deeplearning4j/deeplearning4j

/*
 *  ******************************************************************************
 *  *
 *  *
 *  * This program and the accompanying materials are made available under the
 *  * terms of the Apache License, Version 2.0 which is available at
 *  * https://www.apache.org/licenses/LICENSE-2.0.
 *  *
 *  *  See the NOTICE file distributed with this work for additional
 *  *  information regarding copyright ownership.
 *  * 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.
 *  *
 *  * SPDX-License-Identifier: Apache-2.0
 *  *****************************************************************************
 */

package org.deeplearning4j.nn.conf.memory;

import lombok.EqualsAndHashCode;
import lombok.Getter;
import lombok.NonNull;
import org.deeplearning4j.nn.conf.CacheMode;
import org.deeplearning4j.nn.conf.inputs.InputType;
import org.nd4j.linalg.api.buffer.DataBuffer;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.shade.jackson.annotation.JsonProperty;

import java.text.DecimalFormat;
import java.util.Arrays;
import java.util.LinkedHashMap;
import java.util.Map;

@Getter
@EqualsAndHashCode(callSuper = true)
public class NetworkMemoryReport extends MemoryReport {

    private static final DecimalFormat BYTES_FORMAT = new DecimalFormat("#,###");

    private final Map<String, MemoryReport> layerAndVertexReports;
    private final Class<?> modelClass;
    private final String modelName;
    private final InputType[] networkInputTypes;

    public NetworkMemoryReport(
                    @NonNull @JsonProperty("layerAndVertexReports") Map<String, MemoryReport> layerAndVertexReports,
                    @NonNull @JsonProperty("modelClass") Class<?> modelClass,
                    @JsonProperty("modelName") String modelName,
                    @NonNull @JsonProperty("networkInputTypes") InputType... networkInputTypes) {
        this.layerAndVertexReports = layerAndVertexReports;
        this.modelClass = modelClass;
        this.modelName = modelName;
        this.networkInputTypes = networkInputTypes;
    }


    @Override
    public Class<?> getReportClass() {
        return modelClass;
    }

    @Override
    public String getName() {
        return modelName;
    }

    @Override
    public long getTotalMemoryBytes(int minibatchSize, @NonNull MemoryUseMode memoryUseMode,
                    @NonNull CacheMode cacheMode, @NonNull DataType dataType) {

        //As per MemoryReport javadoc: we need
        // sum_layers (StdFixed + minibatch * StdVariable) + sum_layers (CacheFixed + minibatch * CacheVariable)
        // + max_layers ( WorkingMemoryFixed + minibatch * WorkingMemoryVariable)

        long totalBytes = 0;
        long maxWorking = 0;
        long maxWorkingFixed = 0;
        long maxWorkingVariable = 0;
        for (MemoryReport lmr : layerAndVertexReports.values()) {

            for (MemoryType mt : MemoryType.values()) {
                if (mt == MemoryType.WORKING_MEMORY_FIXED || mt == MemoryType.WORKING_MEMORY_VARIABLE) {
                    continue;
                }
                totalBytes += lmr.getMemoryBytes(mt, minibatchSize, memoryUseMode, cacheMode, dataType);
            }

            long workFixed = lmr.getMemoryBytes(MemoryType.WORKING_MEMORY_FIXED, minibatchSize, memoryUseMode,
                            cacheMode, dataType);
            long workVar = lmr.getMemoryBytes(MemoryType.WORKING_MEMORY_VARIABLE, minibatchSize, memoryUseMode,
                            cacheMode, dataType);
            long currWorking = workFixed + workVar;

            if (currWorking > maxWorking) {
                maxWorking = currWorking;
                maxWorkingFixed = workFixed;
                maxWorkingVariable = workVar;
            }
        }

        return totalBytes + maxWorkingFixed + maxWorkingVariable;
    }

    @Override
    public long getMemoryBytes(MemoryType memoryType, int minibatchSize, MemoryUseMode memoryUseMode,
                    CacheMode cacheMode, DataType dataType) {
        long totalBytes = 0;
        for (MemoryReport lmr : layerAndVertexReports.values()) {

            long bytes = lmr.getMemoryBytes(memoryType, minibatchSize, memoryUseMode, cacheMode, dataType);

