|
| 1 | +/* |
| 2 | + * Licensed to the Apache Software Foundation (ASF) under one |
| 3 | + * or more contributor license agreements. See the NOTICE file |
| 4 | + * distributed with this work for additional information |
| 5 | + * regarding copyright ownership. The ASF licenses this file |
| 6 | + * to you under the Apache License, Version 2.0 (the |
| 7 | + * "License"); you may not use this file except in compliance |
| 8 | + * with the License. You may obtain a copy of the License at |
| 9 | + * |
| 10 | + * http://www.apache.org/licenses/LICENSE-2.0 |
| 11 | + * |
| 12 | + * Unless required by applicable law or agreed to in writing, |
| 13 | + * software distributed under the License is distributed on an |
| 14 | + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
| 15 | + * KIND, either express or implied. See the License for the |
| 16 | + * specific language governing permissions and limitations |
| 17 | + * under the License. |
| 18 | + */ |
| 19 | + |
| 20 | +package org.apache.sysds.runtime.instructions.ooc; |
| 21 | + |
| 22 | +import org.apache.sysds.common.Opcodes; |
| 23 | +import org.apache.sysds.lops.WeightedDivMM.WDivMMType; |
| 24 | +import org.apache.sysds.runtime.controlprogram.context.ExecutionContext; |
| 25 | +import org.apache.sysds.runtime.functionobjects.Multiply; |
| 26 | +import org.apache.sysds.runtime.functionobjects.Plus; |
| 27 | +import org.apache.sysds.runtime.instructions.InstructionUtils; |
| 28 | +import org.apache.sysds.runtime.instructions.cp.CPOperand; |
| 29 | +import org.apache.sysds.runtime.instructions.spark.data.IndexedMatrixValue; |
| 30 | +import org.apache.sysds.runtime.matrix.data.MatrixBlock; |
| 31 | +import org.apache.sysds.runtime.matrix.data.MatrixIndexes; |
| 32 | +import org.apache.sysds.runtime.matrix.operators.AggregateBinaryOperator; |
| 33 | +import org.apache.sysds.runtime.matrix.operators.AggregateOperator; |
| 34 | +import org.apache.sysds.runtime.matrix.operators.BinaryOperator; |
| 35 | +import org.apache.sysds.runtime.matrix.operators.RightScalarOperator; |
| 36 | +import org.apache.sysds.runtime.matrix.operators.QuaternaryOperator; |
| 37 | +import org.apache.sysds.runtime.meta.DataCharacteristics; |
| 38 | + |
| 39 | +import java.util.function.Function; |
| 40 | + |
| 41 | +public class WDivMMOOCInstruction extends QuaternaryOOCInstruction { |
| 42 | + |
| 43 | + protected WDivMMOOCInstruction(QuaternaryOperator op, CPOperand in1, CPOperand in2, CPOperand in3, CPOperand in4, |
| 44 | + CPOperand out, String opcode, String istr) { |
| 45 | + super(op, in1, in2, in3, in4, out, opcode, istr); |
| 46 | + } |
| 47 | + |
| 48 | + public static WDivMMOOCInstruction parseInstruction(QuaternaryOOCInstruction instr) { |
| 49 | + String instrStr = instr.getInstructionString(); |
| 50 | + String opcode = InstructionUtils.getInstructionPartsWithValueType(instr.getInstructionString())[0]; |
| 51 | + return new WDivMMOOCInstruction((QuaternaryOperator) instr.getOperator(), instr.input1, instr.input2, |
| 52 | + instr.input3, instr.input4, instr.output, opcode, instrStr); |
| 53 | + } |
| 54 | + |
| 55 | + @Override |
| 56 | + public void processInstruction(ExecutionContext ec) { |
| 57 | + QuaternaryOperator qop = ((QuaternaryOperator) _optr); |
| 58 | + final WDivMMType wt = qop.wtype3; |
| 59 | + |
| 60 | + CachingStream X = new CachingStream(ec.getMatrixObject(input1).getStreamHandle()); |
| 61 | + CachingStream U = new CachingStream(ec.getMatrixObject(input2).getStreamHandle()); |
| 62 | + CachingStream V = new CachingStream(ec.getMatrixObject(input3).getStreamHandle()); |
| 63 | + |
| 64 | + boolean basic = wt.isBasic(); |
| 65 | + boolean left = wt.isLeft(); |
| 66 | + boolean mult = wt.isMult(); |
| 67 | + boolean minus = wt.isMinus(); |
| 68 | + boolean four = wt.hasFourInputs(); |
| 69 | + boolean scalar = wt.hasScalar(); |
