initial import of Chrome's trace_event into skia framework
authorcommit-bot@chromium.org <commit-bot@chromium.org@2bbb7eff-a529-9590-31e7-b0007b416f81>
Fri, 31 Jan 2014 00:04:25 +0000 (00:04 +0000)
committercommit-bot@chromium.org <commit-bot@chromium.org@2bbb7eff-a529-9590-31e7-b0007b416f81>
Fri, 31 Jan 2014 00:04:25 +0000 (00:04 +0000)
This patch includes a modified version of Chrome's trace_event.h, which provides
tracing macros that can easily integrate into the about://tracing framework.

Currently the macros link to a default implementation of the (narrow) tracing
class SkDefaultEventTracer which does nothing; next step will be to have Chrome
subclass the SkEventTracer with a shim that bolts Skia's trace events to its own,
allowing Skia's trace events to show up in about://tracing.

I've verified that this file builds properly, and when I added a simple scoped
TRACE_EVENT0 to SkCanvas::drawRect, along with some debug prints in the NOP
implementation of tracing, I saw what I expected printed to the screen.

BUG=skia:
R=nduca@chromium.org, reed@google.com, mtklein@google.com, bsalomon@google.com

Author: humper@google.com

Review URL: https://codereview.chromium.org/149563004

git-svn-id: http://skia.googlecode.com/svn/trunk@13256 2bbb7eff-a529-9590-31e7-b0007b416f81

gyp/core.gypi
gyp/tests.gyp
gyp/utils.gyp
include/utils/SkEventTracer.h [new file with mode: 0644]
src/core/SkTraceEvent.h [new file with mode: 0644]
src/utils/SkEventTracer.cpp [new file with mode: 0644]
tests/TracingTest.cpp [new file with mode: 0644]

index fcd5e39..b199868 100644 (file)
         '<(skia_src_path)/core/SkTileGridPicture.cpp',
         '<(skia_src_path)/core/SkTLList.h',
         '<(skia_src_path)/core/SkTLS.cpp',
+        '<(skia_src_path)/core/SkTraceEvent.h',
         '<(skia_src_path)/core/SkTSearch.cpp',
         '<(skia_src_path)/core/SkTSort.h',
         '<(skia_src_path)/core/SkTypeface.cpp',
index 827ba3f..42ddebe 100644 (file)
         '../tests/Test.h',
         '../tests/TestSize.cpp',
         '../tests/TileGridTest.cpp',
+        '../tests/TracingTest.cpp',
         '../tests/ToUnicode.cpp',
         '../tests/Typeface.cpp',
         '../tests/UnicodeTest.cpp',
index e9e9109..3f6e5b5 100644 (file)
@@ -41,6 +41,7 @@
         '../include/utils/SkDebugUtils.h',
         '../include/utils/SkDeferredCanvas.h',
         '../include/utils/SkDumpCanvas.h',
+        '../include/utils/SkEventTracer.h',
         '../include/utils/SkInterpolator.h',
         '../include/utils/SkLayer.h',
         '../include/utils/SkMatrix44.h',
@@ -74,6 +75,7 @@
         '../src/utils/SkCullPoints.cpp',
         '../src/utils/SkDeferredCanvas.cpp',
         '../src/utils/SkDumpCanvas.cpp',
+        '../src/utils/SkEventTracer.cpp',
         '../src/utils/SkFloatUtils.h',
         '../src/utils/SkGatherPixelRefsAndRects.cpp',
         '../src/utils/SkGatherPixelRefsAndRects.h',
diff --git a/include/utils/SkEventTracer.h b/include/utils/SkEventTracer.h
new file mode 100644 (file)
index 0000000..af2f68d
--- /dev/null
@@ -0,0 +1,75 @@
+/*
+ * Copyright (C) 2014 Google Inc. All rights reserved.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#ifndef SkEventTracer_DEFINED
+#define SkEventTracer_DEFINED
+
+// The class in this header defines the interface between Skia's internal
+// tracing macros and an external entity (e.g., Chrome) that will consume them.
+// Such an entity should subclass SkEventTracer and provide an instance of 
+// that event to SkEventTracer::SetInstance.
+
+// If you're looking for the tracing macros to instrument Skia itself, those
+// live in src/core/SkTraceEvent.h
+
+#include "SkTypes.h"
+
+// This will mark the trace event as disabled by default. The user will need
+// to explicitly enable the event.
+#define TRACE_DISABLED_BY_DEFAULT(name) "disabled-by-default-" name
+
+class SK_API SkEventTracer {
+public:
+
+    typedef uint32_t Handle;
+    
+    static SkEventTracer* GetInstance();
+
+    static void SetInstance(SkEventTracer* tracer) { 
+        SkDELETE(SkEventTracer::gInstance);
+        SkEventTracer::gInstance = tracer;
+    }
+
+    virtual ~SkEventTracer() { }
+
+    // The pointer returned from GetCategoryGroupEnabled() points to a
+    // value with zero or more of the following bits. Used in this class only.
+    // The TRACE_EVENT macros should only use the value as a bool.
+    // These values must be in sync with macro values in trace_event.h in chromium.
+    enum CategoryGroupEnabledFlags {
+        // Category group enabled for the recording mode.
+        kEnabledForRecording_CategoryGroupEnabledFlags = 1 << 0,
+        // Category group enabled for the monitoring mode.
+        kEnabledForMonitoring_CategoryGroupEnabledFlags = 1 << 1,
+        // Category group enabled by SetEventCallbackEnabled().
+        kEnabledForEventCallback_CategoryGroupEnabledFlags = 1 << 2,
+    };
+
+    virtual const unsigned char* getCategoryGroupEnabled(const char* name) = 0;
+    virtual const char* getCategoryGroupName(
+      const uint8_t* category_group_enabled) = 0;
+    
+    virtual SkEventTracer::Handle 
+        addTraceEvent(char phase,
+                      const uint8_t* categoryEnabledFlag,
+                      const char* name,
+                      uint64_t id,
+                      int32_t numArgs,
+                      const char** argNames,
+                      const uint8_t* argTypes,
+                      const uint64_t* argValues,
+                      uint8_t flags) = 0;
+    
+    virtual void 
+        updateTraceEventDuration(const uint8_t* categoryEnabledFlag, 
+                                 const char* name, 
+                                 SkEventTracer::Handle) = 0;
+private:
+    static SkEventTracer *gInstance;
+};
+
+#endif // SkEventTracer_DEFINED
diff --git a/src/core/SkTraceEvent.h b/src/core/SkTraceEvent.h
new file mode 100644 (file)
index 0000000..0c7989f
--- /dev/null
@@ -0,0 +1,1245 @@
+// Copyright (c) 2014 Google Inc.
+//
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// This header file defines the set of trace_event macros without specifying
+// how the events actually get collected and stored. If you need to expose trace
+// events to some other universe, you can copy-and-paste this file as well as
+// trace_event.h, modifying the macros contained there as necessary for the
+// target platform. The end result is that multiple libraries can funnel events
+// through to a shared trace event collector.
+
+// Trace events are for tracking application performance and resource usage.
+// Macros are provided to track:
+//    Begin and end of function calls
+//    Counters
+//
+// Events are issued against categories. Whereas LOG's
+// categories are statically defined, TRACE categories are created
+// implicitly with a string. For example:
+//   TRACE_EVENT_INSTANT0("MY_SUBSYSTEM", "SomeImportantEvent",
+//                        TRACE_EVENT_SCOPE_THREAD)
+//
+// It is often the case that one trace may belong in multiple categories at the
+// same time. The first argument to the trace can be a comma-separated list of
+// categories, forming a category group, like:
+//
+// TRACE_EVENT_INSTANT0("input,views", "OnMouseOver", TRACE_EVENT_SCOPE_THREAD)
+//
+// We can enable/disable tracing of OnMouseOver by enabling/disabling either
+// category.
+//
+// Events can be INSTANT, or can be pairs of BEGIN and END in the same scope:
+//   TRACE_EVENT_BEGIN0("MY_SUBSYSTEM", "SomethingCostly")
+//   doSomethingCostly()
+//   TRACE_EVENT_END0("MY_SUBSYSTEM", "SomethingCostly")
+// Note: our tools can't always determine the correct BEGIN/END pairs unless
+// these are used in the same scope. Use ASYNC_BEGIN/ASYNC_END macros if you
+// need them to be in separate scopes.