            if (memoryType == MemoryType.WORKING_MEMORY_FIXED || memoryType == MemoryType.WORKING_MEMORY_VARIABLE) {
                totalBytes = Math.max(totalBytes, bytes);
            } else {
                totalBytes += bytes;
            }
        }

        return totalBytes;
    }

    @Override
    public String toString() {

        long fixedMemBytes = getTotalMemoryBytes(0, MemoryUseMode.INFERENCE, CacheMode.NONE, DataType.FLOAT);
        long perEx = getTotalMemoryBytes(1, MemoryUseMode.INFERENCE, CacheMode.NONE, DataType.FLOAT)
                        - fixedMemBytes;

        long fixedMemBytesTrain = getTotalMemoryBytes(0, MemoryUseMode.TRAINING, CacheMode.NONE, DataType.FLOAT);
        long perExTrain = getTotalMemoryBytes(1, MemoryUseMode.TRAINING, CacheMode.NONE, DataType.FLOAT)
                        - fixedMemBytesTrain;

        Map<Class<?>, Integer> layerCounts = new LinkedHashMap<>();
        for (MemoryReport mr : layerAndVertexReports.values()) {
            if (layerCounts.containsKey(mr.getReportClass())) {
                layerCounts.put(mr.getReportClass(), layerCounts.get(mr.getReportClass()) + 1);
            } else {
                layerCounts.put(mr.getReportClass(), 1);
            }
        }

        StringBuilder sbLayerCounts = new StringBuilder();
        for (Map.Entry<Class<?>, Integer> e : layerCounts.entrySet()) {
            sbLayerCounts.append(e.getValue()).append(" x ").append(e.getKey().getSimpleName()).append(", ");
        }

        StringBuilder sb = new StringBuilder();
        sb.append("----- Network Memory Report -----\n").append("  Model Class:                        ")
                        .append(modelClass.getName()).append("\n").append("  Model Name:                         ")
                        .append(modelName).append("\n").append("  Network Input:                      ")
                        .append(Arrays.toString(networkInputTypes)).append("\n")
                        .append("  # Layers:                           ").append(layerAndVertexReports.size())
                        .append("\n").append("  Layer Types:                        ").append(sbLayerCounts)
                        .append("\n");

        appendFixedPlusVariable(sb, "  Inference Memory (FP32)             ", fixedMemBytes, perEx);
        appendFixedPlusVariable(sb, "  Training Memory (FP32):             ", fixedMemBytesTrain, perExTrain);

        sb.append("  Inference Memory Breakdown (FP32):\n");
        appendBreakDown(sb, MemoryUseMode.INFERENCE, CacheMode.NONE, DataType.FLOAT);

        sb.append("  Training Memory Breakdown (CacheMode = ").append(CacheMode.NONE).append(", FP32):\n");
        appendBreakDown(sb, MemoryUseMode.TRAINING, CacheMode.NONE, DataType.FLOAT);


        return sb.toString();
    }

    private void appendBreakDown(StringBuilder sb, MemoryUseMode useMode, CacheMode cacheMode,
                    DataType dataType) {
        for (MemoryType mt : MemoryType.values()) {
            if (useMode == MemoryUseMode.INFERENCE && !mt.isInference()) {
                continue;
            }

            long bytesFixed = getMemoryBytes(mt, 0, useMode, cacheMode, dataType);
            long bytesPerEx = getMemoryBytes(mt, 1, useMode, cacheMode, dataType) - bytesFixed;

            if (bytesFixed > 0 || bytesPerEx > 0) {
                String formatted = String.format("  - %-34s", mt);
                appendFixedPlusVariable(sb, formatted, bytesFixed, bytesPerEx);
            }
        }
    }

    private void appendFixedPlusVariable(StringBuilder sb, String title, long bytesFixed, long bytesPerEx) {
        sb.append(title);
        if (bytesFixed > 0) {
            sb.append(formatBytes(bytesFixed)).append(" bytes");
        }
        if (bytesPerEx > 0) {
            if (bytesFixed > 0) {
                sb.append(" + ");
            }
            sb.append("nExamples * ").append(formatBytes(bytesPerEx)).append(" bytes");
        }
        sb.append("\n");
    }

    private String formatBytes(long bytes) {
        return BYTES_FORMAT.format(bytes);
    }

}