| 70 | + |
| 71 | + OOCStream<IndexedMatrixValue> mmt = matMultOOC(U.getReadStream(), V.getReadStream(), U.getDataCharacteristics(), |
| 72 | + V.getDataCharacteristics(), false, true); |
| 73 | + OOCStream<IndexedMatrixValue> inter; |
| 74 | + OOCStream<IndexedMatrixValue> out; |
| 75 | + |
| 76 | + if(basic) { |
| 77 | + out = elemMultOOC(X.getReadStream(), mmt); |
| 78 | + ec.getMatrixObject(output).setStreamHandle(out); |
| 79 | + return; |
| 80 | + } |
| 81 | + else if(four) { |
| 82 | + if(scalar) { |
| 83 | + double eps = ec.getScalarInput(input4).getDoubleValue(); |
| 84 | + inter = elemDivOOC(X.getReadStream(), elemPlusOOC(mmt, eps)); |
| 85 | + } |
| 86 | + else { |
| 87 | + CachingStream W = new CachingStream(ec.getMatrixObject(input4).getStreamHandle()); |
| 88 | + inter = elemMultOOC(X.getReadStream(), elemMinusOOC(mmt, W.getReadStream())); |
| 89 | + } |
| 90 | + } |
| 91 | + else { |
| 92 | + if(minus) |
| 93 | + inter = maskOOC(X.getReadStream(), elemMinusOOC(mmt, X.getReadStream())); |
| 94 | + else { |
| 95 | + if(mult) |
| 96 | + inter = elemMultOOC(X.getReadStream(), mmt); |
| 97 | + else |
| 98 | + inter = elemDivOOC(X.getReadStream(), mmt); |
| 99 | + } |
| 100 | + } |
| 101 | + |
| 102 | + if(left) |
| 103 | + out = matMultOOC(inter, U.getReadStream(), X.getDataCharacteristics(), U.getDataCharacteristics(), |
| 104 | + true, false); |
| 105 | + else |
| 106 | + out = matMultOOC(inter, V.getReadStream(), X.getDataCharacteristics(), V.getDataCharacteristics(), |
| 107 | + false, false); |
| 108 | + |
| 109 | + ec.getMatrixObject(output).setStreamHandle(out); |
| 110 | + } |
| 111 | + |
| 112 | + private OOCStream<IndexedMatrixValue> matMultOOC(OOCStream<IndexedMatrixValue> m1, OOCStream<IndexedMatrixValue> m2, |
| 113 | + DataCharacteristics dc1, DataCharacteristics dc2, boolean leftTranspose, boolean rightTranspose) { |
| 114 | + |
| 115 | + int emitLeftThreshold = rightTranspose ? (int) dc2.getNumRowBlocks() : (int) dc2.getNumColBlocks(); |
| 116 | + int emitRightThreshold = leftTranspose ? (int) dc1.getNumColBlocks() : (int) dc1.getNumRowBlocks(); |
| 117 | + |
| 118 | + OOCStream<IndexedMatrixValue> intermediateStream = createWritableStream(); |
| 119 | + OOCStream<IndexedMatrixValue> out = createWritableStream(); |
| 120 | + |
| 121 | + AggregateOperator agg = new AggregateOperator(0, Plus.getPlusFnObject()); |
| 122 | + AggregateBinaryOperator op = new AggregateBinaryOperator(Multiply.getMultiplyFnObject(), agg); |
| 123 | + |
| 124 | + joinManyOOC(m1, m2, intermediateStream, (left, right) -> { |
| 125 | + MatrixBlock leftBlock = (MatrixBlock) left.getValue(); |
| 126 | + MatrixBlock rightBlock = (MatrixBlock) right.getValue(); |
| 127 | + if(leftTranspose) |
| 128 | + leftBlock = leftBlock.transpose(); |
| 129 | + if(rightTranspose) |
| 130 | + rightBlock = rightBlock.transpose(); |
| 131 | + |
| 132 | + MatrixBlock partialResult = leftBlock.aggregateBinaryOperations(leftBlock, rightBlock, new MatrixBlock(), op); |
| 133 | + int lidx = (int) (leftTranspose ? left.getIndexes().getColumnIndex() : left.getIndexes().getRowIndex()); |
| 134 | + int ridx = (int) (rightTranspose ? right.getIndexes().getRowIndex() : right.getIndexes().getColumnIndex()); |
| 135 | + return new IndexedMatrixValue(new MatrixIndexes(lidx, ridx), partialResult); |
| 136 | + }, tmp -> leftTranspose ? tmp.getIndexes().getRowIndex() : tmp.getIndexes().getColumnIndex(), |
| 137 | + tmp -> rightTranspose ? tmp.getIndexes().getColumnIndex() : tmp.getIndexes().getRowIndex(), |
| 138 | + emitLeftThreshold, emitRightThreshold); |
| 139 | + |
| 140 | + BinaryOperator plus = InstructionUtils.parseBinaryOperator(Opcodes.PLUS.toString()); |
| 141 | + int emitAggThreshold = leftTranspose ? (int) dc1.getNumRowBlocks() : (int) dc1.getNumColBlocks(); |
| 142 | + |
| 143 | + groupedReduceOOC(intermediateStream, out, (left, right) -> { |
