diff options
Diffstat (limited to 'src/gui/rhi')
| -rw-r--r-- | src/gui/rhi/qrhi.cpp | 905 | ||||
| -rw-r--r-- | src/gui/rhi/qrhi.h | 72 | ||||
| -rw-r--r-- | src/gui/rhi/qrhi_p.h | 14 | ||||
| -rw-r--r-- | src/gui/rhi/qrhi_platform.h | 8 | ||||
| -rw-r--r-- | src/gui/rhi/qrhid3d11.cpp | 544 | ||||
| -rw-r--r-- | src/gui/rhi/qrhid3d11_p.h | 46 | ||||
| -rw-r--r-- | src/gui/rhi/qrhid3d12.cpp | 1114 | ||||
| -rw-r--r-- | src/gui/rhi/qrhid3d12_p.h | 122 | ||||
| -rw-r--r-- | src/gui/rhi/qrhid3dhelpers.cpp | 172 | ||||
| -rw-r--r-- | src/gui/rhi/qrhid3dhelpers_p.h | 53 | ||||
| -rw-r--r-- | src/gui/rhi/qrhigles2.cpp | 819 | ||||
| -rw-r--r-- | src/gui/rhi/qrhigles2_p.h | 81 | ||||
| -rw-r--r-- | src/gui/rhi/qrhimetal.mm | 541 | ||||
| -rw-r--r-- | src/gui/rhi/qrhimetal_p.h | 3 | ||||
| -rw-r--r-- | src/gui/rhi/qrhinull.cpp | 4 | ||||
| -rw-r--r-- | src/gui/rhi/qrhivulkan.cpp | 785 | ||||
| -rw-r--r-- | src/gui/rhi/qrhivulkan_p.h | 44 | ||||
| -rw-r--r-- | src/gui/rhi/qrhivulkanext_p.h | 48 | ||||
| -rw-r--r-- | src/gui/rhi/qshader.cpp | 73 | ||||
| -rw-r--r-- | src/gui/rhi/qshader.h | 4 | ||||
| -rw-r--r-- | src/gui/rhi/qshader_p.h | 3 | ||||
| -rw-r--r-- | src/gui/rhi/qshaderdescription.cpp | 12 |
22 files changed, 4229 insertions, 1238 deletions
diff --git a/src/gui/rhi/qrhi.cpp b/src/gui/rhi/qrhi.cpp index 58bbc34941..a39709c726 100644 --- a/src/gui/rhi/qrhi.cpp +++ b/src/gui/rhi/qrhi.cpp @@ -16,7 +16,7 @@ #include "qrhid3d11_p.h" #include "qrhid3d12_p.h" #endif -#if defined(Q_OS_MACOS) || defined(Q_OS_IOS) +#if QT_CONFIG(metal) #include "qrhimetal_p.h" #endif @@ -29,7 +29,8 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") /*! \class QRhi \ingroup painting-3D - \inmodule QtGui + \inmodule QtGuiPrivate + \inheaderfile rhi/qrhi.h \since 6.6 \brief Accelerated 2D/3D graphics API abstraction. @@ -71,19 +72,22 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") builds on QOpenGLContext, QOpenGLFunctions, and the related cross-platform infrastructure of the Qt GUI module. - \li Direct3D 11.1 or newer, with Shader Model 5.0 or newer . When the D3D + \li Direct3D 11.1 or newer, with Shader Model 5.0 or newer. When the D3D runtime has no support for 11.1 features or Shader Model 5.0, initialization using an accelerated graphics device will fail, but using the \l{https://learn.microsoft.com/en-us/windows/win32/direct3darticles/directx-warp}{software adapter} is still an option. - \li Direct3D 12.0 or newer. The D3D12 device is by default created with - specifying a minimum feature level of \c{D3D_FEATURE_LEVEL_11_0}. + \li Direct3D 12 on Windows 10 version 1703 and newer, with Shader Model 5.0 + or newer. Qt requires ID3D12Device2 to be present, hence the requirement + for at least version 1703 of Windows 10. The D3D12 device is by default + created with specifying a minimum feature level of + \c{D3D_FEATURE_LEVEL_11_0}. \li Metal 1.2 or newer. - \li Vulkan 1.0 or newer , optionally utilizing some Vulkan 1.1 level + \li Vulkan 1.0 or newer, optionally utilizing some Vulkan 1.1 level features. \li Null, a "dummy" backend that issues no graphics calls at all. @@ -650,7 +654,7 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") mechanism because the cost of maintaining the related data structures is not insignificant with some backends. With Vulkan this feature maps directly to VkPipelineCache, vkGetPipelineCacheData and - VkPipelineCacheCreateInfo::pInitialData. With D3D11 there is no real + VkPipelineCacheCreateInfo::pInitialData. With Direct3D 11 there is no real pipline cache, but the results of HLSL->DXBC compilations are stored and can be serialized/deserialized via this mechanism. This allows skipping the time consuming D3DCompile() in future runs of the applications for shaders @@ -662,6 +666,17 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") mechanisms for shader/program binaries provided by Qt. Writing to those may get disabled whenever this flag is set since storing program binaries to multiple caches is not sensible. + + \value SuppressSmokeTestWarnings Indicates that, with backends where this + is relevant, certain, non-fatal QRhi::create() failures should not + produce qWarning() calls. For example, with D3D11, passing this flag + makes a number of warning messages (that appear due to QRhi::create() + failing) to become categorized debug prints instead under the commonly used + \c{qt.rhi.general} logging category. This can be used by engines, such as + Qt Quick, that feature fallback logic, i.e. they retry calling create() + with a different set of flags (such as, \l PreferSoftwareRenderer), in order + to hide the unconditional warnings from the output that would be printed + when the first create() attempt had failed. */ /*! @@ -699,9 +714,12 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") QRhiCommandBuffer::debugMarkBegin()) are supported. \value Timestamps Indicates that command buffer timestamps are supported. - Relevant for QRhiCommandBuffer::lastCompletedGpuTime(). Can be expected to - be supported on Metal, Vulkan, and Direct 3D, assuming the underlying - implementation supports timestamp queries or similar. + Relevant for QRhiCommandBuffer::lastCompletedGpuTime(). This can be + expected to be supported on Metal, Vulkan, Direct 3D 11 and 12, and OpenGL + contexts of version 3.3 or newer. However, with some of these APIs support + for timestamp queries is technically optional, and therefore it cannot be + guaranteed that this feature is always supported with every implementation + of them. \value Instancing Indicates that instanced drawing is supported. In practice this feature will be unsupported with OpenGL ES 2.0 and OpenGL @@ -771,7 +789,7 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") \value WideLines Indicates that lines with a width other than 1 are supported. When reported as not supported, the line width set on the graphics pipeline state is ignored. This can always be false with some - backends (D3D11, Metal). With Vulkan, the value depends on the + backends (D3D11, D3D12, Metal). With Vulkan, the value depends on the implementation. With OpenGL, wide lines are not supported in core profile contexts. @@ -932,7 +950,7 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") with image load/store. This feature is only available with some backends as it does not map well to all graphics APIs, and it is only meant to provide support for special cases anyhow. In practice the feature can be expected to - be supported with Direct3D 11 and Vulkan. + be supported with Direct3D 11/12 and Vulkan. \value NonFillPolygonMode Indicates that setting a PolygonMode other than the default Fill is supported for QRhiGraphicsPipeline. A common use case @@ -953,8 +971,8 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") When not supported, build() will succeed but just show a warning message and the values of the target attributes will be broken. In practice this feature will be unsupported in some OpenGL ES 2.0 and OpenGL 2.x - implementations. Note that while D3D does support half precision input - attributes, it does not support the half3 type. The D3D backends pass + implementations. Note that while Direct3D 11/12 does support half precision + input attributes, it does not support the half3 type. The D3D backends pass half3 attributes as half4. To ensure cross platform compatibility, half3 inputs should be padded to 8 bytes. @@ -966,6 +984,58 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") \value ThreeDimensionalTextureMipmaps Indicates that generating 3D texture mipmaps are supported. In practice this feature will be unsupported with Direct 3D 12. + + \value MultiView Indicates that multiview, see e.g. + \l{https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VK_KHR_multiview.html}{VK_KHR_multiview} + is supported. With OpenGL ES 2.0, Direct 3D 11, and OpenGL (ES) + implementations without \c{GL_OVR_multiview2} this feature will not be + supported. With Vulkan 1.1 and newer, and Direct 3D 12 multiview is + typically supported. When reported as supported, creating a + QRhiTextureRenderTarget with a QRhiColorAttachment that references a texture + array and has \l{QRhiColorAttachment::setMultiViewCount()}{multiViewCount} + set enables recording a render pass that uses multiview rendering. In addition, + any QRhiGraphicsPipeline used in that render pass must have + \l{QRhiGraphicsPipeline::setMultiViewCount()}{the same view count set}. Note that + multiview is only available in combination with 2D texture arrays. It cannot + be used to optimize the rendering into individual textures (e.g. two, for + the left and right eyes). Rather, the target of a multiview render pass is + always a texture array, automatically rendering to the layer (array element) + corresponding to each view. Therefore this feature implies \l TextureArrays + as well. Multiview rendering is not supported in combination with + tessellation or geometry shaders. See QRhiColorAttachment::setMultiViewCount() + for further details on multiview rendering. This enum value has been introduced in Qt 6.7. + + \value TextureViewFormat Indicates that setting a + \l{QRhiTexture::setWriteViewFormat()}{view format} on a QRhiTexture is + effective. When reported as supported, setting the read (sampling) or write + (render target / image load-store) view mode changes the texture's viewing + format. When unsupported, setting a view format has no effect. Note that Qt + has no knowledge or control over format compatibility or resource view rules + in the underlying 3D API and its implementation. Passing in unsuitable, + incompatible formats may lead to errors and unspecified behavior. This is + provided mainly to allow "casting" rendering into a texture created with an + sRGB format to non-sRGB to avoid the unwanted linear->sRGB conversion on + shader writes. Other types of casting may or may not be functional, + depending on the underlying API. Currently implemented for Vulkan and Direct + 3D 12. With D3D12 the feature is available only if + \c CastingFullyTypedFormatSupported is supported, see + \l{https://microsoft.github.io/DirectX-Specs/d3d/RelaxedCasting.html} (and + note that QRhi always uses fully typed formats for textures.) This enum + value has been introduced in Qt 6.8. + + \value ResolveDepthStencil Indicates that resolving a multisample depth or + depth-stencil texture is supported. Otherwise, + \l{QRhiTextureRenderTargetDescription::setDepthResolveTexture()}{setting a + depth resolve texture} is not functional and must be avoided. Direct 3D 11 + and 12 have no support for resolving depth/depth-stencil formats, and + therefore this feature will never be supported with those. Vulkan 1.0 has no + API to request resolving a depth-stencil attachment. Therefore, with Vulkan + this feature will only be supported with Vulkan 1.2 and up, and on 1.1 + implementations with the appropriate extensions present. This feature is + provided for the rare case when resolving into a non-multisample depth + texture becomes necessary, for example when rendering into an + OpenXR-provided depth texture (XR_KHR_composition_layer_depth). This enum + value has been introduced in Qt 6.8. */ /*! @@ -1083,7 +1153,7 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") Contains fields that are relevant to all backends. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -1093,7 +1163,7 @@ Q_LOGGING_CATEGORY(QRHI_LOG_INFO, "qt.rhi.general") \since 6.6 \brief Specifies clear values for a depth or stencil buffer. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -1188,7 +1258,7 @@ QDebug operator<<(QDebug dbg, const QRhiDepthStencilClearValue &v) // ... \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiCommandBuffer::setViewport(), QRhi::clipSpaceCorrMatrix(), QRhiScissor @@ -1308,7 +1378,7 @@ QDebug operator<<(QDebug dbg, const QRhiViewport &v) appropriate. Therefore, any rendering logic targeting OpenGL can feed scissor rectangles into QRhiScissor as-is, without any adaptation. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiCommandBuffer::setScissor(), QRhiViewport @@ -1436,7 +1506,7 @@ QDebug operator<<(QDebug dbg, const QRhiScissor &s) \note the stride must always be a multiple of 4. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiCommandBuffer::setVertexInput() @@ -1605,7 +1675,7 @@ QDebug operator<<(QDebug dbg, const QRhiVertexInputBinding &b) cb->setVertexInput(0, 1, &vbufBinding); \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiCommandBuffer::setVertexInput() @@ -1634,12 +1704,22 @@ QDebug operator<<(QDebug dbg, const QRhiVertexInputBinding &b) \value Half3 Three component half precision (16 bit) float vector \value Half2 Two component half precision (16 bit) float vector \value Half Half precision (16 bit) float + \value UShort4 Four component unsigned short (16 bit) integer vector + \value UShort3 Three component unsigned short (16 bit) integer vector + \value UShort2 Two component unsigned short (16 bit) integer vector + \value UShort Unsigned short (16 bit) integer + \value SShort4 Four component signed short (16 bit) integer vector + \value SShort3 Three component signed short (16 bit) integer vector + \value SShort2 Two component signed short (16 bit) integer vector + \value SShort Signed short (16 bit) integer \note Support for half precision floating point attributes is indicated at - run time by the QRhi::Feature::HalfAttributes feature flag. Note that D3D - supports half input attributes, but does not support the Half3 type. The - D3D backends pass through Half3 as Half4. To ensure cross platform - compatibility, Half3 inputs should be padded to 8 bytes. + run time by the QRhi::Feature::HalfAttributes feature flag. + + \note Direct3D 11/12 supports 16 bit input attributes, but does not support + the Half3, UShort3 or SShort3 types. The D3D backends pass through Half3 as + Half4, UShort3 as UShort4, and SShort3 as SShort4. To ensure cross platform + compatibility, 16 bit inputs should be padded to 8 bytes. */ /*! @@ -1851,6 +1931,24 @@ quint32 QRhiImplementation::byteSizePerVertexForVertexInputFormat(QRhiVertexInpu case QRhiVertexInputAttribute::Half: return sizeof(qfloat16); + case QRhiVertexInputAttribute::UShort4: + return 4 * sizeof(quint16); + case QRhiVertexInputAttribute::UShort3: + return 4 * sizeof(quint16); // ivec3 still takes 8 bytes + case QRhiVertexInputAttribute::UShort2: + return 2 * sizeof(quint16); + case QRhiVertexInputAttribute::UShort: + return sizeof(quint16); + + case QRhiVertexInputAttribute::SShort4: + return 4 * sizeof(qint16); + case QRhiVertexInputAttribute::SShort3: + return 4 * sizeof(qint16); // uvec3 still takes 8 bytes + case QRhiVertexInputAttribute::SShort2: + return 2 * sizeof(qint16); + case QRhiVertexInputAttribute::SShort: + return sizeof(qint16); + default: Q_UNREACHABLE_RETURN(1); } @@ -1883,7 +1981,7 @@ quint32 QRhiImplementation::byteSizePerVertexForVertexInputFormat(QRhiVertexInpu }); \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -2006,9 +2104,7 @@ QDebug operator<<(QDebug dbg, const QRhiVertexInputLayout &v) QShader getShader(const QString &name) { QFile f(name); - if (f.open(QIODevice::ReadOnly)) - return QShader::fromSerialized(f.readAll()); - return QShader(); + return f.open(QIODevice::ReadOnly) ? QShader::fromSerialized(f.readAll()) : QShader(); } QShader vs = getShader("material.vert.qsb"); @@ -2019,7 +2115,7 @@ QDebug operator<<(QDebug dbg, const QRhiVertexInputLayout &v) }); \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -2169,7 +2265,7 @@ QDebug operator<<(QDebug dbg, const QRhiShaderStage &s) out at all. This means that the multisample texture() must not be used afterwards with shaders for sampling when resolveTexture() is set. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiTextureRenderTargetDescription @@ -2263,16 +2359,40 @@ QRhiColorAttachment::QRhiColorAttachment(QRhiRenderBuffer *renderBuffer) \nullptr if there is none. Setting a non-null resolve texture is applicable when the attachment - references a multisample, color renderbuffer. (i.e., renderBuffer() is set) - The QRhiTexture in the resolveTexture() is then a regular, 2D, - non-multisample texture with the same size (but a sample count of 1). The - multisample content is automatically resolved into this texture at the end - of each render pass. + references a multisample texture or renderbuffer. The QRhiTexture in the + resolveTexture() is then a non-multisample 2D texture (or texture array) + with the same size (but a sample count of 1). The multisample content is + automatically resolved into this texture at the end of each render pass. */ /*! \fn void QRhiColorAttachment::setResolveTexture(QRhiTexture *tex) + Sets the resolve texture \a tex. + + \a tex is expected to be a 2D texture or a 2D texture array. In either + case, resolving targets a single mip level of a single layer (array + element) of \a tex. The mip level and array layer are specified by + resolveLevel() and resolveLayer(). + + An exception is \l{setMultiViewCount()}{multiview}: when the color + attachment is associated with a texture array and multiview is enabled, the + resolve texture must also be a texture array with sufficient elements for + all views. In this case all elements that correspond to views are resolved + automatically; the behavior is similar to the following pseudo-code: + \badcode + for (i = 0; i < multiViewCount(); ++i) + resolve texture's layer() + i into resolveTexture's resolveLayer() + i + \endcode + + Setting a non-multisample texture to resolve a multisample texture or + renderbuffer automatically at the end of the render pass is often + preferable to working with multisample textures (and not setting a resolve + texture), because it avoids the need for writing dedicated fragment shaders + that work exclusively with multisample textures (\c sampler2DMS, \c + texelFetch, etc.), and rather allows using the same shader as one would if + the attachment's texture was not multisampled to begin with. This comes at + the expense of an additional resource (the non-multisample \a tex). */ /*! @@ -2296,6 +2416,71 @@ QRhiColorAttachment::QRhiColorAttachment(QRhiRenderBuffer *renderBuffer) */ /*! + \fn int QRhiColorAttachment::multiViewCount() const + + \return the currently set number of views. Defaults to 0 which indicates + the render target with this color attachment is not going to be used with + multiview rendering. + + \since 6.7 + */ + +/*! + \fn void QRhiColorAttachment::setMultiViewCount(int count) + + Sets the view \a count. Setting a value larger than 1 indicates that the + render target with this color attachment is going to be used with multiview + rendering. The default value is 0. Values smaller than 2 indicate no + multiview rendering. + + When \a count is set to \c 2 or greater, the color attachment must be + associated with a 2D texture array. layer() and multiViewCount() together + define the range of texture array elements that are targeted during + multiview rendering. + + For example, if \c layer is \c 0 and \c multiViewCount is \c 2, the texture + array must have 2 (or more) elements, and the multiview rendering will + target elements 0 and 1. The \c{gl_ViewIndex} variable in the shaders has a + value of \c 0 or \c 1 then, where view \c 0 corresponds to the texture array + element \c 0, and view \c 1 to the array element \c 1. + + \note Setting a \a count larger than 1, using a texture array as texture(), + and calling \l{QRhiCommandBuffer::beginPass()}{beginPass()} on a + QRhiTextureRenderTarget with this color attachment implies multiview + rendering for the entire render pass. multiViewCount() should not be set + unless multiview rendering is wanted. Multiview cannot be used with texture + types other than 2D texture arrays. (although 3D textures may work, + depending on the graphics API and backend; applications are nonetheless + advised not to rely on that and only use 2D texture arrays as the render + targets of multiview rendering) + + See + \l{https://registry.khronos.org/OpenGL/extensions/OVR/OVR_multiview.txt}{GL_OVR_multiview} + for more details regarding multiview rendering. Do note that Qt requires + \l{https://registry.khronos.org/OpenGL/extensions/OVR/OVR_multiview2.txt}{GL_OVR_multiview2} + as well, when running on OpenGL (ES). + + Multiview rendering is available only when the + \l{QRhi::MultiView}{MultiView} feature is reported as supported from + \l{QRhi::isFeatureSupported()}{isFeatureSupported()}. + + \note For portability, be aware of limitations that exist for multiview + rendering with some of the graphics APIs. It is recommended that multiview + render passes do not rely on any of the features that + \l{https://registry.khronos.org/OpenGL/extensions/OVR/OVR_multiview.txt}{GL_OVR_multiview} + declares as unsupported. The one exception is shader stage outputs other + than \c{gl_Position} depending on \c{gl_ViewIndex}: that can be relied on + (even with OpenGL) because QRhi never reports multiview as supported without + \c{GL_OVR_multiview2} also being present. + + \note Multiview rendering is not supported in combination with tessellation + or geometry shaders, even though some implementations of some graphics APIs + may allow this. + + \since 6.7 + */ + +/*! \class QRhiTextureRenderTargetDescription \inmodule QtGui \since 6.6 @@ -2371,7 +2556,26 @@ QRhiColorAttachment::QRhiColorAttachment(QRhiRenderBuffer *renderBuffer) QRhiTextureRenderTarget *rt = rhi->newTextureRenderTarget({ colorAtt, depthStencil }); \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note when multisample resolving is enabled, the multisample data may not be + written out at all. This means that the multisample texture in a color + attachment must not be used afterwards with shaders for sampling (or other + purposes) whenever a resolve texture is set, since the multisample color + buffer is merely an intermediate storage then that gets no data written back + on some GPU architectures at all. See + \l{QRhiTextureRenderTarget::Flag}{PreserveColorContents} for more details. + + \note When using setDepthTexture(), not setDepthStencilBuffer(), and the + depth (stencil) data is not of interest afterwards, set the + DoNotStoreDepthStencilContents flag on the QRhiTextureRenderTarget. This + allows indicating to the underlying 3D API that the depth/stencil data can + be discarded, leading potentially to better performance with tiled GPU + architectures. When the depth-stencil buffer is a QRhiRenderBuffer (and also + for the multisample color texture, see previous note) this is implicit, but + with a depth (stencil) QRhiTexture the intention needs to be declared + explicitly. By default QRhi assumes that the data is of interest (e.g., the + depth texture is sampled in a shader afterwards). + + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiColorAttachment, QRhiTextureRenderTarget @@ -2464,6 +2668,16 @@ QRhiTextureRenderTargetDescription::QRhiTextureRenderTargetDescription(const QRh \note depthStencilBuffer() and depthTexture() cannot be both set (cannot be non-null at the same time). + + Using a QRhiRenderBuffer over a 2D QRhiTexture as the depth or + depth/stencil buffer is very common, and is the recommended approach for + applications. Using a QRhiTexture, and so setDepthTexture() becomes + relevant if the depth data is meant to be accessed (e.g. sampled in a + shader) afterwards, or when + \l{QRhiColorAttachment::setMultiViewCount()}{multiview rendering} is + involved (because then the depth texture must be a texture array). + + \sa setDepthTexture() */ /*! @@ -2480,6 +2694,49 @@ QRhiTextureRenderTargetDescription::QRhiTextureRenderTargetDescription(const QRh \note depthStencilBuffer() and depthTexture() cannot be both set (cannot be non-null at the same time). + + \a texture can either be a 2D texture or a 2D texture array (when texture + arrays are supported). Specifying a texture array is relevant in particular + with + \l{QRhiColorAttachment::setMultiViewCount()}{multiview rendering}. + + \note If \a texture is a format with a stencil component, such as + \l QRhiTexture::D24S8, it will serve as the stencil buffer as well. + + \sa setDepthStencilBuffer() + */ + +/*! + \fn QRhiTexture *QRhiTextureRenderTargetDescription::depthResolveTexture() const + + \return the texture to which a multisample depth (or depth-stencil) texture + (or texture array) is resolved to. \nullptr if there is none, which is the + most common case. + + \since 6.8 + \sa QRhiColorAttachment::resolveTexture(), depthTexture() + */ + +/*! + \fn void QRhiTextureRenderTargetDescription::setDepthResolveTexture(QRhiTexture *tex) + + Sets the depth (or depth-stencil) resolve texture \a tex. + + \a tex is expected to be a 2D texture or a 2D texture array with a format + matching the texture set via setDepthTexture(). + + \note Resolving depth (or depth-stencil) data is only functional when the + \l ResolveDepthStencil feature is reported as supported at run time. Support + for depth-stencil resolve is not universally available among the graphics + APIs. Designs assuming unconditional availability of depth-stencil resolve + are therefore non-portable, and should be avoided. + + \note As an additional limitation for OpenGL ES in particular, setting a + depth resolve texture may only be functional in combination with + setDepthTexture(), not with setDepthStencilBuffer(). + + \since 6.8 + \sa QRhiColorAttachment::setResolveTexture(), setDepthTexture() */ /*! @@ -2543,7 +2800,7 @@ QRhiTextureRenderTargetDescription::QRhiTextureRenderTargetDescription(const QRh caller is strongly encouraged to call QImage::detach() on the image before passing it to uploadTexture(). - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiTextureUploadDescription @@ -2606,6 +2863,7 @@ QRhiTextureSubresourceUploadDescription::QRhiTextureSubresourceUploadDescription \fn void QRhiTextureSubresourceUploadDescription::setImage(const QImage &image) Sets \a image. + Upon textures loading, the image data will be read as is, with no formats conversions. \note image() and data() cannot be both set at the same time. */ @@ -2698,7 +2956,7 @@ QRhiTextureSubresourceUploadDescription::QRhiTextureSubresourceUploadDescription \brief Describes one layer (face for cubemaps, slice for 3D textures, element for texture arrays) in a texture upload operation. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -2838,7 +3096,7 @@ QRhiTextureUploadEntry::QRhiTextureUploadEntry(int layer, int level, resourceUpdates->uploadTexture(texture, desc); \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiResourceUpdateBatch @@ -2927,7 +3185,7 @@ QRhiTextureUploadDescription::QRhiTextureUploadDescription(std::initializer_list differ, but the size and position must be carefully controlled to avoid out of bounds copies, in which case the behavior is undefined. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -3033,7 +3291,7 @@ QRhiTextureUploadDescription::QRhiTextureUploadDescription(std::initializer_list \note Multisample textures cannot be read back. Readbacks are supported for multisample swapchain buffers however. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -3124,7 +3382,7 @@ QRhiReadbackDescription::QRhiReadbackDescription(QRhiTexture *texture) available. \l format and \l pixelSize are set upon completion together with \l data. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -3162,7 +3420,7 @@ QRhiReadbackDescription::QRhiReadbackDescription(QRhiTexture *texture) \since 6.6 \brief Base class for classes exposing backend-specific collections of native resource objects. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -3172,7 +3430,7 @@ QRhiReadbackDescription::QRhiReadbackDescription(QRhiTexture *texture) \since 6.6 \brief Base class for classes encapsulating native resource objects. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -3353,7 +3611,7 @@ QRhi *QRhiResource::rhi() const \since 6.6 \brief Vertex, index, or uniform (constant) buffer resource. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. A QRhiBuffer encapsulates zero, one, or more native buffer objects (such as @@ -3460,7 +3718,7 @@ QRhi *QRhiResource::rhi() const QRhiResourceUpdateBatch *batch = rhi->nextResourceUpdateBatch(); for (int i = 0; i < N; ++i) { batch->updateDynamicBuffer(ubuf, i * ONE_UBUF_SIZE, 64, matrix.constData()); - updates->updateDynamicBuffer(ubuf, i * ONE_UBUF_SIZE + 64, 4, &opacity); + batch->updateDynamicBuffer(ubuf, i * ONE_UBUF_SIZE + 64, 4, &opacity); } // ... // beginPass(), set pipeline, etc., and then: @@ -3541,10 +3799,13 @@ QRhi *QRhiResource::rhi() const objects array are pointers to a GLuint. With Vulkan, the native handle is a VkBuffer, so the elements of the array are pointers to a VkBuffer. With Direct3D 11 and Metal the elements are pointers to a ID3D11Buffer or - MTLBuffer pointer, respectively. + MTLBuffer pointer, respectively. With Direct3D 12, the elements are + pointers to a ID3D12Resource. \note Pay attention to the fact that the elements are always pointers to the native buffer handle type, even if the native type itself is a pointer. + (so the elements are \c{VkBuffer *} on Vulkan, even though VkBuffer itself + is a pointer on 64-bit architectures). */ /*! @@ -3748,7 +4009,7 @@ void QRhiBuffer::endFullDynamicBufferUpdateForCurrentFrame() means calling setPixelSize() and create() are not necessary for such renderbuffers. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -3973,7 +4234,7 @@ bool QRhiRenderBuffer::createFrom(NativeRenderBuffer src) // continue using texture, fill it with new data \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiResourceUpdateBatch, QRhi, QRhiTextureRenderTarget @@ -4140,9 +4401,10 @@ bool QRhiRenderBuffer::createFrom(NativeRenderBuffer src) \brief 64-bit integer containing the native object handle. With OpenGL, the native handle is a GLuint value, so \c object can then be - cast to a GLuint. With Vulkan, the native handle is a VkImage, so \c - object can be cast to a VkImage. With Direct3D 11 and Metal \c - object contains a ID3D11Texture2D or MTLTexture pointer, respectively. + cast to a GLuint. With Vulkan, the native handle is a VkImage, so \c object + can be cast to a VkImage. With Direct3D 11 and Metal \c object contains a + ID3D11Texture2D or MTLTexture pointer, respectively. With Direct3D 12 + \c object contains a ID3D12Resource pointer. */ /*! @@ -4359,12 +4621,93 @@ void QRhiTexture::setNativeLayout(int layout) */ /*! + \struct QRhiTexture::ViewFormat + \inmodule QtGui + \since 6.8 + \brief Specifies the view format for reading or writing from or to the texture. + + \note This is a RHI API with limited compatibility guarantees, see \l QRhi + for details. + */ + +/*! + \variable QRhiTexture::ViewFormat::format + */ + +/*! + \variable QRhiTexture::ViewFormat::srgb + */ + +/*! + \fn QRhiTexture::ViewFormat QRhiTexture::readViewFormat() const + \since 6.8 + \return the view format used when sampling the texture. When not called, the view + format is assumed to be the same as format(). + */ + +/*! + \fn void QRhiTexture::setReadViewFormat(const ViewFormat &fmt) + \since 6.8 + + Sets the shader resource view format (or the format of the view used for + sampling the texture) to \a fmt. By default the same format (and sRGB-ness) + is used as the texture itself, and in most cases this function does not need + to be called. + + This setting is only taken into account when the \l TextureViewFormat + feature is reported as supported. + + \note This functionality is provided to allow "casting" between + non-sRGB and sRGB in order to get the shader reads perform, or not perform, + the implicit sRGB conversions. Other types of casting may or may not be + functional. + */ + +/*! + \fn QRhiTexture::ViewFormat QRhiTexture::writeViewFormat() const + \since 6.8 + \return the view format used when writing to the texture and when using it + with image load/store. When not called, the view format is assumed to be the + same as format(). + */ + +/*! + \fn void QRhiTexture::setWriteViewFormat(const ViewFormat &fmt) + \since 6.8 + + Sets the render target view format to \a fmt. By default the same format + (and sRGB-ness) is used as the texture itself, and in most cases this + function does not need to be called. + + One common use case for providing a write view format is working with + externally provided textures that, outside of our control, use an sRGB + format with 3D APIs such as Vulkan or Direct 3D, but the rendering engine is + already prepared to handle linearization and conversion to sRGB at the end + of its shading pipeline. In this case what is wanted when rendering into + such a texture is a render target view (e.g. VkImageView) that has the same, + but non-sRGB format. (if e.g. from an OpenXR implementation one gets a + VK_FORMAT_R8G8B8A8_SRGB texture, it is likely that rendering into it should + be done using a VK_FORMAT_R8G8B8A8_UNORM view, if that is what the rendering + engine's pipeline requires; in this example one would call this function + with a ViewFormat that has a format of QRhiTexture::RGBA8 and \c srgb set to + \c false). + + This setting is only taken into account when the \l TextureViewFormat + feature is reported as supported. + + \note This functionality is provided to allow "casting" between + non-sRGB and sRGB in order to get the shader write not perform, or perform, + the implicit sRGB conversions. Other types of casting may or may not be + functional. + */ + +/*! \class QRhiSampler \inmodule QtGui \since 6.6 \brief Sampler resource. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -4505,7 +4848,7 @@ QRhiResource::Type QRhiSampler::resourceType() const a collection of attachments (color, depth, stencil) and describes how those attachments are used. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -4624,7 +4967,7 @@ const QRhiNativeHandles *QRhiRenderPassDescriptor::nativeHandles() QRhiSwapChain returns when calling \l{QRhiSwapChain::currentFrameRenderTarget()}{currentFrameRenderTarget()}. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiSwapChainRenderTarget, QRhiTextureRenderTarget @@ -4703,7 +5046,7 @@ QRhiSwapChainRenderTarget::QRhiSwapChainRenderTarget(QRhiImplementation *rhi, QR QRhiSwapChainRenderTarget. This is what QRhiSwapChain::currentFrameRenderTarget() returns. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiSwapChain @@ -4752,7 +5095,7 @@ QRhiResource::Type QRhiSwapChainRenderTarget::resourceType() const // rt can now be used with beginPass() \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -4768,7 +5111,19 @@ QRhiResource::Type QRhiSwapChainRenderTarget::resourceType() const \value PreserveColorContents Indicates that the contents of the color attachments is to be loaded when starting a render pass, instead of clearing. This is potentially more expensive, especially on mobile (tiled) - GPUs, but allows preserving the existing contents between passes. + GPUs, but allows preserving the existing contents between passes. When doing + multisample rendering with a resolve texture set, setting this flag also + requests the multisample color data to be stored (written out) to the + multisample texture or render buffer. (for non-multisample rendering the + color data is always stored, but for MSAA storing the multisample data + decreases efficiency for certain GPU architectures, hence defaulting to not + writing it out) Note however that this is non-portable: in some cases there + is no intermediate multisample texture on the graphics API level, e.g. when + using OpenGL ES's \c{GL_EXT_multisampled_render_to_texture} as it is all + implicit, handled by the OpenGL ES implementation. In that case, + PreserveColorContents will likely have no effect. Therefore, avoid relying + on this flag when using multisample rendering and the color attachment is + using a multisample QRhiTexture (not QRhiRenderBuffer). \value PreserveDepthStencilContents Indicates that the contents of the depth texture is to be loaded when starting a render pass, instead @@ -4776,6 +5131,13 @@ QRhiResource::Type QRhiSwapChainRenderTarget::resourceType() const (QRhiTextureRenderTargetDescription::depthTexture() is set) because depth/stencil renderbuffers may not have any physical backing and data may not be written out in the first place. + + \value DoNotStoreDepthStencilContents Indicates that the contents of the + depth texture does not need to be written out. Relevant only when a + QRhiTexture, not QRhiRenderBuffer, is used as the depth-stencil buffer, + because for QRhiRenderBuffer this is implicit. When a depthResolveTexture is + set, the flag is not relevant, because the behavior is then as if the flag + was set. This enum value is introduced in Qt 6.8. */ /*! @@ -4933,11 +5295,19 @@ QRhiResource::Type QRhiTextureRenderTarget::resourceType() const cb->setShaderResources(srb2); // binds srb2 \endcode - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ /*! + \typedef QRhiShaderResourceBindingSet + \relates QRhi + \since 6.7 + + Synonym for QRhiShaderResourceBindings. +*/ + +/*! \internal */ QRhiShaderResourceBindings::QRhiShaderResourceBindings(QRhiImplementation *rhi) @@ -5064,7 +5434,7 @@ void QRhiImplementation::updateLayoutDesc(QRhiShaderResourceBindings *srb) static functions such as uniformBuffer() or sampledTexture() to get an instance. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -6085,7 +6455,7 @@ QDebug operator<<(QDebug dbg, const QRhiShaderResourceBindings &srb) four channels, depth test/write are disabled, stencil operations are disabled. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiCommandBuffer, QRhi @@ -6258,7 +6628,7 @@ QDebug operator<<(QDebug dbg, const QRhiShaderResourceBindings &srb) mode Qt Quick uses, it is enough to set the \c enable flag to true while leaving other values at their defaults. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -6308,7 +6678,7 @@ QDebug operator<<(QDebug dbg, const QRhiShaderResourceBindings &srb) \li compareOp \l Always \endlist - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -6688,6 +7058,29 @@ QRhiResource::Type QRhiGraphicsPipeline::resourceType() const */ /*! + \fn int QRhiGraphicsPipeline::multiViewCount() const + \return the view count. The default is 0, indicating no multiview rendering. + \since 6.7 + */ + +/*! + \fn void QRhiGraphicsPipeline::setMultiViewCount(int count) + Sets the view \a count for multiview rendering. The default is 0, + indicating no multiview rendering. + \a count must be 2 or larger to trigger multiview rendering. + + Multiview is only available when the \l{QRhi::MultiView}{MultiView feature} + is reported as supported. The render target must be a 2D texture array, and + the color attachment for the render target must have the same \a count set. + + See QRhiColorAttachment::setMultiViewCount() for further details on + multiview rendering. + + \since 6.7 + \sa QRhi::MultiView, QRhiColorAttachment::setMultiViewCount() + */ + +/*! \class QRhiSwapChain \inmodule QtGui \since 6.6 @@ -6825,7 +7218,7 @@ QRhiResource::Type QRhiGraphicsPipeline::resourceType() const appropriate sample count also via QSurfaceFormat, by calling QSurfaceFormat::setDefaultFormat() before initializing the QRhi. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -6885,6 +7278,14 @@ QRhiResource::Type QRhiGraphicsPipeline::resourceType() const \enum QRhiSwapChain::Format Describes the swapchain format. The default format is SDR. + This enum is used with + \l{QRhiSwapChain::isFormatSupported()}{isFormatSupported()} to check + upfront if creating the swapchain with the given format is supported by the + platform and the window's associated screen, and with + \l{QRhiSwapChain::setFormat()}{setFormat()} + to set the requested format in the swapchain before calling + \l{QRhiSwapChain::createOrResize()}{createOrResize()} for the first time. + \value SDR 8-bit RGBA or BGRA, depending on the backend and platform. With OpenGL ES in particular, it could happen that the platform provides less than 8 bits (e.g. due to EGL and the QSurfaceFormat choosing a 565 or 444 @@ -6896,10 +7297,14 @@ QRhiResource::Type QRhiGraphicsPipeline::resourceType() const (same as SDR/sRGB) and linear colors. Conversion to the display's native color space (such as, HDR10) is performed by the windowing system. On Windows this is the canonical color space of the system compositor, and is - the recommended format for HDR swapchains in general. + the recommended format for HDR swapchains in general on desktop platforms. \value HDR10 10-bit unsigned int RGB or BGR with 2 bit alpha, high dynamic range, HDR10 (Rec. 2020) color space with an ST2084 PQ transfer function. + + \value HDRExtendedDisplayP3Linear 16-bit float RGBA, high dynamic range, + extended linear Display P3 color space. The primary choice for HDR on + platforms such as iOS and VisionOS. */ /*! @@ -6993,6 +7398,21 @@ QRhiResource::Type QRhiSwapChain::resourceType() const time. If the result is true for a HDR format, then creating the swapchain with that format is expected to succeed as long as the window is not moved to another screen in the meantime. + + The main use of this function is to call it before the first + createOrResize() after the window is already set. This allow the QRhi + backends to perform platform or windowing system specific queries to + determine if the window (and the screen it is on) is capable of true HDR + output with the specified format. + + When the format is reported as supported, call setFormat() to set the + requested format and call createOrResize(). Be aware of the consequences + however: successfully requesting a HDR format will involve having to deal + with a different color space, possibly doing white level correction for + non-HDR-aware content, adjusting tonemapping methods, adjusting offscreen + render target settings, etc. + + \sa setFormat() */ /*! @@ -7041,10 +7461,9 @@ QRhiResource::Type QRhiSwapChain::resourceType() const is backed by two color buffers, one for each eye, instead of just one. When stereoscopic rendering is not supported, the return value will be - the default target. For the time being the only backend and 3D API where traditional - stereoscopic rendering is supported is OpenGL (excluding OpenGL ES), in + the default target. It is supported by all hardware backends except for Metal, in combination with \l QSurfaceFormat::StereoBuffers, assuming it is supported - by the graphics and display driver stack at run time. All other backends + by the graphics and display driver stack at run time. Metal and Null backends are going to return the default render target from this overload. \note the value must not be cached and reused between frames @@ -7111,6 +7530,15 @@ QRhiRenderTarget *QRhiSwapChain::currentFrameRenderTarget(StereoTargetBuffer tar /*! \fn void QRhiSwapChain::setFormat(Format f) Sets the format \a f. + + Avoid setting formats that are reported as unsupported from + isFormatSupported(). Note that support for a given format may depend on the + screen the swapchain's associated window is opened on. On some platforms, + such as Windows and macOS, for HDR output to work it is necessary to have + HDR output enabled in the display settings. + + See isFormatSupported(), \l QRhiSwapChainHdrInfo, and \l Format for more + information on high dynamic range output. */ /*! @@ -7172,11 +7600,12 @@ QRhiRenderTarget *QRhiSwapChain::currentFrameRenderTarget(StereoTargetBuffer tar \brief Describes the high dynamic range related information of the swapchain's associated output. - To perform tonemapping, one often needs to know the maximum luminance of - the display the swapchain's window is associated with. While this is often - made user-configurable, it can be highly useful to set defaults based on - the values reported by the display itself, thus providing a decent starting - point. + To perform HDR-compatible tonemapping, where the target range is not [0,1], + one often needs to know the maximum luminance of the display the + swapchain's window is associated with. While this is often made + user-configurable (think brightness, gamma and similar settings in games), + it can be highly useful to set defaults based on the values reported by the + display itself, thus providing a decent starting point. There are some problems however: the information is exposed in different forms on different platforms, whereas with cross-platform graphics APIs @@ -7184,11 +7613,6 @@ QRhiRenderTarget *QRhiSwapChain::currentFrameRenderTarget(StereoTargetBuffer tar information is not in the scope of the API (and may rather be retrievable via other platform-specific means, if any). - The struct returned from QRhiSwapChain::hdrInfo() contains either some - hard-coded defaults, indicated by the \c isHardCodedDefaults field, or real - values received from an API such as DXGI (IDXGIOutput6) or Cocoa - (NSScreen). The default is 1000 nits for maximum luminance. - With Metal on macOS/iOS, there is no luminance values exposed in the platform APIs. Instead, the maximum color component value, that would be 1.0 in a non-HDR setup, is provided. The \c limitsType field indicates what @@ -7197,10 +7621,23 @@ QRhiRenderTarget *QRhiSwapChain::currentFrameRenderTarget(StereoTargetBuffer tar fit. With an API like Vulkan, where there is no way to get such information, the - values are always the built-in defaults and \c isHardCodedDefaults is - always true. - - \note This a RHI API with limited compatibility guarantees, see \l QRhi + values are always the built-in defaults. + + Therefore, the struct returned from QRhiSwapChain::hdrInfo() contains + either some hard-coded defaults or real values received from an API such as + DXGI (IDXGIOutput6) or Cocoa (NSScreen). When no platform queries are + available (or needs using platform facilities out of scope for QRhi), the + hard-coded defaults are a maximum luminance of 1000 nits and an SDR white + level of 200. + + The struct also exposes the presumed luminance behavior of the platform and + its compositor, to indicate what a color component value of 1.0 is treated + as in a HDR color buffer. In some cases it will be necessary to perform + color correction of non-HDR content composited with HDR content. To enable + this, the SDR white level is queried from the system on some platforms + (Windows) and exposed here. + + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhiSwapChain::hdrInfo() @@ -7217,16 +7654,20 @@ QRhiRenderTarget *QRhiSwapChain::currentFrameRenderTarget(StereoTargetBuffer tar */ /*! - \variable QRhiSwapChainHdrInfo::isHardCodedDefaults + \enum QRhiSwapChainHdrInfo::LuminanceBehavior - Set to true when the data in the QRhiSwapChainHdrInfo consists entirely of - the hard-coded default values, for example because there is no way to query - the relevant information with a given graphics API or platform. (or because - querying it can be achieved only by means, e.g. platform APIs in some other - area, that are out of scope for the QRhi layer of the Qt graphics stack to - handle) + \value SceneReferred Indicates that the color value of 1.0 is interpreted + as 80 nits. This is the behavior of HDR-enabled windows with the Windows + compositor. See + \l{https://learn.microsoft.com/en-us/windows/win32/direct3darticles/high-dynamic-range}{this + page} for more information on HDR on Windows. - \sa QRhiSwapChain::hdrInfo() + \value DisplayReferred Indicates that the color value of 1.0 is interpreted + as the value of the SDR white. (which can be e.g. 200 nits, but will vary + depending on screen brightness) This is the behavior of HDR-enabled windows + on Apple platforms. See + \l{https://developer.apple.com/documentation/metal/hdr_content/displaying_hdr_content_in_a_metal_layer}{this + page} for more information on Apple's EDR system. */ /*! @@ -7256,23 +7697,92 @@ QRhiRenderTarget *QRhiSwapChain::currentFrameRenderTarget(StereoTargetBuffer tar } luminanceInNits; \endcode - Whereas for macOS/iOS, the maximum color component value (e.g. supposedly - something larger than 1.0f) is provided: + On Windows the minimum and maximum luminance depends on the screen + brightness. While not relevant for desktops, on laptops the screen + brightness may change at any time. Increasing brightness implies decreased + maximum luminance. In addition, the results may also be dependent on the + HDR Content Brightness set in Windows Settings' System/Display/HDR view, + if there is such a setting. + + Note however that the changes made to the laptop screen's brightness or in + the system settings while the application is running are not necessarily + reflected in the returned values, meaning calling hdrInfo() again may still + return the same luminance range as before for the rest of the process' + lifetime. The exact behavior is up to DXGI and Qt has no control over it. + + \note The Windows compositor works in scene-referred mode for HDR content. + A color component value of 1.0 corresponds to a luminance of 80 nits. When + rendering non-HDR content (e.g. 2D UI elements), the correction of the + white level is often necessary. (e.g., outputting the fragment color (1, 1, + 1) will likely lead to showing a shade of white that is too dim on-screen) + See \l sdrWhiteLevel. + + For macOS/iOS, the current maximum and potential maximum color + component values are provided: \code struct { float maxColorComponentValue; + float maxPotentialColorComponentValue; } colorComponentValue; \endcode + The value may depend on the screen brightness, which on laptops means that + the result may change in the next call to hdrInfo() if the brightness was + changed in the meantime. The maximum screen brightness implies a maximum + color value of 1.0. + + \note Apple's EDR is display-referred. 1.0 corresponds to a luminance level + of SDR white (e.g. 200 nits), the value of which varies based on the screen + brightness and possibly other settings. The exact luminance value for that, + or the maximum luminance of the display, are not exposed to the + applications. + + \note It has been observed that the color component values are not set to + the correct larger-than-1 value right away on startup on some macOS + systems, but the values tend to change during or after the first frame. + \sa QRhiSwapChain::hdrInfo() */ /*! + \variable QRhiSwapChainHdrInfo::luminanceBehavior + + Describes the platform's presumed behavior with regards to color values. + + \sa sdrWhiteLevel + */ + +/*! + \variable QRhiSwapChainHdrInfo::sdrWhiteLevel + + On Windows this is the dynamic SDR white level in nits. The value is + dependent on the screen brightness (on laptops), and the SDR or HDR Content + Brightness settings in the Windows settings' System/Display/HDR view. + + To perform white level correction for non-HDR (SDR) content, such as 2D UI + elemenents, multiply the final color with sdrWhiteLevel / 80.0 whenever + \l luminanceBehavior is SceneReferred. (assuming Windows and a linear + extended sRGB (scRGB) color space) + + On other platforms the value is always a pre-defined value, 200. This may + not match the system's actual SDR white level, but the value of this + variable is not relevant in practice when the \l luminanceBehavior is + DisplayReferred, because then the color component value of 1.0 refers to + the SDR white by default. + + \sa luminanceBehavior +*/ + +/*! \return the HDR information for the associated display. - The returned struct is always the default one if createOrResize() has not - been successfully called yet. + Do not assume that this is a cheap operation. Depending on the platform, + this function makes various platform queries which may have a performance + impact. + + \note Can be called before createOrResize() as long as the window is + \l{setWindow()}{set}. \note What happens when moving a window with an initialized swapchain between displays (HDR to HDR with different characteristics, HDR to SDR, @@ -7287,10 +7797,11 @@ QRhiRenderTarget *QRhiSwapChain::currentFrameRenderTarget(StereoTargetBuffer tar QRhiSwapChainHdrInfo QRhiSwapChain::hdrInfo() { QRhiSwapChainHdrInfo info; - info.isHardCodedDefaults = true; info.limitsType = QRhiSwapChainHdrInfo::LuminanceInNits; info.limits.luminanceInNits.minLuminance = 0.0f; info.limits.luminanceInNits.maxLuminance = 1000.0f; + info.luminanceBehavior = QRhiSwapChainHdrInfo::SceneReferred; + info.sdrWhiteLevel = 200.0f; return info; } @@ -7298,7 +7809,7 @@ QRhiSwapChainHdrInfo QRhiSwapChain::hdrInfo() QDebug operator<<(QDebug dbg, const QRhiSwapChainHdrInfo &info) { QDebugStateSaver saver(dbg); - dbg.nospace() << "QRhiSwapChainHdrInfo(" << (info.isHardCodedDefaults ? "with hard-coded defaults" : "queried from system"); + dbg.nospace() << "QRhiSwapChainHdrInfo("; switch (info.limitsType) { case QRhiSwapChainHdrInfo::LuminanceInNits: dbg.nospace() << " minLuminance=" << info.limits.luminanceInNits.minLuminance @@ -7306,6 +7817,15 @@ QDebug operator<<(QDebug dbg, const QRhiSwapChainHdrInfo &info) break; case QRhiSwapChainHdrInfo::ColorComponentValue: dbg.nospace() << " maxColorComponentValue=" << info.limits.colorComponentValue.maxColorComponentValue; + dbg.nospace() << " maxPotentialColorComponentValue=" << info.limits.colorComponentValue.maxPotentialColorComponentValue; + break; + } + switch (info.luminanceBehavior) { + case QRhiSwapChainHdrInfo::SceneReferred: + dbg.nospace() << " scene-referred, SDR white level=" << info.sdrWhiteLevel; + break; + case QRhiSwapChainHdrInfo::DisplayReferred: + dbg.nospace() << " display-referred"; break; } dbg.nospace() << ')'; @@ -7325,7 +7845,7 @@ QDebug operator<<(QDebug dbg, const QRhiSwapChainHdrInfo &info) \note Setting the shader is mandatory. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -7407,7 +7927,7 @@ QRhiComputePipeline::QRhiComputePipeline(QRhiImplementation *rhi) QRhiSwapChain::currentFrameCommandBuffer(), or, in case of rendering completely offscreen, initializing one via QRhi::beginOffscreenFrame(). - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -7733,6 +8253,17 @@ void QRhiImplementation::textureFormatInfo(QRhiTexture::Format format, const QSi *bytesPerPixel = bpc; } +bool QRhiImplementation::isStencilSupportingFormat(QRhiTexture::Format format) const +{ + switch (format) { + case QRhiTexture::D24S8: + return true; + default: + break; + } + return false; +} + bool QRhiImplementation::sanityCheckGraphicsPipeline(QRhiGraphicsPipeline *ps) { if (ps->cbeginShaderStages() == ps->cendShaderStages()) { @@ -7852,6 +8383,41 @@ bool QRhiImplementation::sanityCheckShaderResourceBindings(QRhiShaderResourceBin return true; } +int QRhiImplementation::effectiveSampleCount(int sampleCount) const +{ + // Stay compatible with QSurfaceFormat and friends where samples == 0 means the same as 1. + const int s = qBound(1, sampleCount, 64); + const QList<int> supported = supportedSampleCounts(); + int result = 1; + + // Stay compatible with Qt 5 in that requesting an unsupported sample count + // is not an error (although we still do a categorized debug print about + // this), and rather a supported value, preferably a close one, not just 1, + // is used instead. This is actually deviating from Qt 5 as that performs a + // clamping only and does not handle cases such as when sample count 2 is + // not supported but 4 is. (OpenGL handles things like that gracefully, + // other APIs may not, so improve this by picking the next largest, or in + // absence of that, the largest value; this with the goal to not reduce + // quality by rather picking a larger-than-requested value than a smaller one) + + for (int i = 0, ie = supported.count(); i != ie; ++i) { + // assumes the 'supported' list is sorted + if (supported[i] >= s) { + result = supported[i]; + break; + } + } + + if (result != s) { + if (result == 1 && !supported.isEmpty()) + result = supported.last(); + qCDebug(QRHI_LOG_INFO, "Attempted to set unsupported sample count %d, using %d instead", + sampleCount, result); + } + + return result; +} + /*! \internal */ @@ -7867,11 +8433,11 @@ QRhi::~QRhi() if (!d) return; + runCleanup(); + qDeleteAll(d->pendingDeleteResources); d->pendingDeleteResources.clear(); - runCleanup(); - d->destroy(); delete d; } @@ -7959,7 +8525,7 @@ QRhi *QRhi::create(Implementation impl, QRhiInitParams *params, Flags flags, QRh break; #endif case Metal: -#if defined(Q_OS_MACOS) || defined(Q_OS_IOS) +#if QT_CONFIG(metal) r->d = new QRhiMetal(static_cast<QRhiMetalInitParams *>(params), static_cast<QRhiMetalNativeHandles *>(importDevice)); break; @@ -7969,10 +8535,18 @@ QRhi *QRhi::create(Implementation impl, QRhiInitParams *params, Flags flags, QRh #endif case D3D12: #ifdef Q_OS_WIN +#ifdef QRHI_D3D12_AVAILABLE r->d = new QRhiD3D12(static_cast<QRhiD3D12InitParams *>(params), static_cast<QRhiD3D12NativeHandles *>(importDevice)); break; #else + qWarning("Qt was built without Direct3D 12 support. " + "This is likely due to having ancient SDK headers (such as d3d12.h) in the Qt build environment. " + "Rebuild Qt with an SDK supporting D3D12 features introduced in Windows 10 version 1703, " + "or use an MSVC build as those typically are built with more up-to-date SDKs."); + break; +#endif +#else qWarning("This platform has no Direct3D 12 support"); break; #endif @@ -8010,7 +8584,7 @@ bool QRhi::probe(QRhi::Implementation impl, QRhiInitParams *params) // create() and then drop the result. if (impl == Metal) { -#if defined(Q_OS_MACOS) || defined(Q_OS_IOS) +#if QT_CONFIG(metal) ok = QRhiMetal::probe(static_cast<QRhiMetalInitParams *>(params)); #endif } else { @@ -8028,7 +8602,7 @@ bool QRhi::probe(QRhi::Implementation impl, QRhiInitParams *params) \brief Opaque data describing native objects needed to set up a swapchain. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. \sa QRhi::updateSwapChainProxyData() @@ -8061,7 +8635,7 @@ bool QRhi::probe(QRhi::Implementation impl, QRhiInitParams *params) */ QRhiSwapChainProxyData QRhi::updateSwapChainProxyData(QRhi::Implementation impl, QWindow *window) { -#if defined(Q_OS_MACOS) || defined(Q_OS_IOS) +#if QT_CONFIG(metal) if (impl == Metal) return QRhiMetal::updateSwapChainProxyData(window); #else @@ -8113,9 +8687,11 @@ const char *QRhi::backendName() const /*! \enum QRhiDriverInfo::DeviceType - Specifies the graphics device's type, when the information is available. In - practice this is only applicable with Vulkan and Metal. With others the - value will always be UnknownDevice. + Specifies the graphics device's type, when the information is available. + + In practice this is only applicable with Vulkan and Metal. With Direct 3D + 11 and 12, using an adapter with the software flag set leads to the value + \c CpuDevice. Otherwise, and with OpenGL, the value is always UnknownDevice. \value UnknownDevice \value IntegratedDevice @@ -8141,7 +8717,7 @@ const char *QRhi::backendName() const for OpenGL and Metal. deviceType is always UnknownDevice for OpenGL and Direct 3D. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -8236,6 +8812,33 @@ void QRhi::addCleanupCallback(const CleanupCallback &callback) } /*! + \overload + + Registers \a callback to be invoked either when the QRhi is destroyed or + when runCleanup() is called. This overload takes an opaque pointer, \a key, + that is used to ensure that a given callback is registered (and so called) + only once. + + \sa removeCleanupCallback() + */ +void QRhi::addCleanupCallback(const void *key, const CleanupCallback &callback) +{ + d->addCleanupCallback(key, callback); +} + +/*! + Deregisters the callback with \a key. If no cleanup callback was registered + with \a key, the function does nothing. Callbacks registered without a key + cannot be removed. + + \sa addCleanupCallback() + */ +void QRhi::removeCleanupCallback(const void *key) +{ + d->removeCleanupCallback(key); +} + +/*! Invokes all registered cleanup functions. The list of cleanup callbacks it then cleared. Normally destroying the QRhi does this automatically, but sometimes it can be useful to trigger cleanup in order to release all @@ -8249,6 +8852,11 @@ void QRhi::runCleanup() f(this); d->cleanupCallbacks.clear(); + + for (auto it = d->keyedCleanupCallbacks.cbegin(), end = d->keyedCleanupCallbacks.cend(); it != end; ++it) + it.value()(this); + + d->keyedCleanupCallbacks.clear(); } /*! @@ -8271,7 +8879,7 @@ void QRhi::runCleanup() To get an available, empty batch from the pool, call QRhi::nextResourceUpdateBatch(). - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -8412,6 +9020,8 @@ void QRhiResourceUpdateBatch::uploadStaticBuffer(QRhiBuffer *buf, quint32 offset } /*! + \overload + Enqueues updating the entire QRhiBuffer \a buf created with the type QRhiBuffer::Immutable or QRhiBuffer::Static. */ @@ -9267,9 +9877,9 @@ const QRhiNativeHandles *QRhiCommandBuffer::nativeHandles() called when the pass recording was started with specifying QRhiCommandBuffer::ExternalContent. - With Vulkan or Metal one can query the native command buffer or encoder - objects via nativeHandles() and enqueue commands to them. With OpenGL or - Direct3D 11 the (device) context can be retrieved from + With Vulkan, Metal, or Direct3D 12 one can query the native command buffer + or encoder objects via nativeHandles() and enqueue commands to them. With + OpenGL or Direct3D 11 the (device) context can be retrieved from QRhi::nativeHandles(). However, this must never be done without ensuring the QRhiCommandBuffer's state stays up-to-date. Hence the requirement for wrapping any externally added command recording between beginExternal() and @@ -9317,8 +9927,17 @@ void QRhiCommandBuffer::endExternal() } /*! - \return the last available timestamp, in seconds. The value indicates the - elapsed time on the GPU during the last completed frame. + \return the last available timestamp, in seconds, when + \l QRhi::EnableTimestamps was enabled when creating the QRhi. The value + indicates the elapsed time on the GPU during the last completed frame. + + \note Do not expect results other than 0 when the QRhi::Timestamps feature + is not reported as supported, or when QRhi::EnableTimestamps was not passed + to QRhi::create(). There are exceptions to this, because with some graphics + APIs (Metal) timings are available without having to perform extra + operations (timestamp queries), but portable applications should always + consciously opt-in to timestamp collection when they know it is needed, and + call this function accordingly. Care must be exercised with the interpretation of the value, as its precision and granularity is often not controlled by Qt, and depends on the @@ -9326,25 +9945,47 @@ void QRhiCommandBuffer::endExternal() the values between different graphics APIs and hardware is discouraged and may be meaningless. - The timing values may become available asynchronously. The returned value - may therefore be 0 or the last known value referring to some previous - frame. The value my also become 0 again under certain conditions, such as - when resizing the window. It can be expected that the most up-to-date - available value is retrieved in beginFrame() and becomes queriable via this - function once beginFrame() returns. + When the frame was recorded with \l{QRhi::beginFrame()}{beginFrame()} and + \l{QRhi::endFrame()}{endFrame()}, i.e., with a swapchain, the timing values + will likely become available asynchronously. The returned value may + therefore be 0 (e.g., for the first 1-2 frames) or the last known value + referring to some previous frame. The value my also + become 0 again under certain conditions, such as when resizing the window. + It can be expected that the most up-to-date available value is retrieved in + beginFrame() and becomes queriable via this function once beginFrame() + returns. \note Do not assume that the value refers to the previous (\c{currently_recorded - 1}) frame. It may refer to \c{currently_recorded - 2} or \c{currently_recorded - 3} as well. The exact behavior may depend on the graphics API and its implementation. - \note The result is always 0 when the QRhi::Timestamps feature is not - reported as supported, or when QRhi::EnableTimestamps was not passed to - QRhi::create(). There are exceptions to the latter, because with some - graphics APIs timings are available without having to perform extra - operations, but portable applications should always consciously opt-in to - timestamp collection when they know it is needed, and call this function - accordingly. + On the other hand, with offscreen frames the returned value is up-to-date + once \l{QRhi::endOffscreenFrame()}{endOffscreenFrame()} returns, because + offscreen frames reduce GPU pipelining and wait the the commands to be + complete. + + \note This means that, unlike with swapchain frames, with offscreen frames + the returned value is guaranteed to refer to the frame that has just been + submitted and completed. (assuming this function is called after + endOffscreenFrame() but before the next beginOffscreenFrame()) + + Watch out for the consequences of GPU frequency scaling and GPU clock + changes, depending on the platform. For example, on Windows the returned + timing may vary in a quite wide range between frames with modern graphics + cards, even when submitting frames with a similar, or the same workload. + This is out of scope for Qt to control and solve, generally speaking. + However, the D3D12 backend automatically calls + \l{https://learn.microsoft.com/en-us/windows/win32/api/d3d12/nf-d3d12-id3d12device-setstablepowerstate}{ID3D12Device::SetStablePowerState()} + whenever the environment variable \c QT_D3D_STABLE_POWER_STATE is set to a + non-zero value. This can greatly stabilize the result. It can also have a + non-insignificant effect on the CPU-side timings measured via QElapsedTimer + for example, especially when offscreen frames are involved. + + \note Do not and never ship applications to production with + \c QT_D3D_STABLE_POWER_STATE set. See the Windows API documentation for details. + + \sa QRhi::Timestamps, QRhi::EnableTimestamps */ double QRhiCommandBuffer::lastCompletedGpuTime() { @@ -9693,7 +10334,7 @@ void QRhi::setPipelineCacheData(const QByteArray &data) \brief Statistics provided from the underlying memory allocator. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -10011,6 +10652,14 @@ QRhiTexture *QRhi::newTextureArray(QRhiTexture::Format format, minification filter \a minFilter, mipmapping mode \a mipmapMode, and the addressing (wrap) modes \a addressU, \a addressV, and \a addressW. + \note Setting \a mipmapMode to a value other than \c None implies that + images for all relevant mip levels will be provided either via + \l{QRhiResourceUpdateBatch::uploadTexture()}{texture uploads} or by calling + \l{QRhiResourceUpdateBatch::generateMips()}{generateMips()} on the texture + that is used with this sampler. Attempting to use the sampler with a + texture that has no data for all relevant mip levels will lead to rendering + errors, with the exact behavior dependent on the underlying graphics API. + \sa QRhiResource::destroy() */ QRhiSampler *QRhi::newSampler(QRhiSampler::Filter magFilter, diff --git a/src/gui/rhi/qrhi.h b/src/gui/rhi/qrhi.h index 8ec6630acf..d20b7e00d1 100644 --- a/src/gui/rhi/qrhi.h +++ b/src/gui/rhi/qrhi.h @@ -250,7 +250,15 @@ public: Half4, Half3, Half2, - Half + Half, + UShort4, + UShort3, + UShort2, + UShort, + SShort4, + SShort3, + SShort2, + SShort, }; QRhiVertexInputAttribute() = default; @@ -495,7 +503,7 @@ public: quint32 maybeSize; bool hasDynamicOffset; }; - static const int MAX_TEX_SAMPLER_ARRAY_SIZE = 16; + static constexpr int MAX_TEX_SAMPLER_ARRAY_SIZE = 16; struct TextureAndOrSamplerData { int count; TextureAndSampler texSamplers[MAX_TEX_SAMPLER_ARRAY_SIZE]; @@ -535,7 +543,7 @@ public: } }; - static const int LAYOUT_DESC_ENTRIES_PER_BINDING = 4; + static constexpr int LAYOUT_DESC_ENTRIES_PER_BINDING = 4; template<typename Output> static void serializeLayoutDescription(const QRhiShaderResourceBinding *first, @@ -592,6 +600,9 @@ public: int resolveLevel() const { return m_resolveLevel; } void setResolveLevel(int level) { m_resolveLevel = level; } + int multiViewCount() const { return m_multiViewCount; } + void setMultiViewCount(int count) { m_multiViewCount = count; } + private: QRhiTexture *m_texture = nullptr; QRhiRenderBuffer *m_renderBuffer = nullptr; @@ -600,6 +611,7 @@ private: QRhiTexture *m_resolveTexture = nullptr; int m_resolveLayer = 0; int m_resolveLevel = 0; + int m_multiViewCount = 0; }; Q_DECLARE_TYPEINFO(QRhiColorAttachment, Q_RELOCATABLE_TYPE); @@ -630,10 +642,14 @@ public: QRhiTexture *depthTexture() const { return m_depthTexture; } void setDepthTexture(QRhiTexture *texture) { m_depthTexture = texture; } + QRhiTexture *depthResolveTexture() const { return m_depthResolveTexture; } + void setDepthResolveTexture(QRhiTexture *tex) { m_depthResolveTexture = tex; } + private: QVarLengthArray<QRhiColorAttachment, 8> m_colorAttachments; QRhiRenderBuffer *m_depthStencilBuffer = nullptr; QRhiTexture *m_depthTexture = nullptr; + QRhiTexture *m_depthResolveTexture = nullptr; }; class Q_GUI_EXPORT QRhiTextureSubresourceUploadDescription @@ -979,6 +995,15 @@ public: int sampleCount() const { return m_sampleCount; } void setSampleCount(int s) { m_sampleCount = s; } + struct ViewFormat { + QRhiTexture::Format format; + bool srgb; + }; + ViewFormat readViewFormat() const { return m_readViewFormat; } + void setReadViewFormat(const ViewFormat &fmt) { m_readViewFormat = fmt; } + ViewFormat writeViewFormat() const { return m_writeViewFormat; } + void setWriteViewFormat(const ViewFormat &fmt) { m_writeViewFormat = fmt; } + virtual bool create() = 0; virtual NativeTexture nativeTexture(); virtual bool createFrom(NativeTexture src); @@ -995,6 +1020,8 @@ protected: Flags m_flags; int m_arrayRangeStart = -1; int m_arrayRangeLength = -1; + ViewFormat m_readViewFormat = { UnknownFormat, false }; + ViewFormat m_writeViewFormat = { UnknownFormat, false }; }; Q_DECLARE_OPERATORS_FOR_FLAGS(QRhiTexture::Flags) @@ -1158,7 +1185,8 @@ class Q_GUI_EXPORT QRhiTextureRenderTarget : public QRhiRenderTarget public: enum Flag { PreserveColorContents = 1 << 0, - PreserveDepthStencilContents = 1 << 1 + PreserveDepthStencilContents = 1 << 1, + DoNotStoreDepthStencilContents = 1 << 2 }; Q_DECLARE_FLAGS(Flags, Flag) @@ -1213,7 +1241,7 @@ public: virtual void updateResources(UpdateFlags flags = {}) = 0; protected: - static const int BINDING_PREALLOC = 12; + static constexpr int BINDING_PREALLOC = 12; QRhiShaderResourceBindings(QRhiImplementation *rhi); QVarLengthArray<QRhiShaderResourceBinding, BINDING_PREALLOC> m_bindings; size_t m_layoutDescHash = 0; @@ -1233,6 +1261,11 @@ Q_DECLARE_OPERATORS_FOR_FLAGS(QRhiShaderResourceBindings::UpdateFlags) Q_GUI_EXPORT QDebug operator<<(QDebug, const QRhiShaderResourceBindings &); #endif +// The proper name. Until it gets rolled out universally, have the better name +// as a typedef. Eventually it should be reversed (the old name being a typedef +// to the new one). +using QRhiShaderResourceBindingSet = QRhiShaderResourceBindings; + class Q_GUI_EXPORT QRhiGraphicsPipeline : public QRhiResource { public: @@ -1437,6 +1470,9 @@ public: PolygonMode polygonMode() const {return m_polygonMode; } void setPolygonMode(PolygonMode mode) {m_polygonMode = mode; } + int multiViewCount() const { return m_multiViewCount; } + void setMultiViewCount(int count) { m_multiViewCount = count; } + virtual bool create() = 0; protected: @@ -1460,6 +1496,7 @@ protected: float m_slopeScaledDepthBias = 0.0f; int m_patchControlPointCount = 3; PolygonMode m_polygonMode = Fill; + int m_multiViewCount = 0; QVarLengthArray<QRhiShaderStage, 4> m_shaderStages; QRhiVertexInputLayout m_vertexInputLayout; QRhiShaderResourceBindings *m_shaderResourceBindings = nullptr; @@ -1472,11 +1509,16 @@ Q_DECLARE_TYPEINFO(QRhiGraphicsPipeline::TargetBlend, Q_RELOCATABLE_TYPE); struct QRhiSwapChainHdrInfo { - bool isHardCodedDefaults; enum LimitsType { LuminanceInNits, ColorComponentValue }; + + enum LuminanceBehavior { + SceneReferred, + DisplayReferred + }; + LimitsType limitsType; union { struct { @@ -1485,8 +1527,11 @@ struct QRhiSwapChainHdrInfo } luminanceInNits; struct { float maxColorComponentValue; + float maxPotentialColorComponentValue; } colorComponentValue; } limits; + LuminanceBehavior luminanceBehavior; + float sdrWhiteLevel; }; Q_DECLARE_TYPEINFO(QRhiSwapChainHdrInfo, Q_RELOCATABLE_TYPE); @@ -1516,7 +1561,8 @@ public: enum Format { SDR, HDRExtendedSrgbLinear, - HDR10 + HDR10, + HDRExtendedDisplayP3Linear }; enum StereoTargetBuffer { @@ -1770,7 +1816,8 @@ public: EnableDebugMarkers = 1 << 0, PreferSoftwareRenderer = 1 << 1, EnablePipelineCacheDataSave = 1 << 2, - EnableTimestamps = 1 << 3 + EnableTimestamps = 1 << 3, + SuppressSmokeTestWarnings = 1 << 4 }; Q_DECLARE_FLAGS(Flags, Flag) @@ -1821,7 +1868,10 @@ public: OneDimensionalTextureMipmaps, HalfAttributes, RenderToOneDimensionalTexture, - ThreeDimensionalTextureMipmaps + ThreeDimensionalTextureMipmaps, + MultiView, + TextureViewFormat, + ResolveDepthStencil }; enum BeginFrameFlag { @@ -1866,6 +1916,8 @@ public: using CleanupCallback = std::function<void(QRhi *)>; void addCleanupCallback(const CleanupCallback &callback); + void addCleanupCallback(const void *key, const CleanupCallback &callback); + void removeCleanupCallback(const void *key); void runCleanup(); QRhiGraphicsPipeline *newGraphicsPipeline(); @@ -1942,7 +1994,7 @@ public: const QRhiNativeHandles *nativeHandles(); bool makeThreadLocalNativeContextCurrent(); - static const int MAX_MIP_LEVELS = 16; // -> max width or height is 65536 + static constexpr int MAX_MIP_LEVELS = 16; // -> max width or height is 65536 void releaseCachedResources(); diff --git a/src/gui/rhi/qrhi_p.h b/src/gui/rhi/qrhi_p.h index 05df169a35..b5429372a8 100644 --- a/src/gui/rhi/qrhi_p.h +++ b/src/gui/rhi/qrhi_p.h @@ -149,6 +149,7 @@ public: QSize *blockDim) const; void textureFormatInfo(QRhiTexture::Format format, const QSize &size, quint32 *bpl, quint32 *byteSize, quint32 *bytesPerPixel) const; + bool isStencilSupportingFormat(QRhiTexture::Format format) const; void registerResource(QRhiResource *res, bool ownsNativeResources = true) { @@ -178,6 +179,16 @@ public: cleanupCallbacks.append(callback); } + void addCleanupCallback(const void *key, const QRhi::CleanupCallback &callback) + { + keyedCleanupCallbacks[key] = callback; + } + + void removeCleanupCallback(const void *key) + { + keyedCleanupCallbacks.remove(key); + } + bool sanityCheckGraphicsPipeline(QRhiGraphicsPipeline *ps); bool sanityCheckShaderResourceBindings(QRhiShaderResourceBindings *srb); void updateLayoutDesc(QRhiShaderResourceBindings *srb); @@ -221,6 +232,8 @@ public: return a.d.binding < b.d.binding; } + int effectiveSampleCount(int sampleCount) const; + QRhi *q; static const int MAX_SHADER_CACHE_ENTRIES = 128; @@ -238,6 +251,7 @@ private: QHash<QRhiResource *, bool> resources; QSet<QRhiResource *> pendingDeleteResources; QVarLengthArray<QRhi::CleanupCallback, 4> cleanupCallbacks; + QHash<const void *, QRhi::CleanupCallback> keyedCleanupCallbacks; QElapsedTimer pipelineCreationTimer; qint64 accumulatedPipelineCreationTime = 0; diff --git a/src/gui/rhi/qrhi_platform.h b/src/gui/rhi/qrhi_platform.h index 30676d0da6..e7be522c52 100644 --- a/src/gui/rhi/qrhi_platform.h +++ b/src/gui/rhi/qrhi_platform.h @@ -23,7 +23,7 @@ #include <QtGui/qvulkaninstance.h> #endif -#if defined(Q_OS_MACOS) || defined(Q_OS_IOS) || defined(Q_QDOC) +#if QT_CONFIG(metal) || defined(Q_QDOC) Q_FORWARD_DECLARE_OBJC_CLASS(MTLDevice); Q_FORWARD_DECLARE_OBJC_CLASS(MTLCommandQueue); Q_FORWARD_DECLARE_OBJC_CLASS(MTLCommandBuffer); @@ -66,7 +66,7 @@ struct Q_GUI_EXPORT QRhiGles2NativeHandles : public QRhiNativeHandles #endif // opengl/qdoc -#if QT_CONFIG(vulkan) || defined(Q_QDOC) +#if (QT_CONFIG(vulkan) && __has_include(<vulkan/vulkan.h>)) || defined(Q_QDOC) struct Q_GUI_EXPORT QRhiVulkanInitParams : public QRhiInitParams { @@ -145,12 +145,12 @@ struct Q_GUI_EXPORT QRhiD3D12NativeHandles : public QRhiNativeHandles struct Q_GUI_EXPORT QRhiD3D12CommandBufferNativeHandles : public QRhiNativeHandles { - void *commandList = nullptr; // ID3D12GraphicsCommandList + void *commandList = nullptr; // ID3D12GraphicsCommandList1 }; #endif // WIN/QDOC -#if defined(Q_OS_MACOS) || defined(Q_OS_IOS) || defined(Q_QDOC) +#if QT_CONFIG(metal) || defined(Q_QDOC) struct Q_GUI_EXPORT QRhiMetalInitParams : public QRhiInitParams { diff --git a/src/gui/rhi/qrhid3d11.cpp b/src/gui/rhi/qrhid3d11.cpp index f20c1f4072..b09baf57b2 100644 --- a/src/gui/rhi/qrhid3d11.cpp +++ b/src/gui/rhi/qrhid3d11.cpp @@ -6,11 +6,9 @@ #include "vs_test_p.h" #include <QWindow> #include <qmath.h> -#include <private/qsystemlibrary_p.h> #include <QtCore/qcryptographichash.h> #include <QtCore/private/qsystemerror_p.h> - -#include <d3dcompiler.h> +#include "qrhid3dhelpers_p.h" QT_BEGIN_NAMESPACE @@ -32,7 +30,7 @@ using namespace Qt::StringLiterals; \since 6.6 \brief Direct3D 11 specific initialization parameters. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. A D3D11-based QRhi needs no special parameters for initialization. If @@ -89,28 +87,53 @@ using namespace Qt::StringLiterals; \c{d3d11.h} headers is not acceptable here. The actual types are \c{ID3D11Device *} and \c{ID3D11DeviceContext *}. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ /*! \variable QRhiD3D11NativeHandles::dev + + Points to a + \l{https://learn.microsoft.com/en-us/windows/win32/api/d3d11/nn-d3d11-id3d11device}{ID3D11Device} + or left set to \nullptr if no existing device is to be imported. + + \note When importing a device, both the device and the device context must be set to valid objects. */ /*! \variable QRhiD3D11NativeHandles::context + + Points to a \l{https://learn.microsoft.com/en-us/windows/win32/api/d3d11/nn-d3d11-id3d11devicecontext}{ID3D11DeviceContext} + or left set to \nullptr if no existing device context is to be imported. + + \note When importing a device, both the device and the device context must be set to valid objects. */ /*! \variable QRhiD3D11NativeHandles::featureLevel + + Specifies the feature level passed to + \l{https://learn.microsoft.com/en-us/windows/win32/api/d3d11/nf-d3d11-d3d11createdevice}{D3D11CreateDevice()}. + Relevant only when QRhi creates the device, ignored when importing a device + and device context. When not set, the default rules outlined in the D3D + documentation apply. */ /*! \variable QRhiD3D11NativeHandles::adapterLuidLow + + The low part of the local identifier (LUID) of the DXGI adapter to use. + Relevant only when QRhi creates the device, ignored when importing a device + and device context. */ /*! \variable QRhiD3D11NativeHandles::adapterLuidHigh + + The high part of the local identifier (LUID) of the DXGI adapter to use. + Relevant only when QRhi creates the device, ignored when importing a device + and device context. */ // help mingw with its ancient sdk headers @@ -194,13 +217,14 @@ bool QRhiD3D11::create(QRhi::Flags flags) if (qEnvironmentVariableIntValue("QT_D3D_FLIP_DISCARD")) qWarning("The default swap effect is FLIP_DISCARD, QT_D3D_FLIP_DISCARD is now ignored"); - if (qEnvironmentVariableIntValue("QT_D3D_NO_FLIP")) - qWarning("Non-FLIP swapchains are no longer supported, QT_D3D_NO_FLIP is now ignored"); + // Support for flip model swapchains is required now (since we are + // targeting Windows 10+), but the option for using the old model is still + // there. (some features are not supported then, however) + useLegacySwapchainModel = qEnvironmentVariableIntValue("QT_D3D_NO_FLIP"); - qCDebug(QRHI_LOG_INFO, "FLIP_* swapchain supported = true, ALLOW_TEARING supported = %s", - supportsAllowTearing ? "true" : "false"); - - qCDebug(QRHI_LOG_INFO, "Default swap effect: FLIP_DISCARD"); + qCDebug(QRHI_LOG_INFO, "FLIP_* swapchain supported = true, ALLOW_TEARING supported = %s, use legacy (non-FLIP) model = %s", + supportsAllowTearing ? "true" : "false", + useLegacySwapchainModel ? "true" : "false"); if (!importedDeviceAndContext) { IDXGIAdapter1 *adapter; @@ -249,9 +273,7 @@ bool QRhiD3D11::create(QRhi::Flags flags) if (!activeAdapter && (requestedAdapterIndex < 0 || requestedAdapterIndex == adapterIndex)) { activeAdapter = adapter; adapterLuid = desc.AdapterLuid; - driverInfoStruct.deviceName = name.toUtf8(); - driverInfoStruct.deviceId = desc.DeviceId; - driverInfoStruct.vendorId = desc.VendorId; + QRhiD3D::fillDriverInfo(&driverInfoStruct, desc); qCDebug(QRHI_LOG_INFO, " using this adapter"); } else { adapter->Release(); @@ -294,21 +316,24 @@ bool QRhiD3D11::create(QRhi::Flags flags) return false; } + const bool supports11_1 = SUCCEEDED(ctx->QueryInterface(__uuidof(ID3D11DeviceContext1), reinterpret_cast<void **>(&context))); + ctx->Release(); + if (!supports11_1) { + qWarning("ID3D11DeviceContext1 not supported"); + return false; + } + // Test if creating a Shader Model 5.0 vertex shader works; we want to // fail already in create() if that's not the case. ID3D11VertexShader *testShader = nullptr; if (SUCCEEDED(dev->CreateVertexShader(g_testVertexShader, sizeof(g_testVertexShader), nullptr, &testShader))) { testShader->Release(); } else { - qWarning("D3D11 smoke test failed (failed to create vertex shader)"); - ctx->Release(); - return false; - } - - const bool supports11_1 = SUCCEEDED(ctx->QueryInterface(__uuidof(ID3D11DeviceContext1), reinterpret_cast<void **>(&context))); - ctx->Release(); - if (!supports11_1) { - qWarning("ID3D11DeviceContext1 not supported"); + static const char *msg = "D3D11 smoke test: Failed to create vertex shader"; + if (flags.testFlag(QRhi::SuppressSmokeTestWarnings)) + qCDebug(QRHI_LOG_INFO, "%s", msg); + else + qWarning("%s", msg); return false; } @@ -318,11 +343,19 @@ bool QRhiD3D11::create(QRhi::Flags flags) // still not support this D3D_FEATURE_LEVEL_11_1 feature. (e.g. // because it only does 11_0) if (!features.ConstantBufferOffsetting) { - qWarning("Constant buffer offsetting is not supported by the driver"); + static const char *msg = "D3D11 smoke test: Constant buffer offsetting is not supported by the driver"; + if (flags.testFlag(QRhi::SuppressSmokeTestWarnings)) + qCDebug(QRHI_LOG_INFO, "%s", msg); + else + qWarning("%s", msg); return false; } } else { - qWarning("Failed to query D3D11_FEATURE_D3D11_OPTIONS"); + static const char *msg = "D3D11 smoke test: Failed to query D3D11_FEATURE_D3D11_OPTIONS"; + if (flags.testFlag(QRhi::SuppressSmokeTestWarnings)) + qCDebug(QRHI_LOG_INFO, "%s", msg); + else + qWarning("%s", msg); return false; } } else { @@ -332,12 +365,14 @@ bool QRhiD3D11::create(QRhi::Flags flags) if (SUCCEEDED(dev->QueryInterface(__uuidof(IDXGIDevice), reinterpret_cast<void **>(&dxgiDev)))) { IDXGIAdapter *adapter = nullptr; if (SUCCEEDED(dxgiDev->GetAdapter(&adapter))) { - DXGI_ADAPTER_DESC desc; - adapter->GetDesc(&desc); - adapterLuid = desc.AdapterLuid; - driverInfoStruct.deviceName = QString::fromUtf16(reinterpret_cast<char16_t *>(desc.Description)).toUtf8(); - driverInfoStruct.deviceId = desc.DeviceId; - driverInfoStruct.vendorId = desc.VendorId; + IDXGIAdapter1 *adapter1 = nullptr; + if (SUCCEEDED(adapter->QueryInterface(__uuidof(IDXGIAdapter1), reinterpret_cast<void **>(&adapter1)))) { + DXGI_ADAPTER_DESC1 desc; + adapter1->GetDesc1(&desc); + adapterLuid = desc.AdapterLuid; + QRhiD3D::fillDriverInfo(&driverInfoStruct, desc); + adapter1->Release(); + } adapter->Release(); } dxgiDev->Release(); @@ -348,11 +383,6 @@ bool QRhiD3D11::create(QRhi::Flags flags) if (FAILED(context->QueryInterface(__uuidof(ID3DUserDefinedAnnotation), reinterpret_cast<void **>(&annotations)))) annotations = nullptr; - if (flags.testFlag(QRhi::EnableTimestamps)) { - ofr.timestamps.prepare(2, this); - // timestamp queries are optional so we can go on even if they failed - } - deviceLost = false; nativeHandlesStruct.dev = dev; @@ -378,7 +408,16 @@ void QRhiD3D11::destroy() clearShaderCache(); - ofr.timestamps.destroy(); + if (ofr.tsDisjointQuery) { + ofr.tsDisjointQuery->Release(); + ofr.tsDisjointQuery = nullptr; + } + for (int i = 0; i < 2; ++i) { + if (ofr.tsQueries[i]) { + ofr.tsQueries[i]->Release(); + ofr.tsQueries[i] = nullptr; + } + } if (annotations) { annotations->Release(); @@ -427,19 +466,13 @@ QList<int> QRhiD3D11::supportedSampleCounts() const return { 1, 2, 4, 8 }; } -DXGI_SAMPLE_DESC QRhiD3D11::effectiveSampleCount(int sampleCount) const +DXGI_SAMPLE_DESC QRhiD3D11::effectiveSampleDesc(int sampleCount) const { DXGI_SAMPLE_DESC desc; desc.Count = 1; desc.Quality = 0; - // Stay compatible with QSurfaceFormat and friends where samples == 0 means the same as 1. - int s = qBound(1, sampleCount, 64); - - if (!supportedSampleCounts().contains(s)) { - qWarning("Attempted to set unsupported sample count %d", sampleCount); - return desc; - } + const int s = effectiveSampleCount(sampleCount); desc.Count = UINT(s); if (s > 1) @@ -588,6 +621,12 @@ bool QRhiD3D11::isFeatureSupported(QRhi::Feature feature) const return true; case QRhi::ThreeDimensionalTextureMipmaps: return true; + case QRhi::MultiView: + return false; + case QRhi::TextureViewFormat: + return false; // because we use fully typed formats for textures and relaxed casting is a D3D12 thing + case QRhi::ResolveDepthStencil: + return false; default: Q_UNREACHABLE(); return false; @@ -1258,7 +1297,6 @@ const QRhiNativeHandles *QRhiD3D11::nativeHandles(QRhiCommandBuffer *cb) void QRhiD3D11::beginExternal(QRhiCommandBuffer *cb) { QD3D11CommandBuffer *cbD = QRHI_RES(QD3D11CommandBuffer, cb); - // no timestampSwapChain, in order to avoid timestamp mess executeCommandBuffer(cbD); cbD->resetCommands(); } @@ -1281,6 +1319,19 @@ double QRhiD3D11::lastCompletedGpuTime(QRhiCommandBuffer *cb) return cbD->lastGpuTime; } +static inline QD3D11RenderTargetData *rtData(QRhiRenderTarget *rt) +{ + switch (rt->resourceType()) { + case QRhiResource::SwapChainRenderTarget: + return &QRHI_RES(QD3D11SwapChainRenderTarget, rt)->d; + case QRhiResource::TextureRenderTarget: + return &QRHI_RES(QD3D11TextureRenderTarget, rt)->d; + default: + Q_UNREACHABLE(); + return nullptr; + } +} + QRhi::FrameOpResult QRhiD3D11::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginFrameFlags flags) { Q_UNUSED(flags); @@ -1297,12 +1348,22 @@ QRhi::FrameOpResult QRhiD3D11::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginF finishActiveReadbacks(); - if (swapChainD->timestamps.active[currentFrameSlot]) { + if (swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex]) { double elapsedSec = 0; - if (swapChainD->timestamps.tryQueryTimestamps(currentFrameSlot, context, &elapsedSec)) + if (swapChainD->timestamps.tryQueryTimestamps(swapChainD->currentTimestampPairIndex, context, &elapsedSec)) swapChainD->cb.lastGpuTime = elapsedSec; } + ID3D11Query *tsStart = swapChainD->timestamps.query[swapChainD->currentTimestampPairIndex * 2]; + ID3D11Query *tsDisjoint = swapChainD->timestamps.disjointQuery[swapChainD->currentTimestampPairIndex]; + const bool recordTimestamps = tsStart && tsDisjoint && !swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex]; + + QD3D11CommandBuffer::Command &cmd(swapChainD->cb.commands.get()); + cmd.cmd = QD3D11CommandBuffer::Command::BeginFrame; + cmd.args.beginFrame.tsQuery = recordTimestamps ? tsStart : nullptr; + cmd.args.beginFrame.tsDisjointQuery = recordTimestamps ? tsDisjoint : nullptr; + cmd.args.beginFrame.swapchainData = rtData(&swapChainD->rt); + return QRhi::FrameOpSuccess; } @@ -1312,17 +1373,13 @@ QRhi::FrameOpResult QRhiD3D11::endFrame(QRhiSwapChain *swapChain, QRhi::EndFrame Q_ASSERT(contextState.currentSwapChain = swapChainD); const int currentFrameSlot = swapChainD->currentFrameSlot; - ID3D11Query *tsDisjoint = swapChainD->timestamps.disjointQuery[currentFrameSlot]; - const int tsIdx = QD3D11SwapChain::BUFFER_COUNT * currentFrameSlot; - ID3D11Query *tsStart = swapChainD->timestamps.query[tsIdx]; - ID3D11Query *tsEnd = swapChainD->timestamps.query[tsIdx + 1]; - const bool recordTimestamps = tsDisjoint && tsStart && tsEnd && !swapChainD->timestamps.active[currentFrameSlot]; + QD3D11CommandBuffer::Command &cmd(swapChainD->cb.commands.get()); + cmd.cmd = QD3D11CommandBuffer::Command::EndFrame; + cmd.args.endFrame.tsQuery = nullptr; // done later manually, see below + cmd.args.endFrame.tsDisjointQuery = nullptr; // send all commands to the context - if (recordTimestamps) - executeCommandBuffer(&swapChainD->cb, swapChainD); - else - executeCommandBuffer(&swapChainD->cb); + executeCommandBuffer(&swapChainD->cb); if (swapChainD->sampleDesc.Count > 1) { context->ResolveSubresource(swapChainD->backBufferTex, 0, @@ -1330,17 +1387,25 @@ QRhi::FrameOpResult QRhiD3D11::endFrame(QRhiSwapChain *swapChain, QRhi::EndFrame swapChainD->colorFormat); } - // this is here because we want to include the time spent on the resolve as well + // this is here because we want to include the time spent on the ResolveSubresource as well + ID3D11Query *tsEnd = swapChainD->timestamps.query[swapChainD->currentTimestampPairIndex * 2 + 1]; + ID3D11Query *tsDisjoint = swapChainD->timestamps.disjointQuery[swapChainD->currentTimestampPairIndex]; + const bool recordTimestamps = tsEnd && tsDisjoint && !swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex]; if (recordTimestamps) { context->End(tsEnd); context->End(tsDisjoint); - swapChainD->timestamps.active[currentFrameSlot] = true; + swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex] = true; + swapChainD->currentTimestampPairIndex = (swapChainD->currentTimestampPairIndex + 1) % QD3D11SwapChainTimestamps::TIMESTAMP_PAIRS; } if (!flags.testFlag(QRhi::SkipPresent)) { UINT presentFlags = 0; if (swapChainD->swapInterval == 0 && (swapChainD->swapChainFlags & DXGI_SWAP_CHAIN_FLAG_ALLOW_TEARING)) presentFlags |= DXGI_PRESENT_ALLOW_TEARING; + if (!swapChainD->swapChain) { + qWarning("Failed to present: IDXGISwapChain is unavailable"); + return QRhi::FrameOpError; + } HRESULT hr = swapChainD->swapChain->Present(swapChainD->swapInterval, presentFlags); if (hr == DXGI_ERROR_DEVICE_REMOVED || hr == DXGI_ERROR_DEVICE_RESET) { qWarning("Device loss detected in Present()"); @@ -1375,12 +1440,36 @@ QRhi::FrameOpResult QRhiD3D11::beginOffscreenFrame(QRhiCommandBuffer **cb, QRhi: ofr.cbWrapper.resetState(); *cb = &ofr.cbWrapper; - if (ofr.timestamps.active[ofr.timestampIdx]) { - double elapsedSec = 0; - if (ofr.timestamps.tryQueryTimestamps(ofr.timestampIdx, context, &elapsedSec)) - ofr.cbWrapper.lastGpuTime = elapsedSec; + if (rhiFlags.testFlag(QRhi::EnableTimestamps)) { + D3D11_QUERY_DESC queryDesc = {}; + if (!ofr.tsDisjointQuery) { + queryDesc.Query = D3D11_QUERY_TIMESTAMP_DISJOINT; + HRESULT hr = dev->CreateQuery(&queryDesc, &ofr.tsDisjointQuery); + if (FAILED(hr)) { + qWarning("Failed to create timestamp disjoint query: %s", + qPrintable(QSystemError::windowsComString(hr))); + return QRhi::FrameOpError; + } + } + queryDesc.Query = D3D11_QUERY_TIMESTAMP; + for (int i = 0; i < 2; ++i) { + if (!ofr.tsQueries[i]) { + HRESULT hr = dev->CreateQuery(&queryDesc, &ofr.tsQueries[i]); + if (FAILED(hr)) { + qWarning("Failed to create timestamp query: %s", + qPrintable(QSystemError::windowsComString(hr))); + return QRhi::FrameOpError; + } + } + } } + QD3D11CommandBuffer::Command &cmd(ofr.cbWrapper.commands.get()); + cmd.cmd = QD3D11CommandBuffer::Command::BeginFrame; + cmd.args.beginFrame.tsQuery = ofr.tsQueries[0] ? ofr.tsQueries[0] : nullptr; + cmd.args.beginFrame.tsDisjointQuery = ofr.tsDisjointQuery ? ofr.tsDisjointQuery : nullptr; + cmd.args.beginFrame.swapchainData = nullptr; + return QRhi::FrameOpSuccess; } @@ -1389,25 +1478,39 @@ QRhi::FrameOpResult QRhiD3D11::endOffscreenFrame(QRhi::EndFrameFlags flags) Q_UNUSED(flags); ofr.active = false; - ID3D11Query *tsDisjoint = ofr.timestamps.disjointQuery[ofr.timestampIdx]; - ID3D11Query *tsStart = ofr.timestamps.query[ofr.timestampIdx * 2]; - ID3D11Query *tsEnd = ofr.timestamps.query[ofr.timestampIdx * 2 + 1]; - const bool recordTimestamps = tsDisjoint && tsStart && tsEnd && !ofr.timestamps.active[ofr.timestampIdx]; - if (recordTimestamps) { - context->Begin(tsDisjoint); - context->End(tsStart); // record timestamp; no Begin() for D3D11_QUERY_TIMESTAMP - } + QD3D11CommandBuffer::Command &cmd(ofr.cbWrapper.commands.get()); + cmd.cmd = QD3D11CommandBuffer::Command::EndFrame; + cmd.args.endFrame.tsQuery = ofr.tsQueries[1] ? ofr.tsQueries[1] : nullptr; + cmd.args.endFrame.tsDisjointQuery = ofr.tsDisjointQuery ? ofr.tsDisjointQuery : nullptr; executeCommandBuffer(&ofr.cbWrapper); context->Flush(); finishActiveReadbacks(); - if (recordTimestamps) { - context->End(tsEnd); - context->End(tsDisjoint); - ofr.timestamps.active[ofr.timestampIdx] = true; - ofr.timestampIdx = (ofr.timestampIdx + 1) % 2; + if (ofr.tsQueries[0]) { + quint64 timestamps[2]; + D3D11_QUERY_DATA_TIMESTAMP_DISJOINT dj; + HRESULT hr; + bool ok = true; + do { + hr = context->GetData(ofr.tsDisjointQuery, &dj, sizeof(dj), 0); + } while (hr == S_FALSE); + ok &= hr == S_OK; + do { + hr = context->GetData(ofr.tsQueries[1], ×tamps[1], sizeof(quint64), 0); + } while (hr == S_FALSE); + ok &= hr == S_OK; + do { + hr = context->GetData(ofr.tsQueries[0], ×tamps[0], sizeof(quint64), 0); + } while (hr == S_FALSE); + ok &= hr == S_OK; + if (ok) { + if (!dj.Disjoint && dj.Frequency) { + const float elapsedMs = (timestamps[1] - timestamps[0]) / float(dj.Frequency) * 1000.0f; + ofr.cbWrapper.lastGpuTime = elapsedMs / 1000.0; + } + } } return QRhi::FrameOpSuccess; @@ -1447,9 +1550,9 @@ static inline DXGI_FORMAT toD3DTextureFormat(QRhiTexture::Format format, QRhiTex case QRhiTexture::D16: return DXGI_FORMAT_R16_TYPELESS; case QRhiTexture::D24: - return DXGI_FORMAT_R24_UNORM_X8_TYPELESS; + return DXGI_FORMAT_R24G8_TYPELESS; case QRhiTexture::D24S8: - return DXGI_FORMAT_D24_UNORM_S8_UINT; + return DXGI_FORMAT_R24G8_TYPELESS; case QRhiTexture::D32F: return DXGI_FORMAT_R32_TYPELESS; @@ -1550,7 +1653,7 @@ QRhi::FrameOpResult QRhiD3D11::finish() } else { Q_ASSERT(contextState.currentSwapChain); Q_ASSERT(contextState.currentSwapChain->cb.recordingPass == QD3D11CommandBuffer::NoPass); - executeCommandBuffer(&contextState.currentSwapChain->cb); // no timestampSwapChain, in order to avoid timestamp mess + executeCommandBuffer(&contextState.currentSwapChain->cb); contextState.currentSwapChain->cb.resetCommands(); } } @@ -1925,19 +2028,6 @@ void QRhiD3D11::finishActiveReadbacks() f(); } -static inline QD3D11RenderTargetData *rtData(QRhiRenderTarget *rt) -{ - switch (rt->resourceType()) { - case QRhiResource::SwapChainRenderTarget: - return &QRHI_RES(QD3D11SwapChainRenderTarget, rt)->d; - case QRhiResource::TextureRenderTarget: - return &QRHI_RES(QD3D11TextureRenderTarget, rt)->d; - default: - Q_UNREACHABLE(); - return nullptr; - } -} - void QRhiD3D11::resourceUpdate(QRhiCommandBuffer *cb, QRhiResourceUpdateBatch *resourceUpdates) { Q_ASSERT(QRHI_RES(QD3D11CommandBuffer, cb)->recordingPass == QD3D11CommandBuffer::NoPass); @@ -2056,6 +2146,8 @@ void QRhiD3D11::endPass(QRhiCommandBuffer *cb, QRhiResourceUpdateBatch *resource cmd.args.resolveSubRes.srcSubRes = D3D11CalcSubresource(0, UINT(colorAtt.layer()), 1); cmd.args.resolveSubRes.format = dstTexD->dxgiFormat; } + if (rtTex->m_desc.depthResolveTexture()) + qWarning("Resolving multisample depth-stencil buffers is not supported with D3D"); } cbD->recordingPass = QD3D11CommandBuffer::NoPass; @@ -2643,7 +2735,7 @@ void QRhiD3D11::resetShaderResources() currentShaderMask &= ~StageU##MaskBit; \ } -void QRhiD3D11::executeCommandBuffer(QD3D11CommandBuffer *cbD, QD3D11SwapChain *timestampSwapChain) +void QRhiD3D11::executeCommandBuffer(QD3D11CommandBuffer *cbD) { quint32 stencilRef = 0; float blendConstants[] = { 1, 1, 1, 1 }; @@ -2656,26 +2748,30 @@ void QRhiD3D11::executeCommandBuffer(QD3D11CommandBuffer *cbD, QD3D11SwapChain * }; int currentShaderMask = 0xFF; - if (timestampSwapChain) { - const int currentFrameSlot = timestampSwapChain->currentFrameSlot; - ID3D11Query *tsDisjoint = timestampSwapChain->timestamps.disjointQuery[currentFrameSlot]; - const int tsIdx = QD3D11SwapChain::BUFFER_COUNT * currentFrameSlot; - ID3D11Query *tsStart = timestampSwapChain->timestamps.query[tsIdx]; - if (tsDisjoint && tsStart && !timestampSwapChain->timestamps.active[currentFrameSlot]) { - // The timestamps seem to include vsync time with Present(1), except - // when running on a non-primary gpu. This is not ideal. So try working - // it around by issuing a semi-fake OMSetRenderTargets early and - // writing the first timestamp only afterwards. - context->Begin(tsDisjoint); - QD3D11RenderTargetData *rtD = rtData(×tampSwapChain->rt); - context->OMSetRenderTargets(UINT(rtD->colorAttCount), rtD->colorAttCount ? rtD->rtv : nullptr, rtD->dsv); - context->End(tsStart); // just record a timestamp, no Begin needed - } - } - for (auto it = cbD->commands.cbegin(), end = cbD->commands.cend(); it != end; ++it) { const QD3D11CommandBuffer::Command &cmd(*it); switch (cmd.cmd) { + case QD3D11CommandBuffer::Command::BeginFrame: + if (cmd.args.beginFrame.tsDisjointQuery) + context->Begin(cmd.args.beginFrame.tsDisjointQuery); + if (cmd.args.beginFrame.tsQuery) { + if (cmd.args.beginFrame.swapchainData) { + // The timestamps seem to include vsync time with Present(1), except + // when running on a non-primary gpu. This is not ideal. So try working + // it around by issuing a semi-fake OMSetRenderTargets early and + // writing the first timestamp only afterwards. + QD3D11RenderTargetData *rtD = cmd.args.beginFrame.swapchainData; + context->OMSetRenderTargets(UINT(rtD->colorAttCount), rtD->colorAttCount ? rtD->rtv : nullptr, rtD->dsv); + } + context->End(cmd.args.beginFrame.tsQuery); // no Begin() for D3D11_QUERY_TIMESTAMP + } + break; + case QD3D11CommandBuffer::Command::EndFrame: + if (cmd.args.endFrame.tsQuery) + context->End(cmd.args.endFrame.tsQuery); + if (cmd.args.endFrame.tsDisjointQuery) + context->End(cmd.args.endFrame.tsDisjointQuery); + break; case QD3D11CommandBuffer::Command::ResetShaderResources: resetShaderResources(); break; @@ -3029,7 +3125,7 @@ bool QD3D11RenderBuffer::create() return false; QRHI_RES_RHI(QRhiD3D11); - sampleDesc = rhiD->effectiveSampleCount(m_sampleCount); + sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount); D3D11_TEXTURE2D_DESC desc = {}; desc.Width = UINT(m_pixelSize.width()); @@ -3171,7 +3267,7 @@ static inline DXGI_FORMAT toD3DDepthTextureDSVFormat(QRhiTexture::Format format) case QRhiTexture::Format::D16: return DXGI_FORMAT_D16_UNORM; case QRhiTexture::Format::D24: - return DXGI_FORMAT_R24_UNORM_X8_TYPELESS; + return DXGI_FORMAT_D24_UNORM_S8_UINT; case QRhiTexture::Format::D24S8: return DXGI_FORMAT_D24_UNORM_S8_UINT; case QRhiTexture::Format::D32F: @@ -3200,7 +3296,7 @@ bool QD3D11Texture::prepareCreate(QSize *adjustedSize) QRHI_RES_RHI(QRhiD3D11); dxgiFormat = toD3DTextureFormat(m_format, m_flags); mipLevelCount = uint(hasMipMaps ? rhiD->q->mipLevelsForSize(size) : 1); - sampleDesc = rhiD->effectiveSampleCount(m_sampleCount); + sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount); if (sampleDesc.Count > 1) { if (isCube) { qWarning("Cubemap texture cannot be multisample"); @@ -3834,6 +3930,27 @@ bool QD3D11TextureRenderTarget::create() dsvDesc.Format = toD3DDepthTextureDSVFormat(depthTexD->format()); dsvDesc.ViewDimension = depthTexD->sampleDesc.Count > 1 ? D3D11_DSV_DIMENSION_TEXTURE2DMS : D3D11_DSV_DIMENSION_TEXTURE2D; + if (depthTexD->flags().testFlag(QRhiTexture::TextureArray)) { + if (depthTexD->sampleDesc.Count > 1) { + dsvDesc.ViewDimension = D3D11_DSV_DIMENSION_TEXTURE2DMSARRAY; + if (depthTexD->arrayRangeStart() >= 0 && depthTexD->arrayRangeLength() >= 0) { + dsvDesc.Texture2DMSArray.FirstArraySlice = UINT(depthTexD->arrayRangeStart()); + dsvDesc.Texture2DMSArray.ArraySize = UINT(depthTexD->arrayRangeLength()); + } else { + dsvDesc.Texture2DMSArray.FirstArraySlice = 0; + dsvDesc.Texture2DMSArray.ArraySize = UINT(qMax(0, depthTexD->arraySize())); + } + } else { + dsvDesc.ViewDimension = D3D11_DSV_DIMENSION_TEXTURE2DARRAY; + if (depthTexD->arrayRangeStart() >= 0 && depthTexD->arrayRangeLength() >= 0) { + dsvDesc.Texture2DArray.FirstArraySlice = UINT(depthTexD->arrayRangeStart()); + dsvDesc.Texture2DArray.ArraySize = UINT(depthTexD->arrayRangeLength()); + } else { + dsvDesc.Texture2DArray.FirstArraySlice = 0; + dsvDesc.Texture2DArray.ArraySize = UINT(qMax(0, depthTexD->arraySize())); + } + } + } HRESULT hr = rhiD->dev->CreateDepthStencilView(depthTexD->tex, &dsvDesc, &dsv); if (FAILED(hr)) { qWarning("Failed to create dsv: %s", @@ -4128,6 +4245,22 @@ static inline DXGI_FORMAT toD3DAttributeFormat(QRhiVertexInputAttribute::Format return DXGI_FORMAT_R16G16_FLOAT; case QRhiVertexInputAttribute::Half: return DXGI_FORMAT_R16_FLOAT; + case QRhiVertexInputAttribute::UShort4: + // Note: D3D does not support UShort3. Pass through UShort3 as UShort4. + case QRhiVertexInputAttribute::UShort3: + return DXGI_FORMAT_R16G16B16A16_UINT; + case QRhiVertexInputAttribute::UShort2: + return DXGI_FORMAT_R16G16_UINT; + case QRhiVertexInputAttribute::UShort: + return DXGI_FORMAT_R16_UINT; + case QRhiVertexInputAttribute::SShort4: + // Note: D3D does not support SShort3. Pass through SShort3 as SShort4. + case QRhiVertexInputAttribute::SShort3: + return DXGI_FORMAT_R16G16B16A16_SINT; + case QRhiVertexInputAttribute::SShort2: + return DXGI_FORMAT_R16G16_SINT; + case QRhiVertexInputAttribute::SShort: + return DXGI_FORMAT_R16_SINT; default: Q_UNREACHABLE(); return DXGI_FORMAT_R32G32B32A32_FLOAT; @@ -4240,18 +4373,6 @@ static inline D3D11_BLEND_OP toD3DBlendOp(QRhiGraphicsPipeline::BlendOp op) } } -static pD3DCompile resolveD3DCompile() -{ - for (const wchar_t *libraryName : {L"D3DCompiler_47", L"D3DCompiler_43"}) { - QSystemLibrary library(libraryName); - if (library.load()) { - if (auto symbol = library.resolve("D3DCompile")) - return reinterpret_cast<pD3DCompile>(symbol); - } - } - return nullptr; -} - static inline QByteArray sourceHash(const QByteArray &source) { // taken from the GL backend, use the same mechanism to get a key @@ -4317,7 +4438,7 @@ QByteArray QRhiD3D11::compileHlslShaderSource(const QShader &shader, QShader::Va return cacheIt.value(); } - static const pD3DCompile d3dCompile = resolveD3DCompile(); + static const pD3DCompile d3dCompile = QRhiD3D::resolveD3DCompile(); if (d3dCompile == nullptr) { qWarning("Unable to resolve function D3DCompile()"); return QByteArray(); @@ -4367,7 +4488,7 @@ bool QD3D11GraphicsPipeline::create() rastDesc.SlopeScaledDepthBias = m_slopeScaledDepthBias; rastDesc.DepthClipEnable = true; rastDesc.ScissorEnable = m_flags.testFlag(UsesScissor); - rastDesc.MultisampleEnable = rhiD->effectiveSampleCount(m_sampleCount).Count > 1; + rastDesc.MultisampleEnable = rhiD->effectiveSampleDesc(m_sampleCount).Count > 1; HRESULT hr = rhiD->dev->CreateRasterizerState(&rastDesc, &rastState); if (FAILED(hr)) { qWarning("Failed to create rasterizer state: %s", @@ -4685,14 +4806,13 @@ void QD3D11CommandBuffer::destroy() // nothing to do here } -bool QD3D11Timestamps::prepare(int pairCount, QRhiD3D11 *rhiD) +bool QD3D11SwapChainTimestamps::prepare(QRhiD3D11 *rhiD) { // Creates the query objects if not yet done, but otherwise calling this // function is expected to be a no-op. - Q_ASSERT(pairCount <= MAX_TIMESTAMP_PAIRS); D3D11_QUERY_DESC queryDesc = {}; - for (int i = 0; i < pairCount; ++i) { + for (int i = 0; i < TIMESTAMP_PAIRS; ++i) { if (!disjointQuery[i]) { queryDesc.Query = D3D11_QUERY_TIMESTAMP_DISJOINT; HRESULT hr = rhiD->dev->CreateQuery(&queryDesc, &disjointQuery[i]); @@ -4704,7 +4824,7 @@ bool QD3D11Timestamps::prepare(int pairCount, QRhiD3D11 *rhiD) } queryDesc.Query = D3D11_QUERY_TIMESTAMP; for (int j = 0; j < 2; ++j) { - const int idx = pairCount * i + j; + const int idx = 2 * i + j; if (!query[idx]) { HRESULT hr = rhiD->dev->CreateQuery(&queryDesc, &query[idx]); if (FAILED(hr)) { @@ -4715,20 +4835,19 @@ bool QD3D11Timestamps::prepare(int pairCount, QRhiD3D11 *rhiD) } } } - this->pairCount = pairCount; return true; } -void QD3D11Timestamps::destroy() +void QD3D11SwapChainTimestamps::destroy() { - for (int i = 0; i < MAX_TIMESTAMP_PAIRS; ++i) { + for (int i = 0; i < TIMESTAMP_PAIRS; ++i) { active[i] = false; if (disjointQuery[i]) { disjointQuery[i]->Release(); disjointQuery[i] = nullptr; } for (int j = 0; j < 2; ++j) { - const int idx = MAX_TIMESTAMP_PAIRS * i + j; + const int idx = TIMESTAMP_PAIRS * i + j; if (query[idx]) { query[idx]->Release(); query[idx] = nullptr; @@ -4737,26 +4856,21 @@ void QD3D11Timestamps::destroy() } } -bool QD3D11Timestamps::tryQueryTimestamps(int idx, ID3D11DeviceContext *context, double *elapsedSec) +bool QD3D11SwapChainTimestamps::tryQueryTimestamps(int pairIndex, ID3D11DeviceContext *context, double *elapsedSec) { bool result = false; - if (!active[idx]) + if (!active[pairIndex]) return result; - ID3D11Query *tsDisjoint = disjointQuery[idx]; - const int tsIdx = pairCount * idx; - ID3D11Query *tsStart = query[tsIdx]; - ID3D11Query *tsEnd = query[tsIdx + 1]; + ID3D11Query *tsDisjoint = disjointQuery[pairIndex]; + ID3D11Query *tsStart = query[pairIndex * 2]; + ID3D11Query *tsEnd = query[pairIndex * 2 + 1]; quint64 timestamps[2]; D3D11_QUERY_DATA_TIMESTAMP_DISJOINT dj; bool ok = true; ok &= context->GetData(tsDisjoint, &dj, sizeof(dj), D3D11_ASYNC_GETDATA_DONOTFLUSH) == S_OK; ok &= context->GetData(tsEnd, ×tamps[1], sizeof(quint64), D3D11_ASYNC_GETDATA_DONOTFLUSH) == S_OK; - // this above is often not ready, not even in frame_where_recorded+2, - // not clear why. so make the whole thing async and do not touch the - // queries until they are finally all available in frame this+2 or - // this+4 or ... ok &= context->GetData(tsStart, ×tamps[0], sizeof(quint64), D3D11_ASYNC_GETDATA_DONOTFLUSH) == S_OK; if (ok) { @@ -4765,16 +4879,14 @@ bool QD3D11Timestamps::tryQueryTimestamps(int idx, ID3D11DeviceContext *context, *elapsedSec = elapsedMs / 1000.0; result = true; } - active[idx] = false; - } // else leave active set, will retry in a subsequent beginFrame or similar + active[pairIndex] = false; + } // else leave active set, will retry in a subsequent beginFrame return result; } QD3D11SwapChain::QD3D11SwapChain(QRhiImplementation *rhi) - : QRhiSwapChain(rhi), - rt(rhi, this), - cb(rhi) + : QRhiSwapChain(rhi), rt(rhi, this), rtRight(rhi, this), cb(rhi) { backBufferTex = nullptr; backBufferRtv = nullptr; @@ -4795,6 +4907,10 @@ void QD3D11SwapChain::releaseBuffers() backBufferRtv->Release(); backBufferRtv = nullptr; } + if (backBufferRtvRight) { + backBufferRtvRight->Release(); + backBufferRtvRight = nullptr; + } if (backBufferTex) { backBufferTex->Release(); backBufferTex = nullptr; @@ -4834,8 +4950,12 @@ void QD3D11SwapChain::destroy() } QRHI_RES_RHI(QRhiD3D11); - if (rhiD) + if (rhiD) { rhiD->unregisterResource(this); + // See Deferred Destruction Issues with Flip Presentation Swap Chains in + // https://learn.microsoft.com/en-us/windows/win32/api/d3d11/nf-d3d11-id3d11devicecontext-flush + rhiD->context->Flush(); + } } QRhiCommandBuffer *QD3D11SwapChain::currentFrameCommandBuffer() @@ -4848,48 +4968,15 @@ QRhiRenderTarget *QD3D11SwapChain::currentFrameRenderTarget() return &rt; } -QSize QD3D11SwapChain::surfacePixelSize() +QRhiRenderTarget *QD3D11SwapChain::currentFrameRenderTarget(StereoTargetBuffer targetBuffer) { - Q_ASSERT(m_window); - return m_window->size() * m_window->devicePixelRatio(); + return targetBuffer == StereoTargetBuffer::LeftBuffer? &rt: &rtRight; } -static bool output6ForWindow(QWindow *w, IDXGIAdapter1 *adapter, IDXGIOutput6 **result) -{ - bool ok = false; - QRect wr = w->geometry(); - wr = QRect(wr.topLeft() * w->devicePixelRatio(), wr.size() * w->devicePixelRatio()); - const QPoint center = wr.center(); - IDXGIOutput *currentOutput = nullptr; - IDXGIOutput *output = nullptr; - for (UINT i = 0; adapter->EnumOutputs(i, &output) != DXGI_ERROR_NOT_FOUND; ++i) { - DXGI_OUTPUT_DESC desc; - output->GetDesc(&desc); - const RECT r = desc.DesktopCoordinates; - const QRect dr(QPoint(r.left, r.top), QPoint(r.right - 1, r.bottom - 1)); - if (dr.contains(center)) { - currentOutput = output; - break; - } else { - output->Release(); - } - } - if (currentOutput) { - ok = SUCCEEDED(currentOutput->QueryInterface(__uuidof(IDXGIOutput6), reinterpret_cast<void **>(result))); - currentOutput->Release(); - } - return ok; -} - -static bool outputDesc1ForWindow(QWindow *w, IDXGIAdapter1 *adapter, DXGI_OUTPUT_DESC1 *result) +QSize QD3D11SwapChain::surfacePixelSize() { - bool ok = false; - IDXGIOutput6 *out6 = nullptr; - if (output6ForWindow(w, adapter, &out6)) { - ok = SUCCEEDED(out6->GetDesc1(result)); - out6->Release(); - } - return ok; + Q_ASSERT(m_window); + return m_window->size() * m_window->devicePixelRatio(); } bool QD3D11SwapChain::isFormatSupported(Format f) @@ -4904,7 +4991,7 @@ bool QD3D11SwapChain::isFormatSupported(Format f) QRHI_RES_RHI(QRhiD3D11); DXGI_OUTPUT_DESC1 desc1; - if (outputDesc1ForWindow(m_window, rhiD->activeAdapter, &desc1)) { + if (QRhiD3D::outputDesc1ForWindow(m_window, rhiD->activeAdapter, &desc1)) { if (desc1.ColorSpace == DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020) return f == QRhiSwapChain::HDRExtendedSrgbLinear || f == QRhiSwapChain::HDR10; } @@ -4915,14 +5002,16 @@ bool QD3D11SwapChain::isFormatSupported(Format f) QRhiSwapChainHdrInfo QD3D11SwapChain::hdrInfo() { QRhiSwapChainHdrInfo info = QRhiSwapChain::hdrInfo(); - if (m_format != QRhiSwapChain::SDR && m_window) { + // Must use m_window, not window, given this may be called before createOrResize(). + if (m_window) { QRHI_RES_RHI(QRhiD3D11); DXGI_OUTPUT_DESC1 hdrOutputDesc; - if (outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc)) { - info.isHardCodedDefaults = false; + if (QRhiD3D::outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc)) { info.limitsType = QRhiSwapChainHdrInfo::LuminanceInNits; info.limits.luminanceInNits.minLuminance = hdrOutputDesc.MinLuminance; info.limits.luminanceInNits.maxLuminance = hdrOutputDesc.MaxLuminance; + info.luminanceBehavior = QRhiSwapChainHdrInfo::SceneReferred; // 1.0 = 80 nits + info.sdrWhiteLevel = QRhiD3D::sdrWhiteLevelInNits(hdrOutputDesc); } } return info; @@ -4972,26 +5061,19 @@ bool QD3D11SwapChain::newColorBuffer(const QSize &size, DXGI_FORMAT format, DXGI return true; } -static const DXGI_FORMAT DEFAULT_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM; -static const DXGI_FORMAT DEFAULT_SRGB_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM_SRGB; - bool QRhiD3D11::ensureDirectCompositionDevice() { if (dcompDevice) return true; qCDebug(QRHI_LOG_INFO, "Creating Direct Composition device (needed for semi-transparent windows)"); - - HRESULT hr = DCompositionCreateDevice(nullptr, __uuidof(IDCompositionDevice), reinterpret_cast<void **>(&dcompDevice)); - if (FAILED(hr)) { - qWarning("Failed to Direct Composition device: %s", - qPrintable(QSystemError::windowsComString(hr))); - return false; - } - - return true; + dcompDevice = QRhiD3D::createDirectCompositionDevice(); + return dcompDevice ? true : false; } +static const DXGI_FORMAT DEFAULT_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM; +static const DXGI_FORMAT DEFAULT_SRGB_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM_SRGB; + bool QD3D11SwapChain::createOrResize() { // Can be called multiple times due to window resizes - that is not the @@ -4999,6 +5081,7 @@ bool QD3D11SwapChain::createOrResize() // resize the buffers then. const bool needsRegistration = !window || window != m_window; + const bool stereo = m_window->format().stereo(); // except if the window actually changes if (window && window != m_window) @@ -5017,9 +5100,9 @@ bool QD3D11SwapChain::createOrResize() QRHI_RES_RHI(QRhiD3D11); if (m_flags.testFlag(SurfaceHasPreMulAlpha) || m_flags.testFlag(SurfaceHasNonPreMulAlpha)) { - if (rhiD->ensureDirectCompositionDevice()) { + if (!rhiD->useLegacySwapchainModel && rhiD->ensureDirectCompositionDevice()) { if (!dcompTarget) { - hr = rhiD->dcompDevice->CreateTargetForHwnd(hwnd, true, &dcompTarget); + hr = rhiD->dcompDevice->CreateTargetForHwnd(hwnd, false, &dcompTarget); if (FAILED(hr)) { qWarning("Failed to create Direct Compsition target for the window: %s", qPrintable(QSystemError::windowsComString(hr))); @@ -5050,13 +5133,13 @@ bool QD3D11SwapChain::createOrResize() swapChainFlags |= DXGI_SWAP_CHAIN_FLAG_ALLOW_TEARING; if (!swapChain) { - sampleDesc = rhiD->effectiveSampleCount(m_sampleCount); + sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount); colorFormat = DEFAULT_FORMAT; srgbAdjustedColorFormat = m_flags.testFlag(sRGB) ? DEFAULT_SRGB_FORMAT : DEFAULT_FORMAT; DXGI_COLOR_SPACE_TYPE hdrColorSpace = DXGI_COLOR_SPACE_RGB_FULL_G22_NONE_P709; // SDR DXGI_OUTPUT_DESC1 hdrOutputDesc; - if (outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc) && m_format != SDR) { + if (QRhiD3D::outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc) && m_format != SDR) { // https://docs.microsoft.com/en-us/windows/win32/direct3darticles/high-dynamic-range if (hdrOutputDesc.ColorSpace == DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020) { switch (m_format) { @@ -5096,8 +5179,9 @@ bool QD3D11SwapChain::createOrResize() desc.BufferUsage = DXGI_USAGE_RENDER_TARGET_OUTPUT; desc.BufferCount = BUFFER_COUNT; desc.Flags = swapChainFlags; - desc.Scaling = DXGI_SCALING_NONE; - desc.SwapEffect = DXGI_SWAP_EFFECT_FLIP_DISCARD; + desc.Scaling = rhiD->useLegacySwapchainModel ? DXGI_SCALING_STRETCH : DXGI_SCALING_NONE; + desc.SwapEffect = rhiD->useLegacySwapchainModel ? DXGI_SWAP_EFFECT_DISCARD : DXGI_SWAP_EFFECT_FLIP_DISCARD; + desc.Stereo = stereo; if (dcompVisual) { // With DirectComposition setting AlphaMode to STRAIGHT fails the @@ -5157,14 +5241,19 @@ bool QD3D11SwapChain::createOrResize() qWarning("Failed to set content for Direct Composition visual: %s", qPrintable(QSystemError::windowsComString(hr))); } + } else { + // disable Alt+Enter; not relevant when using DirectComposition + rhiD->dxgiFactory->MakeWindowAssociation(hwnd, DXGI_MWA_NO_WINDOW_CHANGES); } } if (FAILED(hr)) { - qWarning("Failed to create D3D11 swapchain: %s", - qPrintable(QSystemError::windowsComString(hr))); + qWarning("Failed to create D3D11 swapchain: %s" + " (Width=%u Height=%u Format=%u SampleCount=%u BufferCount=%u Scaling=%u SwapEffect=%u Stereo=%u)", + qPrintable(QSystemError::windowsComString(hr)), + desc.Width, desc.Height, UINT(desc.Format), desc.SampleDesc.Count, + desc.BufferCount, UINT(desc.Scaling), UINT(desc.SwapEffect), UINT(desc.Stereo)); return false; } - rhiD->dxgiFactory->MakeWindowAssociation(hwnd, DXGI_MWA_NO_WINDOW_CHANGES); } else { releaseBuffers(); // flip model -> buffer count is the real buffer count, not 1 like with the legacy modes @@ -5211,6 +5300,19 @@ bool QD3D11SwapChain::createOrResize() return false; } + if (stereo) { + // Create a second render target view for the right eye + rtvDesc.ViewDimension = D3D11_RTV_DIMENSION_TEXTURE2DARRAY; + rtvDesc.Texture2DArray.FirstArraySlice = 1; + rtvDesc.Texture2DArray.ArraySize = 1; + hr = rhiD->dev->CreateRenderTargetView(backBufferTex, &rtvDesc, &backBufferRtvRight); + if (FAILED(hr)) { + qWarning("Failed to create rtv for swapchain backbuffer (right eye): %s", + qPrintable(QSystemError::windowsComString(hr))); + return false; + } + } + // Try to reduce stalls by having a dedicated MSAA texture per swapchain buffer. for (int i = 0; i < BUFFER_COUNT; ++i) { if (sampleDesc.Count > 1) { @@ -5249,8 +5351,20 @@ bool QD3D11SwapChain::createOrResize() rtD->d.colorAttCount = 1; rtD->d.dsAttCount = m_depthStencil ? 1 : 0; + if (stereo) { + rtD = QRHI_RES(QD3D11SwapChainRenderTarget, &rtRight); + rtD->d.rp = QRHI_RES(QD3D11RenderPassDescriptor, m_renderPassDesc); + rtD->d.pixelSize = pixelSize; + rtD->d.dpr = float(window->devicePixelRatio()); + rtD->d.sampleCount = int(sampleDesc.Count); + rtD->d.colorAttCount = 1; + rtD->d.dsAttCount = m_depthStencil ? 1 : 0; + rtD->d.rtv[0] = backBufferRtvRight; + rtD->d.dsv = ds ? ds->dsv : nullptr; + } + if (rhiD->rhiFlags.testFlag(QRhi::EnableTimestamps)) { - timestamps.prepare(BUFFER_COUNT, rhiD); + timestamps.prepare(rhiD); // timestamp queries are optional so we can go on even if they failed } diff --git a/src/gui/rhi/qrhid3d11_p.h b/src/gui/rhi/qrhid3d11_p.h index e410375029..7644748407 100644 --- a/src/gui/rhi/qrhid3d11_p.h +++ b/src/gui/rhi/qrhid3d11_p.h @@ -16,7 +16,7 @@ // #include "qrhi_p.h" -#include "qshaderdescription.h" +#include <rhi/qshaderdescription.h> #include <QWindow> #include <d3d11_1.h> @@ -356,6 +356,8 @@ struct QD3D11CommandBuffer : public QRhiCommandBuffer struct Command { enum Cmd { + BeginFrame, + EndFrame, ResetShaderResources, SetRenderTarget, Clear, @@ -386,6 +388,15 @@ struct QD3D11CommandBuffer : public QRhiCommandBuffer // QRhiTexture/Buffer/etc. pointers). union Args { struct { + ID3D11Query *tsQuery; + ID3D11Query *tsDisjointQuery; + QD3D11RenderTargetData *swapchainData; + } beginFrame; + struct { + ID3D11Query *tsQuery; + ID3D11Query *tsDisjointQuery; + } endFrame; + struct { QRhiRenderTarget *rt; } setRenderTarget; struct { @@ -556,17 +567,15 @@ struct QD3D11CommandBuffer : public QRhiCommandBuffer } }; -static const int QD3D11_SWAPCHAIN_BUFFER_COUNT = 2; - -struct QD3D11Timestamps +struct QD3D11SwapChainTimestamps { - static const int MAX_TIMESTAMP_PAIRS = QD3D11_SWAPCHAIN_BUFFER_COUNT; - bool active[MAX_TIMESTAMP_PAIRS] = {}; - ID3D11Query *disjointQuery[MAX_TIMESTAMP_PAIRS] = {}; - ID3D11Query *query[MAX_TIMESTAMP_PAIRS * 2] = {}; - int pairCount = 0; + static const int TIMESTAMP_PAIRS = 2; + + bool active[TIMESTAMP_PAIRS] = {}; + ID3D11Query *disjointQuery[TIMESTAMP_PAIRS] = {}; + ID3D11Query *query[TIMESTAMP_PAIRS * 2] = {}; - bool prepare(int pairCount, QRhiD3D11 *rhiD); + bool prepare(QRhiD3D11 *rhiD); void destroy(); bool tryQueryTimestamps(int idx, ID3D11DeviceContext *context, double *elapsedSec); }; @@ -579,6 +588,7 @@ struct QD3D11SwapChain : public QRhiSwapChain QRhiCommandBuffer *currentFrameCommandBuffer() override; QRhiRenderTarget *currentFrameRenderTarget() override; + QRhiRenderTarget *currentFrameRenderTarget(StereoTargetBuffer targetBuffer) override; QSize surfacePixelSize() override; bool isFormatSupported(Format f) override; @@ -594,6 +604,7 @@ struct QD3D11SwapChain : public QRhiSwapChain QWindow *window = nullptr; QSize pixelSize; QD3D11SwapChainRenderTarget rt; + QD3D11SwapChainRenderTarget rtRight; QD3D11CommandBuffer cb; DXGI_FORMAT colorFormat; DXGI_FORMAT srgbAdjustedColorFormat; @@ -601,7 +612,8 @@ struct QD3D11SwapChain : public QRhiSwapChain UINT swapChainFlags = 0; ID3D11Texture2D *backBufferTex; ID3D11RenderTargetView *backBufferRtv; - static const int BUFFER_COUNT = QD3D11_SWAPCHAIN_BUFFER_COUNT; + ID3D11RenderTargetView *backBufferRtvRight = nullptr; + static const int BUFFER_COUNT = 2; ID3D11Texture2D *msaaTex[BUFFER_COUNT]; ID3D11RenderTargetView *msaaRtv[BUFFER_COUNT]; DXGI_SAMPLE_DESC sampleDesc; @@ -611,7 +623,8 @@ struct QD3D11SwapChain : public QRhiSwapChain UINT swapInterval = 1; IDCompositionTarget *dcompTarget = nullptr; IDCompositionVisual *dcompVisual = nullptr; - QD3D11Timestamps timestamps; + QD3D11SwapChainTimestamps timestamps; + int currentTimestampPairIndex = 0; }; class QRhiD3D11 : public QRhiImplementation @@ -736,8 +749,8 @@ public: const uint *dynOfsPairs, int dynOfsPairCount, bool offsetOnlyChange); void resetShaderResources(); - void executeCommandBuffer(QD3D11CommandBuffer *cbD, QD3D11SwapChain *timestampSwapChain = nullptr); - DXGI_SAMPLE_DESC effectiveSampleCount(int sampleCount) const; + void executeCommandBuffer(QD3D11CommandBuffer *cbD); + DXGI_SAMPLE_DESC effectiveSampleDesc(int sampleCount) const; void finishActiveReadbacks(); void reportLiveObjects(ID3D11Device *device); void clearShaderCache(); @@ -757,6 +770,7 @@ public: IDXGIFactory1 *dxgiFactory = nullptr; IDCompositionDevice *dcompDevice = nullptr; bool supportsAllowTearing = false; + bool useLegacySwapchainModel = false; bool deviceLost = false; QRhiD3D11NativeHandles nativeHandlesStruct; QRhiDriverInfo driverInfoStruct; @@ -778,8 +792,8 @@ public: OffscreenFrame(QRhiImplementation *rhi) : cbWrapper(rhi) { } bool active = false; QD3D11CommandBuffer cbWrapper; - QD3D11Timestamps timestamps; - int timestampIdx = 0; + ID3D11Query *tsQueries[2] = {}; + ID3D11Query *tsDisjointQuery = nullptr; } ofr; struct TextureReadback { diff --git a/src/gui/rhi/qrhid3d12.cpp b/src/gui/rhi/qrhid3d12.cpp index 19843ed82b..0f176c683d 100644 --- a/src/gui/rhi/qrhid3d12.cpp +++ b/src/gui/rhi/qrhid3d12.cpp @@ -2,16 +2,10 @@ // SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only #include "qrhid3d12_p.h" -#include "qshader.h" -#include <QWindow> #include <qmath.h> -#include <private/qsystemlibrary_p.h> -#include <QtCore/qcryptographichash.h> #include <QtCore/private/qsystemerror_p.h> - -#include <d3dcompiler.h> #include <comdef.h> - +#include "qrhid3dhelpers_p.h" #include "cs_mipmap_p.h" #if __has_include(<pix.h>) @@ -19,6 +13,8 @@ #define QRHI_D3D12_HAS_OLD_PIX #endif +#ifdef __ID3D12Device2_INTERFACE_DEFINED__ + QT_BEGIN_NAMESPACE /* @@ -30,7 +26,7 @@ QT_BEGIN_NAMESPACE \inmodule QtGui \brief Direct3D 12 specific initialization parameters. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. A D3D12-based QRhi needs no special parameters for initialization. If @@ -81,34 +77,60 @@ QT_BEGIN_NAMESPACE \c{d3d12.h} headers is not acceptable here. The actual types are \c{ID3D12Device *} and \c{ID3D12CommandQueue *}. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ /*! \variable QRhiD3D12NativeHandles::dev + + Points to a + \l{https://learn.microsoft.com/en-us/windows/win32/api/d3d12/nn-d3d12-id3d12device}{ID3D12Device} + or left set to \nullptr if no existing device is to be imported. */ /*! \variable QRhiD3D12NativeHandles::minimumFeatureLevel + + Specifies the \b minimum feature level passed to + \l{https://learn.microsoft.com/en-us/windows/win32/api/d3d12/nf-d3d12-d3d12createdevice}{D3D12CreateDevice()}. + When not set, \c{D3D_FEATURE_LEVEL_11_0} is used. See + \l{https://learn.microsoft.com/en-us/windows/win32/direct3d12/hardware-feature-levels}{this + page} for details. + + Relevant only when QRhi creates the device, ignored when importing a device + and device context. */ /*! \variable QRhiD3D12NativeHandles::adapterLuidLow + + The low part of the local identifier (LUID) of the DXGI adapter to use. + Relevant only when QRhi creates the device, ignored when importing a device + and device context. */ /*! \variable QRhiD3D12NativeHandles::adapterLuidHigh + + The high part of the local identifier (LUID) of the DXGI adapter to use. + Relevant only when QRhi creates the device, ignored when importing a device + and device context. */ /*! \variable QRhiD3D12NativeHandles::commandQueue + + When set, must point to a + \l{https://learn.microsoft.com/en-us/windows/win32/api/d3d12/nn-d3d12-id3d12commandqueue}{ID3D12CommandQueue}. + It allows to optionally import a command queue as well, in addition to a + device. */ /*! \class QRhiD3D12CommandBufferNativeHandles \inmodule QtGui - \brief Holds the ID3D12GraphicsCommandList object that is backing a QRhiCommandBuffer. + \brief Holds the ID3D12GraphicsCommandList1 object that is backing a QRhiCommandBuffer. \note The command list object is only guaranteed to be valid, and in recording state, while recording a frame. That is, between a @@ -116,7 +138,7 @@ QT_BEGIN_NAMESPACE \l{QRhi::beginOffscreenFrame()}{beginOffscreenFrame()} - \l{QRhi::endOffscreenFrame()}{endOffscreenFrame()} pair. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -132,8 +154,14 @@ QRhiD3D12::QRhiD3D12(QRhiD3D12InitParams *params, QRhiD3D12NativeHandles *import debugLayer = params->enableDebugLayer; if (importParams) { if (importParams->dev) { - dev = reinterpret_cast<ID3D12Device *>(importParams->dev); - importedDevice = true; + ID3D12Device *d3d12Device = reinterpret_cast<ID3D12Device *>(importParams->dev); + if (SUCCEEDED(d3d12Device->QueryInterface(__uuidof(ID3D12Device2), reinterpret_cast<void **>(&dev)))) { + // get rid of the ref added by QueryInterface + d3d12Device->Release(); + importedDevice = true; + } else { + qWarning("ID3D12Device2 not supported, cannot import device"); + } } if (importParams->commandQueue) { cmdQueue = reinterpret_cast<ID3D12CommandQueue *>(importParams->commandQueue); @@ -179,9 +207,20 @@ bool QRhiD3D12::create(QRhi::Flags flags) factoryFlags |= DXGI_CREATE_FACTORY_DEBUG; HRESULT hr = CreateDXGIFactory2(factoryFlags, __uuidof(IDXGIFactory2), reinterpret_cast<void **>(&dxgiFactory)); if (FAILED(hr)) { - qWarning("CreateDXGIFactory2() failed to create DXGI factory: %s", - qPrintable(QSystemError::windowsComString(hr))); - return false; + // retry without debug, if it was requested (to match D3D11 backend behavior) + if (debugLayer) { + qCDebug(QRHI_LOG_INFO, "Debug layer was requested but is not available. " + "Attempting to create DXGIFactory2 without it."); + factoryFlags &= ~DXGI_CREATE_FACTORY_DEBUG; + hr = CreateDXGIFactory2(factoryFlags, __uuidof(IDXGIFactory2), reinterpret_cast<void **>(&dxgiFactory)); + } + if (SUCCEEDED(hr)) { + debugLayer = false; + } else { + qWarning("CreateDXGIFactory2() failed to create DXGI factory: %s", + qPrintable(QSystemError::windowsComString(hr))); + return false; + } } supportsAllowTearing = false; @@ -249,9 +288,7 @@ bool QRhiD3D12::create(QRhi::Flags flags) if (!activeAdapter && (requestedAdapterIndex < 0 || requestedAdapterIndex == adapterIndex)) { activeAdapter = adapter; adapterLuid = desc.AdapterLuid; - driverInfoStruct.deviceName = name.toUtf8(); - driverInfoStruct.deviceId = desc.DeviceId; - driverInfoStruct.vendorId = desc.VendorId; + QRhiD3D::fillDriverInfo(&driverInfoStruct, desc); qCDebug(QRHI_LOG_INFO, " using this adapter"); } else { adapter->Release(); @@ -267,7 +304,7 @@ bool QRhiD3D12::create(QRhi::Flags flags) hr = D3D12CreateDevice(activeAdapter, minimumFeatureLevel, - __uuidof(ID3D12Device), + __uuidof(ID3D12Device2), reinterpret_cast<void **>(&dev)); if (FAILED(hr)) { qWarning("Failed to create D3D12 device: %s", qPrintable(QSystemError::windowsComString(hr))); @@ -281,16 +318,20 @@ bool QRhiD3D12::create(QRhi::Flags flags) for (int adapterIndex = 0; dxgiFactory->EnumAdapters1(UINT(adapterIndex), &adapter) != DXGI_ERROR_NOT_FOUND; ++adapterIndex) { DXGI_ADAPTER_DESC1 desc; adapter->GetDesc1(&desc); - adapter->Release(); if (desc.AdapterLuid.LowPart == adapterLuid.LowPart && desc.AdapterLuid.HighPart == adapterLuid.HighPart) { - driverInfoStruct.deviceName = QString::fromUtf16(reinterpret_cast<char16_t *>(desc.Description)).toUtf8(); - driverInfoStruct.deviceId = desc.DeviceId; - driverInfoStruct.vendorId = desc.VendorId; + activeAdapter = adapter; + QRhiD3D::fillDriverInfo(&driverInfoStruct, desc); break; + } else { + adapter->Release(); } } + if (!activeAdapter) { + qWarning("No adapter"); + return false; + } qCDebug(QRHI_LOG_INFO, "Using imported device %p", dev); } @@ -392,6 +433,9 @@ bool QRhiD3D12::create(QRhi::Flags flags) qWarning("Could not create host-visible staging area"); return false; } + QString decoratedName = QLatin1String("Small staging area buffer/"); + decoratedName += QString::number(i); + smallStagingAreas[i].mem.buffer->SetName(reinterpret_cast<LPCWSTR>(decoratedName.utf16())); } if (!shaderVisibleCbvSrvUavHeap.create(dev, @@ -402,6 +446,53 @@ bool QRhiD3D12::create(QRhi::Flags flags) return false; } + if (flags.testFlag(QRhi::EnableTimestamps)) { + static bool wantsStablePowerState = qEnvironmentVariableIntValue("QT_D3D_STABLE_POWER_STATE"); + // + // https://learn.microsoft.com/en-us/windows/win32/api/d3d12/nf-d3d12-id3d12device-setstablepowerstate + // + // NB! This is a _global_ setting, affecting other processes (and 3D + // APIs such as Vulkan), as long as this application is running. Hence + // making it an env.var. for now. Never enable it in production. But + // extremely useful for the GPU timings with NVIDIA at least; the + // timestamps become stable and smooth, making the number readable and + // actually useful e.g. in Quick 3D's DebugView when this is enabled. + // (otherwise the number's all over the place) + // + // See also + // https://developer.nvidia.com/blog/advanced-api-performance-setstablepowerstate/ + // for possible other approaches. + // + if (wantsStablePowerState) + dev->SetStablePowerState(TRUE); + + hr = cmdQueue->GetTimestampFrequency(×tampTicksPerSecond); + if (FAILED(hr)) { + qWarning("Failed to query timestamp frequency: %s", + qPrintable(QSystemError::windowsComString(hr))); + return false; + } + if (!timestampQueryHeap.create(dev, QD3D12_FRAMES_IN_FLIGHT * 2, D3D12_QUERY_HEAP_TYPE_TIMESTAMP)) { + qWarning("Failed to create timestamp query pool"); + return false; + } + const quint32 readbackBufSize = QD3D12_FRAMES_IN_FLIGHT * 2 * sizeof(quint64); + if (!timestampReadbackArea.create(this, readbackBufSize, D3D12_HEAP_TYPE_READBACK)) { + qWarning("Failed to create timestamp readback buffer"); + return false; + } + timestampReadbackArea.mem.buffer->SetName(L"Timestamp readback buffer"); + memset(timestampReadbackArea.mem.p, 0, readbackBufSize); + } + + caps = {}; + D3D12_FEATURE_DATA_D3D12_OPTIONS3 options3 = {}; + if (SUCCEEDED(dev->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS3, &options3, sizeof(options3)))) { + caps.multiView = options3.ViewInstancingTier != D3D12_VIEW_INSTANCING_TIER_NOT_SUPPORTED; + // https://microsoft.github.io/DirectX-Specs/d3d/RelaxedCasting.html + caps.textureViewFormat = options3.CastingFullyTypedFormatSupported; + } + deviceLost = false; offscreenActive = false; @@ -430,6 +521,9 @@ void QRhiD3D12::destroy() } } + timestampQueryHeap.destroy(); + timestampReadbackArea.destroy(); + shaderVisibleCbvSrvUavHeap.destroy(); for (int i = 0; i < QD3D12_FRAMES_IN_FLIGHT; ++i) @@ -445,8 +539,10 @@ void QRhiD3D12::destroy() cbvSrvUavPool.destroy(); for (int i = 0; i < QD3D12_FRAMES_IN_FLIGHT; ++i) { - cmdAllocators[i]->Release(); - cmdAllocators[i] = nullptr; + if (cmdAllocators[i]) { + cmdAllocators[i]->Release(); + cmdAllocators[i] = nullptr; + } } if (fullFenceEvent) { @@ -565,7 +661,7 @@ bool QRhiD3D12::isFeatureSupported(QRhi::Feature feature) const return false; #endif case QRhi::Timestamps: - return false; // ### + return true; case QRhi::Instancing: return true; case QRhi::CustomInstanceStepRate: @@ -638,6 +734,14 @@ bool QRhiD3D12::isFeatureSupported(QRhi::Feature feature) const return true; case QRhi::ThreeDimensionalTextureMipmaps: return false; // we generate mipmaps ourselves with compute and this is not implemented + case QRhi::MultiView: + return caps.multiView; + case QRhi::TextureViewFormat: + return caps.textureViewFormat; + case QRhi::ResolveDepthStencil: + // there is no Multisample Resolve support for depth/stencil formats + // https://learn.microsoft.com/en-us/windows/win32/direct3ddxgi/hardware-support-for-direct3d-12-1-formats + return false; } return false; } @@ -794,15 +898,18 @@ void QRhiD3D12::setGraphicsPipeline(QRhiCommandBuffer *cb, QRhiGraphicsPipeline } cbD->cmdList->IASetPrimitiveTopology(psD->topology); + + if (psD->viewInstanceMask) + cbD->cmdList->SetViewInstanceMask(psD->viewInstanceMask); } } -void QRhiD3D12::visitUniformBuffer(QD3D12Stage s, - const QRhiShaderResourceBinding::Data::UniformBufferData &d, - int, - int binding, - int dynamicOffsetCount, - const QRhiCommandBuffer::DynamicOffset *dynamicOffsets) +void QD3D12CommandBuffer::visitUniformBuffer(QD3D12Stage s, + const QRhiShaderResourceBinding::Data::UniformBufferData &d, + int, + int binding, + int dynamicOffsetCount, + const QRhiCommandBuffer::DynamicOffset *dynamicOffsets) { QD3D12Buffer *bufD = QRHI_RES(QD3D12Buffer, d.buf); quint32 offset = d.offset; @@ -815,29 +922,30 @@ void QRhiD3D12::visitUniformBuffer(QD3D12Stage s, } } } - visitorData.cbufs[s].append({ bufD->handles[currentFrameSlot], offset }); + QRHI_RES_RHI(QRhiD3D12); + visitorData.cbufs[s].append({ bufD->handles[rhiD->currentFrameSlot], offset }); } -void QRhiD3D12::visitTexture(QD3D12Stage s, - const QRhiShaderResourceBinding::TextureAndSampler &d, - int) +void QD3D12CommandBuffer::visitTexture(QD3D12Stage s, + const QRhiShaderResourceBinding::TextureAndSampler &d, + int) { QD3D12Texture *texD = QRHI_RES(QD3D12Texture, d.tex); visitorData.srvs[s].append(texD->srv); } -void QRhiD3D12::visitSampler(QD3D12Stage s, - const QRhiShaderResourceBinding::TextureAndSampler &d, - int) +void QD3D12CommandBuffer::visitSampler(QD3D12Stage s, + const QRhiShaderResourceBinding::TextureAndSampler &d, + int) { QD3D12Sampler *samplerD = QRHI_RES(QD3D12Sampler, d.sampler); visitorData.samplers[s].append(samplerD->lookupOrCreateShaderVisibleDescriptor()); } -void QRhiD3D12::visitStorageBuffer(QD3D12Stage s, - const QRhiShaderResourceBinding::Data::StorageBufferData &d, - QD3D12ShaderResourceVisitor::StorageOp, - int) +void QD3D12CommandBuffer::visitStorageBuffer(QD3D12Stage s, + const QRhiShaderResourceBinding::Data::StorageBufferData &d, + QD3D12ShaderResourceVisitor::StorageOp, + int) { QD3D12Buffer *bufD = QRHI_RES(QD3D12Buffer, d.buf); // SPIRV-Cross generated HLSL uses RWByteAddressBuffer @@ -850,17 +958,17 @@ void QRhiD3D12::visitStorageBuffer(QD3D12Stage s, visitorData.uavs[s].append({ bufD->handles[0], uavDesc }); } -void QRhiD3D12::visitStorageImage(QD3D12Stage s, - const QRhiShaderResourceBinding::Data::StorageImageData &d, - QD3D12ShaderResourceVisitor::StorageOp, - int) +void QD3D12CommandBuffer::visitStorageImage(QD3D12Stage s, + const QRhiShaderResourceBinding::Data::StorageImageData &d, + QD3D12ShaderResourceVisitor::StorageOp, + int) { QD3D12Texture *texD = QRHI_RES(QD3D12Texture, d.tex); const bool isCube = texD->m_flags.testFlag(QRhiTexture::CubeMap); const bool isArray = texD->m_flags.testFlag(QRhiTexture::TextureArray); const bool is3D = texD->m_flags.testFlag(QRhiTexture::ThreeDimensional); D3D12_UNORDERED_ACCESS_VIEW_DESC uavDesc = {}; - uavDesc.Format = texD->dxgiFormat; + uavDesc.Format = texD->rtFormat; if (isCube) { uavDesc.ViewDimension = D3D12_UAV_DIMENSION_TEXTURE2DARRAY; uavDesc.Texture2DArray.MipSlice = UINT(d.level); @@ -899,8 +1007,8 @@ void QRhiD3D12::setShaderResources(QRhiCommandBuffer *cb, QRhiShaderResourceBind QD3D12ShaderResourceBindings *srbD = QRHI_RES(QD3D12ShaderResourceBindings, srb); - for (int i = 0, ie = srbD->sortedBindings.size(); i != ie; ++i) { - const QRhiShaderResourceBinding::Data *b = shaderResourceBindingData(srbD->sortedBindings[i]); + for (int i = 0, ie = srbD->m_bindings.size(); i != ie; ++i) { + const QRhiShaderResourceBinding::Data *b = shaderResourceBindingData(srbD->m_bindings[i]); switch (b->type) { case QRhiShaderResourceBinding::UniformBuffer: { @@ -1012,14 +1120,15 @@ void QRhiD3D12::setShaderResources(QRhiCommandBuffer *cb, QRhiShaderResourceBind QD3D12ShaderResourceVisitor visitor(srbD, stageData, gfxPsD ? 