+//
+// A common use case is to trace entire function scopes. This
+// issues a trace BEGIN and END automatically:
+//   void doSomethingCostly() {
+//     TRACE_EVENT0("MY_SUBSYSTEM", "doSomethingCostly");
+//     ...
+//   }
+//
+// Additional parameters can be associated with an event:
+//   void doSomethingCostly2(int howMuch) {
+//     TRACE_EVENT1("MY_SUBSYSTEM", "doSomethingCostly",
+//         "howMuch", howMuch);
+//     ...
+//   }
+//
+// The trace system will automatically add to this information the
+// current process id, thread id, and a timestamp in microseconds.
+//
+// To trace an asynchronous procedure such as an IPC send/receive, use
+// ASYNC_BEGIN and ASYNC_END:
+//   [single threaded sender code]
+//     static int send_count = 0;
+//     ++send_count;
+//     TRACE_EVENT_ASYNC_BEGIN0("ipc", "message", send_count);
+//     Send(new MyMessage(send_count));
+//   [receive code]
+//     void OnMyMessage(send_count) {
+//       TRACE_EVENT_ASYNC_END0("ipc", "message", send_count);
+//     }
+// The third parameter is a unique ID to match ASYNC_BEGIN/ASYNC_END pairs.
+// ASYNC_BEGIN and ASYNC_END can occur on any thread of any traced process.
+// Pointers can be used for the ID parameter, and they will be mangled
+// internally so that the same pointer on two different processes will not
+// match. For example:
+//   class MyTracedClass {
+//    public:
+//     MyTracedClass() {
+//       TRACE_EVENT_ASYNC_BEGIN0("category", "MyTracedClass", this);
+//     }
+//     ~MyTracedClass() {
+//       TRACE_EVENT_ASYNC_END0("category", "MyTracedClass", this);
+//     }
+//   }
+//
+// Trace event also supports counters, which is a way to track a quantity
+// as it varies over time. Counters are created with the following macro:
+//   TRACE_COUNTER1("MY_SUBSYSTEM", "myCounter", g_myCounterValue);
+//
+// Counters are process-specific. The macro itself can be issued from any
+// thread, however.
+//
+// Sometimes, you want to track two counters at once. You can do this with two
+// counter macros:
+//   TRACE_COUNTER1("MY_SUBSYSTEM", "myCounter0", g_myCounterValue[0]);
+//   TRACE_COUNTER1("MY_SUBSYSTEM", "myCounter1", g_myCounterValue[1]);
+// Or you can do it with a combined macro:
+//   TRACE_COUNTER2("MY_SUBSYSTEM", "myCounter",
+//       "bytesPinned", g_myCounterValue[0],
+//       "bytesAllocated", g_myCounterValue[1]);
+// This indicates to the tracing UI that these counters should be displayed
+// in a single graph, as a summed area chart.
+//
+// Since counters are in a global namespace, you may want to disambiguate with a
+// unique ID, by using the TRACE_COUNTER_ID* variations.
+//
+// By default, trace collection is compiled in, but turned off at runtime.
+// Collecting trace data is the responsibility of the embedding
+// application. In Chrome's case, navigating to about:tracing will turn on
+// tracing and display data collected across all active processes.
+//
+//
+// Memory scoping note:
+// Tracing copies the pointers, not the string content, of the strings passed
+// in for category_group, name, and arg_names.  Thus, the following code will
+// cause problems:
+//     char* str = strdup("importantName");
+//     TRACE_EVENT_INSTANT0("SUBSYSTEM", str);  // BAD!
+//     free(str);                   // Trace system now has dangling pointer
+//
+// To avoid this issue with the |name| and |arg_name| parameters, use the
+// TRACE_EVENT_COPY_XXX overloads of the macros at additional runtime overhead.
+// Notes: The category must always be in a long-lived char* (i.e. static const).
+//        The |arg_values|, when used, are always deep copied with the _COPY
+//        macros.
+//
+// When are string argument values copied:
+// const char* arg_values are only referenced by default:
+//     TRACE_EVENT1("category", "name",
+//                  "arg1", "literal string is only referenced");
+// Use TRACE_STR_COPY to force copying of a const char*:
+//     TRACE_EVENT1("category", "name",
+//                  "arg1", TRACE_STR_COPY("string will be copied"));
+// std::string arg_values are always copied:
+//     TRACE_EVENT1("category", "name",
+//                  "arg1", std::string("string will be copied"));
+//
+//
+// Thread Safety:
+// A thread safe singleton and mutex are used for thread safety. Category
+// enabled flags are used to limit the performance impact when the system
+// is not enabled.
+//
+// TRACE_EVENT macros first cache a pointer to a category. The categories are
+// statically allocated and safe at all times, even after exit. Fetching a
+// category is protected by the TraceLog::lock_. Multiple threads initializing
+// the static variable is safe, as they will be serialized by the lock and
+// multiple calls will return the same pointer to the category.
+//
+// Then the category_group_enabled flag is checked. This is a unsigned char, and
+// not intended to be multithread safe. It optimizes access to AddTraceEvent
+// which is threadsafe internally via TraceLog::lock_. The enabled flag may
+// cause some threads to incorrectly call or skip calling AddTraceEvent near
+// the time of the system being enabled or disabled. This is acceptable as
+// we tolerate some data loss while the system is being enabled/disabled and
+// because AddTraceEvent is threadsafe internally and checks the enabled state
+// again under lock.
+//
+// Without the use of these static category pointers and enabled flags all
+// trace points would carry a significant performance cost of acquiring a lock
+// and resolving the category.
+
+#ifndef SkTraceEvent_DEFINED
+#define SkTraceEvent_DEFINED
+
+#include "SkEventTracer.h"
+
+// By default, const char* argument values are assumed to have long-lived scope
+// and will not be copied. Use this macro to force a const char* to be copied.
+#define TRACE_STR_COPY(str) \
+    skia::tracing_internals::TraceStringWithCopy(str)
+
+// By default, uint64 ID argument values are not mangled with the Process ID in
+// TRACE_EVENT_ASYNC macros. Use this macro to force Process ID mangling.
+#define TRACE_ID_MANGLE(id) \
+    skia::tracing_internals::TraceID::ForceMangle(id)
+
+// By default, pointers are mangled with the Process ID in TRACE_EVENT_ASYNC
+// macros. Use this macro to prevent Process ID mangling.
+#define TRACE_ID_DONT_MANGLE(id) \
+    skia::tracing_internals::TraceID::DontMangle(id)
+
+// Records a pair of begin and end events called "name" for the current
+// scope, with 0, 1 or 2 associated arguments. If the category is not
+// enabled, then this does nothing.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+#define TRACE_EVENT0(category_group, name) \
+    INTERNAL_TRACE_EVENT_ADD_SCOPED(category_group, name)
+#define TRACE_EVENT1(category_group, name, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_SCOPED(category_group, name, arg1_name, arg1_val)
+#define TRACE_EVENT2( \
+    category_group, name, arg1_name, arg1_val, arg2_name, arg2_val) \
+  INTERNAL_TRACE_EVENT_ADD_SCOPED( \
+      category_group, name, arg1_name, arg1_val, arg2_name, arg2_val)
+
+// Records events like TRACE_EVENT2 but uses |memory_tag| for memory tracing.
+// Use this where |name| is too generic to accurately aggregate allocations.
+#define TRACE_EVENT_WITH_MEMORY_TAG2( \
+    category, name, memory_tag, arg1_name, arg1_val, arg2_name, arg2_val) \
+  INTERNAL_TRACE_EVENT_ADD_SCOPED( \
+      category, name, arg1_name, arg1_val, arg2_name, arg2_val)
+
+// UNSHIPPED_TRACE_EVENT* are like TRACE_EVENT* except that they are not
+// included in official builds.