| 144 | + MatrixBlock mb = ((MatrixBlock) left.getValue()).binaryOperationsInPlace(plus, right.getValue()); |
| 145 | + left.setValue(mb); |
| 146 | + return left; |
| 147 | + }, emitAggThreshold); |
| 148 | + |
| 149 | + return out; |
| 150 | + } |
| 151 | + |
| 152 | + private OOCStream<IndexedMatrixValue> elemOOC(OOCStream<IndexedMatrixValue> m1, OOCStream<IndexedMatrixValue> m2, BinaryOperator bop) { |
| 153 | + SubscribableTaskQueue<IndexedMatrixValue> out = new SubscribableTaskQueue<>(); |
| 154 | + Function<IndexedMatrixValue, MatrixIndexes> key = imv -> |
| 155 | + new MatrixIndexes(imv.getIndexes().getRowIndex(), imv.getIndexes().getColumnIndex()); |
| 156 | + |
| 157 | + joinOOC(m1, m2, out, (left, right) -> { |
| 158 | + MatrixBlock lb = (MatrixBlock) left.getValue(); |
| 159 | + MatrixBlock rb = (MatrixBlock) right.getValue(); |
| 160 | + MatrixBlock combined = lb.binaryOperations(bop, rb); |
| 161 | + return new IndexedMatrixValue( |
| 162 | + new MatrixIndexes(left.getIndexes().getRowIndex(), left.getIndexes().getColumnIndex()), combined); |
| 163 | + }, key); |
| 164 | + |
| 165 | + return out; |
| 166 | + } |
| 167 | + |
| 168 | + private OOCStream<IndexedMatrixValue> elemDivOOC(OOCStream<IndexedMatrixValue> m1, OOCStream<IndexedMatrixValue> m2) { |
| 169 | + BinaryOperator div = InstructionUtils.parseBinaryOperator(Opcodes.DIV.toString()); |
| 170 | + return elemOOC(m1, m2, div); |
| 171 | + } |
| 172 | + |
| 173 | + private OOCStream<IndexedMatrixValue> elemMultOOC(OOCStream<IndexedMatrixValue> m1, OOCStream<IndexedMatrixValue> m2) { |
| 174 | + BinaryOperator div = InstructionUtils.parseBinaryOperator(Opcodes.MULT.toString()); |
| 175 | + return elemOOC(m1, m2, div); |
| 176 | + } |
| 177 | + |
| 178 | + private OOCStream<IndexedMatrixValue> elemMinusOOC(OOCStream<IndexedMatrixValue> m1, OOCStream<IndexedMatrixValue> m2) { |
| 179 | + BinaryOperator div = InstructionUtils.parseBinaryOperator(Opcodes.MINUS.toString()); |
| 180 | + return elemOOC(m1, m2, div); |
| 181 | + } |
| 182 | + |
| 183 | + private OOCStream<IndexedMatrixValue> elemPlusOOC(OOCStream<IndexedMatrixValue> m1, double eps) { |
| 184 | + SubscribableTaskQueue<IndexedMatrixValue> out = new SubscribableTaskQueue<>(); |
| 185 | + mapOOC(m1, out, blk -> { |
| 186 | + MatrixBlock res = ((MatrixBlock) blk.getValue()) |
| 187 | + .scalarOperations(new RightScalarOperator(Plus.getPlusFnObject(), eps), null); |
| 188 | + return new IndexedMatrixValue( |
| 189 | + new MatrixIndexes(blk.getIndexes().getRowIndex(), blk.getIndexes().getColumnIndex()), res); |
| 190 | + }); |
| 191 | + return out; |
| 192 | + } |
| 193 | + |
| 194 | + private OOCStream<IndexedMatrixValue> maskOOC(OOCStream<IndexedMatrixValue> mask, OOCStream<IndexedMatrixValue> m1) { |
| 195 | + SubscribableTaskQueue<IndexedMatrixValue> out = new SubscribableTaskQueue<>(); |
| 196 | + Function<IndexedMatrixValue, MatrixIndexes> key = imv -> |
| 197 | + new MatrixIndexes(imv.getIndexes().getRowIndex(), imv.getIndexes().getColumnIndex()); |
| 198 | + |
| 199 | + joinOOC(mask, m1, out, (left, right) -> { |
| 200 | + MatrixBlock lb = (MatrixBlock) left.getValue(); |
| 201 | + MatrixBlock rb = (MatrixBlock) right.getValue(); |
| 202 | + MatrixBlock combined = mask(lb, rb); |
| 203 | + return new IndexedMatrixValue( |
| 204 | + new MatrixIndexes(left.getIndexes().getRowIndex(), left.getIndexes().getColumnIndex()), combined); |
| 205 | + }, key); |
| 206 | + |
| 207 | + return out; |
| 208 | + } |
| 209 | + |
| 210 | + private MatrixBlock mask(MatrixBlock mask, MatrixBlock blk) { |
| 211 | + for(int i = 0; i < blk.getNumRows(); i++) { |
| 212 | + for(int j = 0; j < blk.getNumColumns(); j++) { |
| 213 | + if(mask.get(i,j) ==0) blk.set(i, j, 0); |
| 214 | + } |
| 215 | + } |
| 216 | + return blk; |
| 217 | + } |
| 218 | +} |
0 commit comments