diff options
Diffstat (limited to 'indra')
114 files changed, 5083 insertions, 2744 deletions
diff --git a/indra/cmake/Copy3rdPartyLibs.cmake b/indra/cmake/Copy3rdPartyLibs.cmake index b20d23cead..1c56f49486 100644 --- a/indra/cmake/Copy3rdPartyLibs.cmake +++ b/indra/cmake/Copy3rdPartyLibs.cmake @@ -104,6 +104,8 @@ if(WINDOWS) set(MSVC_VER 120) elseif (MSVC_VERSION GREATER_EQUAL 1910 AND MSVC_VERSION LESS 1920) # Visual Studio 2017 set(MSVC_VER 140) + elseif (MSVC_VERSION GREATER_EQUAL 1920 AND MSVC_VERSION LESS 1930) # Visual Studio 2019 + set(MSVC_VER 140) else (MSVC80) MESSAGE(WARNING "New MSVC_VERSION ${MSVC_VERSION} of MSVC: adapt Copy3rdPartyLibs.cmake") endif (MSVC80) diff --git a/indra/cmake/LLCommon.cmake b/indra/cmake/LLCommon.cmake index 8900419f9b..34499aaa36 100644 --- a/indra/cmake/LLCommon.cmake +++ b/indra/cmake/LLCommon.cmake @@ -3,12 +3,14 @@ include(APR) include(Boost) include(EXPAT) +include(Tracy) include(ZLIB) set(LLCOMMON_INCLUDE_DIRS ${LIBS_OPEN_DIR}/llcommon ${APRUTIL_INCLUDE_DIR} ${APR_INCLUDE_DIR} + ${TRACY_INCLUDE_DIR} ) set(LLCOMMON_SYSTEM_INCLUDE_DIRS ${Boost_INCLUDE_DIRS} @@ -30,7 +32,8 @@ else (LINUX) ${BOOST_FIBER_LIBRARY} ${BOOST_CONTEXT_LIBRARY} ${BOOST_THREAD_LIBRARY} - ${BOOST_SYSTEM_LIBRARY} ) + ${BOOST_SYSTEM_LIBRARY} + ) endif (LINUX) set(LLCOMMON_LINK_SHARED OFF CACHE BOOL "Build the llcommon target as a static library.") diff --git a/indra/cmake/Tracy.cmake b/indra/cmake/Tracy.cmake new file mode 100644 index 0000000000..1b8c3db2e2 --- /dev/null +++ b/indra/cmake/Tracy.cmake @@ -0,0 +1,29 @@ +# -*- cmake -*- +include(Prebuilt) + +set(USE_TRACY OFF CACHE BOOL "Use Tracy profiler.") + +if (USE_TRACY) + set(TRACY_INCLUDE_DIR ${LIBS_PREBUILT_DIR}/include/tracy) + +# See: indra/llcommon/llprofiler.h + add_definitions(-DLL_PROFILER_CONFIGURATION=2) + use_prebuilt_binary(tracy) + + if (WINDOWS) + MESSAGE(STATUS "Including Tracy for Windows: '${TRACY_INCLUDE_DIR}'") + endif (WINDOWS) + + if (DARWIN) + MESSAGE(STATUS "Including Tracy for Darwin: '${TRACY_INCLUDE_DIR}'") + endif (DARWIN) + + if (LINUX) + MESSAGE(STATUS "Including Tracy for Linux: '${TRACY_INCLUDE_DIR}'") + endif (LINUX) +else (USE_TRACY) + # Tracy.cmake should not set LLCOMMON_INCLUDE_DIRS, let LLCommon.cmake do that + set(TRACY_INCLUDE_DIR "") + set(TRACY_LIBRARY "") +endif (USE_TRACY) + diff --git a/indra/edit-me-to-trigger-new-build.txt b/indra/edit-me-to-trigger-new-build.txt index 5366987cff..eab7c17b71 100644 --- a/indra/edit-me-to-trigger-new-build.txt +++ b/indra/edit-me-to-trigger-new-build.txt @@ -1,3 +1,4 @@ euclid 5/29/2020 euclid 7/23/2020 -euclid 4/29/2021
\ No newline at end of file +euclid 4/29/2021 +euclid 10/5/2021 DRTVWR-546 diff --git a/indra/llcharacter/lljoint.cpp b/indra/llcharacter/lljoint.cpp index dee642310e..d72282ab42 100644 --- a/indra/llcharacter/lljoint.cpp +++ b/indra/llcharacter/lljoint.cpp @@ -922,6 +922,13 @@ const LLMatrix4 &LLJoint::getWorldMatrix() return mXform.getWorldMatrix(); } +const LLMatrix4a& LLJoint::getWorldMatrix4a() +{ + updateWorldMatrixParent(); + + return mWorldMatrix; +} + //-------------------------------------------------------------------- // setWorldMatrix() @@ -1003,6 +1010,7 @@ void LLJoint::updateWorldMatrix() { sNumUpdates++; mXform.updateMatrix(FALSE); + mWorldMatrix.loadu(mXform.getWorldMatrix()); mDirtyFlags = 0x0; } } diff --git a/indra/llcharacter/lljoint.h b/indra/llcharacter/lljoint.h index 1b646b641f..ba821667c7 100644 --- a/indra/llcharacter/lljoint.h +++ b/indra/llcharacter/lljoint.h @@ -38,6 +38,7 @@ #include "m4math.h" #include "llquaternion.h" #include "xform.h" +#include "llmatrix4a.h" const S32 LL_CHARACTER_MAX_JOINTS_PER_MESH = 15; // Need to set this to count of animate-able joints, @@ -123,6 +124,7 @@ protected: // explicit transformation members LLXformMatrix mXform; + LLMatrix4a mWorldMatrix; LLVector3 mDefaultPosition; LLVector3 mDefaultScale; @@ -259,6 +261,8 @@ public: const LLMatrix4 &getWorldMatrix(); void setWorldMatrix( const LLMatrix4& mat ); + const LLMatrix4a& getWorldMatrix4a(); + void updateWorldMatrixChildren(); void updateWorldMatrixParent(); diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 22dfe12e40..78d6ea3090 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -12,6 +12,7 @@ include(JsonCpp) include(Copy3rdPartyLibs) include(ZLIB) include(URIPARSER) +include(Tracy) include_directories( ${EXPAT_INCLUDE_DIRS} @@ -19,6 +20,7 @@ include_directories( ${JSONCPP_INCLUDE_DIR} ${ZLIB_INCLUDE_DIRS} ${URIPARSER_INCLUDE_DIRS} + ${TRACY_INCLUDE_DIR} ) # add_executable(lltreeiterators lltreeiterators.cpp) @@ -117,14 +119,16 @@ set(llcommon_SOURCE_FILES lluriparser.cpp lluuid.cpp llworkerthread.cpp - timing.cpp u64.cpp + threadpool.cpp + workqueue.cpp StackWalker.cpp ) set(llcommon_HEADER_FILES CMakeLists.txt + chrono.h ctype_workaround.h fix_macros.h indra_constants.h @@ -197,6 +201,7 @@ set(llcommon_HEADER_FILES llmortician.h llnametable.h llpointer.h + llprofiler.h llpounceable.h llpredicate.h llpreprocessor.h @@ -251,8 +256,12 @@ set(llcommon_HEADER_FILES lockstatic.h stdtypes.h stringize.h + threadpool.h + threadsafeschedule.h timer.h + tuple.h u64.h + workqueue.h StackWalker.h ) @@ -299,6 +308,7 @@ target_link_libraries( ${BOOST_SYSTEM_LIBRARY} ${GOOGLE_PERFTOOLS_LIBRARIES} ${URIPARSER_LIBRARIES} + ${TRACY_LIBRARY} ) if (DARWIN) @@ -355,6 +365,9 @@ if (LL_TESTS) LL_ADD_INTEGRATION_TEST(lluri "" "${test_libs}") LL_ADD_INTEGRATION_TEST(llunits "" "${test_libs}") LL_ADD_INTEGRATION_TEST(stringize "" "${test_libs}") + LL_ADD_INTEGRATION_TEST(threadsafeschedule "" "${test_libs}") + LL_ADD_INTEGRATION_TEST(tuple "" "${test_libs}") + LL_ADD_INTEGRATION_TEST(workqueue "" "${test_libs}") ## llexception_test.cpp isn't a regression test, and doesn't need to be run ## every build. It's to help a developer make implementation choices about diff --git a/indra/llcommon/chrono.h b/indra/llcommon/chrono.h new file mode 100644 index 0000000000..806e871892 --- /dev/null +++ b/indra/llcommon/chrono.h @@ -0,0 +1,65 @@ +/** + * @file chrono.h + * @author Nat Goodspeed + * @date 2021-10-05 + * @brief supplement <chrono> with utility functions + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_CHRONO_H) +#define LL_CHRONO_H + +#include <chrono> +#include <type_traits> // std::enable_if + +namespace LL +{ + +// time_point_cast() is derived from https://stackoverflow.com/a/35293183 +// without the iteration: we think errors in the ~1 microsecond range are +// probably acceptable. + +// This variant is for the optimal case when the source and dest use the same +// clock: that case is handled by std::chrono. +template <typename DestTimePoint, typename SrcTimePoint, + typename std::enable_if<std::is_same<typename DestTimePoint::clock, + typename SrcTimePoint::clock>::value, + bool>::type = true> +DestTimePoint time_point_cast(const SrcTimePoint& time) +{ + return std::chrono::time_point_cast<typename DestTimePoint::duration>(time); +} + +// This variant is for when the source and dest use different clocks -- see +// the linked StackOverflow answer, also Howard Hinnant's, for more context. +template <typename DestTimePoint, typename SrcTimePoint, + typename std::enable_if<! std::is_same<typename DestTimePoint::clock, + typename SrcTimePoint::clock>::value, + bool>::type = true> +DestTimePoint time_point_cast(const SrcTimePoint& time) +{ + // The basic idea is that we must adjust the passed time_point by the + // difference between the clocks' epochs. But since time_point doesn't + // expose its epoch, we fall back on what each of them thinks is now(). + // However, since we necessarily make sequential calls to those now() + // functions, the answers differ not only by the cycles spent executing + // those calls, but by potential OS interruptions between them. Try to + // reduce that error by capturing the source clock time both before and + // after the dest clock, and splitting the difference. Of course an + // interruption between two of these now() calls without a comparable + // interruption between the other two will skew the result, but better is + // more expensive. + const auto src_before = typename SrcTimePoint::clock::now(); + const auto dest_now = typename DestTimePoint::clock::now(); + const auto src_after = typename SrcTimePoint::clock::now(); + const auto src_diff = src_after - src_before; + const auto src_now = src_before + src_diff / 2; + return dest_now + (time - src_now); +} + +} // namespace LL + +#endif /* ! defined(LL_CHRONO_H) */ diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h index e5a913a6a9..a228fd22be 100644 --- a/indra/llcommon/linden_common.h +++ b/indra/llcommon/linden_common.h @@ -27,6 +27,14 @@ #ifndef LL_LINDEN_COMMON_H #define LL_LINDEN_COMMON_H +#include "llprofiler.h" +#if TRACY_ENABLE && !defined(LL_PROFILER_ENABLE_TRACY_OPENGL) // hooks for memory profiling +void *tracy_aligned_malloc(size_t size, size_t alignment); +void tracy_aligned_free(void *memblock); +#define _aligned_malloc(X, Y) tracy_aligned_malloc((X), (Y)) +#define _aligned_free(X) tracy_aligned_free((X)) +#endif + // *NOTE: Please keep includes here to a minimum! // // Files included here are included in every library .cpp file and diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index 96be913d17..5d4a623bf6 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -33,6 +33,47 @@ #include "lltracethreadrecorder.h" #include "llcleanup.h" +#if (TRACY_ENABLE) +// Override new/delete for tracy memory profiling +void *operator new(size_t size) +{ + auto ptr = (malloc) (size); + if (!ptr) + { + throw std::bad_alloc(); + } + TracyAlloc(ptr, size); + return ptr; +} + +void operator delete(void *ptr) noexcept +{ + TracyFree(ptr); + (free)(ptr); +} + +// C-style malloc/free can't be so easily overridden, so we define tracy versions and use +// a pre-processor #define in linden_common.h to redirect to them. The parens around the native +// functions below prevents recursive substitution by the preprocessor. +// +// Unaligned mallocs are rare in LL code but hooking them causes problems in 3p lib code (looking at +// you, Havok), so we'll only capture the aligned version. + +void *tracy_aligned_malloc(size_t size, size_t alignment) +{ + auto ptr = ll_aligned_malloc_fallback(size, alignment); + if (ptr) TracyAlloc(ptr, size); + return ptr; +} + +void tracy_aligned_free(void *memblock) +{ + TracyFree(memblock); + ll_aligned_free_fallback(memblock); +} + +#endif + //static BOOL LLCommon::sAprInitialized = FALSE; diff --git a/indra/llcommon/llcond.h b/indra/llcommon/llcond.h index e31b67d893..c08acb66a1 100644 --- a/indra/llcommon/llcond.h +++ b/indra/llcommon/llcond.h @@ -53,6 +53,8 @@ private: LLCoros::Mutex mMutex; // Use LLCoros::ConditionVariable for the same reason. LLCoros::ConditionVariable mCond; + using LockType = LLCoros::LockType; + using cv_status = LLCoros::cv_status; public: /// LLCond can be explicitly initialized with a specific value for mData if @@ -65,10 +67,14 @@ public: LLCond(const LLCond&) = delete; LLCond& operator=(const LLCond&) = delete; - /// get() returns a const reference to the stored DATA. The only way to - /// get a non-const reference -- to modify the stored DATA -- is via - /// update_one() or update_all(). - const value_type& get() const { return mData; } + /// get() returns the stored DATA by value -- so to use get(), DATA must + /// be copyable. The only way to get a non-const reference -- to modify + /// the stored DATA -- is via update_one() or update_all(). + value_type get() + { + LockType lk(mMutex); + return mData; + } /** * Pass update_one() an invocable accepting non-const (DATA&). The @@ -83,7 +89,7 @@ public: void update_one(MODIFY modify) { { // scope of lock can/should end before notify_one() - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); modify(mData); } mCond.notify_one(); @@ -102,7 +108,7 @@ public: void update_all(MODIFY modify) { { // scope of lock can/should end before notify_all() - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); modify(mData); } mCond.notify_all(); @@ -118,7 +124,7 @@ public: template <typename Pred> void wait(Pred pred) { - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); // We must iterate explicitly since the predicate accepted by // condition_variable::wait() requires a different signature: // condition_variable::wait() calls its predicate with no arguments. @@ -205,14 +211,14 @@ private: template <typename Clock, typename Duration, typename Pred> bool wait_until(const std::chrono::time_point<Clock, Duration>& timeout_time, Pred pred) { - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); // We advise the caller to pass a predicate accepting (const DATA&). // But what if they instead pass a predicate accepting non-const // (DATA&)? Such a predicate could modify mData, which would be Bad. // Forbid that. while (! pred(const_cast<const value_type&>(mData))) { - if (LLCoros::cv_status::timeout == mCond.wait_until(lk, timeout_time)) + if (cv_status::timeout == mCond.wait_until(lk, timeout_time)) { // It's possible that wait_until() timed out AND the predicate // became true more or less simultaneously. Even though diff --git a/indra/llcommon/llerror.cpp b/indra/llcommon/llerror.cpp index 8355df9045..f7af181927 100644 --- a/indra/llcommon/llerror.cpp +++ b/indra/llcommon/llerror.cpp @@ -109,6 +109,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED int syslogPriority = LOG_CRIT; switch (level) { case LLError::LEVEL_DEBUG: syslogPriority = LOG_DEBUG; break; @@ -166,6 +167,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED if (LLError::getAlwaysFlush()) { mFile << message << std::endl; @@ -208,6 +210,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED static std::string s_ansi_error = createANSI("31"); // red static std::string s_ansi_warn = createANSI("34"); // blue static std::string s_ansi_debug = createANSI("35"); // magenta @@ -220,7 +223,8 @@ namespace { } else { - fprintf(stderr, "%s\n", message.c_str()); + LL_PROFILE_ZONE_NAMED("fprintf"); + fprintf(stderr, "%s\n", message.c_str()); } } @@ -229,6 +233,7 @@ namespace { LL_FORCE_INLINE void writeANSI(const std::string& ansi_code, const std::string& message) { + LL_PROFILE_ZONE_SCOPED static std::string s_ansi_bold = createANSI("1"); // bold static std::string s_ansi_reset = createANSI("0"); // reset // ANSI color code escape sequence, message, and reset in one fprintf call @@ -265,6 +270,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED mBuffer->addLine(message); } @@ -291,6 +297,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED debugger_print(message); } }; @@ -1178,6 +1185,7 @@ namespace void writeToRecorders(const LLError::CallSite& site, const std::string& message) { + LL_PROFILE_ZONE_SCOPED LLError::ELevel level = site.mLevel; LLError::SettingsConfigPtr s = LLError::Settings::getInstance()->getSettingsConfig(); @@ -1311,6 +1319,7 @@ namespace LLError bool Log::shouldLog(CallSite& site) { + LL_PROFILE_ZONE_SCOPED LLMutexTrylock lock(getMutex<LOG_MUTEX>(), 5); if (!lock.isLocked()) { @@ -1354,6 +1363,7 @@ namespace LLError void Log::flush(const std::ostringstream& out, const CallSite& site) { + LL_PROFILE_ZONE_SCOPED LLMutexTrylock lock(getMutex<LOG_MUTEX>(),5); if (!lock.isLocked()) { diff --git a/indra/llcommon/llerror.h b/indra/llcommon/llerror.h index d439136ca8..d06c0e2132 100644 --- a/indra/llcommon/llerror.h +++ b/indra/llcommon/llerror.h @@ -35,7 +35,9 @@ #include "stdtypes.h" +#include "llprofiler.h" #include "llpreprocessor.h" + #include <boost/static_assert.hpp> const int LL_ERR_NOERR = 0; @@ -348,7 +350,8 @@ typedef LLError::NoClassInfo _LL_CLASS_TO_LOG; // if (condition) LL_INFOS() << "True" << LL_ENDL; else LL_INFOS()() << "False" << LL_ENDL; #define lllog(level, once, ...) \ - do { \ + do { \ + LL_PROFILE_ZONE_NAMED("lllog"); \ const char* tags[] = {"", ##__VA_ARGS__}; \ static LLError::CallSite _site(lllog_site_args_(level, once, tags)); \ lllog_test_() diff --git a/indra/llcommon/llerrorcontrol.h b/indra/llcommon/llerrorcontrol.h index e87bb7bf35..57f10b7895 100644 --- a/indra/llcommon/llerrorcontrol.h +++ b/indra/llcommon/llerrorcontrol.h @@ -190,6 +190,7 @@ namespace LLError {} void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED mCallable(level, message); } private: diff --git a/indra/llcommon/lleventfilter.h b/indra/llcommon/lleventfilter.h index 48c2570732..7613850fb2 100644 --- a/indra/llcommon/lleventfilter.h +++ b/indra/llcommon/lleventfilter.h @@ -429,6 +429,8 @@ public: // path, then stores it to mTarget. virtual bool post(const LLSD& event) { + LL_PROFILE_ZONE_SCOPED + // Extract the element specified by 'mPath' from 'event'. To perform a // generic type-appropriate store through mTarget, construct an // LLSDParam<T> and store that, thus engaging LLSDParam's custom diff --git a/indra/llcommon/llfasttimer.h b/indra/llcommon/llfasttimer.h index dfc63d08a2..9bd93d7240 100644 --- a/indra/llcommon/llfasttimer.h +++ b/indra/llcommon/llfasttimer.h @@ -38,7 +38,10 @@ #define LL_FAST_TIMER_ON 1 #define LL_FASTTIMER_USE_RDTSC 1 +// NOTE: Also see llprofiler.h +#if !defined(LL_PROFILER_CONFIGURATION) #define LL_RECORD_BLOCK_TIME(timer_stat) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(timer_stat)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); +#endif // LL_PROFILER_CONFIGURATION namespace LLTrace { diff --git a/indra/llcommon/llframetimer.cpp b/indra/llcommon/llframetimer.cpp index 1e9920746b..c54029e8b4 100644 --- a/indra/llcommon/llframetimer.cpp +++ b/indra/llcommon/llframetimer.cpp @@ -29,6 +29,11 @@ #include "llframetimer.h" +// We don't bother building a stand alone lib; we just need to include the one source file for Tracy support +#if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #include "TracyClient.cpp" +#endif // LL_PROFILER_CONFIGURATION + // Static members //LLTimer LLFrameTimer::sInternalTimer; U64 LLFrameTimer::sStartTotalTime = totalTime(); diff --git a/indra/llcommon/llinstancetracker.h b/indra/llcommon/llinstancetracker.h index 402333cca7..02535a59e7 100644 --- a/indra/llcommon/llinstancetracker.h +++ b/indra/llcommon/llinstancetracker.h @@ -83,13 +83,34 @@ class LLInstanceTracker typedef llthread::LockStatic<StaticData> LockStatic; public: + using ptr_t = std::shared_ptr<T>; + using weak_t = std::weak_ptr<T>; + + /** + * Storing a dumb T* somewhere external is a bad idea, since + * LLInstanceTracker subclasses are explicitly destroyed rather than + * managed by smart pointers. It's legal to declare stack instances of an + * LLInstanceTracker subclass. But it's reasonable to store a + * std::weak_ptr<T>, which will become invalid when the T instance is + * destroyed. + */ + weak_t getWeak() + { + return mSelf; + } + + static S32 instanceCount() + { + return LockStatic()->mMap.size(); + } + // snapshot of std::pair<const KEY, std::shared_ptr<T>> pairs class snapshot { // It's very important that what we store in this snapshot are // weak_ptrs, NOT shared_ptrs. That's how we discover whether any // instance has been deleted during the lifespan of a snapshot. - typedef std::vector<std::pair<const KEY, std::weak_ptr<T>>> VectorType; + typedef std::vector<std::pair<const KEY, weak_t>> VectorType; // Dereferencing our iterator produces a std::shared_ptr for each // instance that still exists. Since we store weak_ptrs, that involves // two chained transformations: @@ -98,7 +119,7 @@ public: // It is very important that we filter lazily, that is, during // traversal. Any one of our stored weak_ptrs might expire during // traversal. - typedef std::pair<const KEY, std::shared_ptr<T>> strong_pair; + typedef std::pair<const KEY, ptr_t> strong_pair; // Note for future reference: nat has not yet had any luck (up to // Boost 1.67) trying to use boost::transform_iterator with a hand- // coded functor, only with actual functions. In my experience, an @@ -202,17 +223,12 @@ public: iterator end() { return iterator(snapshot::end(), key_getter); } }; - static T* getInstance(const KEY& k) + static ptr_t getInstance(const KEY& k) { LockStatic lock; const InstanceMap& map(lock->mMap); typename InstanceMap::const_iterator found = map.find(k); - return (found == map.end()) ? NULL : found->second.get(); - } - - static S32 instanceCount() - { - return LockStatic()->mMap.size(); + return (found == map.end()) ? NULL : found->second; } protected: @@ -222,7 +238,9 @@ protected: // shared_ptr, so give it a no-op deleter. We store shared_ptrs in our // InstanceMap specifically so snapshot can store weak_ptrs so we can // detect deletions during traversals. - std::shared_ptr<T> ptr(static_cast<T*>(this), [](T*){}); + ptr_t ptr(static_cast<T*>(this), [](T*){}); + // save corresponding weak_ptr for future reference + mSelf = ptr; LockStatic lock; add_(lock, key, ptr); } @@ -257,7 +275,7 @@ private: static std::string report(const char* key) { return report(std::string(key)); } // caller must instantiate LockStatic - void add_(LockStatic& lock, const KEY& key, const std::shared_ptr<T>& ptr) + void add_(LockStatic& lock, const KEY& key, const ptr_t& ptr) { mInstanceKey = key; InstanceMap& map = lock->mMap; @@ -281,7 +299,7 @@ private: break; } } - std::shared_ptr<T> remove_(LockStatic& lock) + ptr_t remove_(LockStatic& lock) { InstanceMap& map = lock->mMap; typename InstanceMap::iterator iter = map.find(mInstanceKey); @@ -295,6 +313,9 @@ private: } private: + // Storing a weak_ptr to self is a bit like deriving from + // std::enable_shared_from_this(), except more explicit. + weak_t mSelf; KEY mInstanceKey; }; @@ -326,6 +347,9 @@ class LLInstanceTracker<T, void, KEY_COLLISION_BEHAVIOR> typedef llthread::LockStatic<StaticData> LockStatic; public: + using ptr_t = std::shared_ptr<T>; + using weak_t = std::weak_ptr<T>; + /** * Storing a dumb T* somewhere external is a bad idea, since * LLInstanceTracker subclasses are explicitly destroyed rather than @@ -334,12 +358,15 @@ public: * std::weak_ptr<T>, which will become invalid when the T instance is * destroyed. */ - std::weak_ptr<T> getWeak() + weak_t getWeak() { return mSelf; } - static S32 instanceCount() { return LockStatic()->mSet.size(); } + static S32 instanceCount() + { + return LockStatic()->mSet.size(); + } // snapshot of std::shared_ptr<T> pointers class snapshot @@ -347,7 +374,7 @@ public: // It's very important that what we store in this snapshot are // weak_ptrs, NOT shared_ptrs. That's how we discover whether any // instance has been deleted during the lifespan of a snapshot. - typedef std::vector<std::weak_ptr<T>> VectorType; + typedef std::vector<weak_t> VectorType; // Dereferencing our iterator produces a std::shared_ptr for each // instance that still exists. Since we store weak_ptrs, that involves // two chained transformations: @@ -453,7 +480,7 @@ protected: private: // Storing a weak_ptr to self is a bit like deriving from // std::enable_shared_from_this(), except more explicit. - std::weak_ptr<T> mSelf; + weak_t mSelf; }; #endif diff --git a/indra/llcommon/llleaplistener.cpp b/indra/llcommon/llleaplistener.cpp index 3e6ce9092c..11bfec1b31 100644 --- a/indra/llcommon/llleaplistener.cpp +++ b/indra/llcommon/llleaplistener.cpp @@ -220,7 +220,7 @@ void LLLeapListener::getAPI(const LLSD& request) const { Response reply(LLSD(), request); - LLEventAPI* found = LLEventAPI::getInstance(request["api"]); + auto found = LLEventAPI::getInstance(request["api"]); if (found) { reply["name"] = found->getName(); diff --git a/indra/llcommon/llmemory.cpp b/indra/llcommon/llmemory.cpp index ea84e4c1ea..849867586a 100644 --- a/indra/llcommon/llmemory.cpp +++ b/indra/llcommon/llmemory.cpp @@ -82,6 +82,7 @@ void LLMemory::initMaxHeapSizeGB(F32Gigabytes max_heap_size) //static void LLMemory::updateMemoryInfo() { + LL_PROFILE_ZONE_SCOPED #if LL_WINDOWS PROCESS_MEMORY_COUNTERS counters; @@ -145,6 +146,7 @@ void* LLMemory::tryToAlloc(void* address, U32 size) //static void LLMemory::logMemoryInfo(BOOL update) { + LL_PROFILE_ZONE_SCOPED if(update) { updateMemoryInfo() ; diff --git a/indra/llcommon/llmutex.cpp b/indra/llcommon/llmutex.cpp index 4d73c04d07..a49002b5dc 100644 --- a/indra/llcommon/llmutex.cpp +++ b/indra/llcommon/llmutex.cpp @@ -44,6 +44,7 @@ LLMutex::~LLMutex() void LLMutex::lock() { + LL_PROFILE_ZONE_SCOPED if(isSelfLocked()) { //redundant lock mCount++; @@ -65,6 +66,7 @@ void LLMutex::lock() void LLMutex::unlock() { + LL_PROFILE_ZONE_SCOPED if (mCount > 0) { //not the root unlock mCount--; @@ -85,6 +87,7 @@ void LLMutex::unlock() bool LLMutex::isLocked() { + LL_PROFILE_ZONE_SCOPED if (!mMutex.try_lock()) { return true; @@ -108,6 +111,7 @@ LLThread::id_t LLMutex::lockingThread() const bool LLMutex::trylock() { + LL_PROFILE_ZONE_SCOPED if(isSelfLocked()) { //redundant lock mCount++; @@ -146,17 +150,20 @@ LLCondition::~LLCondition() void LLCondition::wait() { + LL_PROFILE_ZONE_SCOPED std::unique_lock< std::mutex > lock(mMutex); mCond.wait(lock); } void LLCondition::signal() { + LL_PROFILE_ZONE_SCOPED mCond.notify_one(); } void LLCondition::broadcast() { + LL_PROFILE_ZONE_SCOPED mCond.notify_all(); } @@ -166,6 +173,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex) : mMutex(mutex), mLocked(false) { + LL_PROFILE_ZONE_SCOPED if (mMutex) mLocked = mMutex->trylock(); } @@ -174,6 +182,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms) : mMutex(mutex), mLocked(false) { + LL_PROFILE_ZONE_SCOPED if (!mMutex) return; @@ -188,6 +197,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms) LLMutexTrylock::~LLMutexTrylock() { + LL_PROFILE_ZONE_SCOPED if (mMutex && mLocked) mMutex->unlock(); } @@ -199,6 +209,7 @@ LLMutexTrylock::~LLMutexTrylock() // LLScopedLock::LLScopedLock(std::mutex* mutex) : mMutex(mutex) { + LL_PROFILE_ZONE_SCOPED if(mutex) { mutex->lock(); @@ -217,6 +228,7 @@ LLScopedLock::~LLScopedLock() void LLScopedLock::unlock() { + LL_PROFILE_ZONE_SCOPED if(mLocked) { mMutex->unlock(); diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h new file mode 100644 index 0000000000..49510df913 --- /dev/null +++ b/indra/llcommon/llprofiler.h @@ -0,0 +1,103 @@ +/** + * @file llprofiler.h + * @brief Wrapper for Tracy and/or other profilers + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2021, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_PROFILER_H +#define LL_PROFILER_H + +#define LL_PROFILER_CONFIG_NONE 0 // No profiling +#define LL_PROFILER_CONFIG_FAST_TIMER 1 // Profiling on: Only Fast Timers +#define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy +#define LL_PROFILER_CONFIG_TRACY_FAST_TIMER 3 // Profiling on: Fast Timers + Tracy + +#ifndef LL_PROFILER_CONFIGURATION +#define LL_PROFILER_CONFIGURATION LL_PROFILER_CONFIG_FAST_TIMER +#endif + +#if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #define TRACY_ENABLE 1 +// Normally these would be enabled but we want to be able to build any viewer with Tracy enabled and run the Tracy server on another machine +// They must be undefined in order to work across multiple machines +// #define TRACY_NO_BROADCAST 1 +// #define TRACY_ONLY_LOCALHOST 1 + #define TRACY_ONLY_IPV4 1 + #include "Tracy.hpp" + + // Mutually exclusive with detailed memory tracing + #define LL_PROFILER_ENABLE_TRACY_OPENGL 0 + #endif + + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneScoped // Want descriptive names; was: ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, name, true ); + #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB + #define LL_PROFILE_ZONE_SCOPED ZoneScoped + + #define LL_PROFILE_ZONE_NUM( val ) ZoneValue( val ) + #define LL_PROFILE_ZONE_TEXT( text, size ) ZoneText( text, size ) + + #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow + #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan + #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red + #endif + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER + #define LL_PROFILER_FRAME_END + #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) + #define LL_RECORD_BLOCK_TIME(name) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_PROFILE_ZONE_NAMED(name) // LL_PROFILE_ZONE_NAMED is a no-op when Tracy is disabled + #define LL_PROFILE_ZONE_SCOPED // LL_PROFILE_ZONE_SCOPED is a no-op when Tracy is disabled + #define LL_PROFILE_ZONE_COLOR(name,color) // LL_RECORD_BLOCK_TIME(name) + + #define LL_PROFILE_ZONE_NUM( val ) (void)( val ); // Not supported + #define LL_PROFILE_ZONE_TEXT( text, size ) (void)( text ); void( size ); // Not supported + + #define LL_PROFILE_ZONE_ERR(name) (void)(name); // Not supported + #define LL_PROFILE_ZONE_INFO(name) (void)(name); // Not supported + #define LL_PROFILE_ZONE_WARN(name) (void)(name); // Not supported + #endif + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneScoped const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB + #define LL_PROFILE_ZONE_SCOPED ZoneScoped + + #define LL_PROFILE_ZONE_NUM( val ) ZoneValue( val ) + #define LL_PROFILE_ZONE_TEXT( text, size ) ZoneText( text, size ) + + #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow + #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan + #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red + #endif +#else + #define LL_PROFILER_FRAME_END + #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) +#endif // LL_PROFILER + +#endif // LL_PROFILER_H diff --git a/indra/llcommon/llsdutil.cpp b/indra/llcommon/llsdutil.cpp index eb3a96b133..c2fe15e9b7 100644 --- a/indra/llcommon/llsdutil.cpp +++ b/indra/llcommon/llsdutil.cpp @@ -214,6 +214,8 @@ BOOL compare_llsd_with_template( const LLSD& template_llsd, LLSD& resultant_llsd) { + LL_PROFILE_ZONE_SCOPED + if ( llsd_to_test.isUndefined() && template_llsd.isDefined() ) @@ -335,6 +337,8 @@ bool filter_llsd_with_template( const LLSD & template_llsd, LLSD & resultant_llsd) { + LL_PROFILE_ZONE_SCOPED + if (llsd_to_test.isUndefined() && template_llsd.isDefined()) { resultant_llsd = template_llsd; @@ -529,6 +533,8 @@ class TypeLookup public: TypeLookup() { + LL_PROFILE_ZONE_SCOPED + for (const Data *di(boost::begin(typedata)), *dend(boost::end(typedata)); di != dend; ++di) { mMap[di->type] = di->name; @@ -537,6 +543,8 @@ public: std::string lookup(LLSD::Type type) const { + LL_PROFILE_ZONE_SCOPED + MapType::const_iterator found = mMap.find(type); if (found != mMap.end()) { @@ -587,6 +595,8 @@ static std::string match_types(LLSD::Type expect, // prototype.type() LLSD::Type actual, // type we're checking const std::string& pfx) // as for llsd_matches { + LL_PROFILE_ZONE_SCOPED + // Trivial case: if the actual type is exactly what we expect, we're good. if (actual == expect) return ""; @@ -624,6 +634,8 @@ static std::string match_types(LLSD::Type expect, // prototype.type() // see docstring in .h file std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::string& pfx) { + LL_PROFILE_ZONE_SCOPED + // An undefined prototype means that any data is valid. // An undefined slot in an array or map prototype means that any data // may fill that slot. @@ -756,6 +768,8 @@ std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::str bool llsd_equals(const LLSD& lhs, const LLSD& rhs, int bits) { + LL_PROFILE_ZONE_SCOPED + // We're comparing strict equality of LLSD representation rather than // performing any conversions. So if the types aren't equal, the LLSD // values aren't equal. @@ -864,6 +878,8 @@ namespace llsd LLSD& drill(LLSD& blob, const LLSD& rawPath) { + LL_PROFILE_ZONE_SCOPED + // Treat rawPath uniformly as an array. If it's not already an array, // store it as the only entry in one. (But let's say Undefined means an // empty array.) @@ -889,6 +905,8 @@ LLSD& drill(LLSD& blob, const LLSD& rawPath) // path entry that's bad. for (LLSD::Integer i = 0; i < path.size(); ++i) { + LL_PROFILE_ZONE_NUM( i ) + const LLSD& key{path[i]}; if (key.isString()) { @@ -917,6 +935,8 @@ LLSD& drill(LLSD& blob, const LLSD& rawPath) LLSD drill(const LLSD& blob, const LLSD& path) { + LL_PROFILE_ZONE_SCOPED + // non-const drill() does exactly what we want. Temporarily cast away // const-ness and use that. return drill(const_cast<LLSD&>(blob), path); @@ -929,6 +949,8 @@ LLSD drill(const LLSD& blob, const LLSD& path) // filter may be include to exclude/include keys in a map. LLSD llsd_clone(LLSD value, LLSD filter) { + LL_PROFILE_ZONE_SCOPED + LLSD clone; bool has_filter(filter.isMap()); diff --git a/indra/llcommon/llsys.cpp b/indra/llcommon/llsys.cpp index 4e61fb8a58..6d5d043e8d 100644 --- a/indra/llcommon/llsys.cpp +++ b/indra/llcommon/llsys.cpp @@ -861,6 +861,7 @@ LLSD LLMemoryInfo::getStatsMap() const LLMemoryInfo& LLMemoryInfo::refresh() { + LL_PROFILE_ZONE_SCOPED mStatsMap = loadStatsMap(); LL_DEBUGS("LLMemoryInfo") << "Populated mStatsMap:\n"; diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp index 6d531d842d..11f5a015f1 100644 --- a/indra/llcommon/llthread.cpp +++ b/indra/llcommon/llthread.cpp @@ -135,6 +135,8 @@ void LLThread::threadRun() set_thread_name(-1, mName.c_str()); #endif + LL_PROFILER_SET_THREAD_NAME( mName.c_str() ); + // this is the first point at which we're actually running in the new thread mID = currentID(); @@ -331,6 +333,7 @@ bool LLThread::runCondition(void) // Stop thread execution if requested until unpaused. void LLThread::checkPause() { + LL_PROFILE_ZONE_SCOPED mDataLock->lock(); // This is in a while loop because the pthread API allows for spurious wakeups. @@ -362,17 +365,20 @@ void LLThread::setQuitting() // static LLThread::id_t LLThread::currentID() { + LL_PROFILE_ZONE_SCOPED return std::this_thread::get_id(); } // static void LLThread::yield() { + LL_PROFILE_ZONE_SCOPED std::this_thread::yield(); } void LLThread::wake() { + LL_PROFILE_ZONE_SCOPED mDataLock->lock(); if(!shouldSleep()) { @@ -383,6 +389,7 @@ void LLThread::wake() void LLThread::wakeLocked() { + LL_PROFILE_ZONE_SCOPED if(!shouldSleep()) { mRunCondition->signal(); @@ -391,11 +398,13 @@ void LLThread::wakeLocked() void LLThread::lockData() { + LL_PROFILE_ZONE_SCOPED mDataLock->lock(); } void LLThread::unlockData() { + LL_PROFILE_ZONE_SCOPED mDataLock->unlock(); } diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 26e0d71d31..719edcd579 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -1,6 +1,6 @@ /** * @file llthreadsafequeue.h - * @brief Base classes for thread, mutex and condition handling. + * @brief Queue protected with mutexes for cross-thread use * * $LicenseInfo:firstyear=2004&license=viewerlgpl$ * Second Life Viewer Source Code @@ -27,16 +27,19 @@ #ifndef LL_LLTHREADSAFEQUEUE_H #define LL_LLTHREADSAFEQUEUE_H -#include "llexception.h" -#include <deque> -#include <string> -#include <chrono> -#include "mutex.h" #include "llcoros.h" #include LLCOROS_MUTEX_HEADER #include <boost/fiber/timed_mutex.hpp> #include LLCOROS_CONDVAR_HEADER +#include "llexception.h" +#include "mutex.h" +#include <chrono> +#include <queue> +#include <string> +/***************************************************************************** +* LLThreadSafeQueue +*****************************************************************************/ // // A general queue exception. // @@ -66,70 +69,108 @@ public: } }; -// -// Implements a thread safe FIFO. -// -template<typename ElementT> +/** + * Implements a thread safe FIFO. + */ +// Let the default std::queue default to underlying std::deque. Override if +// desired. +template<typename ElementT, typename QueueT=std::queue<ElementT>> class LLThreadSafeQueue { public: typedef ElementT value_type; - - // If the pool is set to NULL one will be allocated and managed by this - // queue. + + // Limiting the number of pending items prevents unbounded growth of the + // underlying queue. LLThreadSafeQueue(U32 capacity = 1024); - - // Add an element to the front of queue (will block if the queue has + virtual ~LLThreadSafeQueue() {} + + // Add an element to the queue (will block if the queue has // reached capacity). // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. - void pushFront(ElementT const & element); - - // Try to add an element to the front of queue without blocking. Returns + template <typename T> + void push(T&& element); + // legacy name + void pushFront(ElementT const & element) { return push(element); } + + // Try to add an element to the queue without blocking. Returns // true only if the element was actually added. - bool tryPushFront(ElementT const & element); + template <typename T> + bool tryPush(T&& element); + // legacy name + bool tryPushFront(ElementT const & element) { return tryPush(element); } - // Try to add an element to the front of queue, blocking if full but with - // timeout. Returns true if the element was added. + // Try to add an element to the queue, blocking if full but with timeout + // after specified duration. Returns true if the element was added. // There are potentially two different timeouts involved: how long to try // to lock the mutex, versus how long to wait for the queue to stop being // full. Careful settings for each timeout might be orders of magnitude // apart. However, this method conflates them. + template <typename Rep, typename Period, typename T> + bool tryPushFor(const std::chrono::duration<Rep, Period>& timeout, + T&& element); + // legacy name template <typename Rep, typename Period> bool tryPushFrontFor(const std::chrono::duration<Rep, Period>& timeout, - ElementT const & element); + ElementT const & element) { return tryPushFor(timeout, element); } - // Pop the element at the end of the queue (will block if the queue is + // Try to add an element to the queue, blocking if full but with + // timeout at specified time_point. Returns true if the element was added. + template <typename Clock, typename Duration, typename T> + bool tryPushUntil(const std::chrono::time_point<Clock, Duration>& until, + T&& element); + // no legacy name because this is a newer method + + // Pop the element at the head of the queue (will block if the queue is // empty). // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. - ElementT popBack(void); - - // Pop an element from the end of the queue if there is one available. + ElementT pop(void); + // legacy name + ElementT popBack(void) { return pop(); } + + // Pop an element from the head of the queue if there is one available. // Returns true only if an element was popped. - bool tryPopBack(ElementT & element); - + bool tryPop(ElementT & element); + // legacy name + bool tryPopBack(ElementT & element) { return tryPop(element); } + + // Pop the element at the head of the queue, blocking if empty, with + // timeout after specified duration. Returns true if an element was popped. + template <typename Rep, typename Period> + bool tryPopFor(const std::chrono::duration<Rep, Period>& timeout, ElementT& element); + // no legacy name because this is a newer method + + // Pop the element at the head of the queue, blocking if empty, with + // timeout at specified time_point. Returns true if an element was popped. + template <typename Clock, typename Duration> + bool tryPopUntil(const std::chrono::time_point<Clock, Duration>& until, + ElementT& element); + // no legacy name because this is a newer method + // Returns the size of the queue. size_t size(); // closes the queue: - // - every subsequent pushFront() call will throw LLThreadSafeQueueInterrupt - // - every subsequent tryPushFront() call will return false - // - popBack() calls will return normally until the queue is drained, then - // every subsequent popBack() will throw LLThreadSafeQueueInterrupt - // - tryPopBack() calls will return normally until the queue is drained, - // then every subsequent tryPopBack() call will return false + // - every subsequent push() call will throw LLThreadSafeQueueInterrupt + // - every subsequent tryPush() call will return false + // - pop() calls will return normally until the queue is drained, then + // every subsequent pop() will throw LLThreadSafeQueueInterrupt + // - tryPop() calls will return normally until the queue is drained, + // then every subsequent tryPop() call will return false void close(); - // detect closed state + // producer end: are we prevented from pushing any additional items? bool isClosed(); - // inverse of isClosed() - explicit operator bool(); + // consumer end: are we done, is the queue entirely drained? + bool done(); -private: - std::deque< ElementT > mStorage; +protected: + typedef QueueT queue_type; + QueueT mStorage; U32 mCapacity; bool mClosed; @@ -137,37 +178,152 @@ private: typedef std::unique_lock<decltype(mLock)> lock_t; boost::fibers::condition_variable_any mCapacityCond; boost::fibers::condition_variable_any mEmptyCond; -}; -// LLThreadSafeQueue -//----------------------------------------------------------------------------- + enum pop_result { EMPTY, DONE, WAITING, POPPED }; + // implementation logic, suitable for passing to tryLockUntil() + template <typename Clock, typename Duration> + pop_result tryPopUntil_(lock_t& lock, + const std::chrono::time_point<Clock, Duration>& until, + ElementT& element); + // if we're able to lock immediately, do so and run the passed callable, + // which must accept lock_t& and return bool + template <typename CALLABLE> + bool tryLock(CALLABLE&& callable); + // if we're able to lock before the passed time_point, do so and run the + // passed callable, which must accept lock_t& and return bool + template <typename Clock, typename Duration, typename CALLABLE> + bool tryLockUntil(const std::chrono::time_point<Clock, Duration>& until, + CALLABLE&& callable); + // while lock is locked, really push the passed element, if we can + template <typename T> + bool push_(lock_t& lock, T&& element); + // while lock is locked, really pop the head element, if we can + pop_result pop_(lock_t& lock, ElementT& element); + // Is the current head element ready to pop? We say yes; subclass can + // override as needed. + virtual bool canPop(const ElementT& head) const { return true; } +}; -template<typename ElementT> -LLThreadSafeQueue<ElementT>::LLThreadSafeQueue(U32 capacity) : +/***************************************************************************** +* PriorityQueueAdapter +*****************************************************************************/ +namespace LL +{ + /** + * std::priority_queue's API is almost like std::queue, intentionally of + * course, but you must access the element about to pop() as top() rather + * than as front(). Make an adapter for use with LLThreadSafeQueue. + */ + template <typename T, typename Container=std::vector<T>, + typename Compare=std::less<typename Container::value_type>> + class PriorityQueueAdapter + { + public: + // publish all the same types + typedef std::priority_queue<T, Container, Compare> queue_type; + typedef typename queue_type::container_type container_type; + typedef typename queue_type::value_compare value_compare; + typedef typename queue_type::value_type value_type; + typedef typename queue_type::size_type size_type; + typedef typename queue_type::reference reference; + typedef typename queue_type::const_reference const_reference; + + // Although std::queue defines both const and non-const front() + // methods, std::priority_queue defines only const top(). + const_reference front() const { return mQ.top(); } + // std::priority_queue has no equivalent to back(), so it's good that + // LLThreadSafeQueue doesn't use it. + + // All the rest of these merely forward to the corresponding + // queue_type methods. + bool empty() const { return mQ.empty(); } + size_type size() const { return mQ.size(); } + void push(const value_type& value) { mQ.push(value); } + void push(value_type&& value) { mQ.push(std::move(value)); } + template <typename... Args> + void emplace(Args&&... args) { mQ.emplace(std::forward<Args>(args)...); } + void pop() { mQ.pop(); } + + private: + queue_type mQ; + }; +} // namespace LL + + +/***************************************************************************** +* LLThreadSafeQueue implementation +*****************************************************************************/ +template<typename ElementT, typename QueueT> +LLThreadSafeQueue<ElementT, QueueT>::LLThreadSafeQueue(U32 capacity) : mCapacity(capacity), mClosed(false) { } -template<typename ElementT> -void LLThreadSafeQueue<ElementT>::pushFront(ElementT const & element) +// if we're able to lock immediately, do so and run the passed callable, which +// must accept lock_t& and return bool +template <typename ElementT, typename QueueT> +template <typename CALLABLE> +bool