5 : 1); + QD3D12CommandBuffer::VisitorData &visitorData(cbD->visitorData); visitorData = {}; using namespace std::placeholders; - visitor.uniformBuffer = std::bind(&QRhiD3D12::visitUniformBuffer, this, _1, _2, _3, _4, dynamicOffsetCount, dynamicOffsets); - visitor.texture = std::bind(&QRhiD3D12::visitTexture, this, _1, _2, _3); - visitor.sampler = std::bind(&QRhiD3D12::visitSampler, this, _1, _2, _3); - visitor.storageBuffer = std::bind(&QRhiD3D12::visitStorageBuffer, this, _1, _2, _3, _4); - visitor.storageImage = std::bind(&QRhiD3D12::visitStorageImage, this, _1, _2, _3, _4); + visitor.uniformBuffer = std::bind(&QD3D12CommandBuffer::visitUniformBuffer, cbD, _1, _2, _3, _4, dynamicOffsetCount, dynamicOffsets); + visitor.texture = std::bind(&QD3D12CommandBuffer::visitTexture, cbD, _1, _2, _3); + visitor.sampler = std::bind(&QD3D12CommandBuffer::visitSampler, cbD, _1, _2, _3); + visitor.storageBuffer = std::bind(&QD3D12CommandBuffer::visitStorageBuffer, cbD, _1, _2, _3, _4); + visitor.storageImage = std::bind(&QD3D12CommandBuffer::visitStorageImage, cbD, _1, _2, _3, _4); visitor.visit(); @@ -1375,8 +1484,24 @@ void QRhiD3D12::endExternal(QRhiCommandBuffer *cb) double QRhiD3D12::lastCompletedGpuTime(QRhiCommandBuffer *cb) { - Q_UNUSED(cb); - return 0; + QD3D12CommandBuffer *cbD = QRHI_RES(QD3D12CommandBuffer, cb); + return cbD->lastGpuTime; +} + +static void calculateGpuTime(QD3D12CommandBuffer *cbD, + int timestampPairStartIndex, + const quint8 *readbackBufPtr, + quint64 timestampTicksPerSecond) +{ + const size_t byteOffset = timestampPairStartIndex * sizeof(quint64); + const quint64 *p = reinterpret_cast<const quint64 *>(readbackBufPtr + byteOffset); + const quint64 startTime = *p++; + const quint64 endTime = *p; + if (startTime < endTime) { + const quint64 ticks = endTime - startTime; + const double timeSec = ticks / double(timestampTicksPerSecond); + cbD->lastGpuTime = timeSec; + } } QRhi::FrameOpResult QRhiD3D12::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginFrameFlags flags) @@ -1400,7 +1525,7 @@ QRhi::FrameOpResult QRhiD3D12::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginF // be in flight anymore). With Qt Quick this situation cannot happen anyway // by design (one QRhi per window). for (QD3D12SwapChain *sc : std::as_const(swapchains)) - sc->waitCommandCompletionForFrameSlot(sc->currentFrameSlot); + sc->waitCommandCompletionForFrameSlot(currentFrameSlot); // note: swapChainD->currentFrameSlot, not sc's HRESULT hr = cmdAllocators[currentFrameSlot]->Reset(); if (FAILED(hr)) { @@ -1422,6 +1547,16 @@ QRhi::FrameOpResult QRhiD3D12::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginF swapChainD->rtWrapper.d.dsv = swapChainD->ds ? swapChainD->ds->dsv.cpuHandle : D3D12_CPU_DESCRIPTOR_HANDLE { 0 }; + if (swapChainD->stereo) { + swapChainD->rtWrapperRight.d.rtv[0] = swapChainD->sampleDesc.Count > 1 + ? swapChainD->msaaRtvs[swapChainD->currentBackBufferIndex].cpuHandle + : swapChainD->rtvsRight[swapChainD->currentBackBufferIndex].cpuHandle; + + swapChainD->rtWrapperRight.d.dsv = + swapChainD->ds ? swapChainD->ds->dsv.cpuHandle : D3D12_CPU_DESCRIPTOR_HANDLE{ 0 }; + } + + // Time to release things that are marked for currentFrameSlot since due to // the wait above we know that the previous commands on the GPU for this // slot must have finished already. @@ -1439,6 +1574,20 @@ QRhi::FrameOpResult QRhiD3D12::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginF finishActiveReadbacks(); // last, in case the readback-completed callback issues rhi calls + if (timestampQueryHeap.isValid() && timestampTicksPerSecond) { + // Read the timestamps for the previous frame for this slot. (the + // ResolveQuery() should have completed by now due to the wait above) + const int timestampPairStartIndex = currentFrameSlot * QD3D12_FRAMES_IN_FLIGHT; + calculateGpuTime(cbD, + timestampPairStartIndex, + timestampReadbackArea.mem.p, + timestampTicksPerSecond); + // Write the start timestamp for this frame for this slot. + cbD->cmdList->EndQuery(timestampQueryHeap.heap, + D3D12_QUERY_TYPE_TIMESTAMP, + timestampPairStartIndex); + } + return QRhi::FrameOpSuccess; } @@ -1463,7 +1612,20 @@ QRhi::FrameOpResult QRhiD3D12::endFrame(QRhiSwapChain *swapChain, QRhi::EndFrame barrierGen.addTransitionBarrier(backBufferResourceHandle, D3D12_RESOURCE_STATE_PRESENT); barrierGen.enqueueBufferedTransitionBarriers(cbD); - ID3D12GraphicsCommandList *cmdList = cbD->cmdList; + if (timestampQueryHeap.isValid()) { + const int timestampPairStartIndex = currentFrameSlot * QD3D12_FRAMES_IN_FLIGHT; + cbD->cmdList->EndQuery(timestampQueryHeap.heap, + D3D12_QUERY_TYPE_TIMESTAMP, + timestampPairStartIndex + 1); + cbD->cmdList->ResolveQueryData(timestampQueryHeap.heap, + D3D12_QUERY_TYPE_TIMESTAMP, + timestampPairStartIndex, + 2, + timestampReadbackArea.mem.buffer, + timestampPairStartIndex * sizeof(quint64)); + } + + ID3D12GraphicsCommandList1 *cmdList = cbD->cmdList; HRESULT hr = cmdList->Close(); if (FAILED(hr)) { qWarning("Failed to close command list: %s", @@ -1481,6 +1643,10 @@ QRhi::FrameOpResult QRhiD3D12::endFrame(QRhiSwapChain *swapChain, QRhi::EndFrame { presentFlags |= DXGI_PRESENT_ALLOW_TEARING; } + if (!swapChainD->swapChain) { + qWarning("Failed to present, no swapchain"); + return QRhi::FrameOpError; + } HRESULT hr = swapChainD->swapChain->Present(swapChainD->swapInterval, presentFlags); if (hr == DXGI_ERROR_DEVICE_REMOVED || hr == DXGI_ERROR_DEVICE_RESET) { qWarning("Device loss detected in Present()"); @@ -1547,6 +1713,12 @@ QRhi::FrameOpResult QRhiD3D12::beginOffscreenFrame(QRhiCommandBuffer **cb, QRhi: bindShaderVisibleHeaps(cbD); + if (timestampQueryHeap.isValid() && timestampTicksPerSecond) { + cbD->cmdList->EndQuery(timestampQueryHeap.heap, + D3D12_QUERY_TYPE_TIMESTAMP, + currentFrameSlot * QD3D12_FRAMES_IN_FLIGHT); + } + offscreenActive = true; *cb = cbD; @@ -1560,7 +1732,20 @@ QRhi::FrameOpResult QRhiD3D12::endOffscreenFrame(QRhi::EndFrameFlags flags) offscreenActive = false; QD3D12CommandBuffer *cbD = offscreenCb[currentFrameSlot]; - ID3D12GraphicsCommandList *cmdList = cbD->cmdList; + if (timestampQueryHeap.isValid()) { + const int timestampPairStartIndex = currentFrameSlot * QD3D12_FRAMES_IN_FLIGHT; + cbD->cmdList->EndQuery(timestampQueryHeap.heap, + D3D12_QUERY_TYPE_TIMESTAMP, + timestampPairStartIndex + 1); + cbD->cmdList->ResolveQueryData(timestampQueryHeap.heap, + D3D12_QUERY_TYPE_TIMESTAMP, + timestampPairStartIndex, + 2, + timestampReadbackArea.mem.buffer, + timestampPairStartIndex * sizeof(quint64)); + } + + ID3D12GraphicsCommandList1 *cmdList = cbD->cmdList; HRESULT hr = cmdList->Close(); if (FAILED(hr)) { qWarning("Failed to close command list: %s", @@ -1580,6 +1765,14 @@ QRhi::FrameOpResult QRhiD3D12::endOffscreenFrame(QRhi::EndFrameFlags flags) // previous) frame is safe since we waited for completion above. finishActiveReadbacks(true); + // the timestamp query results should be available too, given the wait + if (timestampQueryHeap.isValid()) { + calculateGpuTime(cbD, + currentFrameSlot * QD3D12_FRAMES_IN_FLIGHT, + timestampReadbackArea.mem.p, + timestampTicksPerSecond); + } + return QRhi::FrameOpSuccess; } @@ -1601,7 +1794,7 @@ QRhi::FrameOpResult QRhiD3D12::finish() Q_ASSERT(cbD->recordingPass == QD3D12CommandBuffer::NoPass); - ID3D12GraphicsCommandList *cmdList = cbD->cmdList; + ID3D12GraphicsCommandList1 *cmdList = cbD->cmdList; HRESULT hr = cmdList->Close(); if (FAILED(hr)) { qWarning("Failed to close command list: %s", @@ -1786,15 +1979,19 @@ void QRhiD3D12::endPass(QRhiCommandBuffer *cb, QRhiResourceUpdateBatch *resource barrierGen.addTransitionBarrier(dstTexD->handle, D3D12_RESOURCE_STATE_RESOLVE_DEST); barrierGen.enqueueBufferedTransitionBarriers(cbD); - const UINT srcSubresource = calcSubresource(0, UINT(colorAtt.layer()), 1); - const UINT dstSubresource = calcSubresource(UINT(colorAtt.resolveLevel()), - UINT(colorAtt.resolveLayer()), - dstTexD->mipLevelCount); - cbD->cmdList->ResolveSubresource(dstRes->resource, dstSubresource, - srcRes->resource, srcSubresource, - dstTexD->dxgiFormat); + const UINT resolveCount = colorAtt.multiViewCount() >= 2 ? colorAtt.multiViewCount() : 1; + for (UINT resolveIdx = 0; resolveIdx < resolveCount; ++resolveIdx) { + const UINT srcSubresource = calcSubresource(0, UINT(colorAtt.layer()) + resolveIdx, 1); + const UINT dstSubresource = calcSubresource(UINT(colorAtt.resolveLevel()), + UINT(colorAtt.resolveLayer()) + resolveIdx, + dstTexD->mipLevelCount); + cbD->cmdList->ResolveSubresource(dstRes->resource, dstSubresource, + srcRes->resource, srcSubresource, + dstTexD->dxgiFormat); + } } - + if (rtTex->m_desc.depthResolveTexture()) + qWarning("Resolving multisample depth-stencil buffers is not supported with D3D"); } cbD->recordingPass = QD3D12CommandBuffer::NoPass; @@ -2041,6 +2238,36 @@ void QD3D12CpuDescriptorPool::release(const QD3D12Descriptor &descriptor, quint3 quint64(descriptor.cpuHandle.ptr)); } +bool QD3D12QueryHeap::create(ID3D12Device *device, + quint32 queryCount, + D3D12_QUERY_HEAP_TYPE heapType) +{ + capacity = queryCount; + + D3D12_QUERY_HEAP_DESC heapDesc = {}; + heapDesc.Type = heapType; + heapDesc.Count = capacity; + + HRESULT hr = device->CreateQueryHeap(&heapDesc, __uuidof(ID3D12QueryHeap), reinterpret_cast<void **>(&heap)); + if (FAILED(hr)) { + qWarning("Failed to create query heap: %s", qPrintable(QSystemError::windowsComString(hr))); + heap = nullptr; + capacity = 0; + return false; + } + + return true; +} + +void QD3D12QueryHeap::destroy() +{ + if (heap) { + heap->Release(); + heap = nullptr; + } + capacity = 0; +} + bool QD3D12StagingArea::create(QRhiD3D12 *rhi, quint32 capacity, D3D12_HEAP_TYPE heapType) { Q_ASSERT(heapType == D3D12_HEAP_TYPE_UPLOAD || heapType == D3D12_HEAP_TYPE_READBACK); @@ -2381,8 +2608,8 @@ static inline QPair<int, int> mapBinding(int binding, const QShader::NativeResou void QD3D12ShaderResourceVisitor::visit() { - for (int bindingIdx = 0, bindingCount = srb->sortedBindings.count(); bindingIdx != bindingCount; ++bindingIdx) { - const QRhiShaderResourceBinding &b(srb->sortedBindings[bindingIdx]); + for (int bindingIdx = 0, bindingCount = srb->m_bindings.count(); bindingIdx != bindingCount; ++bindingIdx) { + const QRhiShaderResourceBinding &b(srb->m_bindings[bindingIdx]); const QRhiShaderResourceBinding::Data *bd = QRhiImplementation::shaderResourceBindingData(b); for (int stageIdx = 0; stageIdx < stageCount; ++stageIdx) { @@ -2533,6 +2760,7 @@ bool QD3D12MipmapGenerator::create(QRhiD3D12 *rhiD) // b0 rootParams[0].ParameterType = D3D12_ROOT_PARAMETER_TYPE_CBV; rootParams[0].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL; + rootParams[0].Descriptor.Flags = D3D12_ROOT_DESCRIPTOR_FLAG_DATA_STATIC; // t0 descriptorRanges[0].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV; @@ -2879,24 +3107,18 @@ void QRhiD3D12::waitGpu() } } -DXGI_SAMPLE_DESC QRhiD3D12::effectiveSampleCount(int sampleCount, DXGI_FORMAT format) const +DXGI_SAMPLE_DESC QRhiD3D12::effectiveSampleDesc(int sampleCount, DXGI_FORMAT format) const { DXGI_SAMPLE_DESC desc; desc.Count = 1; desc.Quality = 0; - // Stay compatible with QSurfaceFormat and friends where samples == 0 means the same as 1. - int s = qBound(1, sampleCount, 64); - - if (!supportedSampleCounts().contains(s)) { - qWarning("Attempted to set unsupported sample count %d", sampleCount); - return desc; - } + const int s = effectiveSampleCount(sampleCount); if (s > 1) { D3D12_FEATURE_DATA_MULTISAMPLE_QUALITY_LEVELS msaaInfo = {}; msaaInfo.Format = format; - msaaInfo.SampleCount = s; + msaaInfo.SampleCount = UINT(s); if (SUCCEEDED(dev->CheckFeatureSupport(D3D12_FEATURE_MULTISAMPLE_QUALITY_LEVELS, &msaaInfo, sizeof(msaaInfo)))) { if (msaaInfo.NumQualityLevels > 0) { desc.Count = UINT(s); @@ -2910,7 +3132,7 @@ DXGI_SAMPLE_DESC QRhiD3D12::effectiveSampleCount(int sampleCount, DXGI_FORMAT fo return desc; } -bool QRhiD3D12::startCommandListForCurrentFrameSlot(ID3D12GraphicsCommandList **cmdList) +bool QRhiD3D12::startCommandListForCurrentFrameSlot(ID3D12GraphicsCommandList1 **cmdList) { ID3D12CommandAllocator *cmdAlloc = cmdAllocators[currentFrameSlot]; if (!*cmdList) { @@ -2918,7 +3140,7 @@ bool QRhiD3D12::startCommandListForCurrentFrameSlot(ID3D12GraphicsCommandList ** D3D12_COMMAND_LIST_TYPE_DIRECT, cmdAlloc, nullptr, - __uuidof(ID3D12GraphicsCommandList), + __uuidof(ID3D12GraphicsCommandList1), reinterpret_cast<void **>(cmdList)); if (FAILED(hr)) { qWarning("Failed to create command list: %s", qPrintable(QSystemError::windowsComString(hr))); @@ -3076,18 +3298,42 @@ void QRhiD3D12::enqueueResourceUpdates(QD3D12CommandBuffer *cbD, QRhiResourceUpd for (int layer = 0, maxLayer = u.subresDesc.size(); layer < maxLayer; ++layer) { for (int level = 0; level < QRhi::MAX_MIP_LEVELS; ++level) { for (const QRhiTextureSubresourceUploadDescription &subresDesc : std::as_const(u.subresDesc[layer][level])) { - const UINT subresource = calcSubresource(UINT(level), is3D ? 0u : UINT(layer), texD->mipLevelCount); - D3D12_PLACED_SUBRESOURCE_FOOTPRINT layout; - UINT64 totalBytes = 0; - D3D12_RESOURCE_DESC desc = res->desc; - if (is3D) { - desc.Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D; - desc.DepthOrArraySize = 1; + D3D12_SUBRESOURCE_FOOTPRINT footprint = {}; + footprint.Format = res->desc.Format; + footprint.Depth = 1; + quint32 totalBytes = 0; + + const QSize subresSize = subresDesc.sourceSize().isEmpty() ? q->sizeForMipLevel(level, texD->m_pixelSize) + : subresDesc.sourceSize(); + const QPoint srcPos = subresDesc.sourceTopLeft(); + QPoint dstPos = subresDesc.destinationTopLeft(); + + if (!subresDesc.image().isNull()) { + const QImage img = subresDesc.image(); + const int bpl = img.bytesPerLine(); + footprint.RowPitch = aligned<UINT>(bpl, D3D12_TEXTURE_DATA_PITCH_ALIGNMENT); + totalBytes = footprint.RowPitch * img.height(); + } else if (!subresDesc.data().isEmpty() && isCompressedFormat(texD->m_format)) { + QSize blockDim; + quint32 bpl = 0; + compressedFormatInfo(texD->m_format, subresSize, &bpl, nullptr, &blockDim); + footprint.RowPitch = aligned<UINT>(bpl, D3D12_TEXTURE_DATA_PITCH_ALIGNMENT); + const int rowCount = aligned(subresSize.height(), blockDim.height()) / blockDim.height(); + totalBytes = footprint.RowPitch * rowCount; + } else if (!subresDesc.data().isEmpty()) { + quint32 bpl = 0; + if (subresDesc.dataStride()) + bpl = subresDesc.dataStride(); + else + textureFormatInfo(texD->m_format, subresSize, &bpl, nullptr, nullptr); + footprint.RowPitch = aligned<UINT>(bpl, D3D12_TEXTURE_DATA_PITCH_ALIGNMENT); + totalBytes = footprint.RowPitch * subresSize.height(); + } else { + qWarning("Invalid texture upload for %p layer=%d mip=%d", texD, layer, level); + continue; } - dev->GetCopyableFootprints(&desc, subresource, 1, 0, - &layout, nullptr, nullptr, &totalBytes); - const quint32 allocSize = QD3D12StagingArea::allocSizeForArray(quint32(totalBytes), 1); + const quint32 allocSize = QD3D12StagingArea::allocSizeForArray(totalBytes, 1); QD3D12StagingArea::Allocation stagingAlloc; if (smallStagingAreas[currentFrameSlot].remainingCapacity() >= allocSize) stagingAlloc = smallStagingAreas[currentFrameSlot].get(allocSize); @@ -3104,32 +3350,29 @@ void QRhiD3D12::enqueueResourceUpdates(QD3D12CommandBuffer *cbD, QRhiResourceUpd } } - const UINT requiredBytesPerLine = layout.Footprint.RowPitch; // multiple of 256 - const QSize subresSize = subresDesc.sourceSize().isEmpty() ? q->sizeForMipLevel(level, texD->m_pixelSize) - : subresDesc.sourceSize(); - const QPoint srcPos = subresDesc.sourceTopLeft(); - QPoint dstPos = subresDesc.destinationTopLeft(); - D3D12_TEXTURE_COPY_LOCATION dst; dst.pResource = res->resource; dst.Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX; - dst.SubresourceIndex = subresource; + dst.SubresourceIndex = calcSubresource(UINT(level), is3D ? 0u : UINT(layer), texD->mipLevelCount); D3D12_TEXTURE_COPY_LOCATION src; src.pResource = stagingAlloc.buffer; src.Type = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT; src.PlacedFootprint.Offset = stagingAlloc.bufferOffset; - src.PlacedFootprint.Footprint = layout.Footprint; D3D12_BOX srcBox; // back, right, bottom are exclusive if (!subresDesc.image().isNull()) { - QImage img = subresDesc.image(); + const QImage img = subresDesc.image(); const int bpc = qMax(1, img.depth() / 8); const int bpl = img.bytesPerLine(); QSize size = subresDesc.sourceSize().isEmpty() ? img.size() : subresDesc.sourceSize(); size.setWidth(qMin(size.width(), img.width() - srcPos.x())); size.setHeight(qMin(size.height(), img.height() - srcPos.y())); + + footprint.Width = size.width(); + footprint.Height = size.height(); + srcBox.left = 0; srcBox.top = 0; srcBox.right = UINT(size.width()); @@ -3140,7 +3383,7 @@ void QRhiD3D12::enqueueResourceUpdates(QD3D12CommandBuffer *cbD, QRhiResourceUpd const uchar *imgPtr = img.constBits(); const quint32 lineBytes = size.width() * bpc; for (int y = 0, h = size.height(); y < h; ++y) { - memcpy(stagingAlloc.p + y * requiredBytesPerLine, + memcpy(stagingAlloc.p + y * footprint.RowPitch, imgPtr + srcPos.x() * bpc + (y + srcPos.y()) * bpl, lineBytes); } @@ -3157,15 +3400,19 @@ void QRhiD3D12::enqueueResourceUpdates(QD3D12CommandBuffer *cbD, QRhiResourceUpd // width and height must be multiples of the block width and height srcBox.right = aligned(subresSize.width(), blockDim.width()); srcBox.bottom = aligned(subresSize.height(), blockDim.height()); + srcBox.front = 0; srcBox.back = 1; - const quint32 copyBytes = qMin(bpl, requiredBytesPerLine); + footprint.Width = aligned(subresSize.width(), blockDim.width()); + footprint.Height = aligned(subresSize.height(), blockDim.height()); + + const quint32 copyBytes = qMin(bpl, footprint.RowPitch); const QByteArray imgData = subresDesc.data(); const char *imgPtr = imgData.constData(); const int rowCount = aligned(subresSize.height(), blockDim.height()) / blockDim.height(); for (int y = 0; y < rowCount; ++y) - memcpy(stagingAlloc.p + y * requiredBytesPerLine, imgPtr + y * bpl, copyBytes); + memcpy(stagingAlloc.p + y * footprint.RowPitch, imgPtr + y * bpl, copyBytes); } else if (!subresDesc.data().isEmpty()) { srcBox.left = 0; srcBox.top = 0; @@ -3174,24 +3421,24 @@ void QRhiD3D12::enqueueResourceUpdates(QD3D12CommandBuffer *cbD, QRhiResourceUpd srcBox.front = 0; srcBox.back = 1; + footprint.Width = subresSize.width(); + footprint.Height = subresSize.height(); + quint32 bpl = 0; if (subresDesc.dataStride()) bpl = subresDesc.dataStride(); else textureFormatInfo(texD->m_format, subresSize, &bpl, nullptr, nullptr); - const quint32 copyBytes = qMin(bpl, requiredBytesPerLine); + const quint32 copyBytes = qMin(bpl, footprint.RowPitch); const QByteArray data = subresDesc.data(); const char *imgPtr = data.constData(); for (int y = 0, h = subresSize.height(); y < h; ++y) - memcpy(stagingAlloc.p + y * requiredBytesPerLine, imgPtr + y * bpl, copyBytes); - } else { - qWarning("Invalid texture upload for %p layer=%d mip=%d", texD, layer, level); - if (ownStagingArea.has_value()) - ownStagingArea->destroyWithDeferredRelease(&releaseQueue); - continue; + memcpy(stagingAlloc.p + y * footprint.RowPitch, imgPtr + y * bpl, copyBytes); } + src.PlacedFootprint.Footprint = footprint; + cbD->cmdList->CopyTextureRegion(&dst, UINT(dstPos.x()), UINT(dstPos.y()), @@ -3742,7 +3989,7 @@ bool QD3D12RenderBuffer::create() case QRhiRenderBuffer::Color: { dxgiFormat = toD3DTextureFormat(backingFormat(), {}); - sampleDesc = rhiD->effectiveSampleCount(m_sampleCount, dxgiFormat); + sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount, dxgiFormat); D3D12_RESOURCE_DESC resourceDesc = {}; resourceDesc.Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D; resourceDesc.Width = UINT64(m_pixelSize.width()); @@ -3783,7 +4030,7 @@ bool QD3D12RenderBuffer::create() case QRhiRenderBuffer::DepthStencil: { dxgiFormat = DS_FORMAT; - sampleDesc = rhiD->effectiveSampleCount(m_sampleCount, dxgiFormat); + sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount, dxgiFormat); D3D12_RESOURCE_DESC resourceDesc = {}; resourceDesc.Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D; resourceDesc.Width = UINT64(m_pixelSize.width()); @@ -3936,10 +4183,30 @@ bool QD3D12Texture::prepareCreate(QSize *adjustedSize) const QSize size = is1D ? QSize(qMax(1, m_pixelSize.width()), 1) : (m_pixelSize.isEmpty() ? QSize(1, 1) : m_pixelSize); - QRHI_RES_RHI(QRhiD3D12); dxgiFormat = toD3DTextureFormat(m_format, m_flags); + if (isDepth) { + srvFormat = toD3DDepthTextureSRVFormat(m_format); + rtFormat = toD3DDepthTextureDSVFormat(m_format); + } else { + srvFormat = dxgiFormat; + rtFormat = dxgiFormat; + } + if (m_writeViewFormat.format != UnknownFormat) { + if (isDepth) + rtFormat = toD3DDepthTextureDSVFormat(m_writeViewFormat.format); + else + rtFormat = toD3DTextureFormat(m_writeViewFormat.format, m_writeViewFormat.srgb ? sRGB : Flags()); + } + if (m_readViewFormat.format != UnknownFormat) { + if (isDepth) + srvFormat = toD3DDepthTextureSRVFormat(m_readViewFormat.format); + else + srvFormat = toD3DTextureFormat(m_readViewFormat.format, m_readViewFormat.srgb ? sRGB : Flags()); + } + + QRHI_RES_RHI(QRhiD3D12); mipLevelCount = uint(hasMipMaps ? rhiD->q->mipLevelsForSize(size) : 1); - sampleDesc = rhiD->effectiveSampleCount(m_sampleCount, dxgiFormat); + sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount, dxgiFormat); if (sampleDesc.Count > 1) { if (isCube) { qWarning("Cubemap texture cannot be multisample"); @@ -3996,14 +4263,13 @@ bool QD3D12Texture::prepareCreate(QSize *adjustedSize) bool QD3D12Texture::finishCreate() { QRHI_RES_RHI(QRhiD3D12); - const bool isDepth = isDepthTextureFormat(m_format); const bool isCube = m_flags.testFlag(CubeMap); const bool is3D = m_flags.testFlag(ThreeDimensional); const bool isArray = m_flags.testFlag(TextureArray); const bool is1D = m_flags.testFlag(OneDimensional); D3D12_SHADER_RESOURCE_VIEW_DESC srvDesc = {}; - srvDesc.Format = isDepth ? toD3DDepthTextureSRVFormat(m_format) : dxgiFormat; + srvDesc.Format = srvFormat; srvDesc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING; if (isCube) { @@ -4093,7 +4359,7 @@ bool QD3D12Texture::create() bool needsOptimizedClearValueSpecified = false; UINT resourceFlags = 0; - if (m_flags.testFlag(RenderTarget)) { + if (m_flags.testFlag(RenderTarget) || sampleDesc.Count > 1) { if (isDepth) resourceFlags |= D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL; else @@ -4362,7 +4628,7 @@ QRhiRenderPassDescriptor *QD3D12TextureRenderTarget::newCompatibleRenderPassDesc QD3D12Texture *texD = QRHI_RES(QD3D12Texture, it->texture()); QD3D12RenderBuffer *rbD = QRHI_RES(QD3D12RenderBuffer, it->renderBuffer()); if (texD) - rpD->colorFormat[rpD->colorAttachmentCount] = texD->dxgiFormat; + rpD->colorFormat[rpD->colorAttachmentCount] = texD->rtFormat; else if (rbD) rpD->colorFormat[rpD->colorAttachmentCount] = rbD->dxgiFormat; rpD->colorAttachmentCount += 1; @@ -4410,19 +4676,21 @@ bool QD3D12TextureRenderTarget::create() qWarning("Could not look up texture handle for render target"); return false; } + const bool isMultiView = it->multiViewCount() >= 2; + UINT layerCount = isMultiView ? UINT(it->multiViewCount()) : 1; D3D12_RENDER_TARGET_VIEW_DESC rtvDesc = {}; - rtvDesc.Format = toD3DTextureFormat(texD->format(), texD->flags()); + rtvDesc.Format = texD->rtFormat; if (texD->flags().testFlag(QRhiTexture::CubeMap)) { rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2DARRAY; rtvDesc.Texture2DArray.MipSlice = UINT(colorAtt.level()); rtvDesc.Texture2DArray.FirstArraySlice = UINT(colorAtt.layer()); - rtvDesc.Texture2DArray.ArraySize = 1; + rtvDesc.Texture2DArray.ArraySize = layerCount; } else if (texD->flags().testFlag(QRhiTexture::OneDimensional)) { if (texD->flags().testFlag(QRhiTexture::TextureArray)) { rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE1DARRAY; rtvDesc.Texture1DArray.MipSlice = UINT(colorAtt.level()); rtvDesc.Texture1DArray.FirstArraySlice = UINT(colorAtt.layer()); - rtvDesc.Texture1DArray.ArraySize = 1; + rtvDesc.Texture1DArray.ArraySize = layerCount; } else { rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE1D; rtvDesc.Texture1D.MipSlice = UINT(colorAtt.level()); @@ -4431,18 +4699,18 @@ bool QD3D12TextureRenderTarget::create() if (texD->sampleDesc.Count > 1) { rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2DMSARRAY; rtvDesc.Texture2DMSArray.FirstArraySlice = UINT(colorAtt.layer()); - rtvDesc.Texture2DMSArray.ArraySize = 1; + rtvDesc.Texture2DMSArray.ArraySize = layerCount; } else { rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2DARRAY; rtvDesc.Texture2DArray.MipSlice = UINT(colorAtt.level()); rtvDesc.Texture2DArray.FirstArraySlice = UINT(colorAtt.layer()); - rtvDesc.Texture2DArray.ArraySize = 1; + rtvDesc.Texture2DArray.ArraySize = layerCount; } } else if (texD->flags().testFlag(QRhiTexture::ThreeDimensional)) { rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE3D; rtvDesc.Texture3D.MipSlice = UINT(colorAtt.level()); rtvDesc.Texture3D.FirstWSlice = UINT(colorAtt.layer()); - rtvDesc.Texture3D.WSize = 1; + rtvDesc.Texture3D.WSize = layerCount; } else { if (texD->sampleDesc.Count > 1) { rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2DMS; @@ -4485,9 +4753,30 @@ bool QD3D12TextureRenderTarget::create() return false; } D3D12_DEPTH_STENCIL_VIEW_DESC dsvDesc = {}; - dsvDesc.Format = toD3DDepthTextureDSVFormat(depthTexD->format()); + dsvDesc.Format = depthTexD->rtFormat; dsvDesc.ViewDimension = depthTexD->sampleDesc.Count > 1 ? D3D12_DSV_DIMENSION_TEXTURE2DMS : D3D12_DSV_DIMENSION_TEXTURE2D; + if (depthTexD->flags().testFlag(QRhiTexture::TextureArray)) { + if (depthTexD->sampleDesc.Count > 1) { + dsvDesc.ViewDimension = D3D12_DSV_DIMENSION_TEXTURE2DMSARRAY; + if (depthTexD->arrayRangeStart() >= 0 && depthTexD->arrayRangeLength() >= 0) { + dsvDesc.Texture2DMSArray.FirstArraySlice = UINT(depthTexD->arrayRangeStart()); + dsvDesc.Texture2DMSArray.ArraySize = UINT(depthTexD->arrayRangeLength()); + } else { + dsvDesc.Texture2DMSArray.FirstArraySlice = 0; + dsvDesc.Texture2DMSArray.ArraySize = UINT(qMax(0, depthTexD->arraySize())); + } + } else { + dsvDesc.ViewDimension = D3D12_DSV_DIMENSION_TEXTURE2DARRAY; + if (depthTexD->arrayRangeStart() >= 0 && depthTexD->arrayRangeLength() >= 0) { + dsvDesc.Texture2DArray.FirstArraySlice = UINT(depthTexD->arrayRangeStart()); + dsvDesc.Texture2DArray.ArraySize = UINT(depthTexD->arrayRangeLength()); + } else { + dsvDesc.Texture2DArray.FirstArraySlice = 0; + dsvDesc.Texture2DArray.ArraySize = UINT(qMax(0, depthTexD->arraySize())); + } + } + } dsv = rhiD->dsvPool.allocate(1); if (!dsv.isValid()) { qWarning("Failed to allocate DSV for texture render target"); @@ -4554,8 +4843,6 @@ QD3D12ShaderResourceBindings::~QD3D12ShaderResourceBindings() void QD3D12ShaderResourceBindings::destroy() { - sortedBindings.clear(); - QRHI_RES_RHI(QRhiD3D12); if (rhiD) rhiD->unregisterResource(this); @@ -4563,20 +4850,14 @@ void QD3D12ShaderResourceBindings::destroy() bool QD3D12ShaderResourceBindings::create() { - if (!sortedBindings.isEmpty()) - destroy(); - QRHI_RES_RHI(QRhiD3D12); if (!rhiD->sanityCheckShaderResourceBindings(this)) return false; rhiD->updateLayoutDesc(this); - std::copy(m_bindings.cbegin(), m_bindings.cend(), std::back_inserter(sortedBindings)); - std::sort(sortedBindings.begin(), sortedBindings.end(), QRhiImplementation::sortedBindingLessThan); - hasDynamicOffset = false; - for (const QRhiShaderResourceBinding &b : sortedBindings) { + for (const QRhiShaderResourceBinding &b : std::as_const(m_bindings)) { const QRhiShaderResourceBinding::Data *bd = QRhiImplementation::shaderResourceBindingData(b); if (bd->type == QRhiShaderResourceBinding::UniformBuffer && bd->u.ubuf.hasDynamicOffset) { hasDynamicOffset = true; @@ -4599,11 +4880,7 @@ bool QD3D12ShaderResourceBindings::create() void QD3D12ShaderResourceBindings::updateResources(UpdateFlags flags) { - sortedBindings.clear(); - std::copy(m_bindings.cbegin(), m_bindings.cend(), std::back_inserter(sortedBindings)); - if (!flags.testFlag(BindingsAreSorted)) - std::sort(sortedBindings.begin(), sortedBindings.end(), QRhiImplementation::sortedBindingLessThan); - + Q_UNUSED(flags); generation += 1; } @@ -4621,6 +4898,7 @@ void QD3D12ShaderResourceBindings::visitUniformBuffer(QD3D12Stage s, rootParam.ParameterType = D3D12_ROOT_PARAMETER_TYPE_CBV; rootParam.ShaderVisibility = qd3d12_stageToVisibility(s); rootParam.Descriptor.ShaderRegister = shaderRegister; + rootParam.Descriptor.Flags = D3D12_ROOT_DESCRIPTOR_FLAG_DATA_STATIC; visitorData.cbParams[s].append(rootParam); } @@ -4812,21 +5090,14 @@ QD3D12ObjectHandle QD3D12ShaderResourceBindings::createRootSignature(const QD3D1 return QD3D12RootSignature::addToPool(&rhiD->rootSignaturePool, rootSig); } -// For now we mirror exactly what's done in the D3D11 backend, meaning we use -// the old shader compiler (so like fxc, not dxc) to generate shader model 5.0 -// output. Some day this should be moved to the new compiler and DXIL. - -static pD3DCompile resolveD3DCompile() -{ - for (const wchar_t *libraryName : {L"D3DCompiler_47", L"D3DCompiler_43"}) { - QSystemLibrary library(libraryName); - if (library.load()) { - if (auto symbol = library.resolve("D3DCompile")) - return reinterpret_cast<pD3DCompile>(symbol); - } - } - return nullptr; -} +// For shader model < 6.0 we do the same as the D3D11 backend: use the old +// compiler (D3DCompile) to generate DXBC, just as qsb does (when -c is passed) +// by invoking fxc, not dxc. For SM >= 6.0 we have to use the new compiler and +// work with DXIL. And that involves IDxcCompiler and needs the presence of +// dxcompiler.dll and dxil.dll at runtime. Plus there's a chance we have +// ancient SDK headers when not using MSVC. So this is heavily optional, +// meaning support for dxc can be disabled both at build time (no dxcapi.h) and +// at run time (no DLLs). static inline void makeHlslTargetString(char target[7], const char stage[3], int version) { @@ -4841,9 +5112,139 @@ static inline void makeHlslTargetString(char target[7], const char stage[3], int target[6] = '\0'; } +enum class HlslCompileFlag +{ + WithDebugInfo = 0x01 +}; + +static QByteArray legacyCompile(const QShaderCode &hlslSource, const char *target, int flags, QString *error) +{ + static const pD3DCompile d3dCompile = QRhiD3D::resolveD3DCompile(); + if (!d3dCompile) { + qWarning("Unable to resolve function D3DCompile()"); + return QByteArray(); + } + + ID3DBlob *bytecode = nullptr; + ID3DBlob *errors = nullptr; + UINT d3dCompileFlags = 0; + if (flags & int(HlslCompileFlag::WithDebugInfo)) + d3dCompileFlags |= D3DCOMPILE_DEBUG; + + HRESULT hr = d3dCompile(hlslSource.shader().constData(), SIZE_T(hlslSource.shader().size()), + nullptr, nullptr, nullptr, + hlslSource.entryPoint().constData(), target, d3dCompileFlags, 0, &bytecode, &errors); + if (FAILED(hr) || !bytecode) { + qWarning("HLSL shader compilation failed: 0x%x", uint(hr)); + if (errors) { + *error = QString::fromUtf8(static_cast<const char *>(errors->GetBufferPointer()), + int(errors->GetBufferSize())); + errors->Release(); + } + return QByteArray(); + } + + QByteArray result; + result.resize(int(bytecode->GetBufferSize())); + memcpy(result.data(), bytecode->GetBufferPointer(), size_t(result.size())); + bytecode->Release(); + return result; +} + +#ifdef QRHI_D3D12_HAS_DXC + +#ifndef DXC_CP_UTF8 +#define DXC_CP_UTF8 65001 +#endif + +#ifndef DXC_ARG_DEBUG +#define DXC_ARG_DEBUG L"-Zi" +#endif + +static QByteArray dxcCompile(const QShaderCode &hlslSource, const char *target, int flags, QString *error) +{ + static std::pair<IDxcCompiler *, IDxcLibrary *> dxc = QRhiD3D::createDxcCompiler(); + IDxcCompiler *compiler = dxc.first; + if (!compiler) { + qWarning("Unable to instantiate IDxcCompiler. Likely no dxcompiler.dll and dxil.dll present. " + "Use windeployqt or try https://github.com/microsoft/DirectXShaderCompiler/releases"); + return QByteArray(); + } + IDxcLibrary *library = dxc.second; + if (!library) + return QByteArray(); + + IDxcBlobEncoding *sourceBlob = nullptr; + HRESULT hr = library->CreateBlobWithEncodingOnHeapCopy(hlslSource.shader().constData(), + UINT32(hlslSource.shader().size()), + DXC_CP_UTF8, + &sourceBlob); + if (FAILED(hr)) { + qWarning("Failed to create source blob for dxc: 0x%x (%s)", + uint(hr), + qPrintable(QSystemError::windowsComString(hr))); + return QByteArray(); + } + + const QString entryPointStr = QString::fromLatin1(hlslSource.entryPoint()); + const QString targetStr = QString::fromLatin1(target); + + QVarLengthArray<LPCWSTR, 4> argPtrs; + QString debugArg; + if (flags & int(HlslCompileFlag::WithDebugInfo)) { + debugArg = QString::fromUtf16(reinterpret_cast<const char16_t *>(DXC_ARG_DEBUG)); + argPtrs.append(reinterpret_cast<LPCWSTR>(debugArg.utf16())); + } + + IDxcOperationResult *result = nullptr; + hr = compiler->Compile(sourceBlob, + nullptr, + reinterpret_cast<LPCWSTR>(entryPointStr.utf16()), + reinterpret_cast<LPCWSTR>(targetStr.utf16()), + argPtrs.data(), argPtrs.count(), + nullptr, 0, + nullptr, + &result); + sourceBlob->Release(); + if (SUCCEEDED(hr)) + result->GetStatus(&hr); + if (FAILED(hr)) { + qWarning("HLSL shader compilation failed: 0x%x (%s)", + uint(hr), + qPrintable(QSystemError::windowsComString(hr))); + if (result) { + IDxcBlobEncoding *errorsBlob = nullptr; + if (SUCCEEDED(result->GetErrorBuffer(&errorsBlob))) { + if (errorsBlob) { + *error = QString::fromUtf8(static_cast<const char *>(errorsBlob->GetBufferPointer()), + int(errorsBlob->GetBufferSize())); + errorsBlob->Release(); + } + } + } + return QByteArray(); + } + + IDxcBlob *bytecode = nullptr; + if FAILED(result->GetResult(&bytecode)) { + qWarning("No result from IDxcCompiler: 0x%x (%s)", + uint(hr), + qPrintable(QSystemError::windowsComString(hr))); + return QByteArray(); + } + + QByteArray ba; + ba.resize(int(bytecode->GetBufferSize())); + memcpy(ba.data(), bytecode->GetBufferPointer(), size_t(ba.size())); + bytecode->Release(); + return ba; +} + +#endif // QRHI_D3D12_HAS_DXC + static QByteArray compileHlslShaderSource(const QShader &shader, QShader::Variant shaderVariant, - UINT flags, + int flags, QString *error, QShaderKey *usedShaderKey) { @@ -4900,33 +5301,17 @@ static QByteArray compileHlslShaderSource(const QShader &shader, break; } - static const pD3DCompile d3dCompile = resolveD3DCompile(); - if (!d3dCompile) { - qWarning("Unable to resolve function D3DCompile()"); - return QByteArray(); - } - - ID3DBlob *bytecode = nullptr; - ID3DBlob *errors = nullptr; - HRESULT hr = d3dCompile(hlslSource.shader().constData(), SIZE_T(hlslSource.shader().size()), - nullptr, nullptr, nullptr, - hlslSource.entryPoint().constData(), target, flags, 0, &bytecode, &errors); - if (FAILED(hr) || !bytecode) { - qWarning("HLSL shader compilation failed: 0x%x", uint(hr)); - if (errors) { - *error = QString::fromUtf8(static_cast<const char *>(errors->GetBufferPointer()), - int(errors->GetBufferSize())); - errors->Release(); - } - return QByteArray(); + if (key.sourceVersion().version() >= 60) { +#ifdef QRHI_D3D12_HAS_DXC + return dxcCompile(hlslSource, target, flags, error); +#else + qWarning("Attempted to runtime-compile HLSL source code for shader model >= 6.0 " + "but the Qt build has no support for DXC. " + "Rebuild Qt with a recent Windows SDK or switch to an MSVC build."); +#endif } - QByteArray result; - result.resize(int(bytecode->GetBufferSize())); - memcpy(result.data(), bytecode->GetBufferPointer(), size_t(result.size())); - bytecode->Release(); - - return result; + return legacyCompile(hlslSource, target, flags, error); } static inline UINT8 toD3DColorWriteMask(QRhiGraphicsPipeline::ColorMask c) @@ -5161,6 +5546,22 @@ static inline DXGI_FORMAT toD3DAttributeFormat(QRhiVertexInputAttribute::Format return DXGI_FORMAT_R16G16_FLOAT; case QRhiVertexInputAttribute::Half: return DXGI_FORMAT_R16_FLOAT; + case QRhiVertexInputAttribute::UShort4: + // Note: D3D does not support UShort3. Pass through UShort3 as UShort4. + case QRhiVertexInputAttribute::UShort3: + return DXGI_FORMAT_R16G16B16A16_UINT; + case QRhiVertexInputAttribute::UShort2: + return DXGI_FORMAT_R16G16_UINT; + case QRhiVertexInputAttribute::UShort: + return DXGI_FORMAT_R16_UINT; + case QRhiVertexInputAttribute::SShort4: + // Note: D3D does not support SShort3. Pass through SShort3 as SShort4. + case QRhiVertexInputAttribute::SShort3: + return DXGI_FORMAT_R16G16B16A16_SINT; + case QRhiVertexInputAttribute::SShort2: + return DXGI_FORMAT_R16G16_SINT; + case QRhiVertexInputAttribute::SShort: + return DXGI_FORMAT_R16_SINT; } Q_UNREACHABLE_RETURN(DXGI_FORMAT_R32G32B32A32_FLOAT); } @@ -5216,16 +5617,16 @@ bool QD3D12GraphicsPipeline::create() } else { QString error; QShaderKey shaderKey; - UINT compileFlags = 0; + int compileFlags = 0; if (m_flags.testFlag(CompileShadersWithDebugInfo)) - compileFlags |= D3DCOMPILE_DEBUG; + compileFlags |= int(HlslCompileFlag::WithDebugInfo); const QByteArray bytecode = compileHlslShaderSource(shaderStage.shader(), shaderStage.shaderVariant(), compileFlags, &error, &shaderKey); if (bytecode.isEmpty()) { - qWarning("HLSL compute shader compilation failed: %s", qPrintable(error)); + qWarning("HLSL graphics shader compilation failed: %s", qPrintable(error)); return false; } @@ -5253,32 +5654,94 @@ bool QD3D12GraphicsPipeline::create() } QD3D12RenderPassDescriptor *rpD = QRHI_RES(QD3D12RenderPassDescriptor, m_renderPassDesc); - const DXGI_SAMPLE_DESC sampleDesc = rhiD->effectiveSampleCount(m_sampleCount, DXGI_FORMAT(rpD->colorFormat[0])); + const DXGI_SAMPLE_DESC sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount, DXGI_FORMAT(rpD->colorFormat[0])); + + struct { + QD3D12PipelineStateSubObject<ID3D12RootSignature *, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_ROOT_SIGNATURE> rootSig; + QD3D12PipelineStateSubObject<D3D12_INPUT_LAYOUT_DESC, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_INPUT_LAYOUT> inputLayout; + QD3D12PipelineStateSubObject<D3D12_PRIMITIVE_TOPOLOGY_TYPE, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_PRIMITIVE_TOPOLOGY> primitiveTopology; + QD3D12PipelineStateSubObject<D3D12_SHADER_BYTECODE, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_VS> VS; + QD3D12PipelineStateSubObject<D3D12_SHADER_BYTECODE, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_HS> HS; + QD3D12PipelineStateSubObject<D3D12_SHADER_BYTECODE, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_DS> DS; + QD3D12PipelineStateSubObject<D3D12_SHADER_BYTECODE, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_GS> GS; + QD3D12PipelineStateSubObject<D3D12_SHADER_BYTECODE, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_PS> PS; + QD3D12PipelineStateSubObject<D3D12_RASTERIZER_DESC, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_RASTERIZER> rasterizerState; + QD3D12PipelineStateSubObject<D3D12_DEPTH_STENCIL_DESC, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_DEPTH_STENCIL> depthStencilState; + QD3D12PipelineStateSubObject<D3D12_BLEND_DESC, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_BLEND> blendState; + QD3D12PipelineStateSubObject<D3D12_RT_FORMAT_ARRAY, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_RENDER_TARGET_FORMATS> rtFormats; + QD3D12PipelineStateSubObject<DXGI_FORMAT, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_DEPTH_STENCIL_FORMAT> dsFormat; + QD3D12PipelineStateSubObject<DXGI_SAMPLE_DESC, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_SAMPLE_DESC> sampleDesc; + QD3D12PipelineStateSubObject<UINT, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_SAMPLE_MASK> sampleMask; + QD3D12PipelineStateSubObject<D3D12_VIEW_INSTANCING_DESC, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_VIEW_INSTANCING> viewInstancingDesc; + } stream; + + stream.rootSig.object = rootSig; + + QVarLengthArray<D3D12_INPUT_ELEMENT_DESC, 4> inputDescs; + QByteArrayList matrixSliceSemantics; + if (!shaderBytecode[VS].isEmpty()) { + for (auto it = m_vertexInputLayout.cbeginAttributes(), itEnd = m_vertexInputLayout.cendAttributes(); + it != itEnd; ++it) + { + D3D12_INPUT_ELEMENT_DESC desc = {}; + // The output from SPIRV-Cross uses TEXCOORD<location> as the + // semantic, except for matrices that are unrolled into consecutive + // vec2/3/4s attributes and need TEXCOORD<location>_ as + // SemanticName and row/column index as SemanticIndex. + const int matrixSlice = it->matrixSlice(); + if (matrixSlice < 0) { + desc.SemanticName = "TEXCOORD"; + desc.SemanticIndex = UINT(it->location()); + } else { + QByteArray sem; + sem.resize(16); + qsnprintf(sem.data(), sem.size(), "TEXCOORD%d_", it->location() - matrixSlice); + matrixSliceSemantics.append(sem); + desc.SemanticName = matrixSliceSemantics.last().constData(); + desc.SemanticIndex = UINT(matrixSlice); + } + desc.Format = toD3DAttributeFormat(it->format()); + desc.InputSlot = UINT(it->binding()); + desc.AlignedByteOffset = it->offset(); + const QRhiVertexInputBinding *inputBinding = m_vertexInputLayout.bindingAt(it->binding()); + if (inputBinding->classification() == QRhiVertexInputBinding::PerInstance) { + desc.InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_INSTANCE_DATA; + desc.InstanceDataStepRate = inputBinding->instanceStepRate(); + } else { + desc.InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA; + } + inputDescs.append(desc); + } + } + + stream.inputLayout.object.NumElements = inputDescs.count(); + stream.inputLayout.object.pInputElementDescs = inputDescs.isEmpty() ? nullptr : inputDescs.constData(); + + stream.primitiveTopology.object = toD3DTopologyType(m_topology); + topology = toD3DTopology(m_topology, m_patchControlPointCount); - D3D12_GRAPHICS_PIPELINE_STATE_DESC psoDesc = {}; - psoDesc.pRootSignature = rootSig; for (const QRhiShaderStage &shaderStage : std::as_const(m_shaderStages)) { const int d3dStage = qd3d12_stage(shaderStage.type()); switch (d3dStage) { case VS: - psoDesc.VS.pShaderBytecode = shaderBytecode[d3dStage].constData(); - psoDesc.VS.BytecodeLength = shaderBytecode[d3dStage].size(); + stream.VS.object.pShaderBytecode = shaderBytecode[d3dStage].constData(); + stream.VS.object.BytecodeLength = shaderBytecode[d3dStage].size(); break; case HS: - psoDesc.HS.pShaderBytecode = shaderBytecode[d3dStage].constData(); - psoDesc.HS.BytecodeLength = shaderBytecode[d3dStage].size(); + stream.HS.object.pShaderBytecode = shaderBytecode[d3dStage].constData(); + stream.HS.object.BytecodeLength = shaderBytecode[d3dStage].size(); break; case DS: - psoDesc.DS.pShaderBytecode = shaderBytecode[d3dStage].constData(); - psoDesc.DS.BytecodeLength = shaderBytecode[d3dStage].size(); + stream.DS.object.pShaderBytecode = shaderBytecode[d3dStage].constData(); + stream.DS.object.BytecodeLength = shaderBytecode[d3dStage].size(); break; case GS: - psoDesc.GS.pShaderBytecode = shaderBytecode[d3dStage].constData(); - psoDesc.GS.BytecodeLength = shaderBytecode[d3dStage].size(); + stream.GS.object.pShaderBytecode = shaderBytecode[d3dStage].constData(); + stream.GS.object.BytecodeLength = shaderBytecode[d3dStage].size(); break; case PS: - psoDesc.PS.pShaderBytecode = shaderBytecode[d3dStage].constData(); - psoDesc.PS.BytecodeLength = shaderBytecode[d3dStage].size(); + stream.PS.object.pShaderBytecode = shaderBytecode[d3dStage].constData(); + stream.PS.object.BytecodeLength = shaderBytecode[d3dStage].size(); break; default: Q_UNREACHABLE(); @@ -5286,7 +5749,32 @@ bool QD3D12GraphicsPipeline::create() } } - psoDesc.BlendState.IndependentBlendEnable = m_targetBlends.count() > 1; + stream.rasterizerState.object.FillMode = toD3DFillMode(m_polygonMode); + stream.rasterizerState.object.CullMode = toD3DCullMode(m_cullMode); + stream.rasterizerState.object.FrontCounterClockwise = m_frontFace == CCW; + stream.rasterizerState.object.DepthBias = m_depthBias; + stream.rasterizerState.object.SlopeScaledDepthBias = m_slopeScaledDepthBias; + stream.rasterizerState.object.DepthClipEnable = TRUE; + stream.rasterizerState.object.MultisampleEnable = sampleDesc.Count > 1; + + stream.depthStencilState.object.DepthEnable = m_depthTest; + stream.depthStencilState.object.DepthWriteMask = m_depthWrite ? D3D12_DEPTH_WRITE_MASK_ALL : D3D12_DEPTH_WRITE_MASK_ZERO; + stream.depthStencilState.object.DepthFunc = toD3DCompareOp(m_depthOp); + stream.depthStencilState.object.StencilEnable = m_stencilTest; + if (m_stencilTest) { + stream.depthStencilState.object.StencilReadMask = UINT8(m_stencilReadMask); + stream.depthStencilState.object.StencilWriteMask = UINT8(m_stencilWriteMask); + stream.depthStencilState.object.FrontFace.StencilFailOp = toD3DStencilOp(m_stencilFront.failOp); + stream.depthStencilState.object.FrontFace.StencilDepthFailOp = toD3DStencilOp(m_stencilFront.depthFailOp); + stream.depthStencilState.object.FrontFace.StencilPassOp = toD3DStencilOp(m_stencilFront.passOp); + stream.depthStencilState.object.FrontFace.StencilFunc = toD3DCompareOp(m_stencilFront.compareOp); + stream.depthStencilState.object.BackFace.StencilFailOp = toD3DStencilOp(m_stencilBack.failOp); + stream.depthStencilState.object.BackFace.StencilDepthFailOp = toD3DStencilOp(m_stencilBack.depthFailOp); + stream.depthStencilState.object.BackFace.StencilPassOp = toD3DStencilOp(m_stencilBack.passOp); + stream.depthStencilState.object.BackFace.StencilFunc = toD3DCompareOp(m_stencilBack.compareOp); + } + + stream.blendState.object.IndependentBlendEnable = m_targetBlends.count() > 1; for (int i = 0, ie = m_targetBlends.count(); i != ie; ++i) { const QRhiGraphicsPipeline::TargetBlend &b(m_targetBlends[i]); D3D12_RENDER_TARGET_BLEND_DESC blend = {}; @@ -5298,95 +5786,40 @@ bool QD3D12GraphicsPipeline::create() blend.DestBlendAlpha = toD3DBlendFactor(b.dstAlpha, false); blend.BlendOpAlpha = toD3DBlendOp(b.opAlpha); blend.RenderTargetWriteMask = toD3DColorWriteMask(b.colorWrite); - psoDesc.BlendState.RenderTarget[i] = blend; + stream.blendState.object.RenderTarget[i] = blend; } if (m_targetBlends.isEmpty()) { D3D12_RENDER_TARGET_BLEND_DESC blend = {}; blend.RenderTargetWriteMask = D3D12_COLOR_WRITE_ENABLE_ALL; - psoDesc.BlendState.RenderTarget[0] = blend; + stream.blendState.object.RenderTarget[0] = blend; } - psoDesc.SampleMask = 0xFFFFFFFF; + stream.rtFormats.object.NumRenderTargets = rpD->colorAttachmentCount; + for (int i = 0; i < rpD->colorAttachmentCount; ++i) + stream.rtFormats.object.RTFormats[i] = DXGI_FORMAT(rpD->colorFormat[i]); - psoDesc.RasterizerState.FillMode = toD3DFillMode(m_polygonMode); - psoDesc.RasterizerState.CullMode = toD3DCullMode(m_cullMode); - psoDesc.RasterizerState.FrontCounterClockwise = m_frontFace == CCW; - psoDesc.RasterizerState.DepthBias = m_depthBias; - psoDesc.RasterizerState.SlopeScaledDepthBias = m_slopeScaledDepthBias; - psoDesc.RasterizerState.DepthClipEnable = TRUE; - psoDesc.RasterizerState.MultisampleEnable = sampleDesc.Count > 1; + stream.dsFormat.object = rpD->hasDepthStencil ? DXGI_FORMAT(rpD->dsFormat) : DXGI_FORMAT_UNKNOWN; - psoDesc.DepthStencilState.DepthEnable = m_depthTest; - psoDesc.DepthStencilState.DepthWriteMask = m_depthWrite ? D3D12_DEPTH_WRITE_MASK_ALL : D3D12_DEPTH_WRITE_MASK_ZERO; - psoDesc.DepthStencilState.DepthFunc = toD3DCompareOp(m_depthOp); - psoDesc.DepthStencilState.StencilEnable = m_stencilTest; - if (m_stencilTest) { - psoDesc.DepthStencilState.StencilReadMask = UINT8(m_stencilReadMask); - psoDesc.DepthStencilState.StencilWriteMask = UINT8(m_stencilWriteMask); - psoDesc.DepthStencilState.FrontFace.StencilFailOp = toD3DStencilOp(m_stencilFront.failOp); - psoDesc.DepthStencilState.FrontFace.StencilDepthFailOp = toD3DStencilOp(m_stencilFront.depthFailOp); - psoDesc.DepthStencilState.FrontFace.StencilPassOp = toD3DStencilOp(m_stencilFront.passOp); - psoDesc.DepthStencilState.FrontFace.StencilFunc = toD3DCompareOp(m_stencilFront.compareOp); - psoDesc.DepthStencilState.BackFace.StencilFailOp = toD3DStencilOp(m_stencilBack.failOp); - psoDesc.DepthStencilState.BackFace.StencilDepthFailOp = toD3DStencilOp(m_stencilBack.depthFailOp); - psoDesc.DepthStencilState.BackFace.StencilPassOp = toD3DStencilOp(m_stencilBack.passOp); - psoDesc.DepthStencilState.BackFace.StencilFunc = toD3DCompareOp(m_stencilBack.compareOp); - } + stream.sampleDesc.object = sampleDesc; - QVarLengthArray<D3D12_INPUT_ELEMENT_DESC, 4> inputDescs; - QByteArrayList matrixSliceSemantics; - if (!shaderBytecode[VS].isEmpty()) { - for (auto it = m_vertexInputLayout.cbeginAttributes(), itEnd = m_vertexInputLayout.cendAttributes(); - it != itEnd; ++it) - { - D3D12_INPUT_ELEMENT_DESC desc = {}; - // The output from SPIRV-Cross uses TEXCOORD<location> as the - // semantic, except for matrices that are unrolled into consecutive - // vec2/3/4s attributes and need TEXCOORD<location>_ as - // SemanticName and row/column index as SemanticIndex. - const int matrixSlice = it->matrixSlice(); - if (matrixSlice < 0) { - desc.SemanticName = "TEXCOORD"; - desc.SemanticIndex = UINT(it->location()); - } else { - QByteArray sem; - sem.resize(16); - qsnprintf(sem.data(), sem.size(), "TEXCOORD%d_", it->location() - matrixSlice); - matrixSliceSemantics.append(sem); - desc.SemanticName = matrixSliceSemantics.last().constData(); - desc.SemanticIndex = UINT(matrixSlice); - } - desc.Format = toD3DAttributeFormat(it->format()); - desc.InputSlot = UINT(it->binding()); - desc.AlignedByteOffset = it->offset(); - const QRhiVertexInputBinding *inputBinding = m_vertexInputLayout.bindingAt(it->binding()); - if (inputBinding->classification() == QRhiVertexInputBinding::PerInstance) { - desc.InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_INSTANCE_DATA; - desc.InstanceDataStepRate = inputBinding->instanceStepRate(); - } else { - desc.InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA; - } - inputDescs.append(desc); + stream.sampleMask.object = 0xFFFFFFFF; + + viewInstanceMask = 0; + const bool isMultiView = m_multiViewCount >= 2; + stream.viewInstancingDesc.object.ViewInstanceCount = isMultiView ? m_multiViewCount : 0; + QVarLengthArray<D3D12_VIEW_INSTANCE_LOCATION, 4> viewInstanceLocations; + if (isMultiView) { + for (int i = 0; i < m_multiViewCount; ++i) { + viewInstanceMask |= (1 << i); + viewInstanceLocations.append({ 0, UINT(i) }); } + stream.viewInstancingDesc.object.pViewInstanceLocations = viewInstanceLocations.constData(); } - if (!inputDescs.isEmpty()) { - psoDesc.InputLayout.pInputElementDescs = inputDescs.constData(); - psoDesc.InputLayout.NumElements = inputDescs.count(); - } - - psoDesc.PrimitiveTopologyType = toD3DTopologyType(m_topology); - topology = toD3DTopology(m_topology, m_patchControlPointCount); - psoDesc.NumRenderTargets = rpD->colorAttachmentCount; - for (int i = 0; i < rpD->colorAttachmentCount; ++i) - psoDesc.RTVFormats[i] = DXGI_FORMAT(rpD->colorFormat[i]); - psoDesc.DSVFormat = rpD->hasDepthStencil ? DXGI_FORMAT(rpD->dsFormat) : DXGI_FORMAT_UNKNOWN; - psoDesc.SampleDesc = sampleDesc; + const D3D12_PIPELINE_STATE_STREAM_DESC streamDesc = { sizeof(stream), &stream }; ID3D12PipelineState *pso = nullptr; - HRESULT hr = rhiD->dev->CreateGraphicsPipelineState(&psoDesc, - __uuidof(ID3D12PipelineState), - reinterpret_cast<void **>(&pso)); + HRESULT hr = rhiD->dev->CreatePipelineState(&streamDesc, __uuidof(ID3D12PipelineState), reinterpret_cast<void **>(&pso)); if (FAILED(hr)) { qWarning("Failed to create graphics pipeline state: %s", qPrintable(QSystemError::windowsComString(hr))); @@ -5450,9 +5883,9 @@ bool QD3D12ComputePipeline::create() } else { QString error; QShaderKey shaderKey; - UINT compileFlags = 0; + int compileFlags = 0; if (m_flags.testFlag(CompileShadersWithDebugInfo)) - compileFlags |= D3DCOMPILE_DEBUG; + compileFlags |= int(HlslCompileFlag::WithDebugInfo); const QByteArray bytecode = compileHlslShaderSource(m_shaderStage.shader(), m_shaderStage.shaderVariant(), compileFlags, @@ -5485,14 +5918,16 @@ bool QD3D12ComputePipeline::create() return false; } - D3D12_COMPUTE_PIPELINE_STATE_DESC psoDesc = {}; - psoDesc.pRootSignature = rootSig; - psoDesc.CS.pShaderBytecode = shaderBytecode.constData(); - psoDesc.CS.BytecodeLength = shaderBytecode.size(); + struct { + QD3D12PipelineStateSubObject<ID3D12RootSignature *, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_ROOT_SIGNATURE> rootSig; + QD3D12PipelineStateSubObject<D3D12_SHADER_BYTECODE, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE_CS> CS; + } stream; + stream.rootSig.object = rootSig; + stream.CS.object.pShaderBytecode = shaderBytecode.constData(); + stream.CS.object.BytecodeLength = shaderBytecode.size(); + const D3D12_PIPELINE_STATE_STREAM_DESC streamDesc = { sizeof(stream), &stream }; ID3D12PipelineState *pso = nullptr; - HRESULT hr = rhiD->dev->CreateComputePipelineState(&psoDesc, - __uuidof(ID3D12PipelineState), - reinterpret_cast<void **>(&pso)); + HRESULT hr = rhiD->dev->CreatePipelineState(&streamDesc, __uuidof(ID3D12PipelineState), reinterpret_cast<void **>(&pso)); if (FAILED(hr)) { qWarning("Failed to create compute pipeline state: %s", qPrintable(QSystemError::windowsComString(hr))); @@ -5644,6 +6079,7 @@ int QD3D12SwapChainRenderTarget::sampleCount() const QD3D12SwapChain::QD3D12SwapChain(QRhiImplementation *rhi) : QRhiSwapChain(rhi), rtWrapper(rhi, this), + rtWrapperRight(rhi, this), cbWrapper(rhi) { } @@ -5700,6 +6136,8 @@ void QD3D12SwapChain::releaseBuffers() for (UINT i = 0; i < BUFFER_COUNT; ++i) { rhiD->resourcePool.remove(colorBuffers[i]); rhiD->rtvPool.release(rtvs[i], 1); + if (stereo) + rhiD->rtvPool.release(rtvsRight[i], 1); if (!msaaBuffers[i].isNull()) rhiD->resourcePool.remove(msaaBuffers[i]); if (msaaRtvs[i].isValid()) @@ -5734,48 +6172,15 @@ QRhiRenderTarget *QD3D12SwapChain::currentFrameRenderTarget() return &rtWrapper; } -QSize QD3D12SwapChain::surfacePixelSize() +QRhiRenderTarget *QD3D12SwapChain::currentFrameRenderTarget(StereoTargetBuffer targetBuffer) { - Q_ASSERT(m_window); - return m_window->size() * m_window->devicePixelRatio(); + return !stereo || targetBuffer == StereoTargetBuffer::LeftBuffer ? &rtWrapper : &rtWrapperRight; } -static bool output6ForWindow(QWindow *w, IDXGIAdapter1 *adapter, IDXGIOutput6 **result) -{ - bool ok = false; - QRect wr = w->geometry(); - wr = QRect(wr.topLeft() * w->devicePixelRatio(), wr.size() * w->devicePixelRatio()); - const QPoint center = wr.center(); - IDXGIOutput *currentOutput = nullptr; - IDXGIOutput *output = nullptr; - for (UINT i = 0; adapter->EnumOutputs(i, &output) != DXGI_ERROR_NOT_FOUND; ++i) { - DXGI_OUTPUT_DESC desc; - output->GetDesc(&desc); - const RECT r = desc.DesktopCoordinates; - const QRect dr(QPoint(r.left, r.top), QPoint(r.right - 1, r.bottom - 1)); - if (dr.contains(center)) { - currentOutput = output; - break; - } else { - output->Release(); - } - } - if (currentOutput) { - ok = SUCCEEDED(currentOutput->QueryInterface(__uuidof(IDXGIOutput6), reinterpret_cast<void **>(result))); - currentOutput->Release(); - } - return ok; -} - -static bool outputDesc1ForWindow(QWindow *w, IDXGIAdapter1 *adapter, DXGI_OUTPUT_DESC1 *result) +QSize QD3D12SwapChain::surfacePixelSize() { - bool ok = false; - IDXGIOutput6 *out6 = nullptr; - if (output6ForWindow(w, adapter, &out6)) { - ok = SUCCEEDED(out6->GetDesc1(result)); - out6->Release(); - } - return ok; + Q_ASSERT(m_window); + return m_window->size() * m_window->devicePixelRatio(); } bool QD3D12SwapChain::isFormatSupported(Format f) @@ -5790,7 +6195,7 @@ bool QD3D12SwapChain::isFormatSupported(Format f) QRHI_RES_RHI(QRhiD3D12); DXGI_OUTPUT_DESC1 desc1; - if (outputDesc1ForWindow(m_window, rhiD->activeAdapter, &desc1)) { + if (QRhiD3D::outputDesc1ForWindow(m_window, rhiD->activeAdapter, &desc1)) { if (desc1.ColorSpace == DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020) return f == QRhiSwapChain::HDRExtendedSrgbLinear || f == QRhiSwapChain::HDR10; } @@ -5801,14 +6206,16 @@ bool QD3D12SwapChain::isFormatSupported(Format f) QRhiSwapChainHdrInfo QD3D12SwapChain::hdrInfo() { QRhiSwapChainHdrInfo info = QRhiSwapChain::hdrInfo(); - if (m_format != QRhiSwapChain::SDR && m_window) { + // Must use m_window, not window, given this may be called before createOrResize(). + if (m_window) { QRHI_RES_RHI(QRhiD3D12); DXGI_OUTPUT_DESC1 hdrOutputDesc; - if (outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc)) { - info.isHardCodedDefaults = false; + if (QRhiD3D::outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc)) { info.limitsType = QRhiSwapChainHdrInfo::LuminanceInNits; info.limits.luminanceInNits.minLuminance = hdrOutputDesc.MinLuminance; info.limits.luminanceInNits.maxLuminance = hdrOutputDesc.MaxLuminance; + info.luminanceBehavior = QRhiSwapChainHdrInfo::SceneReferred; // 1.0 = 80 nits + info.sdrWhiteLevel = QRhiD3D::sdrWhiteLevelInNits(hdrOutputDesc); } } return info; @@ -5831,25 +6238,19 @@ QRhiRenderPassDescriptor *QD3D12SwapChain::newCompatibleRenderPassDescriptor() return rpD; } -static const DXGI_FORMAT DEFAULT_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM; -static const DXGI_FORMAT DEFAULT_SRGB_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM_SRGB; - bool QRhiD3D12::ensureDirectCompositionDevice() { if (dcompDevice) return true; qCDebug(QRHI_LOG_INFO, "Creating Direct Composition device (needed for semi-transparent windows)"); - - HRESULT hr = DCompositionCreateDevice(nullptr, __uuidof(IDCompositionDevice), reinterpret_cast<void **>(&dcompDevice)); - if (FAILED(hr)) { - qWarning("Failed to Direct Composition device: %s", qPrintable(QSystemError::windowsComString(hr))); - return false; - } - - return true; + dcompDevice = QRhiD3D::createDirectCompositionDevice(); + return dcompDevice ? true : false; } +static const DXGI_FORMAT DEFAULT_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM; +static const DXGI_FORMAT DEFAULT_SRGB_FORMAT = DXGI_FORMAT_R8G8B8A8_UNORM_SRGB; + void QD3D12SwapChain::chooseFormats() { colorFormat = DEFAULT_FORMAT; @@ -5857,7 +6258,7 @@ void QD3D12SwapChain::chooseFormats() hdrColorSpace = DXGI_COLOR_SPACE_RGB_FULL_G22_NONE_P709; // SDR DXGI_OUTPUT_DESC1 hdrOutputDesc; QRHI_RES_RHI(QRhiD3D12); - if (outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc) && m_format != SDR) { + if (QRhiD3D::outputDesc1ForWindow(m_window, rhiD->activeAdapter, &hdrOutputDesc) && m_format != SDR) { // https://docs.microsoft.com/en-us/windows/win32/direct3darticles/high-dynamic-range if (hdrOutputDesc.ColorSpace == DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020) { switch (m_format) { @@ -5882,7 +6283,7 @@ void QD3D12SwapChain::chooseFormats() "(or Use HDR is Off in the Display Settings), ignoring HDR format request"); } } - sampleDesc = rhiD->effectiveSampleCount(m_sampleCount, colorFormat); + sampleDesc = rhiD->effectiveSampleDesc(m_sampleCount, colorFormat); } bool QD3D12SwapChain::createOrResize() @@ -5907,13 +6308,14 @@ bool QD3D12SwapChain::createOrResize() HWND hwnd = reinterpret_cast<HWND>(window->winId()); HRESULT hr; QRHI_RES_RHI(QRhiD3D12); + stereo = m_window->format().stereo() && rhiD->dxgiFactory->IsWindowedStereoEnabled(); if (m_flags.testFlag(SurfaceHasPreMulAlpha) || m_flags.testFlag(SurfaceHasNonPreMulAlpha)) { if (rhiD->ensureDirectCompositionDevice()) { if (!dcompTarget) { - hr = rhiD->dcompDevice->CreateTargetForHwnd(hwnd, true, &dcompTarget); + hr = rhiD->dcompDevice->CreateTargetForHwnd(hwnd, false, &dcompTarget); if (FAILED(hr)) { - qWarning("Failed to create Direct Compsition target for the window: %s", + qWarning("Failed to create Direct Composition target for the window: %s", qPrintable(QSystemError::windowsComString(hr))); } } @@ -5949,6 +6351,7 @@ bool QD3D12SwapChain::createOrResize() desc.Flags = swapChainFlags; desc.Scaling = DXGI_SCALING_NONE; desc.SwapEffect = DXGI_SWAP_EFFECT_FLIP_DISCARD; + desc.Stereo = stereo; if (dcompVisual) { // With DirectComposition setting AlphaMode to STRAIGHT fails the @@ -6003,13 +6406,19 @@ bool QD3D12SwapChain::createOrResize() qWarning("Failed to set content for Direct Composition visual: %s", qPrintable(QSystemError::windowsComString(hr))); } + } else { + // disable Alt+Enter; not relevant when using DirectComposition + rhiD->dxgiFactory->MakeWindowAssociation(hwnd, DXGI_MWA_NO_WINDOW_CHANGES); } } if (FAILED(hr)) { - qWarning("Failed to create D3D12 swapchain: %s", qPrintable(QSystemError::windowsComString(hr))); + qWarning("Failed to create D3D12 swapchain: %s" + " (Width=%u Height=%u Format=%u SampleCount=%u BufferCount=%u Scaling=%u SwapEffect=%u Stereo=%u)", + qPrintable(QSystemError::windowsComString(hr)), + desc.Width, desc.Height, UINT(desc.Format), desc.SampleDesc.Count, + desc.BufferCount, UINT(desc.Scaling), UINT(desc.SwapEffect), UINT(desc.Stereo)); return false; } - rhiD->dxgiFactory->MakeWindowAssociation(hwnd, DXGI_MWA_NO_WINDOW_CHANGES); for (int i = 0; i < QD3D12_FRAMES_IN_FLIGHT; ++i) { hr = rhiD->dev->CreateFence(0, @@ -6056,6 +6465,16 @@ bool QD3D12SwapChain::createOrResize() rtvDesc.Format = srgbAdjustedColorFormat; rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2D; rhiD->dev->CreateRenderTargetView(colorBuffer, &rtvDesc, rtvs[i].cpuHandle); + + if (stereo) { + rtvsRight[i] = rhiD->rtvPool.allocate(1); + D3D12_RENDER_TARGET_VIEW_DESC rtvDesc = {}; + rtvDesc.Format = srgbAdjustedColorFormat; + rtvDesc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2DARRAY; + rtvDesc.Texture2DArray.ArraySize = 1; + rtvDesc.Texture2DArray.FirstArraySlice = 1; + rhiD->dev->CreateRenderTargetView(colorBuffer, &rtvDesc, rtvsRight[i].cpuHandle); + } } if (m_depthStencil && m_depthStencil->sampleCount() != m_sampleCount) { @@ -6128,6 +6547,15 @@ bool QD3D12SwapChain::createOrResize() rtD->d.colorAttCount = 1; rtD->d.dsAttCount = m_depthStencil ? 1 : 0; + rtWrapperRight.setRenderPassDescriptor(m_renderPassDesc); + QD3D12SwapChainRenderTarget *rtDr = QRHI_RES(QD3D12SwapChainRenderTarget, &rtWrapperRight); + rtDr->d.rp = QRHI_RES(QD3D12RenderPassDescriptor, m_renderPassDesc); + rtDr->d.pixelSize = pixelSize; + rtDr->d.dpr = float(window->devicePixelRatio()); + rtDr->d.sampleCount = int(sampleDesc.Count); + rtDr->d.colorAttCount = 1; + rtDr->d.dsAttCount = m_depthStencil ? 