+
+#if OFFICIAL_BUILD
+#undef TRACING_IS_OFFICIAL_BUILD
+#define TRACING_IS_OFFICIAL_BUILD 1
+#elif !defined(TRACING_IS_OFFICIAL_BUILD)
+#define TRACING_IS_OFFICIAL_BUILD 0
+#endif
+
+#if TRACING_IS_OFFICIAL_BUILD
+#define UNSHIPPED_TRACE_EVENT0(category_group, name) (void)0
+#define UNSHIPPED_TRACE_EVENT1(category_group, name, arg1_name, arg1_val) \
+    (void)0
+#define UNSHIPPED_TRACE_EVENT2(category_group, name, arg1_name, arg1_val, \
+                               arg2_name, arg2_val) (void)0
+#define UNSHIPPED_TRACE_EVENT_INSTANT0(category_group, name, scope) (void)0
+#define UNSHIPPED_TRACE_EVENT_INSTANT1(category_group, name, scope, \
+                                       arg1_name, arg1_val) (void)0
+#define UNSHIPPED_TRACE_EVENT_INSTANT2(category_group, name, scope, \
+                                       arg1_name, arg1_val, \
+                                       arg2_name, arg2_val) (void)0
+#else
+#define UNSHIPPED_TRACE_EVENT0(category_group, name) \
+    TRACE_EVENT0(category_group, name)
+#define UNSHIPPED_TRACE_EVENT1(category_group, name, arg1_name, arg1_val) \
+    TRACE_EVENT1(category_group, name, arg1_name, arg1_val)
+#define UNSHIPPED_TRACE_EVENT2(category_group, name, arg1_name, arg1_val, \
+                               arg2_name, arg2_val) \
+    TRACE_EVENT2(category_group, name, arg1_name, arg1_val, arg2_name, arg2_val)
+#define UNSHIPPED_TRACE_EVENT_INSTANT0(category_group, name, scope) \
+    TRACE_EVENT_INSTANT0(category_group, name, scope)
+#define UNSHIPPED_TRACE_EVENT_INSTANT1(category_group, name, scope, \
+                                       arg1_name, arg1_val) \
+    TRACE_EVENT_INSTANT1(category_group, name, scope, arg1_name, arg1_val)
+#define UNSHIPPED_TRACE_EVENT_INSTANT2(category_group, name, scope, \
+                                       arg1_name, arg1_val, \
+                                       arg2_name, arg2_val) \
+    TRACE_EVENT_INSTANT2(category_group, name, scope, arg1_name, arg1_val, \
+                         arg2_name, arg2_val)
+#endif
+
+// Records a single event called "name" immediately, with 0, 1 or 2
+// associated arguments. If the category is not enabled, then this
+// does nothing.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+#define TRACE_EVENT_INSTANT0(category_group, name, scope) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_INSTANT, \
+        category_group, name, TRACE_EVENT_FLAG_NONE | scope)
+#define TRACE_EVENT_INSTANT1(category_group, name, scope, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_INSTANT, \
+        category_group, name, TRACE_EVENT_FLAG_NONE | scope, \
+        arg1_name, arg1_val)
+#define TRACE_EVENT_INSTANT2(category_group, name, scope, arg1_name, arg1_val, \
+                             arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_INSTANT, \
+        category_group, name, TRACE_EVENT_FLAG_NONE | scope, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+#define TRACE_EVENT_COPY_INSTANT0(category_group, name, scope) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_INSTANT, \
+        category_group, name, TRACE_EVENT_FLAG_COPY | scope)
+#define TRACE_EVENT_COPY_INSTANT1(category_group, name, scope, \
+                                  arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_INSTANT, \
+        category_group, name, TRACE_EVENT_FLAG_COPY | scope, arg1_name, \
+        arg1_val)
+#define TRACE_EVENT_COPY_INSTANT2(category_group, name, scope, \
+                                  arg1_name, arg1_val, \
+                                  arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_INSTANT, \
+        category_group, name, TRACE_EVENT_FLAG_COPY | scope, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+
+// Sets the current sample state to the given category and name (both must be
+// constant strings). These states are intended for a sampling profiler.
+// Implementation note: we store category and name together because we don't
+// want the inconsistency/expense of storing two pointers.
+// |thread_bucket| is [0..2] and is used to statically isolate samples in one
+// thread from others.
+#define TRACE_EVENT_SET_SAMPLING_STATE_FOR_BUCKET( \
+    bucket_number, category, name)                 \
+        skia::tracing_internals::                     \
+        TraceEventSamplingStateScope<bucket_number>::Set(category "\0" name)
+
+// Returns a current sampling state of the given bucket.
+#define TRACE_EVENT_GET_SAMPLING_STATE_FOR_BUCKET(bucket_number) \
+    skia::tracing_internals::TraceEventSamplingStateScope<bucket_number>::Current()
+
+// Creates a scope of a sampling state of the given bucket.
+//
+// {  // The sampling state is set within this scope.
+//    TRACE_EVENT_SAMPLING_STATE_SCOPE_FOR_BUCKET(0, "category", "name");
+//    ...;
+// }
+#define TRACE_EVENT_SCOPED_SAMPLING_STATE_FOR_BUCKET(                   \
+    bucket_number, category, name)                                      \
+    skia::tracing_internals::TraceEventSamplingStateScope<bucket_number>   \
+        traceEventSamplingScope(category "\0" name);
+
+// Syntactic sugars for the sampling tracing in the main thread.
+#define TRACE_EVENT_SCOPED_SAMPLING_STATE(category, name) \
+    TRACE_EVENT_SCOPED_SAMPLING_STATE_FOR_BUCKET(0, category, name)
+#define TRACE_EVENT_GET_SAMPLING_STATE() \
+    TRACE_EVENT_GET_SAMPLING_STATE_FOR_BUCKET(0)
+#define TRACE_EVENT_SET_SAMPLING_STATE(category, name) \
+    TRACE_EVENT_SET_SAMPLING_STATE_FOR_BUCKET(0, category, name)
+
+
+// Records a single BEGIN event called "name" immediately, with 0, 1 or 2
+// associated arguments. If the category is not enabled, then this
+// does nothing.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+#define TRACE_EVENT_BEGIN0(category_group, name) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_BEGIN, \
+        category_group, name, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_BEGIN1(category_group, name, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_BEGIN, \
+        category_group, name, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val)
+#define TRACE_EVENT_BEGIN2(category_group, name, arg1_name, arg1_val, \
+        arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_BEGIN, \
+        category_group, name, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val, \
+        arg2_name, arg2_val)
+#define TRACE_EVENT_COPY_BEGIN0(category_group, name) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_BEGIN, \
+        category_group, name, TRACE_EVENT_FLAG_COPY)
+#define TRACE_EVENT_COPY_BEGIN1(category_group, name, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_BEGIN, \
+        category_group, name, TRACE_EVENT_FLAG_COPY, arg1_name, arg1_val)
+#define TRACE_EVENT_COPY_BEGIN2(category_group, name, arg1_name, arg1_val, \
+        arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_BEGIN, \
+        category_group, name, TRACE_EVENT_FLAG_COPY, arg1_name, arg1_val, \
+        arg2_name, arg2_val)
+
+// Similar to TRACE_EVENT_BEGINx but with a custom |at| timestamp provided.
+// - |id| is used to match the _BEGIN event with the _END event.
+//   Events are considered to match if their category_group, name and id values
+//   all match. |id| must either be a pointer or an integer value up to 64 bits.
+//   If it's a pointer, the bits will be xored with a hash of the process ID so
+//   that the same pointer on two different processes will not collide.
+#define TRACE_EVENT_BEGIN_WITH_ID_TID_AND_TIMESTAMP0(category_group, \
+        name, id, thread_id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID_TID_AND_TIMESTAMP( \
+        TRACE_EVENT_PHASE_ASYNC_BEGIN, category_group, name, id, thread_id, \
+        timestamp, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_COPY_BEGIN_WITH_ID_TID_AND_TIMESTAMP0( \
+        category_group, name, id, thread_id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID_TID_AND_TIMESTAMP( \
+        TRACE_EVENT_PHASE_ASYNC_BEGIN, category_group, name, id, thread_id, \
+        timestamp, TRACE_EVENT_FLAG_COPY)
+
+// Records a single END event for "name" immediately. If the category
+// is not enabled, then this does nothing.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+#define TRACE_EVENT_END0(category_group, name) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_END, \
+        category_group, name, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_END1(category_group, name, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_END, \
+        category_group, name, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val)
+#define TRACE_EVENT_END2(category_group, name, arg1_name, arg1_val, \
+        arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_END, \
+        category_group, name, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val, \
+        arg2_name, arg2_val)
+#define TRACE_EVENT_COPY_END0(category_group, name) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_END, \
+        category_group, name, TRACE_EVENT_FLAG_COPY)
+#define TRACE_EVENT_COPY_END1(category_group, name, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_END, \
+        category_group, name, TRACE_EVENT_FLAG_COPY, arg1_name, arg1_val)
+#define TRACE_EVENT_COPY_END2(category_group, name, arg1_name, arg1_val, \
+        arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_END, \
+        category_group, name, TRACE_EVENT_FLAG_COPY, arg1_name, arg1_val, \
+        arg2_name, arg2_val)
+
+// Similar to TRACE_EVENT_ENDx but with a custom |at| timestamp provided.
