rhi: Keep track of pipeline creation times

Make our QRhiMemAllocStats struct a bit more generic, drop the memory
allocation part in the naming, and use the same getter and struct for
reporting some important timings. (we are free to rename for now, there
are no users in other modules yet)

The time spent in graphics (or compute) pipeline creation has a special
relevance in particular with the modern APIs (as it is the single
biggest potentially time consuming blocking operation), but also highly
interesting with others like D3D11 simply because that's where we do the
expensive source-to-intermediate compilation is HLSL source is provided.
In order to see the effects of the various caching mechanisms (of which
there can be confusingly many, on multiple levels), the ability to see
how much time we spent on pipeline creation e.g. until we render the
first view of an application can be pretty essential.

Task-number: QTBUG-103802
Change-Id: I85dd056a39db7e6b25fb1f9d02e4c94298d22b41
Reviewed-by: Andy Nichols <andy.nichols@qt.io>

13 files changed, 110 insertions, 38 deletions

diff --git a/src/gui/rhi/qrhi.cpp b/src/gui/rhi/qrhi.cpp
index 402902e24f..e0194e4f58 100644
--- a/src/gui/rhi/qrhi.cpp
+++ b/src/gui/rhi/qrhi.cpp
@@ -6962,7 +6962,7 @@ void QRhi::setPipelineCacheData(const QByteArray &data)
 }
 
 /*!
-    \struct QRhiMemAllocStats
+    \struct QRhiStats
     \internal
     \inmodule QtGui
 
@@ -6970,10 +6970,12 @@ void QRhi::setPipelineCacheData(const QByteArray &data)
  */
 
 #ifndef QT_NO_DEBUG_STREAM
-QDebug operator<<(QDebug dbg, const QRhiMemAllocStats &info)
+QDebug operator<<(QDebug dbg, const QRhiStats &info)
 {
     QDebugStateSaver saver(dbg);
-    dbg.nospace() << "QRhiMemAllocStats(blockCount=" << info.blockCount
+    dbg.nospace() << "QRhiStats("
+                  << "totalPipelineCreationTime=" << info.totalPipelineCreationTime
+                  << " blockCount=" << info.blockCount
                   << " allocCount=" << info.allocCount
                   << " usedBytes=" << info.usedBytes
                   << " unusedBytes=" << info.unusedBytes
@@ -6983,21 +6985,46 @@ QDebug operator<<(QDebug dbg, const QRhiMemAllocStats &info)
 #endif
 
 /*!
-    Gathers and returns some statistics about the memory allocation of graphics
-    resources. Only supported with some backends. With graphics APIs where
-    there is no lower level control over resource memory allocations, this will
-    never be supported and all fields in the results are 0.
+    Gathers and returns statistics about the timings and allocations of
+    graphics resources.
 
-    With Vulkan, the values are valid always, and are queried from the
-    underlying memory allocator library. This gives an insight into the memory
-    requirements of the active buffers and textures.
+    Data about memory allocations is only available with some backends, where
+    such operations are under Qt's control. With graphics APIs where there is
+    no lower level control over resource memory allocations, this will never be
+    supported and all relevant fields in the results are 0.
 
-    \note Gathering the data may not be free, and therefore the function should
-    not be called at a high frequency.
+    With Vulkan in particular, the values are valid always, and are queried
+    from the underlying memory allocator library. This gives an insight into
+    the memory requirements of the active buffers and textures.
+
+    \warning Gathering some of the data may be an expensive operation, and
+    therefore the function must not be called at a high frequency.
+
+    Additional data, such as the total time in milliseconds spent in graphics
+    and compute pipeline creation (which usually involves shader compilation or
+    cache lookups, and potentially expensive processing) is available with most
+    backends.
+
+    \note The elapsed times for operations such as pipeline creation may be
+    affected by various factors. The results should not be compared between
+    different backends since the concept of "pipelines" and what exactly
+    happens under the hood during, for instance, a call to
+    QRhiGraphicsPipeline::create(), differ greatly between graphics APIs and
+    their implementations.
+
+    \note Additionally, many drivers will likely employ various caching
+    strategies for shaders, programs, pipelines. (independently of Qt's own
+    similar facilities, such as setPipelineCacheData() or the OpenGL-specific
+    program binary disk cache). Because such internal behavior is transparent
+    to the API client, Qt and QRhi have no knowledge or control over the exact
+    caching strategy, persistency, invalidation of the cached data, etc. When
+    reading timings, such as the time spent on pipeline creation, the potential
+    presence and unspecified behavior of driver-level caching mechanisms should
+    be kept in mind.
  */
-QRhiMemAllocStats QRhi::graphicsMemoryAllocationStatistics() const
+QRhiStats QRhi::statistics() const
 {
-    return d->graphicsMemoryAllocationStatistics();
+    return d->statistics();
 }
 
