From 8be5717a60a4c4fbd29a0968a9f331152ff581e3 Mon Sep 17 00:00:00 2001 From: hthh Date: Mon, 13 Jun 2016 15:41:29 +1000 Subject: [PATCH] Improve PerfQuery accuracy In TimeSplitters: Future Perfect, PerfQuery is used to detect the visibility of lights and draw coronas. 25 points are drawn for each light. However, the returned count was incorrectly being divided by four leading to dim coronas. Using 4x antialiasing was a workaround because of a bug where antialiasing multiplied the PerfQuery results. This commit fixes that bug too (but only for OpenGL). --- Source/Core/VideoBackends/D3D/PerfQuery.cpp | 4 +++- Source/Core/VideoBackends/D3D12/PerfQuery.cpp | 6 +++++- Source/Core/VideoBackends/OGL/PerfQuery.cpp | 21 ++++++++++++++----- 3 files changed, 24 insertions(+), 7 deletions(-) diff --git a/Source/Core/VideoBackends/D3D/PerfQuery.cpp b/Source/Core/VideoBackends/D3D/PerfQuery.cpp index 2eb3c6d89d..15574a9c76 100644 --- a/Source/Core/VideoBackends/D3D/PerfQuery.cpp +++ b/Source/Core/VideoBackends/D3D/PerfQuery.cpp @@ -86,7 +86,7 @@ u32 PerfQuery::GetQueryResult(PerfQueryType type) else if (type == PQ_EFB_COPY_CLOCKS) result = m_results[PQG_EFB_COPY_CLOCKS]; - return result / 4; + return result; } void PerfQuery::FlushOne() @@ -102,6 +102,8 @@ void PerfQuery::FlushOne() } // NOTE: Reported pixel metrics should be referenced to native resolution + // TODO: Dropping the lower 2 bits from this count should be closer to actual + // hardware behavior when drawing triangles. m_results[entry.query_type] += (u32)(result * EFB_WIDTH / g_renderer->GetTargetWidth() * EFB_HEIGHT / g_renderer->GetTargetHeight()); diff --git a/Source/Core/VideoBackends/D3D12/PerfQuery.cpp b/Source/Core/VideoBackends/D3D12/PerfQuery.cpp index 4794b134c1..df5b5ce224 100644 --- a/Source/Core/VideoBackends/D3D12/PerfQuery.cpp +++ b/Source/Core/VideoBackends/D3D12/PerfQuery.cpp @@ -98,7 +98,7 @@ u32 PerfQuery::GetQueryResult(PerfQueryType type) else if (type == PQ_EFB_COPY_CLOCKS) result = m_results[PQG_EFB_COPY_CLOCKS]; - return result / 4; + return result; } void PerfQuery::FlushOne() @@ -126,6 +126,8 @@ void PerfQuery::FlushOne() m_query_readback_buffer->Unmap(0, &write_range); // NOTE: Reported pixel metrics should be referenced to native resolution + // TODO: Dropping the lower 2 bits from this count should be closer to actual + // hardware behavior when drawing triangles. m_results[entry.query_type] += (u32)(result * EFB_WIDTH / g_renderer->GetTargetWidth() * EFB_HEIGHT / g_renderer->GetTargetHeight()); @@ -179,6 +181,8 @@ void PerfQuery::FlushResults() sizeof(UINT64)); // NOTE: Reported pixel metrics should be referenced to native resolution + // TODO: Dropping the lower 2 bits from this count should be closer to actual + // hardware behavior when drawing triangles. m_results[entry.query_type] += (u32)(result * EFB_WIDTH / g_renderer->GetTargetWidth() * EFB_HEIGHT / g_renderer->GetTargetHeight()); diff --git a/Source/Core/VideoBackends/OGL/PerfQuery.cpp b/Source/Core/VideoBackends/OGL/PerfQuery.cpp index cbe4ff7056..b3b5df2265 100644 --- a/Source/Core/VideoBackends/OGL/PerfQuery.cpp +++ b/Source/Core/VideoBackends/OGL/PerfQuery.cpp @@ -11,6 +11,7 @@ #include "VideoBackends/OGL/PerfQuery.h" #include "VideoCommon/RenderBase.h" +#include "VideoCommon/VideoConfig.h" namespace OGL { @@ -79,7 +80,7 @@ u32 PerfQuery::GetQueryResult(PerfQueryType type) result = m_results[PQG_EFB_COPY_CLOCKS]; } - return result / 4; + return result; } // Implementations @@ -155,8 +156,16 @@ void PerfQueryGL::FlushOne() glGetQueryObjectuiv(entry.query_id, GL_QUERY_RESULT, &result); // NOTE: Reported pixel metrics should be referenced to native resolution - m_results[entry.query_type] += (u64)result * EFB_WIDTH / g_renderer->GetTargetWidth() * - EFB_HEIGHT / g_renderer->GetTargetHeight(); + // TODO: Dropping the lower 2 bits from this count should be closer to actual + // hardware behavior when drawing triangles. + result = static_cast(result) * EFB_WIDTH * EFB_HEIGHT / + (g_renderer->GetTargetWidth() * g_renderer->GetTargetHeight()); + + // Adjust for multisampling + if (g_ActiveConfig.iMultisamples > 1) + result /= g_ActiveConfig.iMultisamples; + + m_results[entry.query_type] += result; m_query_read_pos = (m_query_read_pos + 1) % m_query_buffer.size(); --m_query_count; @@ -241,8 +250,10 @@ void PerfQueryGLESNV::FlushOne() glGetOcclusionQueryuivNV(entry.query_id, GL_OCCLUSION_TEST_RESULT_HP, &result); // NOTE: Reported pixel metrics should be referenced to native resolution - m_results[entry.query_type] += (u64)result * EFB_WIDTH / g_renderer->GetTargetWidth() * - EFB_HEIGHT / g_renderer->GetTargetHeight(); + // TODO: Dropping the lower 2 bits from this count should be closer to actual + // hardware behavior when drawing triangles. + m_results[entry.query_type] += static_cast(result) * EFB_WIDTH * EFB_HEIGHT / + (g_renderer->GetTargetWidth() * g_renderer->GetTargetHeight()); m_query_read_pos = (m_query_read_pos + 1) % m_query_buffer.size(); --m_query_count;