1 : 0; + if (needsRegistration) { rhiD->swapchains.insert(this); rhiD->registerResource(this); @@ -6137,3 +6565,5 @@ bool QD3D12SwapChain::createOrResize() } QT_END_NAMESPACE + +#endif // __ID3D12Device2_INTERFACE_DEFINED__ diff --git a/src/gui/rhi/qrhid3d12_p.h b/src/gui/rhi/qrhid3d12_p.h index d40046355b..3f9abbb5ac 100644 --- a/src/gui/rhi/qrhid3d12_p.h +++ b/src/gui/rhi/qrhid3d12_p.h @@ -16,7 +16,6 @@ // #include "qrhi_p.h" -#include "qshaderdescription.h" #include <QWindow> #include <QBitArray> @@ -30,6 +29,15 @@ #include "D3D12MemAlloc.h" +// ID3D12Device2 and ID3D12GraphicsCommandList1 and types and enums introduced +// with those are hard requirements now. These should be declared in any +// moderately recent d3d12.h, but if it is an SDK from before Windows 10 +// version 1703 then these types could be missing. In the absence of other +// options, handle this by skipping all the code and making QRhi::create() fail +// in such builds. +#ifdef __ID3D12Device2_INTERFACE_DEFINED__ +#define QRHI_D3D12_AVAILABLE + QT_BEGIN_NAMESPACE static const int QD3D12_FRAMES_IN_FLIGHT = 2; @@ -110,6 +118,18 @@ struct QD3D12CpuDescriptorPool const char *debugName; }; +struct QD3D12QueryHeap +{ + bool isValid() const { return heap && capacity; } + bool create(ID3D12Device *device, + quint32 queryCount, + D3D12_QUERY_HEAP_TYPE heapType); + void destroy(); + + ID3D12QueryHeap *heap = nullptr; + quint32 capacity = 0; +}; + struct QD3D12StagingArea { static const quint32 ALIGNMENT = D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT; // 512 so good enough both for cb and texdata @@ -667,6 +687,7 @@ struct QD3D12Buffer : public QRhiBuffer }; QVarLengthArray<HostWrite, 16> pendingHostWrites[QD3D12_FRAMES_IN_FLIGHT]; friend class QRhiD3D12; + friend struct QD3D12CommandBuffer; }; struct QD3D12RenderBuffer : public QRhiRenderBuffer @@ -710,10 +731,13 @@ struct QD3D12Texture : public QRhiTexture QD3D12ObjectHandle handle; QD3D12Descriptor srv; DXGI_FORMAT dxgiFormat; + DXGI_FORMAT srvFormat; + DXGI_FORMAT rtFormat; // RTV/DSV/UAV uint mipLevelCount; DXGI_SAMPLE_DESC sampleDesc; uint generation = 0; friend class QRhiD3D12; + friend struct QD3D12CommandBuffer; }; struct QD3D12Sampler : public QRhiSampler @@ -847,9 +871,11 @@ struct QD3D12ShaderResourceBindings : public QRhiShaderResourceBindings QD3D12ShaderResourceVisitor::StorageOp op, int shaderRegister); - QVarLengthArray<QRhiShaderResourceBinding, 8> sortedBindings; bool hasDynamicOffset = false; uint generation = 0; + + friend class QRhiD3D12; + friend struct QD3D12ShaderResourceVisitor; }; struct QD3D12GraphicsPipeline : public QRhiGraphicsPipeline @@ -863,6 +889,7 @@ struct QD3D12GraphicsPipeline : public QRhiGraphicsPipeline QD3D12ObjectHandle rootSigHandle; std::array<QD3D12ShaderStageData, 5> stageData; D3D12_PRIMITIVE_TOPOLOGY topology; + UINT viewInstanceMask = 0; uint generation = 0; friend class QRhiD3D12; }; @@ -889,7 +916,7 @@ struct QD3D12CommandBuffer : public QRhiCommandBuffer const QRhiNativeHandles *nativeHandles(); - ID3D12GraphicsCommandList *cmdList = nullptr; // not owned + ID3D12GraphicsCommandList1 *cmdList = nullptr; // not owned QRhiD3D12CommandBufferNativeHandles nativeHandlesStruct; enum PassType { @@ -921,9 +948,11 @@ struct QD3D12CommandBuffer : public QRhiCommandBuffer currentVertexOffsets = {}; } + // per-frame PassType recordingPass; QRhiRenderTarget *currentTarget; + // per-pass QD3D12GraphicsPipeline *currentGraphicsPipeline; QD3D12ComputePipeline *currentComputePipeline; uint currentPipelineGeneration; @@ -935,6 +964,38 @@ struct QD3D12CommandBuffer : public QRhiCommandBuffer DXGI_FORMAT currentIndexFormat; std::array<QD3D12ObjectHandle, D3D12_IA_VERTEX_INPUT_RESOURCE_SLOT_COUNT> currentVertexBuffers; std::array<quint32, D3D12_IA_VERTEX_INPUT_RESOURCE_SLOT_COUNT> currentVertexOffsets; + + // global + double lastGpuTime = 0; + + // per-setShaderResources + struct VisitorData { + QVarLengthArray<QPair<QD3D12ObjectHandle, quint32>, 4> cbufs[6]; + QVarLengthArray<QD3D12Descriptor, 8> srvs[6]; + QVarLengthArray<QD3D12Descriptor, 8> samplers[6]; + QVarLengthArray<QPair<QD3D12ObjectHandle, D3D12_UNORDERED_ACCESS_VIEW_DESC>, 4> uavs[6]; + } visitorData; + + void visitUniformBuffer(QD3D12Stage s, + const QRhiShaderResourceBinding::Data::UniformBufferData &d, + int shaderRegister, + int binding, + int dynamicOffsetCount, + const QRhiCommandBuffer::DynamicOffset *dynamicOffsets); + void visitTexture(QD3D12Stage s, + const QRhiShaderResourceBinding::TextureAndSampler &d, + int shaderRegister); + void visitSampler(QD3D12Stage s, + const QRhiShaderResourceBinding::TextureAndSampler &d, + int shaderRegister); + void visitStorageBuffer(QD3D12Stage s, + const QRhiShaderResourceBinding::Data::StorageBufferData &d, + QD3D12ShaderResourceVisitor::StorageOp op, + int shaderRegister); + void visitStorageImage(QD3D12Stage s, + const QRhiShaderResourceBinding::Data::StorageImageData &d, + QD3D12ShaderResourceVisitor::StorageOp op, + int shaderRegister); }; struct QD3D12SwapChain : public QRhiSwapChain @@ -945,6 +1006,7 @@ struct QD3D12SwapChain : public QRhiSwapChain QRhiCommandBuffer *currentFrameCommandBuffer() override; QRhiRenderTarget *currentFrameRenderTarget() override; + QRhiRenderTarget *currentFrameRenderTarget(StereoTargetBuffer targetBuffer) override; QSize surfacePixelSize() override; bool isFormatSupported(Format f) override; @@ -964,6 +1026,7 @@ struct QD3D12SwapChain : public QRhiSwapChain QSize pixelSize; UINT swapInterval = 1; UINT swapChainFlags = 0; + BOOL stereo = false; DXGI_FORMAT colorFormat; DXGI_FORMAT srgbAdjustedColorFormat; DXGI_COLOR_SPACE_TYPE hdrColorSpace; @@ -972,24 +1035,33 @@ struct QD3D12SwapChain : public QRhiSwapChain static const UINT BUFFER_COUNT = 3; QD3D12ObjectHandle colorBuffers[BUFFER_COUNT]; QD3D12Descriptor rtvs[BUFFER_COUNT]; + QD3D12Descriptor rtvsRight[BUFFER_COUNT]; DXGI_SAMPLE_DESC sampleDesc; QD3D12ObjectHandle msaaBuffers[BUFFER_COUNT]; QD3D12Descriptor msaaRtvs[BUFFER_COUNT]; QD3D12RenderBuffer *ds = nullptr; UINT currentBackBufferIndex = 0; QD3D12SwapChainRenderTarget rtWrapper; + QD3D12SwapChainRenderTarget rtWrapperRight; QD3D12CommandBuffer cbWrapper; struct FrameResources { ID3D12Fence *fence = nullptr; HANDLE fenceEvent = nullptr; UINT64 fenceCounter = 0; - ID3D12GraphicsCommandList *cmdList = nullptr; + ID3D12GraphicsCommandList1 *cmdList = nullptr; } frameRes[QD3D12_FRAMES_IN_FLIGHT]; int currentFrameSlot = 0; // index in frameRes }; +template<typename T, D3D12_PIPELINE_STATE_SUBOBJECT_TYPE Type> +struct alignas(void*) QD3D12PipelineStateSubObject +{ + D3D12_PIPELINE_STATE_SUBOBJECT_TYPE type = Type; + T object = {}; +}; + class QRhiD3D12 : public QRhiImplementation { public: @@ -1109,9 +1181,9 @@ public: void setPipelineCacheData(const QByteArray &data) override; void waitGpu(); - DXGI_SAMPLE_DESC effectiveSampleCount(int sampleCount, DXGI_FORMAT format) const; + DXGI_SAMPLE_DESC effectiveSampleDesc(int sampleCount, DXGI_FORMAT format) const; bool ensureDirectCompositionDevice(); - bool startCommandListForCurrentFrameSlot(ID3D12GraphicsCommandList **cmdList); + bool startCommandListForCurrentFrameSlot(ID3D12GraphicsCommandList1 **cmdList); void enqueueResourceUpdates(QD3D12CommandBuffer *cbD, QRhiResourceUpdateBatch *resourceUpdates); void finishActiveReadbacks(bool forced = false); bool ensureShaderVisibleDescriptorHeapCapacity(QD3D12ShaderVisibleDescriptorHeap *h, @@ -1122,7 +1194,7 @@ public: void bindShaderVisibleHeaps(QD3D12CommandBuffer *cbD); bool debugLayer = false; - ID3D12Device *dev = nullptr; + ID3D12Device2 *dev = nullptr; D3D_FEATURE_LEVEL minimumFeatureLevel = D3D_FEATURE_LEVEL(0); LUID adapterLuid = {}; bool importedDevice = false; @@ -1152,6 +1224,9 @@ public: QD3D12MipmapGenerator mipmapGen; QD3D12StagingArea smallStagingAreas[QD3D12_FRAMES_IN_FLIGHT]; QD3D12ShaderVisibleDescriptorHeap shaderVisibleCbvSrvUavHeap; + UINT64 timestampTicksPerSecond = 0; + QD3D12QueryHeap timestampQueryHeap; + QD3D12StagingArea timestampReadbackArea; IDCompositionDevice *dcompDevice = nullptr; QD3D12SwapChain *currentSwapChain = nullptr; QSet<QD3D12SwapChain *> swapchains; @@ -1160,35 +1235,14 @@ public: bool offscreenActive = false; QD3D12CommandBuffer *offscreenCb[QD3D12_FRAMES_IN_FLIGHT] = {}; - struct VisitorData { - QVarLengthArray<QPair<QD3D12ObjectHandle, quint32>, 4> cbufs[6]; - QVarLengthArray<QD3D12Descriptor, 8> srvs[6]; - QVarLengthArray<QD3D12Descriptor, 8> samplers[6]; - QVarLengthArray<QPair<QD3D12ObjectHandle, D3D12_UNORDERED_ACCESS_VIEW_DESC>, 4> uavs[6]; - } visitorData; - - void visitUniformBuffer(QD3D12Stage s, - const QRhiShaderResourceBinding::Data::UniformBufferData &d, - int shaderRegister, - int binding, - int dynamicOffsetCount, - const QRhiCommandBuffer::DynamicOffset *dynamicOffsets); - void visitTexture(QD3D12Stage s, - const QRhiShaderResourceBinding::TextureAndSampler &d, - int shaderRegister); - void visitSampler(QD3D12Stage s, - const QRhiShaderResourceBinding::TextureAndSampler &d, - int shaderRegister); - void visitStorageBuffer(QD3D12Stage s, - const QRhiShaderResourceBinding::Data::StorageBufferData &d, - QD3D12ShaderResourceVisitor::StorageOp op, - int shaderRegister); - void visitStorageImage(QD3D12Stage s, - const QRhiShaderResourceBinding::Data::StorageImageData &d, - QD3D12ShaderResourceVisitor::StorageOp op, - int shaderRegister); + struct { + bool multiView = false; + bool textureViewFormat = false; + } caps; }; QT_END_NAMESPACE +#endif // __ID3D12Device2_INTERFACE_DEFINED__ + #endif diff --git a/src/gui/rhi/qrhid3dhelpers.cpp b/src/gui/rhi/qrhid3dhelpers.cpp new file mode 100644 index 0000000000..216c358cbe --- /dev/null +++ b/src/gui/rhi/qrhid3dhelpers.cpp @@ -0,0 +1,172 @@ +// Copyright (C) 2023 The Qt Company Ltd. +// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only + +#include "qrhid3dhelpers_p.h" +#include <QtCore/private/qsystemlibrary_p.h> +#include <QtCore/private/qsystemerror_p.h> + +QT_BEGIN_NAMESPACE + +namespace QRhiD3D { + +bool output6ForWindow(QWindow *w, IDXGIAdapter1 *adapter, IDXGIOutput6 **result) +{ + bool ok = false; + QRect wr = w->geometry(); + wr = QRect(wr.topLeft() * w->devicePixelRatio(), wr.size() * w->devicePixelRatio()); + const QPoint center = wr.center(); + IDXGIOutput *currentOutput = nullptr; + IDXGIOutput *output = nullptr; + for (UINT i = 0; adapter->EnumOutputs(i, &output) != DXGI_ERROR_NOT_FOUND; ++i) { + DXGI_OUTPUT_DESC desc; + output->GetDesc(&desc); + const RECT r = desc.DesktopCoordinates; + const QRect dr(QPoint(r.left, r.top), QPoint(r.right - 1, r.bottom - 1)); + if (dr.contains(center)) { + currentOutput = output; + break; + } else { + output->Release(); + } + } + if (currentOutput) { + ok = SUCCEEDED(currentOutput->QueryInterface(__uuidof(IDXGIOutput6), reinterpret_cast<void **>(result))); + currentOutput->Release(); + } + return ok; +} + +bool outputDesc1ForWindow(QWindow *w, IDXGIAdapter1 *adapter, DXGI_OUTPUT_DESC1 *result) +{ + bool ok = false; + IDXGIOutput6 *out6 = nullptr; + if (output6ForWindow(w, adapter, &out6)) { + ok = SUCCEEDED(out6->GetDesc1(result)); + out6->Release(); + } + return ok; +} + +float sdrWhiteLevelInNits(const DXGI_OUTPUT_DESC1 &outputDesc) +{ + QVector<DISPLAYCONFIG_PATH_INFO> pathInfos; + uint32_t pathInfoCount, modeInfoCount; + LONG result; + do { + if (GetDisplayConfigBufferSizes(QDC_ONLY_ACTIVE_PATHS, &pathInfoCount, &modeInfoCount) == ERROR_SUCCESS) { + pathInfos.resize(pathInfoCount); + QVector<DISPLAYCONFIG_MODE_INFO> modeInfos(modeInfoCount); + result = QueryDisplayConfig(QDC_ONLY_ACTIVE_PATHS, &pathInfoCount, pathInfos.data(), &modeInfoCount, modeInfos.data(), nullptr); + } else { + return 200.0f; + } + } while (result == ERROR_INSUFFICIENT_BUFFER); + + MONITORINFOEX monitorInfo = {}; + monitorInfo.cbSize = sizeof(monitorInfo); + GetMonitorInfo(outputDesc.Monitor, &monitorInfo); + + for (const DISPLAYCONFIG_PATH_INFO &info : pathInfos) { + DISPLAYCONFIG_SOURCE_DEVICE_NAME deviceName = {}; + deviceName.header.type = DISPLAYCONFIG_DEVICE_INFO_GET_SOURCE_NAME; + deviceName.header.size = sizeof(deviceName); + deviceName.header.adapterId = info.sourceInfo.adapterId; + deviceName.header.id = info.sourceInfo.id; + if (DisplayConfigGetDeviceInfo(&deviceName.header) == ERROR_SUCCESS) { + if (!wcscmp(monitorInfo.szDevice, deviceName.viewGdiDeviceName)) { + DISPLAYCONFIG_SDR_WHITE_LEVEL whiteLevel = {}; + whiteLevel.header.type = DISPLAYCONFIG_DEVICE_INFO_GET_SDR_WHITE_LEVEL; + whiteLevel.header.size = sizeof(DISPLAYCONFIG_SDR_WHITE_LEVEL); + whiteLevel.header.adapterId = info.targetInfo.adapterId; + whiteLevel.header.id = info.targetInfo.id; + if (DisplayConfigGetDeviceInfo(&whiteLevel.header) == ERROR_SUCCESS) + return whiteLevel.SDRWhiteLevel * 80 / 1000.0f; + } + } + } + + return 200.0f; +} + +pD3DCompile resolveD3DCompile() +{ + for (const wchar_t *libraryName : {L"D3DCompiler_47", L"D3DCompiler_43"}) { + QSystemLibrary library(libraryName); + if (library.load()) { + if (auto symbol = library.resolve("D3DCompile")) + return reinterpret_cast<pD3DCompile>(symbol); + } else { + qWarning("Failed to load D3DCompiler_47/43.dll"); + } + } + return nullptr; +} + +IDCompositionDevice *createDirectCompositionDevice() +{ + QSystemLibrary dcomplib(QStringLiteral("dcomp")); + typedef HRESULT (__stdcall *DCompositionCreateDeviceFuncPtr)( + _In_opt_ IDXGIDevice *dxgiDevice, + _In_ REFIID iid, + _Outptr_ void **dcompositionDevice); + DCompositionCreateDeviceFuncPtr func = reinterpret_cast<DCompositionCreateDeviceFuncPtr>( + dcomplib.resolve("DCompositionCreateDevice")); + if (!func) { + qWarning("Unable to resolve DCompositionCreateDevice, perhaps dcomp.dll is missing?"); + return nullptr; + } + IDCompositionDevice *device = nullptr; + HRESULT hr = func(nullptr, __uuidof(IDCompositionDevice), reinterpret_cast<void **>(&device)); + if (FAILED(hr)) { + qWarning("Failed to create Direct Composition device: %s", + qPrintable(QSystemError::windowsComString(hr))); + return nullptr; + } + return device; +} + +#ifdef QRHI_D3D12_HAS_DXC +std::pair<IDxcCompiler *, IDxcLibrary *> createDxcCompiler() +{ + QSystemLibrary dxclib(QStringLiteral("dxcompiler")); + // this will not be in the system library location, hence onlySystemDirectory==false + if (!dxclib.load(false)) { + qWarning("Failed to load dxcompiler.dll"); + return {}; + } + DxcCreateInstanceProc func = reinterpret_cast<DxcCreateInstanceProc>(dxclib.resolve("DxcCreateInstance")); + if (!func) { + qWarning("Unable to resolve DxcCreateInstance"); + return {}; + } + IDxcCompiler *compiler = nullptr; + HRESULT hr = func(CLSID_DxcCompiler, __uuidof(IDxcCompiler), reinterpret_cast<void**>(&compiler)); + if (FAILED(hr)) { + qWarning("Failed to create dxc compiler instance: %s", + qPrintable(QSystemError::windowsComString(hr))); + return {}; + } + IDxcLibrary *library = nullptr; + hr = func(CLSID_DxcLibrary, __uuidof(IDxcLibrary), reinterpret_cast<void**>(&library)); + if (FAILED(hr)) { + qWarning("Failed to create dxc library instance: %s", + qPrintable(QSystemError::windowsComString(hr))); + return {}; + } + return { compiler, library }; +} +#endif + +void fillDriverInfo(QRhiDriverInfo *info, const DXGI_ADAPTER_DESC1 &desc) +{ + const QString name = QString::fromUtf16(reinterpret_cast<const char16_t *>(desc.Description)); + info->deviceName = name.toUtf8(); + info->deviceId = desc.DeviceId; + info->vendorId = desc.VendorId; + info->deviceType = (desc.Flags & DXGI_ADAPTER_FLAG_SOFTWARE) ? QRhiDriverInfo::CpuDevice + : QRhiDriverInfo::UnknownDevice; +} + +} // namespace + +QT_END_NAMESPACE diff --git a/src/gui/rhi/qrhid3dhelpers_p.h b/src/gui/rhi/qrhid3dhelpers_p.h new file mode 100644 index 0000000000..f31cdc8d11 --- /dev/null +++ b/src/gui/rhi/qrhid3dhelpers_p.h @@ -0,0 +1,53 @@ +// Copyright (C) 2023 The Qt Company Ltd. +// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only + +#ifndef QRHID3DHELPERS_P_H +#define QRHID3DHELPERS_P_H + +// +// W A R N I N G +// ------------- +// +// This file is not part of the Qt API. It exists purely as an +// implementation detail. This header file may change from version to +// version without notice, or even be removed. +// +// We mean it. +// + +#include <rhi/qrhi.h> + +#include <QtGui/qwindow.h> + +#include <dxgi1_6.h> +#include <dcomp.h> +#include <d3dcompiler.h> + +#if __has_include(<dxcapi.h>) +#include <dxcapi.h> +#define QRHI_D3D12_HAS_DXC +#endif + +QT_BEGIN_NAMESPACE + +namespace QRhiD3D { + +bool output6ForWindow(QWindow *w, IDXGIAdapter1 *adapter, IDXGIOutput6 **result); +bool outputDesc1ForWindow(QWindow *w, IDXGIAdapter1 *adapter, DXGI_OUTPUT_DESC1 *result); +float sdrWhiteLevelInNits(const DXGI_OUTPUT_DESC1 &outputDesc); + +pD3DCompile resolveD3DCompile(); + +IDCompositionDevice *createDirectCompositionDevice(); + +#ifdef QRHI_D3D12_HAS_DXC +std::pair<IDxcCompiler *, IDxcLibrary *> createDxcCompiler(); +#endif + +void fillDriverInfo(QRhiDriverInfo *info, const DXGI_ADAPTER_DESC1 &desc); + +} // namespace + +QT_END_NAMESPACE + +#endif diff --git a/src/gui/rhi/qrhigles2.cpp b/src/gui/rhi/qrhigles2.cpp index 774dec310f..7e886a5d00 100644 --- a/src/gui/rhi/qrhigles2.cpp +++ b/src/gui/rhi/qrhigles2.cpp @@ -7,6 +7,7 @@ #include <QtCore/qmap.h> #include <QtGui/private/qopenglextensions_p.h> #include <QtGui/private/qopenglprogrambinarycache_p.h> +#include <QtGui/private/qwindow_p.h> #include <qpa/qplatformopenglcontext.h> #include <qmath.h> @@ -31,7 +32,7 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief OpenGL specific initialization parameters. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. An OpenGL-based QRhi needs an already created QSurface that can be used in @@ -143,7 +144,7 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief Holds the OpenGL context used by the QRhi. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -215,6 +216,10 @@ QT_BEGIN_NAMESPACE #define GL_DEPTH_COMPONENT32F 0x8CAC #endif +#ifndef GL_UNSIGNED_INT_24_8 +#define GL_UNSIGNED_INT_24_8 0x84FA +#endif + #ifndef GL_STENCIL_INDEX #define GL_STENCIL_INDEX 0x1901 #endif @@ -240,7 +245,7 @@ QT_BEGIN_NAMESPACE #endif #ifndef GL_FRAMEBUFFER_SRGB -#define GL_FRAMEBUFFER_SRGB 0x8DB9 +#define GL_FRAMEBUFFER_SRGB 0x8DB9 #endif #ifndef GL_READ_FRAMEBUFFER @@ -355,6 +360,10 @@ QT_BEGIN_NAMESPACE #define GL_TEXTURE_2D_MULTISAMPLE 0x9100 #endif +#ifndef GL_TEXTURE_2D_MULTISAMPLE_ARRAY +#define GL_TEXTURE_2D_MULTISAMPLE_ARRAY 0x9102 +#endif + #ifndef GL_TEXTURE_EXTERNAL_OES #define GL_TEXTURE_EXTERNAL_OES 0x8D65 #endif @@ -484,21 +493,41 @@ QT_BEGIN_NAMESPACE #endif #ifndef GL_TEXTURE_1D -# define GL_TEXTURE_1D 0x0DE0 +# define GL_TEXTURE_1D 0x0DE0 #endif #ifndef GL_TEXTURE_1D_ARRAY -# define GL_TEXTURE_1D_ARRAY 0x8C18 +# define GL_TEXTURE_1D_ARRAY 0x8C18 #endif #ifndef GL_HALF_FLOAT -#define GL_HALF_FLOAT 0x140B +#define GL_HALF_FLOAT 0x140B #endif #ifndef GL_MAX_VERTEX_OUTPUT_COMPONENTS #define GL_MAX_VERTEX_OUTPUT_COMPONENTS 0x9122 #endif +#ifndef GL_TIMESTAMP +#define GL_TIMESTAMP 0x8E28 +#endif + +#ifndef GL_QUERY_RESULT +#define GL_QUERY_RESULT 0x8866 +#endif + +#ifndef GL_QUERY_RESULT_AVAILABLE +#define GL_QUERY_RESULT_AVAILABLE 0x8867 +#endif + +#ifndef GL_BUFFER +#define GL_BUFFER 0x82E0 +#endif + +#ifndef GL_PROGRAM +#define GL_PROGRAM 0x82E2 +#endif + /*! Constructs a new QRhiGles2InitParams. @@ -825,8 +854,8 @@ bool QRhiGles2::create(QRhi::Flags flags) caps.maxDrawBuffers = 1; caps.hasDrawBuffersFunc = false; // This does not mean MSAA is not supported, just that we cannot query - // the supported sample counts. - caps.maxSamples = 1; + // the supported sample counts. Assume that 4x is always supported. + caps.maxSamples = 4; } caps.msaaRenderBuffer = f->hasOpenGLExtension(QOpenGLExtensions::FramebufferMultisample) @@ -863,7 +892,13 @@ bool QRhiGles2::create(QRhi::Flags flags) #else caps.needsDepthStencilCombinedAttach = false; #endif - caps.srgbCapableDefaultFramebuffer = f->hasOpenGLExtension(QOpenGLExtensions::SRGBFrameBuffer); + + // QOpenGLExtensions::SRGBFrameBuffer is not useful here. We need to know if + // controlling the sRGB-on-shader-write state is supported, not that if the + // default framebuffer is sRGB-capable. And there are two different + // extensions for desktop and ES. + caps.srgbWriteControl = ctx->hasExtension("GL_EXT_framebuffer_sRGB") || ctx->hasExtension("GL_EXT_sRGB_write_control"); + caps.coreProfile = actualFormat.profile() == QSurfaceFormat::CoreProfile; if (caps.gles) @@ -1009,6 +1044,60 @@ bool QRhiGles2::create(QRhi::Flags flags) caps.halfAttributes = f->hasOpenGLExtension(QOpenGLExtensions::HalfFloatVertex); + // We always require GL_OVR_multiview2 for symmetry with other backends. + caps.multiView = f->hasOpenGLExtension(QOpenGLExtensions::MultiView) + && f->hasOpenGLExtension(QOpenGLExtensions::MultiViewExtended); + if (caps.multiView) { + glFramebufferTextureMultiviewOVR = + reinterpret_cast<void(QOPENGLF_APIENTRYP)(GLenum, GLenum, GLuint, GLint, GLint, GLsizei)>( + ctx->getProcAddress(QByteArrayLiteral("glFramebufferTextureMultiviewOVR"))); + } + + // Only do timestamp queries on OpenGL 3.3+. + caps.timestamps = !caps.gles && (caps.ctxMajor > 3 || (caps.ctxMajor == 3 && caps.ctxMinor >= 3)); + if (caps.timestamps) { + glQueryCounter = reinterpret_cast<void(QOPENGLF_APIENTRYP)(GLuint, GLenum)>( + ctx->getProcAddress(QByteArrayLiteral("glQueryCounter"))); + glGetQueryObjectui64v = reinterpret_cast<void(QOPENGLF_APIENTRYP)(GLuint, GLenum, quint64 *)>( + ctx->getProcAddress(QByteArrayLiteral("glGetQueryObjectui64v"))); + if (!glQueryCounter || !glGetQueryObjectui64v) + caps.timestamps = false; + } + + // glObjectLabel is available on OpenGL ES 3.2+ and OpenGL 4.3+ + if (caps.gles) + caps.objectLabel = caps.ctxMajor > 3 || (caps.ctxMajor == 3 && caps.ctxMinor >= 2); + else + caps.objectLabel = caps.ctxMajor > 4 || (caps.ctxMajor == 4 && caps.ctxMinor >= 3); + if (caps.objectLabel) { + glObjectLabel = reinterpret_cast<void(QOPENGLF_APIENTRYP)(GLenum, GLuint, GLsizei, const GLchar *)>( + ctx->getProcAddress(QByteArrayLiteral("glObjectLabel"))); + } + + if (caps.gles) { + // This is the third way to get multisample rendering with GLES. (1. is + // multisample render buffer -> resolve to texture; 2. is multisample + // texture with GLES 3.1; 3. is this, avoiding the explicit multisample + // buffer and should be more efficient with tiled architectures. + // Interesting also because 2. does not seem to work in practice on + // devices such as the Quest 3) + caps.glesMultisampleRenderToTexture = ctx->hasExtension("GL_EXT_multisampled_render_to_texture"); + if (caps.glesMultisampleRenderToTexture) { + glFramebufferTexture2DMultisampleEXT = reinterpret_cast<void(QOPENGLF_APIENTRYP)(GLenum, GLenum, GLenum, GLuint, GLint, GLsizei)>( + ctx->getProcAddress(QByteArrayLiteral("glFramebufferTexture2DMultisampleEXT"))); + } + caps.glesMultiviewMultisampleRenderToTexture = ctx->hasExtension("GL_OVR_multiview_multisampled_render_to_texture"); + if (caps.glesMultiviewMultisampleRenderToTexture) { + glFramebufferTextureMultisampleMultiviewOVR = reinterpret_cast<void(QOPENGLF_APIENTRYP)(GLenum, GLenum, GLuint, GLint, GLsizei, GLint, GLsizei)>( + ctx->getProcAddress(QByteArrayLiteral("glFramebufferTextureMultisampleMultiviewOVR"))); + } + } else { + caps.glesMultisampleRenderToTexture = false; + caps.glesMultiviewMultisampleRenderToTexture = false; + } + + caps.unpackRowLength = !caps.gles || caps.ctxMajor >= 3; + nativeHandlesStruct.context = ctx; contextLost = false; @@ -1024,6 +1113,11 @@ void QRhiGles2::destroy() ensureContext(); executeDeferredReleases(); + if (ofr.tsQueries[0]) { + f->glDeleteQueries(2, ofr.tsQueries); + ofr.tsQueries[0] = ofr.tsQueries[1] = 0; + } + if (vao) { f->glDeleteVertexArrays(1, &vao); vao = 0; @@ -1061,6 +1155,7 @@ void QRhiGles2::executeDeferredReleases() break; case QRhiGles2::DeferredReleaseEntry::TextureRenderTarget: f->glDeleteFramebuffers(1, &e.textureRenderTarget.framebuffer); + f->glDeleteTextures(1, &e.textureRenderTarget.nonMsaaThrowawayDepthTexture); break; default: Q_UNREACHABLE(); @@ -1080,17 +1175,6 @@ QList<int> QRhiGles2::supportedSampleCounts() const return supportedSampleCountList; } -int QRhiGles2::effectiveSampleCount(int sampleCount) const -{ - // Stay compatible with QSurfaceFormat and friends where samples == 0 means the same as 1. - const int s = qBound(1, sampleCount, 64); - if (!supportedSampleCounts().contains(s)) { - qWarning("Attempted to set unsupported sample count %d", sampleCount); - return 1; - } - return s; -} - QRhiSwapChain *QRhiGles2::createSwapChain() { return new QGles2SwapChain(this); @@ -1215,13 +1299,13 @@ static inline void toGlTextureFormat(QRhiTexture::Format format, const QRhiGles2 *glintformat = GL_DEPTH_COMPONENT24; *glsizedintformat = *glintformat; *glformat = GL_DEPTH_COMPONENT; - *gltype = GL_UNSIGNED_SHORT; + *gltype = GL_UNSIGNED_INT; break; case QRhiTexture::D24S8: *glintformat = GL_DEPTH24_STENCIL8; *glsizedintformat = *glintformat; *glformat = GL_DEPTH_STENCIL; - *gltype = GL_UNSIGNED_SHORT; + *gltype = GL_UNSIGNED_INT_24_8; break; case QRhiTexture::D32F: *glintformat = GL_DEPTH_COMPONENT32F; @@ -1298,7 +1382,7 @@ bool QRhiGles2::isFeatureSupported(QRhi::Feature feature) const case QRhi::DebugMarkers: return false; case QRhi::Timestamps: - return false; + return caps.timestamps; case QRhi::Instancing: return caps.instancing; case QRhi::CustomInstanceStepRate: @@ -1344,7 +1428,7 @@ bool QRhiGles2::isFeatureSupported(QRhi::Feature feature) const case QRhi::PipelineCacheDataLoadSave: return caps.programBinary; case QRhi::ImageDataStride: - return !caps.gles || caps.ctxMajor >= 3; + return caps.unpackRowLength; case QRhi::RenderBufferImport: return true; case QRhi::ThreeDimensionalTextures: @@ -1371,6 +1455,12 @@ bool QRhiGles2::isFeatureSupported(QRhi::Feature feature) const return caps.texture1D; case QRhi::ThreeDimensionalTextureMipmaps: return caps.texture3D; + case QRhi::MultiView: + return caps.multiView && caps.maxTextureArraySize > 0; + case QRhi::TextureViewFormat: + return false; + case QRhi::ResolveDepthStencil: + return true; default: Q_UNREACHABLE_RETURN(false); } @@ -1945,10 +2035,14 @@ const QRhiNativeHandles *QRhiGles2::nativeHandles(QRhiCommandBuffer *cb) return nullptr; } -static inline void addBoundaryCommand(QGles2CommandBuffer *cbD, QGles2CommandBuffer::Command::Cmd type) +static inline void addBoundaryCommand(QGles2CommandBuffer *cbD, QGles2CommandBuffer::Command::Cmd type, GLuint tsQuery = 0) { QGles2CommandBuffer::Command &cmd(cbD->commands.get()); cmd.cmd = type; + if (type == QGles2CommandBuffer::Command::BeginFrame) + cmd.args.beginFrame.timestampQuery = tsQuery; + else if (type == QGles2CommandBuffer::Command::EndFrame) + cmd.args.endFrame.timestampQuery = tsQuery; } void QRhiGles2::beginExternal(QRhiCommandBuffer *cb) @@ -2010,8 +2104,8 @@ void QRhiGles2::endExternal(QRhiCommandBuffer *cb) double QRhiGles2::lastCompletedGpuTime(QRhiCommandBuffer *cb) { - Q_UNUSED(cb); - return 0; + QGles2CommandBuffer *cbD = QRHI_RES(QGles2CommandBuffer, cb); + return cbD->lastGpuTime; } QRhi::FrameOpResult QRhiGles2::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginFrameFlags) @@ -2027,7 +2121,17 @@ QRhi::FrameOpResult QRhiGles2::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginF executeDeferredReleases(); swapChainD->cb.resetState(); - addBoundaryCommand(&swapChainD->cb, QGles2CommandBuffer::Command::BeginFrame); + if (swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex]) { + double elapsedSec = 0; + if (swapChainD->timestamps.tryQueryTimestamps(swapChainD->currentTimestampPairIndex, this, &elapsedSec)) + swapChainD->cb.lastGpuTime = elapsedSec; + } + + GLuint tsStart = swapChainD->timestamps.query[swapChainD->currentTimestampPairIndex * 2]; + GLuint tsEnd = swapChainD->timestamps.query[swapChainD->currentTimestampPairIndex * 2 + 1]; + const bool recordTimestamps = tsStart && tsEnd && !swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex]; + + addBoundaryCommand(&swapChainD->cb, QGles2CommandBuffer::Command::BeginFrame, recordTimestamps ? tsStart : 0); return QRhi::FrameOpSuccess; } @@ -2037,7 +2141,15 @@ QRhi::FrameOpResult QRhiGles2::endFrame(QRhiSwapChain *swapChain, QRhi::EndFrame QGles2SwapChain *swapChainD = QRHI_RES(QGles2SwapChain, swapChain); Q_ASSERT(currentSwapChain == swapChainD); - addBoundaryCommand(&swapChainD->cb, QGles2CommandBuffer::Command::EndFrame); + GLuint tsStart = swapChainD->timestamps.query[swapChainD->currentTimestampPairIndex * 2]; + GLuint tsEnd = swapChainD->timestamps.query[swapChainD->currentTimestampPairIndex * 2 + 1]; + const bool recordTimestamps = tsStart && tsEnd && !swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex]; + if (recordTimestamps) { + swapChainD->timestamps.active[swapChainD->currentTimestampPairIndex] = true; + swapChainD->currentTimestampPairIndex = (swapChainD->currentTimestampPairIndex + 1) % QGles2SwapChainTimestamps::TIMESTAMP_PAIRS; + } + + addBoundaryCommand(&swapChainD->cb, QGles2CommandBuffer::Command::EndFrame, recordTimestamps ? tsEnd : 0); if (!ensureContext(swapChainD->surface)) return contextLost ? QRhi::FrameOpDeviceLost : QRhi::FrameOpError; @@ -2069,7 +2181,12 @@ QRhi::FrameOpResult QRhiGles2::beginOffscreenFrame(QRhiCommandBuffer **cb, QRhi: executeDeferredReleases(); ofr.cbWrapper.resetState(); - addBoundaryCommand(&ofr.cbWrapper, QGles2CommandBuffer::Command::BeginFrame); + if (rhiFlags.testFlag(QRhi::EnableTimestamps) && caps.timestamps) { + if (!ofr.tsQueries[0]) + f->glGenQueries(2, ofr.tsQueries); + } + + addBoundaryCommand(&ofr.cbWrapper, QGles2CommandBuffer::Command::BeginFrame, ofr.tsQueries[0]); *cb = &ofr.cbWrapper; return QRhi::FrameOpSuccess; @@ -2081,7 +2198,7 @@ QRhi::FrameOpResult QRhiGles2::endOffscreenFrame(QRhi::EndFrameFlags flags) Q_ASSERT(ofr.active); ofr.active = false; - addBoundaryCommand(&ofr.cbWrapper, QGles2CommandBuffer::Command::EndFrame); + addBoundaryCommand(&ofr.cbWrapper, QGles2CommandBuffer::Command::EndFrame, ofr.tsQueries[1]); if (!ensureContext()) return contextLost ? QRhi::FrameOpDeviceLost : QRhi::FrameOpError; @@ -2094,6 +2211,16 @@ QRhi::FrameOpResult QRhiGles2::endOffscreenFrame(QRhi::EndFrameFlags flags) // another, sharing context. f->glFlush(); + if (ofr.tsQueries[0]) { + quint64 timestamps[2]; + glGetQueryObjectui64v(ofr.tsQueries[1], GL_QUERY_RESULT, ×tamps[1]); + glGetQueryObjectui64v(ofr.tsQueries[0], GL_QUERY_RESULT, ×tamps[0]); + if (timestamps[1] >= timestamps[0]) { + const quint64 nanoseconds = timestamps[1] - timestamps[0]; + ofr.cbWrapper.lastGpuTime = nanoseconds / 1000000000.0; // seconds + } + } + return QRhi::FrameOpSuccess; } @@ -2213,17 +2340,15 @@ void QRhiGles2::enqueueSubresUpload(QGles2Texture *texD, QGles2CommandBuffer *cb const GLenum effectiveTarget = faceTargetBase + (isCubeMap ? uint(layer) : 0u); const QPoint dp = subresDesc.destinationTopLeft(); const QByteArray rawData = subresDesc.data(); - if (!subresDesc.image().isNull()) { - QImage img = subresDesc.image(); - QSize size = img.size(); + + auto setCmdByNotCompressedData = [&](const void* data, QSize size, quint32 dataStride) + { + quint32 bytesPerLine = 0; + quint32 bytesPerPixel = 0; + textureFormatInfo(texD->m_format, size, &bytesPerLine, nullptr, &bytesPerPixel); + QGles2CommandBuffer::Command &cmd(cbD->commands.get()); cmd.cmd = QGles2CommandBuffer::Command::SubImage; - if (!subresDesc.sourceSize().isEmpty() || !subresDesc.sourceTopLeft().isNull()) { - const QPoint sp = subresDesc.sourceTopLeft(); - if (!subresDesc.sourceSize().isEmpty()) - size = subresDesc.sourceSize(); - img = img.copy(sp.x(), sp.y(), size.width(), size.height()); - } cmd.args.subImage.target = texD->target; cmd.args.subImage.texture = texD->texture; cmd.args.subImage.faceTarget = effectiveTarget; @@ -2235,9 +2360,35 @@ void QRhiGles2::enqueueSubresUpload(QGles2Texture *texD, QGles2CommandBuffer *cb cmd.args.subImage.h = size.height(); cmd.args.subImage.glformat = texD->glformat; cmd.args.subImage.gltype = texD->gltype; - cmd.args.subImage.rowStartAlign = 4; - cmd.args.subImage.rowLength = 0; - cmd.args.subImage.data = cbD->retainImage(img); + + if (dataStride == 0) + dataStride = bytesPerLine; + + cmd.args.subImage.rowStartAlign = (dataStride & 3) ? 1 : 4; + cmd.args.subImage.rowLength = caps.unpackRowLength ? (bytesPerPixel ? dataStride / bytesPerPixel : 0) : 0; + + cmd.args.subImage.data = data; + }; + + if (!subresDesc.image().isNull()) { + QImage img = subresDesc.image(); + QSize size = img.size(); + if (!subresDesc.sourceSize().isEmpty() || !subresDesc.sourceTopLeft().isNull()) { + const QPoint sp = subresDesc.sourceTopLeft(); + if (!subresDesc.sourceSize().isEmpty()) + size = subresDesc.sourceSize(); + + if (caps.unpackRowLength) { + cbD->retainImage(img); + // create a non-owning wrapper for the subimage + const uchar *data = img.constBits() + sp.y() * img.bytesPerLine() + sp.x() * (qMax(1, img.depth() / 8)); + img = QImage(data, size.width(), size.height(), img.bytesPerLine(), img.format()); + } else { + img = img.copy(sp.x(), sp.y(), size.width(), size.height()); + } + } + + setCmdByNotCompressedData(cbD->retainImage(img), size, img.bytesPerLine()); } else if (!rawData.isEmpty() && isCompressed) { const int depth = qMax(1, texD->m_depth); const int arraySize = qMax(0, texD->m_arraySize); @@ -2305,31 +2456,8 @@ void QRhiGles2::enqueueSubresUpload(QGles2Texture *texD, QGles2CommandBuffer *cb } else if (!rawData.isEmpty()) { const QSize size = subresDesc.sourceSize().isEmpty() ? q->sizeForMipLevel(level, texD->m_pixelSize) : subresDesc.sourceSize(); - quint32 bytesPerLine = 0; - quint32 bytesPerPixel = 0; - textureFormatInfo(texD->m_format, size, &bytesPerLine, nullptr, &bytesPerPixel); - QGles2CommandBuffer::Command &cmd(cbD->commands.get()); - cmd.cmd = QGles2CommandBuffer::Command::SubImage; - cmd.args.subImage.target = texD->target; - cmd.args.subImage.texture = texD->texture; - cmd.args.subImage.faceTarget = effectiveTarget; - cmd.args.subImage.level = level; - cmd.args.subImage.dx = dp.x(); - cmd.args.subImage.dy = is1D && isArray ? layer : dp.y(); - cmd.args.subImage.dz = is3D || isArray ? layer : 0; - cmd.args.subImage.w = size.width(); - cmd.args.subImage.h = size.height(); - cmd.args.subImage.glformat = texD->glformat; - cmd.args.subImage.gltype = texD->gltype; - // Default unpack alignment (row start alignment - // requirement) is 4. QImage guarantees 4 byte aligned - // row starts, but our raw data here does not. - cmd.args.subImage.rowStartAlign = (bytesPerLine & 3) ? 