 /*!
diff --git a/src/gui/rhi/qrhi_p.h b/src/gui/rhi/qrhi_p.h
index 6e95e4542b..053049eaf8 100644
--- a/src/gui/rhi/qrhi_p.h
+++ b/src/gui/rhi/qrhi_p.h
@@ -1583,18 +1583,19 @@ Q_DECLARE_TYPEINFO(QRhiDriverInfo, Q_RELOCATABLE_TYPE);
 Q_GUI_EXPORT QDebug operator<<(QDebug, const QRhiDriverInfo &);
 #endif
 
-struct Q_GUI_EXPORT QRhiMemAllocStats
+struct Q_GUI_EXPORT QRhiStats
 {
+    qint64 totalPipelineCreationTime = 0;
     quint32 blockCount = 0;
     quint32 allocCount = 0;
     quint64 usedBytes = 0;
     quint64 unusedBytes = 0;
 };
 
-Q_DECLARE_TYPEINFO(QRhiMemAllocStats, Q_RELOCATABLE_TYPE);
+Q_DECLARE_TYPEINFO(QRhiStats, Q_RELOCATABLE_TYPE);
 
 #ifndef QT_NO_DEBUG_STREAM
-Q_GUI_EXPORT QDebug operator<<(QDebug, const QRhiMemAllocStats &);
+Q_GUI_EXPORT QDebug operator<<(QDebug, const QRhiStats &);
 #endif
 
 struct Q_GUI_EXPORT QRhiInitParams
@@ -1795,7 +1796,7 @@ public:
     QByteArray pipelineCacheData();
     void setPipelineCacheData(const QByteArray &data);
 
-    QRhiMemAllocStats graphicsMemoryAllocationStatistics() const;
+    QRhiStats statistics() const;
 
 protected:
     QRhi();
diff --git a/src/gui/rhi/qrhi_p_p.h b/src/gui/rhi/qrhi_p_p.h
index 2a0cb7d631..516536575c 100644
--- a/src/gui/rhi/qrhi_p_p.h
+++ b/src/gui/rhi/qrhi_p_p.h
@@ -18,6 +18,7 @@
 #include "qrhi_p.h"
 #include <QBitArray>
 #include <QAtomicInt>
+#include <QElapsedTimer>
 #include <QLoggingCategory>
 #include <QtCore/qset.h>
 #include <QtCore/qvarlengtharray.h>
@@ -131,7 +132,7 @@ public:
     virtual int resourceLimit(QRhi::ResourceLimit limit) const = 0;
     virtual const QRhiNativeHandles *nativeHandles() = 0;
     virtual QRhiDriverInfo driverInfo() const = 0;
-    virtual QRhiMemAllocStats graphicsMemoryAllocationStatistics() = 0;
+    virtual QRhiStats statistics() = 0;
     virtual bool makeThreadLocalNativeContextCurrent() = 0;
     virtual void releaseCachedResources() = 0;
     virtual bool isDeviceLost() const = 0;
@@ -201,6 +202,21 @@ public:
         return (quint32(implType) << 24) | ver;
     }
 
+    void pipelineCreationStart()
+    {
+        pipelineCreationTimer.start();
+    }
+
+    void pipelineCreationEnd()
+    {
+        accumulatedPipelineCreationTime += pipelineCreationTimer.elapsed();
+    }
+
+    qint64 totalPipelineCreationTime() const
+    {
+        return accumulatedPipelineCreationTime;
+    }
+
     QRhi *q;
 
     static const int MAX_SHADER_CACHE_ENTRIES = 128;
@@ -219,6 +235,8 @@ private:
     QSet<QRhiResource *> pendingDeleteResources;
     QVarLengthArray<QRhi::CleanupCallback, 4> cleanupCallbacks;
     QVarLengthArray<QRhi::GpuFrameTimeCallback, 4> gpuFrameTimeCallbacks;
+    QElapsedTimer pipelineCreationTimer;
+    qint64 accumulatedPipelineCreationTime = 0;
 
     friend class QRhi;
     friend class QRhiResourceUpdateBatchPrivate;
diff --git a/src/gui/rhi/qrhid3d11.cpp b/src/gui/rhi/qrhid3d11.cpp
index 0ac92e2266..fddef52cee 100644
--- a/src/gui/rhi/qrhid3d11.cpp
+++ b/src/gui/rhi/qrhid3d11.cpp
@@ -615,9 +615,11 @@ QRhiDriverInfo QRhiD3D11::driverInfo() const
     return driverInfoStruct;
 }
 
