Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(20)

Unified Diff: tools/binary_size/java/src/org/chromium/tools/binary_size/ParallelAddress2Line.java

Issue 119083006: Add tool to help analyze binary size (Closed) Base URL: https://chromium.googlesource.com/chromium/src.git@master
Patch Set: Rebase onto master for a clean commit of gyp change Created 6 years, 11 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
Index: tools/binary_size/java/src/org/chromium/tools/binary_size/ParallelAddress2Line.java
diff --git a/tools/binary_size/java/src/org/chromium/tools/binary_size/ParallelAddress2Line.java b/tools/binary_size/java/src/org/chromium/tools/binary_size/ParallelAddress2Line.java
new file mode 100644
index 0000000000000000000000000000000000000000..8e49c43cfaf5890df67923e752a6018641372a8e
--- /dev/null
+++ b/tools/binary_size/java/src/org/chromium/tools/binary_size/ParallelAddress2Line.java
@@ -0,0 +1,526 @@
+// Copyright 2014 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+package org.chromium.tools.binary_size;
+
+import java.io.BufferedReader;
+import java.io.File;
+import java.io.FileNotFoundException;
+import java.io.FileOutputStream;
+import java.io.FileReader;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.OutputStream;
+import java.util.Timer;
+import java.util.TimerTask;
+import java.util.concurrent.CountDownLatch;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+
+/**
+ * A tool for parallelizing "addr2line" against a given binary.
+ * The tool runs "nm" to dump the symbols from a library, then spawns a pool
+ * of addr2name workers that resolve addresses to name in parallel.
+ * <p>
+ * This tool is intentionally written to be standalone so that it can be
+ * compiled without reliance upon any other libraries. All that is required
+ * is a vanilla installation of the Java Runtime Environment, 1.5 or later.
+ */
+// TODO(andrewhayden): Rewrite entire tool in Python
+public class ParallelAddress2Line {
+ private final AtomicBoolean mStillEnqueuing = new AtomicBoolean(true);
+ private final AtomicInteger mEnqueuedCount =
+ new AtomicInteger(Integer.MAX_VALUE);
+ private final AtomicInteger mDoneCount = new AtomicInteger(0);
+ private final AtomicInteger mSuccessCount = new AtomicInteger(0);
+ private final AtomicInteger mAddressSkipCount = new AtomicInteger(0);
+ private final String mLibraryPath;
+ private final String mNmPath;
+ private final String mNmInPath;
+ private final String mAddr2linePath;
+ private final boolean mVerbose;
+ private final boolean mNoProgress;
+ private Addr2LineWorkerPool mPool;
+ private final boolean mNoDedupe;
+ private final boolean mDisambiguate;
+ private final NmDumper mNmDumper;
+
+ private static final String USAGE =
+ "--addr2line [ARG] instead of the 'addr2line' in $PATH, use this (e.g.,\n" +
+ " arch-specific binary) (optional)\n" +
+ "--disambiguate create a listing of all source files which can be used\n" +
+ " to disambiguate some percentage of ambiguous source\n" +
+ " references; only useful on some architectures and adds\n" +
+ " significant startup cost (optional)\n" +
+ "--failfile [ARG] output symbols from failed lookups to the specified\n" +
+ " file (optional)\n" +
+ "--library [ARG] path to the library to process, e.g.\n" +
+ " out/Release/lib/libchromeview.so (required)\n" +
+ "--nm [ARG] instead of the 'nm' in $PATH, use this (e.g.,\n" +
+ " arch-specific binary) (optional)\n" +
+ "--nm-infile [ARG] instead of running nm on the specified library,\n" +
+ " ingest the specified nm file. (optional)\n" +
+ "--no-dedupe don't de-dupe symbols that live at the same address;\n" +
+ " deduping more accurately describes the use of space\n" +
+ " within the binary; if spatial analysis is your goal,\n" +
+ " leave deduplication on. (optional)\n" +
+ "--no-progress don't output periodic progress reports (optional)\n" +
+ "--outfile [ARG] output results into the specified file (required)\n" +
+ "--skipfile [ARG] output skipped symbols to the specified file (optional)\n" +
+ "--threads [ARG] number of parallel worker threads to create. Start low\n" +
+ " and watch your memory, defaults to 1 (optional)\n" +
+ "--verbose be verbose (optional)\n";
+
+ // Regex for parsing "nm" output. A sample line looks like this:
+ // 0167b39c 00000018 t ACCESS_DESCRIPTION_free /path/file.c:95
+ //
+ // The fields are: address, size, type, name, source location
+ // Regular expression explained ( see also: https://xkcd.com/208 ):
+ // ([0-9a-f]{8}+) The address
+ // [\\s]+ Whitespace separator
+ // ([0-9a-f]{8}+) The size. From here on out it's all optional.
