From 417a4f3d126b48817e3aed000341f0fc4189be36 Mon Sep 17 00:00:00 2001
From: Kenneth Heafield <github@kheafield.com>
Date: Sun, 19 May 2013 10:02:45 -0400
Subject: KenLM 10ddf7d92335 Comment some asserts, update uusage

---
 klm/lm/search_trie.hh              |  3 +-
 klm/util/double-conversion/utils.h |  9 ++++--
 klm/util/usage.cc                  | 65 +++++++++++++++++++++++++++-----------
 3 files changed, 55 insertions(+), 22 deletions(-)

(limited to 'klm')

diff --git a/klm/lm/search_trie.hh b/klm/lm/search_trie.hh
index 1264baf5..60be416b 100644
--- a/klm/lm/search_trie.hh
+++ b/klm/lm/search_trie.hh
@@ -41,7 +41,8 @@ template <class Quant, class Bhiksha> class TrieSearch {
     static void UpdateConfigFromBinary(int fd, const std::vector<uint64_t> &counts, Config &config) {
       Quant::UpdateConfigFromBinary(fd, counts, config);
       util::AdvanceOrThrow(fd, Quant::Size(counts.size(), config) + Unigram::Size(counts[0]));
-      Bhiksha::UpdateConfigFromBinary(fd, config);
+      // Currently the unigram pointers are not compresssed, so there will only be a header for order > 2.
+      if (counts.size() > 2) Bhiksha::UpdateConfigFromBinary(fd, config);
     }
 
     static uint64_t Size(const std::vector<uint64_t> &counts, const Config &config) {
diff --git a/klm/util/double-conversion/utils.h b/klm/util/double-conversion/utils.h
index 767094b8..2bd71605 100644
--- a/klm/util/double-conversion/utils.h
+++ b/klm/util/double-conversion/utils.h
@@ -218,7 +218,8 @@ class StringBuilder {
   // 0-characters; use the Finalize() method to terminate the string
   // instead.
   void AddCharacter(char c) {
-    ASSERT(c != '\0');
+    // I just extract raw data not a cstr so null is fine.
+    //ASSERT(c != '\0');
     ASSERT(!is_finalized() && position_ < buffer_.length());
     buffer_[position_++] = c;
   }
@@ -233,7 +234,8 @@ class StringBuilder {
   // builder. The input string must have enough characters.
   void AddSubstring(const char* s, int n) {
     ASSERT(!is_finalized() && position_ + n < buffer_.length());
-    ASSERT(static_cast<size_t>(n) <= strlen(s));
+    // I just extract raw data not a cstr so null is fine.
+    //ASSERT(static_cast<size_t>(n) <= strlen(s));
     memmove(&buffer_[position_], s, n * kCharSize);
     position_ += n;
   }
@@ -253,7 +255,8 @@ class StringBuilder {
     buffer_[position_] = '\0';
     // Make sure nobody managed to add a 0-character to the
     // buffer while building the string.
-    ASSERT(strlen(buffer_.start()) == static_cast<size_t>(position_));
+    // I just extract raw data not a cstr so null is fine.
+    //ASSERT(strlen(buffer_.start()) == static_cast<size_t>(position_));
     position_ = -1;
     ASSERT(is_finalized());
     return buffer_.start();
diff --git a/klm/util/usage.cc b/klm/util/usage.cc
index ad4dc7b4..5fa3cc9a 100644
--- a/klm/util/usage.cc
+++ b/klm/util/usage.cc
@@ -5,51 +5,80 @@
 #include <fstream>
 #include <ostream>
 #include <sstream>
+#include <set>
+#include <string>
 
 #include <string.h>
 #include <ctype.h>
 #if !defined(_WIN32) && !defined(_WIN64)
 #include <sys/resource.h>
 #include <sys/time.h>
+#include <time.h>
 #include <unistd.h>
 #endif
 
 namespace util {
 
-namespace {
 #if !defined(_WIN32) && !defined(_WIN64)
+namespace {
 float FloatSec(const struct timeval &tv) {
   return static_cast<float>(tv.tv_sec) + (static_cast<float>(tv.tv_usec) / 1000000.0);
 }
-#endif
+float FloatSec(const struct timespec &tv) {
+  return static_cast<float>(tv.tv_sec) + (static_cast<float>(tv.tv_nsec) / 1000000000.0);
+}
 
 const char *SkipSpaces(const char *at) {
-  for (; *at == ' '; ++at) {}
+  for (; *at == ' ' || *at == '\t'; ++at) {}
   return at;
 }
+
+class RecordStart {
+  public:
+    RecordStart() {
+      clock_gettime(CLOCK_MONOTONIC, &started_);
+    }
+
+    const struct timespec &Started() const {
+      return started_;
+    }
+
+  private:
+    struct timespec started_;
+};
+
+const RecordStart kRecordStart;
 } // namespace
+#endif
 
 void PrintUsage(std::ostream &out) {
 #if !defined(_WIN32) && !defined(_WIN64)
+  // Linux doesn't set memory usage in getrusage :-(
+  std::set<std::string> headers;
+  headers.insert("VmPeak:");
+  headers.insert("VmRSS:");
+  headers.insert("Name:");
+
+  std::ifstream status("/proc/self/status", std::ios::in);
+  std::string header, value;
+  while ((status >> header) && getline(status, value)) {
+    if (headers.find(header) != headers.end()) {
+      out << header << SkipSpaces(value.c_str()) << '\t';
+    }
+  }
+
   struct rusage usage;
-  if (getrusage(RUSAGE_SELF, &usage)) {
+  if (getrusage(RUSAGE_CHILDREN, &usage)) {
     perror("getrusage");
     return;
   }
-  out << "user\t" << FloatSec(usage.ru_utime) << "\nsys\t" << FloatSec(usage.ru_stime) << '\n';
-  out << "CPU\t" << (FloatSec(usage.ru_utime) + FloatSec(usage.ru_stime)) << '\n';
-  // Linux doesn't set memory usage :-(.  
-  std::ifstream status("/proc/self/status", std::ios::in);
-  std::string line;
-  while (getline(status, line)) {
-    if (!strncmp(line.c_str(), "VmRSS:\t", 7)) {
-      out << "RSSCur\t" << SkipSpaces(line.c_str() + 7) << '\n';
-      break;
-    } else if (!strncmp(line.c_str(), "VmPeak:\t", 8)) {
-      out << "VmPeak\t" << SkipSpaces(line.c_str() + 8) << '\n';
-    }
-  }
-  out << "RSSMax\t" << usage.ru_maxrss << " kB" << '\n';
+  out << "RSSMax:" << usage.ru_maxrss << " kB" << '\t';
+  out << "user:" << FloatSec(usage.ru_utime) << "\tsys:" << FloatSec(usage.ru_stime) << '\t';
+  out << "CPU:" << (FloatSec(usage.ru_utime) + FloatSec(usage.ru_stime));
+
+  struct timespec current;
+  clock_gettime(CLOCK_MONOTONIC, &current);
+  out << "\treal:" << (FloatSec(current) - FloatSec(kRecordStart.Started())) << '\n';
 #endif
 }
 
-- 
cgit v1.2.3