-QRhiMemAllocStats QRhiD3D11::graphicsMemoryAllocationStatistics()
+QRhiStats QRhiD3D11::statistics()
 {
-    return {};
+    QRhiStats result;
+    result.totalPipelineCreationTime = totalPipelineCreationTime();
+    return result;
 }
 
 bool QRhiD3D11::makeThreadLocalNativeContextCurrent()
@@ -4252,6 +4254,7 @@ bool QD3D11GraphicsPipeline::create()
         destroy();
 
     QRHI_RES_RHI(QRhiD3D11);
+    rhiD->pipelineCreationStart();
     if (!rhiD->sanityCheckGraphicsPipeline(this))
         return false;
 
@@ -4438,6 +4441,7 @@ bool QD3D11GraphicsPipeline::create()
         } // else leave inputLayout set to nullptr; that's valid and it avoids a debug layer warning about an input layout with 0 elements
     }
 
+    rhiD->pipelineCreationEnd();
     generation += 1;
     rhiD->registerResource(this);
     return true;
@@ -4473,6 +4477,7 @@ bool QD3D11ComputePipeline::create()
         destroy();
 
     QRHI_RES_RHI(QRhiD3D11);
+    rhiD->pipelineCreationStart();
 
     auto cacheIt = rhiD->m_shaderCache.constFind(m_shaderStage);
     if (cacheIt != rhiD->m_shaderCache.constEnd()) {
@@ -4508,6 +4513,7 @@ bool QD3D11ComputePipeline::create()
 
     cs.shader->AddRef();
 
+    rhiD->pipelineCreationEnd();
     generation += 1;
     rhiD->registerResource(this);
     return true;
diff --git a/src/gui/rhi/qrhid3d11_p_p.h b/src/gui/rhi/qrhid3d11_p_p.h
index 76a8337f16..3dcaf20928 100644
--- a/src/gui/rhi/qrhid3d11_p_p.h
+++ b/src/gui/rhi/qrhid3d11_p_p.h
@@ -656,7 +656,7 @@ public:
     int resourceLimit(QRhi::ResourceLimit limit) const override;
     const QRhiNativeHandles *nativeHandles() override;
     QRhiDriverInfo driverInfo() const override;
-    QRhiMemAllocStats graphicsMemoryAllocationStatistics() override;
+    QRhiStats statistics() override;
     bool makeThreadLocalNativeContextCurrent() override;
     void releaseCachedResources() override;
     bool isDeviceLost() const override;
diff --git a/src/gui/rhi/qrhigles2.cpp b/src/gui/rhi/qrhigles2.cpp
index 13a7b997ea..6b033f1ed2 100644
--- a/src/gui/rhi/qrhigles2.cpp
+++ b/src/gui/rhi/qrhigles2.cpp
@@ -1302,9 +1302,11 @@ QRhiDriverInfo QRhiGles2::driverInfo() const
     return driverInfoStruct;
 }
 
-QRhiMemAllocStats QRhiGles2::graphicsMemoryAllocationStatistics()
+QRhiStats QRhiGles2::statistics()
 {
-    return {};
+    QRhiStats result;
+    result.totalPipelineCreationTime = totalPipelineCreationTime();
+    return result;
 }
 
 bool QRhiGles2::makeThreadLocalNativeContextCurrent()
@@ -5454,6 +5456,7 @@ bool QGles2GraphicsPipeline::create()
     if (!rhiD->ensureContext())
         return false;
 
+    rhiD->pipelineCreationStart();
     if (!rhiD->sanityCheckGraphicsPipeline(this))
         return false;
 
@@ -5574,6 +5577,7 @@ bool QGles2GraphicsPipeline::create()
     currentSrb = nullptr;
     currentSrbGeneration = 0;
 
+    rhiD->pipelineCreationEnd();
     generation += 1;
     rhiD->registerResource(this);
     return true;
@@ -5620,6 +5624,8 @@ bool QGles2ComputePipeline::create()
     if (!rhiD->ensureContext())
         return false;
 