LLThreadSafeQueue<ElementT, QueueT>::tryLock(CALLABLE&& callable) +{ + lock_t lock1(mLock, std::defer_lock); + if (!lock1.try_lock()) + return false; + + return std::forward<CALLABLE>(callable)(lock1); +} + + +// if we're able to lock before the passed time_point, do so and run the +// passed callable, which must accept lock_t& and return bool +template <typename ElementT, typename QueueT> +template <typename Clock, typename Duration, typename CALLABLE> +bool LLThreadSafeQueue<ElementT, QueueT>::tryLockUntil( + const std::chrono::time_point<Clock, Duration>& until, + CALLABLE&& callable) +{ + lock_t lock1(mLock, std::defer_lock); + if (!lock1.try_lock_until(until)) + return false; + + return std::forward<CALLABLE>(callable)(lock1); +} + + +// while lock is locked, really push the passed element, if we can +template <typename ElementT, typename QueueT> +template <typename T> +bool LLThreadSafeQueue<ElementT, QueueT>::push_(lock_t& lock, T&& element) +{ + if (mStorage.size() >= mCapacity) + return false; + + mStorage.push(std::forward<T>(element)); + lock.unlock(); + // now that we've pushed, if somebody's been waiting to pop, signal them + mEmptyCond.notify_one(); + return true; +} + + +template <typename ElementT, typename QueueT> +template<typename T> +void LLThreadSafeQueue<ElementT, QueueT>::push(T&& element) { lock_t lock1(mLock); while (true) { + // On the producer side, it doesn't matter whether the queue has been + // drained or not: the moment either end calls close(), further push() + // operations will fail. if (mClosed) { LLTHROW(LLThreadSafeQueueInterrupt()); } - if (mStorage.size() < mCapacity) - { - mStorage.push_front(element); - lock1.unlock(); - mEmptyCond.notify_one(); + if (push_(lock1, std::forward<T>(element))) return; - } // Storage Full. Wait for signal. mCapacityCond.wait(lock1); @@ -175,142 +331,225 @@ void LLThreadSafeQueue<ElementT>::pushFront(ElementT const & element) } -template <typename ElementT> -template <typename Rep, typename Period> -bool LLThreadSafeQueue<ElementT>::tryPushFrontFor(const std::chrono::duration<Rep, Period>& timeout, - ElementT const & element) +template<typename ElementT, typename QueueT> +template<typename T> +bool LLThreadSafeQueue<ElementT, QueueT>::tryPush(T&& element) { - // Convert duration to time_point: passing the same timeout duration to - // each of multiple calls is wrong. - auto endpoint = std::chrono::steady_clock::now() + timeout; + return tryLock( + [this, element=std::move(element)](lock_t& lock) + { + if (mClosed) + return false; + return push_(lock, std::move(element)); + }); +} - lock_t lock1(mLock, std::defer_lock); - if (!lock1.try_lock_until(endpoint)) - return false; - while (true) - { - if (mClosed) - { - return false; - } +template <typename ElementT, typename QueueT> +template <typename Rep, typename Period, typename T> +bool LLThreadSafeQueue<ElementT, QueueT>::tryPushFor( + const std::chrono::duration<Rep, Period>& timeout, + T&& element) +{ + // Convert duration to time_point: passing the same timeout duration to + // each of multiple calls is wrong. + return tryPushUntil(std::chrono::steady_clock::now() + timeout, + std::forward<T>(element)); +} - if (mStorage.size() < mCapacity) - { - mStorage.push_front(element); - lock1.unlock(); - mEmptyCond.notify_one(); - return true; - } - // Storage Full. Wait for signal. - if (LLCoros::cv_status::timeout == mCapacityCond.wait_until(lock1, endpoint)) +template <typename ElementT, typename QueueT> +template <typename Clock, typename Duration, typename T> +bool LLThreadSafeQueue<ElementT, QueueT>::tryPushUntil( + const std::chrono::time_point<Clock, Duration>& until, + T&& element) +{ + return tryLockUntil( + until, + [this, until, element=std::move(element)](lock_t& lock) { - // timed out -- formally we might recheck both conditions above - return false; - } - // If we didn't time out, we were notified for some reason. Loop back - // to check. - } + while (true) + { + if (mClosed) + { + return false; + } + + if (push_(lock, std::move(element))) + return true; + + // Storage Full. Wait for signal. + if (LLCoros::cv_status::timeout == mCapacityCond.wait_until(lock, until)) + { + // timed out -- formally we might recheck both conditions above + return false; + } + // If we didn't time out, we were notified for some reason. Loop back + // to check. + } + }); } -template<typename ElementT> -bool LLThreadSafeQueue<ElementT>::tryPushFront(ElementT const & element) +// while lock is locked, really pop the head element, if we can +template <typename ElementT, typename QueueT> +typename LLThreadSafeQueue<ElementT, QueueT>::pop_result +LLThreadSafeQueue<ElementT, QueueT>::pop_(lock_t& lock, ElementT& element) { - lock_t lock1(mLock, std::defer_lock); - if (!lock1.try_lock()) - return false; - - if (mClosed) - return false; + // If mStorage is empty, there's no head element. + if (mStorage.empty()) + return mClosed? DONE : EMPTY; - if (mStorage.size() >= mCapacity) - return false; + // If there's a head element, pass it to canPop() to see if it's ready to pop. + if (! canPop(mStorage.front())) + return WAITING; - mStorage.push_front(element); - lock1.unlock(); - mEmptyCond.notify_one(); - return true; + // std::queue::front() is the element about to pop() + element = mStorage.front(); + mStorage.pop(); + lock.unlock(); + // now that we've popped, if somebody's been waiting to push, signal them + mCapacityCond.notify_one(); + return POPPED; } -template<typename ElementT> -ElementT LLThreadSafeQueue<ElementT>::popBack(void) +template<typename ElementT, typename QueueT> +ElementT LLThreadSafeQueue<ElementT, QueueT>::pop(void) { lock_t lock1(mLock); + ElementT value; while (true) { - if (!mStorage.empty()) - { - ElementT value = mStorage.back(); - mStorage.pop_back(); - lock1.unlock(); - mCapacityCond.notify_one(); - return value; - } - - if (mClosed) + // On the consumer side, we always try to pop before checking mClosed + // so we can finish draining the queue. + pop_result popped = pop_(lock1, value); + if (popped == POPPED) + return std::move(value); + + // Once the queue is DONE, there will never be any more coming. + if (popped == DONE) { LLTHROW(LLThreadSafeQueueInterrupt()); } - // Storage empty. Wait for signal. + // If we didn't pop because WAITING, i.e. canPop() returned false, + // then even if the producer end has been closed, there's still at + // least one item to drain: wait for it. Or we might be EMPTY, with + // the queue still open. Either way, wait for signal. mEmptyCond.wait(lock1); } } -template<typename ElementT> -bool LLThreadSafeQueue<ElementT>::tryPopBack(ElementT & element) +template<typename ElementT, typename QueueT> +bool LLThreadSafeQueue<ElementT, QueueT>::tryPop(ElementT & element) { - lock_t lock1(mLock, std::defer_lock); - if (!lock1.try_lock()) - return false; + return tryLock( + [this, &element](lock_t& lock) + { + // conflate EMPTY, DONE, WAITING: tryPop() behavior when the queue + // is closed is implemented by simple inability to push any new + // elements + return pop_(lock, element) == POPPED; + }); +} - // no need to check mClosed: tryPopBack() behavior when the queue is - // closed is implemented by simple inability to push any new elements - if (mStorage.empty()) - return false; - element = mStorage.back(); - mStorage.pop_back(); - lock1.unlock(); - mCapacityCond.notify_one(); - return true; +template <typename ElementT, typename QueueT> +template <typename Rep, typename Period> +bool LLThreadSafeQueue<ElementT, QueueT>::tryPopFor( + const std::chrono::duration<Rep, Period>& timeout, + ElementT& element) +{ + // Convert duration to time_point: passing the same timeout duration to + // each of multiple calls is wrong. + return tryPopUntil(std::chrono::steady_clock::now() + timeout, element); } -template<typename ElementT> -size_t LLThreadSafeQueue<ElementT>::size(void) +template <typename ElementT, typename QueueT> +template <typename Clock, typename Duration> +bool LLThreadSafeQueue<ElementT, QueueT>::tryPopUntil( + const std::chrono::time_point<Clock, Duration>& until, + ElementT& element) +{ + return tryLockUntil( + until, + [this, until, &element](lock_t& lock) + { + // conflate EMPTY, DONE, WAITING + return tryPopUntil_(lock, until, element) == POPPED; + }); +} + + +// body of tryPopUntil(), called once we have the lock +template <typename ElementT, typename QueueT> +template <typename Clock, typename Duration> +typename LLThreadSafeQueue<ElementT, QueueT>::pop_result +LLThreadSafeQueue<ElementT, QueueT>::tryPopUntil_( + lock_t& lock, + const std::chrono::time_point<Clock, Duration>& until, + ElementT& element) +{ + while (true) + { + pop_result popped = pop_(lock, element); + if (popped == POPPED || popped == DONE) + { + // If we succeeded, great! If we've drained the last item, so be + // it. Either way, break the loop and tell caller. + return popped; + } + + // EMPTY or WAITING: wait for signal. + if (LLCoros::cv_status::timeout == mEmptyCond.wait_until(lock, until)) + { + // timed out -- formally we might recheck + // as it is, break loop + return popped; + } + // If we didn't time out, we were notified for some reason. Loop back + // to check. + } +} + + +template<typename ElementT, typename QueueT> +size_t LLThreadSafeQueue<ElementT, QueueT>::size(void) { lock_t lock(mLock); return mStorage.size(); } -template<typename ElementT> -void LLThreadSafeQueue<ElementT>::close() + +template<typename ElementT, typename QueueT> +void LLThreadSafeQueue<ElementT, QueueT>::close() { lock_t lock(mLock); mClosed = true; lock.unlock(); - // wake up any blocked popBack() calls + // wake up any blocked pop() calls mEmptyCond.notify_all(); - // wake up any blocked pushFront() calls + // wake up any blocked push() calls mCapacityCond.notify_all(); } -template<typename ElementT> -bool LLThreadSafeQueue<ElementT>::isClosed() + +template<typename ElementT, typename QueueT> +bool LLThreadSafeQueue<ElementT, QueueT>::isClosed() { lock_t lock(mLock); - return mClosed && mStorage.size() == 0; + return mClosed; } -template<typename ElementT> -LLThreadSafeQueue<ElementT>::operator bool() + +template<typename ElementT, typename QueueT> +bool LLThreadSafeQueue<ElementT, QueueT>::done() { - return ! isClosed(); + lock_t lock(mLock); + return mClosed && mStorage.empty(); } #endif diff --git a/indra/llcommon/lluuid.h b/indra/llcommon/lluuid.h index fe7482ba29..86a396ab06 100644 --- a/indra/llcommon/lluuid.h +++ b/indra/llcommon/lluuid.h @@ -184,6 +184,17 @@ struct boost::hash<LLUUID> } }; +// Adapt boost hash to std hash +namespace std +{ + template<> struct hash<LLUUID> + { + std::size_t operator()(LLUUID const& s) const noexcept + { + return boost::hash<LLUUID>()(s); + } + }; +} #endif diff --git a/indra/llcommon/tests/llinstancetracker_test.cpp b/indra/llcommon/tests/llinstancetracker_test.cpp index 9b89159625..5daa29adf4 100644 --- a/indra/llcommon/tests/llinstancetracker_test.cpp +++ b/indra/llcommon/tests/llinstancetracker_test.cpp @@ -90,19 +90,19 @@ namespace tut { Keyed one("one"); ensure_equals(Keyed::instanceCount(), 1); - Keyed* found = Keyed::getInstance("one"); - ensure("couldn't find stack Keyed", found); - ensure_equals("found wrong Keyed instance", found, &one); + auto found = Keyed::getInstance("one"); + ensure("couldn't find stack Keyed", bool(found)); + ensure_equals("found wrong Keyed instance", found.get(), &one); { boost::scoped_ptr<Keyed> two(new Keyed("two")); ensure_equals(Keyed::instanceCount(), 2); - Keyed* found = Keyed::getInstance("two"); - ensure("couldn't find heap Keyed", found); - ensure_equals("found wrong Keyed instance", found, two.get()); + auto found = Keyed::getInstance("two"); + ensure("couldn't find heap Keyed", bool(found)); + ensure_equals("found wrong Keyed instance", found.get(), two.get()); } ensure_equals(Keyed::instanceCount(), 1); } - Keyed* found = Keyed::getInstance("one"); + auto found = Keyed::getInstance("one"); ensure("Keyed key lives too long", ! found); ensure_equals(Keyed::instanceCount(), 0); } diff --git a/indra/llcommon/tests/threadsafeschedule_test.cpp b/indra/llcommon/tests/threadsafeschedule_test.cpp new file mode 100644 index 0000000000..af67b9f492 --- /dev/null +++ b/indra/llcommon/tests/threadsafeschedule_test.cpp @@ -0,0 +1,69 @@ +/** + * @file threadsafeschedule_test.cpp + * @author Nat Goodspeed + * @date 2021-10-04 + * @brief Test for threadsafeschedule. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "threadsafeschedule.h" +// STL headers +// std headers +#include <chrono> +// external library headers +// other Linden headers +#include "../test/lltut.h" + +using namespace std::literals::chrono_literals; // ms suffix +using namespace std::literals::string_literals; // s suffix +using Queue = LL::ThreadSafeSchedule<std::string>; + +/***************************************************************************** +* TUT +*****************************************************************************/ +namespace tut +{ + struct threadsafeschedule_data + { + Queue queue; + }; + typedef test_group<threadsafeschedule_data> threadsafeschedule_group; + typedef threadsafeschedule_group::object object; + threadsafeschedule_group threadsafeschedulegrp("threadsafeschedule"); + + template<> template<> + void object::test<1>() + { + set_test_name("push"); + // Simply calling push() a few times might result in indeterminate + // delivery order if the resolution of steady_clock is coarser than + // the real time required for each push() call. Explicitly increment + // the timestamp for each one -- but since we're passing explicit + // timestamps, make the queue reorder them. + queue.push(Queue::TimeTuple(Queue::Clock::now() + 20ms, "ghi")); + // Given the various push() overloads, you have to match the type + // exactly: conversions are ambiguous. + queue.push("abc"s); + queue.push(Queue::Clock::now() + 10ms, "def"); + queue.close(); + auto entry = queue.pop(); + ensure_equals("failed to pop first", std::get<0>(entry), "abc"s); + entry = queue.pop(); + ensure_equals("failed to pop second", std::get<0>(entry), "def"s); + ensure("queue not closed", queue.isClosed()); + ensure("queue prematurely done", ! queue.done()); + std::string s; + bool popped = queue.tryPopFor(1s, s); + ensure("failed to pop third", popped); + ensure_equals("third is wrong", s, "ghi"s); + popped = queue.tryPop(s); + ensure("queue not empty", ! popped); + ensure("queue not done", queue.done()); + } +} // namespace tut diff --git a/indra/llcommon/tests/tuple_test.cpp b/indra/llcommon/tests/tuple_test.cpp new file mode 100644 index 0000000000..af94e2086c --- /dev/null +++ b/indra/llcommon/tests/tuple_test.cpp @@ -0,0 +1,47 @@ +/** + * @file tuple_test.cpp + * @author Nat Goodspeed + * @date 2021-10-04 + * @brief Test for tuple. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "tuple.h" +// STL headers +// std headers +// external library headers +// other Linden headers +#include "../test/lltut.h" + +/***************************************************************************** +* TUT +*****************************************************************************/ +namespace tut +{ + struct tuple_data + { + }; + typedef test_group<tuple_data> tuple_group; + typedef tuple_group::object object; + tuple_group tuplegrp("tuple"); + + template<> template<> + void object::test<1>() + { + set_test_name("tuple"); + std::tuple<std::string, int> tup{ "abc", 17 }; + std::tuple<int, std::string, int> ptup{ tuple_cons(34, tup) }; + std::tuple<std::string, int> tup2; + int i; + std::tie(i, tup2) = tuple_split(ptup); + ensure_equals("tuple_car() fail", i, 34); + ensure_equals("tuple_cdr() (0) fail", std::get<0>(tup2), "abc"); + ensure_equals("tuple_cdr() (1) fail", std::get<1>(tup2), 17); + } +} // namespace tut diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp new file mode 100644 index 0000000000..d5405400fd --- /dev/null +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -0,0 +1,159 @@ +/** + * @file workqueue_test.cpp + * @author Nat Goodspeed + * @date 2021-10-07 + * @brief Test for workqueue. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "workqueue.h" +// STL headers +// std headers +#include <chrono> +#include <deque> +// external library headers +// other Linden headers +#include "../test/lltut.h" +#include "llcond.h" +#include "llstring.h" +#include "stringize.h" + +using namespace LL; +using namespace std::literals::chrono_literals; // ms suffix +using namespace std::literals::string_literals; // s suffix + +/***************************************************************************** +* TUT +*****************************************************************************/ +namespace tut +{ + struct workqueue_data + { + WorkQueue queue{"queue"}; + }; + typedef test_group<workqueue_data> workqueue_group; + typedef workqueue_group::object object; + workqueue_group workqueuegrp("workqueue"); + + template<> template<> + void object::test<1>() + { + set_test_name("name"); + ensure_equals("didn't capture name", queue.getKey(), "queue"); + ensure("not findable", WorkQueue::getInstance("queue") == queue.getWeak().lock()); + WorkQueue q2; + ensure("has no name", LLStringUtil::startsWith(q2.getKey(), "WorkQueue")); + } + + template<> template<> + void object::test<2>() + { + set_test_name("post"); + bool wasRun{ false }; + // We only get away with binding a simple bool because we're running + // the work on the same thread. + queue.post([&wasRun](){ wasRun = true; }); + queue.close(); + ensure("ran too soon", ! wasRun); + queue.runUntilClose(); + ensure("didn't run", wasRun); + } + + template<> template<> + void object::test<3>() + { + set_test_name("postEvery"); + // record of runs + using Shared = std::deque<WorkQueue::TimePoint>; + // This is an example of how to share data between the originator of + // postEvery(work) and the work item itself, since usually a WorkQueue + // is used to dispatch work to a different thread. Neither of them + // should call any of LLCond's wait methods: you don't want to stall + // either the worker thread or the originating thread (conventionally + // main). Use LLCond or a subclass even if all you want to do is + // signal the work item that it can quit; consider LLOneShotCond. + LLCond<Shared> data; + auto start = WorkQueue::TimePoint::clock::now(); + auto interval = 100ms; + queue.postEvery( + interval, + [&data, count = 0] + () mutable + { + // record the timestamp at which this instance is running + data.update_one( + [](Shared& data) + { + data.push_back(WorkQueue::TimePoint::clock::now()); + }); + // by the 3rd call, return false to stop + return (++count < 3); + }); + // no convenient way to close() our queue while we've got a + // postEvery() running, so run until we think we should have exhausted + // the iterations + queue.runFor(10*interval); + // Take a copy of the captured deque. + Shared result = data.get(); + ensure_equals("called wrong number of times", result.size(), 3); + // postEvery() assumes you want the first call to happen right away. + // Pretend our start time was (interval) earlier than that, to make + // our too early/too late tests uniform for all entries. + start -= interval; + for (size_t i = 0; i < result.size(); ++i) + { + auto diff = result[i] - start; + start += interval; + try + { + ensure(STRINGIZE("call " << i << " too soon"), diff >= interval); + ensure(STRINGIZE("call " << i << " too late"), diff < interval*1.5); + } + catch (const tut::failure&) + { + auto interval_ms = interval / 1ms; + auto diff_ms = diff / 1ms; + std::cerr << "interval " << interval_ms + << "ms; diff " << diff_ms << "ms" << std::endl; + throw; + } + } + } + + template<> template<> + void object::test<4>() + { + set_test_name("postTo"); + WorkQueue main("main"); + auto qptr = WorkQueue::getInstance("queue"); + int result = 0; + main.postTo( + qptr, + [](){ return 17; }, + // Note that a postTo() *callback* can safely bind a reference to + // a variable on the invoking thread, because the callback is run + // on the invoking thread. + [&result](int i){ result = i; }); + // this should post the callback to main + qptr->runOne(); + // this should run the callback + main.runOne(); + ensure_equals("failed to run int callback", result, 17); + + std::string alpha; + // postTo() handles arbitrary return types + main.postTo( + qptr, + [](){ return "abc"s; }, + [&alpha](const std::string& s){ alpha = s; }); + qptr->runPending(); + main.runPending(); + ensure_equals("failed to run string callback", alpha, "abc"); + } +} // namespace tut diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp new file mode 100644 index 0000000000..aa7d4179a2 --- /dev/null +++ b/indra/llcommon/threadpool.cpp @@ -0,0 +1,75 @@ +/** + * @file threadpool.cpp + * @author Nat Goodspeed + * @date 2021-10-21 + * @brief Implementation for threadpool. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "threadpool.h" +// STL headers +// std headers +// external library headers +// other Linden headers +#include "llerror.h" +#include "llevents.h" +#include "stringize.h" + +LL::ThreadPool::ThreadPool(const std::string& name, size_t threads): + mQueue(name), + mName("ThreadPool:" + name) +{ + for (size_t i = 0; i < threads; ++i) + { + std::string tname{ STRINGIZE(mName << ':' << (i+i) << '/' << threads) }; + mThreads.emplace_back(tname, [this, tname](){ run(tname); }); + } + // Listen on "LLApp", and when the app is shutting down, close the queue + // and join the workers. + LLEventPumps::instance().obtain("LLApp").listen( + mName, + [this](const LLSD& stat) + { + std::string status(stat["status"]); + if (status != "running") + { + // viewer is starting shutdown -- proclaim the end is nigh! + LL_DEBUGS("ThreadPool") << mName << " saw " << status << LL_ENDL; + close(); + } + return false; + }); +} + +LL::ThreadPool::~ThreadPool() +{ + close(); +} + +void LL::ThreadPool::close() +{ + if (! mQueue.isClosed()) + { + LL_DEBUGS("ThreadPool") << mName << " closing queue and joining threads" << LL_ENDL; + mQueue.close(); + for (auto& pair: mThreads) + { + LL_DEBUGS("ThreadPool") << mName << " waiting on thread " << pair.first << LL_ENDL; + pair.second.join(); + } + LL_DEBUGS("ThreadPool") << mName << " shutdown complete" << LL_ENDL; + } +} + +void LL::ThreadPool::run(const std::string& name) +{ + LL_DEBUGS("ThreadPool") << name << " starting" << LL_ENDL; + mQueue.runUntilClose(); + LL_DEBUGS("ThreadPool") << name << " stopping" << LL_ENDL; +} diff --git a/indra/llcommon/threadpool.h b/indra/llcommon/threadpool.h new file mode 100644 index 0000000000..8f3c8514b5 --- /dev/null +++ b/indra/llcommon/threadpool.h @@ -0,0 +1,46 @@ +/** + * @file threadpool.h + * @author Nat Goodspeed + * @date 2021-10-21 + * @brief ThreadPool configures a WorkQueue along with a pool of threads to + * service it. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_THREADPOOL_H) +#define LL_THREADPOOL_H + +#include "workqueue.h" +#include <string> +#include <thread> +#include <utility> // std::pair +#include <vector> + +namespace LL +{ + + class ThreadPool + { + public: + /** + * Pass ThreadPool a string name. This can be used to look up the + * relevant WorkQueue. + */ + ThreadPool(const std::string& name, size_t threads=1); + ~ThreadPool(); + void close(); + + private: + void run(const std::string& name); + + WorkQueue mQueue; + std::string mName; + std::vector<std::pair<std::string, std::thread>> mThreads; + }; + +} // namespace LL + +#endif /* ! defined(LL_THREADPOOL_H) */ diff --git a/indra/llcommon/threadsafeschedule.h b/indra/llcommon/threadsafeschedule.h new file mode 100644 index 0000000000..c8ad23532b --- /dev/null +++ b/indra/llcommon/threadsafeschedule.h @@ -0,0 +1,373 @@ +/** + * @file threadsafeschedule.h + * @author Nat Goodspeed + * @date 2021-10-02 + * @brief ThreadSafeSchedule is an ordered queue in which every item has an + * associated timestamp. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_THREADSAFESCHEDULE_H) +#define LL_THREADSAFESCHEDULE_H + +#include "chrono.h" +#include "llexception.h" +#include "llthreadsafequeue.h" +#include "tuple.h" +#include <chrono> +#include <tuple> + +namespace LL +{ + namespace ThreadSafeSchedulePrivate + { + using TimePoint = std::chrono::steady_clock::time_point; + // Bundle consumer's data with a TimePoint to order items by timestamp. + template <typename... Args> + using TimestampedTuple = std::tuple<TimePoint, Args...>; + + // comparison functor for TimedTuples -- see TimedQueue comments + struct ReverseTupleOrder + { + template <typename Tuple> + bool operator()(const Tuple& left, const Tuple& right) const + { + return std::get<0>(left) > std::get<0>(right); + } + }; + + template <typename... Args> + using TimedQueue = PriorityQueueAdapter< + TimestampedTuple<Args...>, + // std::vector is the default storage for std::priority_queue, + // have to restate to specify comparison template parameter + std::vector<TimestampedTuple<Args...>>, + // std::priority_queue uses a counterintuitive comparison + // behavior: the default std::less comparator is used to present + // the *highest* value as top(). So to sort by earliest timestamp, + // we must invert by using >. + ReverseTupleOrder>; + } // namespace ThreadSafeSchedulePrivate + + /** + * ThreadSafeSchedule is an ordered LLThreadSafeQueue in which every item + * is given an associated timestamp. That is, TimePoint is implicitly + * prepended to the std::tuple with the specified types. + * + * Items are popped in increasing chronological order. Moreover, any item + * with a timestamp in the future is held back until + * std::chrono::steady_clock reaches that timestamp. + */ + template <typename... Args> + class ThreadSafeSchedule: + public LLThreadSafeQueue<ThreadSafeSchedulePrivate::TimestampedTuple<Args...>, + ThreadSafeSchedulePrivate::TimedQueue<Args...>> + { + public: + using DataTuple = std::tuple<Args...>; + using TimeTuple = ThreadSafeSchedulePrivate::TimestampedTuple<Args...>; + + private: + using super = LLThreadSafeQueue<TimeTuple, ThreadSafeSchedulePrivate::TimedQueue<Args...>>; + using lock_t = typename super::lock_t; + // VS 2017 needs this due to a bug: + // https://developercommunity.visualstudio.com/t/cannot-access-protected-enumerator-of-enclosing-cl/203430 + enum pop_result { EMPTY=super::EMPTY, DONE=super::DONE, WAITING=super::WAITING, POPPED=super::POPPED }; + + public: + using Closed = LLThreadSafeQueueInterrupt; + using TimePoint = ThreadSafeSchedulePrivate::TimePoint; + using Clock = TimePoint::clock; + + ThreadSafeSchedule(U32 capacity=1024): + super(capacity) + {} + + /*----------------------------- push() -----------------------------*/ + /// explicitly pass TimeTuple + using super::push; + + /// pass DataTuple with implicit now + // This could be ambiguous for Args with a single type. Unfortunately + // we can't enable_if an individual method with a condition based on + // the *class* template arguments, only on that method's template + // arguments. We could specialize this class for the single-Args case; + // we could minimize redundancy by breaking out a common base class... + void push(const DataTuple& tuple) + { + push(tuple_cons(Clock::now(), tuple)); + } + + /// individually pass each component of the TimeTuple + void push(const TimePoint& time, Args&&... args) + { + push(TimeTuple(time, std::forward<Args>(args)...)); + } + + /// individually pass every component except the TimePoint (implies now) + // This could be ambiguous if the first specified template parameter + // type is also TimePoint. We could try to disambiguate, but a simpler + // approach would be for the caller to explicitly construct DataTuple + // and call that overload. + void push(Args&&... args) + { + push(Clock::now(), std::forward<Args>(args)...); + } + + /*--------------------------- tryPush() ----------------------------*/ + /// explicit TimeTuple + using super::tryPush; + + /// DataTuple with implicit now + bool tryPush(const DataTuple& tuple) + { + return tryPush(tuple_cons(Clock::now(), tuple)); + } + + /// individually pass components + bool tryPush(const TimePoint& time, Args&&... args) + { + return tryPush(TimeTuple(time, std::forward<Args>(args)...)); + } + + /// individually pass components with implicit now + bool tryPush(Args&&... args) + { + return tryPush(Clock::now(), std::forward<Args>(args)...); + } + + /*-------------------------- tryPushFor() --------------------------*/ + /// explicit TimeTuple + using super::tryPushFor; + + /// DataTuple with implicit now + template <typename Rep, typename Period> + bool tryPushFor(const std::chrono::duration<Rep, Period>& timeout, + const DataTuple& tuple) + { + return tryPushFor(timeout, tuple_cons(Clock::now(), tuple)); + } + + /// individually pass components + template <typename Rep, typename Period> + bool tryPushFor(const std::chrono::duration<Rep, Period>& timeout, + const TimePoint& time, Args&&... args) + { + return tryPushFor(TimeTuple(time, std::forward<Args>(args)...)); + } + + /// individually pass components with implicit now + template <typename Rep, typename Period> + bool tryPushFor(const std::chrono::duration<Rep, Period>& timeout, + Args&&... args) + { + return tryPushFor(Clock::now(), std::forward<Args>(args)...); + } + + /*------------------------- tryPushUntil() -------------------------*/ + /// explicit TimeTuple + using super::tryPushUntil; + + /// DataTuple with implicit now + template <typename Clock, typename Duration> + bool tryPushUntil(const std::chrono::time_point<Clock, Duration>& until, + const DataTuple& tuple) + { + return tryPushUntil(until, tuple_cons(Clock::now(), tuple)); + } + + /// individually pass components + template <typename Clock, typename Duration> + bool tryPushUntil(const std::chrono::time_point<Clock, Duration>& until, + const TimePoint& time, Args&&... args) + { + return tryPushUntil(until, TimeTuple(time, std::forward<Args>(args)...)); + } + + /// individually pass components with implicit now + template <typename Clock, typename Duration> + bool tryPushUntil(const std::chrono::time_point<Clock, Duration>& until, + Args&&... args) + { + return tryPushUntil(until, Clock::now(), std::forward<Args>(args)...); + } + + /*----------------------------- pop() ------------------------------*/ + // Our consumer may or may not care about the timestamp associated + // with each popped item, so we allow retrieving either DataTuple or + // TimeTuple. One potential use would be to observe, and possibly + // adjust for, the time lag between the item time and the actual + // current time. + + /// pop DataTuple by value + // It would be great to notice when sizeof...(Args) == 1 and directly + // return the first (only) value, instead of making pop()'s caller + // call std::get<0>(value). See push(DataTuple) remarks for why we + // haven't yet jumped through those hoops. + DataTuple pop() + { + return tuple_cdr(popWithTime()); + } + + /// pop TimeTuple by value + TimeTuple popWithTime() + { + lock_t lock(super::mLock); + // We can't just sit around waiting forever, given that there may + // be items in the queue that are not yet ready but will *become* + // ready in the near future. So in fact, with this class, every + // pop() becomes a tryPopUntil(), constrained to the timestamp of + // the head item. It almost doesn't matter what we specify for the + // caller's time constraint -- all we really care about is the + // head item's timestamp. Since pop() and popWithTime() are + // defined to wait until either an item becomes available or the + // queue is closed, loop until one of those things happens. The + // constraint we pass just determines how often we'll loop while + // waiting. + TimeTuple tt; + while (true) + { + // Pick a point suitably far into the future. + TimePoint until = TimePoint::clock::now() + std::chrono::hours(24); + pop_result popped = tryPopUntil_(lock, until, tt); + if (popped == POPPED) + return std::move(tt); + + // DONE: throw, just as super::pop() does + if (popped == DONE) + { + LLTHROW(LLThreadSafeQueueInterrupt()); + } + // WAITING: we've still got items to drain. + // EMPTY: not closed, so it's worth waiting for more items. + // Either way, loop back to wait. + } + } + + // We can use tryPop(TimeTuple&) just as it stands; the only behavior + // difference is in our canPop() override method. + using super::tryPop; + + /// tryPop(DataTuple&) + bool tryPop(DataTuple& tuple) + { + TimeTuple tt; + if (! super::tryPop(tt)) + return false; + tuple = tuple_cdr(std::move(tt)); + return true; + } + + /// for when Args has exactly one type + bool tryPop(typename std::tuple_element<1, TimeTuple>::type& value) + { + TimeTuple tt; + if (! super::tryPop(tt)) + return false; + value = std::get<1>(std::move(tt)); + return true; + } + + /// tryPopFor() + template <typename Rep, typename Period, typename Tuple> + bool tryPopFor(const std::chrono::duration<Rep, Period>& timeout, Tuple& tuple) + { + // It's important to use OUR tryPopUntil() implementation, rather + // than delegating immediately to our base class. + return tryPopUntil(Clock::now() + timeout, tuple); + } + + /// tryPopUntil(TimeTuple&) + template <typename Clock, typename Duration> + bool tryPopUntil(const std::chrono::time_point<Clock, Duration>& until, + TimeTuple& tuple) + { + // super::tryPopUntil() wakes up when an item becomes available or + // we hit 'until', whichever comes first. Thing is, the current + // head of the queue could become ready sooner than either of + // those events, and we need to deliver it as soon as it does. + // Don't wait past the TimePoint of the head item. + // Naturally, lock the queue before peeking at mStorage. + return super::tryLockUntil( + until, + [this, until, &tuple](lock_t& lock) + { + // Use our time_point_cast to allow for 'until' that's a + // time_point type other than TimePoint. + return POPPED == + tryPopUntil_(lock, LL::time_point_cast<TimePoint>(until), tuple); + }); + } + + pop_result tryPopUntil_(lock_t& lock, const TimePoint& until, TimeTuple& tuple) + { + TimePoint adjusted = until; + if (! super::mStorage.empty()) + { + // use whichever is earlier: the head item's timestamp, or + // the caller's limit + adjusted = min(std::get<0>(super::mStorage.front()), adjusted); + } + // now delegate to base-class tryPopUntil_() + pop_result popped; + while ((popped = pop_result(super::tryPopUntil_(lock, adjusted, tuple))) == WAITING) + { + // If super::tryPopUntil_() returns WAITING, it means there's + // a head item, but it's not yet time. But it's worth looping + // back to recheck. + } + return popped; + } + + /// tryPopUntil(DataTuple&) + template <typename Clock, typename Duration> + bool tryPopUntil(const std::chrono::time_point<Clock, Duration>& until, + DataTuple& tuple) + { + TimeTuple tt; + if (! tryPopUntil(until, tt)) + return false; + tuple = tuple_cdr(std::move(tt)); + return true; + } + + /// for when Args has exactly one type + template <typename Clock, typename Duration> + bool tryPopUntil(const std::chrono::time_point<Clock, Duration>& until, + typename std::tuple_element<1, TimeTuple>::type& value) + { + TimeTuple tt; + if (! tryPopUntil(until, tt)) + return false; + value = std::get<1>(std::move(tt)); + return true; + } + + /*------------------------------ etc. ------------------------------*/ + // We can't hide items that aren't yet ready because we can't traverse + // the underlying priority_queue: it has no iterators, only top(). So + // a consumer could observe size() > 0 and yet tryPop() returns false. + // Shrug, in a multi-consumer scenario that would be expected behavior. + using super::size; + // open/closed state + using super::close; + using super::isClosed; + using super::done; + + private: + // this method is called by base class pop_() every time we're + // considering whether to deliver the current head element + bool canPop(const TimeTuple& head) const override + { + // an item with a future timestamp isn't yet ready to pop + // (should we add some slop for overhead?) + return std::get<0>(head) <= Clock::now(); + } + }; + +} // namespace LL + +#endif /* ! defined(LL_THREADSAFESCHEDULE_H) */ diff --git a/indra/llcommon/timing.cpp b/indra/llcommon/timing.cpp deleted file mode 100644 index c2dc695ef3..0000000000 --- a/indra/llcommon/timing.cpp +++ /dev/null @@ -1,25 +0,0 @@ -/** - * @file timing.cpp - * @brief This file will be deprecated in the future. - * - * $LicenseInfo:firstyear=2000&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ diff --git a/indra/llcommon/tuple.h b/indra/llcommon/tuple.h new file mode 100644 index 0000000000..bfe7e3c2ba --- /dev/null +++ b/indra/llcommon/tuple.h @@ -0,0 +1,84 @@ +/** + * @file tuple.h + * @author Nat Goodspeed + * @date 2021-10-04 + * @brief A couple tuple utilities + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_TUPLE_H) +#define LL_TUPLE_H + +#include <tuple> +#include <type_traits> // std::remove_reference +#include <utility> // std::pair + +/** + * tuple_cons() behaves like LISP cons: it uses std::tuple_cat() to prepend a + * new item of arbitrary type to an existing std::tuple. + */ +template <typename First, typename... Rest, typename Tuple_=std::tuple<Rest...>> +auto tuple_cons(First&& first, Tuple_&& rest) +{ + // All we need to do is make a tuple containing 'first', and let + // tuple_cat() do the hard part. + return std::tuple_cat(std::tuple<First>(std::forward<First>(first)), + std::forward<Tuple_>(rest)); +} + +/** + * tuple_car() behaves like LISP car: it extracts the first item from a + * std::tuple. + */ +template <typename... Args, typename Tuple_=std::tuple<Args...>> +auto tuple_car(Tuple_&& tuple) +{ + return std::get<0>(std::forward<Tuple_>(tuple)); +} + +/** + * tuple_cdr() behaves like LISP cdr: it returns a new tuple containing + * everything BUT the first item. + */ +// derived from https://stackoverflow.com/a/24046437 +template <typename Tuple, std::size_t... Indices> +auto tuple_cdr_(Tuple&& tuple, const std::index_sequence<Indices...>) +{ + // Given an index sequence from [0..N-1), extract tuple items [1..N) + return std::make_tuple(std::get<Indices+1u>(std::forward<Tuple>(tuple))...); +} + +template <typename Tuple> +auto tuple_cdr(Tuple&& tuple) +{ + return tuple_cdr_( + std::forward<Tuple>(tuple), + // Pass helper function an index sequence one item shorter than tuple + std::make_index_sequence< + std::tuple_size< + // tuple_size doesn't like reference types + typename std::remove_reference<Tuple>::type + >::value - 1u> + ()); +} + +/** + * tuple_split(), the opposite of tuple_cons(), has no direct analog in LISP. + * It returns a std::pair of tuple_car(), tuple_cdr(). We could call this + * function tuple_car_cdr(), or tuple_slice() or some such. But tuple_split() + * feels more descriptive. + */ +template <typename... Args, typename Tuple_=std::tuple<Args...>> +auto tuple_split(Tuple_&& tuple) +{ + // We're not really worried about forwarding multiple times a tuple that + // might contain move-only items, because the implementation above only + // applies std::get() exactly once to each item. + return std::make_pair(tuple_car(std::forward<Tuple_>(tuple)), + tuple_cdr(std::forward<Tuple_>(tuple))); +} + +#endif /* ! defined(LL_TUPLE_H) */ diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp new file mode 100644 index 0000000000..114aeea1f3 --- /dev/null +++ b/indra/llcommon/workqueue.cpp @@ -0,0 +1,138 @@ +/** + * @file workqueue.cpp + * @author Nat Goodspeed + * @date 2021-10-06 + * @brief Implementation for WorkQueue. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "workqueue.h" +// STL headers +// std headers +// external library headers +// other Linden headers +#include "llcoros.h" +#include LLCOROS_MUTEX_HEADER +#include "llerror.h" +#include "llexception.h" +#include "stringize.h" + +using Mutex = LLCoros::Mutex; +using Lock = LLCoros::LockType; + +LL::WorkQueue::WorkQueue(const std::string& name): + super(makeName(name)) +{ + // TODO: register for "LLApp" events so we can implicitly close() on + // viewer shutdown. +} + +void LL::WorkQueue::close() +{ + mQueue.close(); +} + +bool LL::WorkQueue::isClosed() +{ + return mQueue.isClosed(); +} + +bool LL::WorkQueue::done() +{ + return mQueue.done(); +} + +void LL::WorkQueue::runUntilClose() +{ + try + { + for (;;) + { + callWork(mQueue.pop()); + } + } + catch (const Queue::Closed&) + { + } +} + +bool LL::WorkQueue::runPending() +{ + for (Work work; mQueue.tryPop(work); ) + { + callWork(work); + } + return ! mQueue.done(); +} + +bool LL::WorkQueue::runOne() +{ + Work work; + if (mQueue.tryPop(work)) + { + callWork(work); + } + return ! mQueue.done(); +} + +bool LL::WorkQueue::runUntil(const TimePoint& until) +{ + // Should we subtract some slop to allow for typical Work execution time? + // How much slop? + Work work; + while (TimePoint::clock::now() < until && mQueue.tryPopUntil(until, work)) + { + callWork(work); + } + return ! mQueue.done(); +} + +std::string LL::WorkQueue::makeName(const std::string& name) +{ + if (! name.empty()) + return name; + + static U32 discriminator = 0; + static Mutex mutex; + U32 num; + { + // Protect discriminator from concurrent access by different threads. + // It can't be thread_local, else two racing threads will come up with + // the same name. + Lock lk(mutex); + num = discriminator++; + } + return STRINGIZE("WorkQueue" << num); +} + +void LL::WorkQueue::callWork(const Queue::DataTuple& work) +{ + // ThreadSafeSchedule::pop() always delivers a tuple, even when + // there's only one data field per item, as for us. + callWork(std::get<0>(work)); +} + +void LL::WorkQueue::callWork(const Work& work) +{ + try + { + work(); + } + catch (...) + { + // No matter what goes wrong with any individual work item, the worker + // thread must go on! Log our own instance name with the exception. + LOG_UNHANDLED_EXCEPTION(getKey()); + } +} + +void LL::WorkQueue::error(const std::string& msg) +{ + LL_ERRS("WorkQueue") << msg << LL_ENDL; +} diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h new file mode 100644 index 0000000000..cfae2019dc --- /dev/null +++ b/indra/llcommon/workqueue.h @@ -0,0 +1,334 @@ +/** + * @file workqueue.h + * @author Nat Goodspeed + * @date 2021-09-30 + * @brief Queue used for inter-thread work passing. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_WORKQUEUE_H) +#define LL_WORKQUEUE_H + +#include "llinstancetracker.h" +#include "threadsafeschedule.h" +#include <chrono> +#include <functional> // std::function +#include <queue> +#include <string> +#include <utility> // std::pair +#include <vector> + +namespace LL +{ + /** + * A typical WorkQueue has a string name that can be used to find it. + */ + class WorkQueue: public LLInstanceTracker<WorkQueue, std::string> + { + private: + using super = LLInstanceTracker<WorkQueue, std::string>; + + public: + using Work = std::function<void()>; + + private: + using Queue = ThreadSafeSchedule<Work>; + // helper for postEvery() + template <typename Rep, typename Period, typename CALLABLE> + class BackJack; + + public: + using TimePoint = Queue::TimePoint; + using TimedWork = Queue::TimeTuple; + using Closed = Queue::Closed; + + /** + * You may omit the WorkQueue name, in which case a unique name is + * synthesized; for practical purposes that makes it anonymous. + */ + WorkQueue(const std::string& name = std::string()); + + /** + * Since the point of WorkQueue is to pass work to some other worker + * thread(s) asynchronously, it's important that the WorkQueue continue + * to exist until the worker thread(s) have drained it. To communicate + * that it's time for them to quit, close() the queue. + */ + void close(); + + /// producer end: are we prevented from pushing any additional items? + bool isClosed(); + /// consumer end: are we done, is the queue entirely drained? + bool done(); + + /*---------------------- fire and forget API -----------------------*/ + + /// fire-and-forget, but at a particular (future?) time + template <typename CALLABLE> + void post(const TimePoint& time, CALLABLE&& callable) + { + // Defer reifying an arbitrary CALLABLE until we hit this method. + // All other methods should accept CALLABLEs of arbitrary type to + // avoid multiple levels of std::function indirection. + mQueue.push(TimedWork(time, std::move(callable))); + } + + /// fire-and-forget + template <typename CALLABLE> + void post(CALLABLE&& callable) + { + // We use TimePoint::clock::now() instead of TimePoint's + // representation of the epoch because this WorkQueue may contain + // a mix of past-due TimedWork items and TimedWork items scheduled + // for the future. Sift this new item into the correct place. + post(TimePoint::clock::now(), std::move(callable)); + } + + /** + * Launch a callable returning bool that will trigger repeatedly at + * specified interval, until the callable returns false. + * + * If you need to signal that callable from outside, DO NOT bind a + * reference to a simple bool! That's not thread-safe. Instead, bind + * an LLCond variant, e.g. LLOneShotCond or LLBoolCond. + */ + template <typename Rep, typename Period, typename CALLABLE> + void postEvery(const std::chrono::duration<Rep, Period>& interval, + CALLABLE&& callable); + + /*------------------------- handshake API --------------------------*/ + + /** + * Post work to another WorkQueue to be run at a specified time, + * requesting a specific callback to be run on this WorkQueue on + * completion. + * + * Returns true if able to post, false if the other WorkQueue is + * inaccessible. + */ + // Apparently some Microsoft header file defines a macro CALLBACK? The + // natural template argument name CALLBACK produces very weird Visual + // Studio compile errors that seem utterly unrelated to this source + // code. + template <typename CALLABLE, typename FOLLOWUP> + bool postTo(WorkQueue::weak_t target, + const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) + { + // We're being asked to post to the WorkQueue at target. + // target is a weak_ptr: have to lock it to check it. + auto tptr = target.lock(); + if (! tptr) + // can't post() if the target WorkQueue has been destroyed + return false; + + // Here we believe target WorkQueue still exists. Post to it a + // lambda that packages our callable, our callback and a weak_ptr + // to this originating WorkQueue. + tptr->post( + time, + [reply = super::getWeak(), + callable = std::move(callable), + callback = std::move(callback)] + () + { + // Call the callable in any case -- but to minimize + // copying the result, immediately bind it into a reply + // lambda. The reply lambda also binds the original + // callback, so that when we, the originating WorkQueue, + // finally receive and process the reply lambda, we'll + // call the bound callback with the bound result -- on the + // same thread that originally called postTo(). + auto rlambda = + [result = callable(), + callback = std::move(callback)] + () + { callback(std::move(result)); }; + // Check if this originating WorkQueue still exists. + // Remember, the outer lambda is now running on a thread + // servicing the target WorkQueue, and real time has + // elapsed since postTo()'s tptr->post() call. + // reply is a weak_ptr: have to lock it to check it. + auto rptr = reply.lock(); + if (rptr) + { + // Only post reply lambda if the originating WorkQueue + // still exists. If not -- who would we tell? Log it? + try + { + rptr->post(std::move(rlambda)); + } + catch (const Closed&) + { + // Originating WorkQueue might still exist, but + // might be Closed. Same thing: just discard the + // callback. + } + } + }); + // looks like we were able to post() + return true; + } + + /** + * Post work to another WorkQueue, requesting a specific callback to + * be run on this WorkQueue on completion. + * + * Returns true if able to post, false if the other WorkQueue is + * inaccessible. + */ + template <typename CALLABLE, typename FOLLOWUP> + bool postTo(WorkQueue::weak_t target, + CALLABLE&& callable, FOLLOWUP&& callback) + { + return postTo(target, TimePoint::clock::now(), std::move(callable), std::move(callback)); + } + + /*--------------------------- worker API ---------------------------*/ + + /** + * runUntilClose() pulls TimedWork items off this WorkQueue until the + * queue is closed, at which point it returns. This would be the + * typical entry point for a simple worker thread. + */ + void runUntilClose(); + + /** + * runPending() runs all TimedWork items that are ready to run. It + * returns true if the queue remains open, false if the queue has been + * closed. This could be used by a thread whose primary purpose is to + * serve the queue, but also wants to do other things with its idle time. + */ + bool runPending(); + + /** + * runOne() runs at most one ready TimedWork item -- zero if none are + * ready. It returns true if the queue remains open, false if the + * queue has been closed. + */ + bool runOne(); + + /** + * runFor() runs a subset of ready TimedWork items, until the + * timeslice has been exceeded. It returns true if the queue remains + * open, false if the queue has been closed. This could be used by a + * busy main thread to lend a bounded few CPU cycles to this WorkQueue + * without risking the WorkQueue blowing out the length of any one + * frame. + */ + template <typename Rep, typename Period> + bool runFor(const std::chrono::duration<Rep, Period>& timeslice) + { + return runUntil(TimePoint::clock::now() + timeslice); + } + + /** + * runUntil() is just like runFor(), only with a specific end time + * instead of a timeslice duration. + */ + bool runUntil(const TimePoint& until); + + private: + static void error(const std::string& msg); + static std::string makeName(const std::string& name); + void callWork(const Queue::DataTuple& work); + void callWork(const Work& work); + Queue mQueue; + }; + + /** + * BackJack is, in effect, a hand-rolled lambda, binding a WorkQueue, a + * CALLABLE that returns bool, a TimePoint and an interval at which to + * relaunch it. As long as the callable continues returning true, BackJack + * keeps resubmitting it to the target WorkQueue. + */ + // Why is BackJack a class and not a lambda? Because, unlike a lambda, a + // class method gets its own 'this' pointer -- which we need to resubmit + // the whole BackJack callable. + template <typename Rep, typename Period, typename CALLABLE> + class WorkQueue::BackJack + { + public: + // bind the desired data + BackJack(WorkQueue::weak_t target, + const WorkQueue::TimePoint& start, + const std::chrono::duration<Rep, Period>& interval, + CALLABLE&& callable): + mTarget(target), + mStart(start), + mInterval(interval), + mCallable(std::move(callable)) + {} + + // Call by target WorkQueue -- note that although WE require a + // callable returning bool, WorkQueue wants a void callable. We + // consume the bool. + void operator()() + { + // If mCallable() throws an exception, don't catch it here: if it + // throws once, it's likely to throw every time, so it's a waste + // of time to arrange to call it again. + if (mCallable()) + { + // Modify mStart to the new start time we desire. If we simply + // added mInterval to now, we'd get actual timings of + // (mInterval + slop), where 'slop' is the latency between the + // previous mStart and the WorkQueue actually calling us. + // Instead, add mInterval to mStart so that at least we + // register our intent to fire at exact mIntervals. + mStart += mInterval; + + // We're being called at this moment by the target WorkQueue. + // Assume it still exists, rather than checking the result of + // lock(). + // Resubmit the whole *this callable: that's why we're a class + // rather than a lambda. Allow moving *this so we can carry a + // move-only callable; but naturally this statement must be + // the last time we reference this instance, which may become + // moved-from. + try + { + mTarget.lock()->post(mStart, std::move(*this)); + } + catch (const Closed&) + { + // Once this queue is closed, oh well, just stop + } + } + } + + private: + WorkQueue::weak_t mTarget; + WorkQueue::TimePoint mStart; + std::chrono::duration<Rep, Period> mInterval; + CALLABLE mCallable; + }; + + template <typename Rep, typename Period, typename CALLABLE> + void WorkQueue::postEvery(const std::chrono::duration<Rep, Period>& interval, + CALLABLE&& callable) + { + if (interval.count() <= 0) + { + // It's essential that postEvery() be called with a positive + // interval, since each call to BackJack posts another instance of + // itself at (start + interval) and we order by target time. A + // zero or negative interval would result in that BackJack + // instance going to the head of the queue every time, immediately + // ready to run. Effectively that would produce an infinite loop, + // a denial of service on this WorkQueue. + error("postEvery(interval) may not be 0"); + } + // Instantiate and post a suitable BackJack, binding a weak_ptr to + // self, the current time, the desired interval and the desired + // callable. + post( + BackJack<Rep, Period, CALLABLE>( + getWeak(), TimePoint::clock::now(), interval, std::move(callable))); + } + +} // namespace LL + +#endif /* ! defined(LL_WORKQUEUE_H) */ diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h index 7ba347062f..5291a05607 100644 --- a/indra/llmath/llmatrix4a.h +++ b/indra/llmath/llmatrix4a.h @@ -36,6 +36,26 @@ class LLMatrix4a public: LL_ALIGN_16(LLVector4a mMatrix[4]); + LLMatrix4a() + { + + } + + explicit LLMatrix4a(const LLMatrix4& val) + { + loadu(val); + } + + inline F32* getF32ptr() + { + return (F32*) &mMatrix; + } + + inline const F32* getF32ptr() const + { + return (F32*)&mMatrix; + } + inline void clear() { mMatrix[0].clear(); @@ -44,6 +64,14 @@ public: mMatrix[3].clear(); } + inline void setIdentity() + { + mMatrix[0].set(1.f, 0.f, 0.f, 0.f); + mMatrix[1].set(0.f, 1.f, 0.f, 0.f); + mMatrix[2].set(0.f, 0.f, 1.f, 0.f); + mMatrix[3].set(0.f, 0.f, 0.f, 1.f); + } + inline void loadu(const LLMatrix4& src) { mMatrix[0] = _mm_loadu_ps(src.mMatrix[0]); @@ -105,7 +133,7 @@ public: mMatrix[3].setAdd(a.mMatrix[3],d3); } - inline void rotate(const LLVector4a& v, LLVector4a& res) + inline void rotate(const LLVector4a& v, LLVector4a& res) const { LLVector4a y,z; @@ -151,6 +179,8 @@ public: { affineTransformSSE(v,res); } + + const LLVector4a& getTranslation() const { return mMatrix[3]; } }; inline LLVector4a rowMul(const LLVector4a &row, const LLMatrix4a &mat) @@ -176,6 +206,15 @@ inline void matMul(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res) res.mMatrix[3] = row3; } +//Faster version of matMul wehere res must not be a or b +inline void matMulUnsafe(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res) +{ + res.mMatrix[0] = rowMul(a.mMatrix[0], b); + res.mMatrix[1] = rowMul(a.mMatrix[1], b); + res.mMatrix[2] = rowMul(a.mMatrix[2], b); + res.mMatrix[3] = rowMul(a.mMatrix[3], b); +} + inline std::ostream& operator<<(std::ostream& s, const LLMatrix4a& m) { s << "[" << m.mMatrix[0] << ", " << m.mMatrix[1] << ", " << m.mMatrix[2] << ", " << m.mMatrix[3] << "]"; diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h index 27abf39537..5a02928374 100644 --- a/indra/llmath/llvector4a.h +++ b/indra/llmath/llvector4a.h @@ -46,10 +46,9 @@ class LLRotation; // of this writing, July 08, 2010) about getting it implemented before you resort to // LLVector3/LLVector4. ///////////////////////////////// -struct LLVector4a; LL_ALIGN_PREFIX(16) -struct LLVector4a +class LLVector4a { public: @@ -138,10 +137,10 @@ public: // BASIC GET/SET //////////////////////////////////// - // Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) + // Return a "this" as an F32 pointer. inline F32* getF32ptr(); - // Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) + // Return a "this" as a const F32 pointer. inline const F32* const getF32ptr() const; // Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl index 69d3d01efe..8be1c1b114 100644 --- a/indra/llmath/llvector4a.inl +++ b/indra/llmath/llvector4a.inl @@ -58,13 +58,13 @@ inline void LLVector4a::store4a(F32* dst) const // BASIC GET/SET //////////////////////////////////// -// Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) +// Return a "this" as an F32 pointer. F32* LLVector4a::getF32ptr() { return (F32*) &mQ; } -// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) +// Return a "this" as a const F32 pointer. const F32* const LLVector4a::getF32ptr() const { return (const F32* const) &mQ; diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index e085fa6ada..130f30bedc 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -383,6 +383,7 @@ public: virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch) { //this is a depth first traversal, so it's safe to assum all children have complete //bounding data + LL_PROFILE_ZONE_SCOPED LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0); @@ -822,6 +823,8 @@ S32 LLProfile::getNumPoints(const LLProfileParams& params, BOOL path_open,F32 de BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split, BOOL is_sculpted, S32 sculpt_size) { + LL_PROFILE_ZONE_SCOPED + if ((!mDirty) && (!is_sculpted)) { return FALSE; @@ -1302,6 +1305,8 @@ S32 LLPath::getNumNGonPoints(const LLPathParams& params, S32 sides, F32 startOff void LLPath::genNGon(const LLPathParams& params, S32 sides, F32 startOff, F32 end_scale, F32 twist_scale) { + LL_PROFILE_ZONE_SCOPED + // Generates a circular path, starting at (1, 0, 0), counterclockwise along the xz plane. static const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f }; @@ -1536,6 +1541,8 @@ S32 LLPath::getNumPoints(const LLPathParams& params, F32 detail) BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, BOOL is_sculpted, S32 sculpt_size) { + LL_PROFILE_ZONE_SCOPED + if ((!mDirty) && (!is_sculpted)) { return FALSE; @@ -2112,6 +2119,8 @@ LLVolume::~LLVolume() BOOL LLVolume::generate() { + LL_PROFILE_ZONE_SCOPED + LL_CHECK_MEMORY llassert_always(mProfilep); @@ -2370,6 +2379,8 @@ bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) { + LL_PROFILE_ZONE_SCOPED + //input stream is now pointing at a zlib compressed block of LLSD //decompress block LLSD mdl; @@ -2755,6 +2766,8 @@ S32 LLVolume::getNumFaces() const void LLVolume::createVolumeFaces() { + LL_PROFILE_ZONE_SCOPED + if (mGenerateSingleFace) { // do nothing @@ -3720,6 +3733,8 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices, const LLMatrix3& norm_mat_in, S32 face_mask) { + LL_PROFILE_ZONE_SCOPED + LLMatrix4a mat; mat.loadu(mat_in); @@ -4846,6 +4861,8 @@ void LLVolumeFace::freeData() BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) { + LL_PROFILE_ZONE_SCOPED + //tree for this face is no longer valid delete mOctree; mOctree = NULL; @@ -5514,6 +5531,8 @@ bool LLVolumeFace::cacheOptimize() void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size) { + LL_PROFILE_ZONE_SCOPED + if (mOctree) { return; @@ -6287,6 +6306,8 @@ void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVe void LLVolumeFace::createTangents() { + LL_PROFILE_ZONE_SCOPED + if (!mTangents) { allocateTangents(mNumVertices); @@ -6482,6 +6503,8 @@ void LLVolumeFace::fillFromLegacyData(std::vector<LLVolumeFace::VertexData>& v, BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { + LL_PROFILE_ZONE_SCOPED + LL_CHECK_MEMORY BOOL flat = mTypeMask & FLAT_MASK; @@ -6974,6 +6997,8 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVector4a *normal, const LLVector2 *texcoord, U32 triangleCount, const U16* index_array, LLVector4a *tangent) { + LL_PROFILE_ZONE_SCOPED + //LLVector4a *tan1 = new LLVector4a[vertexCount * 2]; LLVector4a* tan1 = (LLVector4a*) ll_aligned_malloc_16(vertexCount*2*sizeof(LLVector4a)); // new(tan1) LLVector4a; diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp index 3baf1bad18..6e40dae30b 100644 --- a/indra/llmath/m4math.cpp +++ b/indra/llmath/m4math.cpp @@ -32,8 +32,7 @@ #include "m4math.h" #include "m3math.h" #include "llquaternion.h" - - +#include "llmatrix4a.h" // LLMatrix4 @@ -115,6 +114,12 @@ LLMatrix4::LLMatrix4(const LLQuaternion &q) *this = initRotation(q); } +LLMatrix4::LLMatrix4(const LLMatrix4a& mat) + : LLMatrix4(mat.getF32ptr()) +{ + +} + LLMatrix4::LLMatrix4(const LLQuaternion &q, const LLVector4 &pos) { *this = initRotTrans(q, pos); diff --git a/indra/llmath/m4math.h b/indra/llmath/m4math.h index bf60adb9b6..b9da970cde 100644 --- a/indra/llmath/m4math.h +++ b/indra/llmath/m4math.h @@ -32,6 +32,7 @@ class LLVector4; class LLMatrix3; class LLQuaternion; +class LLMatrix4a; // NOTA BENE: Currently assuming a right-handed, x-forward, y-left, z-up universe @@ -104,6 +105,7 @@ public: explicit LLMatrix4(const F32 *mat); // Initializes Matrix to values in mat explicit LLMatrix4(const LLMatrix3 &mat); // Initializes Matrix to values in mat and sets position to (0,0,0) explicit LLMatrix4(const LLQuaternion &q); // Initializes Matrix with rotation q and sets position to (0,0,0) + explicit LLMatrix4(const LLMatrix4a& mat); LLMatrix4(const LLMatrix3 &mat, const LLVector4 &pos); // Initializes Matrix to values in mat and pos diff --git a/indra/llmath/v3math.cpp b/indra/llmath/v3math.cpp index b04c67d926..93010d2250 100644 --- a/indra/llmath/v3math.cpp +++ b/indra/llmath/v3math.cpp @@ -316,6 +316,12 @@ LLVector3::LLVector3(const LLVector4 &vec) mV[VZ] = (F32)vec.mV[VZ]; } +LLVector3::LLVector3(const LLVector4a& vec) + : LLVector3(vec.getF32ptr()) +{ + +} + LLVector3::LLVector3(const LLSD& sd) { setValue(sd); diff --git a/indra/llmath/v3math.h b/indra/llmath/v3math.h index 6f857d7061..068f489020 100644 --- a/indra/llmath/v3math.h +++ b/indra/llmath/v3math.h @@ -33,6 +33,7 @@ #include "llsd.h" class LLVector2; class LLVector4; +class LLVector4a; class LLMatrix3; class LLMatrix4; class LLVector3d; @@ -62,7 +63,9 @@ class LLVector3 explicit LLVector3(const LLVector2 &vec); // Initializes LLVector3 to (vec[0]. vec[1], 0) explicit LLVector3(const LLVector3d &vec); // Initializes LLVector3 to (vec[0]. vec[1], vec[2]) explicit LLVector3(const LLVector4 &vec); // Initializes LLVector4 to (vec[0]. vec[1], vec[2]) - explicit LLVector3(const LLSD& sd); + explicit LLVector3(const LLVector4a& vec); // Initializes LLVector4 to (vec[0]. vec[1], vec[2]) + explicit LLVector3(const LLSD& sd); + LLSD getValue() const; diff --git a/indra/llprimitive/lldaeloader.cpp b/indra/llprimitive/lldaeloader.cpp index dfa29fb539..8343de0cbc 100644 --- a/indra/llprimitive/lldaeloader.cpp +++ b/indra/llprimitive/lldaeloader.cpp @@ -1173,17 +1173,19 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do LLMeshSkinInfo& skin_info = model->mSkinInfo; + LLMatrix4 mat; for (int i = 0; i < 4; i++) { for(int j = 0; j < 4; j++) { - skin_info.mBindShapeMatrix.mMatrix[i][j] = dom_value[i + j*4]; + mat.mMatrix[i][j] = dom_value[i + j*4]; } } - LLMatrix4 trans = normalized_transformation; - trans *= skin_info.mBindShapeMatrix; - skin_info.mBindShapeMatrix = trans; + skin_info.mBindShapeMatrix.loadu(mat); + + LLMatrix4a trans(normalized_transformation); + matMul(trans, skin_info.mBindShapeMatrix, skin_info.mBindShapeMatrix); } @@ -1401,7 +1403,7 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do mat.mMatrix[i][j] = transform[k*16 + i + j*4]; } } - model->mSkinInfo.mInvBindMatrix.push_back(mat); + model->mSkinInfo.mInvBindMatrix.push_back(LLMatrix4a(mat)); } } } @@ -1475,9 +1477,9 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do if (mJointMap.find(lookingForJoint) != mJointMap.end() && model->mSkinInfo.mInvBindMatrix.size() > i) { - LLMatrix4 newInverse = model->mSkinInfo.mInvBindMatrix[i]; + LLMatrix4 newInverse = LLMatrix4(model->mSkinInfo.mInvBindMatrix[i].getF32ptr()); newInverse.setTranslation( mJointList[lookingForJoint].getTranslation() ); - model->mSkinInfo.mAlternateBindMatrix.push_back( newInverse ); + model->mSkinInfo.mAlternateBindMatrix.push_back( LLMatrix4a(newInverse) ); } else { diff --git a/indra/llprimitive/llmodel.cpp b/indra/llprimitive/llmodel.cpp index 702a1b5238..a23b991f1d 100644 --- a/indra/llprimitive/llmodel.cpp +++ b/indra/llprimitive/llmodel.cpp @@ -1396,7 +1396,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin) } } - mInvBindMatrix.push_back(mat); + mInvBindMatrix.push_back(LLMatrix4a(mat)); } if (mJointNames.size() != mInvBindMatrix.size()) @@ -1410,13 +1410,15 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin) if (skin.has("bind_shape_matrix")) { + LLMatrix4 mat; for (U32 j = 0; j < 4; j++) { for (U32 k = 0; k < 4; k++) { - mBindShapeMatrix.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal(); + mat.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal(); } } + mBindShapeMatrix.loadu(mat); } if (skin.has("alt_inverse_bind_matrix")) @@ -1432,7 +1434,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin) } } - mAlternateBindMatrix.push_back(mat); + mAlternateBindMatrix.push_back(LLMatrix4a(mat)); } } diff --git a/indra/llprimitive/llmodel.h b/indra/llprimitive/llmodel.h index 51fa2f8079..96d4582b4f 100644 --- a/indra/llprimitive/llmodel.h +++ b/indra/llprimitive/llmodel.h @@ -33,6 +33,8 @@ #include "m4math.h" #include <queue> +#include <boost/align/aligned_allocator.hpp> + class daeElement; class domMesh; @@ -49,10 +51,11 @@ public: LLUUID mMeshID; std::vector<std::string> mJointNames; mutable std::vector<S32> mJointNums; - std::vector<LLMatrix4> mInvBindMatrix; - std::vector<LLMatrix4> mAlternateBindMatrix; + typedef std::vector<LLMatrix4a, boost::alignment::aligned_allocator<LLMatrix4a, 16>> matrix_list_t; + matrix_list_t mInvBindMatrix; + matrix_list_t mAlternateBindMatrix; - LLMatrix4 mBindShapeMatrix; + LLMatrix4a mBindShapeMatrix; float mPelvisOffset; bool mLockScaleIfJointPosition; bool mInvalidJointsScrubbed; diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp index 43fedeca64..673f6cb6df 100644 --- a/indra/llrender/llgl.cpp +++ b/indra/llrender/llgl.cpp @@ -434,9 +434,6 @@ LLGLManager::LLGLManager() : mHasMapBufferRange(FALSE), mHasFlushBufferRange(FALSE), mHasPBuffer(FALSE), - mHasShaderObjects(FALSE), - mHasVertexShader(FALSE), - mHasFragmentShader(FALSE), mNumTextureImageUnits(0), mHasOcclusionQuery(FALSE), mHasTimerQuery(FALSE), @@ -775,14 +772,9 @@ bool LLGLManager::initGL() stop_glerror(); - stop_glerror(); - - if (mHasFragmentShader) - { - GLint num_tex_image_units; - glGetIntegerv(GL_MAX_TEXTURE_IMAGE_UNITS_ARB, &num_tex_image_units); - mNumTextureImageUnits = llmin(num_tex_image_units, 32); - } + GLint num_tex_image_units; + glGetIntegerv(GL_MAX_TEXTURE_IMAGE_UNITS_ARB, &num_tex_image_units); + mNumTextureImageUnits = llmin(num_tex_image_units, 32); if (LLRender::sGLCoreProfile) { @@ -975,9 +967,9 @@ void LLGLManager::asLLSD(LLSD& info) info["has_map_buffer_range"] = mHasMapBufferRange; info["has_flush_buffer_range"] = mHasFlushBufferRange; info["has_pbuffer"] = mHasPBuffer; - info["has_shader_objects"] = mHasShaderObjects; - info["has_vertex_shader"] = mHasVertexShader; - info["has_fragment_shader"] = mHasFragmentShader; + info["has_shader_objects"] = std::string("Assumed TRUE"); // was mHasShaderObjects; + info["has_vertex_shader"] = std::string("Assumed TRUE"); // was mHasVertexShader; + info["has_fragment_shader"] = std::string("Assumed TRUE"); // was mHasFragmentShader; info["num_texture_image_units"] = mNumTextureImageUnits; info["has_occlusion_query"] = mHasOcclusionQuery; info["has_timer_query"] = mHasTimerQuery; @@ -1083,9 +1075,6 @@ void LLGLManager::initExtensions() mHasCubeMap = FALSE; mHasOcclusionQuery = FALSE; mHasPointParameters = FALSE; - mHasShaderObjects = FALSE; - mHasVertexShader = FALSE; - mHasFragmentShader = FALSE; mHasTextureRectangle = FALSE; #else // LL_MESA_HEADLESS //important, gGLHExts.mSysExts is uninitialized until after glh_init_extensions is called mHasMultitexture = glh_init_extensions("GL_ARB_multitexture"); @@ -1143,10 +1132,6 @@ void LLGLManager::initExtensions() #if !LL_DARWIN mHasPointParameters = !mIsATI && ExtensionExists("GL_ARB_point_parameters", gGLHExts.mSysExts); #endif - mHasShaderObjects = ExtensionExists("GL_ARB_shader_objects", gGLHExts.mSysExts) && (LLRender::sGLCoreProfile || ExtensionExists("GL_ARB_shading_language_100", gGLHExts.mSysExts)); - mHasVertexShader = ExtensionExists("GL_ARB_vertex_program", gGLHExts.mSysExts) && ExtensionExists("GL_ARB_vertex_shader", gGLHExts.mSysExts) - && (LLRender::sGLCoreProfile || ExtensionExists("GL_ARB_shading_language_100", gGLHExts.mSysExts)); - mHasFragmentShader = ExtensionExists("GL_ARB_fragment_shader", gGLHExts.mSysExts) && (LLRender::sGLCoreProfile || ExtensionExists("GL_ARB_shading_language_100", gGLHExts.mSysExts)); #endif #if LL_LINUX @@ -1169,9 +1154,6 @@ void LLGLManager::initExtensions() mHasCubeMap = FALSE; mHasOcclusionQuery = FALSE; mHasPointParameters = FALSE; - mHasShaderObjects = FALSE; - mHasVertexShader = FALSE; - mHasFragmentShader = FALSE; LL_WARNS("RenderInit") << "GL extension support DISABLED via LL_GL_NOEXT" << LL_ENDL; } else if (getenv("LL_GL_BASICEXT")) /* Flawfinder: ignore */ @@ -1184,9 +1166,6 @@ void LLGLManager::initExtensions() mHasAnisotropic = FALSE; //mHasCubeMap = FALSE; // apparently fatal on Intel 915 & similar //mHasOcclusionQuery = FALSE; // source of many ATI system hangs - mHasShaderObjects = FALSE; - mHasVertexShader = FALSE; - mHasFragmentShader = FALSE; mHasBlendFuncSeparate = FALSE; LL_WARNS("RenderInit") << "GL extension support forced to SIMPLE level via LL_GL_BASICEXT" << LL_ENDL; } @@ -1208,9 +1187,6 @@ void LLGLManager::initExtensions() if (strchr(blacklist,'j')) mHasCubeMap = FALSE;//S // if (strchr(blacklist,'k')) mHasATIVAO = FALSE;//S if (strchr(blacklist,'l')) mHasOcclusionQuery = FALSE; - if (strchr(blacklist,'m')) mHasShaderObjects = FALSE;//S - if (strchr(blacklist,'n')) mHasVertexShader = FALSE;//S - if (strchr(blacklist,'o')) mHasFragmentShader = FALSE;//S if (strchr(blacklist,'p')) mHasPointParameters = FALSE;//S if (strchr(blacklist,'q')) mHasFramebufferObject = FALSE;//S if (strchr(blacklist,'r')) mHasDrawBuffers = FALSE;//S @@ -1257,18 +1233,6 @@ void LLGLManager::initExtensions() { LL_INFOS("RenderInit") << "Couldn't initialize GL_ARB_point_parameters" << LL_ENDL; } - if (!mHasShaderObjects) - { - LL_INFOS("RenderInit") << "Couldn't initialize GL_ARB_shader_objects" << LL_ENDL; - } - if (!mHasVertexShader) - { - LL_INFOS("RenderInit") << "Couldn't initialize GL_ARB_vertex_shader" << LL_ENDL; - } - if (!mHasFragmentShader) - { - LL_INFOS("RenderInit") << "Couldn't initialize GL_ARB_fragment_shader" << LL_ENDL; - } if (!mHasBlendFuncSeparate) { LL_INFOS("RenderInit") << "Couldn't initialize GL_EXT_blend_func_separate" << LL_ENDL; @@ -1436,134 +1400,132 @@ void LLGLManager::initExtensions() glPointParameterfARB = (PFNGLPOINTPARAMETERFARBPROC)GLH_EXT_GET_PROC_ADDRESS("glPointParameterfARB"); glPointParameterfvARB = (PFNGLPOINTPARAMETERFVARBPROC)GLH_EXT_GET_PROC_ADDRESS("glPointParameterfvARB"); } - if (mHasShaderObjects) - { - glDeleteObjectARB = (PFNGLDELETEOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDeleteObjectARB"); - glGetHandleARB = (PFNGLGETHANDLEARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetHandleARB"); - glDetachObjectARB = (PFNGLDETACHOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDetachObjectARB"); - glCreateShaderObjectARB = (PFNGLCREATESHADEROBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glCreateShaderObjectARB"); - glShaderSourceARB = (PFNGLSHADERSOURCEARBPROC) GLH_EXT_GET_PROC_ADDRESS("glShaderSourceARB"); - glCompileShaderARB = (PFNGLCOMPILESHADERARBPROC) GLH_EXT_GET_PROC_ADDRESS("glCompileShaderARB"); - glCreateProgramObjectARB = (PFNGLCREATEPROGRAMOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glCreateProgramObjectARB"); - glAttachObjectARB = (PFNGLATTACHOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glAttachObjectARB"); - glLinkProgramARB = (PFNGLLINKPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glLinkProgramARB"); - glUseProgramObjectARB = (PFNGLUSEPROGRAMOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUseProgramObjectARB"); - glValidateProgramARB = (PFNGLVALIDATEPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glValidateProgramARB"); - glUniform1fARB = (PFNGLUNIFORM1FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1fARB"); - glUniform2fARB = (PFNGLUNIFORM2FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2fARB"); - glUniform3fARB = (PFNGLUNIFORM3FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3fARB"); - glUniform4fARB = (PFNGLUNIFORM4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4fARB"); - glUniform1iARB = (PFNGLUNIFORM1IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1iARB"); - glUniform2iARB = (PFNGLUNIFORM2IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2iARB"); - glUniform3iARB = (PFNGLUNIFORM3IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3iARB"); - glUniform4iARB = (PFNGLUNIFORM4IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4iARB"); - glUniform1fvARB = (PFNGLUNIFORM1FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1fvARB"); - glUniform2fvARB = (PFNGLUNIFORM2FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2fvARB"); - glUniform3fvARB = (PFNGLUNIFORM3FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3fvARB"); - glUniform4fvARB = (PFNGLUNIFORM4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4fvARB"); - glUniform1ivARB = (PFNGLUNIFORM1IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1ivARB"); - glUniform2ivARB = (PFNGLUNIFORM2IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2ivARB"); - glUniform3ivARB = (PFNGLUNIFORM3IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3ivARB"); - glUniform4ivARB = (PFNGLUNIFORM4IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4ivARB"); - glUniformMatrix2fvARB = (PFNGLUNIFORMMATRIX2FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix2fvARB"); - glUniformMatrix3fvARB = (PFNGLUNIFORMMATRIX3FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix3fvARB"); - glUniformMatrix3x4fv = (PFNGLUNIFORMMATRIX3X4FVPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix3x4fv"); - glUniformMatrix4fvARB = (PFNGLUNIFORMMATRIX4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix4fvARB"); - glGetObjectParameterfvARB = (PFNGLGETOBJECTPARAMETERFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetObjectParameterfvARB"); - glGetObjectParameterivARB = (PFNGLGETOBJECTPARAMETERIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetObjectParameterivARB"); - glGetInfoLogARB = (PFNGLGETINFOLOGARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetInfoLogARB"); - glGetAttachedObjectsARB = (PFNGLGETATTACHEDOBJECTSARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetAttachedObjectsARB"); - glGetUniformLocationARB = (PFNGLGETUNIFORMLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetUniformLocationARB"); - glGetActiveUniformARB = (PFNGLGETACTIVEUNIFORMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetActiveUniformARB"); - glGetUniformfvARB = (PFNGLGETUNIFORMFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetUniformfvARB"); - glGetUniformivARB = (PFNGLGETUNIFORMIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetUniformivARB"); - glGetShaderSourceARB = (PFNGLGETSHADERSOURCEARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetShaderSourceARB"); - } - if (mHasVertexShader) - { - LL_INFOS() << "initExtensions() VertexShader-related procs..." << LL_ENDL; - - // nSight doesn't support use of ARB funcs that have been normalized in the API - if (!LLRender::sNsightDebugSupport) - { - glGetAttribLocationARB = (PFNGLGETATTRIBLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetAttribLocationARB"); - glBindAttribLocationARB = (PFNGLBINDATTRIBLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glBindAttribLocationARB"); - } - else - { - glGetAttribLocationARB = (PFNGLGETATTRIBLOCATIONARBPROC)GLH_EXT_GET_PROC_ADDRESS("glGetAttribLocation"); - glBindAttribLocationARB = (PFNGLBINDATTRIBLOCATIONARBPROC)GLH_EXT_GET_PROC_ADDRESS("glBindAttribLocation"); - } - - glGetActiveAttribARB = (PFNGLGETACTIVEATTRIBARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetActiveAttribARB"); - glVertexAttrib1dARB = (PFNGLVERTEXATTRIB1DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1dARB"); - glVertexAttrib1dvARB = (PFNGLVERTEXATTRIB1DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1dvARB"); - glVertexAttrib1fARB = (PFNGLVERTEXATTRIB1FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1fARB"); - glVertexAttrib1fvARB = (PFNGLVERTEXATTRIB1FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1fvARB"); - glVertexAttrib1sARB = (PFNGLVERTEXATTRIB1SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1sARB"); - glVertexAttrib1svARB = (PFNGLVERTEXATTRIB1SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1svARB"); - glVertexAttrib2dARB = (PFNGLVERTEXATTRIB2DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2dARB"); - glVertexAttrib2dvARB = (PFNGLVERTEXATTRIB2DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2dvARB"); - glVertexAttrib2fARB = (PFNGLVERTEXATTRIB2FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2fARB"); - glVertexAttrib2fvARB = (PFNGLVERTEXATTRIB2FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2fvARB"); - glVertexAttrib2sARB = (PFNGLVERTEXATTRIB2SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2sARB"); - glVertexAttrib2svARB = (PFNGLVERTEXATTRIB2SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2svARB"); - glVertexAttrib3dARB = (PFNGLVERTEXATTRIB3DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3dARB"); - glVertexAttrib3dvARB = (PFNGLVERTEXATTRIB3DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3dvARB"); - glVertexAttrib3fARB = (PFNGLVERTEXATTRIB3FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3fARB"); - glVertexAttrib3fvARB = (PFNGLVERTEXATTRIB3FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3fvARB"); - glVertexAttrib3sARB = (PFNGLVERTEXATTRIB3SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3sARB"); - glVertexAttrib3svARB = (PFNGLVERTEXATTRIB3SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3svARB"); - glVertexAttrib4nbvARB = (PFNGLVERTEXATTRIB4NBVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nbvARB"); - glVertexAttrib4nivARB = (PFNGLVERTEXATTRIB4NIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nivARB"); - glVertexAttrib4nsvARB = (PFNGLVERTEXATTRIB4NSVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nsvARB"); - glVertexAttrib4nubARB = (PFNGLVERTEXATTRIB4NUBARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nubARB"); - glVertexAttrib4nubvARB = (PFNGLVERTEXATTRIB4NUBVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nubvARB"); - glVertexAttrib4nuivARB = (PFNGLVERTEXATTRIB4NUIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nuivARB"); - glVertexAttrib4nusvARB = (PFNGLVERTEXATTRIB4NUSVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nusvARB"); - glVertexAttrib4bvARB = (PFNGLVERTEXATTRIB4BVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4bvARB"); - glVertexAttrib4dARB = (PFNGLVERTEXATTRIB4DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4dARB"); - glVertexAttrib4dvARB = (PFNGLVERTEXATTRIB4DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4dvARB"); - glVertexAttrib4fARB = (PFNGLVERTEXATTRIB4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4fARB"); - glVertexAttrib4fvARB = (PFNGLVERTEXATTRIB4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4fvARB"); - glVertexAttrib4ivARB = (PFNGLVERTEXATTRIB4IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4ivARB"); - glVertexAttrib4sARB = (PFNGLVERTEXATTRIB4SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4sARB"); - glVertexAttrib4svARB = (PFNGLVERTEXATTRIB4SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4svARB"); - glVertexAttrib4ubvARB = (PFNGLVERTEXATTRIB4UBVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4ubvARB"); - glVertexAttrib4uivARB = (PFNGLVERTEXATTRIB4UIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4uivARB"); - glVertexAttrib4usvARB = (PFNGLVERTEXATTRIB4USVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4usvARB"); - glVertexAttribPointerARB = (PFNGLVERTEXATTRIBPOINTERARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttribPointerARB"); - glVertexAttribIPointer = (PFNGLVERTEXATTRIBIPOINTERPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttribIPointer"); - glEnableVertexAttribArrayARB = (PFNGLENABLEVERTEXATTRIBARRAYARBPROC) GLH_EXT_GET_PROC_ADDRESS("glEnableVertexAttribArrayARB"); - glDisableVertexAttribArrayARB = (PFNGLDISABLEVERTEXATTRIBARRAYARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDisableVertexAttribArrayARB"); - glProgramStringARB = (PFNGLPROGRAMSTRINGARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramStringARB"); - glBindProgramARB = (PFNGLBINDPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glBindProgramARB"); - glDeleteProgramsARB = (PFNGLDELETEPROGRAMSARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDeleteProgramsARB"); - glGenProgramsARB = (PFNGLGENPROGRAMSARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGenProgramsARB"); - glProgramEnvParameter4dARB = (PFNGLPROGRAMENVPARAMETER4DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4dARB"); - glProgramEnvParameter4dvARB = (PFNGLPROGRAMENVPARAMETER4DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4dvARB"); - glProgramEnvParameter4fARB = (PFNGLPROGRAMENVPARAMETER4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4fARB"); - glProgramEnvParameter4fvARB = (PFNGLPROGRAMENVPARAMETER4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4fvARB"); - glProgramLocalParameter4dARB = (PFNGLPROGRAMLOCALPARAMETER4DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4dARB"); - glProgramLocalParameter4dvARB = (PFNGLPROGRAMLOCALPARAMETER4DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4dvARB"); - glProgramLocalParameter4fARB = (PFNGLPROGRAMLOCALPARAMETER4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4fARB"); - glProgramLocalParameter4fvARB = (PFNGLPROGRAMLOCALPARAMETER4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4fvARB"); - glGetProgramEnvParameterdvARB = (PFNGLGETPROGRAMENVPARAMETERDVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramEnvParameterdvARB"); - glGetProgramEnvParameterfvARB = (PFNGLGETPROGRAMENVPARAMETERFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramEnvParameterfvARB"); - glGetProgramLocalParameterdvARB = (PFNGLGETPROGRAMLOCALPARAMETERDVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramLocalParameterdvARB"); - glGetProgramLocalParameterfvARB = (PFNGLGETPROGRAMLOCALPARAMETERFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramLocalParameterfvARB"); - glGetProgramivARB = (PFNGLGETPROGRAMIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramivARB"); - glGetProgramStringARB = (PFNGLGETPROGRAMSTRINGARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramStringARB"); - glGetVertexAttribdvARB = (PFNGLGETVERTEXATTRIBDVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetVertexAttribdvARB"); - glGetVertexAttribfvARB = (PFNGLGETVERTEXATTRIBFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetVertexAttribfvARB"); - glGetVertexAttribivARB = (PFNGLGETVERTEXATTRIBIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetVertexAttribivARB"); - glGetVertexAttribPointervARB = (PFNGLGETVERTEXATTRIBPOINTERVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glgetVertexAttribPointervARB"); - glIsProgramARB = (PFNGLISPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glIsProgramARB"); - } - LL_DEBUGS("RenderInit") << "GL Probe: Got symbols" << LL_ENDL; + + // Assume shader capabilities + glDeleteObjectARB = (PFNGLDELETEOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDeleteObjectARB"); + glGetHandleARB = (PFNGLGETHANDLEARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetHandleARB"); + glDetachObjectARB = (PFNGLDETACHOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDetachObjectARB"); + glCreateShaderObjectARB = (PFNGLCREATESHADEROBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glCreateShaderObjectARB"); + glShaderSourceARB = (PFNGLSHADERSOURCEARBPROC) GLH_EXT_GET_PROC_ADDRESS("glShaderSourceARB"); + glCompileShaderARB = (PFNGLCOMPILESHADERARBPROC) GLH_EXT_GET_PROC_ADDRESS("glCompileShaderARB"); + glCreateProgramObjectARB = (PFNGLCREATEPROGRAMOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glCreateProgramObjectARB"); + glAttachObjectARB = (PFNGLATTACHOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glAttachObjectARB"); + glLinkProgramARB = (PFNGLLINKPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glLinkProgramARB"); + glUseProgramObjectARB = (PFNGLUSEPROGRAMOBJECTARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUseProgramObjectARB"); + glValidateProgramARB = (PFNGLVALIDATEPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glValidateProgramARB"); + glUniform1fARB = (PFNGLUNIFORM1FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1fARB"); + glUniform2fARB = (PFNGLUNIFORM2FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2fARB"); + glUniform3fARB = (PFNGLUNIFORM3FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3fARB"); + glUniform4fARB = (PFNGLUNIFORM4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4fARB"); + glUniform1iARB = (PFNGLUNIFORM1IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1iARB"); + glUniform2iARB = (PFNGLUNIFORM2IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2iARB"); + glUniform3iARB = (PFNGLUNIFORM3IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3iARB"); + glUniform4iARB = (PFNGLUNIFORM4IARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4iARB"); + glUniform1fvARB = (PFNGLUNIFORM1FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1fvARB"); + glUniform2fvARB = (PFNGLUNIFORM2FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2fvARB"); + glUniform3fvARB = (PFNGLUNIFORM3FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3fvARB"); + glUniform4fvARB = (PFNGLUNIFORM4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4fvARB"); + glUniform1ivARB = (PFNGLUNIFORM1IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform1ivARB"); + glUniform2ivARB = (PFNGLUNIFORM2IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform2ivARB"); + glUniform3ivARB = (PFNGLUNIFORM3IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform3ivARB"); + glUniform4ivARB = (PFNGLUNIFORM4IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniform4ivARB"); + glUniformMatrix2fvARB = (PFNGLUNIFORMMATRIX2FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix2fvARB"); + glUniformMatrix3fvARB = (PFNGLUNIFORMMATRIX3FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix3fvARB"); + glUniformMatrix3x4fv = (PFNGLUNIFORMMATRIX3X4FVPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix3x4fv"); + glUniformMatrix4fvARB = (PFNGLUNIFORMMATRIX4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glUniformMatrix4fvARB"); + glGetObjectParameterfvARB = (PFNGLGETOBJECTPARAMETERFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetObjectParameterfvARB"); + glGetObjectParameterivARB = (PFNGLGETOBJECTPARAMETERIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetObjectParameterivARB"); + glGetInfoLogARB = (PFNGLGETINFOLOGARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetInfoLogARB"); + glGetAttachedObjectsARB = (PFNGLGETATTACHEDOBJECTSARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetAttachedObjectsARB"); + glGetUniformLocationARB = (PFNGLGETUNIFORMLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetUniformLocationARB"); + glGetActiveUniformARB = (PFNGLGETACTIVEUNIFORMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetActiveUniformARB"); + glGetUniformfvARB = (PFNGLGETUNIFORMFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetUniformfvARB"); + glGetUniformivARB = (PFNGLGETUNIFORMIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetUniformivARB"); + glGetShaderSourceARB = (PFNGLGETSHADERSOURCEARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetShaderSourceARB"); + + LL_INFOS() << "initExtensions() VertexShader-related procs..." << LL_ENDL; + + // nSight doesn't support use of ARB funcs that have been normalized in the API + if (!LLRender::sNsightDebugSupport) + { + glGetAttribLocationARB = (PFNGLGETATTRIBLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetAttribLocationARB"); + glBindAttribLocationARB = (PFNGLBINDATTRIBLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glBindAttribLocationARB"); + } + else + { + glGetAttribLocationARB = (PFNGLGETATTRIBLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetAttribLocation"); + glBindAttribLocationARB = (PFNGLBINDATTRIBLOCATIONARBPROC) GLH_EXT_GET_PROC_ADDRESS("glBindAttribLocation"); + } + + glGetActiveAttribARB = (PFNGLGETACTIVEATTRIBARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetActiveAttribARB"); + glVertexAttrib1dARB = (PFNGLVERTEXATTRIB1DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1dARB"); + glVertexAttrib1dvARB = (PFNGLVERTEXATTRIB1DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1dvARB"); + glVertexAttrib1fARB = (PFNGLVERTEXATTRIB1FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1fARB"); + glVertexAttrib1fvARB = (PFNGLVERTEXATTRIB1FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1fvARB"); + glVertexAttrib1sARB = (PFNGLVERTEXATTRIB1SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1sARB"); + glVertexAttrib1svARB = (PFNGLVERTEXATTRIB1SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib1svARB"); + glVertexAttrib2dARB = (PFNGLVERTEXATTRIB2DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2dARB"); + glVertexAttrib2dvARB = (PFNGLVERTEXATTRIB2DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2dvARB"); + glVertexAttrib2fARB = (PFNGLVERTEXATTRIB2FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2fARB"); + glVertexAttrib2fvARB = (PFNGLVERTEXATTRIB2FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2fvARB"); + glVertexAttrib2sARB = (PFNGLVERTEXATTRIB2SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2sARB"); + glVertexAttrib2svARB = (PFNGLVERTEXATTRIB2SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib2svARB"); + glVertexAttrib3dARB = (PFNGLVERTEXATTRIB3DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3dARB"); + glVertexAttrib3dvARB = (PFNGLVERTEXATTRIB3DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3dvARB"); + glVertexAttrib3fARB = (PFNGLVERTEXATTRIB3FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3fARB"); + glVertexAttrib3fvARB = (PFNGLVERTEXATTRIB3FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3fvARB"); + glVertexAttrib3sARB = (PFNGLVERTEXATTRIB3SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3sARB"); + glVertexAttrib3svARB = (PFNGLVERTEXATTRIB3SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib3svARB"); + glVertexAttrib4nbvARB = (PFNGLVERTEXATTRIB4NBVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nbvARB"); + glVertexAttrib4nivARB = (PFNGLVERTEXATTRIB4NIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nivARB"); + glVertexAttrib4nsvARB = (PFNGLVERTEXATTRIB4NSVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nsvARB"); + glVertexAttrib4nubARB = (PFNGLVERTEXATTRIB4NUBARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nubARB"); + glVertexAttrib4nubvARB = (PFNGLVERTEXATTRIB4NUBVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nubvARB"); + glVertexAttrib4nuivARB = (PFNGLVERTEXATTRIB4NUIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nuivARB"); + glVertexAttrib4nusvARB = (PFNGLVERTEXATTRIB4NUSVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4nusvARB"); + glVertexAttrib4bvARB = (PFNGLVERTEXATTRIB4BVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4bvARB"); + glVertexAttrib4dARB = (PFNGLVERTEXATTRIB4DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4dARB"); + glVertexAttrib4dvARB = (PFNGLVERTEXATTRIB4DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4dvARB"); + glVertexAttrib4fARB = (PFNGLVERTEXATTRIB4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4fARB"); + glVertexAttrib4fvARB = (PFNGLVERTEXATTRIB4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4fvARB"); + glVertexAttrib4ivARB = (PFNGLVERTEXATTRIB4IVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4ivARB"); + glVertexAttrib4sARB = (PFNGLVERTEXATTRIB4SARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4sARB"); + glVertexAttrib4svARB = (PFNGLVERTEXATTRIB4SVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4svARB"); + glVertexAttrib4ubvARB = (PFNGLVERTEXATTRIB4UBVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4ubvARB"); + glVertexAttrib4uivARB = (PFNGLVERTEXATTRIB4UIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4uivARB"); + glVertexAttrib4usvARB = (PFNGLVERTEXATTRIB4USVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttrib4usvARB"); + glVertexAttribPointerARB = (PFNGLVERTEXATTRIBPOINTERARBPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttribPointerARB"); + glVertexAttribIPointer = (PFNGLVERTEXATTRIBIPOINTERPROC) GLH_EXT_GET_PROC_ADDRESS("glVertexAttribIPointer"); + glEnableVertexAttribArrayARB = (PFNGLENABLEVERTEXATTRIBARRAYARBPROC) GLH_EXT_GET_PROC_ADDRESS("glEnableVertexAttribArrayARB"); + glDisableVertexAttribArrayARB = (PFNGLDISABLEVERTEXATTRIBARRAYARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDisableVertexAttribArrayARB"); + glProgramStringARB = (PFNGLPROGRAMSTRINGARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramStringARB"); + glBindProgramARB = (PFNGLBINDPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glBindProgramARB"); + glDeleteProgramsARB = (PFNGLDELETEPROGRAMSARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDeleteProgramsARB"); + glGenProgramsARB = (PFNGLGENPROGRAMSARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGenProgramsARB"); + glProgramEnvParameter4dARB = (PFNGLPROGRAMENVPARAMETER4DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4dARB"); + glProgramEnvParameter4dvARB = (PFNGLPROGRAMENVPARAMETER4DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4dvARB"); + glProgramEnvParameter4fARB = (PFNGLPROGRAMENVPARAMETER4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4fARB"); + glProgramEnvParameter4fvARB = (PFNGLPROGRAMENVPARAMETER4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramEnvParameter4fvARB"); + glProgramLocalParameter4dARB = (PFNGLPROGRAMLOCALPARAMETER4DARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4dARB"); + glProgramLocalParameter4dvARB = (PFNGLPROGRAMLOCALPARAMETER4DVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4dvARB"); + glProgramLocalParameter4fARB = (PFNGLPROGRAMLOCALPARAMETER4FARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4fARB"); + glProgramLocalParameter4fvARB = (PFNGLPROGRAMLOCALPARAMETER4FVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glProgramLocalParameter4fvARB"); + glGetProgramEnvParameterdvARB = (PFNGLGETPROGRAMENVPARAMETERDVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramEnvParameterdvARB"); + glGetProgramEnvParameterfvARB = (PFNGLGETPROGRAMENVPARAMETERFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramEnvParameterfvARB"); + glGetProgramLocalParameterdvARB = (PFNGLGETPROGRAMLOCALPARAMETERDVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramLocalParameterdvARB"); + glGetProgramLocalParameterfvARB = (PFNGLGETPROGRAMLOCALPARAMETERFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramLocalParameterfvARB"); + glGetProgramivARB = (PFNGLGETPROGRAMIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramivARB"); + glGetProgramStringARB = (PFNGLGETPROGRAMSTRINGARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetProgramStringARB"); + glGetVertexAttribdvARB = (PFNGLGETVERTEXATTRIBDVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetVertexAttribdvARB"); + glGetVertexAttribfvARB = (PFNGLGETVERTEXATTRIBFVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetVertexAttribfvARB"); + glGetVertexAttribivARB = (PFNGLGETVERTEXATTRIBIVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glGetVertexAttribivARB"); + glGetVertexAttribPointervARB = (PFNGLGETVERTEXATTRIBPOINTERVARBPROC) GLH_EXT_GET_PROC_ADDRESS("glgetVertexAttribPointervARB"); + glIsProgramARB = (PFNGLISPROGRAMARBPROC) GLH_EXT_GET_PROC_ADDRESS("glIsProgramARB"); + + LL_DEBUGS("RenderInit") << "GL Probe: Got symbols" << LL_ENDL; #endif - mInited = TRUE; + mInited = TRUE; } void rotate_quat(LLQuaternion& rotation) @@ -2116,7 +2078,7 @@ void LLGLState::checkClientArrays(const std::string& msg, U32 data_mask) glClientActiveTextureARB(GL_TEXTURE0_ARB); gGL.getTexUnit(0)->activate(); - if (gGLManager.mHasVertexShader && LLGLSLShader::sNoFixedFunction) + if (LLGLSLShader::sNoFixedFunction) { //make sure vertex attribs are all disabled GLint count; glGetIntegerv(GL_MAX_VERTEX_ATTRIBS_ARB, &count); diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h index a07e2d9bb0..a03d5352be 100644 --- a/indra/llrender/llgl.h +++ b/indra/llrender/llgl.h @@ -94,9 +94,6 @@ public: BOOL mHasMapBufferRange; BOOL mHasFlushBufferRange; BOOL mHasPBuffer; - BOOL mHasShaderObjects; - BOOL mHasVertexShader; - BOOL mHasFragmentShader; S32 mNumTextureImageUnits; BOOL mHasOcclusionQuery; BOOL mHasTimerQuery; diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h index 6bca3623e0..3d93cc0762 100644 --- a/indra/llrender/llglheaders.h +++ b/indra/llrender/llglheaders.h @@ -812,4 +812,23 @@ extern void glGetBufferPointervARB (GLenum, GLenum, GLvoid* *); #define GL_RENDERBUFFER_FREE_MEMORY_ATI 0x87FD #endif +#if defined(TRACY_ENABLE) && LL_PROFILER_ENABLE_TRACY_OPENGL + // Tracy uses the following: + // glGenQueries + // glGetQueryiv + // glGetQueryObjectiv + #define glGenQueries glGenQueriesARB + #define glGetQueryiv glGetQueryivARB + #define glGetQueryObjectiv glGetQueryObjectivARB + #include <tracy/TracyOpenGL.hpp> + + #define LL_PROFILER_GPU_ZONEC(name,color) TracyGpuZoneC(name,color); + #define LL_PROFILER_GPU_COLLECT TracyGpuCollect + #define LL_PROFILER_GPU_CONTEXT TracyGpuContext +#else + #define LL_PROFILER_GPU_ZONEC(name,color) (void)name;(void)color; + #define LL_PROFILER_GPU_COLLECT + #define LL_PROFILER_GPU_CONTEXT +#endif + #endif // LL_LLGLHEADERS_H diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp index 4351f6e2c8..2fb3b8257d 100644 --- a/indra/llrender/llglslshader.cpp +++ b/indra/llrender/llglslshader.cpp @@ -384,6 +384,8 @@ BOOL LLGLSLShader::createShader(std::vector<LLStaticHashedString> * attributes, U32 varying_count, const char** varyings) { + LL_PROFILE_ZONE_SCOPED; + unloadInternal(); sInstances.insert(this); @@ -588,6 +590,8 @@ void LLGLSLShader::attachObjects(GLhandleARB* objects, S32 count) BOOL LLGLSLShader::mapAttributes(const std::vector<LLStaticHashedString> * attributes) { + LL_PROFILE_ZONE_SCOPED; + //before linking, make sure reserved attributes always have consistent locations for (U32 i = 0; i < LLShaderMgr::instance()->mReservedAttribs.size(); i++) { @@ -649,6 +653,8 @@ BOOL LLGLSLShader::mapAttributes(const std::vector<LLStaticHashedString> * attri void LLGLSLShader::mapUniform(GLint index, const vector<LLStaticHashedString> * uniforms) { + LL_PROFILE_ZONE_SCOPED; + if (index == -1) { return; @@ -770,6 +776,8 @@ void LLGLSLShader::removePermutation(std::string name) GLint LLGLSLShader::mapUniformTextureChannel(GLint location, GLenum type) { + LL_PROFILE_ZONE_SCOPED; + if ((type >= GL_SAMPLER_1D_ARB && type <= GL_SAMPLER_2D_RECT_SHADOW_ARB) || type == GL_SAMPLER_2D_MULTISAMPLE) { //this here is a texture @@ -782,7 +790,9 @@ GLint LLGLSLShader::mapUniformTextureChannel(GLint location, GLenum type) BOOL LLGLSLShader::mapUniforms(const vector<LLStaticHashedString> * uniforms) { - BOOL res = TRUE; + LL_PROFILE_ZONE_SCOPED; + + BOOL res = TRUE; mTotalUniformSize = 0; mActiveTextureChannels = 0; @@ -925,6 +935,8 @@ BOOL LLGLSLShader::mapUniforms(const vector<LLStaticHashedString> * uniforms) BOOL LLGLSLShader::link(BOOL suppress_errors) { + LL_PROFILE_ZONE_SCOPED; + BOOL success = LLShaderMgr::instance()->linkProgramObject(mProgramObject, suppress_errors); if (!success && !suppress_errors) @@ -937,56 +949,52 @@ BOOL LLGLSLShader::link(BOOL suppress_errors) void LLGLSLShader::bind() { + LL_PROFILE_ZONE_SCOPED; + gGL.flush(); - if (gGLManager.mHasShaderObjects) + + if (sCurBoundShader != mProgramObject) // Don't re-bind current shader { LLVertexBuffer::unbind(); glUseProgramObjectARB(mProgramObject); sCurBoundShader = mProgramObject; sCurBoundShaderPtr = this; - if (mUniformsDirty) - { - LLShaderMgr::instance()->updateShaderUniforms(this); - mUniformsDirty = FALSE; - } + } + + if (mUniformsDirty) + { + LLShaderMgr::instance()->updateShaderUniforms(this); + mUniformsDirty = FALSE; } } void LLGLSLShader::unbind() { + LL_PROFILE_ZONE_SCOPED; + gGL.flush(); - if (gGLManager.mHasShaderObjects) - { - stop_glerror(); - if (gGLManager.mIsNVIDIA) - { - for (U32 i = 0; i < mAttribute.size(); ++i) - { - vertexAttrib4f(i, 0,0,0,1); - stop_glerror(); - } - } - LLVertexBuffer::unbind(); - glUseProgramObjectARB(0); - sCurBoundShader = 0; - sCurBoundShaderPtr = NULL; - stop_glerror(); - } + stop_glerror(); + LLVertexBuffer::unbind(); + glUseProgramObjectARB(0); + sCurBoundShader = 0; + sCurBoundShaderPtr = NULL; + stop_glerror(); } void LLGLSLShader::bindNoShader(void) { + LL_PROFILE_ZONE_SCOPED; + LLVertexBuffer::unbind(); - if (gGLManager.mHasShaderObjects) - { - glUseProgramObjectARB(0); - sCurBoundShader = 0; - sCurBoundShaderPtr = NULL; - } + glUseProgramObjectARB(0); + sCurBoundShader = 0; + sCurBoundShaderPtr = NULL; } S32 LLGLSLShader::bindTexture(const std::string &uniform, LLTexture *texture, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace colorspace) { + LL_PROFILE_ZONE_SCOPED; + S32 channel = 0; channel = getUniformLocation(uniform); @@ -995,6 +1003,8 @@ S32 LLGLSLShader::bindTexture(const std::string &uniform, LLTexture *texture, LL S32 LLGLSLShader::bindTexture(S32 uniform, LLTexture *texture, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace colorspace) { + LL_PROFILE_ZONE_SCOPED; + if (uniform < 0 || uniform >= (S32)mTexture.size()) { LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL; @@ -1014,6 +1024,8 @@ S32 LLGLSLShader::bindTexture(S32 uniform, LLTexture *texture, LLTexUnit::eTextu S32 LLGLSLShader::unbindTexture(const std::string &uniform, LLTexUnit::eTextureType mode) { + LL_PROFILE_ZONE_SCOPED; + S32 channel = 0; channel = getUniformLocation(uniform); @@ -1022,6 +1034,8 @@ S32 LLGLSLShader::unbindTexture(const std::string &uniform, LLTexUnit::eTextureT S32 LLGLSLShader::unbindTexture(S32 uniform, LLTexUnit::eTextureType mode) { + LL_PROFILE_ZONE_SCOPED; + if (uniform < 0 || uniform >= (S32)mTexture.size()) { LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL; @@ -1040,6 +1054,8 @@ S32 LLGLSLShader::unbindTexture(S32 uniform, LLTexUnit::eTextureType mode) S32 LLGLSLShader::enableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace space) { + LL_PROFILE_ZONE_SCOPED; + if (uniform < 0 || uniform >= (S32)mTexture.size()) { LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL; @@ -1057,6 +1073,8 @@ S32 LLGLSLShader::enableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTex S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace space) { + LL_PROFILE_ZONE_SCOPED; + if (uniform < 0 || uniform >= (S32)mTexture.size()) { LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL; @@ -1346,6 +1364,8 @@ void LLGLSLShader::uniformMatrix3fv(U32 index, U32 count, GLboolean transpose, c void LLGLSLShader::uniformMatrix3x4fv(U32 index, U32 count, GLboolean transpose, const GLfloat *v) { + LL_PROFILE_ZONE_SCOPED; + if (mProgramObject) { if (mUniform.size() <= index) @@ -1380,6 +1400,8 @@ void LLGLSLShader::uniformMatrix4fv(U32 index, U32 count, GLboolean transpose, c GLint LLGLSLShader::getUniformLocation(const LLStaticHashedString& uniform) { + LL_PROFILE_ZONE_SCOPED; + GLint ret = -1; if (mProgramObject) { @@ -1404,6 +1426,8 @@ GLint LLGLSLShader::getUniformLocation(const LLStaticHashedString& uniform) GLint LLGLSLShader::getUniformLocation(U32 index) { + LL_PROFILE_ZONE_SCOPED; + GLint ret = -1; if (mProgramObject) { @@ -1416,6 +1440,8 @@ GLint LLGLSLShader::getUniformLocation(U32 index) GLint LLGLSLShader::getAttribLocation(U32 attrib) { + LL_PROFILE_ZONE_SCOPED; + if (attrib < mAttribute.size()) { return mAttribute[attrib]; diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 0151d20128..276fa55e15 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -1304,7 +1304,10 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt } stop_glerror(); - glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels); + { + LL_PROFILE_ZONE_NAMED("glTexImage2D"); + glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels); + } stop_glerror(); if (use_scratch) diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 03b6aac20c..43b4441ea8 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1848,6 +1848,7 @@ LLLightState* LLRender::getLight(U32 index) void LLRender::setAmbientLightColor(const LLColor4& color) { + LL_PROFILE_ZONE_SCOPED if (color != mAmbientLightColor) { ++mLightHash; diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp index e3c0255290..401085a00b 100644 --- a/indra/llrender/llrendertarget.cpp +++ b/indra/llrender/llrendertarget.cpp @@ -437,11 +437,13 @@ void LLRenderTarget::bindTarget() GL_COLOR_ATTACHMENT1, GL_COLOR_ATTACHMENT2, GL_COLOR_ATTACHMENT3}; + LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x4000FF ) glDrawBuffersARB(mTex.size(), drawbuffers); } if (mTex.empty()) { //no color buffer to draw to + LL_PROFILER_GPU_ZONEC( "gl.DrawBuffer", 0x0000FF ) glDrawBuffer(GL_NONE); glReadBuffer(GL_NONE); } diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 7d2b09ca4a..0449ac392c 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -91,6 +91,8 @@ LLVBOPool LLVertexBuffer::sDynamicIBOPool(GL_DYNAMIC_DRAW_ARB, GL_ELEMENT_ARRAY_ U32 LLVBOPool::sBytesPooled = 0; U32 LLVBOPool::sIndexBytesPooled = 0; +U32 LLVBOPool::sNameIdx = 0; +U32 LLVBOPool::sNamePool[1024]; std::list<U32> LLVertexBuffer::sAvailableVAOName; U32 LLVertexBuffer::sCurVAOName = 1; @@ -121,15 +123,20 @@ bool LLVertexBuffer::sPreferStreamDraw = false; U32 LLVBOPool::genBuffer() { - U32 ret = 0; + LL_PROFILE_ZONE_SCOPED - glGenBuffersARB(1, &ret); - - return ret; + if (sNameIdx == 0) + { + glGenBuffersARB(1024, sNamePool); + sNameIdx = 1024; + } + + return sNamePool[--sNameIdx]; } void LLVBOPool::deleteBuffer(U32 name) { + LL_PROFILE_ZONE_SCOPED if (gGLManager.mInited) { LLVertexBuffer::unbind(); @@ -152,6 +159,7 @@ LLVBOPool::LLVBOPool(U32 vboUsage, U32 vboType) volatile U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed) { + LL_PROFILE_ZONE_SCOPED llassert(vbo_block_size(size) == size); volatile U8* ret = NULL; @@ -267,10 +275,12 @@ void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size) void LLVBOPool::seedPool() { + LL_PROFILE_ZONE_SCOPED U32 dummy_name = 0; if (mFreeList.size() < LL_VBO_POOL_SEED_COUNT) { + LL_PROFILE_ZONE_NAMED("VBOPool Resize"); mFreeList.resize(LL_VBO_POOL_SEED_COUNT); } @@ -411,6 +421,7 @@ void LLVertexBuffer::releaseVAOName(U32 name) //static void LLVertexBuffer::seedPools() { + LL_PROFILE_ZONE_SCOPED sStreamVBOPool.seedPool(); sDynamicVBOPool.seedPool(); sDynamicCopyVBOPool.seedPool(); @@ -615,6 +626,7 @@ void LLVertexBuffer::drawArrays(U32 mode, const std::vector<LLVector3>& pos, con glNormalPointer(GL_FLOAT, 0, norm[0].mV); } LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF0000 ) glDrawArrays(sGLMode[mode], 0, count); LLGLSLShader::stopProfile(count, mode); } @@ -654,6 +666,7 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto } LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x80FF80 ) glDrawElements(sGLMode[mode], num_indices, GL_UNSIGNED_SHORT, indicesp); LLGLSLShader::stopProfile(num_indices, mode); } @@ -763,6 +776,7 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi stop_glerror(); LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawRangeElements", 0xFFFF00 ) glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, idx); LLGLSLShader::stopProfile(count, mode); @@ -814,6 +828,7 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const stop_glerror(); LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0xA0FFA0 ) glDrawElements(sGLMode[mode], count, GL_UNSIGNED_SHORT, ((U16*) getIndicesPointer()) + indices_offset); LLGLSLShader::stopProfile(count, mode); @@ -861,6 +876,7 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const stop_glerror(); LLGLSLShader::startProfile(); stop_glerror(); + LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF4040 ) glDrawArrays(sGLMode[mode], first, count); stop_glerror(); LLGLSLShader::stopProfile(count, mode); diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h index dbe1a3687f..1d60970df4 100644 --- a/indra/llrender/llvertexbuffer.h +++ b/indra/llrender/llvertexbuffer.h @@ -89,6 +89,9 @@ public: std::vector<record_list_t> mFreeList; std::vector<U32> mMissCount; + //used to avoid calling glGenBuffers for every VBO creation + static U32 sNamePool[1024]; + static U32 sNameIdx; }; @@ -127,7 +130,7 @@ public: static LLVBOPool sDynamicCopyVBOPool; static LLVBOPool sStreamIBOPool; static LLVBOPool sDynamicIBOPool; - + static std::list<U32> sAvailableVAOName; static U32 sCurVAOName; diff --git a/indra/llui/llnotifications.cpp b/indra/llui/llnotifications.cpp index b791a19c2b..88eda1c172 100644 --- a/indra/llui/llnotifications.cpp +++ b/indra/llui/llnotifications.cpp @@ -1387,7 +1387,7 @@ bool LLNotifications::failedUniquenessTest(const LLSD& payload) LLNotificationChannelPtr LLNotifications::getChannel(const std::string& channelName) { - return LLNotificationChannelPtr(LLNotificationChannel::getInstance(channelName)); + return LLNotificationChannelPtr(LLNotificationChannel::getInstance(channelName).get()); } diff --git a/indra/llui/llstatbar.cpp b/indra/llui/llstatbar.cpp index 6c8e63442b..8adcd664df 100644 --- a/indra/llui/llstatbar.cpp +++ b/indra/llui/llstatbar.cpp @@ -554,29 +554,25 @@ void LLStatBar::draw() void LLStatBar::setStat(const std::string& stat_name) { using namespace LLTrace; - const StatType<CountAccumulator>* count_stat; - const StatType<EventAccumulator>* event_stat; - const StatType<SampleAccumulator>* sample_stat; - const StatType<MemAccumulator>* mem_stat; - if ((count_stat = StatType<CountAccumulator>::getInstance(stat_name))) + if (auto count_stat = StatType<CountAccumulator>::getInstance(stat_name)) { - mStat.countStatp = count_stat; + mStat.countStatp = count_stat.get(); mStatType = STAT_COUNT; } - else if ((event_stat = StatType<EventAccumulator>::getInstance(stat_name))) + else if (auto event_stat = StatType<EventAccumulator>::getInstance(stat_name)) { - mStat.eventStatp = event_stat; + mStat.eventStatp = event_stat.get(); mStatType = STAT_EVENT; } - else if ((sample_stat = StatType<SampleAccumulator>::getInstance(stat_name))) + else if (auto sample_stat = StatType<SampleAccumulator>::getInstance(stat_name)) { - mStat.sampleStatp = sample_stat; + mStat.sampleStatp = sample_stat.get(); mStatType = STAT_SAMPLE; } - else if ((mem_stat = StatType<MemAccumulator>::getInstance(stat_name))) + else if (auto mem_stat = StatType<MemAccumulator>::getInstance(stat_name)) { - mStat.memStatp = mem_stat; + mStat.memStatp = mem_stat.get(); mStatType = STAT_MEM; } } diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index b2b123f0da..4ce7c30bef 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -28,6 +28,8 @@ #if LL_WINDOWS && !LL_MESA_HEADLESS +#define LL_WINDOW_SINGLE_THREADED 0 + #include "llwindowwin32.h" // LLWindow library includes @@ -45,6 +47,7 @@ #include "lldir.h" #include "llsdutil.h" #include "llglslshader.h" +#include "llthreadsafequeue.h" // System includes #include <commdlg.h> @@ -79,6 +82,18 @@ const F32 ICON_FLASH_TIME = 0.5f; extern BOOL gDebugWindowProc; +static std::thread::id sWindowThreadId; +static std::thread::id sMainThreadId; + +#if 1 || LL_WINDOW_SINGLE_THREADED +#define ASSERT_MAIN_THREAD() +#define ASSERT_WINDOW_THREAD() +#else +#define ASSERT_MAIN_THREAD() llassert(LLThread::currentID() == sMainThreadId) +#define ASSERT_WINDOW_THREAD() llassert(LLThread::currentID() == sWindowThreadId) +#endif + + LPWSTR gIconResource = IDI_APPLICATION; LPDIRECTINPUT8 gDirectInput8; @@ -294,7 +309,7 @@ LLWinImm::LLWinImm() : mHImmDll(NULL) // static -BOOL LLWinImm::isIME(HKL hkl) +BOOL LLWinImm::isIME(HKL hkl) { if ( sTheInstance.mImmIsIME ) return sTheInstance.mImmIsIME(hkl); @@ -326,7 +341,7 @@ BOOL LLWinImm::getOpenStatus(HIMC himc) } // static -BOOL LLWinImm::setOpenStatus(HIMC himc, BOOL status) +BOOL LLWinImm::setOpenStatus(HIMC himc, BOOL status) { if ( sTheInstance.mImmSetOpenStatus ) return sTheInstance.mImmSetOpenStatus(himc, status); @@ -454,6 +469,8 @@ private: static LLMonitorInfo sMonitorInfo; + + LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, const std::string& title, const std::string& name, S32 x, S32 y, S32 width, S32 height, U32 flags, @@ -463,7 +480,11 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, U32 fsaa_samples) : LLWindow(callbacks, fullscreen, flags) { - + sMainThreadId = LLThread::currentID(); + mWindowThread = new LLWindowWin32Thread(this); +#if !LL_WINDOW_SINGLE_THREADED + mWindowThread->start(); +#endif //MAINT-516 -- force a load of opengl32.dll just in case windows went sideways LoadLibrary(L"opengl32.dll"); @@ -784,7 +805,7 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, LLCoordScreen windowPos(x,y); LLCoordScreen windowSize(window_rect.right - window_rect.left, window_rect.bottom - window_rect.top); - if (!switchContext(mFullscreen, windowSize, TRUE, &windowPos)) + if (!switchContext(mFullscreen, windowSize, disable_vsync, &windowPos)) { return; } @@ -811,6 +832,8 @@ LLWindowWin32::~LLWindowWin32() delete [] mWindowClassName; mWindowClassName = NULL; + + delete mWindowThread; } void LLWindowWin32::show() @@ -930,26 +953,35 @@ void LLWindowWin32::close() LL_DEBUGS("Window") << "Destroying Window" << LL_ENDL; - if (IsWindow(mWindowHandle)) - { - // Make sure we don't leave a blank toolbar button. - ShowWindow(mWindowHandle, SW_HIDE); - - // This causes WM_DESTROY to be sent *immediately* - if (!destroy_window_handler(mWindowHandle)) + mWindowThread->post([=]() { - OSMessageBox(mCallbacks->translateString("MBDestroyWinFailed"), - mCallbacks->translateString("MBShutdownErr"), - OSMB_OK); - } - } - else + if (IsWindow(mWindowHandle)) + { + // Make sure we don't leave a blank toolbar button. + ShowWindow(mWindowHandle, SW_HIDE); + + // This causes WM_DESTROY to be sent *immediately* + if (!destroy_window_handler(mWindowHandle)) + { + OSMessageBox(mCallbacks->translateString("MBDestroyWinFailed"), + mCallbacks->translateString("MBShutdownErr"), + OSMB_OK); + } + } + else + { + // Something killed the window while we were busy destroying gl or handle somehow got broken + LL_WARNS("Window") << "Failed to destroy Window, invalid handle!" << LL_ENDL; + } + mWindowHandle = NULL; + + mWindowThread->mFinished = true; + }); + + while (!mWindowThread->isStopped()) { - // Something killed the window while we were busy destroying gl or handle somehow got broken - LL_WARNS("Window") << "Failed to destroy Window, invalid handle!" << LL_ENDL; + std::this_thread::sleep_for(std::chrono::milliseconds(1)); } - - mWindowHandle = NULL; } BOOL LLWindowWin32::isValid() @@ -1090,171 +1122,203 @@ BOOL LLWindowWin32::setSizeImpl(const LLCoordWindow size) } // changing fullscreen resolution -BOOL LLWindowWin32::switchContext(BOOL fullscreen, const LLCoordScreen &size, BOOL disable_vsync, const LLCoordScreen * const posp) +BOOL LLWindowWin32::switchContext(BOOL fullscreen, const LLCoordScreen& size, BOOL disable_vsync, const LLCoordScreen* const posp) { - GLuint pixel_format; - DEVMODE dev_mode; - ::ZeroMemory(&dev_mode, sizeof(DEVMODE)); - dev_mode.dmSize = sizeof(DEVMODE); - DWORD current_refresh; - DWORD dw_ex_style; - DWORD dw_style; - RECT window_rect = {0, 0, 0, 0}; - S32 width = size.mX; - S32 height = size.mY; - BOOL auto_show = FALSE; - - if (mhRC) - { - auto_show = TRUE; - resetDisplayResolution(); - } - - if (EnumDisplaySettings(NULL, ENUM_CURRENT_SETTINGS, &dev_mode)) - { - current_refresh = dev_mode.dmDisplayFrequency; - } - else - { - current_refresh = 60; - } - - gGLManager.shutdownGL(); - //destroy gl context - if (mhRC) - { - if (!wglMakeCurrent(NULL, NULL)) - { - LL_WARNS("Window") << "Release of DC and RC failed" << LL_ENDL; - } + //called from main thread + GLuint pixel_format; + DEVMODE dev_mode; + ::ZeroMemory(&dev_mode, sizeof(DEVMODE)); + dev_mode.dmSize = sizeof(DEVMODE); + DWORD current_refresh; + DWORD dw_ex_style; + DWORD dw_style; + RECT window_rect = { 0, 0, 0, 0 }; + S32 width = size.mX; + S32 height = size.mY; + BOOL auto_show = FALSE; + + if (mhRC) + { + auto_show = TRUE; + resetDisplayResolution(); + } - if (!wglDeleteContext(mhRC)) - { - LL_WARNS("Window") << "Release of rendering context failed" << LL_ENDL; - } + if (EnumDisplaySettings(NULL, ENUM_CURRENT_SETTINGS, &dev_mode)) + { + current_refresh = dev_mode.dmDisplayFrequency; + } + else + { + current_refresh = 60; + } - mhRC = NULL; - } + gGLManager.shutdownGL(); + //destroy gl context + if (mhRC) + { + if (!wglMakeCurrent(NULL, NULL)) + { + LL_WARNS("Window") << "Release of DC and RC failed" << LL_ENDL; + } - if (fullscreen) - { - mFullscreen = TRUE; - BOOL success = FALSE; - DWORD closest_refresh = 0; + if (!wglDeleteContext(mhRC)) + { + LL_WARNS("Window") << "Release of rendering context failed" << LL_ENDL; + } - for (S32 mode_num = 0;; mode_num++) - { - if (!EnumDisplaySettings(NULL, mode_num, &dev_mode)) - { - break; - } + mhRC = NULL; + } - if (dev_mode.dmPelsWidth == width && - dev_mode.dmPelsHeight == height && - dev_mode.dmBitsPerPel == BITS_PER_PIXEL) - { - success = TRUE; - if ((dev_mode.dmDisplayFrequency - current_refresh) - < (closest_refresh - current_refresh)) - { - closest_refresh = dev_mode.dmDisplayFrequency; - } - } - } + if (fullscreen) + { + mFullscreen = TRUE; + BOOL success = FALSE; + DWORD closest_refresh = 0; - if (closest_refresh == 0) - { - LL_WARNS("Window") << "Couldn't find display mode " << width << " by " << height << " at " << BITS_PER_PIXEL << " bits per pixel" << LL_ENDL; - return FALSE; - } + for (S32 mode_num = 0;; mode_num++) + { + if (!EnumDisplaySettings(NULL, mode_num, &dev_mode)) + { + break; + } - // If we found a good resolution, use it. - if (success) - { - success = setDisplayResolution(width, height, BITS_PER_PIXEL, closest_refresh); - } + if (dev_mode.dmPelsWidth == width && + dev_mode.dmPelsHeight == height && + dev_mode.dmBitsPerPel == BITS_PER_PIXEL) + { + success = TRUE; + if ((dev_mode.dmDisplayFrequency - current_refresh) + < (closest_refresh - current_refresh)) + { + closest_refresh = dev_mode.dmDisplayFrequency; + } + } + } - // Keep a copy of the actual current device mode in case we minimize - // and change the screen resolution. JC - EnumDisplaySettings(NULL, ENUM_CURRENT_SETTINGS, &dev_mode); + if (closest_refresh == 0) + { + LL_WARNS("Window") << "Couldn't find display mode " << width << " by " << height << " at " << BITS_PER_PIXEL << " bits per pixel" << LL_ENDL; + return FALSE; + } - if (success) - { - mFullscreen = TRUE; - mFullscreenWidth = dev_mode.dmPelsWidth; - mFullscreenHeight = dev_mode.dmPelsHeight; - mFullscreenBits = dev_mode.dmBitsPerPel; - mFullscreenRefresh = dev_mode.dmDisplayFrequency; + // If we found a good resolution, use it. + if (success) + { + success = setDisplayResolution(width, height, BITS_PER_PIXEL, closest_refresh); + } - LL_INFOS("Window") << "Running at " << dev_mode.dmPelsWidth - << "x" << dev_mode.dmPelsHeight - << "x" << dev_mode.dmBitsPerPel - << " @ " << dev_mode.dmDisplayFrequency - << LL_ENDL; + // Keep a copy of the actual current device mode in case we minimize + // and change the screen resolution. JC + EnumDisplaySettings(NULL, ENUM_CURRENT_SETTINGS, &dev_mode); - window_rect.left = (long) 0; - window_rect.right = (long) width; // Windows GDI rects don't include rightmost pixel - window_rect.top = (long) 0; - window_rect.bottom = (long) height; - dw_ex_style = WS_EX_APPWINDOW; - dw_style = WS_POPUP; + if (success) + { + mFullscreen = TRUE; + mFullscreenWidth = dev_mode.dmPelsWidth; + mFullscreenHeight = dev_mode.dmPelsHeight; + mFullscreenBits = dev_mode.dmBitsPerPel; + mFullscreenRefresh = dev_mode.dmDisplayFrequency; + + LL_INFOS("Window") << "Running at " << dev_mode.dmPelsWidth + << "x" << dev_mode.dmPelsHeight + << "x" << dev_mode.dmBitsPerPel + << " @ " << dev_mode.dmDisplayFrequency + << LL_ENDL; + + window_rect.left = (long)0; + window_rect.right = (long)width; // Windows GDI rects don't include rightmost pixel + window_rect.top = (long)0; + window_rect.bottom = (long)height; + dw_ex_style = WS_EX_APPWINDOW; + dw_style = WS_POPUP; + + // Move window borders out not to cover window contents. + // This converts client rect to window rect, i.e. expands it by the window border size. + AdjustWindowRectEx(&window_rect, dw_style, FALSE, dw_ex_style); + } + // If it failed, we don't want to run fullscreen + else + { + mFullscreen = FALSE; + mFullscreenWidth = -1; + mFullscreenHeight = -1; + mFullscreenBits = -1; + mFullscreenRefresh = -1; - // Move window borders out not to cover window contents. - // This converts client rect to window rect, i.e. expands it by the window border size. - AdjustWindowRectEx(&window_rect, dw_style, FALSE, dw_ex_style); - } - // If it failed, we don't want to run fullscreen - else - { - mFullscreen = FALSE; - mFullscreenWidth = -1; - mFullscreenHeight = -1; - mFullscreenBits = -1; - mFullscreenRefresh = -1; + LL_INFOS("Window") << "Unable to run fullscreen at " << width << "x" << height << LL_ENDL; + return FALSE; + } + } + else + { + mFullscreen = FALSE; + window_rect.left = (long)(posp ? posp->mX : 0); + window_rect.right = (long)width + window_rect.left; // Windows GDI rects don't include rightmost pixel + window_rect.top = (long)(posp ? posp->mY : 0); + window_rect.bottom = (long)height + window_rect.top; + // Window with an edge + dw_ex_style = WS_EX_APPWINDOW | WS_EX_WINDOWEDGE; + dw_style = WS_OVERLAPPEDWINDOW; + } - LL_INFOS("Window") << "Unable to run fullscreen at " << width << "x" << height << LL_ENDL; - return FALSE; - } - } - else - { - mFullscreen = FALSE; - window_rect.left = (long) (posp ? posp->mX : 0); - window_rect.right = (long) width + window_rect.left; // Windows GDI rects don't include rightmost pixel - window_rect.top = (long) (posp ? posp->mY : 0); - window_rect.bottom = (long) height + window_rect.top; - // Window with an edge - dw_ex_style = WS_EX_APPWINDOW | WS_EX_WINDOWEDGE; - dw_style = WS_OVERLAPPEDWINDOW; - } + // don't post quit messages when destroying old windows + mPostQuit = FALSE; - // don't post quit messages when destroying old windows - mPostQuit = FALSE; - // create window + // create window LL_DEBUGS("Window") << "Creating window with X: " << window_rect.left << " Y: " << window_rect.top << " Width: " << (window_rect.right - window_rect.left) << " Height: " << (window_rect.bottom - window_rect.top) << " Fullscreen: " << mFullscreen << LL_ENDL; - if (mWindowHandle && !destroy_window_handler(mWindowHandle)) + + auto oldHandle = mWindowHandle; + + //zero out mWindowHandle and mhDC before destroying window so window thread falls back to peekmessage + mWindowHandle = 0; + mhDC = 0; + + if (oldHandle && !destroy_window_handler(oldHandle)) { LL_WARNS("Window") << "Failed to properly close window before recreating it!" << LL_ENDL; - } - mWindowHandle = CreateWindowEx(dw_ex_style, - mWindowClassName, - mWindowTitle, - WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, - window_rect.left, // x pos - window_rect.top, // y pos - window_rect.right - window_rect.left, // width - window_rect.bottom - window_rect.top, // height - NULL, - NULL, - mhInstance, - NULL); + } + + mWindowHandle = NULL; + mhDC = 0; + + mWindowThread->post( + [this, window_rect, dw_ex_style, dw_style]() + { + mWindowHandle = CreateWindowEx(dw_ex_style, + mWindowClassName, + mWindowTitle, + WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, + window_rect.left, // x pos + window_rect.top, // y pos + window_rect.right - window_rect.left, // width + window_rect.bottom - window_rect.top, // height + NULL, + NULL, + mhInstance, + NULL); + + if (mWindowHandle) + { + mhDC = GetDC(mWindowHandle); + } + } + ); + + // HACK wait for above handle to become populated + // TODO: use a future + int count = 1024; + while (!mhDC && count > 0) + { + Sleep(10); + --count; + } if (mWindowHandle) { @@ -1288,7 +1352,7 @@ BOOL LLWindowWin32::switchContext(BOOL fullscreen, const LLCoordScreen &size, BO 0, 0, 0 }; - if (!(mhDC = GetDC(mWindowHandle))) + if (!mhDC) { close(); OSMessageBox(mCallbacks->translateString("MBDevContextErr"), @@ -1582,25 +1646,48 @@ const S32 max_format = (S32)num_formats - 1; mhDC = 0; // Zero The Device Context } + auto oldHandle = mWindowHandle; + mWindowHandle = NULL; + mhDC = 0; + // Destroy The Window - if (mWindowHandle && !destroy_window_handler(mWindowHandle)) + if (oldHandle && !destroy_window_handler(oldHandle)) { LL_WARNS("Window") << "Failed to properly close window!" << LL_ENDL; } - mWindowHandle = CreateWindowEx(dw_ex_style, - mWindowClassName, - mWindowTitle, - WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, - window_rect.left, // x pos - window_rect.top, // y pos - window_rect.right - window_rect.left, // width - window_rect.bottom - window_rect.top, // height - NULL, - NULL, - mhInstance, - NULL); + mWindowThread->post( + [this, window_rect, dw_ex_style, dw_style]() + { + mWindowHandle = CreateWindowEx(dw_ex_style, + mWindowClassName, + mWindowTitle, + WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, + window_rect.left, // x pos + window_rect.top, // y pos + window_rect.right - window_rect.left, // width + window_rect.bottom - window_rect.top, // height + NULL, + NULL, + mhInstance, + NULL); + + if (mWindowHandle) + { + mhDC = GetDC(mWindowHandle); + } + } + ); + // HACK wait for above handle to become populated + // TODO: use a future + int count = 1024; + while (!mhDC && count > 0) + { + PostMessage(oldHandle, WM_USER + 8, 0x1717, 0x3b3b); + Sleep(10); + --count; + } if (mWindowHandle) { @@ -1612,7 +1699,7 @@ const S32 max_format = (S32)num_formats - 1; LL_WARNS("Window") << "Window recreation failed, code: " << GetLastError() << LL_ENDL; } - if (!(mhDC = GetDC(mWindowHandle))) + if (!mhDC) { close(); OSMessageBox(mCallbacks->translateString("MBDevContextErr"), mCallbacks->translateString("MBError"), OSMB_OK); @@ -1748,6 +1835,8 @@ const S32 max_format = (S32)num_formats - 1; return FALSE; } + LL_PROFILER_GPU_CONTEXT + if (!gGLManager.initGL()) { close(); @@ -1764,6 +1853,7 @@ const S32 max_format = (S32)num_formats - 1; else { LL_DEBUGS("Window") << "Keeping vertical sync" << LL_ENDL; + wglSwapIntervalEXT(1); } SetWindowLongPtr(mWindowHandle, GWLP_USERDATA, (LONG_PTR)this); @@ -1813,31 +1903,41 @@ void LLWindowWin32::moveWindow( const LLCoordScreen& position, const LLCoordScre BOOL LLWindowWin32::setCursorPosition(const LLCoordWindow position) { - mMousePositionModified = TRUE; + ASSERT_MAIN_THREAD(); + if (!mWindowHandle) { return FALSE; } - - // Inform the application of the new mouse position (needed for per-frame + // Inform the application of the new mouse position (needed for per-frame // hover/picking to function). mCallbacks->handleMouseMove(this, position.convert(), (MASK)0); - // DEV-18951 VWR-8524 Camera moves wildly when alt-clicking. - // Because we have preemptively notified the application of the new - // mouse position via handleMouseMove() above, we need to clear out - // any stale mouse move events. RN/JC - MSG msg; - while (PeekMessage(&msg, NULL, WM_MOUSEMOVE, WM_MOUSEMOVE, PM_REMOVE)) - { } - - LLCoordScreen screen_pos(position.convert()); - return ::SetCursorPos(screen_pos.mX, screen_pos.mY); + mMousePositionModified = TRUE; + LLCoordScreen screen_pos(position.convert()); + + mWindowThread->post([=] + { + SetCursorPos(screen_pos.mX, screen_pos.mY); + // DEV-18951 VWR-8524 Camera moves wildly when alt-clicking. + // Because we have preemptively notified the application of the new + // mouse position via handleMouseMove() above, we need to clear out + // any stale mouse move events. RN/JC + MSG msg; + while (PeekMessage(&msg, NULL, WM_MOUSEMOVE, WM_MOUSEMOVE, PM_REMOVE)) + { + } + + mMousePositionModified = FALSE; + }); + + return TRUE; } BOOL LLWindowWin32::getCursorPosition(LLCoordWindow *position) { + ASSERT_MAIN_THREAD(); POINT cursor_point; if (!mWindowHandle @@ -1853,21 +1953,35 @@ BOOL LLWindowWin32::getCursorPosition(LLCoordWindow *position) void LLWindowWin32::hideCursor() { - while (ShowCursor(FALSE) >= 0) - { - // nothing, wait for cursor to push down - } + ASSERT_MAIN_THREAD(); + + mWindowThread->post([=]() + { + while (ShowCursor(FALSE) >= 0) + { + // nothing, wait for cursor to push down + } + }); + mCursorHidden = TRUE; mHideCursorPermanent = TRUE; } void LLWindowWin32::showCursor() { - // makes sure the cursor shows up - while (ShowCursor(TRUE) < 0) - { - // do nothing, wait for cursor to pop out - } + LL_PROFILE_ZONE_SCOPED; + + ASSERT_MAIN_THREAD(); + + mWindowThread->post([=]() + { + // makes sure the cursor shows up + while (ShowCursor(TRUE) < 0) + { + // do nothing, wait for cursor to pop out + } + }); + mCursorHidden = FALSE; mHideCursorPermanent = FALSE; } @@ -1969,6 +2083,8 @@ void LLWindowWin32::initCursors() void LLWindowWin32::updateCursor() { + ASSERT_MAIN_THREAD(); + LL_PROFILE_ZONE_SCOPED if (mNextCursor == UI_CURSOR_ARROW && mBusyCount > 0) { @@ -1978,7 +2094,11 @@ void LLWindowWin32::updateCursor() if( mCurrentCursor != mNextCursor ) { mCurrentCursor = mNextCursor; - SetCursor( mCursor[mNextCursor] ); + auto nextCursor = mCursor[mNextCursor]; + mWindowThread->post([=]() + { + SetCursor(nextCursor); + }); } } @@ -1994,13 +2114,8 @@ void LLWindowWin32::captureMouse() void LLWindowWin32::releaseMouse() { - // *NOTE:Mani ReleaseCapture will spawn new windows messages... - // which will in turn call our MainWindowProc. It therefore requires - // pausing *and more importantly resumption* of the mainlooptimeout... - // just like DispatchMessage below. - mCallbacks->handlePauseWatchdog(this); + LL_PROFILE_ZONE_SCOPED; ReleaseCapture(); - mCallbacks->handleResumeWatchdog(this); } @@ -2009,1003 +2124,1129 @@ void LLWindowWin32::delayInputProcessing() mInputProcessingPaused = TRUE; } + void LLWindowWin32::gatherInput() { - MSG msg; - int msg_count = 0; + ASSERT_MAIN_THREAD(); + LL_PROFILE_ZONE_SCOPED + MSG msg; - while ((msg_count < MAX_MESSAGE_PER_UPDATE) && PeekMessage(&msg, NULL, 0, 0, PM_REMOVE)) - { - mCallbacks->handlePingWatchdog(this, "Main:TranslateGatherInput"); - TranslateMessage(&msg); +#if LL_WINDOW_SINGLE_THREADED + int msg_count = 0; - // turn watchdog off in here to not fail if windows is doing something wacky - mCallbacks->handlePauseWatchdog(this); - DispatchMessage(&msg); - mCallbacks->handleResumeWatchdog(this); - msg_count++; + while ((msg_count < MAX_MESSAGE_PER_UPDATE)) + { + LL_PROFILE_ZONE_NAMED("gi - loop"); + ++msg_count; + { + LL_PROFILE_ZONE_NAMED("gi - PeekMessage"); + if (!PeekMessage(&msg, NULL, 0, 0, PM_REMOVE)) + { + break; + } + } - if ( mInputProcessingPaused ) - { - break; - } - /* Attempted workaround for problem where typing fast and hitting - return would result in only part of the text being sent. JC + { + LL_PROFILE_ZONE_NAMED("gi - translate"); + TranslateMessage(&msg); + } + + { + LL_PROFILE_ZONE_NAMED("gi - dispatch"); + DispatchMessage(&msg); + } - BOOL key_posted = TranslateMessage(&msg); - DispatchMessage(&msg); - msg_count++; + if (mInputProcessingPaused) + { + break; + } - // If a key was translated, a WM_CHAR might have been posted to the end - // of the event queue. We need it immediately. - if (key_posted && msg.message == WM_KEYDOWN) - { - if (PeekMessage(&msg, NULL, WM_CHAR, WM_CHAR, PM_REMOVE)) - { - TranslateMessage(&msg); - DispatchMessage(&msg); - msg_count++; - } - } - */ - mCallbacks->handlePingWatchdog(this, "Main:AsyncCallbackGatherInput"); - // For async host by name support. Really hacky. - if (gAsyncMsgCallback && (LL_WM_HOST_RESOLVED == msg.message)) - { - gAsyncMsgCallback(msg); - } - } + // For async host by name support. Really hacky. + if (gAsyncMsgCallback && (LL_WM_HOST_RESOLVED == msg.message)) + { + LL_PROFILE_ZONE_NAMED("gi - callback"); + gAsyncMsgCallback(msg); + } + } +#else //multi-threaded window impl + { + if (mWindowThread->mFunctionQueue.size() > 0) + { + LL_PROFILE_ZONE_NAMED("gi - PostMessage"); + if (mWindowHandle) + { // post a nonsense user message to wake up the Window Thread in case any functions are pending + // and no windows events came through this frame + PostMessage(mWindowHandle, WM_USER + 0x0017, 0xB0B0, 0x1337); + } + } + + while (mWindowThread->mMessageQueue.tryPopBack(msg)) + { + LL_PROFILE_ZONE_NAMED("gi - message queue"); + if (mInputProcessingPaused) + { + continue; + } + + // For async host by name support. Really hacky. + if (gAsyncMsgCallback && (LL_WM_HOST_RESOLVED == msg.message)) + { + LL_PROFILE_ZONE_NAMED("gi - callback"); + gAsyncMsgCallback(msg); + } + } + } + + { + LL_PROFILE_ZONE_NAMED("gi - function queue"); + //process any pending functions + std::function<void()> curFunc; + while (mFunctionQueue.tryPopBack(curFunc)) + { + curFunc(); + } + } +#endif mInputProcessingPaused = FALSE; updateCursor(); - - // clear this once we've