+ // [\\s]+ Whitespace separator
+ // (\\S?) The symbol type, which is any non-whitespace char
+ // [\\s*] Whitespace separator
+ // ([^\\t]*) Symbol name, any non-tab character (spaces ok!)
+ // [\\t]? Tab separator
+ // (.*) The location (filename[:linennum|?][ (discriminator n)]
+ private static final Pattern sNmPattern = Pattern.compile(
+ "([0-9a-f]{8}+)[\\s]+([0-9a-f]{8}+)[\\s]*(\\S?)[\\s*]([^\\t]*)[\\t]?(.*)");
+
+ private ParallelAddress2Line(
+ final String libraryPath,
+ final String nmPath,
+ final String nmInPath,
+ final String addr2linePath,
+ final String outPath,
+ final String skipPath,
+ final String failPath,
+ final boolean verbose,
+ final boolean noProgress,
+ final boolean noDedupe,
+ final boolean disambiguate) {
+ this.mLibraryPath = libraryPath;
+ this.mNmPath = nmPath;
+ this.mNmInPath = nmInPath;
+ this.mAddr2linePath = addr2linePath;
+ this.mVerbose = verbose;
+ this.mNoProgress = noProgress;
+ this.mNoDedupe = noDedupe;
+ this.mDisambiguate = disambiguate;
+ this.mNmDumper = new NmDumper(outPath, failPath, skipPath);
+
+ final File libraryFile = new File(libraryPath);
+ if (!(libraryFile.exists() && libraryFile.canRead())) {
+ throw new IllegalStateException("Can't read library file: " + libraryPath);
+ }
+ }
+
+ private static final File findFile(File directory, String target) {
+ for (File file : directory.listFiles()) {
+ if (file.isDirectory() && file.canRead()) {
+ File result = findFile(file, target);
+ if (result != null) return result;
+ } else {
+ if (target.equals(file.getName())) return file;
+ }
+ }
+ return null;
+ }
+
+ private void run(final int addr2linePoolSize) throws InterruptedException {
+ try {
+ runInternal(addr2linePoolSize);
+ } finally {
+ mNmDumper.close();
+ }
+ }
+
+ private void runInternal(final int addr2linePoolSize) throws InterruptedException {
+ // Step 1: Dump symbols with nm
+ final String nmOutputPath;
+ if (mNmInPath == null) {
+ // Generate nm output with nm binary
+ logVerbose("Running nm to dump symbols from " + mLibraryPath);
+ try {
+ nmOutputPath = dumpSymbols();
+ } catch (Exception e) {
+ throw new RuntimeException("nm failed", e);
+ }
+ } else {
+ // Use user-supplied nm output
+ logVerbose("Using user-supplied nm file: " + mNmInPath);
+ nmOutputPath = mNmInPath;
+ }
+
+ // Step 2: Prepare addr2line worker pool to process nm output
+ try {
+ logVerbose("Creating " + addr2linePoolSize + " workers for " + mAddr2linePath);
+ mPool = new Addr2LineWorkerPool(addr2linePoolSize,
+ mAddr2linePath, mLibraryPath, mDisambiguate, !mNoDedupe);
+ } catch (IOException e) {
+ throw new RuntimeException("Couldn't initialize name2address pool!", e);
+ }
+
+ // Step 3: Spool symbol-processing tasks to workers
+ final long startTime = System.currentTimeMillis();
+ Timer timer = null;
+ if (!mNoProgress) {
+ timer = startTaskMonitor(startTime);
+ }
+ final int queued = spoolTasks(nmOutputPath);
+
+ // All tasks have been enqueued.