+    rhiD->pipelineCreationStart();
+
     const QShaderDescription csDesc = m_shaderStage.shader().description();
     QShader::SeparateToCombinedImageSamplerMappingList csSamplerMappingList;
     QShaderVersion shaderVersion;
@@ -5675,6 +5681,7 @@ bool QGles2ComputePipeline::create()
     currentSrb = nullptr;
     currentSrbGeneration = 0;
 
+    rhiD->pipelineCreationEnd();
     generation += 1;
     rhiD->registerResource(this);
     return true;
diff --git a/src/gui/rhi/qrhigles2_p_p.h b/src/gui/rhi/qrhigles2_p_p.h
index 64a22c83c5..7a1576f0c7 100644
--- a/src/gui/rhi/qrhigles2_p_p.h
+++ b/src/gui/rhi/qrhigles2_p_p.h
@@ -807,7 +807,7 @@ public:
     int resourceLimit(QRhi::ResourceLimit limit) const override;
     const QRhiNativeHandles *nativeHandles() override;
     QRhiDriverInfo driverInfo() const override;
-    QRhiMemAllocStats graphicsMemoryAllocationStatistics() override;
+    QRhiStats statistics() override;
     bool makeThreadLocalNativeContextCurrent() override;
     void releaseCachedResources() override;
     bool isDeviceLost() const override;
diff --git a/src/gui/rhi/qrhimetal.mm b/src/gui/rhi/qrhimetal.mm
index 6e469f5b6d..1a9ae651a3 100644
--- a/src/gui/rhi/qrhimetal.mm
+++ b/src/gui/rhi/qrhimetal.mm
@@ -658,9 +658,11 @@ QRhiDriverInfo QRhiMetal::driverInfo() const
     return driverInfoStruct;
 }
 
-QRhiMemAllocStats QRhiMetal::graphicsMemoryAllocationStatistics()
+QRhiStats QRhiMetal::statistics()
 {
-    return {};
+    QRhiStats result;
+    result.totalPipelineCreationTime = totalPipelineCreationTime();
+    return result;
 }
 
 bool QRhiMetal::makeThreadLocalNativeContextCurrent()
@@ -3586,6 +3588,7 @@ bool QMetalGraphicsPipeline::create()
         destroy();
 
     QRHI_RES_RHI(QRhiMetal);
+    rhiD->pipelineCreationStart();
     if (!rhiD->sanityCheckGraphicsPipeline(this))
         return false;
 
@@ -3772,6 +3775,7 @@ bool QMetalGraphicsPipeline::create()
     d->depthBias = float(m_depthBias);
     d->slopeScaledDepthBias = m_slopeScaledDepthBias;
 
+    rhiD->pipelineCreationEnd();
     lastActiveFrameSlot = -1;
     generation += 1;
     rhiD->registerResource(this);
@@ -3811,6 +3815,7 @@ bool QMetalComputePipeline::create()
         destroy();
 
     QRHI_RES_RHI(QRhiMetal);
+    rhiD->pipelineCreationStart();
 
     auto cacheIt = rhiD->d->shaderCache.constFind(m_shaderStage);
     if (cacheIt != rhiD->d->shaderCache.constEnd()) {
@@ -3858,6 +3863,7 @@ bool QMetalComputePipeline::create()
         return false;
     }
 
+    rhiD->pipelineCreationEnd();
     lastActiveFrameSlot = -1;
     generation += 1;
     rhiD->registerResource(this);
diff --git a/src/gui/rhi/qrhimetal_p_p.h b/src/gui/rhi/qrhimetal_p_p.h
index ef9b9c93ca..6ac91159e9 100644
--- a/src/gui/rhi/qrhimetal_p_p.h
+++ b/src/gui/rhi/qrhimetal_p_p.h
@@ -417,7 +417,7 @@ public:
     int resourceLimit(QRhi::ResourceLimit limit) const override;
     const QRhiNativeHandles *nativeHandles() override;
     QRhiDriverInfo driverInfo() const override;
-    QRhiMemAllocStats graphicsMemoryAllocationStatistics() override;
+    QRhiStats statistics() override;
     bool makeThreadLocalNativeContextCurrent() override;
     void releaseCachedResources() override;
     bool isDeviceLost() const override;
diff --git a/src/gui/rhi/qrhinull.cpp b/src/gui/rhi/qrhinull.cpp
index 2a3412a69c..bd66f33c83 100644
--- a/src/gui/rhi/qrhinull.cpp
+++ b/src/gui/rhi/qrhinull.cpp
@@ -151,7 +151,7 @@ QRhiDriverInfo QRhiNull::driverInfo() const
     return info;
 }
 