processed all mouse messages that might have occurred after - // we slammed the mouse position - mMousePositionModified = FALSE; } static LLTrace::BlockTimerStatHandle FTM_KEYHANDLER("Handle Keyboard"); static LLTrace::BlockTimerStatHandle FTM_MOUSEHANDLER("Handle Mouse"); +#if LL_WINDOW_SINGLE_THREADED +#define WINDOW_IMP_POST(x) x +#else +#define WINDOW_IMP_POST(x) window_imp->post([=]() { x; }) +#endif + LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_param, LPARAM l_param) { - // Ignore clicks not originated in the client area, i.e. mouse-up events not preceded with a WM_LBUTTONDOWN. - // This helps prevent avatar walking after maximizing the window by double-clicking the title bar. - static bool sHandleLeftMouseUp = true; + ASSERT_WINDOW_THREAD(); + LL_PROFILE_ZONE_SCOPED; - // Ignore the double click received right after activating app. - // This is to avoid triggering double click teleport after returning focus (see MAINT-3786). - static bool sHandleDoubleClick = true; + // Ignore clicks not originated in the client area, i.e. mouse-up events not preceded with a WM_LBUTTONDOWN. + // This helps prevent avatar walking after maximizing the window by double-clicking the title bar. + static bool sHandleLeftMouseUp = true; - LLWindowWin32 *window_imp = (LLWindowWin32 *)GetWindowLongPtr( h_wnd, GWLP_USERDATA ); + // Ignore the double click received right after activating app. + // This is to avoid triggering double click teleport after returning focus (see MAINT-3786). + static bool sHandleDoubleClick = true; - bool debug_window_proc = gDebugWindowProc || debugLoggingEnabled("Window"); + LLWindowWin32* window_imp = (LLWindowWin32*)GetWindowLongPtr(h_wnd, GWLP_USERDATA); + bool debug_window_proc = false; // gDebugWindowProc || debugLoggingEnabled("Window"); - if (NULL != window_imp) - { - window_imp->mCallbacks->handleResumeWatchdog(window_imp); - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:StartWndProc"); - // Has user provided their own window callback? - if (NULL != window_imp->mWndProc) - { - if (!window_imp->mWndProc(h_wnd, u_msg, w_param, l_param)) - { - // user has handled window message - return 0; - } - } - - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:PreSwitchWndProc"); - - // Juggle to make sure we can get negative positions for when - // mouse is outside window. - LLCoordWindow window_coord((S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param)); - - // This doesn't work, as LOWORD returns unsigned short. - //LLCoordWindow window_coord(LOWORD(l_param), HIWORD(l_param)); - LLCoordGL gl_coord; - - // pass along extended flag in mask - MASK mask = (l_param>>16 & KF_EXTENDED) ? MASK_EXTENDED : 0x0; - BOOL eat_keystroke = TRUE; - - switch(u_msg) - { - RECT update_rect; - S32 update_width; - S32 update_height; - - case WM_TIMER: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_TIMER"); - window_imp->mCallbacks->handleTimerEvent(window_imp); - break; - - case WM_DEVICECHANGE: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_DEVICECHANGE"); - if (debug_window_proc) - { - LL_INFOS("Window") << " WM_DEVICECHANGE: wParam=" << w_param - << "; lParam=" << l_param << LL_ENDL; - } - if (w_param == DBT_DEVNODES_CHANGED || w_param == DBT_DEVICEARRIVAL) - { - if (window_imp->mCallbacks->handleDeviceChange(window_imp)) - { - return 0; - } - } - break; - - case WM_PAINT: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_PAINT"); - GetUpdateRect(window_imp->mWindowHandle, &update_rect, FALSE); - update_width = update_rect.right - update_rect.left + 1; - update_height = update_rect.bottom - update_rect.top + 1; - window_imp->mCallbacks->handlePaint(window_imp, update_rect.left, update_rect.top, - update_width, update_height); - break; - case WM_PARENTNOTIFY: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_PARENTNOTIFY"); - u_msg = u_msg; - break; - - case WM_SETCURSOR: - // This message is sent whenever the cursor is moved in a window. - // You need to set the appropriate cursor appearance. - - // Only take control of cursor over client region of window - // This allows Windows(tm) to handle resize cursors, etc. - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_SETCURSOR"); - if (LOWORD(l_param) == HTCLIENT) - { - SetCursor(window_imp->mCursor[ window_imp->mCurrentCursor] ); - return 0; - } - break; - - case WM_ENTERMENULOOP: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_ENTERMENULOOP"); - window_imp->mCallbacks->handleWindowBlock(window_imp); - break; - - case WM_EXITMENULOOP: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_EXITMENULOOP"); - window_imp->mCallbacks->handleWindowUnblock(window_imp); - break; - - case WM_ACTIVATEAPP: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_ACTIVATEAPP"); - { - // This message should be sent whenever the app gains or loses focus. - BOOL activating = (BOOL) w_param; - BOOL minimized = window_imp->getMinimized(); - - if (debug_window_proc) - { - LL_INFOS("Window") << "WINDOWPROC ActivateApp " - << " activating " << S32(activating) - << " minimized " << S32(minimized) - << " fullscreen " << S32(window_imp->mFullscreen) - << LL_ENDL; - } - - if (window_imp->mFullscreen) - { - // When we run fullscreen, restoring or minimizing the app needs - // to switch the screen resolution - if (activating) - { - window_imp->setFullscreenResolution(); - window_imp->restore(); - } - else - { - window_imp->minimize(); - window_imp->resetDisplayResolution(); - } - } - - if (!activating) - { - sHandleDoubleClick = false; - } - - window_imp->mCallbacks->handleActivateApp(window_imp, activating); - - break; - } - - case WM_ACTIVATE: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_ACTIVATE"); - { - // Can be one of WA_ACTIVE, WA_CLICKACTIVE, or WA_INACTIVE - BOOL activating = (LOWORD(w_param) != WA_INACTIVE); - - BOOL minimized = BOOL(HIWORD(w_param)); - - if (!activating && LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->interruptLanguageTextInput(); - } - - // JC - I'm not sure why, but if we don't report that we handled the - // WM_ACTIVATE message, the WM_ACTIVATEAPP messages don't work - // properly when we run fullscreen. - if (debug_window_proc) - { - LL_INFOS("Window") << "WINDOWPROC Activate " - << " activating " << S32(activating) - << " minimized " << S32(minimized) - << LL_ENDL; - } - - // Don't handle this. - break; - } - - case WM_QUERYOPEN: - // TODO: use this to return a nice icon - break; - - case WM_SYSCOMMAND: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_SYSCOMMAND"); - switch(w_param) - { - case SC_KEYMENU: - // Disallow the ALT key from triggering the default system menu. - return 0; - - case SC_SCREENSAVE: - case SC_MONITORPOWER: - // eat screen save messages and prevent them! - return 0; - } - break; - - case WM_CLOSE: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_CLOSE"); - // Will the app allow the window to close? - if (window_imp->mCallbacks->handleCloseRequest(window_imp)) - { - // Get the app to initiate cleanup. - window_imp->mCallbacks->handleQuit(window_imp); - // The app is responsible for calling destroyWindow when done with GL - } - return 0; - - case WM_DESTROY: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_DESTROY"); - if (window_imp->shouldPostQuit()) - { - PostQuitMessage(0); // Posts WM_QUIT with an exit code of 0 - } - return 0; - - case WM_COMMAND: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_COMMAND"); - if (!HIWORD(w_param)) // this message is from a menu - { - window_imp->mCallbacks->handleMenuSelect(window_imp, LOWORD(w_param)); - } - break; - - case WM_SYSKEYDOWN: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_SYSKEYDOWN"); - // allow system keys, such as ALT-F4 to be processed by Windows - eat_keystroke = FALSE; - case WM_KEYDOWN: - window_imp->mKeyCharCode = 0; // don't know until wm_char comes in next - window_imp->mKeyScanCode = ( l_param >> 16 ) & 0xff; - window_imp->mKeyVirtualKey = w_param; - window_imp->mRawMsg = u_msg; - window_imp->mRawWParam = w_param; - window_imp->mRawLParam = l_param; - - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_KEYDOWN"); - { - if (debug_window_proc) - { - LL_INFOS("Window") << "Debug WindowProc WM_KEYDOWN " - << " key " << S32(w_param) - << LL_ENDL; - } - if(gKeyboard->handleKeyDown(w_param, mask) && eat_keystroke) - { - return 0; - } - // pass on to windows if we didn't handle it - break; - } - case WM_SYSKEYUP: - eat_keystroke = FALSE; - case WM_KEYUP: - { - window_imp->mKeyScanCode = ( l_param >> 16 ) & 0xff; - window_imp->mKeyVirtualKey = w_param; - window_imp->mRawMsg = u_msg; - window_imp->mRawWParam = w_param; - window_imp->mRawLParam = l_param; - - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_KEYUP"); - LL_RECORD_BLOCK_TIME(FTM_KEYHANDLER); - - if (debug_window_proc) - { - LL_INFOS("Window") << "Debug WindowProc WM_KEYUP " - << " key " << S32(w_param) - << LL_ENDL; - } - if (gKeyboard->handleKeyUp(w_param, mask) && eat_keystroke) - { - return 0; - } + if (NULL != window_imp) + { + // Has user provided their own window callback? + if (NULL != window_imp->mWndProc) + { + LL_PROFILE_ZONE_NAMED("mwp - WndProc"); + if (!window_imp->mWndProc(h_wnd, u_msg, w_param, l_param)) + { + // user has handled window message + return 0; + } + } - // pass on to windows - break; - } - case WM_IME_SETCONTEXT: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_IME_SETCONTEXT"); - if (debug_window_proc) - { - LL_INFOS("Window") << "WM_IME_SETCONTEXT" << LL_ENDL; - } - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - l_param &= ~ISC_SHOWUICOMPOSITIONWINDOW; - // Invoke DefWinProc with the modified LPARAM. - } - break; + // Juggle to make sure we can get negative positions for when + // mouse is outside window. + LLCoordWindow window_coord((S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param)); - case WM_IME_STARTCOMPOSITION: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_IME_STARTCOMPOSITION"); - if (debug_window_proc) - { - LL_INFOS() << "WM_IME_STARTCOMPOSITION" << LL_ENDL; - } - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->handleStartCompositionMessage(); - return 0; - } - break; + // This doesn't work, as LOWORD returns unsigned short. + //LLCoordWindow window_coord(LOWORD(l_param), HIWORD(l_param)); + LLCoordGL gl_coord; - case WM_IME_ENDCOMPOSITION: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_IME_ENDCOMPOSITION"); - if (debug_window_proc) - { - LL_INFOS() << "WM_IME_ENDCOMPOSITION" << LL_ENDL; - } - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - return 0; - } - break; + // pass along extended flag in mask + MASK mask = (l_param >> 16 & KF_EXTENDED) ? MASK_EXTENDED : 0x0; + BOOL eat_keystroke = TRUE; - case WM_IME_COMPOSITION: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_IME_COMPOSITION"); - if (debug_window_proc) - { - LL_INFOS() << "WM_IME_COMPOSITION" << LL_ENDL; - } - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->handleCompositionMessage(l_param); - return 0; - } - break; + switch (u_msg) + { + RECT update_rect; + S32 update_width; + S32 update_height; - case WM_IME_REQUEST: - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_IME_REQUEST"); - if (debug_window_proc) - { - LL_INFOS() << "WM_IME_REQUEST" << LL_ENDL; - } - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - LRESULT result = 0; - if (window_imp->handleImeRequests(w_param, l_param, &result)) - { - return result; - } - } - break; + case WM_TIMER: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_TIMER"); + WINDOW_IMP_POST(window_imp->mCallbacks->handleTimerEvent(window_imp)); + break; + } - case WM_CHAR: - window_imp->mKeyCharCode = w_param; - window_imp->mRawMsg = u_msg; - window_imp->mRawWParam = w_param; - window_imp->mRawLParam = l_param; - - // Should really use WM_UNICHAR eventually, but it requires a specific Windows version and I need - // to figure out how that works. - Doug - // - // ... Well, I don't think so. - // How it works is explained in Win32 API document, but WM_UNICHAR didn't work - // as specified at least on Windows XP SP1 Japanese version. I have never used - // it since then, and I'm not sure whether it has been fixed now, but I don't think - // it is worth trying. The good old WM_CHAR works just fine even for supplementary - // characters. We just need to take care of surrogate pairs sent as two WM_CHAR's - // by ourselves. It is not that tough. -- Alissa Sabre @ SL - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_CHAR"); - if (debug_window_proc) - { - LL_INFOS("Window") << "Debug WindowProc WM_CHAR " - << " key " << S32(w_param) - << LL_ENDL; - } - // Even if LLWindowCallbacks::handleUnicodeChar(llwchar, BOOL) returned FALSE, - // we *did* processed the event, so I believe we should not pass it to DefWindowProc... - window_imp->handleUnicodeUTF16((U16)w_param, gKeyboard->currentMask(FALSE)); - return 0; + case WM_DEVICECHANGE: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_DEVICECHANGE"); + if (debug_window_proc) + { + LL_INFOS("Window") << " WM_DEVICECHANGE: wParam=" << w_param + << "; lParam=" << l_param << LL_ENDL; + } + if (w_param == DBT_DEVNODES_CHANGED || w_param == DBT_DEVICEARRIVAL) + { + WINDOW_IMP_POST(window_imp->mCallbacks->handleDeviceChange(window_imp)); + + return TRUE; + } + break; + } - case WM_NCLBUTTONDOWN: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_NCLBUTTONDOWN"); - // A click in a non-client area, e.g. title bar or window border. - sHandleLeftMouseUp = false; - sHandleDoubleClick = true; - } - break; + case WM_PAINT: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_PAINT"); + GetUpdateRect(window_imp->mWindowHandle, &update_rect, FALSE); + update_width = update_rect.right - update_rect.left + 1; + update_height = update_rect.bottom - update_rect.top + 1; + + WINDOW_IMP_POST(window_imp->mCallbacks->handlePaint(window_imp, update_rect.left, update_rect.top, + update_width, update_height)); + break; + } + case WM_PARENTNOTIFY: + { + break; + } - case WM_LBUTTONDOWN: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_LBUTTONDOWN"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - sHandleLeftMouseUp = true; + case WM_SETCURSOR: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_SETCURSOR"); + // This message is sent whenever the cursor is moved in a window. + // You need to set the appropriate cursor appearance. - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->interruptLanguageTextInput(); - } + // Only take control of cursor over client region of window + // This allows Windows(tm) to handle resize cursors, etc. + if (LOWORD(l_param) == HTCLIENT) + { + SetCursor(window_imp->mCursor[window_imp->mCurrentCursor]); + return 0; + } + break; + } + case WM_ENTERMENULOOP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_ENTERMENULOOP"); + WINDOW_IMP_POST(window_imp->mCallbacks->handleWindowBlock(window_imp)); + break; + } - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleMouseDown(window_imp, gl_coord, mask)) - { - return 0; - } - } - break; + case WM_EXITMENULOOP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_EXITMENULOOP"); + WINDOW_IMP_POST(window_imp->mCallbacks->handleWindowUnblock(window_imp)); + break; + } - case WM_LBUTTONDBLCLK: - //RN: ignore right button double clicks for now - //case WM_RBUTTONDBLCLK: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_LBUTTONDBLCLK"); + case WM_ACTIVATEAPP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_ACTIVATEAPP"); + window_imp->post([=]() + { + // This message should be sent whenever the app gains or loses focus. + BOOL activating = (BOOL)w_param; + BOOL minimized = window_imp->getMinimized(); + + if (debug_window_proc) + { + LL_INFOS("Window") << "WINDOWPROC ActivateApp " + << " activating " << S32(activating) + << " minimized " << S32(minimized) + << " fullscreen " << S32(window_imp->mFullscreen) + << LL_ENDL; + } + + if (window_imp->mFullscreen) + { + // When we run fullscreen, restoring or minimizing the app needs + // to switch the screen resolution + if (activating) + { + window_imp->setFullscreenResolution(); + window_imp->restore(); + } + else + { + window_imp->minimize(); + window_imp->resetDisplayResolution(); + } + } + + if (!activating) + { + sHandleDoubleClick = false; + } + + window_imp->mCallbacks->handleActivateApp(window_imp, activating); + }); + break; + } + case WM_ACTIVATE: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_ACTIVATE"); + window_imp->post([=]() + { + // Can be one of WA_ACTIVE, WA_CLICKACTIVE, or WA_INACTIVE + BOOL activating = (LOWORD(w_param) != WA_INACTIVE); + + BOOL minimized = BOOL(HIWORD(w_param)); + + if (!activating && LLWinImm::isAvailable() && window_imp->mPreeditor) + { + window_imp->interruptLanguageTextInput(); + } + + // JC - I'm not sure why, but if we don't report that we handled the + // WM_ACTIVATE message, the WM_ACTIVATEAPP messages don't work + // properly when we run fullscreen. + if (debug_window_proc) + { + LL_INFOS("Window") << "WINDOWPROC Activate " + << " activating " << S32(activating) + << " minimized " << S32(minimized) + << LL_ENDL; + } + }); + + break; + } - if (!sHandleDoubleClick) - { - sHandleDoubleClick = true; - break; - } + case WM_QUERYOPEN: + // TODO: use this to return a nice icon + break; - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleDoubleClick(window_imp, gl_coord, mask) ) - { - return 0; - } - } - break; + case WM_SYSCOMMAND: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_SYSCOMMAND"); + switch (w_param) + { + case SC_KEYMENU: + // Disallow the ALT key from triggering the default system menu. + return 0; + + case SC_SCREENSAVE: + case SC_MONITORPOWER: + // eat screen save messages and prevent them! + return 0; + } + break; + } + case WM_CLOSE: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_CLOSE"); + window_imp->post([=]() + { + // Will the app allow the window to close? + if (window_imp->mCallbacks->handleCloseRequest(window_imp)) + { + // Get the app to initiate cleanup. + window_imp->mCallbacks->handleQuit(window_imp); + // The app is responsible for calling destroyWindow when done with GL + } + }); + return 0; + } + case WM_DESTROY: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_DESTROY"); + if (window_imp->shouldPostQuit()) + { + PostQuitMessage(0); // Posts WM_QUIT with an exit code of 0 + } + return 0; + } + case WM_COMMAND: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_COMMAND"); + if (!HIWORD(w_param)) // this message is from a menu + { + WINDOW_IMP_POST(window_imp->mCallbacks->handleMenuSelect(window_imp, LOWORD(w_param))); + } + break; + } + case WM_SYSKEYDOWN: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_SYSKEYDOWN"); + // allow system keys, such as ALT-F4 to be processed by Windows + eat_keystroke = FALSE; + } + case WM_KEYDOWN: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_KEYDOWN"); + window_imp->post([=]() + { + window_imp->mKeyCharCode = 0; // don't know until wm_char comes in next + window_imp->mKeyScanCode = (l_param >> 16) & 0xff; + window_imp->mKeyVirtualKey = w_param; + window_imp->mRawMsg = u_msg; + window_imp->mRawWParam = w_param; + window_imp->mRawLParam = l_param; + + { + if (debug_window_proc) + { + LL_INFOS("Window") << "Debug WindowProc WM_KEYDOWN " + << " key " << S32(w_param) + << LL_ENDL; + } + + gKeyboard->handleKeyDown(w_param, mask); + } + }); + return eat_keystroke; + } + case WM_SYSKEYUP: + eat_keystroke = FALSE; + case WM_KEYUP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_KEYUP"); + window_imp->post([=]() + { + window_imp->mKeyScanCode = (l_param >> 16) & 0xff; + window_imp->mKeyVirtualKey = w_param; + window_imp->mRawMsg = u_msg; + window_imp->mRawWParam = w_param; + window_imp->mRawLParam = l_param; + + { + LL_RECORD_BLOCK_TIME(FTM_KEYHANDLER); + + if (debug_window_proc) + { + LL_INFOS("Window") << "Debug WindowProc WM_KEYUP " + << " key " << S32(w_param) + << LL_ENDL; + } + gKeyboard->handleKeyUp(w_param, mask); + } + }); + return eat_keystroke; + } + case WM_IME_SETCONTEXT: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_IME_SETCONTEXT"); + if (debug_window_proc) + { + LL_INFOS("Window") << "WM_IME_SETCONTEXT" << LL_ENDL; + } + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + l_param &= ~ISC_SHOWUICOMPOSITIONWINDOW; + // Invoke DefWinProc with the modified LPARAM. + } + break; + } + case WM_IME_STARTCOMPOSITION: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_IME_STARTCOMPOSITION"); + if (debug_window_proc) + { + LL_INFOS() << "WM_IME_STARTCOMPOSITION" << LL_ENDL; + } + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + WINDOW_IMP_POST(window_imp->handleStartCompositionMessage()); + return 0; + } + break; + } + case WM_IME_ENDCOMPOSITION: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_IME_ENDCOMPOSITION"); + if (debug_window_proc) + { + LL_INFOS() << "WM_IME_ENDCOMPOSITION" << LL_ENDL; + } + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + return 0; + } + break; + } + case WM_IME_COMPOSITION: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_IME_COMPOSITION"); + if (debug_window_proc) + { + LL_INFOS() << "WM_IME_COMPOSITION" << LL_ENDL; + } + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + WINDOW_IMP_POST(window_imp->handleCompositionMessage(l_param)); + return 0; + } + break; + } + case WM_IME_REQUEST: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_IME_REQUEST"); + if (debug_window_proc) + { + LL_INFOS() << "WM_IME_REQUEST" << LL_ENDL; + } + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + LRESULT result; + window_imp->handleImeRequests(w_param, l_param, &result); + return result; + } + break; + } + case WM_CHAR: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_CHAR"); + window_imp->post([=]() + { + window_imp->mKeyCharCode = w_param; + window_imp->mRawMsg = u_msg; + window_imp->mRawWParam = w_param; + window_imp->mRawLParam = l_param; + + // Should really use WM_UNICHAR eventually, but it requires a specific Windows version and I need + // to figure out how that works. - Doug + // + // ... Well, I don't think so. + // How it works is explained in Win32 API document, but WM_UNICHAR didn't work + // as specified at least on Windows XP SP1 Japanese version. I have never used + // it since then, and I'm not sure whether it has been fixed now, but I don't think + // it is worth trying. The good old WM_CHAR works just fine even for supplementary + // characters. We just need to take care of surrogate pairs sent as two WM_CHAR's + // by ourselves. It is not that tough. -- Alissa Sabre @ SL + if (debug_window_proc) + { + LL_INFOS("Window") << "Debug WindowProc WM_CHAR " + << " key " << S32(w_param) + << LL_ENDL; + } + // Even if LLWindowCallbacks::handleUnicodeChar(llwchar, BOOL) returned FALSE, + // we *did* processed the event, so I believe we should not pass it to DefWindowProc... + window_imp->handleUnicodeUTF16((U16)w_param, gKeyboard->currentMask(FALSE)); + }); + return 0; + } + case WM_NCLBUTTONDOWN: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_NCLBUTTONDOWN"); + { + // A click in a non-client area, e.g. title bar or window border. + window_imp->post([=]() + { + sHandleLeftMouseUp = false; + sHandleDoubleClick = true; + }); + } + break; + } + case WM_LBUTTONDOWN: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_LBUTTONDOWN"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + window_imp->post([=]() + { + auto glc = gl_coord; + sHandleLeftMouseUp = true; + + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + window_imp->interruptLanguageTextInput(); + } + + // Because we move the cursor position in the app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + glc = cursor_coord_window.convert(); + } + else + { + glc = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->mCallbacks->handleMouseMove(window_imp, glc, mask); + window_imp->mCallbacks->handleMouseDown(window_imp, glc, mask); + }); + + return 0; + } + break; + } - case WM_LBUTTONUP: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_LBUTTONUP"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + case WM_LBUTTONDBLCLK: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_LBUTTONDBLCLK"); + //RN: ignore right button double clicks for now + //case WM_RBUTTONDBLCLK: + if (!sHandleDoubleClick) + { + sHandleDoubleClick = true; + return 0; + } - if (!sHandleLeftMouseUp) - { - sHandleLeftMouseUp = true; - break; - } - sHandleDoubleClick = true; - - //if (gDebugClicks) - //{ - // LL_INFOS("Window") << "WndProc left button up" << LL_ENDL; - //} - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleMouseUp(window_imp, gl_coord, mask)) - { - return 0; - } - } - break; + // Because we move the cursor position in the app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + gl_coord = cursor_coord_window.convert(); + } + else + { + gl_coord = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->post([=]() + { + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + window_imp->mCallbacks->handleDoubleClick(window_imp, gl_coord, mask); + }); + return 0; + } + case WM_LBUTTONUP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_LBUTTONUP"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + + if (!sHandleLeftMouseUp) + { + sHandleLeftMouseUp = true; + return 0; + } + sHandleDoubleClick = true; + window_imp->post([=]() + { + auto glc = gl_coord; + + //if (gDebugClicks) + //{ + // LL_INFOS("Window") << "WndProc left button up" << LL_ENDL; + //} + // Because we move the cursor position in the app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + glc = cursor_coord_window.convert(); + } + else + { + glc = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->mCallbacks->handleMouseMove(window_imp, glc, mask); + window_imp->mCallbacks->handleMouseUp(window_imp, glc, mask); + }); + } + return 0; + } + case WM_RBUTTONDBLCLK: + case WM_RBUTTONDOWN: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_RBUTTONDOWN"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + WINDOW_IMP_POST(window_imp->interruptLanguageTextInput()); + } + + // Because we move the cursor position in the llviewerapp, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + gl_coord = cursor_coord_window.convert(); + } + else + { + gl_coord = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->post([=]() + { + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + window_imp->mCallbacks->handleRightMouseDown(window_imp, gl_coord, mask); + }); + } + return 0; + } + break; - case WM_RBUTTONDBLCLK: - case WM_RBUTTONDOWN: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_RBUTTONDOWN"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->interruptLanguageTextInput(); - } + case WM_RBUTTONUP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_RBUTTONUP"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + // Because we move the cursor position in the app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + gl_coord = cursor_coord_window.convert(); + } + else + { + gl_coord = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + if (window_imp->mCallbacks->handleRightMouseUp(window_imp, gl_coord, mask)) + { + return 0; + } + } + } + break; - // Because we move the cursor position in the llviewerapp, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleRightMouseDown(window_imp, gl_coord, mask)) - { - return 0; - } - } - break; + case WM_MBUTTONDOWN: + // case WM_MBUTTONDBLCLK: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_MBUTTONDOWN"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + window_imp->interruptLanguageTextInput(); + } + + // Because we move the cursor position in tllviewerhe app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + gl_coord = cursor_coord_window.convert(); + } + else + { + gl_coord = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + if (window_imp->mCallbacks->handleMiddleMouseDown(window_imp, gl_coord, mask)) + { + return 0; + } + } + } + break; - case WM_RBUTTONUP: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_RBUTTONUP"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleRightMouseUp(window_imp, gl_coord, mask)) - { - return 0; - } - } - break; + case WM_MBUTTONUP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_MBUTTONUP"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + // Because we move the cursor position in the llviewer app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + gl_coord = cursor_coord_window.convert(); + } + else + { + gl_coord = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + if (window_imp->mCallbacks->handleMiddleMouseUp(window_imp, gl_coord, mask)) + { + return 0; + } + } + } + break; + case WM_XBUTTONDOWN: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_XBUTTONDOWN"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + S32 button = GET_XBUTTON_WPARAM(w_param); + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + window_imp->interruptLanguageTextInput(); + } + + // Because we move the cursor position in tllviewerhe app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + gl_coord = cursor_coord_window.convert(); + } + else + { + gl_coord = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 + if (window_imp->mCallbacks->handleOtherMouseDown(window_imp, gl_coord, mask, button + 3)) + { + return 0; + } + } + } + break; - case WM_MBUTTONDOWN: -// case WM_MBUTTONDBLCLK: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_MBUTTONDOWN"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->interruptLanguageTextInput(); - } + case WM_XBUTTONUP: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_XBUTTONUP"); + { + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + S32 button = GET_XBUTTON_WPARAM(w_param); + // Because we move the cursor position in the llviewer app, we need to query + // to find out where the cursor at the time the event is handled. + // If we don't do this, many clicks could get buffered up, and if the + // first click changes the cursor position, all subsequent clicks + // will occur at the wrong location. JC + if (window_imp->mMousePositionModified) + { + LLCoordWindow cursor_coord_window; + window_imp->getCursorPosition(&cursor_coord_window); + gl_coord = cursor_coord_window.convert(); + } + else + { + gl_coord = window_coord.convert(); + } + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 + if (window_imp->mCallbacks->handleOtherMouseUp(window_imp, gl_coord, mask, button + 3)) + { + return 0; + } + } + } + break; - // Because we move the cursor position in tllviewerhe app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleMiddleMouseDown(window_imp, gl_coord, mask)) - { - return 0; - } - } - break; + case WM_MOUSEWHEEL: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_MOUSEWHEEL"); + static short z_delta = 0; - case WM_MBUTTONUP: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_MBUTTONUP"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - // Because we move the cursor position in the llviewer app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleMiddleMouseUp(window_imp, gl_coord, mask)) - { - return 0; - } - } - break; - case WM_XBUTTONDOWN: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_MBUTTONDOWN"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - S32 button = GET_XBUTTON_WPARAM(w_param); - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->interruptLanguageTextInput(); - } + RECT client_rect; - // Because we move the cursor position in tllviewerhe app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 - if (window_imp->mCallbacks->handleOtherMouseDown(window_imp, gl_coord, mask, button + 3)) - { - return 0; - } - } - break; + // eat scroll events that occur outside our window, since we use mouse position to direct scroll + // instead of keyboard focus + // NOTE: mouse_coord is in *window* coordinates for scroll events + POINT mouse_coord = { (S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param) }; - case WM_XBUTTONUP: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_MBUTTONUP"); - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - S32 button = GET_XBUTTON_WPARAM(w_param); - // Because we move the cursor position in the llviewer app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 - if (window_imp->mCallbacks->handleOtherMouseUp(window_imp, gl_coord, mask, button + 3)) - { - return 0; - } - } - break; + if (ScreenToClient(window_imp->mWindowHandle, &mouse_coord) + && GetClientRect(window_imp->mWindowHandle, &client_rect)) + { + // we have a valid mouse point and client rect + if (mouse_coord.x < client_rect.left || client_rect.right < mouse_coord.x + || mouse_coord.y < client_rect.top || client_rect.bottom < mouse_coord.y) + { + // mouse is outside of client rect, so don't do anything + return 0; + } + } - case WM_MOUSEWHEEL: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_MOUSEWHEEL"); - static short z_delta = 0; + S16 incoming_z_delta = HIWORD(w_param); + z_delta += incoming_z_delta; + // cout << "z_delta " << z_delta << endl; + + // current mouse wheels report changes in increments of zDelta (+120, -120) + // Future, higher resolution mouse wheels may report smaller deltas. + // So we sum the deltas and only act when we've exceeded WHEEL_DELTA + // + // If the user rapidly spins the wheel, we can get messages with + // large deltas, like 480 or so. Thus we need to scroll more quickly. + if (z_delta <= -WHEEL_DELTA || WHEEL_DELTA <= z_delta) + { + window_imp->mCallbacks->handleScrollWheel(window_imp, -z_delta / WHEEL_DELTA); + z_delta = 0; + } + return 0; + } + /* + // TODO: add this after resolving _WIN32_WINNT issue + case WM_MOUSELEAVE: + { + window_imp->mCallbacks->handleMouseLeave(window_imp); + + // TRACKMOUSEEVENT track_mouse_event; + // track_mouse_event.cbSize = sizeof( TRACKMOUSEEVENT ); + // track_mouse_event.dwFlags = TME_LEAVE; + // track_mouse_event.hwndTrack = h_wnd; + // track_mouse_event.dwHoverTime = HOVER_DEFAULT; + // TrackMouseEvent( &track_mouse_event ); + return 0; + } + */ + case WM_MOUSEHWHEEL: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_MOUSEHWHEEL"); + static short h_delta = 0; - RECT client_rect; + RECT client_rect; - // eat scroll events that occur outside our window, since we use mouse position to direct scroll - // instead of keyboard focus - // NOTE: mouse_coord is in *window* coordinates for scroll events - POINT mouse_coord = {(S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param)}; + // eat scroll events that occur outside our window, since we use mouse position to direct scroll + // instead of keyboard focus + // NOTE: mouse_coord is in *window* coordinates for scroll events + POINT mouse_coord = { (S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param) }; - if (ScreenToClient(window_imp->mWindowHandle, &mouse_coord) - && GetClientRect(window_imp->mWindowHandle, &client_rect)) - { - // we have a valid mouse point and client rect - if (mouse_coord.x < client_rect.left || client_rect.right < mouse_coord.x - || mouse_coord.y < client_rect.top || client_rect.bottom < mouse_coord.y) - { - // mouse is outside of client rect, so don't do anything - return 0; - } - } + if (ScreenToClient(window_imp->mWindowHandle, &mouse_coord) + && GetClientRect(window_imp->mWindowHandle, &client_rect)) + { + // we have a valid mouse point and client rect + if (mouse_coord.x < client_rect.left || client_rect.right < mouse_coord.x + || mouse_coord.y < client_rect.top || client_rect.bottom < mouse_coord.y) + { + // mouse is outside of client rect, so don't do anything + return 0; + } + } - S16 incoming_z_delta = HIWORD(w_param); - z_delta += incoming_z_delta; - // cout << "z_delta " << z_delta << endl; - - // current mouse wheels report changes in increments of zDelta (+120, -120) - // Future, higher resolution mouse wheels may report smaller deltas. - // So we sum the deltas and only act when we've exceeded WHEEL_DELTA - // - // If the user rapidly spins the wheel, we can get messages with - // large deltas, like 480 or so. Thus we need to scroll more quickly. - if (z_delta <= -WHEEL_DELTA || WHEEL_DELTA <= z_delta) - { - window_imp->mCallbacks->handleScrollWheel(window_imp, -z_delta / WHEEL_DELTA); - z_delta = 0; - } - return 0; - } - /* - // TODO: add this after resolving _WIN32_WINNT issue - case WM_MOUSELEAVE: - { - window_imp->mCallbacks->handleMouseLeave(window_imp); - - // TRACKMOUSEEVENT track_mouse_event; - // track_mouse_event.cbSize = sizeof( TRACKMOUSEEVENT ); - // track_mouse_event.dwFlags = TME_LEAVE; - // track_mouse_event.hwndTrack = h_wnd; - // track_mouse_event.dwHoverTime = HOVER_DEFAULT; - // TrackMouseEvent( &track_mouse_event ); - return 0; - } - */ - case WM_MOUSEHWHEEL: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_MOUSEHWHEEL"); - static short h_delta = 0; + S16 incoming_h_delta = HIWORD(w_param); + h_delta += incoming_h_delta; - RECT client_rect; + // If the user rapidly spins the wheel, we can get messages with + // large deltas, like 480 or so. Thus we need to scroll more quickly. + if (h_delta <= -WHEEL_DELTA || WHEEL_DELTA <= h_delta) + { + WINDOW_IMP_POST(window_imp->mCallbacks->handleScrollHWheel(window_imp, h_delta / WHEEL_DELTA)); + h_delta = 0; + } + return 0; + } + // Handle mouse movement within the window + case WM_MOUSEMOVE: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_MOUSEMOVE"); + if (!window_imp->mMousePositionModified) + { + MASK mask = gKeyboard->currentMask(TRUE); + WINDOW_IMP_POST(window_imp->mCallbacks->handleMouseMove(window_imp, window_coord.convert(), mask)); + } + return 0; + } - // eat scroll events that occur outside our window, since we use mouse position to direct scroll - // instead of keyboard focus - // NOTE: mouse_coord is in *window* coordinates for scroll events - POINT mouse_coord = {(S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param)}; + case WM_GETMINMAXINFO: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_GETMINMAXINFO"); + LPMINMAXINFO min_max = (LPMINMAXINFO)l_param; + min_max->ptMinTrackSize.x = window_imp->mMinWindowWidth; + min_max->ptMinTrackSize.y = window_imp->mMinWindowHeight; + return 0; + } - if (ScreenToClient(window_imp->mWindowHandle, &mouse_coord) - && GetClientRect(window_imp->mWindowHandle, &client_rect)) - { - // we have a valid mouse point and client rect - if (mouse_coord.x < client_rect.left || client_rect.right < mouse_coord.x - || mouse_coord.y < client_rect.top || client_rect.bottom < mouse_coord.y) - { - // mouse is outside of client rect, so don't do anything - return 0; - } - } + case WM_SIZE: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_SIZE"); + S32 width = S32(LOWORD(l_param)); + S32 height = S32(HIWORD(l_param)); - S16 incoming_h_delta = HIWORD(w_param); - h_delta += incoming_h_delta; + if (debug_window_proc) + { + BOOL maximized = (w_param == SIZE_MAXIMIZED); + BOOL restored = (w_param == SIZE_RESTORED); + BOOL minimized = (w_param == SIZE_MINIMIZED); + + LL_INFOS("Window") << "WINDOWPROC Size " + << width << "x" << height + << " max " << S32(maximized) + << " min " << S32(minimized) + << " rest " << S32(restored) + << LL_ENDL; + } - // If the user rapidly spins the wheel, we can get messages with - // large deltas, like 480 or so. Thus we need to scroll more quickly. - if (h_delta <= -WHEEL_DELTA || WHEEL_DELTA <= h_delta) - { - window_imp->mCallbacks->handleScrollHWheel(window_imp, h_delta / WHEEL_DELTA); - h_delta = 0; - } - return 0; - } - // Handle mouse movement within the window - case WM_MOUSEMOVE: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_MOUSEMOVE"); - MASK mask = gKeyboard->currentMask(TRUE); - window_imp->mCallbacks->handleMouseMove(window_imp, window_coord.convert(), mask); - return 0; - } + // There's an odd behavior with WM_SIZE that I would call a bug. If + // the window is maximized, and you call MoveWindow() with a size smaller + // than a maximized window, it ends up sending WM_SIZE with w_param set + // to SIZE_MAXIMIZED -- which isn't true. So the logic below doesn't work. + // (SL-44655). Fixed it by calling ShowWindow(SW_RESTORE) first (see + // LLWindowWin32::moveWindow in this file). - case WM_GETMINMAXINFO: - { - LPMINMAXINFO min_max = (LPMINMAXINFO)l_param; - min_max->ptMinTrackSize.x = window_imp->mMinWindowWidth; - min_max->ptMinTrackSize.y = window_imp->mMinWindowHeight; - return 0; - } + // If we are now restored, but we weren't before, this + // means that the window was un-minimized. + if (w_param == SIZE_RESTORED && window_imp->mLastSizeWParam != SIZE_RESTORED) + { + WINDOW_IMP_POST(window_imp->mCallbacks->handleActivate(window_imp, TRUE)); + } - case WM_SIZE: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_SIZE"); - S32 width = S32( LOWORD(l_param) ); - S32 height = S32( HIWORD(l_param) ); + // handle case of window being maximized from fully minimized state + if (w_param == SIZE_MAXIMIZED && window_imp->mLastSizeWParam != SIZE_MAXIMIZED) + { + WINDOW_IMP_POST(window_imp->mCallbacks->handleActivate(window_imp, TRUE)); + } - if (debug_window_proc) - { - BOOL maximized = ( w_param == SIZE_MAXIMIZED ); - BOOL restored = ( w_param == SIZE_RESTORED ); - BOOL minimized = ( w_param == SIZE_MINIMIZED ); - - LL_INFOS("Window") << "WINDOWPROC Size " - << width << "x" << height - << " max " << S32(maximized) - << " min " << S32(minimized) - << " rest " << S32(restored) - << LL_ENDL; - } + // Also handle the minimization case + if (w_param == SIZE_MINIMIZED && window_imp->mLastSizeWParam != SIZE_MINIMIZED) + { + WINDOW_IMP_POST(window_imp->mCallbacks->handleActivate(window_imp, FALSE)); + } - // There's an odd behavior with WM_SIZE that I would call a bug. If - // the window is maximized, and you call MoveWindow() with a size smaller - // than a maximized window, it ends up sending WM_SIZE with w_param set - // to SIZE_MAXIMIZED -- which isn't true. So the logic below doesn't work. - // (SL-44655). Fixed it by calling ShowWindow(SW_RESTORE) first (see - // LLWindowWin32::moveWindow in this file). + // Actually resize all of our views + if (w_param != SIZE_MINIMIZED) + { + // Ignore updates for minimizing and minimized "windows" + WINDOW_IMP_POST(window_imp->mCallbacks->handleResize(window_imp, + LOWORD(l_param), + HIWORD(l_param))); + } - // If we are now restored, but we weren't before, this - // means that the window was un-minimized. - if (w_param == SIZE_RESTORED && window_imp->mLastSizeWParam != SIZE_RESTORED) - { - window_imp->mCallbacks->handleActivate(window_imp, TRUE); - } + window_imp->mLastSizeWParam = w_param; - // handle case of window being maximized from fully minimized state - if (w_param == SIZE_MAXIMIZED && window_imp->mLastSizeWParam != SIZE_MAXIMIZED) - { - window_imp->mCallbacks->handleActivate(window_imp, TRUE); - } + return 0; + } - // Also handle the minimization case - if (w_param == SIZE_MINIMIZED && window_imp->mLastSizeWParam != SIZE_MINIMIZED) - { - window_imp->mCallbacks->handleActivate(window_imp, FALSE); - } + case WM_DPICHANGED: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_DPICHANGED"); + LPRECT lprc_new_scale; + F32 new_scale = F32(LOWORD(w_param)) / F32(USER_DEFAULT_SCREEN_DPI); + lprc_new_scale = (LPRECT)l_param; + S32 new_width = lprc_new_scale->right - lprc_new_scale->left; + S32 new_height = lprc_new_scale->bottom - lprc_new_scale->top; + WINDOW_IMP_POST(window_imp->mCallbacks->handleDPIChanged(window_imp, new_scale, new_width, new_height)); + + SetWindowPos(h_wnd, + HWND_TOP, + lprc_new_scale->left, + lprc_new_scale->top, + new_width, + new_height, + SWP_NOZORDER | SWP_NOACTIVATE); + + return 0; + } - // Actually resize all of our views - if (w_param != SIZE_MINIMIZED) - { - // Ignore updates for minimizing and minimized "windows" - window_imp->mCallbacks->handleResize( window_imp, - LOWORD(l_param), - HIWORD(l_param) ); - } + case WM_SETFOCUS: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_SETFOCUS"); + if (debug_window_proc) + { + LL_INFOS("Window") << "WINDOWPROC SetFocus" << LL_ENDL; + } + WINDOW_IMP_POST(window_imp->mCallbacks->handleFocus(window_imp)); + return 0; + } - window_imp->mLastSizeWParam = w_param; + case WM_KILLFOCUS: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_KILLFOCUS"); + if (debug_window_proc) + { + LL_INFOS("Window") << "WINDOWPROC KillFocus" << LL_ENDL; + } + WINDOW_IMP_POST(window_imp->mCallbacks->handleFocusLost(window_imp)); + return 0; + } - return 0; - } + case WM_COPYDATA: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_COPYDATA"); + { + // received a URL + PCOPYDATASTRUCT myCDS = (PCOPYDATASTRUCT)l_param; + void* data = new U8[myCDS->cbData]; + memcpy(data, myCDS->lpData, myCDS->cbData); + auto myType = myCDS->dwData; + + window_imp->post([=]() + { + window_imp->mCallbacks->handleDataCopy(window_imp, myType, data); + delete[] data; + }); + }; + return 0; + + break; + } + case WM_SETTINGCHANGE: + { + LL_PROFILE_ZONE_NAMED("mwp - WM_SETTINGCHANGE"); + if (w_param == SPI_SETMOUSEVANISH) + { + if (!SystemParametersInfo(SPI_GETMOUSEVANISH, 0, &window_imp->mMouseVanish, 0)) + { + WINDOW_IMP_POST(window_imp->mMouseVanish = TRUE); + } + } + } + break; - case WM_DPICHANGED: - { - LPRECT lprc_new_scale; - F32 new_scale = F32(LOWORD(w_param)) / F32(USER_DEFAULT_SCREEN_DPI); - lprc_new_scale = (LPRECT)l_param; - S32 new_width = lprc_new_scale->right - lprc_new_scale->left; - S32 new_height = lprc_new_scale->bottom - lprc_new_scale->top; - if (window_imp->mCallbacks->handleDPIChanged(window_imp, new_scale, new_width, new_height)) - { - SetWindowPos(h_wnd, - HWND_TOP, - lprc_new_scale->left, - lprc_new_scale->top, - new_width, - new_height, - SWP_NOZORDER | SWP_NOACTIVATE); - } - return 0; - } - - case WM_SETFOCUS: - if (debug_window_proc) - { - LL_INFOS("Window") << "WINDOWPROC SetFocus" << LL_ENDL; - } - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_SETFOCUS"); - window_imp->mCallbacks->handleFocus(window_imp); - return 0; - - case WM_KILLFOCUS: - if (debug_window_proc) - { - LL_INFOS("Window") << "WINDOWPROC KillFocus" << LL_ENDL; - } - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_KILLFOCUS"); - window_imp->mCallbacks->handleFocusLost(window_imp); - return 0; - - case WM_COPYDATA: - { - window_imp->mCallbacks->handlePingWatchdog(window_imp, "Main:WM_COPYDATA"); - // received a URL - PCOPYDATASTRUCT myCDS = (PCOPYDATASTRUCT) l_param; - window_imp->mCallbacks->handleDataCopy(window_imp, myCDS->dwData, myCDS->lpData); - }; - return 0; - - break; - - case WM_SETTINGCHANGE: - { - if (w_param == SPI_SETMOUSEVANISH) - { - if (!SystemParametersInfo(SPI_GETMOUSEVANISH, 0, &window_imp->mMouseVanish, 0)) - { - window_imp->mMouseVanish = TRUE; - } - } - } - break; - default: - { - if (debug_window_proc) - { - LL_INFOS("Window") << "Unhandled windows message code: " << U32(u_msg) << LL_ENDL; - } - } - break; - } - - window_imp->mCallbacks->handlePauseWatchdog(window_imp); - } + //list of messages we get often that we don't care to log about + case WM_NCHITTEST: + case WM_NCMOUSEMOVE: + case WM_NCMOUSELEAVE: + case WM_MOVING: + case WM_MOVE: + case WM_WINDOWPOSCHANGING: + case WM_WINDOWPOSCHANGED: + break; + + default: + { + LL_PROFILE_ZONE_NAMED("mwp - default"); + if (debug_window_proc) + { + LL_INFOS("Window") << "Unhandled windows message code: 0x" << std::hex << U32(u_msg) << LL_ENDL; + } + } + break; + } + } else { // (NULL == window_imp) LL_DEBUGS("Window") << "No window implementation to handle message with, message code: " << U32(u_msg) << LL_ENDL; } - // pass unhandled messages down to Windows - return DefWindowProc(h_wnd, u_msg, w_param, l_param); + // pass unhandled messages down to Windows + LRESULT ret; + { + LL_PROFILE_ZONE_NAMED("mwp - DefWindowProc"); + ret = DefWindowProc(h_wnd, u_msg, w_param, l_param); + } + return ret; } BOOL LLWindowWin32::convertCoords(LLCoordGL from, LLCoordWindow *to) @@ -3184,6 +3425,8 @@ BOOL LLWindowWin32::copyTextToClipboard(const LLWString& wstr) // Constrains the mouse to the window. void LLWindowWin32::setMouseClipping( BOOL b ) { + LL_PROFILE_ZONE_SCOPED; + ASSERT_MAIN_THREAD(); if( b != mIsMouseClipping ) { BOOL success = FALSE; @@ -3260,6 +3503,7 @@ F32 LLWindowWin32::getGamma() BOOL LLWindowWin32::restoreGamma() { + ASSERT_MAIN_THREAD(); if (mCustomGammaSet != FALSE) { LL_DEBUGS("Window") << "Restoring gamma" << LL_ENDL; @@ -3271,6 +3515,7 @@ BOOL LLWindowWin32::restoreGamma() BOOL LLWindowWin32::setGamma(const F32 gamma) { + ASSERT_MAIN_THREAD(); mCurrentGamma = gamma; //Get the previous gamma ramp to restore later. @@ -3309,6 +3554,7 @@ BOOL LLWindowWin32::setGamma(const F32 gamma) void LLWindowWin32::setFSAASamples(const U32 fsaa_samples) { + ASSERT_MAIN_THREAD(); mFSAASamples = fsaa_samples; } @@ -3319,6 +3565,7 @@ U32 LLWindowWin32::getFSAASamples() LLWindow::LLWindowResolution* LLWindowWin32::getSupportedResolutions(S32 &num_resolutions) { + ASSERT_MAIN_THREAD(); if (!mSupportedResolutions) { mSupportedResolutions = new LLWindowResolution[MAX_NUM_RESOLUTIONS]; @@ -3473,7 +3720,10 @@ BOOL LLWindowWin32::resetDisplayResolution() void LLWindowWin32::swapBuffers() { + ASSERT_MAIN_THREAD(); SwapBuffers(mhDC); + + LL_PROFILER_GPU_COLLECT } @@ -3946,6 +4196,7 @@ void LLWindowWin32::updateLanguageTextInputArea() void LLWindowWin32::interruptLanguageTextInput() { + ASSERT_MAIN_THREAD(); if (mPreeditor && LLWinImm::isAvailable()) { HIMC himc = LLWinImm::getContext(mWindowHandle); @@ -4148,6 +4399,7 @@ static LLWString find_context(const LLWString & wtext, S32 focus, S32 focus_leng // for files and via IDropTarget interface requests. LLWindowCallbacks::DragNDropResult LLWindowWin32::completeDragNDropRequest( const LLCoordGL gl_coord, const MASK mask, LLWindowCallbacks::DragNDropAction action, const std::string url ) { + ASSERT_MAIN_THREAD(); return mCallbacks->handleDragNDrop( this, gl_coord, mask, action, url ); } @@ -4196,6 +4448,7 @@ BOOL LLWindowWin32::handleImeRequests(WPARAM request, LPARAM param, LRESULT *res LL_WARNS("Window") << "*** IMR_QUERYCHARPOSITON called but getPreeditLocation failed." << LL_ENDL; return FALSE; } + fillCharPosition(caret_coord, preedit_bounds, text_control, char_position); *result = 1; @@ -4403,3 +4656,79 @@ std::vector<std::string> LLWindowWin32::getDynamicFallbackFontList() #endif // LL_WINDOWS + +inline LLWindowWin32Thread::LLWindowWin32Thread(LLWindowWin32* window) + : LLThread("Window Thread"), + mWindow(window), + mFunctionQueue(MAX_QUEUE_SIZE) +{ + +} + +inline void LLWindowWin32Thread::run() +{ + sWindowThreadId = getID(); + while (!mFinished) + { + LL_PROFILE_ZONE_SCOPED; + + + if (mWindow && mWindow->mWindowHandle != 0) + { + MSG msg; + BOOL status; + if (mWindow->mhDC == 0) + { + LL_PROFILE_ZONE_NAMED("w32t - PeekMessage"); + status = PeekMessage(&msg, mWindow->mWindowHandle, 0, 0, PM_REMOVE); + } + else + { + LL_PROFILE_ZONE_NAMED("w32t - GetMessage"); + status = GetMessage(&msg, mWindow->mWindowHandle, 0, 0); + } + if (status > 0) + { + TranslateMessage(&msg); + DispatchMessage(&msg); + + mMessageQueue.pushFront(msg); + } + } + + { + LL_PROFILE_ZONE_NAMED("w32t - Function Queue"); + //process any pending functions + std::function<void()> curFunc; + while (mFunctionQueue.tryPopBack(curFunc)) + { + curFunc(); + } + } + +#if 0 + { + LL_PROFILE_ZONE_NAMED("w32t - Sleep"); + std::this_thread::sleep_for(std::chrono::milliseconds(1)); + } +#endif + } +} + +void LLWindowWin32Thread::post(const std::function<void()>& func) +{ +#if LL_WINDOW_SINGLE_THREADED + func(); +#else + mFunctionQueue.pushFront(func); +#endif +} + +void LLWindowWin32::post(const std::function<void()>& func) +{ +#if LL_WINDOW_SINGLE_THREADED + func(); +#else + mFunctionQueue.pushFront(func); +#endif +}
\ No newline at end of file diff --git a/indra/llwindow/llwindowwin32.h b/indra/llwindow/llwindowwin32.h index 0b3d14fb16..66647459b2 100644 --- a/indra/llwindow/llwindowwin32.h +++ b/indra/llwindow/llwindowwin32.h @@ -33,11 +33,46 @@ #include "llwindow.h" #include "llwindowcallbacks.h" #include "lldragdropwin32.h" +#include "llthread.h" +#include "llthreadsafequeue.h" // Hack for async host by name #define LL_WM_HOST_RESOLVED (WM_APP + 1) typedef void (*LLW32MsgCallback)(const MSG &msg); +class LLWindowWin32; + +// Thread that owns the Window Handle +class LLWindowWin32Thread : public LLThread +{ +public: + class Message + { + public: + LRESULT mMsg; + }; + + static const int MAX_QUEUE_SIZE = 2048; + + LLThreadSafeQueue<MSG> mMessageQueue; + LLThreadSafeQueue<std::function<void()>> mFunctionQueue; + + bool mFinished = false; + + LLWindowWin32Thread(LLWindowWin32* window); + + void run() override; + + void post(const std::function<void()>& func); + +private: + + // call PeekMessage and pull enqueue messages for later processing + void gatherInput(); + LLWindowWin32* mWindow = nullptr; + +}; + class LLWindowWin32 : public LLWindow { public: @@ -172,9 +207,9 @@ protected: WCHAR *mWindowTitle; WCHAR *mWindowClassName; - HWND mWindowHandle; // window handle - HGLRC mhRC; // OpenGL rendering context - HDC mhDC; // Windows Device context handle + HWND mWindowHandle = 0; // window handle + HGLRC mhRC = 0; // OpenGL rendering context + HDC mhDC = 0; // Windows Device context handle HINSTANCE mhInstance; // handle to application instance WNDPROC mWndProc; // user-installable window proc RECT mOldMouseClip; // Screen rect to which the mouse cursor was globally constrained before we changed it in clipMouse() @@ -221,7 +256,12 @@ protected: BOOL mMouseVanish; + LLWindowWin32Thread* mWindowThread = nullptr; + LLThreadSafeQueue<std::function<void()>> mFunctionQueue; + void post(const std::function<void()>& func); + friend class LLWindowManager; + friend class LLWindowWin32Thread; }; class LLSplashScreenWin32 : public LLSplashScreen diff --git a/indra/llxml/llcontrol.h b/indra/llxml/llcontrol.h index 19508becc3..5da13f5010 100644 --- a/indra/llxml/llcontrol.h +++ b/indra/llxml/llcontrol.h @@ -405,8 +405,8 @@ public: const T& default_value, const std::string& comment = "Declared In Code") { - mCachedControlPtr = LLControlCache<T>::getInstance(name); - if (mCachedControlPtr.isNull()) + mCachedControlPtr = LLControlCache<T>::getInstance(name).get(); + if (! mCachedControlPtr) { mCachedControlPtr = new LLControlCache<T>(group, name, default_value, comment); } @@ -415,8 +415,8 @@ public: LLCachedControl(LLControlGroup& group, const std::string& name) { - mCachedControlPtr = LLControlCache<T>::getInstance(name); - if (mCachedControlPtr.isNull()) + mCachedControlPtr = LLControlCache<T>::getInstance(name).get(); + if (! mCachedControlPtr) { mCachedControlPtr = new LLControlCache<T>(group, name); } diff --git a/indra/newview/CMakeLists.txt b/indra/newview/CMakeLists.txt index 68b5969ff1..0144cff4b2 100644 --- a/indra/newview/CMakeLists.txt +++ b/indra/newview/CMakeLists.txt @@ -47,6 +47,7 @@ include(OpenGL) include(OpenSSL) include(PNG) include(TemplateCheck) +include(Tracy) include(UI) include(UnixInstall) include(ViewerMiscLibs) @@ -92,6 +93,7 @@ include_directories( ${LIBS_PREBUILT_DIR}/include/collada/1.4 ${LLAPPEARANCE_INCLUDE_DIRS} ${CMAKE_CURRENT_SOURCE_DIR} + ${TRACY_INCLUDE_DIR} ) include_directories(SYSTEM @@ -391,7 +393,6 @@ set(viewer_SOURCE_FILES llloginhandler.cpp lllogininstance.cpp llmachineid.cpp - llmainlooprepeater.cpp llmanip.cpp llmaniprotate.cpp llmanipscale.cpp @@ -587,7 +588,6 @@ set(viewer_SOURCE_FILES llsyntaxid.cpp llsyswellitem.cpp llsyswellwindow.cpp - lltelemetry.cpp llteleporthistory.cpp llteleporthistorystorage.cpp lltextureatlas.cpp @@ -1031,7 +1031,6 @@ set(viewer_HEADER_FILES llloginhandler.h lllogininstance.h llmachineid.h - llmainlooprepeater.h llmanip.h llmaniprotate.h llmanipscale.h @@ -2065,6 +2064,7 @@ target_link_libraries(${VIEWER_BINARY_NAME} ${LLPHYSICS_LIBRARIES} ${LLPHYSICSEXTENSIONS_LIBRARIES} ${LLAPPEARANCE_LIBRARIES} + ${TRACY_LIBRARY} ) if (USE_BUGSPLAT) diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index b1120c18b2..3c7fe174fd 100644 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -812,17 +812,6 @@ <key>Value</key> <integer>0</integer> </map> - <key>FramePerSecondLimit</key> - <map> - <key>Comment</key> - <string>Controls upper limit of frames per second</string> - <key>Persist</key> - <integer>1</integer> - <key>Type</key> - <string>U32</string> - <key>Value</key> - <integer>120</integer> - </map> <key>BackgroundYieldTime</key> <map> <key>Comment</key> @@ -3366,13 +3355,13 @@ <key>DisableVerticalSync</key> <map> <key>Comment</key> - <string>Update frames as fast as possible (FALSE = update frames between display scans)</string> + <string>Update frames as fast as possible (FALSE = update frames between display scans). Requires restart.</string> <key>Persist</key> <integer>1</integer> <key>Type</key> <string>Boolean</string> <key>Value</key> - <integer>1</integer> + <integer>0</integer> </map> <key>EnableGroupChatPopups</key> <map> @@ -3869,6 +3858,17 @@ <key>Value</key> <integer>1</integer> </map> + <key>MainWorkTime</key> + <map> + <key>Comment</key> + <string>Max time per frame devoted to mainloop work queue (in milliseconds)</string> + <key>Persist</key> + <integer>1</integer> + <key>Type</key> + <string>F32</string> + <key>Value</key> + <real>0.1</real> + </map> <key>QueueInventoryFetchTimeout</key> <map> <key>Comment</key> @@ -12663,6 +12663,20 @@ <key>Value</key> <integer>50</integer> </map> + <key>ThreadPoolSizes</key> + <map> + <key>Comment</key> + <string>Map of size overrides for specific thread pools.</string> + <key>Persist</key> + <integer>1</integer> + <key>Type</key> + <string>LLSD</string> + <key>Value</key> + <map> + <key>General</key> + <integer>4</integer> + </map> + </map> <key>ThrottleBandwidthKBPS</key> <map> <key>Comment</key> diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 69606793db..a016e658ef 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -91,7 +91,6 @@ #include "llsdutil_math.h" #include "lllocationhistory.h" #include "llfasttimerview.h" -#include "lltelemetry.h" #include "llvector4a.h" #include "llviewermenufile.h" #include "llvoicechannel.h" @@ -234,11 +233,12 @@ #include "llavatariconctrl.h" #include "llgroupiconctrl.h" #include "llviewerassetstats.h" +#include "workqueue.h" +using namespace LL; // Include for security api initialization #include "llsecapi.h" #include "llmachineid.h" -#include "llmainlooprepeater.h" #include "llcleanup.h" #include "llcoproceduremanager.h" @@ -367,6 +367,8 @@ BOOL gLogoutInProgress = FALSE; BOOL gSimulateMemLeak = FALSE; +WorkQueue gMainloopWork("mainloop"); + //////////////////////////////////////////////////////////// // Internal globals... that should be removed. static std::string gArgs; @@ -382,42 +384,6 @@ static std::string gLaunchFileOnQuit; // Used on Win32 for other apps to identify our window (eg, win_setup) const char* const VIEWER_WINDOW_CLASSNAME = "Second Life"; -//-- LLDeferredTaskList ------------------------------------------------------ - -/** - * A list of deferred tasks. - * - * We sometimes need to defer execution of some code until the viewer gets idle, - * e.g. removing an inventory item from within notifyObservers() may not work out. - * - * Tasks added to this list will be executed in the next LLAppViewer::idle() iteration. - * All tasks are executed only once. - */ -class LLDeferredTaskList: public LLSingleton<LLDeferredTaskList> -{ - LLSINGLETON_EMPTY_CTOR(LLDeferredTaskList); - LOG_CLASS(LLDeferredTaskList); - - friend class LLAppViewer; - typedef boost::signals2::signal<void()> signal_t; - - void addTask(const signal_t::slot_type& cb) - { - mSignal.connect(cb); - } - - void run() - { - if (!mSignal.empty()) - { - mSignal(); - mSignal.disconnect_all_slots(); - } - } - - signal_t mSignal; -}; - //---------------------------------------------------------------------------- // List of entries from strings.xml to always replace @@ -695,8 +661,7 @@ LLAppViewer::LLAppViewer() mPeriodicSlowFrame(LLCachedControl<bool>(gSavedSettings,"Periodic Slow Frame", FALSE)), mFastTimerLogThread(NULL), mSettingsLocationList(NULL), - mIsFirstRun(false), - mMinMicroSecPerFrame(0.f) + mIsFirstRun(false) { if(NULL != sInstance) { @@ -978,9 +943,6 @@ bool LLAppViewer::init() } LL_INFOS("InitInfo") << "Cache initialization is done." << LL_ENDL ; - // Initialize the repeater service. - LLMainLoopRepeater::instance().start(); - // // Initialize the window // @@ -1012,19 +974,6 @@ bool LLAppViewer::init() return 0; } - // If we don't have the right shader requirements. - if (!gGLManager.mHasShaderObjects - || !gGLManager.mHasVertexShader - || !gGLManager.mHasFragmentShader) - { - LLUIString details = LLNotifications::instance().getGlobalString("UnsupportedShaderRequirements"); - OSMessageBox( - details.getString(), - LLStringUtil::null, - OSMB_OK); - return 0; - } - // Without SSE2 support we will crash almost immediately, warn here. if (!gSysCPU.hasSSE2()) { @@ -1327,10 +1276,6 @@ bool LLAppViewer::init() joystick = LLViewerJoystick::getInstance(); joystick->setNeedsReset(true); /*----------------------------------------------------------------------*/ - - gSavedSettings.getControl("FramePerSecondLimit")->getSignal()->connect(boost::bind(&LLAppViewer::onChangeFrameLimit, this, _2)); - onChangeFrameLimit(gSavedSettings.getLLSD("FramePerSecondLimit")); - // Load User's bindings loadKeyBindings(); @@ -1359,7 +1304,8 @@ void LLAppViewer::initMaxHeapSize() } static LLTrace::BlockTimerStatHandle FTM_MESSAGES("System Messages"); -static LLTrace::BlockTimerStatHandle FTM_SLEEP("Sleep"); +static LLTrace::BlockTimerStatHandle FTM_SLEEP1("Sleep1"); +static LLTrace::BlockTimerStatHandle FTM_SLEEP2("Sleep2"); static LLTrace::BlockTimerStatHandle FTM_YIELD("Yield"); static LLTrace::BlockTimerStatHandle FTM_TEXTURE_CACHE("Texture Cache"); @@ -1421,13 +1367,17 @@ bool LLAppViewer::frame() bool LLAppViewer::doFrame() { + LL_RECORD_BLOCK_TIME(FTM_FRAME); + LLEventPump& mainloop(LLEventPumps::instance().obtain("mainloop")); LLSD newFrame; - LL_RECORD_BLOCK_TIME(FTM_FRAME); - LLTrace::BlockTimer::processTimes(); - LLTrace::get_frame_recording().nextPeriod(); - LLTrace::BlockTimer::logStats(); + { + LL_PROFILE_ZONE_NAMED( "df blocktimer" ) + LLTrace::BlockTimer::processTimes(); + LLTrace::get_frame_recording().nextPeriod(); + LLTrace::BlockTimer::logStats(); + } LLTrace::get_thread_recorder()->pullFromChildren(); @@ -1435,6 +1385,7 @@ bool LLAppViewer::doFrame() LL_CLEAR_CALLSTACKS(); { + LL_PROFILE_ZONE_NAMED( "df processMiscNativeEvents" ) pingMainloopTimeout("Main:MiscNativeWindowEvents"); if (gViewerWindow) @@ -1443,7 +1394,10 @@ bool LLAppViewer::doFrame() gViewerWindow->getWindow()->processMiscNativeEvents(); } - pingMainloopTimeout("Main:GatherInput"); + { + LL_PROFILE_ZONE_NAMED( "df gatherInput" ) + pingMainloopTimeout("Main:GatherInput"); + } if (gViewerWindow) { @@ -1467,13 +1421,21 @@ bool LLAppViewer::doFrame() } } - // canonical per-frame event - mainloop.post(newFrame); - // give listeners a chance to run - llcoro::suspend(); + { + LL_PROFILE_ZONE_NAMED( "df mainloop" ) + // canonical per-frame event + mainloop.post(newFrame); + } + + { + LL_PROFILE_ZONE_NAMED( "df suspend" ) + // give listeners a chance to run + llcoro::suspend(); + } if (!LLApp::isExiting()) { + LL_PROFILE_ZONE_NAMED( "df JoystickKeyboard" ) pingMainloopTimeout("Main:JoystickKeyboard"); // Scan keyboard for movement keys. Command keys and typing @@ -1494,12 +1456,18 @@ bool LLAppViewer::doFrame() // Update state based on messages, user input, object idle. { - pauseMainloopTimeout(); // *TODO: Remove. Messages shouldn't be stalling for 20+ seconds! + { + LL_PROFILE_ZONE_NAMED( "df pauseMainloopTimeout" ) + pauseMainloopTimeout(); // *TODO: Remove. Messages shouldn't be stalling for 20+ seconds! + } LL_RECORD_BLOCK_TIME(FTM_IDLE); idle(); - resumeMainloopTimeout(); + { + LL_PROFILE_ZONE_NAMED( "df resumeMainloopTimeout" ) + resumeMainloopTimeout(); + } } if (gDoDisconnect && (LLStartUp::getStartupState() == STATE_STARTED)) @@ -1514,46 +1482,40 @@ bool LLAppViewer::doFrame() // *TODO: Should we run display() even during gHeadlessClient? DK 2011-02-18 if (!LLApp::isExiting() && !gHeadlessClient && gViewerWindow) { + LL_PROFILE_ZONE_NAMED( "df Display" ) pingMainloopTimeout("Main:Display"); gGLActive = TRUE; display(); - static U64 last_call = 0; - if (!gTeleportDisplay) { - // Frame/draw throttling, controlled by FramePerSecondLimit - U64 elapsed_time = LLTimer::getTotalTime() - last_call; - if (elapsed_time < mMinMicroSecPerFrame) - { - LL_RECORD_BLOCK_TIME(FTM_SLEEP); - // llclamp for when time function gets funky - U64 sleep_time = llclamp(mMinMicroSecPerFrame - elapsed_time, (U64)1, (U64)1e6); - micro_sleep(sleep_time, 0); - } - } - last_call = LLTimer::getTotalTime(); - - pingMainloopTimeout("Main:Snapshot"); - LLFloaterSnapshot::update(); // take snapshots + LL_PROFILE_ZONE_NAMED( "df Snapshot" ) + pingMainloopTimeout("Main:Snapshot"); + LLFloaterSnapshot::update(); // take snapshots LLFloaterOutfitSnapshot::update(); - gGLActive = FALSE; + gGLActive = FALSE; + } } } - pingMainloopTimeout("Main:Sleep"); + { + LL_PROFILE_ZONE_NAMED( "df pauseMainloopTimeout" ) + pingMainloopTimeout("Main:Sleep"); - pauseMainloopTimeout(); + pauseMainloopTimeout(); + } // Sleep and run background threads { - LL_RECORD_BLOCK_TIME(FTM_SLEEP); + //LL_RECORD_BLOCK_TIME(SLEEP2); + LL_PROFILE_ZONE_WARN( "Sleep2" ) // yield some time to the os based on command line option static LLCachedControl<S32> yield_time(gSavedSettings, "YieldTime", -1); if(yield_time >= 0) { LL_RECORD_BLOCK_TIME(FTM_YIELD); + LL_PROFILE_ZONE_NUM( yield_time ) ms_sleep(yield_time); } @@ -1616,16 +1578,22 @@ bool LLAppViewer::doFrame() total_io_pending += io_pending ; } - gMeshRepo.update() ; + + { + LL_PROFILE_ZONE_NAMED( "df gMeshRepo" ) + gMeshRepo.update() ; + } if(!total_work_pending) //pause texture fetching threads if nothing to process. { + LL_PROFILE_ZONE_NAMED( "df getTextureCache" ) LLAppViewer::getTextureCache()->pause(); LLAppViewer::getImageDecodeThread()->pause(); LLAppViewer::getTextureFetch()->pause(); } if(!total_io_pending) //pause file threads if nothing to process. { + LL_PROFILE_ZONE_NAMED( "df LLVFSThread" ) LLVFSThread::sLocal->pause(); LLLFSThread::sLocal->pause(); } @@ -1633,6 +1601,7 @@ bool LLAppViewer::doFrame() //texture fetching debugger if(LLTextureFetchDebugger::isEnabled()) { + LL_PROFILE_ZONE_NAMED( "df tex_fetch_debugger_instance" ) LLFloaterTextureFetchDebugger* tex_fetch_debugger_instance = LLFloaterReg::findTypedInstance<LLFloaterTextureFetchDebugger>("tex_fetch_debugger"); if(tex_fetch_debugger_instance) @@ -1641,8 +1610,10 @@ bool LLAppViewer::doFrame() } } - resumeMainloopTimeout(); - + { + LL_PROFILE_ZONE_NAMED( "df resumeMainloopTimeout" ) + resumeMainloopTimeout(); + } pingMainloopTimeout("Main:End"); } } @@ -1668,7 +1639,7 @@ bool LLAppViewer::doFrame() LL_INFOS() << "Exiting main_loop" << LL_ENDL; } - LLPROFILE_UPDATE(); + LL_PROFILER_FRAME_END return ! LLApp::isRunning(); } @@ -2162,8 +2133,6 @@ bool LLAppViewer::cleanup() SUBSYSTEM_CLEANUP(LLProxy); LLCore::LLHttp::cleanup(); - LLMainLoopRepeater::instance().stop(); - ll_close_fail_log(); LLError::LLCallStacks::cleanup(); @@ -2338,7 +2307,7 @@ bool LLAppViewer::loadSettingsFromDirectory(const std::string& location_key, LL_INFOS("Settings") << "Attempting to load settings for the group " << file.name() << " - from location " << location_key << LL_ENDL; - LLControlGroup* settings_group = LLControlGroup::getInstance(file.name); + auto settings_group = LLControlGroup::getInstance(file.name); if(!settings_group) { LL_WARNS("Settings") << "No matching settings group for name " << file.name() << LL_ENDL; @@ -2483,12 +2452,7 @@ bool LLAppViewer::initConfiguration() #ifndef LL_RELEASE_FOR_DOWNLOAD // provide developer build only overrides for these control variables that are not // persisted to settings.xml - LLControlVariable* c = gSavedSettings.getControl("ShowConsoleWindow"); - if (c) - { - c->setValue(true, false); - } - c = gSavedSettings.getControl("AllowMultipleViewers"); + LLControlVariable* c = gSavedSettings.getControl("AllowMultipleViewers"); if (c) { c->setValue(true, false); @@ -2640,7 +2604,7 @@ bool LLAppViewer::initConfiguration() group_part = name.substr(0, pos); name_part = name.substr(pos+1); LL_INFOS() << "Setting " << group_part << "." << name_part << " to " << value << LL_ENDL; - LLControlGroup* g = LLControlGroup::getInstance(group_part); + auto g = LLControlGroup::getInstance(group_part); if (g) control = g->getControl(name_part); } else @@ -4433,7 +4397,7 @@ bool LLAppViewer::initCache() void LLAppViewer::addOnIdleCallback(const boost::function<void()>& cb) { - LLDeferredTaskList::instance().addTask(cb); + gMainloopWork.post(cb); } void LLAppViewer::loadKeyBindings() @@ -5207,8 +5171,19 @@ void LLAppViewer::idle() } } - // Execute deferred tasks. - LLDeferredTaskList::instance().run(); + // Service the WorkQueue we use for replies from worker threads. + // Use function statics for the timeslice setting so we only have to fetch + // and convert MainWorkTime once. + static F32 MainWorkTimeRaw = gSavedSettings.getF32("MainWorkTime"); + static F32Milliseconds MainWorkTimeMs(MainWorkTimeRaw); + // MainWorkTime is specified in fractional milliseconds, but std::chrono + // uses integer representations. What if we want less than a microsecond? + // Use nanoseconds. We're very sure we will never need to specify a + // MainWorkTime that would be larger than we could express in + // std::chrono::nanoseconds. + static std::chrono::nanoseconds MainWorkTimeNanoSec{ + std::chrono::nanoseconds::rep(MainWorkTimeMs.value() * 1000000)}; + gMainloopWork.runFor(MainWorkTimeNanoSec); // Handle shutdown process, for example, // wait for floaters to close, send quit message, @@ -5616,19 +5591,6 @@ void LLAppViewer::disconnectViewer() LLUrlEntryParcel::setDisconnected(gDisconnected); } -bool LLAppViewer::onChangeFrameLimit(LLSD const & evt) -{ - if (evt.asInteger() > 0) - { - mMinMicroSecPerFrame = (U64)(1000000.0f / F32(evt.asInteger())); - } - else - { - mMinMicroSecPerFrame = 0; - } - return false; -} - void LLAppViewer::forceErrorLLError() { LL_ERRS() << "This is a deliberate llerror" << LL_ENDL; diff --git a/indra/newview/llappviewer.h b/indra/newview/llappviewer.h index 95f6efa29a..37119aeef9 100644 --- a/indra/newview/llappviewer.h +++ b/indra/newview/llappviewer.h @@ -261,8 +261,6 @@ private: void sendLogoutRequest(); void disconnectViewer(); - bool onChangeFrameLimit(LLSD const & evt); - // *FIX: the app viewer class should be some sort of singleton, no? // Perhaps its child class is the singleton and this should be an abstract base. static LLAppViewer* sInstance; @@ -318,10 +316,7 @@ private: // llcorehttp library init/shutdown helper LLAppCoreHttp mAppCoreHttp; - bool mIsFirstRun; - U64 mMinMicroSecPerFrame; // frame throttling - - + bool mIsFirstRun; }; // consts from viewer.h diff --git a/indra/newview/llappviewerwin32.cpp b/indra/newview/llappviewerwin32.cpp index 25d18fa11f..758bd73cb0 100644 --- a/indra/newview/llappviewerwin32.cpp +++ b/indra/newview/llappviewerwin32.cpp @@ -323,6 +323,10 @@ int APIENTRY WINMAIN(HINSTANCE hInstance, PWSTR pCmdLine, int nCmdShow) { + // Call Tracy first thing to have it allocate memory + // https://github.com/wolfpld/tracy/issues/196 + LL_PROFILER_FRAME_END + const S32 MAX_HEAPS = 255; DWORD heap_enable_lfh_error[MAX_HEAPS]; S32 num_heaps = 0; diff --git a/indra/newview/llbrowsernotification.cpp b/indra/newview/llbrowsernotification.cpp index 0460bff1b4..30ac35fff7 100644 --- a/indra/newview/llbrowsernotification.cpp +++ b/indra/newview/llbrowsernotification.cpp @@ -43,14 +43,14 @@ LLBrowserNotification::LLBrowserNotification() bool LLBrowserNotification::processNotification(const LLNotificationPtr& notification) { LLUUID media_id = notification->getPayload()["media_id"].asUUID(); - LLMediaCtrl* media_instance = LLMediaCtrl::getInstance(media_id); + auto media_instance = LLMediaCtrl::getInstance(media_id); if (media_instance) { media_instance->showNotification(notification); } else if (LLViewerMediaFocus::instance().getControlsMediaID() == media_id) { - LLViewerMediaImpl* impl = LLViewerMedia::getInstance()->getMediaImplFromTextureID(media_id); + auto impl = LLViewerMedia::getInstance()->getMediaImplFromTextureID(media_id); if (impl) { impl->showNotification(notification); diff --git a/indra/newview/llcontrolavatar.cpp b/indra/newview/llcontrolavatar.cpp index fab249f988..606e670805 100644 --- a/indra/newview/llcontrolavatar.cpp +++ b/indra/newview/llcontrolavatar.cpp @@ -241,7 +241,7 @@ void LLControlAvatar::matchVolumeTransform() if (skin_info) { LL_DEBUGS("BindShape") << getFullname() << " bind shape " << skin_info->mBindShapeMatrix << LL_ENDL; - bind_rot = LLSkinningUtil::getUnscaledQuaternion(skin_info->mBindShapeMatrix); + bind_rot = LLSkinningUtil::getUnscaledQuaternion(LLMatrix4(skin_info->mBindShapeMatrix)); } #endif setRotation(bind_rot*obj_rot); diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp index 507af56cb0..30c4a21e1c 100644 --- a/indra/newview/lldrawable.cpp +++ b/indra/newview/lldrawable.cpp @@ -101,6 +101,8 @@ LLDrawable::LLDrawable(LLViewerObject *vobj, bool new_entry) void LLDrawable::init(bool new_entry) { + LL_PROFILE_ZONE_SCOPED + // mXform mParent = NULL; mRenderType = 0; @@ -232,6 +234,8 @@ void LLDrawable::markDead() LLVOVolume* LLDrawable::getVOVolume() const { + LL_PROFILE_ZONE_SCOPED + LLViewerObject* objectp = mVObjp; if ( !isDead() && objectp && (objectp->getPCode() == LL_PCODE_VOLUME)) { @@ -335,6 +339,7 @@ static LLTrace::BlockTimerStatHandle FTM_ALLOCATE_FACE("Allocate Face"); LLFace* LLDrawable::addFace(LLFacePool *poolp, LLViewerTexture *texturep) { + LL_PROFILE_ZONE_SCOPED LLFace *face; { @@ -363,6 +368,8 @@ LLFace* LLDrawable::addFace(LLFacePool *poolp, LLViewerTexture *texturep) LLFace* LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep) { + LL_PROFILE_ZONE_SCOPED + LLFace *face; { @@ -387,6 +394,8 @@ LLFace* LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep) LLFace* LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep, LLViewerTexture *normalp) { + LL_PROFILE_ZONE_SCOPED + LLFace *face; face = new LLFace(this, mVObjp); @@ -408,6 +417,8 @@ LLFace* LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep, LLFace* LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep, LLViewerTexture *normalp, LLViewerTexture *specularp) { + LL_PROFILE_ZONE_SCOPED + LLFace *face; face = new LLFace(this, mVObjp); @@ -430,6 +441,8 @@ LLFace* LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep, void LLDrawable::setNumFaces(const S32 newFaces, LLFacePool *poolp, LLViewerTexture *texturep) { + LL_PROFILE_ZONE_SCOPED + if (newFaces == (S32)mFaces.size()) { return; @@ -453,6 +466,8 @@ void LLDrawable::setNumFaces(const S32 newFaces, LLFacePool *poolp, LLViewerText void LLDrawable::setNumFacesFast(const S32 newFaces, LLFacePool *poolp, LLViewerTexture *texturep) { + LL_PROFILE_ZONE_SCOPED + if (newFaces <= (S32)mFaces.size() && newFaces >= (S32)mFaces.size()/2) { return; @@ -476,6 +491,8 @@ void LLDrawable::setNumFacesFast(const S32 newFaces, LLFacePool *poolp, LLViewer void LLDrawable::mergeFaces(LLDrawable* src) { + LL_PROFILE_ZONE_SCOPED + U32 face_count = mFaces.size() + src->mFaces.size(); mFaces.reserve(face_count); @@ -509,6 +526,8 @@ void LLDrawable::updateMaterial() void LLDrawable::makeActive() { + LL_PROFILE_ZONE_SCOPED + #if !LL_RELEASE_FOR_DOWNLOAD if (mVObjp.notNull()) { @@ -572,6 +591,8 @@ void LLDrawable::makeActive() void LLDrawable::makeStatic(BOOL warning_enabled) { + LL_PROFILE_ZONE_SCOPED + if (isState(ACTIVE) && !isState(ACTIVE_CHILD) && !mVObjp->isAttachment() && @@ -618,6 +639,8 @@ void LLDrawable::makeStatic(BOOL warning_enabled) // Returns "distance" between target destination and resulting xfrom F32 LLDrawable::updateXform(BOOL undamped) { + LL_PROFILE_ZONE_SCOPED + BOOL damped = !undamped; // Position @@ -769,6 +792,8 @@ void LLDrawable::moveUpdatePipeline(BOOL moved) void LLDrawable::movePartition() { + LL_PROFILE_ZONE_SCOPED + LLSpatialPartition* part = getSpatialPartition(); if (part) { @@ -813,6 +838,8 @@ BOOL LLDrawable::updateMoveUndamped() void LLDrawable::updatePartition() { + LL_PROFILE_ZONE_SCOPED + if (!getVOVolume()) { movePartition(); @@ -830,6 +857,8 @@ void LLDrawable::updatePartition() BOOL LLDrawable::updateMoveDamped() { + LL_PROFILE_ZONE_SCOPED + F32 dist_squared = updateXform(FALSE); mGeneration++; @@ -853,6 +882,8 @@ BOOL LLDrawable::updateMoveDamped() void LLDrawable::updateDistance(LLCamera& camera, bool force_update) { + LL_PROFILE_ZONE_SCOPED + if (LLViewerCamera::sCurCameraID != LLViewerCamera::CAMERA_WORLD) { LL_WARNS() << "Attempted to update distance for non-world camera." << LL_ENDL; @@ -957,6 +988,8 @@ void LLDrawable::updateTexture() BOOL LLDrawable::updateGeometry(BOOL priority) { + LL_PROFILE_ZONE_SCOPED + llassert(mVObjp.notNull()); BOOL res = mVObjp->updateGeometry(this); return res; @@ -1034,6 +1067,8 @@ const LLVector3& LLDrawable::getBounds(LLVector3& min, LLVector3& max) const void LLDrawable::updateSpatialExtents() { + LL_PROFILE_ZONE_SCOPED + if (mVObjp) { const LLVector4a* exts = getSpatialExtents(); @@ -1164,6 +1199,8 @@ void LLDrawable::setGroup(LLViewerOctreeGroup *groupp) LLSpatialPartition* LLDrawable::getSpatialPartition() { + LL_PROFILE_ZONE_SCOPED + LLSpatialPartition* retval = NULL; if (!mVObjp || @@ -1247,6 +1284,8 @@ LLSpatialBridge::LLSpatialBridge(LLDrawable* root, BOOL render_by_group, U32 dat LLDrawable(root->getVObj(), true), LLSpatialPartition(data_mask, render_by_group, GL_STREAM_DRAW_ARB, regionp) { + LL_PROFILE_ZONE_SCOPED + mBridge = this; mDrawable = root; root->setSpatialBridge(this); @@ -1292,6 +1331,8 @@ void LLSpatialBridge::destroyTree() void LLSpatialBridge::updateSpatialExtents() { + LL_PROFILE_ZONE_SCOPED + LLSpatialGroup* root = (LLSpatialGroup*) mOctree->getListener(0); { @@ -1455,6 +1496,8 @@ public: void LLSpatialBridge::setVisible(LLCamera& camera_in, std::vector<LLDrawable*>* results, BOOL for_select) { + LL_PROFILE_ZONE_SCOPED + if (!gPipeline.hasRenderType(mDrawableType)) { return; @@ -1552,6 +1595,8 @@ void LLSpatialBridge::setVisible(LLCamera& camera_in, std::vector<LLDrawable*>* void LLSpatialBridge::updateDistance(LLCamera& camera_in, bool force_update) { + LL_PROFILE_ZONE_SCOPED + if (mDrawable == NULL) { markDead(); diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index 687b13d2c8..02ab316256 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -121,6 +121,8 @@ LLDrawPoolAvatar::~LLDrawPoolAvatar() // virtual BOOL LLDrawPoolAvatar::isDead() { + LL_PROFILE_ZONE_SCOPED + if (!LLFacePool::isDead()) { return FALSE; @@ -138,11 +140,15 @@ BOOL LLDrawPoolAvatar::isDead() S32 LLDrawPoolAvatar::getShaderLevel() const { + LL_PROFILE_ZONE_SCOPED + return (S32) LLViewerShaderMgr::instance()->getShaderLevel(LLViewerShaderMgr::SHADER_AVATAR); } void LLDrawPoolAvatar::prerender() { + LL_PROFILE_ZONE_SCOPED + mShaderLevel = LLViewerShaderMgr::instance()->getShaderLevel(LLViewerShaderMgr::SHADER_AVATAR); sShaderLevel = mShaderLevel; @@ -163,12 +169,15 @@ void LLDrawPoolAvatar::prerender() { LLVOAvatar* avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); updateRiggedVertexBuffers(avatarp); + updateSkinInfoMatrixPalettes(avatarp); } } } LLMatrix4& LLDrawPoolAvatar::getModelView() { + LL_PROFILE_ZONE_SCOPED + static LLMatrix4 ret; ret.initRows(LLVector4(gGLModelView+0), @@ -257,6 +266,8 @@ void LLDrawPoolAvatar::endDeferredPass(S32 pass) void LLDrawPoolAvatar::renderDeferred(S32 pass) { + LL_PROFILE_ZONE_SCOPED + render(pass); } @@ -267,6 +278,8 @@ S32 LLDrawPoolAvatar::getNumPostDeferredPasses() void LLDrawPoolAvatar::beginPostDeferredPass(S32 pass) { + LL_PROFILE_ZONE_SCOPED + switch (pass) { case 0: @@ -295,6 +308,8 @@ void LLDrawPoolAvatar::beginPostDeferredPass(S32 pass) void LLDrawPoolAvatar::beginPostDeferredAlpha() { + LL_PROFILE_ZONE_SCOPED + sSkipOpaque = TRUE; sShaderLevel = mShaderLevel; sVertexProgram = &gDeferredAvatarAlphaProgram; @@ -309,6 +324,8 @@ void LLDrawPoolAvatar::beginPostDeferredAlpha() void LLDrawPoolAvatar::beginDeferredRiggedAlpha() { + LL_PROFILE_ZONE_SCOPED + sVertexProgram = &gDeferredSkinnedAlphaProgram; gPipeline.bindDeferredShader(*sVertexProgram); sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); @@ -317,6 +334,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedAlpha() void LLDrawPoolAvatar::beginDeferredRiggedMaterialAlpha(S32 pass) { + LL_PROFILE_ZONE_SCOPED + switch (pass) { case 0: pass = 1; break; @@ -343,6 +362,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedMaterialAlpha(S32 pass) void LLDrawPoolAvatar::endDeferredRiggedAlpha() { + LL_PROFILE_ZONE_SCOPED + LLVertexBuffer::unbind(); gPipeline.unbindDeferredShader(*sVertexProgram); sDiffuseChannel = 0; @@ -353,6 +374,8 @@ void LLDrawPoolAvatar::endDeferredRiggedAlpha() void LLDrawPoolAvatar::endPostDeferredPass(S32 pass) { + LL_PROFILE_ZONE_SCOPED + switch (pass) { case 0: @@ -381,6 +404,8 @@ void LLDrawPoolAvatar::endPostDeferredPass(S32 pass) void LLDrawPoolAvatar::endPostDeferredAlpha() { + LL_PROFILE_ZONE_SCOPED + // if we're in software-blending, remember to set the fence _after_ we draw so we wait till this rendering is done sRenderingSkinned = FALSE; sSkipOpaque = FALSE; @@ -392,6 +417,8 @@ void LLDrawPoolAvatar::endPostDeferredAlpha() void LLDrawPoolAvatar::renderPostDeferred(S32 pass) { + LL_PROFILE_ZONE_SCOPED + static const S32 actual_pass[] = { //map post deferred pass numbers to what render() expects 2, //skinned @@ -428,225 +455,235 @@ S32 LLDrawPoolAvatar::getNumShadowPasses() void LLDrawPoolAvatar::beginShadowPass(S32 pass) { LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + { + LL_PROFILE_ZONE_SCOPED; - if (pass == SHADOW_PASS_AVATAR_OPAQUE) - { - sVertexProgram = &gDeferredAvatarShadowProgram; - - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + if (pass == SHADOW_PASS_AVATAR_OPAQUE) + { + sVertexProgram = &gDeferredAvatarShadowProgram; - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) - { - sVertexProgram = &gDeferredAvatarAlphaShadowProgram; + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + sVertexProgram = &gDeferredAvatarAlphaShadowProgram; - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) - { - sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram; + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram; - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) - { - sVertexProgram = &gDeferredAttachmentAlphaShadowProgram; + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } - - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + sVertexProgram = &gDeferredAttachmentAlphaShadowProgram; - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) - { - sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram; + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } + + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram; - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - gGL.diffuseColor4f(1,1,1,1); - } - else // SHADOW_PASS_ATTACHMENT_OPAQUE - { - sVertexProgram = &gDeferredAttachmentShadowProgram; - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } + + gGL.diffuseColor4f(1, 1, 1, 1); + } + else // SHADOW_PASS_ATTACHMENT_OPAQUE { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } - sVertexProgram->bind(); - } + sVertexProgram = &gDeferredAttachmentShadowProgram; + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } + sVertexProgram->bind(); + } + } } void LLDrawPoolAvatar::endShadowPass(S32 pass) { LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + { + LL_PROFILE_ZONE_SCOPED; - if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE) - { - LLVertexBuffer::unbind(); - } + if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE) + { + LLVertexBuffer::unbind(); + } - if (sShaderLevel > 0) - { - sVertexProgram->unbind(); - } - sVertexProgram = NULL; - sRenderingSkinned = FALSE; - LLDrawPoolAvatar::sShadowPass = -1; + if (sShaderLevel > 0) + { + sVertexProgram->unbind(); + } + sVertexProgram = NULL; + sRenderingSkinned = FALSE; + LLDrawPoolAvatar::sShadowPass = -1; + } } void LLDrawPoolAvatar::renderShadow(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + { + LL_PROFILE_ZONE_SCOPED; - if (mDrawFace.empty()) - { - return; - } + if (mDrawFace.empty()) + { + return; + } - const LLFace *facep = mDrawFace[0]; - if (!facep->getDrawable()) - { - return; - } - LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); + const LLFace *facep = mDrawFace[0]; + if (!facep->getDrawable()) + { + return; + } + LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); - if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull()) - { - return; - } - LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance(); - BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor(); - if (oa == LLVOAvatar::AOA_INVISIBLE || - (impostor && oa == LLVOAvatar::AOA_JELLYDOLL)) - { - // No shadows for jellydolled or invisible avs. - return; - } - - LLDrawPoolAvatar::sShadowPass = pass; + if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull()) + { + return; + } + LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance(); + BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor(); + if (impostor || (oa == LLVOAvatar::AOA_INVISIBLE)) + { + // No shadows for impostored (including jellydolled) or invisible avs. + return; + } - if (pass == SHADOW_PASS_AVATAR_OPAQUE) - { - LLDrawPoolAvatar::sSkipTransparent = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipTransparent = false; - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) - { - LLDrawPoolAvatar::sSkipOpaque = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) - { - LLDrawPoolAvatar::sSkipOpaque = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha - { - LLDrawPoolAvatar::sSkipOpaque = true; - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA); - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE); - renderRigged(avatarp, RIGGED_ALPHA); - renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA); - renderRigged(avatarp, RIGGED_GLOW); - renderRigged(avatarp, RIGGED_SPECMAP_BLEND); - renderRigged(avatarp, RIGGED_NORMMAP_BLEND); - renderRigged(avatarp, RIGGED_NORMSPEC_BLEND); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask - { - LLDrawPoolAvatar::sSkipOpaque = true; - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK); - renderRigged(avatarp, RIGGED_NORMMAP_MASK); - renderRigged(avatarp, RIGGED_SPECMAP_MASK); - renderRigged(avatarp, RIGGED_NORMSPEC_MASK); - renderRigged(avatarp, RIGGED_GLOW); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE - { - LLDrawPoolAvatar::sSkipTransparent = true; - renderRigged(avatarp, RIGGED_MATERIAL); - renderRigged(avatarp, RIGGED_SPECMAP); - renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE); - renderRigged(avatarp, RIGGED_NORMMAP); - renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE); - renderRigged(avatarp, RIGGED_NORMSPEC); - renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE); - renderRigged(avatarp, RIGGED_SIMPLE); - renderRigged(avatarp, RIGGED_FULLBRIGHT); - renderRigged(avatarp, RIGGED_SHINY); - renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY); - renderRigged(avatarp, RIGGED_GLOW); - renderRigged(avatarp, RIGGED_DEFERRED_BUMP); - renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE); - LLDrawPoolAvatar::sSkipTransparent = false; - } + LLDrawPoolAvatar::sShadowPass = pass; + + if (pass == SHADOW_PASS_AVATAR_OPAQUE) + { + LLDrawPoolAvatar::sSkipTransparent = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipTransparent = false; + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) + { + LLDrawPoolAvatar::sSkipOpaque = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) + { + LLDrawPoolAvatar::sSkipOpaque = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha + { + LLDrawPoolAvatar::sSkipOpaque = true; + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA); + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE); + renderRigged(avatarp, RIGGED_ALPHA); + renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA); + renderRigged(avatarp, RIGGED_GLOW); + renderRigged(avatarp, RIGGED_SPECMAP_BLEND); + renderRigged(avatarp, RIGGED_NORMMAP_BLEND); + renderRigged(avatarp, RIGGED_NORMSPEC_BLEND); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask + { + LLDrawPoolAvatar::sSkipOpaque = true; + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK); + renderRigged(avatarp, RIGGED_NORMMAP_MASK); + renderRigged(avatarp, RIGGED_SPECMAP_MASK); + renderRigged(avatarp, RIGGED_NORMSPEC_MASK); + renderRigged(avatarp, RIGGED_GLOW); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE + { + LLDrawPoolAvatar::sSkipTransparent = true; + renderRigged(avatarp, RIGGED_MATERIAL); + renderRigged(avatarp, RIGGED_SPECMAP); + renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE); + renderRigged(avatarp, RIGGED_NORMMAP); + renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE); + renderRigged(avatarp, RIGGED_NORMSPEC); + renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE); + renderRigged(avatarp, RIGGED_SIMPLE); + renderRigged(avatarp, RIGGED_FULLBRIGHT); + renderRigged(avatarp, RIGGED_SHINY); + renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY); + renderRigged(avatarp, RIGGED_GLOW); + renderRigged(avatarp, RIGGED_DEFERRED_BUMP); + renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE); + LLDrawPoolAvatar::sSkipTransparent = false; + } + } } S32 LLDrawPoolAvatar::getNumPasses() { + LL_PROFILE_ZONE_SCOPED + if (LLPipeline::sImpostorRender) { return 8; @@ -660,6 +697,8 @@ S32 LLDrawPoolAvatar::getNumPasses() S32 LLDrawPoolAvatar::getNumDeferredPasses() { + LL_PROFILE_ZONE_SCOPED + if (LLPipeline::sImpostorRender) { return 19; @@ -780,6 +819,8 @@ void LLDrawPoolAvatar::endRenderPass(S32 pass) void LLDrawPoolAvatar::beginImpostor() { + LL_PROFILE_ZONE_SCOPED + if (!LLPipeline::sReflectionRender) { LLVOAvatar::sRenderDistance = llclamp(LLVOAvatar::sRenderDistance, 16.f, 256.f); @@ -798,6 +839,8 @@ void LLDrawPoolAvatar::beginImpostor() void LLDrawPoolAvatar::endImpostor() { + LL_PROFILE_ZONE_SCOPED + if (LLGLSLShader::sNoFixedFunction) { gImpostorProgram.unbind(); @@ -807,6 +850,8 @@ void LLDrawPoolAvatar::endImpostor() void LLDrawPoolAvatar::beginRigid() { + LL_PROFILE_ZONE_SCOPED + if (gPipeline.canUseVertexShaders()) { if (LLPipeline::sUnderWaterRender) @@ -840,6 +885,8 @@ void LLDrawPoolAvatar::beginRigid() void LLDrawPoolAvatar::endRigid() { + LL_PROFILE_ZONE_SCOPED + sShaderLevel = mShaderLevel; if (sVertexProgram != NULL) { @@ -849,6 +896,8 @@ void LLDrawPoolAvatar::endRigid() void LLDrawPoolAvatar::beginDeferredImpostor() { + LL_PROFILE_ZONE_SCOPED + if (!LLPipeline::sReflectionRender) { LLVOAvatar::sRenderDistance = llclamp(LLVOAvatar::sRenderDistance, 16.f, 256.f); @@ -865,6 +914,8 @@ void LLDrawPoolAvatar::beginDeferredImpostor() void LLDrawPoolAvatar::endDeferredImpostor() { + LL_PROFILE_ZONE_SCOPED + sShaderLevel = mShaderLevel; sVertexProgram->disableTexture(LLViewerShaderMgr::DEFERRED_NORMAL); sVertexProgram->disableTexture(LLViewerShaderMgr::SPECULAR_MAP); @@ -876,6 +927,8 @@ void LLDrawPoolAvatar::endDeferredImpostor() void LLDrawPoolAvatar::beginDeferredRigid() { + LL_PROFILE_ZONE_SCOPED + sVertexProgram = &gDeferredNonIndexedDiffuseAlphaMaskNoColorProgram; sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); sVertexProgram->bind(); @@ -892,6 +945,8 @@ void LLDrawPoolAvatar::beginDeferredRigid() void LLDrawPoolAvatar::endDeferredRigid() { + LL_PROFILE_ZONE_SCOPED + sShaderLevel = mShaderLevel; sVertexProgram->disableTexture(LLViewerShaderMgr::DIFFUSE_MAP); sVertexProgram->unbind(); @@ -901,6 +956,8 @@ void LLDrawPoolAvatar::endDeferredRigid() void LLDrawPoolAvatar::beginSkinned() { + LL_PROFILE_ZONE_SCOPED + if (sShaderLevel > 0) { if (LLPipeline::sUnderWaterRender) @@ -967,6 +1024,8 @@ void LLDrawPoolAvatar::beginSkinned() void LLDrawPoolAvatar::endSkinned() { + LL_PROFILE_ZONE_SCOPED + // if we're in software-blending, remember to set the fence _after_ we draw so we wait till this rendering is done if (sShaderLevel > 0) { @@ -991,6 +1050,8 @@ void LLDrawPoolAvatar::endSkinned() void LLDrawPoolAvatar::beginRiggedSimple() { + LL_PROFILE_ZONE_SCOPED + if (sShaderLevel > 0) { if (LLPipeline::sUnderWaterRender) @@ -1031,6 +1092,8 @@ void LLDrawPoolAvatar::beginRiggedSimple() void LLDrawPoolAvatar::endRiggedSimple() { + LL_PROFILE_ZONE_SCOPED + LLVertexBuffer::unbind(); if (sShaderLevel > 0 || gPipeline.canUseVertexShaders()) { @@ -1041,27 +1104,37 @@ void LLDrawPoolAvatar::endRiggedSimple() void LLDrawPoolAvatar::beginRiggedAlpha() { + LL_PROFILE_ZONE_SCOPED + beginRiggedSimple(); } void LLDrawPoolAvatar::endRiggedAlpha() { + LL_PROFILE_ZONE_SCOPED + endRiggedSimple(); } void LLDrawPoolAvatar::beginRiggedFullbrightAlpha() { + LL_PROFILE_ZONE_SCOPED + beginRiggedFullbright(); } void LLDrawPoolAvatar::endRiggedFullbrightAlpha() { + LL_PROFILE_ZONE_SCOPED + endRiggedFullbright(); } void LLDrawPoolAvatar::beginRiggedGlow() { + LL_PROFILE_ZONE_SCOPED + if (sShaderLevel > 0) { if (LLPipeline::sUnderWaterRender) @@ -1108,11 +1181,15 @@ void LLDrawPoolAvatar::beginRiggedGlow() void LLDrawPoolAvatar::endRiggedGlow() { + LL_PROFILE_ZONE_SCOPED + endRiggedFullbright(); } void LLDrawPoolAvatar::beginRiggedFullbright() { + LL_PROFILE_ZONE_SCOPED + if (sShaderLevel > 0) { if (LLPipeline::sUnderWaterRender) @@ -1170,6 +1247,8 @@ void LLDrawPoolAvatar::beginRiggedFullbright() void LLDrawPoolAvatar::endRiggedFullbright() { + LL_PROFILE_ZONE_SCOPED + LLVertexBuffer::unbind(); if (sShaderLevel > 0 || gPipeline.canUseVertexShaders()) { @@ -1180,6 +1259,8 @@ void LLDrawPoolAvatar::endRiggedFullbright() void LLDrawPoolAvatar::beginRiggedShinySimple() { + LL_PROFILE_ZONE_SCOPED + if (sShaderLevel > 0) { if (LLPipeline::sUnderWaterRender) @@ -1220,6 +1301,8 @@ void LLDrawPoolAvatar::beginRiggedShinySimple() void LLDrawPoolAvatar::endRiggedShinySimple() { + LL_PROFILE_ZONE_SCOPED + LLVertexBuffer::unbind(); if (sShaderLevel > 0 || gPipeline.canUseVertexShaders()) { @@ -1231,6 +1314,8 @@ void LLDrawPoolAvatar::endRiggedShinySimple() void LLDrawPoolAvatar::beginRiggedFullbrightShiny() { + LL_PROFILE_ZONE_SCOPED + if (sShaderLevel > 0) { if (LLPipeline::sUnderWaterRender) @@ -1296,6 +1381,8 @@ void LLDrawPoolAvatar::beginRiggedFullbrightShiny() void LLDrawPoolAvatar::endRiggedFullbrightShiny() { + LL_PROFILE_ZONE_SCOPED + LLVertexBuffer::unbind(); if (sShaderLevel > 0 || gPipeline.canUseVertexShaders()) { @@ -1308,6 +1395,8 @@ void LLDrawPoolAvatar::endRiggedFullbrightShiny() void LLDrawPoolAvatar::beginDeferredRiggedSimple() { + LL_PROFILE_ZONE_SCOPED + sVertexProgram = &gDeferredSkinnedDiffuseProgram; sDiffuseChannel = 0; sVertexProgram->bind(); @@ -1323,6 +1412,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedSimple() void LLDrawPoolAvatar::endDeferredRiggedSimple() { + LL_PROFILE_ZONE_SCOPED + LLVertexBuffer::unbind(); sVertexProgram->unbind(); sVertexProgram = NULL; @@ -1330,6 +1421,8 @@ void LLDrawPoolAvatar::endDeferredRiggedSimple() void LLDrawPoolAvatar::beginDeferredRiggedBump() { + LL_PROFILE_ZONE_SCOPED + sVertexProgram = &gDeferredSkinnedBumpProgram; sVertexProgram->bind(); if (LLPipeline::sRenderingHUDs) @@ -1346,6 +1439,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedBump() void LLDrawPoolAvatar::endDeferredRiggedBump() { + LL_PROFILE_ZONE_SCOPED + LLVertexBuffer::unbind(); sVertexProgram->disableTexture(LLViewerShaderMgr::BUMP_MAP); sVertexProgram->disableTexture(LLViewerShaderMgr::DIFFUSE_MAP); @@ -1357,6 +1452,8 @@ void LLDrawPoolAvatar::endDeferredRiggedBump() void LLDrawPoolAvatar::beginDeferredRiggedMaterial(S32 pass) { + LL_PROFILE_ZONE_SCOPED + if (pass == 1 || pass == 5 || pass == 9 || @@ -1387,6 +1484,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedMaterial(S32 pass) void LLDrawPoolAvatar::endDeferredRiggedMaterial(S32 pass) { + LL_PROFILE_ZONE_SCOPED + if (pass == 1 || pass == 5 || pass == 9 || @@ -1407,6 +1506,8 @@ void LLDrawPoolAvatar::endDeferredRiggedMaterial(S32 pass) void LLDrawPoolAvatar::beginDeferredSkinned() { + LL_PROFILE_ZONE_SCOPED + sShaderLevel = mShaderLevel; sVertexProgram = &gDeferredAvatarProgram; sRenderingSkinned = TRUE; @@ -1428,6 +1529,8 @@ void LLDrawPoolAvatar::beginDeferredSkinned() void LLDrawPoolAvatar::endDeferredSkinned() { + LL_PROFILE_ZONE_SCOPED + // if we're in software-blending, remember to set the fence _after_ we draw so we wait till this rendering is done sRenderingSkinned = FALSE; sVertexProgram->unbind(); @@ -1740,6 +1843,8 @@ void LLDrawPoolAvatar::getRiggedGeometry( LLVolume* volume, const LLVolumeFace& vol_face) { + LL_PROFILE_ZONE_SCOPED + face->setGeomIndex(0); face->setIndicesIndex(0); @@ -1794,7 +1899,7 @@ void LLDrawPoolAvatar::getRiggedGeometry( U16 offset = 0; - LLMatrix4 mat_vert = skin->mBindShapeMatrix; + LLMatrix4 mat_vert = LLMatrix4(skin->mBindShapeMatrix); glh::matrix4f m((F32*) mat_vert.mMatrix); m = m.inverse().transpose(); @@ -1836,24 +1941,26 @@ void LLDrawPoolAvatar::getRiggedGeometry( void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( LLVOAvatar* avatar, LLFace* face, - const LLMeshSkinInfo* skin, + const LLVOVolume* vobj, LLVolume* volume, LLVolumeFace& vol_face) { + LL_PROFILE_ZONE_SCOPED; + LLVector4a* weights = vol_face.mWeights; if (!weights) { return; } + if (!vobj || vobj->isNoLOD()) + { + return; + } + LLPointer<LLVertexBuffer> buffer = face->getVertexBuffer(); LLDrawable* drawable = face->getDrawable(); - if (drawable->getVOVolume() && drawable->getVOVolume()->isNoLOD()) - { - return; - } - const U32 max_joints = LLSkinningUtil::getMaxJointCount(); #if USE_SEPARATE_JOINT_INDICES_AND_WEIGHTS @@ -1893,23 +2000,26 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( } #endif - // FIXME ugly const cast - LLSkinningUtil::scrubInvalidJoints(avatar, const_cast<LLMeshSkinInfo*>(skin)); - - U32 data_mask = face->getRiggedVertexBufferDataMask(); + U32 data_mask = face->getRiggedVertexBufferDataMask(); + const LLMeshSkinInfo* skin = nullptr; - if (!vol_face.mWeightsScrubbed) - { - LLSkinningUtil::scrubSkinWeights(weights, vol_face.mNumVertices, skin); - vol_face.mWeightsScrubbed = TRUE; - } - if (buffer.isNull() || buffer->getTypeMask() != data_mask || buffer->getNumVerts() != vol_face.mNumVertices || buffer->getNumIndices() != vol_face.mNumIndices || (drawable && drawable->isState(LLDrawable::REBUILD_ALL))) { + LL_PROFILE_ZONE_NAMED("Rigged VBO Rebuild"); + skin = vobj->getSkinInfo(); + // FIXME ugly const cast + LLSkinningUtil::scrubInvalidJoints(avatar, const_cast<LLMeshSkinInfo*>(skin)); + + if (!vol_face.mWeightsScrubbed) + { + LLSkinningUtil::scrubSkinWeights(weights, vol_face.mNumVertices, skin); + vol_face.mWeightsScrubbed = TRUE; + } + if (drawable && drawable->isState(LLDrawable::REBUILD_ALL)) { //rebuild EVERY face in the drawable, not just this one, to avoid missing drawable wide rebuild issues @@ -1935,18 +2045,13 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( } } - if (buffer.isNull() || - buffer->getNumVerts() != vol_face.mNumVertices || - buffer->getNumIndices() != vol_face.mNumIndices) - { - // Allocation failed - return; - } - - if (!buffer.isNull() && - sShaderLevel <= 0 && - face->mLastSkinTime < avatar->getLastSkinTime()) + if (sShaderLevel <= 0 && + face->mLastSkinTime < avatar->getLastSkinTime() && + !buffer.isNull() && + buffer->getNumVerts() == vol_face.mNumVertices && + buffer->getNumIndices() == vol_face.mNumIndices) { + LL_PROFILE_ZONE_NAMED("Software Skinning"); //perform software vertex skinning for this face LLStrider<LLVector3> position; LLStrider<LLVector3> normal; @@ -1963,14 +2068,16 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( LLVector4a* norm = has_normal ? (LLVector4a*) normal.get() : NULL; - //build matrix palette - LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT]; - U32 count = LLSkinningUtil::getMeshJointCount(skin); - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, count, skin, avatar); - LLSkinningUtil::checkSkinWeights(weights, buffer->getNumVerts(), skin); + if (skin == nullptr) + { + skin = vobj->getSkinInfo(); + } - LLMatrix4a bind_shape_matrix; - bind_shape_matrix.loadu(skin->mBindShapeMatrix); + const MatrixPaletteCache& mpc = updateSkinInfoMatrixPalette(avatar, skin); + const LLMatrix4a* mat = &(mpc.mMatrixPalette[0]); + + LLSkinningUtil::checkSkinWeights(weights, buffer->getNumVerts(), skin); + const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix; #if USE_SEPARATE_JOINT_INDICES_AND_WEIGHTS U8* joint_indices_cursor = vol_face.mJointIndices; @@ -2038,6 +2145,8 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) { + LL_PROFILE_ZONE_SCOPED + if (!avatar->shouldRenderRigged()) { return; @@ -2045,15 +2154,18 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) stop_glerror(); + const LLMeshSkinInfo* lastSkin = nullptr; + for (U32 i = 0; i < mRiggedFace[type].size(); ++i) { + LL_PROFILE_ZONE_NAMED("Render Rigged Face"); LLFace* face = mRiggedFace[type][i]; S32 offset = face->getIndicesStart(); U32 count = face->getIndicesCount(); U16 start = face->getGeomStart(); - U16 end = start + face->getGeomCount()-1; + U16 end = start + face->getGeomCount()-1; LLDrawable* drawable = face->getDrawable(); if (!drawable) @@ -2175,52 +2287,32 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) } if (buff) - { + { if (sShaderLevel > 0) { - // upload matrix palette to shader - LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT]; - U32 count = LLSkinningUtil::getMeshJointCount(skin); - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, count, skin, avatar); - - stop_glerror(); - - F32 mp[LL_MAX_JOINTS_PER_MESH_OBJECT*12]; - - for (U32 i = 0; i < count; ++i) - { - F32* m = (F32*) mat[i].mMatrix[0].getF32ptr(); - - U32 idx = i*12; - - mp[idx+0] = m[0]; - mp[idx+1] = m[1]; - mp[idx+2] = m[2]; - mp[idx+3] = m[12]; + if (lastSkin != skin) // <== only upload matrix palette to GL if the skininfo changed + { + // upload matrix palette to shader + const MatrixPaletteCache& mpc = updateSkinInfoMatrixPalette(avatar, skin); + U32 count = mpc.mMatrixPalette.size(); - mp[idx+4] = m[4]; - mp[idx+5] = m[5]; - mp[idx+6] = m[6]; - mp[idx+7] = m[13]; + stop_glerror(); - mp[idx+8] = m[8]; - mp[idx+9] = m[9]; - mp[idx+10] = m[10]; - mp[idx+11] = m[14]; - } + LLDrawPoolAvatar::sVertexProgram->uniformMatrix3x4fv(LLViewerShaderMgr::AVATAR_MATRIX, + count, + FALSE, + (GLfloat*) &(mpc.mGLMp[0])); - LLDrawPoolAvatar::sVertexProgram->uniformMatrix3x4fv(LLViewerShaderMgr::AVATAR_MATRIX, - count, - FALSE, - (GLfloat*) mp); - - stop_glerror(); + stop_glerror(); + } } else { data_mask &= ~LLVertexBuffer::MAP_WEIGHT4; } + lastSkin = skin; + /*if (glow) { gGL.diffuseColor4f(0,0,0,face->getTextureEntry()->getGlow()); @@ -2330,16 +2422,22 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) void LLDrawPoolAvatar::renderDeferredRiggedSimple(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + renderRigged(avatar, RIGGED_DEFERRED_SIMPLE); } void LLDrawPoolAvatar::renderDeferredRiggedBump(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + renderRigged(avatar, RIGGED_DEFERRED_BUMP); } void LLDrawPoolAvatar::renderDeferredRiggedMaterial(LLVOAvatar* avatar, S32 pass) { + LL_PROFILE_ZONE_SCOPED + renderRigged(avatar, pass); } @@ -2352,8 +2450,10 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar) //update rigged vertex buffers for (U32 type = 0; type < NUM_RIGGED_PASSES; ++type) { + LL_PROFILE_ZONE_NAMED("Pass"); for (U32 i = 0; i < mRiggedFace[type].size(); ++i) { + LL_PROFILE_ZONE_NAMED("Face"); LLFace* face = mRiggedFace[type][i]; LLDrawable* drawable = face->getDrawable(); if (!drawable) @@ -2376,43 +2476,111 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar) continue; } - const LLMeshSkinInfo* skin = vobj->getSkinInfo(); - if (!skin) - { - continue; - } - stop_glerror(); LLVolumeFace& vol_face = volume->getVolumeFace(te); - updateRiggedFaceVertexBuffer(avatar, face, skin, volume, vol_face); + updateRiggedFaceVertexBuffer(avatar, face, vobj, volume, vol_face); } } } +void LLDrawPoolAvatar::updateSkinInfoMatrixPalettes(LLVOAvatar* avatarp) +{ + LL_PROFILE_ZONE_SCOPED; + //evict matrix palettes from the cache that haven't been updated in 10 frames + for (matrix_palette_cache_t::iterator iter = mMatrixPaletteCache.begin(); iter != mMatrixPaletteCache.end(); ) + { + if (gFrameCount - iter->second.mFrame > 10) + { + iter = mMatrixPaletteCache.erase(iter); + } + else + { + ++iter; + } + } +} + +const LLDrawPoolAvatar::MatrixPaletteCache& LLDrawPoolAvatar::updateSkinInfoMatrixPalette(LLVOAvatar * avatarp, const LLMeshSkinInfo* skin) +{ + MatrixPaletteCache& entry = mMatrixPaletteCache[skin]; + + if (entry.mFrame != gFrameCount) + { + LL_PROFILE_ZONE_SCOPED; + entry.mFrame = gFrameCount; + //build matrix palette + U32 count = LLSkinningUtil::getMeshJointCount(skin); + entry.mMatrixPalette.resize(count); + LLSkinningUtil::initSkinningMatrixPalette(&(entry.mMatrixPalette[0]), count, skin, avatarp); + + const LLMatrix4a* mat = &(entry.mMatrixPalette[0]); + + stop_glerror(); + + entry.mGLMp.resize(count * 12); + + F32* mp = &(entry.mGLMp[0]); + + for (U32 i = 0; i < count; ++i) + { + F32* m = (F32*)mat[i].mMatrix[0].getF32ptr(); + + U32 idx = i * 12; + + mp[idx + 0] = m[0]; + mp[idx + 1] = m[1]; + mp[idx + 2] = m[2]; + mp[idx + 3] = m[12]; + + mp[idx + 4] = m[4]; + mp[idx + 5] = m[5]; + mp[idx + 6] = m[6]; + mp[idx + 7] = m[13]; + + mp[idx + 8] = m[8]; + mp[idx + 9] = m[9]; + mp[idx + 10] = m[10]; + mp[idx + 11] = m[14]; + } + } + + return entry; +} + void LLDrawPoolAvatar::renderRiggedSimple(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + renderRigged(avatar, RIGGED_SIMPLE); } void LLDrawPoolAvatar::renderRiggedFullbright(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + renderRigged(avatar, RIGGED_FULLBRIGHT); } void LLDrawPoolAvatar::renderRiggedShinySimple(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + renderRigged(avatar, RIGGED_SHINY); } void LLDrawPoolAvatar::renderRiggedFullbrightShiny(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + renderRigged(avatar, RIGGED_FULLBRIGHT_SHINY); } void LLDrawPoolAvatar::renderRiggedAlpha(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + if (!mRiggedFace[RIGGED_ALPHA].empty()) { LLGLEnable blend(GL_BLEND); @@ -2430,6 +2598,8 @@ void LLDrawPoolAvatar::renderRiggedAlpha(LLVOAvatar* avatar) void LLDrawPoolAvatar::renderRiggedFullbrightAlpha(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + if (!mRiggedFace[RIGGED_FULLBRIGHT_ALPHA].empty()) { LLGLEnable blend(GL_BLEND); @@ -2447,6 +2617,8 @@ void LLDrawPoolAvatar::renderRiggedFullbrightAlpha(LLVOAvatar* avatar) void LLDrawPoolAvatar::renderRiggedGlow(LLVOAvatar* avatar) { + LL_PROFILE_ZONE_SCOPED + if (!mRiggedFace[RIGGED_GLOW].empty()) { LLGLEnable blend(GL_BLEND); @@ -2474,6 +2646,8 @@ void LLDrawPoolAvatar::renderRiggedGlow(LLVOAvatar* avatar) //----------------------------------------------------------------------------- LLViewerTexture *LLDrawPoolAvatar::getDebugTexture() { + LL_PROFILE_ZONE_SCOPED + if (mReferences.empty()) { return NULL; @@ -2497,6 +2671,8 @@ LLColor3 LLDrawPoolAvatar::getDebugColor() const void LLDrawPoolAvatar::addRiggedFace(LLFace* facep, U32 type) { + LL_PROFILE_ZONE_SCOPED + llassert (facep->isState(LLFace::RIGGED)); llassert(getType() == LLDrawPool::POOL_AVATAR || getType() == LLDrawPool::POOL_CONTROL_AV); if (facep->getPool() && facep->getPool() != this) @@ -2519,6 +2695,8 @@ void LLDrawPoolAvatar::addRiggedFace(LLFace* facep, U32 type) void LLDrawPoolAvatar::removeRiggedFace(LLFace* facep) { + LL_PROFILE_ZONE_SCOPED + llassert (facep->isState(LLFace::RIGGED)); llassert(getType() == LLDrawPool::POOL_AVATAR || getType() == LLDrawPool::POOL_CONTROL_AV); if (facep->getPool() != this) @@ -2556,7 +2734,7 @@ LLVertexBufferAvatar::LLVertexBufferAvatar() : LLVertexBuffer(sDataMask, GL_STREAM_DRAW_ARB) //avatars are always stream draw due to morph targets { - + LL_PROFILE_ZONE_SCOPED } diff --git a/indra/newview/lldrawpoolavatar.h b/indra/newview/lldrawpoolavatar.h index 92a8538958..0c1ee2cced 100644 --- a/indra/newview/lldrawpoolavatar.h +++ b/indra/newview/lldrawpoolavatar.h @@ -28,14 +28,18 @@ #define LL_LLDRAWPOOLAVATAR_H #include "lldrawpool.h" +#include "llmodel.h" + +#include <unordered_map> class LLVOAvatar; +class LLVOVolume; class LLGLSLShader; class LLFace; -class LLMeshSkinInfo; class LLVolume; class LLVolumeFace; +extern U32 gFrameCount; class LLDrawPoolAvatar : public LLFacePool { @@ -253,11 +257,13 @@ typedef enum void getRiggedGeometry(LLFace* face, LLPointer<LLVertexBuffer>& buffer, U32 data_mask, const LLMeshSkinInfo* skin, LLVolume* volume, const LLVolumeFace& vol_face); void updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* facep, - const LLMeshSkinInfo* skin, + const LLVOVolume* vobj, LLVolume* volume, LLVolumeFace& vol_face); void updateRiggedVertexBuffers(LLVOAvatar* avatar); + void updateSkinInfoMatrixPalettes(LLVOAvatar* avatarp); + void renderRigged(LLVOAvatar* avatar, U32 type, bool glow = false); void renderRiggedSimple(LLVOAvatar* avatar); void renderRiggedAlpha(LLVOAvatar* avatar); @@ -277,6 +283,26 @@ typedef enum std::vector<LLFace*> mRiggedFace[NUM_RIGGED_PASSES]; + class MatrixPaletteCache + { + public: + U32 mFrame; + LLMeshSkinInfo::matrix_list_t mMatrixPalette; + + // Float array ready to be sent to GL + std::vector<F32> mGLMp; + + MatrixPaletteCache() : + mFrame(gFrameCount-1) + { + } + }; + + const MatrixPaletteCache& updateSkinInfoMatrixPalette(LLVOAvatar* avatarp, const LLMeshSkinInfo* skin); + + typedef std::unordered_map<const LLMeshSkinInfo*, MatrixPaletteCache> matrix_palette_cache_t; + matrix_palette_cache_t mMatrixPaletteCache; + /*virtual*/ LLViewerTexture *getDebugTexture(); /*virtual*/ LLColor3 getDebugColor() const; // For AGP debug display diff --git a/indra/newview/lldrawpoolsimple.cpp b/indra/newview/lldrawpoolsimple.cpp index f211cf6e27..74e6665a96 100644 --- a/indra/newview/lldrawpoolsimple.cpp +++ b/indra/newview/lldrawpoolsimple.cpp @@ -199,11 +199,7 @@ void LLDrawPoolSimple::beginRenderPass(S32 pass) } else { - // don't use shaders! - if (gGLManager.mHasShaderObjects) - { - LLGLSLShader::bindNoShader(); - } + LLGLSLShader::bindNoShader(); } } @@ -301,11 +297,7 @@ void LLDrawPoolAlphaMask::beginRenderPass(S32 pass) } else { - // don't use shaders! - if (gGLManager.mHasShaderObjects) - { - LLGLSLShader::bindNoShader(); - } + LLGLSLShader::bindNoShader(); } } @@ -392,11 +384,7 @@ void LLDrawPoolFullbrightAlphaMask::beginRenderPass(S32 pass) } else { - // don't use shaders! - if (gGLManager.mHasShaderObjects) - { - LLGLSLShader::bindNoShader(); - } + LLGLSLShader::bindNoShader(); } } @@ -567,11 +555,7 @@ void LLDrawPoolGrass::beginRenderPass(S32 pass) else { gGL.setAlphaRejectSettings(LLRender::CF_GREATER, 0.5f); - // don't use shaders! - if (gGLManager.mHasShaderObjects) - { - LLGLSLShader::bindNoShader(); - } + LLGLSLShader::bindNoShader(); } } diff --git a/indra/newview/lldrawpoolsky.cpp b/indra/newview/lldrawpoolsky.cpp index b6f55e800a..b1eefaab81 100644 --- a/indra/newview/lldrawpoolsky.cpp +++ b/indra/newview/lldrawpoolsky.cpp @@ -82,13 +82,7 @@ void LLDrawPoolSky::render(S32 pass) } else { - // don't use shaders! - if (gGLManager.mHasShaderObjects) - { - // Ironically, we must support shader objects to be - // able to use this call. - LLGLSLShader::bindNoShader(); - } + LLGLSLShader::bindNoShader(); mShader = NULL; } diff --git a/indra/newview/llexperiencelog.cpp b/indra/newview/llexperiencelog.cpp index ee5d561927..c441fbc09f 100644 --- a/indra/newview/llexperiencelog.cpp +++ b/indra/newview/llexperiencelog.cpp @@ -149,10 +149,6 @@ std::string LLExperienceLog::getPermissionString( const LLSD& message, const std { buf.str(entry); } - else - { - buf.str(); - } } if(buf.str().empty()) diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 4a802ad9aa..34448a780d 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -241,6 +241,8 @@ void LLFace::setPool(LLFacePool* pool) void LLFace::setPool(LLFacePool* new_pool, LLViewerTexture *texturep) { + LL_PROFILE_ZONE_SCOPED + if (!new_pool) { LL_ERRS() << "Setting pool to null!" << LL_ENDL; @@ -320,6 +322,8 @@ void LLFace::setSpecularMap(LLViewerTexture* tex) void LLFace::dirtyTexture() { + LL_PROFILE_ZONE_SCOPED + LLDrawable* drawablep = getDrawable(); if (mVObjp.notNull() && mVObjp->getVolume()) @@ -535,6 +539,8 @@ void LLFace::updateCenterAgent() void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) { + LL_PROFILE_ZONE_SCOPED + if (mDrawablep == NULL || mDrawablep->getSpatialGroup() == NULL) { return; @@ -585,6 +591,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) glTexCoordPointer(2, GL_FLOAT, 8, vol_face.mTexCoords); } gGL.syncMatrices(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x00FF00 ); glDrawElements(GL_TRIANGLES, vol_face.mNumIndices, GL_UNSIGNED_SHORT, vol_face.mIndices); glDisableClientState(GL_TEXTURE_COORD_ARRAY); } @@ -605,6 +612,8 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) void renderFace(LLDrawable* drawable, LLFace *face) { + LL_PROFILE_ZONE_SCOPED + LLVOVolume* vobj = drawable->getVOVolume(); if (vobj) { @@ -891,6 +900,8 @@ bool less_than_max_mag(const LLVector4a& vec) BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f, const LLMatrix4& mat_vert_in, BOOL global_volume) { + LL_PROFILE_ZONE_SCOPED + //get bounding box if (mDrawablep->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION | LLDrawable::REBUILD_RIGGED)) { @@ -2375,6 +2386,8 @@ F32 LLFace::getTextureVirtualSize() BOOL LLFace::calcPixelArea(F32& cos_angle_to_view_dir, F32& radius) { + LL_PROFILE_ZONE_SCOPED + //VECTORIZE THIS //get area of circle around face LLVector4a center; @@ -2654,6 +2667,8 @@ const LLMatrix4& LLFace::getRenderMatrix() const S32 LLFace::renderElements(const U16 *index_array) const { + LL_PROFILE_ZONE_SCOPED + S32 ret = 0; if (isState(GLOBAL)) @@ -2673,6 +2688,8 @@ S32 LLFace::renderElements(const U16 *index_array) const S32 LLFace::renderIndexed() { + LL_PROFILE_ZONE_SCOPED + if(mDrawablep == NULL || mDrawPoolp == NULL) { return 0; @@ -2683,6 +2700,8 @@ S32 LLFace::renderIndexed() S32 LLFace::renderIndexed(U32 mask) { + LL_PROFILE_ZONE_SCOPED + if (mVertexBuffer.isNull()) { return 0; diff --git a/indra/newview/llfeaturemanager.cpp b/indra/newview/llfeaturemanager.cpp index e6bbe234b3..98c8531cd6 100644 --- a/indra/newview/llfeaturemanager.cpp +++ b/indra/newview/llfeaturemanager.cpp @@ -670,11 +670,7 @@ void LLFeatureManager::applyBaseMasks() } // now all those wacky ones - if (!gGLManager.mHasFragmentShader) - { - maskFeatures("NoPixelShaders"); - } - if (!gGLManager.mHasVertexShader || !mGPUSupported) + if (!mGPUSupported) { maskFeatures("NoVertexShaders"); } diff --git a/indra/newview/llfloatermodelpreview.cpp b/indra/newview/llfloatermodelpreview.cpp index d9edd4dc30..0e54b66ea9 100644 --- a/indra/newview/llfloatermodelpreview.cpp +++ b/indra/newview/llfloatermodelpreview.cpp @@ -1357,31 +1357,31 @@ void LLFloaterModelPreview::clearAvatarTab() } void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides) - { +{ S32 display_lod = mModelPreview->mPreviewLOD; if (mModelPreview->mModel[display_lod].empty()) - { + { mSelectedJointName.clear(); return; - } + } // Joints will be listed as long as they are listed in mAlternateBindMatrix // even if they are for some reason identical to defaults. // Todo: Are overrides always identical for all lods? They normally are, but there might be situations where they aren't. if (mJointOverrides[display_lod].empty()) - { + { // populate map for (LLModelLoader::scene::iterator iter = mModelPreview->mScene[display_lod].begin(); iter != mModelPreview->mScene[display_lod].end(); ++iter) - { + { for (LLModelLoader::model_instance_list::iterator model_iter = iter->second.begin(); model_iter != iter->second.end(); ++model_iter) - { + { LLModelInstance& instance = *model_iter; LLModel* model = instance.mModel; const LLMeshSkinInfo *skin = &model->mSkinInfo; U32 joint_count = LLSkinningUtil::getMeshJointCount(skin); U32 bind_count = highlight_overrides ? skin->mAlternateBindMatrix.size() : 0; // simply do not include overrides if data is not needed if (bind_count > 0 && bind_count != joint_count) - { + { std::ostringstream out; out << "Invalid joint overrides for model " << model->getName(); out << ". Amount of joints " << joint_count; @@ -1390,68 +1390,68 @@ void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides) addStringToLog(out.str(), true); // Disable overrides for this model bind_count = 0; - } + } if (bind_count > 0) - { + { for (U32 j = 0; j < joint_count; ++j) - { - const LLVector3& joint_pos = skin->mAlternateBindMatrix[j].getTranslation(); + { + const LLVector3& joint_pos = LLVector3(skin->mAlternateBindMatrix[j].getTranslation()); LLJointOverrideData &data = mJointOverrides[display_lod][skin->mJointNames[j]]; LLJoint* pJoint = LLModelPreview::lookupJointByName(skin->mJointNames[j], mModelPreview); if (pJoint) - { + { // see how voavatar uses aboveJointPosThreshold if (pJoint->aboveJointPosThreshold(joint_pos)) - { + { // valid override if (data.mPosOverrides.size() > 0 && (data.mPosOverrides.begin()->second - joint_pos).lengthSquared() > (LL_JOINT_TRESHOLD_POS_OFFSET * LL_JOINT_TRESHOLD_POS_OFFSET)) - { + { // File contains multiple meshes with conflicting joint offsets // preview may be incorrect, upload result might wary (depends onto // mesh_id that hasn't been generated yet). data.mHasConflicts = true; - } + } data.mPosOverrides[model->getName()] = joint_pos; - } - else - { + } + else + { // default value, it won't be accounted for by avatar data.mModelsNoOverrides.insert(model->getName()); - } - } - } - } - else - { + } + } + } + } + else + { for (U32 j = 0; j < joint_count; ++j) - { + { LLJointOverrideData &data = mJointOverrides[display_lod][skin->mJointNames[j]]; data.mModelsNoOverrides.insert(model->getName()); } } - } - } - } + } + } + } LLPanel *panel = mTabContainer->getPanelByName("rigging_panel"); LLScrollListCtrl *joints_list = panel->getChild<LLScrollListCtrl>("joints_list"); if (joints_list->isEmpty()) - { + { // Populate table - std::map<std::string, std::string> joint_alias_map; + std::map<std::string, std::string> joint_alias_map; mModelPreview->getJointAliases(joint_alias_map); - + S32 conflicts = 0; joint_override_data_map_t::iterator joint_iter = mJointOverrides[display_lod].begin(); joint_override_data_map_t::iterator joint_end = mJointOverrides[display_lod].end(); while (joint_iter != joint_end) - { + { const std::string& listName = joint_iter->first; - + LLScrollListItem::Params item_params; item_params.value(listName); @@ -1459,38 +1459,38 @@ void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides) cell_params.font = LLFontGL::getFontSansSerif(); cell_params.value = listName; if (joint_alias_map.find(listName) == joint_alias_map.end()) - { + { // Missing names cell_params.color = LLColor4::red; - } + } if (joint_iter->second.mHasConflicts) - { + { // Conflicts cell_params.color = LLColor4::orange; conflicts++; - } + } if (highlight_overrides && joint_iter->second.mPosOverrides.size() > 0) - { + { cell_params.font.style = "BOLD"; - } + } item_params.columns.add(cell_params); joints_list->addRow(item_params, ADD_BOTTOM); joint_iter++; - } + } joints_list->selectFirstItem(); LLScrollListItem *selected = joints_list->getFirstSelected(); if (selected) -{ + { mSelectedJointName = selected->getValue().asString(); - } + } LLTextBox *joint_conf_descr = panel->getChild<LLTextBox>("conflicts_description"); joint_conf_descr->setTextArg("[CONFLICTS]", llformat("%d", conflicts)); joint_conf_descr->setTextArg("[JOINTS_COUNT]", llformat("%d", mJointOverrides[display_lod].size())); - } - } + } +} //----------------------------------------------------------------------------- // addStringToLogTab() diff --git a/indra/newview/llfloaterwebcontent.cpp b/indra/newview/llfloaterwebcontent.cpp index 23fd6d9c8e..ceab472c55 100644 --- a/indra/newview/llfloaterwebcontent.cpp +++ b/indra/newview/llfloaterwebcontent.cpp @@ -159,7 +159,7 @@ LLFloater* LLFloaterWebContent::create( Params p) //static void LLFloaterWebContent::closeRequest(const std::string &uuid) { - LLFloaterWebContent* floaterp = instance_tracker_t::getInstance(uuid); + auto floaterp = instance_tracker_t::getInstance(uuid); if (floaterp) { floaterp->closeFloater(false); @@ -169,7 +169,7 @@ void LLFloaterWebContent::closeRequest(const std::string &uuid) //static void LLFloaterWebContent::geometryChanged(const std::string &uuid, S32 x, S32 y, S32 width, S32 height) { - LLFloaterWebContent* floaterp = instance_tracker_t::getInstance(uuid); + auto floaterp = instance_tracker_t::getInstance(uuid); if (floaterp) { floaterp->geometryChanged(x, y, width, height); diff --git a/indra/newview/llglsandbox.cpp b/indra/newview/llglsandbox.cpp index 698c15bd2d..0f288e05ca 100644 --- a/indra/newview/llglsandbox.cpp +++ b/indra/newview/llglsandbox.cpp @@ -993,9 +993,8 @@ private: //----------------------------------------------------------------------------- F32 gpu_benchmark() { - if (!gGLManager.mHasShaderObjects || !gGLManager.mHasTimerQuery) - { // don't bother benchmarking the fixed function - // or venerable drivers which don't support accurate timing anyway + if (!gGLManager.mHasTimerQuery) + { // don't bother benchmarking venerable drivers which don't support accurate timing anyway // and are likely to be correctly identified by the GPU table already. return -1.f; } diff --git a/indra/newview/llhudnametag.cpp b/indra/newview/llhudnametag.cpp index 9d49c30a49..55a4b5a457 100644 --- a/indra/newview/llhudnametag.cpp +++ b/indra/newview/llhudnametag.cpp @@ -224,6 +224,7 @@ BOOL LLHUDNameTag::lineSegmentIntersect(const LLVector4a& start, const LLVector4 void LLHUDNameTag::render() { + LL_PROFILE_ZONE_SCOPED; if (sDisplayText) { LLGLDepthTest gls_depth(GL_TRUE, GL_FALSE); @@ -731,6 +732,7 @@ void LLHUDNameTag::updateSize() void LLHUDNameTag::updateAll() { + LL_PROFILE_ZONE_SCOPED; // iterate over all text objects, calculate their restoration forces, // and add them to the visible set if they are on screen and close enough sVisibleTextObjects.clear(); diff --git a/indra/newview/llmainlooprepeater.cpp b/indra/newview/llmainlooprepeater.cpp deleted file mode 100644 index 6736e9a950..0000000000 --- a/indra/newview/llmainlooprepeater.cpp +++ /dev/null @@ -1,88 +0,0 @@ -/** - * @file llmachineid.cpp - * @brief retrieves unique machine ids - * - * $LicenseInfo:firstyear=2009&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#include "llviewerprecompiledheaders.h" -#include "llapr.h" -#include "llevents.h" -#include "llmainlooprepeater.h" - - - -// LLMainLoopRepeater -//----------------------------------------------------------------------------- - - -LLMainLoopRepeater::LLMainLoopRepeater(void): - mQueue(0) -{ - ; // No op. -} - - -void LLMainLoopRepeater::start(void) -{ - if(mQueue != 0) return; - - mQueue = new LLThreadSafeQueue<LLSD>(1024); - mMainLoopConnection = LLEventPumps::instance(). - obtain("mainloop").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMainLoop, this, _1)); - mRepeaterConnection = LLEventPumps::instance(). - obtain("mainlooprepeater").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMessage, this, _1)); -} - - -void LLMainLoopRepeater::stop(void) -{ - mMainLoopConnection.release(); - mRepeaterConnection.release(); - - delete mQueue; - mQueue = 0; -} - - -bool LLMainLoopRepeater::onMainLoop(LLSD const &) -{ - LLSD message; - while(mQueue->tryPopBack(message)) { - std::string pump = message["pump"].asString(); - if(pump.length() == 0 ) continue; // No pump. - LLEventPumps::instance().obtain(pump).post(message["payload"]); - } - return false; -} - - -bool LLMainLoopRepeater::onMessage(LLSD const & event) -{ - try { - mQueue->pushFront(event); - } catch(LLThreadSafeQueueError & e) { - LL_WARNS() << "could not repeat message (" << e.what() << ")" << - event.asString() << LL_ENDL; - } - return false; -} diff --git a/indra/newview/llmainlooprepeater.h b/indra/newview/llmainlooprepeater.h deleted file mode 100644 index 2ec3a74e4a..0000000000 --- a/indra/newview/llmainlooprepeater.h +++ /dev/null @@ -1,64 +0,0 @@ -/** - * @file llmainlooprepeater.h - * @brief a service for repeating messages on the main loop. - * - * $LicenseInfo:firstyear=2010&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#ifndef LL_LLMAINLOOPREPEATER_H -#define LL_LLMAINLOOPREPEATER_H - - -#include "llsd.h" -#include "llthreadsafequeue.h" - - -// -// A service which creates the pump 'mainlooprepeater' to which any thread can -// post a message that will be re-posted on the main loop. -// -// The posted message should contain two map elements: pump and payload. The -// pump value is a string naming the pump to which the message should be -// re-posted. The payload value is what will be posted to the designated pump. -// -class LLMainLoopRepeater: - public LLSingleton<LLMainLoopRepeater> -{ - LLSINGLETON(LLMainLoopRepeater); -public: - // Start the repeater service. - void start(void); - - // Stop the repeater service. - void stop(void); - -private: - LLTempBoundListener mMainLoopConnection; - LLTempBoundListener mRepeaterConnection; - LLThreadSafeQueue<LLSD> * mQueue; - - bool onMainLoop(LLSD const &); - bool onMessage(LLSD const & event); -}; - - -#endif diff --git a/indra/newview/llmeshrepository.cpp b/indra/newview/llmeshrepository.cpp index 2c1c1191da..8ac64dbd15 100644 --- a/indra/newview/llmeshrepository.cpp +++ b/indra/newview/llmeshrepository.cpp @@ -4046,28 +4046,26 @@ S32 LLMeshRepository::getActualMeshLOD(const LLVolumeParams& mesh_params, S32 lo const LLMeshSkinInfo* LLMeshRepository::getSkinInfo(const LLUUID& mesh_id, const LLVOVolume* requesting_obj) { LL_RECORD_BLOCK_TIME(FTM_MESH_FETCH); + if (mesh_id.notNull()) + { + skin_map::iterator iter = mSkinMap.find(mesh_id); + if (iter != mSkinMap.end()) + { + return &(iter->second); + } - if (mesh_id.notNull()) - { - skin_map::iterator iter = mSkinMap.find(mesh_id); - if (iter != mSkinMap.end()) - { - return &(iter->second); - } - - //no skin info known about given mesh, try to fetch it - { - LLMutexLock lock(mMeshMutex); - //add volume to list of loading meshes - skin_load_map::iterator iter = mLoadingSkins.find(mesh_id); - if (iter == mLoadingSkins.end()) - { //no request pending for this skin info - mPendingSkinRequests.push(mesh_id); - } - mLoadingSkins[mesh_id].insert(requesting_obj->getID()); - } - } - + //no skin info known about given mesh, try to fetch it + { + LLMutexLock lock(mMeshMutex); + //add volume to list of loading meshes + skin_load_map::iterator iter = mLoadingSkins.find(mesh_id); + if (iter == mLoadingSkins.end()) + { //no request pending for this skin info + mPendingSkinRequests.push(mesh_id); + } + mLoadingSkins[mesh_id].insert(requesting_obj->getID()); + } + } return NULL; } diff --git a/indra/newview/llmeshrepository.h b/indra/newview/llmeshrepository.h index 81e49cb1d8..c1698194cb 100644 --- a/indra/newview/llmeshrepository.h +++ b/indra/newview/llmeshrepository.h @@ -27,6 +27,7 @@ #ifndef LL_MESH_REPOSITORY_H #define LL_MESH_REPOSITORY_H +#include <unordered_map> #include "llassettype.h" #include "llmodel.h" #include "lluuid.h" @@ -613,7 +614,7 @@ public: typedef std::map<LLVolumeParams, std::set<LLUUID> > mesh_load_map; mesh_load_map mLoadingMeshes[4]; - typedef std::map<LLUUID, LLMeshSkinInfo> skin_map; + typedef std::unordered_map<LLUUID, LLMeshSkinInfo> skin_map; skin_map mSkinMap; typedef std::map<LLUUID, LLModel::Decomposition*> decomposition_map; diff --git a/indra/newview/llmodelpreview.cpp b/indra/newview/llmodelpreview.cpp index a9e80ab5da..01bddd781d 100644 --- a/indra/newview/llmodelpreview.cpp +++ b/indra/newview/llmodelpreview.cpp @@ -591,7 +591,7 @@ void LLModelPreview::rebuildUploadData() bool upload_skinweights = fmp && fmp->childGetValue("upload_skin").asBoolean(); if (upload_skinweights && high_lod_model->mSkinInfo.mJointNames.size() > 0) { - LLQuaternion bind_rot = LLSkinningUtil::getUnscaledQuaternion(high_lod_model->mSkinInfo.mBindShapeMatrix); + LLQuaternion bind_rot = LLSkinningUtil::getUnscaledQuaternion(LLMatrix4(high_lod_model->mSkinInfo.mBindShapeMatrix)); LLQuaternion identity; if (!bind_rot.isEqualEps(identity, 0.01)) { @@ -3298,7 +3298,7 @@ BOOL LLModelPreview::render() LLJoint *joint = getPreviewAvatar()->getJoint(skin->mJointNums[j]); if (joint) { - const LLVector3& jointPos = skin->mAlternateBindMatrix[j].getTranslation(); + const LLVector3& jointPos = LLVector3(skin->mAlternateBindMatrix[j].getTranslation()); if (joint->aboveJointPosThreshold(jointPos)) { bool override_changed; @@ -3340,11 +3340,10 @@ BOOL LLModelPreview::render() //build matrix palette LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT]; - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, joint_count, + LLSkinningUtil::initSkinningMatrixPalette(mat, joint_count, skin, getPreviewAvatar()); - LLMatrix4a bind_shape_matrix; - bind_shape_matrix.loadu(skin->mBindShapeMatrix); + const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix; U32 max_joints = LLSkinningUtil::getMaxJointCount(); for (U32 j = 0; j < buffer->getNumVerts(); ++j) { diff --git a/indra/newview/llnotificationofferhandler.cpp b/indra/newview/llnotificationofferhandler.cpp index a9678b1e93..d9359d20cf 100644 --- a/indra/newview/llnotificationofferhandler.cpp +++ b/indra/newview/llnotificationofferhandler.cpp @@ -166,14 +166,14 @@ bool LLOfferHandler::processNotification(const LLNotificationPtr& notification) /*virtual*/ void LLOfferHandler::onChange(LLNotificationPtr p) { - LLToastNotifyPanel* panelp = LLToastNotifyPanel::getInstance(p->getID()); + auto panelp = LLToastNotifyPanel::getInstance(p->getID()); if (panelp) { // // HACK: if we're dealing with a notification embedded in IM, update it // otherwise remove its toast // - if (dynamic_cast<LLIMToastNotifyPanel*>(panelp)) + if (dynamic_cast<LLIMToastNotifyPanel*>(panelp.get())) { panelp->updateNotification(); } diff --git a/indra/newview/llskinningutil.cpp b/indra/newview/llskinningutil.cpp index e02b21f036..dc12de29fb 100644 --- a/indra/newview/llskinningutil.cpp +++ b/indra/newview/llskinningutil.cpp @@ -35,7 +35,6 @@ #include "llrigginginfo.h" #define DEBUG_SKINNING LL_DEBUG -#define MAT_USE_SSE 1 void dump_avatar_and_skin_state(const std::string& reason, LLVOAvatar *avatar, const LLMeshSkinInfo *skin) { @@ -120,36 +119,26 @@ void LLSkinningUtil::scrubInvalidJoints(LLVOAvatar *avatar, LLMeshSkinInfo* skin skin->mInvalidJointsScrubbed = true; } -#define MAT_USE_SSE 1 - void LLSkinningUtil::initSkinningMatrixPalette( - LLMatrix4* mat, + LLMatrix4a* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar) { + LL_PROFILE_ZONE_SCOPED; + initJointNums(const_cast<LLMeshSkinInfo*>(skin), avatar); + + LLMatrix4a world[LL_CHARACTER_MAX_ANIMATED_JOINTS]; + for (U32 j = 0; j < count; ++j) { S32 joint_num = skin->mJointNums[j]; - LLJoint *joint = NULL; - if (joint_num >= 0 && joint_num < LL_CHARACTER_MAX_ANIMATED_JOINTS) - { - joint = avatar->getJoint(joint_num); - } - llassert(joint); + LLJoint *joint = avatar->getJoint(joint_num); + if (joint) { -#ifdef MAT_USE_SSE - LLMatrix4a bind, world, res; - bind.loadu(skin->mInvBindMatrix[j]); - world.loadu(joint->getWorldMatrix()); - matMul(bind,world,res); - memcpy(mat[j].mMatrix,res.mMatrix,16*sizeof(float)); -#else - mat[j] = skin->mInvBindMatrix[j]; - mat[j] *= joint->getWorldMatrix(); -#endif + world[j] = joint->getWorldMatrix4a(); } else { @@ -159,16 +148,27 @@ void LLSkinningUtil::initSkinningMatrixPalette( // rendering should be disabled unless all joints are // valid. In other cases of skinned rendering, invalid // joints should already have been removed during scrubInvalidJoints(). - LL_WARNS_ONCE("Avatar") << avatar->getFullname() - << " rigged to invalid joint name " << skin->mJointNames[j] - << " num " << skin->mJointNums[j] << LL_ENDL; - LL_WARNS_ONCE("Avatar") << avatar->getFullname() - << " avatar build state: isBuilt() " << avatar->isBuilt() - << " mInitFlags " << avatar->mInitFlags << LL_ENDL; + LL_WARNS_ONCE("Avatar") << avatar->getFullname() + << " rigged to invalid joint name " << skin->mJointNames[j] + << " num " << skin->mJointNums[j] << LL_ENDL; + LL_WARNS_ONCE("Avatar") << avatar->getFullname() + << " avatar build state: isBuilt() " << avatar->isBuilt() + << " mInitFlags " << avatar->mInitFlags << LL_ENDL; #endif dump_avatar_and_skin_state("initSkinningMatrixPalette joint not found", avatar, skin); } } + + //NOTE: pointer striders used here as a micro-optimization over vector/array lookups + const LLMatrix4a* invBind = &(skin->mInvBindMatrix[0]); + const LLMatrix4a* w = world; + LLMatrix4a* m = mat; + LLMatrix4a* end = m + count; + + while (m < end) + { + matMulUnsafe(*(invBind++), *(w++), *(m++)); + } } void LLSkinningUtil::checkSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin) @@ -212,7 +212,7 @@ void LLSkinningUtil::scrubSkinWeights(LLVector4a* weights, U32 num_vertices, con void LLSkinningUtil::getPerVertexSkinMatrix( F32* weights, - LLMatrix4a* mat, + const LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints) @@ -270,6 +270,7 @@ void LLSkinningUtil::initJointNums(LLMeshSkinInfo* skin, LLVOAvatar *avatar) { if (!skin->mJointNumsInitialized) { + LL_PROFILE_ZONE_SCOPED; for (U32 j = 0; j < skin->mJointNames.size(); ++j) { #if DEBUG_SKINNING @@ -357,13 +358,11 @@ void LLSkinningUtil::updateRiggingInfo(const LLMeshSkinInfo* skin, LLVOAvatar *a rig_info_tab[joint_num].setIsRiggedTo(true); // FIXME could precompute these matMuls. - LLMatrix4a bind_shape; - LLMatrix4a inv_bind; + const LLMatrix4a& bind_shape = skin->mBindShapeMatrix; + const LLMatrix4a& inv_bind = skin->mInvBindMatrix[joint_index]; LLMatrix4a mat; LLVector4a pos_joint_space; - bind_shape.loadu(skin->mBindShapeMatrix); - inv_bind.loadu(skin->mInvBindMatrix[joint_index]); matMul(bind_shape, inv_bind, mat); mat.affineTransform(pos, pos_joint_space); @@ -426,3 +425,4 @@ LLQuaternion LLSkinningUtil::getUnscaledQuaternion(const LLMatrix4& mat4) bind_rot.normalize(); return bind_rot; } + diff --git a/indra/newview/llskinningutil.h b/indra/newview/llskinningutil.h index efe7c85997..807418f983 100644 --- a/indra/newview/llskinningutil.h +++ b/indra/newview/llskinningutil.h @@ -42,10 +42,10 @@ namespace LLSkinningUtil S32 getMaxJointCount(); U32 getMeshJointCount(const LLMeshSkinInfo *skin); void scrubInvalidJoints(LLVOAvatar *avatar, LLMeshSkinInfo* skin); - void initSkinningMatrixPalette(LLMatrix4* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar); + void initSkinningMatrixPalette(LLMatrix4a* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar); void checkSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin); void scrubSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin); - void getPerVertexSkinMatrix(F32* weights, LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints); + void getPerVertexSkinMatrix(F32* weights, const LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints); LL_FORCE_INLINE void getPerVertexSkinMatrixWithIndices( F32* weights, diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index efa4a7fd66..253b6b9953 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -503,7 +503,9 @@ LLSpatialGroup* LLSpatialGroup::getParent() } BOOL LLSpatialGroup::removeObject(LLDrawable *drawablep, BOOL from_octree) - { +{ + LL_PROFILE_ZONE_SCOPED + if(!drawablep) { return FALSE; @@ -591,6 +593,8 @@ public: void LLSpatialGroup::setState(U32 state, S32 mode) { + LL_PROFILE_ZONE_SCOPED + llassert(state <= LLSpatialGroup::STATE_MASK); if (mode > STATE_MODE_SINGLE) @@ -638,6 +642,8 @@ public: void LLSpatialGroup::clearState(U32 state, S32 mode) { + LL_PROFILE_ZONE_SCOPED + llassert(state <= LLSpatialGroup::STATE_MASK); if (mode > STATE_MODE_SINGLE) @@ -724,6 +730,8 @@ void LLSpatialGroup::updateDistance(LLCamera &camera) F32 LLSpatialPartition::calcDistance(LLSpatialGroup* group, LLCamera& camera) { + LL_PROFILE_ZONE_SCOPED + LLVector4a eye; LLVector4a origin; origin.load3(camera.getOrigin().mV); @@ -815,6 +823,8 @@ F32 LLSpatialGroup::getUpdateUrgency() const BOOL LLSpatialGroup::changeLOD() { + LL_PROFILE_ZONE_SCOPED + if (hasState(ALPHA_DIRTY | OBJECT_DIRTY)) { //a rebuild is going to happen, update distance and LoD @@ -907,6 +917,8 @@ void LLSpatialGroup::handleDestruction(const TreeNode* node) void LLSpatialGroup::handleChildAddition(const OctreeNode* parent, OctreeNode* child) { + LL_PROFILE_ZONE_SCOPED + if (child->getListenerCount() == 0) { new LLSpatialGroup(child, getSpatialPartition()); @@ -2700,11 +2712,17 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume) glVertexPointer(3, GL_FLOAT, 16, phys_volume->mHullPoints); gGL.diffuseColor4fv(line_color.mV); gGL.syncMatrices(); - glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + { + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x20FF20 ) + glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + } gGL.diffuseColor4fv(color.mV); glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); - glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + { + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x40FF40 ) + glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + } } else { @@ -3222,6 +3240,7 @@ void renderRaycast(LLDrawable* drawablep) gGL.diffuseColor4f(0,1,1,0.5f); glVertexPointer(3, GL_FLOAT, sizeof(LLVector4a), face.mPositions); gGL.syncMatrices(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x60FF60 ); glDrawElements(GL_TRIANGLES, face.mNumIndices, GL_UNSIGNED_SHORT, face.mIndices); } diff --git a/indra/newview/llstartup.cpp b/indra/newview/llstartup.cpp index 1242131534..13e7fcb6e4 100644 --- a/indra/newview/llstartup.cpp +++ b/indra/newview/llstartup.cpp @@ -179,7 +179,6 @@ #include "pipeline.h" #include "llappviewer.h" #include "llfasttimerview.h" -#include "lltelemetry.h" #include "llfloatermap.h" #include "llweb.h" #include "llvoiceclient.h" @@ -206,6 +205,9 @@ #include "llstacktrace.h" +#include "threadpool.h" + + #if LL_WINDOWS #include "lldxhardware.h" #endif @@ -302,6 +304,18 @@ void callback_cache_name(const LLUUID& id, const std::string& full_name, bool is // local classes // +void launchThreadPool() +{ + LLSD poolSizes{ gSavedSettings.getLLSD("ThreadPoolSizes") }; + LLSD sizeSpec{ poolSizes["General"] }; + LLSD::Integer size{ sizeSpec.isInteger()? sizeSpec.asInteger() : 3 }; + LL_DEBUGS("ThreadPool") << "Instantiating General pool with " + << size << " threads" << LL_ENDL; + // Use a function-static ThreadPool: static duration, but instantiated + // only on demand. + static LL::ThreadPool pool("General", size); +} + void update_texture_fetch() { LLAppViewer::getTextureCache()->update(1); // unpauses the texture cache thread @@ -530,8 +544,6 @@ bool idle_startup() } #if LL_WINDOWS - LLPROFILE_STARTUP(); - // On the windows dev builds, unpackaged, the message.xml file will // be located in indra/build-vc**/newview/<config>/app_settings. std::string message_path = gDirUtilp->getExpandedFilename(LL_PATH_APP_SETTINGS,"message.xml"); @@ -1492,6 +1504,9 @@ bool idle_startup() gAgentCamera.resetCamera(); display_startup(); + // start up the ThreadPool we'll use for textures et al. + launchThreadPool(); + // Initialize global class data needed for surfaces (i.e. textures) LL_DEBUGS("AppInit") << "Initializing sky..." << LL_ENDL; // Initialize all of the viewer object classes for the first time (doing things like texture fetches. diff --git a/indra/newview/llsurface.cpp b/indra/newview/llsurface.cpp index cb356726e6..ea36e1d7be 100644 --- a/indra/newview/llsurface.cpp +++ b/indra/newview/llsurface.cpp @@ -1214,6 +1214,7 @@ F32 LLSurface::getWaterHeight() const BOOL LLSurface::generateWaterTexture(const F32 x, const F32 y, const F32 width, const F32 height) { + LL_PROFILE_ZONE_SCOPED if (!getWaterTexture()) { return FALSE; diff --git a/indra/newview/llsurfacepatch.cpp b/indra/newview/llsurfacepatch.cpp index 5e056944e9..aeefcd6fb8 100644 --- a/indra/newview/llsurfacepatch.cpp +++ b/indra/newview/llsurfacepatch.cpp @@ -728,6 +728,7 @@ BOOL LLSurfacePatch::updateTexture() void LLSurfacePatch::updateGL() { + LL_PROFILE_ZONE_SCOPED F32 meters_per_grid = getSurface()->getMetersPerGrid(); F32 grids_per_patch_edge = (F32)getSurface()->getGridsPerPatchEdge(); diff --git a/indra/newview/lltelemetry.cpp b/indra/newview/lltelemetry.cpp deleted file mode 100644 index 0c63e2fede..0000000000 --- a/indra/newview/lltelemetry.cpp +++ /dev/null @@ -1,145 +0,0 @@ - /** - * @file lltelemetry.cpp - * @brief Wrapper for Rad Game Tools Telemetry - * - * $LicenseInfo:firstyear=2020&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2020, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#include "llviewerprecompiledheaders.h" - -#include "lltelemetry.h" - -#if LLPROFILE_USE_RAD_TELEMETRY_PROFILER - #if LL_WINDOWS - #include "llwin32headers.h" - - // build-vc120-64\packages\lib\release - // build-vc150-64\packages\lib\release - #ifdef _MSC_VER - #pragma comment(lib,"rad_tm_win64.lib") - #else - #pragma message "NOTE: Rad GameTools Telemetry requested but non-MSVC compiler not yet supported on Windows" - #endif - #endif // LL_WINDOWS - - #if LL_DARWIN - #pragma message "NOTE: Rad Game Tools Telemetry requested but not yet supported on Darwin" - #endif - - #if LL_LINUX - #pragma message "NOTE: Rad Game Tools Telemetry requested but not yet supported on Linux" - #endif - -// -// local consts -// -static const tm_int32 TELEMETRY_BUFFER_SIZE = 8 * 1024 * 1024; - -// -// local globals -// -static char *gTelemetryBufferPtr = NULL; // Telemetry - -static const char *tm_status[ TMERR_INIT_NETWORKING_FAILED + 1 ] = -{ - "Telemetry pass: connected" // TM_OK - , "Telemetry FAIL: disabled via #define NTELEMETRY" // TMERR_DISABLED - , "Telemetry FAIL: invalid paramater" // TMERR_INVALID_PARAM - , "Telemetry FAIL: DLL not found" // TMERR_NULL_API - , "Telemetry FAIL: out of resources" // TMERR_OUT_OF_RESOURCES - , "Telemetry FAIL: tmInitialize() not called" // TMERR_UNINITIALIZED - , "Telemetry FAIL: bad hostname" // TMERR_BAD_HOSTNAME - , "Telemetry FAIL: couldn't connect to server" // TMERR_COULD_NOT_CONNECT - , "Telemetry FAIL: unknown network error" // TMERR_UNKNOWN_NETWORK - , "Telemetry FAIL: tmShutdown() already called" // TMERR_ALREADY_SHUTDOWN - , "Telemetry FAIL: memory buffer too small" // TMERR_ARENA_TOO_SMALL - , "Telemetry FAIL: server handshake error" // TMERR_BAD_HANDSHAKE - , "Telemetry FAIL: unaligned parameters" // TMERR_UNALIGNED - , "Telemetry FAIL: network not initialized" // TMERR_NETWORK_NOT_INITIALIZED -- WSAStartup not called before tmOpen() - , "Telemetry FAIL: bad version" // TMERR_BAD_VERSION - , "Telemetry FAIL: timer too large" // TMERR_BAD_TIMER - , "Telemetry FAIL: tmOpen() already called" // TMERR_ALREADY_OPENED - , "Telemetry FAIL: tmInitialize() already called" // TMERR_ALREADY_INITIALIZED - , "Telemetry FAIL: could't open file" // TMERR_FILE_OPEN_FAILED - , "Telemetry FAIL: tmOpen() failed networking" // TMERR_INIT_NETWORKING_FAILED -}; - -// -// exported functionality -// - -void telemetry_shutdown() -{ - #if LL_WINDOWS - if (gTelemetryBufferPtr) - { - tmClose(0); - tmShutdown(); - - delete[] gTelemetryBufferPtr; - gTelemetryBufferPtr = NULL; - } - #endif -} - -void telemetry_startup() -{ - #if LL_WINDOWS - tmLoadLibrary(TM_RELEASE); // Loads .dll - - gTelemetryBufferPtr = new char[ TELEMETRY_BUFFER_SIZE ]; - tmInitialize(TELEMETRY_BUFFER_SIZE, gTelemetryBufferPtr); - - tm_error telemetry_status = tmOpen( - 0, // unused - "SecondLife", // app name - __DATE__ " " __TIME__, // build identifier - "localhost", // server name (or filename) - TMCT_TCP, // connection type (or TMCT_FILE) - 4719, // port - TMOF_INIT_NETWORKING, // open flags - 250 ); // timeout ms - - if (telemetry_status == TMERR_UNKNOWN) - { - LL_ERRS() << "Telemetry FAIL: unknown error" << LL_ENDL; - } - else if (telemetry_status && (telemetry_status <= TMERR_INIT_NETWORKING_FAILED)) - { - LL_INFOS() << tm_status[ telemetry_status ] << LL_ENDL; - free(gTelemetryBufferPtr); - gTelemetryBufferPtr = NULL; - } - #endif // LL_WINDOWS -} - -// Called after we render a frame -void telemetry_update() -{ - #if LL_WINDOWS - if (gTelemetryBufferPtr) - { - tmTick(0); - } - #endif -} -#endif // LLPROFILE_USE_RAD_TELEMETRY_PROFILER diff --git a/indra/newview/lltelemetry.h b/indra/newview/lltelemetry.h deleted file mode 100644 index a73e5fcfa2..0000000000 --- a/indra/newview/lltelemetry.h +++ /dev/null @@ -1,81 +0,0 @@ -/** - * @file lltelemetry.h - * @brief Wrapper for Rad Game Tools Telemetry - * - * $LicenseInfo:firstyear=2020&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2020, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -/* -To use: - -1. Uncomment #define LLPROFILE_USE_RAD_TELEMETRY_PROFILER below - -2. Include this header file - #include "lltelemetry.h" - -3. Add zones to the functions you wish to profile - void onFoo() - { - LLPROFILE_ZONE("Foo"); - } -*/ -//#define LLPROFILE_USE_RAD_TELEMETRY_PROFILER 1 - -// Default NO local telemetry profiling -#ifndef LLPROFILE_USE_RAD_TELEMETRY_PROFILER - #define LLPROFILE_USE_RAD_TELEMETRY_PROFILER 0 - #define LLPROFILE_SHUTDOWN( ...) {} - #define LLPROFILE_STARTUP( ...) {} - #define LLPROFILE_UPDATE( ...) {} - - #define LLPROFILE_AUTO_CPU_MARKER_COLOR(r, g, b) - #define LLPROFILE_ENTER(name) - #define LLPROFILE_ENTER_FORMAT(format, ...) - #define LLPROFILE_FUNCTION - #define LLPROFILE_LEAVE() - #define LLPROFILE_THREAD_NAME(name) - #define LLPROFILE_ZONE(name) - #define LLPROFILE_ZONE_FORMAT(format, ...) -#else - #include <rad_tm.h> - - #define LLPROFILE_SHUTDOWN telemetry_shutdown - #define LLPROFILE_STARTUP telemetry_startup - #define LLPROFILE_UPDATE telemetry_update - - #define LLPROFILE_AUTO_CPU_MARKER_COLOR(r, g, b) tmZoneColor(r, g, b) - #define LLPROFILE_ENTER(name) tmEnter(0, 0, name) - #define LLPROFILE_ENTER_FORMAT(format, ...) tmEnter(0, 0, format, __VA_ARGS__) - #define LLPROFILE_FUNCTION tmFunction(0, 0) - #define LLPROFILE_LEAVE() tmLeave(0) - #define LLPROFILE_THREAD_NAME(name) tmThreadName(0, 0, name) - #define LLPROFILE_ZONE(name) tmZone(0, 0, name) - #define LLPROFILE_ZONE_FORMAT(format, ...) tmZone(0, 0, format, __VA_ARGS__) -#endif // LLPROFILE_USE_RAD_TELEMETRY_PROFILER - -// -// exported functionality -// - -extern void telemetry_startup(); -extern void telemetry_shutdown(); -extern void telemetry_update(); // called after every frame update diff --git a/indra/newview/llviewercontrollistener.cpp b/indra/newview/llviewercontrollistener.cpp index 3443bb644a..8820f9ec56 100644 --- a/indra/newview/llviewercontrollistener.cpp +++ b/indra/newview/llviewercontrollistener.cpp @@ -127,7 +127,7 @@ struct Info LLEventAPI::Response response; std::string groupname; - LLControlGroup* group; + LLControlGroup::ptr_t group; std::string key; LLControlVariable* control; }; @@ -187,7 +187,7 @@ void LLViewerControlListener::groups(LLSD const & request) struct CollectVars: public LLControlGroup::ApplyFunctor { - CollectVars(LLControlGroup* g): + CollectVars(LLControlGroup::ptr_t g): mGroup(g) {} @@ -200,7 +200,7 @@ struct CollectVars: public LLControlGroup::ApplyFunctor ("comment", control->getComment())); } - LLControlGroup* mGroup; + LLControlGroup::ptr_t mGroup; LLSD vars; }; @@ -210,7 +210,7 @@ void LLViewerControlListener::vars(LLSD const & request) // control name. Response response(LLSD(), request); std::string groupname(request["group"]); - LLControlGroup* group(LLControlGroup::getInstance(groupname)); + auto group(LLControlGroup::getInstance(groupname)); if (! group) { return response.error(STRINGIZE("Unrecognized group '" << groupname << "'")); diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp index 109dc93261..33842497d1 100644 --- a/indra/newview/llviewerdisplay.cpp +++ b/indra/newview/llviewerdisplay.cpp @@ -208,9 +208,11 @@ void display_update_camera() // Write some stats to LL_INFOS() void display_stats() { + LL_PROFILE_ZONE_SCOPED F32 fps_log_freq = gSavedSettings.getF32("FPSLogFrequency"); if (fps_log_freq > 0.f && gRecentFPSTime.getElapsedTimeF32() >= fps_log_freq) { + LL_PROFILE_ZONE_NAMED("DS - FPS"); F32 fps = gRecentFrameCount / fps_log_freq; LL_INFOS() << llformat("FPS: %.02f", fps) << LL_ENDL; gRecentFrameCount = 0; @@ -219,6 +221,7 @@ void display_stats() F32 mem_log_freq = gSavedSettings.getF32("MemoryLogFrequency"); if (mem_log_freq > 0.f && gRecentMemoryTime.getElapsedTimeF32() >= mem_log_freq) { + LL_PROFILE_ZONE_NAMED("DS - Memory"); gMemoryAllocated = U64Bytes(LLMemory::getCurrentRSS()); U32Megabytes memory = gMemoryAllocated; LL_INFOS() << "MEMORY: " << memory << LL_ENDL; @@ -228,6 +231,7 @@ void display_stats() F32 asset_storage_log_freq = gSavedSettings.getF32("AssetStorageLogFrequency"); if (asset_storage_log_freq > 0.f && gAssetStorageLogTime.getElapsedTimeF32() >= asset_storage_log_freq) { + LL_PROFILE_ZONE_NAMED("DS - Asset Storage"); gAssetStorageLogTime.reset(); gAssetStorage->logAssetStorageInfo(); } @@ -630,6 +634,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) if (!gDisconnected) { + LL_PROFILE_ZONE_NAMED("display - 1"); LLAppViewer::instance()->pingMainloopTimeout("Display:Update"); if (gPipeline.hasRenderType(LLPipeline::RENDER_TYPE_HUD)) { //don't draw hud objects in this frame @@ -722,6 +727,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) LLAppViewer::instance()->pingMainloopTimeout("Display:Swap"); { + LL_PROFILE_ZONE_NAMED("display - 2") if (gResizeScreenTexture) { gResizeScreenTexture = FALSE; @@ -777,6 +783,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) //if (!for_snapshot) { + LL_PROFILE_ZONE_NAMED("display - 3") LLAppViewer::instance()->pingMainloopTimeout("Display:Imagery"); gPipeline.generateWaterReflection(*LLViewerCamera::getInstance()); gPipeline.generateHighlight(*LLViewerCamera::getInstance()); @@ -825,7 +832,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) LLImageGL::deleteDeadTextures(); stop_glerror(); }*/ - } + } LLGLState::checkStates(); LLGLState::checkClientArrays(); @@ -840,6 +847,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) // LLAppViewer::instance()->pingMainloopTimeout("Display:StateSort"); { + LL_PROFILE_ZONE_NAMED("display - 3") LLViewerCamera::sCurCameraID = LLViewerCamera::CAMERA_WORLD; gPipeline.stateSort(*LLViewerCamera::getInstance(), result); stop_glerror(); @@ -948,6 +956,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) if (!(LLAppViewer::instance()->logoutRequestSent() && LLAppViewer::instance()->hasSavedFinalSnapshot()) && !gRestoreGL) { + LL_PROFILE_ZONE_NAMED("display - 4") LLViewerCamera::sCurCameraID = LLViewerCamera::CAMERA_WORLD; if (gSavedSettings.getBOOL("RenderDepthPrePass") && LLGLSLShader::sNoFixedFunction) @@ -1259,7 +1268,7 @@ bool setup_hud_matrices(const LLRect& screen_region) void render_ui(F32 zoom_factor, int subfield) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI); LLGLState::checkStates(); @@ -1274,7 +1283,7 @@ void render_ui(F32 zoom_factor, int subfield) if(LLSceneMonitor::getInstance()->needsUpdate()) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_SCENE_MON); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_SCENE_MON); gGL.pushMatrix(); gViewerWindow->setup2DRender(); LLSceneMonitor::getInstance()->compare(); @@ -1282,55 +1291,64 @@ void render_ui(F32 zoom_factor, int subfield) gGL.popMatrix(); } - // Finalize scene - gPipeline.renderFinalize(); + // Finalize scene + gPipeline.renderFinalize(); - LL_RECORD_BLOCK_TIME(FTM_RENDER_HUD); - render_hud_elements(); - render_hud_attachments(); - - LLGLSDefault gls_default; - LLGLSUIDefault gls_ui; { - gPipeline.disableLights(); - } + // SL-15709 + // NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_RENDER_HUD); + render_hud_elements(); + render_hud_attachments(); - { - gGL.color4f(1,1,1,1); - if (gPipeline.hasRenderDebugFeatureMask(LLPipeline::RENDER_DEBUG_FEATURE_UI)) + LLGLSDefault gls_default; + LLGLSUIDefault gls_ui; { - if (!gDisconnected) + gPipeline.disableLights(); + } + + { + gGL.color4f(1,1,1,1); + if (gPipeline.hasRenderDebugFeatureMask(LLPipeline::RENDER_DEBUG_FEATURE_UI)) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_3D); - render_ui_3d(); + if (!gDisconnected) + { + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_3D); + render_ui_3d(); + LLGLState::checkStates(); + } + else + { + render_disconnected_background(); + } + + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_2D); + render_ui_2d(); LLGLState::checkStates(); } - else + gGL.flush(); + { - render_disconnected_background(); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_DEBUG_TEXT); + gViewerWindow->setup2DRender(); + gViewerWindow->updateDebugText(); + gViewerWindow->drawDebugText(); } - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_2D); - render_ui_2d(); - LLGLState::checkStates(); + LLVertexBuffer::unbind(); } - gGL.flush(); + if (!gSnapshot) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_DEBUG_TEXT); - gViewerWindow->setup2DRender(); - gViewerWindow->updateDebugText(); - gViewerWindow->drawDebugText(); + set_current_modelview(saved_view); + gGL.popMatrix(); } - LLVertexBuffer::unbind(); - } - - if (!gSnapshot) - { - set_current_modelview(saved_view); - gGL.popMatrix(); - } + } // Tracy integration } static LLTrace::BlockTimerStatHandle FTM_SWAP("Swap"); diff --git a/indra/newview/llviewermessage.cpp b/indra/newview/llviewermessage.cpp index 39c891c9c1..94d2d216b9 100644 --- a/indra/newview/llviewermessage.cpp +++ b/indra/newview/llviewermessage.cpp @@ -3978,8 +3978,8 @@ void process_sim_stats(LLMessageSystem *msg, void **user_data) F32 stat_value; msg->getU32("Stat", "StatID", stat_id, i); msg->getF32("Stat", "StatValue", stat_value, i); - LLStatViewer::SimMeasurementSampler* measurementp = LLStatViewer::SimMeasurementSampler::getInstance((ESimStatID)stat_id); - + auto measurementp = LLStatViewer::SimMeasurementSampler::getInstance((ESimStatID)stat_id); + if (measurementp ) { measurementp->sample(stat_value); diff --git a/indra/newview/llviewerobjectlist.cpp b/indra/newview/llviewerobjectlist.cpp index 63e48d1dd0..0b20556104 100644 --- a/indra/newview/llviewerobjectlist.cpp +++ b/indra/newview/llviewerobjectlist.cpp @@ -168,6 +168,8 @@ U64 LLViewerObjectList::getIndex(const U32 local_id, BOOL LLViewerObjectList::removeFromLocalIDTable(const LLViewerObject* objectp) { + LL_PROFILE_ZONE_SCOPED + if(objectp && objectp->getRegion()) { U32 local_id = objectp->mLocalID; @@ -303,6 +305,8 @@ static LLTrace::BlockTimerStatHandle FTM_PROCESS_OBJECTS("Process Objects"); LLViewerObject* LLViewerObjectList::processObjectUpdateFromCache(LLVOCacheEntry* entry, LLViewerRegion* regionp) { + LL_PROFILE_ZONE_SCOPED + LLDataPacker *cached_dpp = entry->getDP(); if (!cached_dpp) @@ -848,6 +852,8 @@ static LLTrace::BlockTimerStatHandle FTM_IDLE_COPY("Idle Copy"); void LLViewerObjectList::update(LLAgent &agent) { + LL_PROFILE_ZONE_SCOPED + // Update globals LLViewerObject::setVelocityInterpolate( gSavedSettings.getBOOL("VelocityInterpolate") ); LLViewerObject::setPingInterpolate( gSavedSettings.getBOOL("PingInterpolate") ); @@ -1293,6 +1299,8 @@ void LLViewerObjectList::clearDebugText() void LLViewerObjectList::cleanupReferences(LLViewerObject *objectp) { + LL_PROFILE_ZONE_SCOPED + bool new_dead_object = true; if (mDeadObjects.find(objectp->mID) != mDeadObjects.end()) { @@ -1523,6 +1531,8 @@ void LLViewerObjectList::removeFromActiveList(LLViewerObject* objectp) void LLViewerObjectList::updateActive(LLViewerObject *objectp) { + LL_PROFILE_ZONE_SCOPED + if (objectp->isDead()) { return; // We don't update dead objects! @@ -1843,6 +1853,8 @@ void LLViewerObjectList::renderObjectBounds(const LLVector3 ¢er) void LLViewerObjectList::generatePickList(LLCamera &camera) { + LL_PROFILE_ZONE_SCOPED + LLViewerObject *objectp; S32 i; // Reset all of the GL names to zero. @@ -2104,6 +2116,8 @@ LLViewerObject *LLViewerObjectList::replaceObject(const LLUUID &id, const LLPCod S32 LLViewerObjectList::findReferences(LLDrawable *drawablep) const { + LL_PROFILE_ZONE_SCOPED + LLViewerObject *objectp; S32 num_refs = 0; @@ -2167,6 +2181,8 @@ void LLViewerObjectList::orphanize(LLViewerObject *childp, U32 parent_id, U32 ip void LLViewerObjectList::findOrphans(LLViewerObject* objectp, U32 ip, U32 port) { + LL_PROFILE_ZONE_SCOPED + if (objectp->isDead()) { LL_WARNS() << "Trying to find orphans for dead obj " << objectp->mID diff --git a/indra/newview/llviewerparceloverlay.cpp b/indra/newview/llviewerparceloverlay.cpp index 7c3dd00e1a..02f7bbeed8 100644 --- a/indra/newview/llviewerparceloverlay.cpp +++ b/indra/newview/llviewerparceloverlay.cpp @@ -847,6 +847,7 @@ void LLViewerParcelOverlay::setDirty() void LLViewerParcelOverlay::updateGL() { + LL_PROFILE_ZONE_SCOPED updateOverlayTexture(); } diff --git a/indra/newview/llviewershadermgr.cpp b/indra/newview/llviewershadermgr.cpp index be5c22e7c3..7dcf29eb75 100644 --- a/indra/newview/llviewershadermgr.cpp +++ b/indra/newview/llviewershadermgr.cpp @@ -384,7 +384,7 @@ void LLViewerShaderMgr::initAttribsAndUniforms(void) S32 LLViewerShaderMgr::getShaderLevel(S32 type) { - return LLPipeline::sDisableShaders ? 0 : mShaderLevel[type]; + return mShaderLevel[type]; } //============================================================================ @@ -400,15 +400,6 @@ void LLViewerShaderMgr::setShaders() return; } - if (!gGLManager.mHasShaderObjects - || !gGLManager.mHasVertexShader - || !gGLManager.mHasFragmentShader) - { - // Viewer will show 'hardware requirements' warning later - LL_INFOS("ShaderLoading") << "Shaders not supported" << LL_ENDL; - return; - } - static LLCachedControl<U32> max_texture_index(gSavedSettings, "RenderMaxTextureIndex", 16); LLGLSLShader::sIndexedTextureChannels = llmax(llmin(gGLManager.mNumTextureImageUnits, (S32) max_texture_index), 1); diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index ca01bb46aa..af55f68cd2 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -208,6 +208,7 @@ void LLViewerTextureManager::findTextures(const LLUUID& id, std::vector<LLViewe LLViewerFetchedTexture* LLViewerTextureManager::findFetchedTexture(const LLUUID& id, S32 tex_type) { + LL_PROFILE_ZONE_SCOPED; return gTextureList.findImage(id, (ETexListType)tex_type); } diff --git a/indra/newview/llviewertexturelist.cpp b/indra/newview/llviewertexturelist.cpp index 561319ca5d..12495078e9 100644 --- a/indra/newview/llviewertexturelist.cpp +++ b/indra/newview/llviewertexturelist.cpp @@ -620,6 +620,7 @@ void LLViewerTextureList::findTexturesByID(const LLUUID &image_id, std::vector<L LLViewerFetchedTexture *LLViewerTextureList::findImage(const LLTextureKey &search_key) { + LL_PROFILE_ZONE_SCOPED; uuid_map_t::iterator iter = mUUIDMap.find(search_key); if (iter == mUUIDMap.end()) return NULL; diff --git a/indra/newview/llviewerwindow.cpp b/indra/newview/llviewerwindow.cpp index 1d13a306ef..2157585364 100644 --- a/indra/newview/llviewerwindow.cpp +++ b/indra/newview/llviewerwindow.cpp @@ -5280,6 +5280,7 @@ void LLViewerWindow::setup3DRender() void LLViewerWindow::setup3DViewport(S32 x_offset, S32 y_offset) { + LL_PROFILE_ZONE_SCOPED gGLViewport[0] = mWorldViewRectRaw.mLeft + x_offset; gGLViewport[1] = mWorldViewRectRaw.mBottom + y_offset; gGLViewport[2] = mWorldViewRectRaw.getWidth(); diff --git a/indra/newview/llvlcomposition.cpp b/indra/newview/llvlcomposition.cpp index c63c5f6b23..46beac8255 100644 --- a/indra/newview/llvlcomposition.cpp +++ b/indra/newview/llvlcomposition.cpp @@ -254,6 +254,7 @@ BOOL LLVLComposition::generateComposition() BOOL LLVLComposition::generateTexture(const F32 x, const F32 y, const F32 width, const F32 height) { + LL_PROFILE_ZONE_SCOPED llassert(mSurfacep); llassert(x >= 0.f); llassert(y >= 0.f); diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp index e085a945a8..a588d05ff7 100644 --- a/indra/newview/llvoavatar.cpp +++ b/indra/newview/llvoavatar.cpp @@ -3136,6 +3136,8 @@ void LLVOAvatar::idleUpdateWindEffect() void LLVOAvatar::idleUpdateNameTag(const LLVector3& root_pos_last) { + LL_PROFILE_ZONE_SCOPED; + // update chat bubble //-------------------------------------------------------------------- // draw text label over character's head @@ -4884,6 +4886,8 @@ bool LLVOAvatar::shouldAlphaMask() //----------------------------------------------------------------------------- U32 LLVOAvatar::renderSkinned() { + LL_PROFILE_ZONE_SCOPED; + U32 num_indices = 0; if (!mIsBuilt) @@ -6158,27 +6162,29 @@ LLJoint *LLVOAvatar::getJoint( const std::string &name ) LLJoint *LLVOAvatar::getJoint( S32 joint_num ) { LLJoint *pJoint = NULL; - S32 collision_start = mNumBones; - S32 attachment_start = mNumBones + mNumCollisionVolumes; - if (joint_num>=attachment_start) + if (joint_num >= 0) { - // Attachment IDs start at 1 - S32 attachment_id = joint_num - attachment_start + 1; - attachment_map_t::iterator iter = mAttachmentPoints.find(attachment_id); - if (iter != mAttachmentPoints.end()) + if (joint_num < mNumBones) { - pJoint = iter->second; + pJoint = mSkeleton[joint_num]; + } + else if (joint_num < mNumBones + mNumCollisionVolumes) + { + S32 collision_id = joint_num - mNumBones; + pJoint = &mCollisionVolumes[collision_id]; + } + else + { + // Attachment IDs start at 1 + S32 attachment_id = joint_num - (mNumBones + mNumCollisionVolumes) + 1; + attachment_map_t::iterator iter = mAttachmentPoints.find(attachment_id); + if (iter != mAttachmentPoints.end()) + { + pJoint = iter->second; + } } } - else if (joint_num>=collision_start) - { - S32 collision_id = joint_num-collision_start; - pJoint = &mCollisionVolumes[collision_id]; - } - else if (joint_num>=0) - { - pJoint = mSkeleton[joint_num]; - } + llassert(!pJoint || pJoint->getJointNum() == joint_num); return pJoint; } @@ -6513,7 +6519,7 @@ void LLVOAvatar::addAttachmentOverridesForObject(LLViewerObject *vo, std::set<LL LLJoint* pJoint = getJoint( lookingForJoint ); if (pJoint) { - const LLVector3& jointPos = pSkinData->mAlternateBindMatrix[i].getTranslation(); + const LLVector3& jointPos = LLVector3(pSkinData->mAlternateBindMatrix[i].getTranslation()); if (pJoint->aboveJointPosThreshold(jointPos)) { bool override_changed; @@ -7117,6 +7123,7 @@ void LLVOAvatar::updateGL() { if (mMeshTexturesDirty) { + LL_PROFILE_ZONE_SCOPED updateMeshTextures(); mMeshTexturesDirty = FALSE; } @@ -7862,6 +7869,8 @@ void LLVOAvatar::onGlobalColorChanged(const LLTexGlobalColor* global_color) // Do rigged mesh attachments display with this av? bool LLVOAvatar::shouldRenderRigged() const { + LL_PROFILE_ZONE_SCOPED; + if (getOverallAppearance() == AOA_NORMAL) { return true; @@ -8353,6 +8362,7 @@ void LLVOAvatar::updateMeshVisibility() // virtual void LLVOAvatar::updateMeshTextures() { + LL_PROFILE_ZONE_SCOPED static S32 update_counter = 0; mBakedTextureDebugText.clear(); @@ -10949,6 +10959,7 @@ void LLVOAvatar::updateOverallAppearanceAnimations() // Based on isVisuallyMuted(), but has 3 possible results. LLVOAvatar::AvatarOverallAppearance LLVOAvatar::getOverallAppearance() const { + LL_PROFILE_ZONE_SCOPED; AvatarOverallAppearance result = AOA_NORMAL; // Priority order (highest priority first) diff --git a/indra/newview/llvoavatar.h b/indra/newview/llvoavatar.h index 74ef589ca4..39adaab8ca 100644 --- a/indra/newview/llvoavatar.h +++ b/indra/newview/llvoavatar.h @@ -209,6 +209,11 @@ public: virtual LLJoint* getJoint(const std::string &name); LLJoint* getJoint(S32 num); + //if you KNOW joint_num is a valid animated joint index, use getSkeletonJoint for efficiency + inline LLJoint* getSkeletonJoint(S32 joint_num) { return mSkeleton[joint_num]; } + inline size_t getSkeletonJointCount() const { return mSkeleton.size(); } + + void addAttachmentOverridesForObject(LLViewerObject *vo, std::set<LLUUID>* meshes_seen = NULL, bool recursive = true); void removeAttachmentOverridesForObject(const LLUUID& mesh_id); void removeAttachmentOverridesForObject(LLViewerObject *vo); diff --git a/indra/newview/llvosurfacepatch.cpp b/indra/newview/llvosurfacepatch.cpp index 897bace4e1..b5560d3d79 100644 --- a/indra/newview/llvosurfacepatch.cpp +++ b/indra/newview/llvosurfacepatch.cpp @@ -218,6 +218,7 @@ void LLVOSurfacePatch::updateGL() { if (mPatchp) { + LL_PROFILE_ZONE_SCOPED mPatchp->updateGL(); } } diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index f063800587..e5a4b0f374 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -4807,7 +4807,7 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons LLMatrix4a mat[kMaxJoints]; U32 maxJoints = LLSkinningUtil::getMeshJointCount(skin); - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, maxJoints, skin, avatar); + LLSkinningUtil::initSkinningMatrixPalette(mat, maxJoints, skin, avatar); S32 rigged_vert_count = 0; S32 rigged_face_count = 0; @@ -4823,8 +4823,7 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons if ( weight ) { LLSkinningUtil::checkSkinWeights(weight, dst_face.mNumVertices, skin); - LLMatrix4a bind_shape_matrix; - bind_shape_matrix.loadu(skin->mBindShapeMatrix); + const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix; LLVector4a* pos = dst_face.mPositions; @@ -6045,123 +6044,130 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) if (group && group->hasState(LLSpatialGroup::MESH_DIRTY) && !group->hasState(LLSpatialGroup::GEOM_DIRTY)) { LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_VB); - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers + { + // SL-15709 -- NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers - group->mBuilt = 1.f; + group->mBuilt = 1.f; - S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ; + S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ; - const U32 MAX_BUFFER_COUNT = 4096; - LLVertexBuffer* locked_buffer[MAX_BUFFER_COUNT]; - - U32 buffer_count = 0; + const U32 MAX_BUFFER_COUNT = 4096; + LLVertexBuffer* locked_buffer[MAX_BUFFER_COUNT]; - for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) - { - LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); + U32 buffer_count = 0; - if (drawablep && !drawablep->isDead() && drawablep->isState(LLDrawable::REBUILD_ALL) && !drawablep->isState(LLDrawable::RIGGED) ) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { - LLVOVolume* vobj = drawablep->getVOVolume(); - if (debugLoggingEnabled("AnimatedObjectsLinkset")) - { - if (vobj->isAnimatedObject() && vobj->isRiggedMesh()) - { - std::string vobj_name = llformat("Vol%p", vobj); - F32 est_tris = vobj->getEstTrianglesMax(); - LL_DEBUGS("AnimatedObjectsLinkset") << vobj_name << " rebuildMesh, tris " << est_tris << LL_ENDL; - } - } - if (vobj->isNoLOD()) continue; - - vobj->preRebuild(); - - if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) - { - vobj->updateRelativeXform(true); - } + LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); - LLVolume* volume = vobj->getVolume(); - for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + if (drawablep && !drawablep->isDead() && drawablep->isState(LLDrawable::REBUILD_ALL) && !drawablep->isState(LLDrawable::RIGGED) ) { - LLFace* face = drawablep->getFace(i); - if (face) + LLVOVolume* vobj = drawablep->getVOVolume(); + if (debugLoggingEnabled("AnimatedObjectsLinkset")) { - LLVertexBuffer* buff = face->getVertexBuffer(); - if (buff) + if (vobj->isAnimatedObject() && vobj->isRiggedMesh()) { - llassert(!face->isState(LLFace::RIGGED)); + std::string vobj_name = llformat("Vol%p", vobj); + F32 est_tris = vobj->getEstTrianglesMax(); + LL_DEBUGS("AnimatedObjectsLinkset") << vobj_name << " rebuildMesh, tris " << est_tris << LL_ENDL; + } + } + if (vobj->isNoLOD()) continue; - if (!face->getGeometryVolume(*volume, face->getTEOffset(), - vobj->getRelativeXform(), vobj->getRelativeXformInvTrans(), face->getGeomIndex())) - { //something's gone wrong with the vertex buffer accounting, rebuild this group - group->dirtyGeom(); - gPipeline.markRebuild(group, TRUE); - } + vobj->preRebuild(); + if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + vobj->updateRelativeXform(true); + } - if (buff->isLocked() && buffer_count < MAX_BUFFER_COUNT) + LLVolume* volume = vobj->getVolume(); + for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + { + LLFace* face = drawablep->getFace(i); + if (face) + { + LLVertexBuffer* buff = face->getVertexBuffer(); + if (buff) { - locked_buffer[buffer_count++] = buff; + llassert(!face->isState(LLFace::RIGGED)); + + if (!face->getGeometryVolume(*volume, face->getTEOffset(), + vobj->getRelativeXform(), vobj->getRelativeXformInvTrans(), face->getGeomIndex())) + { //something's gone wrong with the vertex buffer accounting, rebuild this group + group->dirtyGeom(); + gPipeline.markRebuild(group, TRUE); + } + + + if (buff->isLocked() && buffer_count < MAX_BUFFER_COUNT) + { + locked_buffer[buffer_count++] = buff; + } } } } + + if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + vobj->updateRelativeXform(); + } + + drawablep->clearState(LLDrawable::REBUILD_ALL); } + } - if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + LL_RECORD_BLOCK_TIME(FTM_REBUILD_MESH_FLUSH); + for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter) { - vobj->updateRelativeXform(); + (*iter)->flush(); } - - drawablep->clearState(LLDrawable::REBUILD_ALL); + // don't forget alpha + if(group != NULL && + !group->mVertexBuffer.isNull() && + group->mVertexBuffer->isLocked()) + { + group->mVertexBuffer->flush(); + } } - } - - { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_MESH_FLUSH); - for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter) - { - (*iter)->flush(); - } - - // don't forget alpha - if(group != NULL && - !group->mVertexBuffer.isNull() && - group->mVertexBuffer->isLocked()) - { - group->mVertexBuffer->flush(); - } - } - //if not all buffers are unmapped - if(num_mapped_vertex_buffer != LLVertexBuffer::sMappedCount) - { - LL_WARNS() << "Not all mapped vertex buffers are unmapped!" << LL_ENDL ; - for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) + //if not all buffers are unmapped + if(num_mapped_vertex_buffer != LLVertexBuffer::sMappedCount) { - LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); - if(!drawablep) - { - continue; - } - for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + LL_WARNS() << "Not all mapped vertex buffers are unmapped!" << LL_ENDL ; + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { - LLFace* face = drawablep->getFace(i); - if (face) + LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); + if(!drawablep) + { + continue; + } + for (S32 i = 0; i < drawablep->getNumFaces(); ++i) { - LLVertexBuffer* buff = face->getVertexBuffer(); - if (buff && buff->isLocked()) + LLFace* face = drawablep->getFace(i); + if (face) { - buff->flush(); + LLVertexBuffer* buff = face->getVertexBuffer(); + if (buff && buff->isLocked()) + { + buff->flush(); + } } } } - } + } + + group->clearState(LLSpatialGroup::MESH_DIRTY | LLSpatialGroup::NEW_DRAWINFO); } - group->clearState(LLSpatialGroup::MESH_DIRTY | LLSpatialGroup::NEW_DRAWINFO); - } + } // Tracy integration // llassert(!group || !group->isState(LLSpatialGroup::NEW_DRAWINFO)); } diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp index a1a1db35d6..c7b0a2bfb4 100644 --- a/indra/newview/llworld.cpp +++ b/indra/newview/llworld.cpp @@ -1284,6 +1284,7 @@ void send_agent_pause() void send_agent_resume() { + LL_PROFILE_ZONE_SCOPED // Note: used to check for LLWorld initialization before it became a singleton. // Rather than just remove this check I'm changing it to assure that the message // system has been initialized. -MG diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index cd1b9c7c69..c0b469af81 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -338,7 +338,6 @@ S32 LLPipeline::sUseOcclusion = 0; bool LLPipeline::sDelayVBUpdate = true; bool LLPipeline::sAutoMaskAlphaDeferred = true; bool LLPipeline::sAutoMaskAlphaNonDeferred = false; -bool LLPipeline::sDisableShaders = false; bool LLPipeline::sRenderTransparentWater = true; bool LLPipeline::sRenderBump = true; bool LLPipeline::sBakeSunlight = false; @@ -1393,10 +1392,7 @@ void LLPipeline::restoreGL() bool LLPipeline::canUseVertexShaders() { - if (sDisableShaders || - !gGLManager.mHasVertexShader || - !gGLManager.mHasFragmentShader || - (assertInitialized() && mVertexShadersLoaded != 1) ) + if ((assertInitialized() && mVertexShadersLoaded != 1) ) { return false; } @@ -1408,8 +1404,7 @@ bool LLPipeline::canUseVertexShaders() bool LLPipeline::canUseWindLightShaders() const { - return (!LLPipeline::sDisableShaders && - gWLSkyProgram.mProgramObject != 0 && + return (gWLSkyProgram.mProgramObject != 0 && LLViewerShaderMgr::instance()->getShaderLevel(LLViewerShaderMgr::SHADER_WINDLIGHT) > 1); } @@ -1897,6 +1892,7 @@ void LLPipeline::createObject(LLViewerObject* vobj) void LLPipeline::resetFrameStats() { + LL_PROFILE_ZONE_SCOPED assertInitialized(); sCompiles = 0; @@ -2125,6 +2121,7 @@ void LLPipeline::grabReferences(LLCullResult& result) void LLPipeline::clearReferences() { + LL_PROFILE_ZONE_SCOPED sCull = NULL; mGroupSaveQ1.clear(); } @@ -4564,92 +4561,99 @@ void LLPipeline::renderGeomDeferred(LLCamera& camera) LLAppViewer::instance()->pingMainloopTimeout("Pipeline:RenderGeomDeferred"); LL_RECORD_BLOCK_TIME(FTM_RENDER_GEOMETRY); + { + // SL-15709 -- NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLS); - LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLS); - - LLGLEnable cull(GL_CULL_FACE); + LLGLEnable cull(GL_CULL_FACE); - for (pool_set_t::iterator iter = mPools.begin(); iter != mPools.end(); ++iter) - { - LLDrawPool *poolp = *iter; - if (hasRenderType(poolp->getType())) + for (pool_set_t::iterator iter = mPools.begin(); iter != mPools.end(); ++iter) { - poolp->prerender(); + LLDrawPool *poolp = *iter; + if (hasRenderType(poolp->getType())) + { + poolp->prerender(); + } } - } - LLGLEnable multisample(RenderFSAASamples > 0 ? GL_MULTISAMPLE_ARB : 0); + LLGLEnable multisample(RenderFSAASamples > 0 ? GL_MULTISAMPLE_ARB : 0); - LLVertexBuffer::unbind(); + LLVertexBuffer::unbind(); - LLGLState::checkStates(); - LLGLState::checkTextureChannels(); - LLGLState::checkClientArrays(); + LLGLState::checkStates(); + LLGLState::checkTextureChannels(); + LLGLState::checkClientArrays(); - U32 cur_type = 0; + U32 cur_type = 0; - gGL.setColorMask(true, true); + gGL.setColorMask(true, true); - pool_set_t::iterator iter1 = mPools.begin(); + pool_set_t::iterator iter1 = mPools.begin(); - while ( iter1 != mPools.end() ) - { - LLDrawPool *poolp = *iter1; + while ( iter1 != mPools.end() ) + { + LLDrawPool *poolp = *iter1; - cur_type = poolp->getType(); + cur_type = poolp->getType(); - pool_set_t::iterator iter2 = iter1; - if (hasRenderType(poolp->getType()) && poolp->getNumDeferredPasses() > 0) - { - LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLRENDER); + pool_set_t::iterator iter2 = iter1; + if (hasRenderType(poolp->getType()) && poolp->getNumDeferredPasses() > 0) + { + LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLRENDER); - gGLLastMatrix = NULL; - gGL.loadMatrix(gGLModelView); + gGLLastMatrix = NULL; + gGL.loadMatrix(gGLModelView); - for( S32 i = 0; i < poolp->getNumDeferredPasses(); i++ ) - { - LLVertexBuffer::unbind(); - poolp->beginDeferredPass(i); - for (iter2 = iter1; iter2 != mPools.end(); iter2++) + for( S32 i = 0; i < poolp->getNumDeferredPasses(); i++ ) { - LLDrawPool *p = *iter2; - if (p->getType() != cur_type) + LLVertexBuffer::unbind(); + poolp->beginDeferredPass(i); + for (iter2 = iter1; iter2 != mPools.end(); iter2++) { - break; + LLDrawPool *p = *iter2; + if (p->getType() != cur_type) + { + break; + } + + if ( !p->getSkipRenderFlag() ) { p->renderDeferred(i); } } - - if ( !p->getSkipRenderFlag() ) { p->renderDeferred(i); } - } - poolp->endDeferredPass(i); - LLVertexBuffer::unbind(); + poolp->endDeferredPass(i); + LLVertexBuffer::unbind(); - if (gDebugGL || gDebugPipeline) - { - LLGLState::checkStates(); + if (gDebugGL || gDebugPipeline) + { + LLGLState::checkStates(); + } } } - } - else - { - // Skip all pools of this type - for (iter2 = iter1; iter2 != mPools.end(); iter2++) + else { - LLDrawPool *p = *iter2; - if (p->getType() != cur_type) + // Skip all pools of this type + for (iter2 = iter1; iter2 != mPools.end(); iter2++) { - break; + LLDrawPool *p = *iter2; + if (p->getType() != cur_type) + { + break; + } } } + iter1 = iter2; + stop_glerror(); } - iter1 = iter2; - stop_glerror(); - } - gGLLastMatrix = NULL; - gGL.matrixMode(LLRender::MM_MODELVIEW); - gGL.loadMatrix(gGLModelView); + gGLLastMatrix = NULL; + gGL.matrixMode(LLRender::MM_MODELVIEW); + gGL.loadMatrix(gGLModelView); - gGL.setColorMask(true, false); + gGL.setColorMask(true, false); + + } // Tracy ZoneScoped } void LLPipeline::renderGeomPostDeferred(LLCamera& camera, bool do_occlusion) @@ -11139,6 +11143,7 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) if (LLPipeline::sRenderDeferred) { GLuint buff = GL_COLOR_ATTACHMENT0; + LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x8000FF ); glDrawBuffersARB(1, &buff); } diff --git a/indra/newview/pipeline.h b/indra/newview/pipeline.h index 0eaa6b141d..5605d26410 100644 --- a/indra/newview/pipeline.h +++ b/indra/newview/pipeline.h @@ -574,7 +574,6 @@ public: static bool sDelayVBUpdate; static bool sAutoMaskAlphaDeferred; static bool sAutoMaskAlphaNonDeferred; - static bool sDisableShaders; // if true, rendering will be done without shaders static bool sRenderTransparentWater; static bool sRenderBump; static bool sBakeSunlight; diff --git a/indra/test/test.cpp b/indra/test/test.cpp index 87c4a8d8a3..bb48216b2b 100644 --- a/indra/test/test.cpp +++ b/indra/test/test.cpp @@ -112,6 +112,7 @@ public: virtual void recordMessage(LLError::ELevel level, const std::string& message) { + LL_PROFILE_ZONE_SCOPED mFile << message << std::endl; } |