+ mEnqueuedCount.set(queued);
+ mStillEnqueuing.set(false);
+ mPool.allRecordsSubmitted();
+ float percentAddressesSkipped = 100f * (mAddressSkipCount.floatValue()
+ / (queued + mAddressSkipCount.get()));
+ float percentAddressesQueued = 100f - percentAddressesSkipped;
+ int totalAddresses = mAddressSkipCount.get() + queued;
+ logVerbose("All addresses have been enqueued (total " + queued + ").");
+ // Remember that the queue to which the addresses was enqueued is of a
+ // small fixed size; by the time this code executes, there is very
+ // little work left to do. Await the termination of the pool with a
+ // reasonable timeout for safety purposes.
+ boolean timedOut = !mPool.await(5, TimeUnit.MINUTES);
+ if (timedOut) {
+ throw new RuntimeException("Worker pool did not terminate!");
+ }
+ if (!mNoProgress) timer.cancel();
+ log(totalAddresses + " addresses discovered; " +
+ queued + " queued for processing (" +
+ String.format("%.2f", percentAddressesQueued) + "%), " +
+ mAddressSkipCount.get() + " skipped (" +
+ String.format("%.2f", percentAddressesSkipped) + "%)");
+ dumpStats(startTime);
+ log("Done.");
+ }
+
+ /**
+ * Monitors the pool periodically printing status updates to stdout.
+ * @param addressProcessingStartTime the time address processing began
+ * @return the daemon timer that is generating the status updates
+ */
+ private final Timer startTaskMonitor(
+ final long addressProcessingStartTime) {
+ Runnable monitorTask = new OutputSpooler();
+ Thread monitor = new Thread(monitorTask, "progress monitor");
+ monitor.setDaemon(true);
+ monitor.start();
+
+ TimerTask task = new TimerTask() {
+ @Override
+ public void run() {
+ dumpStats(addressProcessingStartTime);
+ }
+ };
+ Timer timer = new Timer(true);
+ timer.schedule(task, 1000L, 1000L);
+ return timer;
+ }
+
+ /**
+ * Spools address-lookup tasks to the addr2line workers.
+ * This method will block until most of (or possibly all of) the tasks
+ * have been spooled.
+ * If a skip path is set, any line in the input file that doesn't have
+ * an address will be copied into the skip file.
+ *
+ * @param inputPath the path to the dump produced by nm
+ * @return the number of tasks spooled
+ */
+ private final int spoolTasks(final String inputPath) {
+ FileReader inputReader = null;
+ try {
+ inputReader = new FileReader(inputPath);
+ } catch (IOException e) {
+ throw new RuntimeException("Can't open input file: " + inputPath, e);
+ }
+ final BufferedReader bufferedReader = new BufferedReader(inputReader);
+
+ String currentLine = null;
+ int numSpooled = 0;
+ try {
+ while ((currentLine = bufferedReader.readLine()) != null) {
+ try {
+ final Matcher matcher = sNmPattern.matcher(currentLine);
+ if (!matcher.matches()) {
+ // HACK: Special case for ICU data.
+ // This thing is HUGE (5+ megabytes) and is currently
+ // missed because there is no size information.
+ // torne@ has volunteered to patch the generation code
+ // so that the generated ASM includes a size attribute
+ // so that this hard-coding can go away in the future.
