incl allocation if required) else _rows[r] = (deep && row != null) ? new SparseRowVector(row) : row; } @Override public void append(int r, int c, double v) { if( _rows[r] == null ) _rows[r] = new SparseRowScalar(); else if( _rows[r] instanceof SparseRowScalar && !_rows[r].isEmpty() ) _rows[r] = new SparseRowVector(_rows[r]); _rows[r].append(c, v); } @Override public void setIndexRange(int r, int cl, int cu, double[] v, int vix, int len) { if( _rows[r] == null ) _rows[r] = new SparseRowVector(); else if( _rows[r] instanceof SparseRowScalar ) _rows[r] = new SparseRowVector(_rows[r]); //different sparse row semantics: upper bound inclusive ((SparseRowVector)_rows[r]).setIndexRange(cl, cu-1, v, vix, len); } @Override public void deleteIndexRange(int r, int cl, int cu) { //prior check with isEmpty(r) expected //different sparse row semantics: upper bound inclusive if( _rows[r] instanceof SparseRowScalar ) _rows[r] = new SparseRowVector(_rows[r]); ((SparseRowVector)_rows[r]).deleteIndexRange(cl, cu-1); } @Override public void sort() { for( SparseRow row : _rows ) if( row != null && !row.isEmpty() ) row.sort(); } @Override public void sort(int r) { //prior check with isEmpty(r) expected _rows[r].sort(); } @Override public double get(int r, int c) { if( _rows[r] == null ) return 0; return _rows[r].get(c); } @Override public SparseRow get(int r) { return _rows[r]; } @Override public int posFIndexLTE(int r, int c) { //prior check with isEmpty(r) expected if( _rows[r] instanceof SparseRowScalar ) _rows[r] = new SparseRowVector(_rows[r]); return ((SparseRowVector)_rows[r]).searchIndexesFirstLTE(c); } @Override public int posFIndexGTE(int r, int c) { //prior check with isEmpty(r) expected if( _rows[r] instanceof SparseRowScalar ) _rows[r] = new SparseRowVector(_rows[r]); return ((SparseRowVector)_rows[r]).searchIndexesFirstGTE(c); } @Override public int posFIndexGT(int r, int c) { //prior check with isEmpty(r) expected if( _rows[r] instanceof SparseRowScalar ) _rows[r] = new SparseRowVector(_rows[r]); return ((SparseRowVector)_rows[r]).searchIndexesFirstGT(c); } @Override public String toString() { StringBuilder sb = new StringBuilder(); sb.append("SparseBlockMCSR: rlen="); sb.append(numRows()); sb.append(", nnz="); sb.append(size()); sb.append("\n"); for( int i=0; i<numRows(); i++ ) { sb.append("row +"); sb.append(i); sb.append(": "); sb.append(_rows[i]); sb.append("\n"); } return sb.toString(); } /** * Helper function for MCSR -> {COO, CSR} * @return the underlying array of {@link SparseRow} */ public SparseRow[] getRows() { return _rows; } }