+// - |id| is used to match the _BEGIN event with the _END event.
+//   Events are considered to match if their category_group, name and id values
+//   all match. |id| must either be a pointer or an integer value up to 64 bits.
+//   If it's a pointer, the bits will be xored with a hash of the process ID so
+//   that the same pointer on two different processes will not collide.
+#define TRACE_EVENT_END_WITH_ID_TID_AND_TIMESTAMP0(category_group, \
+        name, id, thread_id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID_TID_AND_TIMESTAMP( \
+        TRACE_EVENT_PHASE_ASYNC_END, category_group, name, id, thread_id, \
+        timestamp, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_COPY_END_WITH_ID_TID_AND_TIMESTAMP0( \
+        category_group, name, id, thread_id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID_TID_AND_TIMESTAMP( \
+        TRACE_EVENT_PHASE_ASYNC_END, category_group, name, id, thread_id, \
+        timestamp, TRACE_EVENT_FLAG_COPY)
+
+// Records the value of a counter called "name" immediately. Value
+// must be representable as a 32 bit integer.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+#define TRACE_COUNTER1(category_group, name, value) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, TRACE_EVENT_FLAG_NONE, \
+        "value", static_cast<int>(value))
+#define TRACE_COPY_COUNTER1(category_group, name, value) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, TRACE_EVENT_FLAG_COPY, \
+        "value", static_cast<int>(value))
+
+// Records the values of a multi-parted counter called "name" immediately.
+// The UI will treat value1 and value2 as parts of a whole, displaying their
+// values as a stacked-bar chart.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+#define TRACE_COUNTER2(category_group, name, value1_name, value1_val, \
+        value2_name, value2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, TRACE_EVENT_FLAG_NONE, \
+        value1_name, static_cast<int>(value1_val), \
+        value2_name, static_cast<int>(value2_val))
+#define TRACE_COPY_COUNTER2(category_group, name, value1_name, value1_val, \
+        value2_name, value2_val) \
+    INTERNAL_TRACE_EVENT_ADD(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, TRACE_EVENT_FLAG_COPY, \
+        value1_name, static_cast<int>(value1_val), \
+        value2_name, static_cast<int>(value2_val))
+
+// Records the value of a counter called "name" immediately. Value
+// must be representable as a 32 bit integer.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+// - |id| is used to disambiguate counters with the same name. It must either
+//   be a pointer or an integer value up to 64 bits. If it's a pointer, the bits
+//   will be xored with a hash of the process ID so that the same pointer on
+//   two different processes will not collide.
+#define TRACE_COUNTER_ID1(category_group, name, id, value) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, \
+        "value", static_cast<int>(value))
+#define TRACE_COPY_COUNTER_ID1(category_group, name, id, value) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        "value", static_cast<int>(value))
+
+// Records the values of a multi-parted counter called "name" immediately.
+// The UI will treat value1 and value2 as parts of a whole, displaying their
+// values as a stacked-bar chart.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+// - |id| is used to disambiguate counters with the same name. It must either
+//   be a pointer or an integer value up to 64 bits. If it's a pointer, the bits
+//   will be xored with a hash of the process ID so that the same pointer on
+//   two different processes will not collide.
+#define TRACE_COUNTER_ID2(category_group, name, id, value1_name, value1_val, \
+        value2_name, value2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, \
+        value1_name, static_cast<int>(value1_val), \
+        value2_name, static_cast<int>(value2_val))
+#define TRACE_COPY_COUNTER_ID2(category_group, name, id, value1_name, \
+        value1_val, value2_name, value2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_COUNTER, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        value1_name, static_cast<int>(value1_val), \
+        value2_name, static_cast<int>(value2_val))
+
+
+// Records a single ASYNC_BEGIN event called "name" immediately, with 0, 1 or 2
+// associated arguments. If the category is not enabled, then this
+// does nothing.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+// - |id| is used to match the ASYNC_BEGIN event with the ASYNC_END event. ASYNC
+//   events are considered to match if their category_group, name and id values
+//   all match. |id| must either be a pointer or an integer value up to 64 bits.
+//   If it's a pointer, the bits will be xored with a hash of the process ID so
+//   that the same pointer on two different processes will not collide.
+//
+// An asynchronous operation can consist of multiple phases. The first phase is
+// defined by the ASYNC_BEGIN calls. Additional phases can be defined using the
+// ASYNC_STEP_INTO or ASYNC_STEP_PAST macros. The ASYNC_STEP_INTO macro will
+// annotate the block following the call. The ASYNC_STEP_PAST macro will
+// annotate the block prior to the call. Note that any particular event must use
+// only STEP_INTO or STEP_PAST macros; they can not mix and match. When the
+// operation completes, call ASYNC_END.
+//
+// An ASYNC trace typically occurs on a single thread (if not, they will only be
+// drawn on the thread defined in the ASYNC_BEGIN event), but all events in that
+// operation must use the same |name| and |id|. Each step can have its own
+// args.
+#define TRACE_EVENT_ASYNC_BEGIN0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_ASYNC_BEGIN1(category_group, name, id, arg1_name, \
+        arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val)
+#define TRACE_EVENT_ASYNC_BEGIN2(category_group, name, id, arg1_name, \
+        arg1_val, arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+#define TRACE_EVENT_COPY_ASYNC_BEGIN0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY)
+#define TRACE_EVENT_COPY_ASYNC_BEGIN1(category_group, name, id, arg1_name, \
+        arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val)
+#define TRACE_EVENT_COPY_ASYNC_BEGIN2(category_group, name, id, arg1_name, \
+        arg1_val, arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+
+// Records a single ASYNC_STEP_INTO event for |step| immediately. If the
+// category is not enabled, then this does nothing. The |name| and |id| must
+// match the ASYNC_BEGIN event above. The |step| param identifies this step
+// within the async event. This should be called at the beginning of the next
+// phase of an asynchronous operation. The ASYNC_BEGIN event must not have any
+// ASYNC_STEP_PAST events.
+#define TRACE_EVENT_ASYNC_STEP_INTO0(category_group, name, id, step) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_STEP_INTO, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, "step", step)
+#define TRACE_EVENT_ASYNC_STEP_INTO1(category_group, name, id, step, \
+                                     arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_STEP_INTO, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, "step", step, \
+        arg1_name, arg1_val)
+
+// Records a single ASYNC_STEP_PAST event for |step| immediately. If the
+// category is not enabled, then this does nothing. The |name| and |id| must
+// match the ASYNC_BEGIN event above. The |step| param identifies this step
+// within the async event. This should be called at the beginning of the next
+// phase of an asynchronous operation. The ASYNC_BEGIN event must not have any
+// ASYNC_STEP_INTO events.
+#define TRACE_EVENT_ASYNC_STEP_PAST0(category_group, name, id, step) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_STEP_PAST, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, "step", step)
+#define TRACE_EVENT_ASYNC_STEP_PAST1(category_group, name, id, step, \
+                                     arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_STEP_PAST, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, "step", step, \
+        arg1_name, arg1_val)
+
+// Records a single ASYNC_END event for "name" immediately. If the category
+// is not enabled, then this does nothing.
+#define TRACE_EVENT_ASYNC_END0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_ASYNC_END1(category_group, name, id, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val)
+#define TRACE_EVENT_ASYNC_END2(category_group, name, id, arg1_name, arg1_val, \
+        arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+#define TRACE_EVENT_COPY_ASYNC_END0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY)
+#define TRACE_EVENT_COPY_ASYNC_END1(category_group, name, id, arg1_name, \
+        arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val)
+#define TRACE_EVENT_COPY_ASYNC_END2(category_group, name, id, arg1_name, \
+        arg1_val, arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_ASYNC_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+
+
+// Records a single FLOW_BEGIN event called "name" immediately, with 0, 1 or 2
+// associated arguments. If the category is not enabled, then this
+// does nothing.
+// - category and name strings must have application lifetime (statics or
+//   literals). They may not include " chars.
+// - |id| is used to match the FLOW_BEGIN event with the FLOW_END event. FLOW
+//   events are considered to match if their category_group, name and id values
+//   all match. |id| must either be a pointer or an integer value up to 64 bits.
+//   If it's a pointer, the bits will be xored with a hash of the process ID so
+//   that the same pointer on two different processes will not collide.