+ if (currentLine.endsWith("icudt46_dat")) {
+ Record record = getIcuRecord(currentLine);
+ if (record != null) {
+ numSpooled++;
+ mPool.submit(record);
+ continue;
+ }
+ }
+ mNmDumper.skipped(currentLine);
+ mAddressSkipCount.incrementAndGet();
+ continue;
+ }
+ final Record record = new Record();
+ record.address = matcher.group(1);
+ record.size = matcher.group(2);
+ if (matcher.groupCount() >= 3) {
+ record.symbolType = matcher.group(3).charAt(0);
+ }
+ if (matcher.groupCount() >= 4) {
+ // May or may not be present
+ record.symbolName = matcher.group(4);
+ }
+ numSpooled++;
+ mPool.submit(record);
+ } catch (Exception e) {
+ throw new RuntimeException("Error processing line: '" + currentLine + "'", e);
+ }
+ }
+ } catch (Exception e) {
+ throw new RuntimeException("Input processing failed", e);
+ } finally {
+ try {
+ bufferedReader.close();
+ } catch (Exception ignored) {
+ // Nothing to be done
+ }
+ try {
+ inputReader.close();
+ } catch (Exception ignored) {
+ // Nothing to be done
+ }
+ }
+ return numSpooled;
+ }
+
+ private Record getIcuRecord(String line) throws IOException {
+ // Line looks like this:
+ // 01c9ee00 r icudt46_dat
+ String[] parts = line.split("\\s");
+ if (parts.length != 3) return null;
+
+ // Convert /src/out/Release/lib/[libraryfile] -> /src/out/Release
+ final File libraryOutputDirectory = new File(mLibraryPath)
+ .getParentFile().getParentFile().getCanonicalFile();
+ final File icuDir = new File(
+ libraryOutputDirectory.getAbsolutePath() +
+ "/obj/third_party/icu");
+ final File icuFile = findFile(icuDir, "icudata.icudt46l_dat.o");
+ if (!icuFile.exists()) return null;
+ final Record record = new Record();
+ record.address = parts[0];
+ record.symbolType = parts[1].charAt(0);
+ record.symbolName = parts[2];
+ record.size = Integer.toHexString((int) icuFile.length());
+ record.location = icuFile.getCanonicalPath() + ":0";
+ record.resolvedSuccessfully = true;
+ while (record.size.length() < 8) {
+ record.size = "0" + record.size;
+ }
+ return record;
+ }
+
+ /**
+ * @return the path to the file that nm wrote
+ * @throws Exception
+ * @throws FileNotFoundException
+ * @throws InterruptedException
+ */
+ private String dumpSymbols() throws Exception, FileNotFoundException, InterruptedException {
+ final Process process = createNmProcess();
+ final File tempFile = File.createTempFile("ParallelAddress2Line", "nm");
+ tempFile.deleteOnExit();
+ final CountDownLatch completionLatch = sink(
+ process.getInputStream(), new FileOutputStream(tempFile), true);
+ sink(process.getErrorStream(), System.err, false);
+ logVerbose("Dumping symbols to: " + tempFile.getAbsolutePath());
+ final int nmRc = process.waitFor();
+ if (nmRc != 0) {
+ throw new RuntimeException("nm process returned " + nmRc);
+ }
+ completionLatch.await(); // wait for output to be done
+ return tempFile.getAbsolutePath();
+ }
+
+ private void dumpStats(final long startTime) {
+ long successful = mSuccessCount.get();
+ long doneNow = mDoneCount.get();
+ long unsuccessful = doneNow - successful;
+ float successPercent = doneNow == 0 ? 100f : 100f * ((float)successful / (float)doneNow);
+ long elapsedMillis = System.currentTimeMillis() - startTime;
+ float elapsedSeconds = elapsedMillis / 1000f;
+ long throughput = doneNow / (elapsedMillis / 1000);
+ final int mapLookupSuccess = mPool.getDisambiguationSuccessCount();
+ final int mapLookupFailure = mPool.getDisambiguationFailureCount();
+ final int mapLookupTotal = mapLookupSuccess + mapLookupFailure;
+ float mapLookupSuccessPercent = 0f;
+ if (mapLookupTotal != 0 && mapLookupSuccess != 0) {
+ mapLookupSuccessPercent = 100f *
+ ((float) mapLookupSuccess / (float) mapLookupTotal);
+ }
+
+ log(doneNow + " addresses processed (" +
+ mSuccessCount.get() + " ok, " + unsuccessful + " failed)" +
+ ", avg " + throughput + " addresses/sec, " +
+ String.format("%.2f", successPercent) + "% success" +
+ ", " + mapLookupTotal + " ambiguous path" +
+ (!mDisambiguate ? "" :
+ ", (" + String.format("%.2f", mapLookupSuccessPercent) + "% disambiguated)") +
+ (mNoDedupe ? "" : ", " + mPool.getDedupeCount() + " deduped") +
+ ", elapsed time " + String.format("%.3f", elapsedSeconds) + " seconds");
+ }
+
+ private Process createNmProcess() throws Exception {
+ ProcessBuilder builder = new ProcessBuilder(
+ mNmPath,
+ "-C", // demangle (for the humans)
+ "-S", // print size
+ mLibraryPath);
+ logVerbose("Creating process: " + builder.command());
+ return builder.start();
+ }
+
+ /**
+ * Make a pipe to drain the specified input stream into the specified
+ * output stream asynchronously.
