Move a bunch of stattimer stuff into header files

This roughly halves stattimer overhead, from +80% to +40%
(on a very small benchmark; more complex benchmarks probably
have lower overhead)

src/core/stats.cpp

@@ -36,70 +36,6 @@ extern "C" const char* getStatTimerName() {
 }
 
 __thread StatTimer* StatTimer::stack;
-StatTimer::StatTimer(int statid, bool push) {
-    uint64_t at_time = getCPUTicks();
-    _start_time = 0;
-    _statid = statid;
-
-    if (!push) {
-        _prev = NULL;
-        return;
-    }
-
-    _prev = stack;
-    stack = this;
-    if (_prev) {
-        _prev->pause(at_time);
-    }
-    resume(at_time);
-}
-
-StatTimer::StatTimer(int statid, uint64_t at_time) {
-    _start_time = 0;
-    _statid = statid;
-    _prev = stack;
-    stack = this;
-    if (_prev) {
-        _prev->pause(at_time);
-    }
-    resume(at_time);
-}
-
-StatTimer::~StatTimer() {
-    assert(stack == this);
-
-    uint64_t at_time;
-    if (!isPaused()) {
-        at_time = getCPUTicks();
-        pause(at_time);
-    } else {
-        // fprintf (stderr, "WARNING: timer was paused.\n");
-        at_time = _last_pause_time;
-    }
-
-    stack = _prev;
-    if (stack) {
-        stack->resume(at_time);
-    }
-}
-
-void StatTimer::pause(uint64_t at_time) {
-    assert(!isPaused());
-    assert(at_time > _start_time);
-
-    uint64_t _duration = at_time - _start_time;
-    Stats::log(_statid, _duration);
-
-    _start_time = 0;
-    _last_pause_time = at_time;
-    // fprintf (stderr, "paused  %d at %lu\n", _statid, at_time);
-}
-
-void StatTimer::resume(uint64_t at_time) {
-    assert(isPaused());
-    _start_time = at_time;
-    // fprintf (stderr, "resumed %d at %lu\n", _statid, at_time);
-}
 
 StatTimer* StatTimer::swapStack(StatTimer* s, uint64_t at_time) {
     StatTimer* prev_stack = stack;

src/core/stats.h

@@ -126,12 +126,69 @@ private:
     int _statid;
 
 public:
-    StatTimer(int statid, bool push = true);
-    StatTimer(int statid, uint64_t at_time);
-    ~StatTimer();
-
-    void pause(uint64_t at_time);
-    void resume(uint64_t at_time);
+    StatTimer(int statid, bool push = true) {
+        uint64_t at_time = getCPUTicks();
+        _start_time = 0;
+        _statid = statid;
+
+        if (!push) {
+            _prev = NULL;
+            return;
+        }
+
+        _prev = stack;
+        stack = this;
+        if (_prev) {
+            _prev->pause(at_time);
+        }
+        resume(at_time);
+    }
+
+    StatTimer(int statid, uint64_t at_time) {
+        _start_time = 0;
+        _statid = statid;
+        _prev = stack;
+        stack = this;
+        if (_prev) {
+            _prev->pause(at_time);
+        }
+        resume(at_time);
+    }
+
+    ~StatTimer() {
+        assert(stack == this);
+
+        uint64_t at_time;
+        if (!isPaused()) {
+            at_time = getCPUTicks();
+            pause(at_time);
+        } else {
+            // fprintf (stderr, "WARNING: timer was paused.\n");
+            at_time = _last_pause_time;
+        }
+
+        stack = _prev;
+        if (stack) {
+            stack->resume(at_time);
+        }
+    }
+
+    void pause(uint64_t at_time) {
+        assert(!isPaused());
+        assert(at_time > _start_time);
+
+        uint64_t _duration = at_time - _start_time;
+        Stats::log(_statid, _duration);
+
+        _start_time = 0;
+        _last_pause_time = at_time;
+        // fprintf (stderr, "paused  %d at %lu\n", _statid, at_time);
+    }
+    void resume(uint64_t at_time) {
+        assert(isPaused());
+        _start_time = at_time;
+        // fprintf (stderr, "resumed %d at %lu\n", _statid, at_time);
+    }
 
     bool isPaused() const { return _start_time == 0; }
     int getId() const { return _statid; }

src/core/util.cpp

@@ -29,18 +29,6 @@
 
 namespace pyston {
 
-static inline uint64_t rdtsc() {
-    unsigned long lo, hi;
-    asm("rdtsc" : "=a"(lo), "=d"(hi));
-    return (lo | (hi << 32));
-}
-
-uint64_t getCPUTicks() {
-    // unsigned int _unused;
-    // return __rdtscp(&_unused);
-    return rdtsc();
-}
-
 #if !DISABLE_TIMERS
 
 int Timer::level = 0;

src/core/util.h

@@ -24,7 +24,11 @@
 
 namespace pyston {
 
-uint64_t getCPUTicks();
+inline uint64_t getCPUTicks() {
+    unsigned long lo, hi;
+    asm("rdtsc" : "=a"(lo), "=d"(hi));
+    return (lo | (hi << 32));
+}
 
 #define DISABLE_TIMERS 0