-QRhiMemAllocStats QRhiNull::graphicsMemoryAllocationStatistics()
+QRhiStats QRhiNull::statistics()
 {
     return {};
 }
diff --git a/src/gui/rhi/qrhinull_p_p.h b/src/gui/rhi/qrhinull_p_p.h
index 7c2703cada..a6ec4752c5 100644
--- a/src/gui/rhi/qrhinull_p_p.h
+++ b/src/gui/rhi/qrhinull_p_p.h
@@ -273,7 +273,7 @@ public:
     int resourceLimit(QRhi::ResourceLimit limit) const override;
     const QRhiNativeHandles *nativeHandles() override;
     QRhiDriverInfo driverInfo() const override;
-    QRhiMemAllocStats graphicsMemoryAllocationStatistics() override;
+    QRhiStats statistics() override;
     bool makeThreadLocalNativeContextCurrent() override;
     void releaseCachedResources() override;
     bool isDeviceLost() const override;
diff --git a/src/gui/rhi/qrhivulkan.cpp b/src/gui/rhi/qrhivulkan.cpp
index 26fc3e05ed..f1a700f926 100644
--- a/src/gui/rhi/qrhivulkan.cpp
+++ b/src/gui/rhi/qrhivulkan.cpp
@@ -4353,16 +4353,19 @@ QRhiDriverInfo QRhiVulkan::driverInfo() const
     return driverInfoStruct;
 }
 
-QRhiMemAllocStats QRhiVulkan::graphicsMemoryAllocationStatistics()
+QRhiStats QRhiVulkan::statistics()
 {
     VmaStats stats;
     vmaCalculateStats(toVmaAllocator(allocator), &stats);
-    return {
-        stats.total.blockCount,
-        stats.total.allocationCount,
-        stats.total.usedBytes,
-        stats.total.unusedBytes
-    };
+
+    QRhiStats result;
+    result.totalPipelineCreationTime = totalPipelineCreationTime();
+    result.blockCount = stats.total.blockCount;
+    result.allocCount = stats.total.allocationCount;
+    result.usedBytes = stats.total.usedBytes;
+    result.unusedBytes = stats.total.unusedBytes;
+
+    return result;
 }
 
 bool QRhiVulkan::makeThreadLocalNativeContextCurrent()
@@ -6844,6 +6847,7 @@ bool QVkGraphicsPipeline::create()
         destroy();
 
     QRHI_RES_RHI(QRhiVulkan);
+    rhiD->pipelineCreationStart();
     if (!rhiD->sanityCheckGraphicsPipeline(this))
         return false;
 
@@ -7081,6 +7085,7 @@ bool QVkGraphicsPipeline::create()
         return false;
     }
 
+    rhiD->pipelineCreationEnd();
     lastActiveFrameSlot = -1;
     generation += 1;
     rhiD->registerResource(this);
@@ -7125,6 +7130,7 @@ bool QVkComputePipeline::create()
         destroy();
 
     QRHI_RES_RHI(QRhiVulkan);
+    rhiD->pipelineCreationStart();
     if (!rhiD->ensurePipelineCache())
         return false;
 
@@ -7176,6 +7182,7 @@ bool QVkComputePipeline::create()
         return false;
     }
 
+    rhiD->pipelineCreationEnd();
     lastActiveFrameSlot = -1;
     generation += 1;
     rhiD->registerResource(this);
diff --git a/src/gui/rhi/qrhivulkan_p_p.h b/src/gui/rhi/qrhivulkan_p_p.h
index 0a035e3f34..ad3c6bc05b 100644
--- a/src/gui/rhi/qrhivulkan_p_p.h
+++ b/src/gui/rhi/qrhivulkan_p_p.h
@@ -730,7 +730,7 @@ public:
     int resourceLimit(QRhi::ResourceLimit limit) const override;
     const QRhiNativeHandles *nativeHandles() override;
     QRhiDriverInfo driverInfo() const override;
-    QRhiMemAllocStats graphicsMemoryAllocationStatistics() override;
+    QRhiStats statistics() override;
     bool makeThreadLocalNativeContextCurrent() override;
     void releaseCachedResources() override;
     bool isDeviceLost() const override;