+// FLOW events are different from ASYNC events in how they are drawn by the
+// tracing UI. A FLOW defines asynchronous data flow, such as posting a task
+// (FLOW_BEGIN) and later executing that task (FLOW_END). Expect FLOWs to be
+// drawn as lines or arrows from FLOW_BEGIN scopes to FLOW_END scopes. Similar
+// to ASYNC, a FLOW can consist of multiple phases. The first phase is defined
+// by the FLOW_BEGIN calls. Additional phases can be defined using the FLOW_STEP
+// macros. When the operation completes, call FLOW_END. An async operation can
+// span threads and processes, but all events in that operation must use the
+// same |name| and |id|. Each event can have its own args.
+#define TRACE_EVENT_FLOW_BEGIN0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_FLOW_BEGIN1(category_group, name, id, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val)
+#define TRACE_EVENT_FLOW_BEGIN2(category_group, name, id, arg1_name, arg1_val, \
+        arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+#define TRACE_EVENT_COPY_FLOW_BEGIN0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY)
+#define TRACE_EVENT_COPY_FLOW_BEGIN1(category_group, name, id, arg1_name, \
+        arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val)
+#define TRACE_EVENT_COPY_FLOW_BEGIN2(category_group, name, id, arg1_name, \
+        arg1_val, arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_BEGIN, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+
+// Records a single FLOW_STEP event for |step| immediately. If the category
+// is not enabled, then this does nothing. The |name| and |id| must match the
+// FLOW_BEGIN event above. The |step| param identifies this step within the
+// async event. This should be called at the beginning of the next phase of an
+// asynchronous operation.
+#define TRACE_EVENT_FLOW_STEP0(category_group, name, id, step) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_STEP, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, "step", step)
+#define TRACE_EVENT_FLOW_STEP1(category_group, name, id, step, \
+        arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_STEP, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, "step", step, \
+        arg1_name, arg1_val)
+#define TRACE_EVENT_COPY_FLOW_STEP0(category_group, name, id, step) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_STEP, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, "step", step)
+#define TRACE_EVENT_COPY_FLOW_STEP1(category_group, name, id, step, \
+        arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_STEP, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, "step", step, \
+        arg1_name, arg1_val)
+
+// Records a single FLOW_END event for "name" immediately. If the category
+// is not enabled, then this does nothing.
+#define TRACE_EVENT_FLOW_END0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE)
+#define TRACE_EVENT_FLOW_END1(category_group, name, id, arg1_name, arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, arg1_name, arg1_val)
+#define TRACE_EVENT_FLOW_END2(category_group, name, id, arg1_name, arg1_val, \
+        arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_NONE, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+#define TRACE_EVENT_COPY_FLOW_END0(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY)
+#define TRACE_EVENT_COPY_FLOW_END1(category_group, name, id, arg1_name, \
+        arg1_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val)
+#define TRACE_EVENT_COPY_FLOW_END2(category_group, name, id, arg1_name, \
+        arg1_val, arg2_name, arg2_val) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_FLOW_END, \
+        category_group, name, id, TRACE_EVENT_FLAG_COPY, \
+        arg1_name, arg1_val, arg2_name, arg2_val)
+
+// Macros to track the life time and value of arbitrary client objects.
+// See also TraceTrackableObject.
+#define TRACE_EVENT_OBJECT_CREATED_WITH_ID(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_CREATE_OBJECT, \
+        category_group, name, TRACE_ID_DONT_MANGLE(id), TRACE_EVENT_FLAG_NONE)
+
+#define TRACE_EVENT_OBJECT_SNAPSHOT_WITH_ID(category_group, name, id, snapshot) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_SNAPSHOT_OBJECT, \
+        category_group, name, TRACE_ID_DONT_MANGLE(id), TRACE_EVENT_FLAG_NONE,\
+        "snapshot", snapshot)
+
+#define TRACE_EVENT_OBJECT_DELETED_WITH_ID(category_group, name, id) \
+    INTERNAL_TRACE_EVENT_ADD_WITH_ID(TRACE_EVENT_PHASE_DELETE_OBJECT, \
+        category_group, name, TRACE_ID_DONT_MANGLE(id), TRACE_EVENT_FLAG_NONE)
+
+#define INTERNAL_TRACE_EVENT_CATEGORY_GROUP_ENABLED_FOR_RECORDING_MODE() \
+    *INTERNAL_TRACE_EVENT_UID(category_group_enabled) & \
+        (SkEventTracer::kEnabledForRecording_CategoryGroupEnabledFlags | \
+         SkEventTracer::kEnabledForEventCallback_CategoryGroupEnabledFlags)
+
+// Macro to efficiently determine if a given category group is enabled.
+#define TRACE_EVENT_CATEGORY_GROUP_ENABLED(category_group, ret) \
+    do { \
+      INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO(category_group); \
+      if (INTERNAL_TRACE_EVENT_CATEGORY_GROUP_ENABLED_FOR_RECORDING_MODE()) { \
+        *ret = true; \
+      } else { \
+        *ret = false; \
+      } \
+    } while (0)
+
+// Macro to efficiently determine, through polling, if a new trace has begun.
+#define TRACE_EVENT_IS_NEW_TRACE(ret) \
+    do { \
+      static int INTERNAL_TRACE_EVENT_UID(lastRecordingNumber) = 0; \
+      int num_traces_recorded = TRACE_EVENT_API_GET_NUM_TRACES_RECORDED(); \
+      if (num_traces_recorded != -1 && \
+          num_traces_recorded != \
+          INTERNAL_TRACE_EVENT_UID(lastRecordingNumber)) { \
+        INTERNAL_TRACE_EVENT_UID(lastRecordingNumber) = \
+            num_traces_recorded; \
+        *ret = true; \
+      } else { \
+        *ret = false; \
+      } \
+    } while (0)
+
+////////////////////////////////////////////////////////////////////////////////
+// Implementation specific tracing API definitions.
+
+// Get a pointer to the enabled state of the given trace category. Only
+// long-lived literal strings should be given as the category group. The
+// returned pointer can be held permanently in a local static for example. If
+// the unsigned char is non-zero, tracing is enabled. If tracing is enabled,
+// TRACE_EVENT_API_ADD_TRACE_EVENT can be called. It's OK if tracing is disabled
+// between the load of the tracing state and the call to
+// TRACE_EVENT_API_ADD_TRACE_EVENT, because this flag only provides an early out
+// for best performance when tracing is disabled.
+// const uint8_t*
+//     TRACE_EVENT_API_GET_CATEGORY_GROUP_ENABLED(const char* category_group)
+#define TRACE_EVENT_API_GET_CATEGORY_GROUP_ENABLED \
+    SkEventTracer::GetInstance()->getCategoryGroupEnabled
+
+// Get the number of times traces have been recorded. This is used to implement
+// the TRACE_EVENT_IS_NEW_TRACE facility.
+// unsigned int TRACE_EVENT_API_GET_NUM_TRACES_RECORDED()
+#define TRACE_EVENT_API_GET_NUM_TRACES_RECORDED \
+    SkEventTracer::GetInstance()->getNumTracesRecorded
+
+// Add a trace event to the platform tracing system.
+// SkEventTracer::Handle TRACE_EVENT_API_ADD_TRACE_EVENT(
+//                    char phase,
+//                    const uint8_t* category_group_enabled,
+//                    const char* name,
+//                    uint64_t id,
+//                    int num_args,
+//                    const char** arg_names,
+//                    const uint8_t* arg_types,
+//                    const uint64_t* arg_values,
+//                    unsigned char flags)
+#define TRACE_EVENT_API_ADD_TRACE_EVENT \
+    SkEventTracer::GetInstance()->addTraceEvent
+
+// Set the duration field of a COMPLETE trace event.
+// void TRACE_EVENT_API_UPDATE_TRACE_EVENT_DURATION(
+//     const uint8_t* category_group_enabled,
+//     const char* name,
+//     SkEventTracer::Handle id)
+#define TRACE_EVENT_API_UPDATE_TRACE_EVENT_DURATION \
+    SkEventTracer::GetInstance()->updateTraceEventDuration
+
+// These operations are atomic in the Chrome tracing implementation
+// to cater to ARM's weak memory consistency; we're just doing read/
+// write here because it's not strictly needed for correctness.
+// So says Nat.
