diff options
| author | Konstantin <const@const.me> | 2023-01-21 21:23:08 +0100 |
|---|---|---|
| committer | Konstantin <const@const.me> | 2023-01-21 21:23:08 +0100 |
| commit | cacec67bb649702db7a877de1b6482a46123f175 (patch) | |
| tree | 07456739a57274cfb8cf6750d2acd480b23d0de1 /Whisper/Utils/GpuProfiler.cpp | |
| parent | 040281bccc424341c964788aa7ca17876a5ac6a4 (diff) | |
Experimental, alternative busy wait implementation
Disabled with a `constexpr` flag because on a desktop with discrete GPU
this slowed down by about 20%. But the CPU load is about zero. Need to
test on iGPUs, thermal shenanigans might make a difference there.
Diffstat (limited to 'Whisper/Utils/GpuProfiler.cpp')
| -rw-r--r-- | Whisper/Utils/GpuProfiler.cpp | 15 |
1 files changed, 5 insertions, 10 deletions
diff --git a/Whisper/Utils/GpuProfiler.cpp b/Whisper/Utils/GpuProfiler.cpp index 6f19415..f7d8dc3 100644 --- a/Whisper/Utils/GpuProfiler.cpp +++ b/Whisper/Utils/GpuProfiler.cpp @@ -45,7 +45,7 @@ HRESULT GpuProfiler::Queue::create() namespace { - static uint64_t getTimestamp( ID3D11Query* query ) + static uint64_t getTimestamp( ID3D11Query* query, const DelayExecution& delay ) { ID3D11DeviceContext* const ctx = context(); @@ -56,12 +56,7 @@ namespace check( hr ); if( S_OK == hr ) return res; -#if 0 - Sleep( 1 ); -#else - for( size_t i = 0; i < 1024; i++ ) - _mm_pause(); -#endif + delay.delay(); } } @@ -86,7 +81,7 @@ void GpuProfiler::Queue::Entry::join( GpuProfiler& owner ) { assert( nullptr != block ); - uint64_t res = getTimestamp( query ); + uint64_t res = getTimestamp( query, owner.delay ); #if PROFILER_COLLECT_TAGS block->haveTimestamp( event, shader, tag, res, owner ); #else @@ -350,8 +345,8 @@ HRESULT GpuProfilerSimple::time( uint64_t& rdi ) const try { const D3D11_QUERY_DATA_TIMESTAMP_DISJOINT dtsd = waitForDisjointData( disjoint ); - const uint64_t t1 = getTimestamp( begin ); - const uint64_t t2 = getTimestamp( end ); + const uint64_t t2 = getTimestamp( end, delay ); + const uint64_t t1 = getTimestamp( begin, delay ); if( !dtsd.Disjoint ) { |