+ * @param in read from here
+ * @param out and write to here
+ * @param closeWhenDone whether or not to close the target output stream
+ * when the input stream terminates
+ * @return a latch that can be used to await the final write to the
+ * output stream, which occurs when either of the streams closes
+ */
+ private static final CountDownLatch sink(final InputStream in,
+ final OutputStream out, final boolean closeWhenDone) {
+ final CountDownLatch latch = new CountDownLatch(1);
+ final Runnable task = new Runnable() {
+ @Override
+ public void run() {
+ byte[] buffer = new byte[4096];
+ try {
+ int numRead = 0;
+ do {
+ numRead = in.read(buffer);
+ if (numRead > 0) {
+ out.write(buffer, 0, numRead);
+ out.flush();
+ }
+ } while (numRead >= 0);
+ } catch (Exception e) {
+ e.printStackTrace();
+ } finally {
+ try { out.flush(); } catch (Exception ignored) {
+ // Nothing to be done
+ }
+ if (closeWhenDone) {
+ try { out.close(); } catch (Exception ignored) {
+ // Nothing to be done
+ }
+ }
+ latch.countDown();
+ }
+ }
+ };
+ final Thread worker = new Thread(task, "pipe " + in + "->" + out);
+ worker.setDaemon(true);
+ worker.start();
+ return latch;
+ }
+
+ private final class OutputSpooler implements Runnable {
+ @Override
+ public void run() {
+ do {
+ readRecord();
+ } while (mStillEnqueuing.get() || (mDoneCount.get() < mEnqueuedCount.get()));
+ }
+
+ /**
+ * Read a record and process it.
+ */
+ private void readRecord() {
+ Record record = mPool.poll();
+ if (record != null) {
+ mDoneCount.incrementAndGet();
+ if (record.resolvedSuccessfully) {
+ mSuccessCount.incrementAndGet();
+ mNmDumper.succeeded(record);
+ } else {
+ mNmDumper.failed(record);
+ }
+ } else {
+ try {
+ // wait to keep going
+ Thread.sleep(100);
+ } catch (InterruptedException e) {
+ e.printStackTrace();
+ }
+ }
+ }
+ }
+
+ /**
+ * Log a message to the console.
+ * @param message the message to log
+ */
+ private final void log(String message) {
+ System.out.println(message);
+ }
+
+ /**
+ * Log a message to the console iff verbose logging is enabled.
+ * @param message the message to log
+ */
+ private final void logVerbose(String message) {
+ if (mVerbose) log(message);
+ }
+
+ /**
+ * Runs the tool. Run with --help for limited help.
+ * @param args
+ * @throws Exception if anything explodes
+ */
+ public static void main(String[] args) throws Exception {
+ ParallelAddress2Line tool = new ParallelAddress2Line(
+ getArg(args, "--library"),
+ getArg(args, "--nm", "nm"),
+ getArg(args, "--nm-infile", null),
+ getArg(args, "--addr2line", "addr2line"),
+ getArg(args, "--outfile"),
+ getArg(args, "--skipfile", null),
+ getArg(args, "--failfile", null),
+ hasFlag(args, "--verbose"),
+ hasFlag(args, "--no-progress"),
+ hasFlag(args, "--no-dedupe"),
+ hasFlag(args, "--disambiguate"));
+ tool.run(Integer.parseInt(getArg(args, "--threads", "1")));
+ }
+
+ private static boolean hasFlag(String[] args, String name) {
+ for (int x = 0; x < args.length; x++) if (name.equals(args[x])) return true;
+ return false;
+ }
+
+ private static String getArg(String[] args, String name, String defaultValue) {
+ for (int x = 0; x < args.length; x++) {
+ if (name.equals(args[x])) {
+ if (x < args.length - 1) return args[x + 1];
+ throw new RuntimeException(name + " is missing a value\n" + USAGE);
+ }
+ }
+ return defaultValue;
+ }
+
+ private static String getArg(String[] args, String name) {
+ String result = getArg(args, name, null);
+ if (result == null) throw new RuntimeException(name + " is required\n" + USAGE);
+ return result;
+ }
+}

Powered by Google App Engine
This is Rietveld 408576698