+// FIXME
+
+#define TRACE_EVENT_API_ATOMIC_WORD intptr_t
+#define TRACE_EVENT_API_ATOMIC_LOAD(var) (*(&var))
+#define TRACE_EVENT_API_ATOMIC_STORE(var, value) (var=value)
+
+// Defines visibility for classes in trace_event.h
+#define TRACE_EVENT_API_CLASS_EXPORT SK_API
+
+// The thread buckets for the sampling profiler.
+TRACE_EVENT_API_CLASS_EXPORT extern \
+    TRACE_EVENT_API_ATOMIC_WORD g_trace_state[3];
+
+#define TRACE_EVENT_API_THREAD_BUCKET(thread_bucket)                           \
+    g_trace_state[thread_bucket]
+
+////////////////////////////////////////////////////////////////////////////////
+
+// Implementation detail: trace event macros create temporary variables
+// to keep instrumentation overhead low. These macros give each temporary
+// variable a unique name based on the line number to prevent name collisions.
+#define INTERNAL_TRACE_EVENT_UID3(a,b) \
+    trace_event_unique_##a##b
+#define INTERNAL_TRACE_EVENT_UID2(a,b) \
+    INTERNAL_TRACE_EVENT_UID3(a,b)
+#define INTERNAL_TRACE_EVENT_UID(name_prefix) \
+    INTERNAL_TRACE_EVENT_UID2(name_prefix, __LINE__)
+
+// Implementation detail: internal macro to create static category.
+// No barriers are needed, because this code is designed to operate safely
+// even when the unsigned char* points to garbage data (which may be the case
+// on processors without cache coherency).
+#define INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO_CUSTOM_VARIABLES( \
+    category_group, atomic, category_group_enabled) \
+    category_group_enabled = \
+        reinterpret_cast<const uint8_t*>(TRACE_EVENT_API_ATOMIC_LOAD( \
+            atomic)); \
+    if (!category_group_enabled) { \
+      category_group_enabled = \
+          TRACE_EVENT_API_GET_CATEGORY_GROUP_ENABLED(category_group); \
+      TRACE_EVENT_API_ATOMIC_STORE(atomic, \
+          reinterpret_cast<TRACE_EVENT_API_ATOMIC_WORD>( \
+              category_group_enabled)); \
+    }
+
+#define INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO(category_group) \
+    static TRACE_EVENT_API_ATOMIC_WORD INTERNAL_TRACE_EVENT_UID(atomic) = 0; \
+    const uint8_t* INTERNAL_TRACE_EVENT_UID(category_group_enabled); \
+    INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO_CUSTOM_VARIABLES(category_group, \
+        INTERNAL_TRACE_EVENT_UID(atomic), \
+        INTERNAL_TRACE_EVENT_UID(category_group_enabled));
+
+// Implementation detail: internal macro to create static category and add
+// event if the category is enabled.
+#define INTERNAL_TRACE_EVENT_ADD(phase, category_group, name, flags, ...) \
+    do { \
+      INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO(category_group); \
+      if (INTERNAL_TRACE_EVENT_CATEGORY_GROUP_ENABLED_FOR_RECORDING_MODE()) { \
+        skia::tracing_internals::AddTraceEvent( \
+            phase, INTERNAL_TRACE_EVENT_UID(category_group_enabled), name, \
+            skia::tracing_internals::kNoEventId, flags, ##__VA_ARGS__); \
+      } \
+    } while (0)
+
+// Implementation detail: internal macro to create static category and add begin
+// event if the category is enabled. Also adds the end event when the scope
+// ends.
+#define INTERNAL_TRACE_EVENT_ADD_SCOPED(category_group, name, ...) \
+    INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO(category_group); \
+    skia::tracing_internals::ScopedTracer INTERNAL_TRACE_EVENT_UID(tracer); \
+    if (INTERNAL_TRACE_EVENT_CATEGORY_GROUP_ENABLED_FOR_RECORDING_MODE()) { \
+      SkEventTracer::Handle h = skia::tracing_internals::AddTraceEvent( \
+          TRACE_EVENT_PHASE_COMPLETE, \
+          INTERNAL_TRACE_EVENT_UID(category_group_enabled), \
+          name, skia::tracing_internals::kNoEventId, \
+          TRACE_EVENT_FLAG_NONE, ##__VA_ARGS__); \
+      INTERNAL_TRACE_EVENT_UID(tracer).Initialize( \
+          INTERNAL_TRACE_EVENT_UID(category_group_enabled), name, h); \
+    }
+
+// Implementation detail: internal macro to create static category and add
+// event if the category is enabled.
+#define INTERNAL_TRACE_EVENT_ADD_WITH_ID(phase, category_group, name, id, \
+                                         flags, ...) \
+    do { \
+      INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO(category_group); \
+      if (INTERNAL_TRACE_EVENT_CATEGORY_GROUP_ENABLED_FOR_RECORDING_MODE()) { \
+        unsigned char trace_event_flags = flags | TRACE_EVENT_FLAG_HAS_ID; \
+        skia::tracing_internals::TraceID trace_event_trace_id( \
+            id, &trace_event_flags); \
+        skia::tracing_internals::AddTraceEvent( \
+            phase, INTERNAL_TRACE_EVENT_UID(category_group_enabled), \
+            name, trace_event_trace_id.data(), trace_event_flags, \
+            ##__VA_ARGS__); \
+      } \
+    } while (0)
+
+// Implementation detail: internal macro to create static category and add
+// event if the category is enabled.
+#define INTERNAL_TRACE_EVENT_ADD_WITH_ID_TID_AND_TIMESTAMP(phase, \
+        category_group, name, id, thread_id, flags, ...) \
+    do { \
+      INTERNAL_TRACE_EVENT_GET_CATEGORY_INFO(category_group); \
+      if (INTERNAL_TRACE_EVENT_CATEGORY_GROUP_ENABLED_FOR_RECORDING_MODE()) { \
+        unsigned char trace_event_flags = flags | TRACE_EVENT_FLAG_HAS_ID; \
+        skia::tracing_internals::TraceID trace_event_trace_id( \
+            id, &trace_event_flags); \
+        skia::tracing_internals::AddTraceEventWithThreadIdAndTimestamp( \
+            phase, INTERNAL_TRACE_EVENT_UID(category_group_enabled), \
+            name, trace_event_trace_id.data(), \
+            thread_id, base::TimeTicks::FromInternalValue(timestamp), \
+            trace_event_flags, ##__VA_ARGS__); \
+      } \
+    } while (0)
+
+// Notes regarding the following definitions:
+// New values can be added and propagated to third party libraries, but existing
+// definitions must never be changed, because third party libraries may use old
+// definitions.
+
+// Phase indicates the nature of an event entry. E.g. part of a begin/end pair.
+#define TRACE_EVENT_PHASE_BEGIN    ('B')
+#define TRACE_EVENT_PHASE_END      ('E')
+#define TRACE_EVENT_PHASE_COMPLETE ('X')
+#define TRACE_EVENT_PHASE_INSTANT  ('i')
+#define TRACE_EVENT_PHASE_ASYNC_BEGIN ('S')
+#define TRACE_EVENT_PHASE_ASYNC_STEP_INTO  ('T')
+#define TRACE_EVENT_PHASE_ASYNC_STEP_PAST  ('p')
+#define TRACE_EVENT_PHASE_ASYNC_END   ('F')
+#define TRACE_EVENT_PHASE_FLOW_BEGIN ('s')
+#define TRACE_EVENT_PHASE_FLOW_STEP  ('t')
+#define TRACE_EVENT_PHASE_FLOW_END   ('f')
+#define TRACE_EVENT_PHASE_METADATA ('M')
+#define TRACE_EVENT_PHASE_COUNTER  ('C')
+#define TRACE_EVENT_PHASE_SAMPLE  ('P')
+#define TRACE_EVENT_PHASE_CREATE_OBJECT ('N')
+#define TRACE_EVENT_PHASE_SNAPSHOT_OBJECT ('O')
+#define TRACE_EVENT_PHASE_DELETE_OBJECT ('D')
+
+// Flags for changing the behavior of TRACE_EVENT_API_ADD_TRACE_EVENT.
+#define TRACE_EVENT_FLAG_NONE         (static_cast<unsigned char>(0))
+#define TRACE_EVENT_FLAG_COPY         (static_cast<unsigned char>(1 << 0))
+#define TRACE_EVENT_FLAG_HAS_ID       (static_cast<unsigned char>(1 << 1))
+#define TRACE_EVENT_FLAG_MANGLE_ID    (static_cast<unsigned char>(1 << 2))
+#define TRACE_EVENT_FLAG_SCOPE_OFFSET (static_cast<unsigned char>(1 << 3))
+
+#define TRACE_EVENT_FLAG_SCOPE_MASK   (static_cast<unsigned char>( \
+    TRACE_EVENT_FLAG_SCOPE_OFFSET | (TRACE_EVENT_FLAG_SCOPE_OFFSET << 1)))
+
+// Type values for identifying types in the TraceValue union.
+#define TRACE_VALUE_TYPE_BOOL         (static_cast<unsigned char>(1))
+#define TRACE_VALUE_TYPE_UINT         (static_cast<unsigned char>(2))
+#define TRACE_VALUE_TYPE_INT          (static_cast<unsigned char>(3))
+#define TRACE_VALUE_TYPE_DOUBLE       (static_cast<unsigned char>(4))
+#define TRACE_VALUE_TYPE_POINTER      (static_cast<unsigned char>(5))
+#define TRACE_VALUE_TYPE_STRING       (static_cast<unsigned char>(6))
+#define TRACE_VALUE_TYPE_COPY_STRING  (static_cast<unsigned char>(7))
+#define TRACE_VALUE_TYPE_CONVERTABLE  (static_cast<unsigned char>(8))
+
+// Enum reflecting the scope of an INSTANT event. Must fit within
+// TRACE_EVENT_FLAG_SCOPE_MASK.
+#define TRACE_EVENT_SCOPE_GLOBAL  (static_cast<unsigned char>(0 << 3))
+#define TRACE_EVENT_SCOPE_PROCESS (static_cast<unsigned char>(1 << 3))
+#define TRACE_EVENT_SCOPE_THREAD  (static_cast<unsigned char>(2 << 3))
+
+#define TRACE_EVENT_SCOPE_NAME_GLOBAL  ('g')
+#define TRACE_EVENT_SCOPE_NAME_PROCESS ('p')
+#define TRACE_EVENT_SCOPE_NAME_THREAD  ('t')
+
+namespace skia {
+namespace tracing_internals {
+
+// Specify these values when the corresponding argument of AddTraceEvent is not
+// used.
+const int kZeroNumArgs = 0;
+const uint64_t kNoEventId = 0;
+
+// TraceID encapsulates an ID that can either be an integer or pointer. Pointers
+// are by default mangled with the Process ID so that they are unlikely to
+// collide when the same pointer is used on different processes.
+class TraceID {
+ public:
+  class DontMangle {
+   public:
+    explicit DontMangle(const void* id)
+        : data_(static_cast<uint64_t>(
+              reinterpret_cast<unsigned long>(id))) {}
+    explicit DontMangle(uint64_t id) : data_(id) {}
+    explicit DontMangle(unsigned int id) : data_(id) {}
+    explicit DontMangle(unsigned short id) : data_(id) {}
+    explicit DontMangle(unsigned char id) : data_(id) {}
+    explicit DontMangle(long long id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit DontMangle(long id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit DontMangle(int id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit DontMangle(short id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit DontMangle(signed char id)
+        : data_(static_cast<uint64_t>(id)) {}
+    uint64_t data() const { return data_; }
+   private:
+    uint64_t data_;
+  };
+
+  class ForceMangle {
+   public:
+    explicit ForceMangle(uint64_t id) : data_(id) {}
+    explicit ForceMangle(unsigned int id) : data_(id) {}
+    explicit ForceMangle(unsigned short id) : data_(id) {}
+    explicit ForceMangle(unsigned char id) : data_(id) {}
+    explicit ForceMangle(long long id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit ForceMangle(long id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit ForceMangle(int id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit ForceMangle(short id)
+        : data_(static_cast<uint64_t>(id)) {}
+    explicit ForceMangle(signed char id)
+        : data_(static_cast<uint64_t>(id)) {}
+    uint64_t data() const { return data_; }
+   private:
+    uint64_t data_;
+  };
+
+  TraceID(const void* id, unsigned char* flags)
+      : data_(static_cast<uint64_t>(
+              reinterpret_cast<unsigned long>(id))) {
+    *flags |= TRACE_EVENT_FLAG_MANGLE_ID;
+  }
+  TraceID(ForceMangle id, unsigned char* flags) : data_(id.data()) {
+    *flags |= TRACE_EVENT_FLAG_MANGLE_ID;
+  }
+  TraceID(DontMangle id, unsigned char* flags) : data_(id.data()) {
+  }
+  TraceID(uint64_t id, unsigned char* flags)
+      : data_(id) { (void)flags; }
+  TraceID(unsigned int id, unsigned char* flags)
+      : data_(id) { (void)flags; }
+  TraceID(unsigned short id, unsigned char* flags)
+      : data_(id) { (void)flags; }
+  TraceID(unsigned char id, unsigned char* flags)
+      : data_(id) { (void)flags; }
+  TraceID(long long id, unsigned char* flags)
+      : data_(static_cast<uint64_t>(id)) { (void)flags; }
+  TraceID(long id, unsigned char* flags)
+      : data_(static_cast<uint64_t>(id)) { (void)flags; }
+  TraceID(int id, unsigned char* flags)
+      : data_(static_cast<uint64_t>(id)) { (void)flags; }
+  TraceID(short id, unsigned char* flags)
+      : data_(static_cast<uint64_t>(id)) { (void)flags; }
+  TraceID(signed char id, unsigned char* flags)
+      : data_(static_cast<uint64_t>(id)) { (void)flags; }
+
+  uint64_t data() const { return data_; }
+
+ private:
+  uint64_t data_;
+};
+
+// Simple union to store various types as uint64_t.
+union TraceValueUnion {
+  bool as_bool;
+  uint64_t as_uint;
+  long long as_int;
+  double as_double;
+  const void* as_pointer;
+  const char* as_string;
+};
+
+// Simple container for const char* that should be copied instead of retained.
+class TraceStringWithCopy {
+ public:
+  explicit TraceStringWithCopy(const char* str) : str_(str) {}
+  operator const char* () const { return str_; }
+ private:
+  const char* str_;
+};
+
+// Define SetTraceValue for each allowed type. It stores the type and
+// value in the return arguments. This allows this API to avoid declaring any
+// structures so that it is portable to third_party libraries.
+#define INTERNAL_DECLARE_SET_TRACE_VALUE(actual_type, \
+                                         union_member, \
+                                         value_type_id) \
+    static inline void SetTraceValue( \
+        actual_type arg, \
+        unsigned char* type, \
+        uint64_t* value) { \
+      TraceValueUnion type_value; \
+      type_value.union_member = arg; \
+      *type = value_type_id; \
+      *value = type_value.as_uint; \
+    }
+// Simpler form for int types that can be safely casted.
+#define INTERNAL_DECLARE_SET_TRACE_VALUE_INT(actual_type, \
+                                             value_type_id) \
+    static inline void SetTraceValue( \
+        actual_type arg, \
+        unsigned char* type, \
+        uint64_t* value) { \
+      *type = value_type_id; \
+      *value = static_cast<uint64_t>(arg); \
+    }
+
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(uint64_t, TRACE_VALUE_TYPE_UINT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(unsigned int, TRACE_VALUE_TYPE_UINT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(unsigned short, TRACE_VALUE_TYPE_UINT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(unsigned char, TRACE_VALUE_TYPE_UINT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(long long, TRACE_VALUE_TYPE_INT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(long, TRACE_VALUE_TYPE_INT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(int, TRACE_VALUE_TYPE_INT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(short, TRACE_VALUE_TYPE_INT)
+INTERNAL_DECLARE_SET_TRACE_VALUE_INT(signed char, TRACE_VALUE_TYPE_INT)
+INTERNAL_DECLARE_SET_TRACE_VALUE(bool, as_bool, TRACE_VALUE_TYPE_BOOL)
+INTERNAL_DECLARE_SET_TRACE_VALUE(double, as_double, TRACE_VALUE_TYPE_DOUBLE)
+INTERNAL_DECLARE_SET_TRACE_VALUE(const void*, as_pointer,
+                                 TRACE_VALUE_TYPE_POINTER)
+INTERNAL_DECLARE_SET_TRACE_VALUE(const char*, as_string,
+                                 TRACE_VALUE_TYPE_STRING)
+INTERNAL_DECLARE_SET_TRACE_VALUE(const TraceStringWithCopy&, as_string,
+                                 TRACE_VALUE_TYPE_COPY_STRING)
+
+#undef INTERNAL_DECLARE_SET_TRACE_VALUE
+#undef INTERNAL_DECLARE_SET_TRACE_VALUE_INT
+
+// These AddTraceEvent and AddTraceEvent template
+// functions are defined here instead of in the macro, because the arg_values
+// could be temporary objects, such as std::string. In order to store
+// pointers to the internal c_str and pass through to the tracing API,
+// the arg_values must live throughout these procedures.
+
+static inline SkEventTracer::Handle
+AddTraceEvent(
+    char phase,
+    const uint8_t* category_group_enabled,
+    const char* name,
+    uint64_t id,
+    unsigned char flags) {
+  return TRACE_EVENT_API_ADD_TRACE_EVENT(
+      phase, category_group_enabled, name, id,
+      kZeroNumArgs, NULL, NULL, NULL, flags);
+}
+
+template<class ARG1_TYPE>
+static inline SkEventTracer::Handle
+AddTraceEvent(
+    char phase,
+    const uint8_t* category_group_enabled,
+    const char* name,
+    uint64_t id,
+    unsigned char flags,
+    const char* arg1_name,
+    const ARG1_TYPE& arg1_val) {
+  const int num_args = 1;
+  uint8_t arg_types[1];
+  uint64_t arg_values[1];
+  SetTraceValue(arg1_val, &arg_types[0], &arg_values[0]);
+  return TRACE_EVENT_API_ADD_TRACE_EVENT(
+      phase, category_group_enabled, name, id,
+      num_args, &arg1_name, arg_types, arg_values, flags);
+}
+
+template<class ARG1_TYPE, class ARG2_TYPE>
+static inline SkEventTracer::Handle
+AddTraceEvent(
+    char phase,
+    const uint8_t* category_group_enabled,
+    const char* name,
+    uint64_t id,
+    unsigned char flags,
+    const char* arg1_name,
+    const ARG1_TYPE& arg1_val,
+    const char* arg2_name,
+    const ARG2_TYPE& arg2_val) {
+  const int num_args = 2;
+  const char* arg_names[2] = { arg1_name, arg2_name };
+  unsigned char arg_types[2];
+  uint64_t arg_values[2];
+  SetTraceValue(arg1_val, &arg_types[0], &arg_values[0]);
+  SetTraceValue(arg2_val, &arg_types[1], &arg_values[1]);
+  return TRACE_EVENT_API_ADD_TRACE_EVENT(
+      phase, category_group_enabled, name, id,
+      num_args, arg_names, arg_types, arg_values, flags);
+}
+
+// Used by TRACE_EVENTx macros. Do not use directly.
+class TRACE_EVENT_API_CLASS_EXPORT ScopedTracer {
+ public:
+  // Note: members of data_ intentionally left uninitialized. See Initialize.
+  ScopedTracer() : p_data_(NULL) {}
+
+  ~ScopedTracer() {
+    if (p_data_ && *data_.category_group_enabled)
+      TRACE_EVENT_API_UPDATE_TRACE_EVENT_DURATION(
+          data_.category_group_enabled, data_.name, data_.event_handle);
+  }
+
+  void Initialize(const uint8_t* category_group_enabled,
+                  const char* name,
+                  SkEventTracer::Handle event_handle) {
+    data_.category_group_enabled = category_group_enabled;
+    data_.name = name;
+    data_.event_handle = event_handle;
+    p_data_ = &data_;
+  }
+
+ private:
+  // This Data struct workaround is to avoid initializing all the members
+  // in Data during construction of this object, since this object is always
+  // constructed, even when tracing is disabled. If the members of Data were
+  // members of this class instead, compiler warnings occur about potential
+  // uninitialized accesses.
+  struct Data {
+    const uint8_t* category_group_enabled;
+    const char* name;
+    SkEventTracer::Handle event_handle;
+  };
+  Data* p_data_;
+  Data data_;
+};
+
+// Used by TRACE_EVENT_BINARY_EFFICIENTx macro. Do not use directly.
+class TRACE_EVENT_API_CLASS_EXPORT ScopedTraceBinaryEfficient {
+ public:
+  ScopedTraceBinaryEfficient(const char* category_group, const char* name);
+  ~ScopedTraceBinaryEfficient();
+
+ private:
+  const uint8_t* category_group_enabled_;
+  const char* name_;
+  SkEventTracer::Handle event_handle_;
+};
+
+// This macro generates less code then TRACE_EVENT0 but is also
+// slower to execute when tracing is off. It should generally only be
+// used with code that is seldom executed or conditionally executed
+// when debugging.
+// For now the category_group must be "gpu".
+#define TRACE_EVENT_BINARY_EFFICIENT0(category_group, name) \
+    skia::tracing_internals::ScopedTraceBinaryEfficient \
+        INTERNAL_TRACE_EVENT_UID(scoped_trace)(category_group, name);
+
+// TraceEventSamplingStateScope records the current sampling state
+// and sets a new sampling state. When the scope exists, it restores
+// the sampling state having recorded.
+template<size_t BucketNumber>
+class TraceEventSamplingStateScope {
+ public:
+  TraceEventSamplingStateScope(const char* category_and_name) {
+    previous_state_ = TraceEventSamplingStateScope<BucketNumber>::Current();
+    TraceEventSamplingStateScope<BucketNumber>::Set(category_and_name);
+  }
+
+  ~TraceEventSamplingStateScope() {
+    TraceEventSamplingStateScope<BucketNumber>::Set(previous_state_);
+  }
+
+  static inline const char* Current() {
+    return reinterpret_cast<const char*>(TRACE_EVENT_API_ATOMIC_LOAD(
+      g_trace_state[BucketNumber]));
+  }
+
+  static inline void Set(const char* category_and_name) {
+    TRACE_EVENT_API_ATOMIC_STORE(
+      g_trace_state[BucketNumber],
+      reinterpret_cast<TRACE_EVENT_API_ATOMIC_WORD>(
+        const_cast<char*>(category_and_name)));
+  }
+
+ private:
+  const char* previous_state_;
+};
+
+}  // namespace tracing_internals
+}  // namespace skia
+
+#endif
diff --git a/src/utils/SkEventTracer.cpp b/src/utils/SkEventTracer.cpp
new file mode 100644 (file)
index 0000000..e00c3e3
--- /dev/null
@@ -0,0 +1,55 @@
+/*
+ * Copyright 2014 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#include "SkEventTracer.h"
+#include "SkOnce.h"
+
+class SkDefaultEventTracer: public SkEventTracer {
+    virtual SkEventTracer::Handle 
+        addTraceEvent(char phase,
+                      const uint8_t* categoryEnabledFlag,
+                      const char* name,
+                      uint64_t id,
+                      int numArgs,
+                      const char** argNames,
+                      const uint8_t* argTypes,
+                      const uint64_t* argValues,
+                      uint8_t flags) SK_OVERRIDE { return 0; }
+    
+    virtual void 
+        updateTraceEventDuration(const uint8_t* categoryEnabledFlag, 
+                                 const char* name, 
+                                 SkEventTracer::Handle) SK_OVERRIDE {};
+
+    virtual const uint8_t* getCategoryGroupEnabled(const char* name) SK_OVERRIDE {
+        static uint8_t no = 0;
+        return &no;
+    };
+    virtual const char* getCategoryGroupName(
+      const uint8_t* category_group_enabled) SK_OVERRIDE {
+        static const char* dummy = "dummy";
+        return dummy;
+    };
+};
+
+SkEventTracer *SkEventTracer::gInstance;
+
+static void intiailize_default_tracer(int) {
+    SkEventTracer::SetInstance(SkNEW(SkDefaultEventTracer));
+}
+
+static void cleanup_tracer() {
+    // calling SetInstance will delete the existing instance.
+    SkEventTracer::SetInstance(NULL);
+}
+
+SkEventTracer* SkEventTracer::GetInstance() {
+    SK_DECLARE_STATIC_ONCE(once);
+    SkOnce(&once, intiailize_default_tracer, 0, cleanup_tracer);
+    SkASSERT(NULL != SkEventTracer::gInstance);
+    return SkEventTracer::gInstance;
+}
diff --git a/tests/TracingTest.cpp b/tests/TracingTest.cpp
new file mode 100644 (file)
index 0000000..730a46c
--- /dev/null
@@ -0,0 +1,13 @@
+/*
+ * Copyright 2014 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#include "SkTraceEvent.h"
+#include "Test.h"
+
+DEF_TEST(Tracing, reporter) {
+  TRACE_EVENT0("skia.testing", "just a test");
+}