1 : 4; - if (subresDesc.dataStride() && bytesPerPixel) - cmd.args.subImage.rowLength = subresDesc.dataStride() / bytesPerPixel; - else - cmd.args.subImage.rowLength = 0; - cmd.args.subImage.data = cbD->retainData(rawData); + + setCmdByNotCompressedData(cbD->retainData(rawData), size, subresDesc.dataStride()); } else { qWarning("Invalid texture upload for %p layer=%d mip=%d", texD, layer, level); } @@ -2848,6 +2976,8 @@ void QRhiGles2::executeCommandBuffer(QRhiCommandBuffer *cb) const QGles2CommandBuffer::Command &cmd(*it); switch (cmd.cmd) { case QGles2CommandBuffer::Command::BeginFrame: + if (cmd.args.beginFrame.timestampQuery) + glQueryCounter(cmd.args.beginFrame.timestampQuery, GL_TIMESTAMP); if (caps.coreProfile) { if (!vao) f->glGenVertexArrays(1, &vao); @@ -2874,6 +3004,8 @@ void QRhiGles2::executeCommandBuffer(QRhiCommandBuffer *cb) #endif if (vao) f->glBindVertexArray(0); + if (cmd.args.endFrame.timestampQuery) + glQueryCounter(cmd.args.endFrame.timestampQuery, GL_TIMESTAMP); break; case QGles2CommandBuffer::Command::ResetFrame: if (vao) @@ -3018,6 +3150,38 @@ void QRhiGles2::executeCommandBuffer(QRhiCommandBuffer *cb) type = GL_HALF_FLOAT; size = 1; break; + case QRhiVertexInputAttribute::UShort4: + type = GL_UNSIGNED_SHORT; + size = 4; + break; + case QRhiVertexInputAttribute::UShort3: + type = GL_UNSIGNED_SHORT; + size = 3; + break; + case QRhiVertexInputAttribute::UShort2: + type = GL_UNSIGNED_SHORT; + size = 2; + break; + case QRhiVertexInputAttribute::UShort: + type = GL_UNSIGNED_SHORT; + size = 1; + break; + case QRhiVertexInputAttribute::SShort4: + type = GL_SHORT; + size = 4; + break; + case QRhiVertexInputAttribute::SShort3: + type = GL_SHORT; + size = 3; + break; + case QRhiVertexInputAttribute::SShort2: + type = GL_SHORT; + size = 2; + break; + case QRhiVertexInputAttribute::SShort: + type = GL_SHORT; + size = 1; + break; default: break; } @@ -3160,7 +3324,7 @@ void QRhiGles2::executeCommandBuffer(QRhiCommandBuffer *cb) } if (caps.hasDrawBuffersFunc) f->glDrawBuffers(bufs.count(), bufs.constData()); - if (caps.srgbCapableDefaultFramebuffer) { + if (caps.srgbWriteControl) { if (cmd.args.bindFramebuffer.srgb) f->glEnable(GL_FRAMEBUFFER_SRGB); else @@ -3315,6 +3479,14 @@ void QRhiGles2::executeCommandBuffer(QRhiCommandBuffer *cb) result->data.resize(w * h * 8); f->glReadPixels(0, 0, w, h, GL_RGBA, GL_HALF_FLOAT, result->data.data()); break; + case QRhiTexture::R16F: + result->data.resize(w * h * 2); + f->glReadPixels(0, 0, w, h, GL_RED, GL_HALF_FLOAT, result->data.data()); + break; + case QRhiTexture::R32F: + result->data.resize(w * h * 4); + f->glReadPixels(0, 0, w, h, GL_RED, GL_FLOAT, result->data.data()); + break; case QRhiTexture::RGBA32F: result->data.resize(w * h * 16); f->glReadPixels(0, 0, w, h, GL_RGBA, GL_FLOAT, result->data.data()); @@ -3420,19 +3592,120 @@ void QRhiGles2::executeCommandBuffer(QRhiCommandBuffer *cb) GLuint fbo[2]; f->glGenFramebuffers(2, fbo); f->glBindFramebuffer(GL_READ_FRAMEBUFFER, fbo[0]); - f->glFramebufferRenderbuffer(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, - GL_RENDERBUFFER, cmd.args.blitFromRb.renderbuffer); + const bool ds = cmd.args.blitFromRenderbuffer.isDepthStencil; + if (ds) { + f->glFramebufferRenderbuffer(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, + GL_RENDERBUFFER, cmd.args.blitFromRenderbuffer.renderbuffer); + f->glFramebufferRenderbuffer(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, + GL_RENDERBUFFER, cmd.args.blitFromRenderbuffer.renderbuffer); + } else { + f->glFramebufferRenderbuffer(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, + GL_RENDERBUFFER, cmd.args.blitFromRenderbuffer.renderbuffer); + } + f->glBindFramebuffer(GL_DRAW_FRAMEBUFFER, fbo[1]); + if (cmd.args.blitFromRenderbuffer.target == GL_TEXTURE_3D || cmd.args.blitFromRenderbuffer.target == GL_TEXTURE_2D_ARRAY) { + if (ds) { + f->glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, + cmd.args.blitFromRenderbuffer.dstTexture, + cmd.args.blitFromRenderbuffer.dstLevel, + cmd.args.blitFromRenderbuffer.dstLayer); + f->glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, + cmd.args.blitFromRenderbuffer.dstTexture, + cmd.args.blitFromRenderbuffer.dstLevel, + cmd.args.blitFromRenderbuffer.dstLayer); + } else { + f->glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, + cmd.args.blitFromRenderbuffer.dstTexture, + cmd.args.blitFromRenderbuffer.dstLevel, + cmd.args.blitFromRenderbuffer.dstLayer); + } + } else { + if (ds) { + f->glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, cmd.args.blitFromRenderbuffer.target, + cmd.args.blitFromRenderbuffer.dstTexture, cmd.args.blitFromRenderbuffer.dstLevel); + f->glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, cmd.args.blitFromRenderbuffer.target, + cmd.args.blitFromRenderbuffer.dstTexture, cmd.args.blitFromRenderbuffer.dstLevel); + } else { + f->glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, cmd.args.blitFromRenderbuffer.target, + cmd.args.blitFromRenderbuffer.dstTexture, cmd.args.blitFromRenderbuffer.dstLevel); + } + } + f->glBlitFramebuffer(0, 0, cmd.args.blitFromRenderbuffer.w, cmd.args.blitFromRenderbuffer.h, + 0, 0, cmd.args.blitFromRenderbuffer.w, cmd.args.blitFromRenderbuffer.h, + ds ? GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT : GL_COLOR_BUFFER_BIT, + GL_NEAREST); // Qt 5 used Nearest when resolving samples, stick to that + f->glBindFramebuffer(GL_FRAMEBUFFER, ctx->defaultFramebufferObject()); + f->glDeleteFramebuffers(2, fbo); + } + break; + case QGles2CommandBuffer::Command::BlitFromTexture: + { + // Altering the scissor state, so reset the stored state, although + // not strictly required as long as blit is done in endPass() only. + cbD->graphicsPassState.reset(); + f->glDisable(GL_SCISSOR_TEST); + GLuint fbo[2]; + f->glGenFramebuffers(2, fbo); + f->glBindFramebuffer(GL_READ_FRAMEBUFFER, fbo[0]); + const bool ds = cmd.args.blitFromTexture.isDepthStencil; + if (cmd.args.blitFromTexture.srcTarget == GL_TEXTURE_2D_MULTISAMPLE_ARRAY) { + if (ds) { + f->glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, + cmd.args.blitFromTexture.srcTexture, + cmd.args.blitFromTexture.srcLevel, + cmd.args.blitFromTexture.srcLayer); + f->glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, + cmd.args.blitFromTexture.srcTexture, + cmd.args.blitFromTexture.srcLevel, + cmd.args.blitFromTexture.srcLayer); + } else { + f->glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, + cmd.args.blitFromTexture.srcTexture, + cmd.args.blitFromTexture.srcLevel, + cmd.args.blitFromTexture.srcLayer); + } + } else { + if (ds) { + f->glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, cmd.args.blitFromTexture.srcTarget, + cmd.args.blitFromTexture.srcTexture, cmd.args.blitFromTexture.srcLevel); + f->glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, cmd.args.blitFromTexture.srcTarget, + cmd.args.blitFromTexture.srcTexture, cmd.args.blitFromTexture.srcLevel); + } else { + f->glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, cmd.args.blitFromTexture.srcTarget, + cmd.args.blitFromTexture.srcTexture, cmd.args.blitFromTexture.srcLevel); + } + } f->glBindFramebuffer(GL_DRAW_FRAMEBUFFER, fbo[1]); - if (cmd.args.blitFromRb.target == GL_TEXTURE_3D || cmd.args.blitFromRb.target == GL_TEXTURE_2D_ARRAY) { - f->glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, - cmd.args.blitFromRb.texture, cmd.args.blitFromRb.dstLevel, cmd.args.blitFromRb.dstLayer); + if (cmd.args.blitFromTexture.dstTarget == GL_TEXTURE_3D || cmd.args.blitFromTexture.dstTarget == GL_TEXTURE_2D_ARRAY) { + if (ds) { + f->glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, + cmd.args.blitFromTexture.dstTexture, + cmd.args.blitFromTexture.dstLevel, + cmd.args.blitFromTexture.dstLayer); + f->glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, + cmd.args.blitFromTexture.dstTexture, + cmd.args.blitFromTexture.dstLevel, + cmd.args.blitFromTexture.dstLayer); + } else { + f->glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, + cmd.args.blitFromTexture.dstTexture, + cmd.args.blitFromTexture.dstLevel, + cmd.args.blitFromTexture.dstLayer); + } } else { - f->glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, cmd.args.blitFromRb.target, - cmd.args.blitFromRb.texture, cmd.args.blitFromRb.dstLevel); + if (ds) { + f->glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, cmd.args.blitFromTexture.dstTarget, + cmd.args.blitFromTexture.dstTexture, cmd.args.blitFromTexture.dstLevel); + f->glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, cmd.args.blitFromTexture.dstTarget, + cmd.args.blitFromTexture.dstTexture, cmd.args.blitFromTexture.dstLevel); + } else { + f->glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, cmd.args.blitFromTexture.dstTarget, + cmd.args.blitFromTexture.dstTexture, cmd.args.blitFromTexture.dstLevel); + } } - f->glBlitFramebuffer(0, 0, cmd.args.blitFromRb.w, cmd.args.blitFromRb.h, - 0, 0, cmd.args.blitFromRb.w, cmd.args.blitFromRb.h, - GL_COLOR_BUFFER_BIT, + f->glBlitFramebuffer(0, 0, cmd.args.blitFromTexture.w, cmd.args.blitFromTexture.h, + 0, 0, cmd.args.blitFromTexture.w, cmd.args.blitFromTexture.h, + ds ? GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT : GL_COLOR_BUFFER_BIT, GL_NEAREST); // Qt 5 used Nearest when resolving samples, stick to that f->glBindFramebuffer(GL_FRAMEBUFFER, ctx->defaultFramebufferObject()); f->glDeleteFramebuffers(2, fbo); @@ -3482,6 +3755,13 @@ void QRhiGles2::executeCommandBuffer(QRhiCommandBuffer *cb) if (caps.compute) f->glMemoryBarrier(cmd.args.barrier.barriers); break; + case QGles2CommandBuffer::Command::InvalidateFramebuffer: + if (caps.gles && caps.ctxMajor >= 3) { + f->glInvalidateFramebuffer(GL_DRAW_FRAMEBUFFER, + cmd.args.invalidateFramebuffer.attCount, + cmd.args.invalidateFramebuffer.att); + } + break; default: break; } @@ -3737,7 +4017,7 @@ void QRhiGles2::bindCombinedSampler(QGles2CommandBuffer *cbD, QGles2Texture *tex f->glTexParameteri(texD->target, GL_TEXTURE_MAG_FILTER, GLint(samplerD->d.glmagfilter)); f->glTexParameteri(texD->target, GL_TEXTURE_WRAP_S, GLint(samplerD->d.glwraps)); f->glTexParameteri(texD->target, GL_TEXTURE_WRAP_T, GLint(samplerD->d.glwrapt)); - if (caps.texture3D) + if (caps.texture3D && texD->target == GL_TEXTURE_3D) f->glTexParameteri(texD->target, GL_TEXTURE_WRAP_R, GLint(samplerD->d.glwrapr)); if (caps.textureCompareMode) { if (samplerD->d.gltexcomparefunc != GL_NEVER) { @@ -4280,32 +4560,129 @@ void QRhiGles2::endPass(QRhiCommandBuffer *cb, QRhiResourceUpdateBatch *resource if (cbD->currentTarget->resourceType() == QRhiResource::TextureRenderTarget) { QGles2TextureRenderTarget *rtTex = QRHI_RES(QGles2TextureRenderTarget, cbD->currentTarget); - if (rtTex->m_desc.cbeginColorAttachments() != rtTex->m_desc.cendColorAttachments()) { - // handle only 1 color attachment and only (msaa) renderbuffer - const QRhiColorAttachment &colorAtt(*rtTex->m_desc.cbeginColorAttachments()); - if (colorAtt.resolveTexture()) { - Q_ASSERT(colorAtt.renderBuffer()); + for (auto it = rtTex->m_desc.cbeginColorAttachments(), itEnd = rtTex->m_desc.cendColorAttachments(); + it != itEnd; ++it) + { + const QRhiColorAttachment &colorAtt(*it); + if (!colorAtt.resolveTexture()) + continue; + + QGles2Texture *resolveTexD = QRHI_RES(QGles2Texture, colorAtt.resolveTexture()); + const QSize size = resolveTexD->pixelSize(); + if (colorAtt.renderBuffer()) { QGles2RenderBuffer *rbD = QRHI_RES(QGles2RenderBuffer, colorAtt.renderBuffer()); - const QSize size = colorAtt.resolveTexture()->pixelSize(); if (rbD->pixelSize() != size) { qWarning("Resolve source (%dx%d) and target (%dx%d) size does not match", rbD->pixelSize().width(), rbD->pixelSize().height(), size.width(), size.height()); } QGles2CommandBuffer::Command &cmd(cbD->commands.get()); cmd.cmd = QGles2CommandBuffer::Command::BlitFromRenderbuffer; - cmd.args.blitFromRb.renderbuffer = rbD->renderbuffer; - cmd.args.blitFromRb.w = size.width(); - cmd.args.blitFromRb.h = size.height(); - QGles2Texture *colorTexD = QRHI_RES(QGles2Texture, colorAtt.resolveTexture()); - if (colorTexD->m_flags.testFlag(QRhiTexture::CubeMap)) - cmd.args.blitFromRb.target = GL_TEXTURE_CUBE_MAP_POSITIVE_X + uint(colorAtt.resolveLayer()); + cmd.args.blitFromRenderbuffer.renderbuffer = rbD->renderbuffer; + cmd.args.blitFromRenderbuffer.w = size.width(); + cmd.args.blitFromRenderbuffer.h = size.height(); + if (resolveTexD->m_flags.testFlag(QRhiTexture::CubeMap)) + cmd.args.blitFromRenderbuffer.target = GL_TEXTURE_CUBE_MAP_POSITIVE_X + uint(colorAtt.resolveLayer()); else - cmd.args.blitFromRb.target = colorTexD->target; - cmd.args.blitFromRb.texture = colorTexD->texture; - cmd.args.blitFromRb.dstLevel = colorAtt.resolveLevel(); - const bool hasZ = colorTexD->m_flags.testFlag(QRhiTexture::ThreeDimensional) - || colorTexD->m_flags.testFlag(QRhiTexture::TextureArray); - cmd.args.blitFromRb.dstLayer = hasZ ? colorAtt.resolveLayer() : 0; + cmd.args.blitFromRenderbuffer.target = resolveTexD->target; + cmd.args.blitFromRenderbuffer.dstTexture = resolveTexD->texture; + cmd.args.blitFromRenderbuffer.dstLevel = colorAtt.resolveLevel(); + const bool hasZ = resolveTexD->m_flags.testFlag(QRhiTexture::ThreeDimensional) + || resolveTexD->m_flags.testFlag(QRhiTexture::TextureArray); + cmd.args.blitFromRenderbuffer.dstLayer = hasZ ? colorAtt.resolveLayer() : 0; + cmd.args.blitFromRenderbuffer.isDepthStencil = false; + } else if (caps.glesMultisampleRenderToTexture) { + // Nothing to do, resolving into colorAtt.resolveTexture() is automatic, + // colorAtt.texture() is in fact not used for anything. + } else { + Q_ASSERT(colorAtt.texture()); + QGles2Texture *texD = QRHI_RES(QGles2Texture, colorAtt.texture()); + if (texD->pixelSize() != size) { + qWarning("Resolve source (%dx%d) and target (%dx%d) size does not match", + texD->pixelSize().width(), texD->pixelSize().height(), size.width(), size.height()); + } + const int resolveCount = colorAtt.multiViewCount() >= 2 ? colorAtt.multiViewCount() : 1; + for (int resolveIdx = 0; resolveIdx < resolveCount; ++resolveIdx) { + const int srcLayer = colorAtt.layer() + resolveIdx; + const int dstLayer = colorAtt.resolveLayer() + resolveIdx; + QGles2CommandBuffer::Command &cmd(cbD->commands.get()); + cmd.cmd = QGles2CommandBuffer::Command::BlitFromTexture; + if (texD->m_flags.testFlag(QRhiTexture::CubeMap)) + cmd.args.blitFromTexture.srcTarget = GL_TEXTURE_CUBE_MAP_POSITIVE_X + uint(srcLayer); + else + cmd.args.blitFromTexture.srcTarget = texD->target; + cmd.args.blitFromTexture.srcTexture = texD->texture; + cmd.args.blitFromTexture.srcLevel = colorAtt.level(); + cmd.args.blitFromTexture.srcLayer = 0; + if (texD->m_flags.testFlag(QRhiTexture::ThreeDimensional) || texD->m_flags.testFlag(QRhiTexture::TextureArray)) + cmd.args.blitFromTexture.srcLayer = srcLayer; + cmd.args.blitFromTexture.w = size.width(); + cmd.args.blitFromTexture.h = size.height(); + if (resolveTexD->m_flags.testFlag(QRhiTexture::CubeMap)) + cmd.args.blitFromTexture.dstTarget = GL_TEXTURE_CUBE_MAP_POSITIVE_X + uint(dstLayer); + else + cmd.args.blitFromTexture.dstTarget = resolveTexD->target; + cmd.args.blitFromTexture.dstTexture = resolveTexD->texture; + cmd.args.blitFromTexture.dstLevel = colorAtt.resolveLevel(); + cmd.args.blitFromTexture.dstLayer = 0; + if (resolveTexD->m_flags.testFlag(QRhiTexture::ThreeDimensional) || resolveTexD->m_flags.testFlag(QRhiTexture::TextureArray)) + cmd.args.blitFromTexture.dstLayer = dstLayer; + cmd.args.blitFromTexture.isDepthStencil = false; + } + } + } + + if (rtTex->m_desc.depthResolveTexture()) { + QGles2Texture *depthResolveTexD = QRHI_RES(QGles2Texture, rtTex->m_desc.depthResolveTexture()); + const QSize size = depthResolveTexD->pixelSize(); + if (rtTex->m_desc.depthStencilBuffer()) { + QGles2RenderBuffer *rbD = QRHI_RES(QGles2RenderBuffer, rtTex->m_desc.depthStencilBuffer()); + QGles2CommandBuffer::Command &cmd(cbD->commands.get()); + cmd.cmd = QGles2CommandBuffer::Command::BlitFromRenderbuffer; + cmd.args.blitFromRenderbuffer.renderbuffer = rbD->renderbuffer; + cmd.args.blitFromRenderbuffer.w = size.width(); + cmd.args.blitFromRenderbuffer.h = size.height(); + cmd.args.blitFromRenderbuffer.target = depthResolveTexD->target; + cmd.args.blitFromRenderbuffer.dstTexture = depthResolveTexD->texture; + cmd.args.blitFromRenderbuffer.dstLevel = 0; + cmd.args.blitFromRenderbuffer.dstLayer = 0; + cmd.args.blitFromRenderbuffer.isDepthStencil = true; + } else if (caps.glesMultisampleRenderToTexture) { + // Nothing to do, resolving into depthResolveTexture() is automatic. + } else { + QGles2Texture *depthTexD = QRHI_RES(QGles2Texture, rtTex->m_desc.depthTexture()); + const int resolveCount = depthTexD->arraySize() >= 2 ? depthTexD->arraySize() : 1; + for (int resolveIdx = 0; resolveIdx < resolveCount; ++resolveIdx) { + QGles2CommandBuffer::Command &cmd(cbD->commands.get()); + cmd.cmd = QGles2CommandBuffer::Command::BlitFromTexture; + cmd.args.blitFromTexture.srcTarget = depthTexD->target; + cmd.args.blitFromTexture.srcTexture = depthTexD->texture; + cmd.args.blitFromTexture.srcLevel = 0; + cmd.args.blitFromTexture.srcLayer = resolveIdx; + cmd.args.blitFromTexture.w = size.width(); + cmd.args.blitFromTexture.h = size.height(); + cmd.args.blitFromTexture.dstTarget = depthResolveTexD->target; + cmd.args.blitFromTexture.dstTexture = depthResolveTexD->texture; + cmd.args.blitFromTexture.dstLevel = 0; + cmd.args.blitFromTexture.dstLayer = resolveIdx; + cmd.args.blitFromTexture.isDepthStencil = true; + } + } + } + + const bool mayDiscardDepthStencil = + (rtTex->m_desc.depthStencilBuffer() + || (rtTex->m_desc.depthTexture() && rtTex->m_flags.testFlag(QRhiTextureRenderTarget::DoNotStoreDepthStencilContents))) + && !rtTex->m_desc.depthResolveTexture(); + if (mayDiscardDepthStencil) { + QGles2CommandBuffer::Command &cmd(cbD->commands.get()); + cmd.cmd = QGles2CommandBuffer::Command::InvalidateFramebuffer; + if (caps.needsDepthStencilCombinedAttach) { + cmd.args.invalidateFramebuffer.attCount = 1; + cmd.args.invalidateFramebuffer.att[0] = GL_DEPTH_STENCIL_ATTACHMENT; + } else { + cmd.args.invalidateFramebuffer.attCount = 2; + cmd.args.invalidateFramebuffer.att[0] = GL_DEPTH_ATTACHMENT; + cmd.args.invalidateFramebuffer.att[1] = GL_STENCIL_ATTACHMENT; } } } @@ -4940,6 +5317,9 @@ bool QGles2Buffer::create() rhiD->f->glBindBuffer(targetForDataOps, buffer); rhiD->f->glBufferData(targetForDataOps, nonZeroSize, nullptr, m_type == Dynamic ? GL_DYNAMIC_DRAW : GL_STATIC_DRAW); + if (rhiD->glObjectLabel) + rhiD->glObjectLabel(GL_BUFFER, buffer, -1, m_objectName.constData()); + usageState.access = AccessNone; rhiD->registerResource(this); @@ -5094,6 +5474,9 @@ bool QGles2RenderBuffer::create() break; } + if (rhiD->glObjectLabel) + rhiD->glObjectLabel(GL_RENDERBUFFER, renderbuffer, -1, m_objectName.constData()); + owns = true; generation += 1; rhiD->registerResource(this); @@ -5224,7 +5607,7 @@ bool QGles2Texture::prepareCreate(QSize *adjustedSize) } target = isCube ? GL_TEXTURE_CUBE_MAP - : m_sampleCount > 1 ? GL_TEXTURE_2D_MULTISAMPLE + : m_sampleCount > 1 ? (isArray ? GL_TEXTURE_2D_MULTISAMPLE_ARRAY : GL_TEXTURE_2D_MULTISAMPLE) : (is3D ? GL_TEXTURE_3D : (is1D ? (isArray ? GL_TEXTURE_1D_ARRAY : GL_TEXTURE_1D) : (isArray ? GL_TEXTURE_2D_ARRAY : GL_TEXTURE_2D))); @@ -5316,8 +5699,16 @@ bool QGles2Texture::create() } } } else { - rhiD->f->glTexImage2D(target, 0, GLint(glintformat), size.width(), size.height(), - 0, glformat, gltype, nullptr); + // 2D texture. For multisample textures the GLES 3.1 + // glStorage2DMultisample must be used for portability. + if (m_sampleCount > 1 && rhiD->caps.multisampledTexture) { + // internal format must be sized + rhiD->f->glTexStorage2DMultisample(target, m_sampleCount, glsizedintformat, + size.width(), size.height(), GL_TRUE); + } else { + rhiD->f->glTexImage2D(target, 0, GLint(glintformat), size.width(), size.height(), + 0, glformat, gltype, nullptr); + } } } else { // Must be specified with immutable storage functions otherwise @@ -5328,6 +5719,9 @@ bool QGles2Texture::create() else if (!is1D && (is3D || isArray)) rhiD->f->glTexStorage3D(target, mipLevelCount, glsizedintformat, size.width(), size.height(), is3D ? qMax(1, m_depth) : qMax(0, m_arraySize)); + else if (m_sampleCount > 1) + rhiD->f->glTexStorage2DMultisample(target, m_sampleCount, glsizedintformat, + size.width(), size.height(), GL_TRUE); else rhiD->f->glTexStorage2D(target, mipLevelCount, glsizedintformat, size.width(), is1D ? qMax(0, m_arraySize) : size.height()); @@ -5340,6 +5734,9 @@ bool QGles2Texture::create() specified = false; } + if (rhiD->glObjectLabel) + rhiD->glObjectLabel(GL_TEXTURE, texture, -1, m_objectName.constData()); + owns = true; generation += 1; @@ -5496,8 +5893,10 @@ void QGles2TextureRenderTarget::destroy() e.type = QRhiGles2::DeferredReleaseEntry::TextureRenderTarget; e.textureRenderTarget.framebuffer = framebuffer; + e.textureRenderTarget.nonMsaaThrowawayDepthTexture = nonMsaaThrowawayDepthTexture; framebuffer = 0; + nonMsaaThrowawayDepthTexture = 0; QRHI_RES_RHI(QRhiGles2); if (rhiD) { @@ -5544,6 +5943,7 @@ bool QGles2TextureRenderTarget::create() d.colorAttCount = 0; int attIndex = 0; + int multiViewCount = 0; for (auto it = m_desc.cbeginColorAttachments(), itEnd = m_desc.cendColorAttachments(); it != itEnd; ++it, ++attIndex) { d.colorAttCount += 1; const QRhiColorAttachment &colorAtt(*it); @@ -5554,20 +5954,56 @@ bool QGles2TextureRenderTarget::create() QGles2Texture *texD = QRHI_RES(QGles2Texture, texture); Q_ASSERT(texD->texture && texD->specified); if (texD->flags().testFlag(QRhiTexture::ThreeDimensional) || texD->flags().testFlag(QRhiTexture::TextureArray)) { - rhiD->f->glFramebufferTextureLayer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + uint(attIndex), texD->texture, - colorAtt.level(), colorAtt.layer()); + if (colorAtt.multiViewCount() < 2) { + rhiD->f->glFramebufferTextureLayer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + uint(attIndex), texD->texture, + colorAtt.level(), colorAtt.layer()); + } else { + multiViewCount = colorAtt.multiViewCount(); + if (texD->sampleCount() > 1 && rhiD->caps.glesMultiviewMultisampleRenderToTexture && colorAtt.resolveTexture()) { + // Special path for GLES and GL_OVR_multiview_multisampled_render_to_texture: + // ignore the color attachment's (multisample) texture + // array and give the resolve texture array to GL. (no + // explicit resolving is needed by us later on) + QGles2Texture *resolveTexD = QRHI_RES(QGles2Texture, colorAtt.resolveTexture()); + rhiD->glFramebufferTextureMultisampleMultiviewOVR(GL_FRAMEBUFFER, + GL_COLOR_ATTACHMENT0 + uint(attIndex), + resolveTexD->texture, + colorAtt.resolveLevel(), + texD->sampleCount(), + colorAtt.resolveLayer(), + multiViewCount); + } else { + rhiD->glFramebufferTextureMultiviewOVR(GL_FRAMEBUFFER, + GL_COLOR_ATTACHMENT0 + uint(attIndex), + texD->texture, + colorAtt.level(), + colorAtt.layer(), + multiViewCount); + } + } } else if (texD->flags().testFlag(QRhiTexture::OneDimensional)) { rhiD->glFramebufferTexture1D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + uint(attIndex), texD->target + uint(colorAtt.layer()), texD->texture, colorAtt.level()); } else { - const GLenum faceTargetBase = texD->flags().testFlag(QRhiTexture::CubeMap) ? GL_TEXTURE_CUBE_MAP_POSITIVE_X : texD->target; - rhiD->f->glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + uint(attIndex), faceTargetBase + uint(colorAtt.layer()), - texD->texture, colorAtt.level()); + if (texD->sampleCount() > 1 && rhiD->caps.glesMultisampleRenderToTexture && colorAtt.resolveTexture()) { + // Special path for GLES and GL_EXT_multisampled_render_to_texture: + // ignore the color attachment's (multisample) texture and + // give the resolve texture to GL. (no explicit resolving is + // needed by us later on) + QGles2Texture *resolveTexD = QRHI_RES(QGles2Texture, colorAtt.resolveTexture()); + const GLenum faceTargetBase = resolveTexD->flags().testFlag(QRhiTexture::CubeMap) ? GL_TEXTURE_CUBE_MAP_POSITIVE_X : resolveTexD->target; + rhiD->glFramebufferTexture2DMultisampleEXT(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + uint(attIndex), faceTargetBase + uint(colorAtt.resolveLayer()), + resolveTexD->texture, colorAtt.level(), texD->sampleCount()); + } else { + const GLenum faceTargetBase = texD->flags().testFlag(QRhiTexture::CubeMap) ? GL_TEXTURE_CUBE_MAP_POSITIVE_X : texD->target; + rhiD->f->glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + uint(attIndex), faceTargetBase + uint(colorAtt.layer()), + texD->texture, colorAtt.level()); + } } if (attIndex == 0) { d.pixelSize = rhiD->q->sizeForMipLevel(colorAtt.level(), texD->pixelSize()); - d.sampleCount = 1; + d.sampleCount = texD->sampleCount(); } } else if (renderBuffer) { QGles2RenderBuffer *rbD = QRHI_RES(QGles2RenderBuffer, renderBuffer); @@ -5588,12 +6024,14 @@ bool QGles2TextureRenderTarget::create() } else { rhiD->f->glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_RENDERBUFFER, depthRbD->renderbuffer); - if (depthRbD->stencilRenderbuffer) + if (depthRbD->stencilRenderbuffer) { rhiD->f->glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_RENDERBUFFER, depthRbD->stencilRenderbuffer); - else // packed + } else { + // packed depth-stencil rhiD->f->glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_RENDERBUFFER, depthRbD->renderbuffer); + } } if (d.colorAttCount == 0) { d.pixelSize = depthRbD->pixelSize(); @@ -5601,11 +6039,105 @@ bool QGles2TextureRenderTarget::create() } } else { QGles2Texture *depthTexD = QRHI_RES(QGles2Texture, m_desc.depthTexture()); - rhiD->f->glFramebufferTexture2D(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, depthTexD->target, - depthTexD->texture, 0); + if (multiViewCount < 2) { + if (depthTexD->sampleCount() > 1 && rhiD->caps.glesMultisampleRenderToTexture && m_desc.depthResolveTexture()) { + // Special path for GLES and + // GL_EXT_multisampled_render_to_texture, for depth-stencil. + // Relevant only when depthResolveTexture is set. + QGles2Texture *depthResolveTexD = QRHI_RES(QGles2Texture, m_desc.depthResolveTexture()); + rhiD->glFramebufferTexture2DMultisampleEXT(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, depthResolveTexD->target, + depthResolveTexD->texture, 0, depthTexD->sampleCount()); + if (rhiD->isStencilSupportingFormat(depthResolveTexD->format())) { + rhiD->glFramebufferTexture2DMultisampleEXT(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, depthResolveTexD->target, + depthResolveTexD->texture, 0, depthTexD->sampleCount()); + } + } else { + rhiD->f->glFramebufferTexture2D(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, depthTexD->target, + depthTexD->texture, 0); + if (rhiD->isStencilSupportingFormat(depthTexD->format())) { + rhiD->f->glFramebufferTexture2D(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, depthTexD->target, + depthTexD->texture, 0); + } + } + } else { + if (depthTexD->sampleCount() > 1 && rhiD->caps.glesMultiviewMultisampleRenderToTexture) { + // And so it turns out + // https://registry.khronos.org/OpenGL/extensions/OVR/OVR_multiview.txt + // does not work with multisample 2D texture arrays. (at least + // that's what Issue 30 in the extension spec seems to imply) + // + // There is https://registry.khronos.org/OpenGL/extensions/EXT/EXT_multiview_texture_multisample.txt + // that seems to resolve that, but that does not seem to + // work (or not available) on GLES devices such as the Quest 3. + // + // So instead, on GLES we can use the + // multisample-multiview-auto-resolving version (which in + // turn is not supported on desktop GL e.g. by NVIDIA), too + // bad we have a multisample depth texture array here as + // every other API out there requires that. So, in absence + // of a depthResolveTexture, create a temporary one ignoring + // what the user has already created. + // + if (!m_flags.testFlag(DoNotStoreDepthStencilContents) && !m_desc.depthResolveTexture()) { + qWarning("Attempted to create a multiview+multisample QRhiTextureRenderTarget, but DoNotStoreDepthStencilContents was not set." + " This path has no choice but to behave as if DoNotStoreDepthStencilContents was set, because QRhi is forced to create" + " a throwaway non-multisample depth texture here. Set the flag to silence this warning, or set a depthResolveTexture."); + } + if (m_desc.depthResolveTexture()) { + QGles2Texture *depthResolveTexD = QRHI_RES(QGles2Texture, m_desc.depthResolveTexture()); + rhiD->glFramebufferTextureMultisampleMultiviewOVR(GL_FRAMEBUFFER, + GL_DEPTH_ATTACHMENT, + depthResolveTexD->texture, + 0, + depthTexD->sampleCount(), + 0, + multiViewCount); + if (rhiD->isStencilSupportingFormat(depthResolveTexD->format())) { + rhiD->glFramebufferTextureMultisampleMultiviewOVR(GL_FRAMEBUFFER, + GL_STENCIL_ATTACHMENT, + depthResolveTexD->texture, + 0, + depthTexD->sampleCount(), + 0, + multiViewCount); + } + } else { + if (!nonMsaaThrowawayDepthTexture) { + rhiD->f->glGenTextures(1, &nonMsaaThrowawayDepthTexture); + rhiD->f->glBindTexture(GL_TEXTURE_2D_ARRAY, nonMsaaThrowawayDepthTexture); + rhiD->f->glTexStorage3D(GL_TEXTURE_2D_ARRAY, 1, GL_DEPTH24_STENCIL8, + depthTexD->pixelSize().width(), depthTexD->pixelSize().height(), multiViewCount); + } + rhiD->glFramebufferTextureMultisampleMultiviewOVR(GL_FRAMEBUFFER, + GL_DEPTH_ATTACHMENT, + nonMsaaThrowawayDepthTexture, + 0, + depthTexD->sampleCount(), + 0, + multiViewCount); + rhiD->glFramebufferTextureMultisampleMultiviewOVR(GL_FRAMEBUFFER, + GL_STENCIL_ATTACHMENT, + nonMsaaThrowawayDepthTexture, + 0, + depthTexD->sampleCount(), + 0, + multiViewCount); + } + } else { + // The depth texture here must be an array with at least + // multiViewCount elements, and the format should be D24 or D32F + // for depth only, or D24S8 for depth and stencil. + rhiD->glFramebufferTextureMultiviewOVR(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, depthTexD->texture, + 0, 0, multiViewCount); + if (rhiD->isStencilSupportingFormat(depthTexD->format())) { + rhiD->glFramebufferTextureMultiviewOVR(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, depthTexD->texture, + 0, 0, multiViewCount); + } + } + } if (d.colorAttCount == 0) { d.pixelSize = depthTexD->pixelSize(); - d.sampleCount = 1; + d.sampleCount = depthTexD->sampleCount(); } } d.dsAttCount = 1; @@ -5622,6 +6154,9 @@ bool QGles2TextureRenderTarget::create() return false; } + if (rhiD->glObjectLabel) + rhiD->glObjectLabel(GL_FRAMEBUFFER, framebuffer, -1, m_objectName.constData()); + QRhiRenderTargetAttachmentTracker::updateResIdList<QGles2Texture, QGles2RenderBuffer>(m_desc, &d.currentResIdList); rhiD->registerResource(this); @@ -5868,6 +6403,9 @@ bool QGles2GraphicsPipeline::create() currentSrb = nullptr; currentSrbGeneration = 0; + if (rhiD->glObjectLabel) + rhiD->glObjectLabel(GL_PROGRAM, program, -1, m_objectName.constData()); + rhiD->pipelineCreationEnd(); generation += 1; rhiD->registerResource(this); @@ -6038,7 +6576,12 @@ QRhiRenderTarget *QGles2SwapChain::currentFrameRenderTarget(StereoTargetBuffer t QSize QGles2SwapChain::surfacePixelSize() { Q_ASSERT(m_window); - return m_window->size() * m_window->devicePixelRatio(); + if (QPlatformWindow *platformWindow = m_window->handle()) + // Prefer using QPlatformWindow geometry and DPR in order to avoid + // errors due to rounded QWindow geometry. + return platformWindow->geometry().size() * platformWindow->devicePixelRatio(); + else + return m_window->size() * m_window->devicePixelRatio(); } bool QGles2SwapChain::isFormatSupported(Format f) @@ -6095,15 +6638,59 @@ bool QGles2SwapChain::createOrResize() frameCount = 0; + QRHI_RES_RHI(QRhiGles2); + if (rhiD->rhiFlags.testFlag(QRhi::EnableTimestamps) && rhiD->caps.timestamps) + timestamps.prepare(rhiD); + // The only reason to register this fairly fake gl swapchain // object with no native resources underneath is to be able to // implement a safe destroy(). - if (needsRegistration) { - QRHI_RES_RHI(QRhiGles2); + if (needsRegistration) rhiD->registerResource(this, false); - } return true; } +void QGles2SwapChainTimestamps::prepare(QRhiGles2 *rhiD) +{ + if (!query[0]) + rhiD->f->glGenQueries(TIMESTAMP_PAIRS * 2, query); +} + +void QGles2SwapChainTimestamps::destroy(QRhiGles2 *rhiD) +{ + rhiD->f->glDeleteQueries(TIMESTAMP_PAIRS * 2, query); + memset(active, 0, sizeof(active)); + memset(query, 0, sizeof(query)); +} + +bool QGles2SwapChainTimestamps::tryQueryTimestamps(int pairIndex, QRhiGles2 *rhiD, double *elapsedSec) +{ + if (!active[pairIndex]) + return false; + + GLuint tsStart = query[pairIndex * 2]; + GLuint tsEnd = query[pairIndex * 2 + 1]; + + GLuint ready = GL_FALSE; + rhiD->f->glGetQueryObjectuiv(tsEnd, GL_QUERY_RESULT_AVAILABLE, &ready); + + if (!ready) + return false; + + bool result = false; + quint64 timestamps[2]; + rhiD->glGetQueryObjectui64v(tsStart, GL_QUERY_RESULT, ×tamps[0]); + rhiD->glGetQueryObjectui64v(tsEnd, GL_QUERY_RESULT, ×tamps[1]); + + if (timestamps[1] >= timestamps[0]) { + const quint64 nanoseconds = timestamps[1] - timestamps[0]; + *elapsedSec = nanoseconds / 1000000000.0; + result = true; + } + + active[pairIndex] = false; + return result; +} + QT_END_NAMESPACE diff --git a/src/gui/rhi/qrhigles2_p.h b/src/gui/rhi/qrhigles2_p.h index 47d6c3f6d7..4139579864 100644 --- a/src/gui/rhi/qrhigles2_p.h +++ b/src/gui/rhi/qrhigles2_p.h @@ -16,7 +16,7 @@ // #include "qrhi_p.h" -#include "qshaderdescription.h" +#include <rhi/qshaderdescription.h> #include <qopengl.h> #include <QByteArray> #include <QWindow> @@ -27,6 +27,7 @@ QT_BEGIN_NAMESPACE class QOpenGLExtensions; +class QRhiGles2; struct QGles2Buffer : public QRhiBuffer { @@ -214,6 +215,7 @@ struct QGles2TextureRenderTarget : public QRhiTextureRenderTarget QGles2RenderTargetData d; GLuint framebuffer = 0; + GLuint nonMsaaThrowawayDepthTexture = 0; friend class QRhiGles2; }; @@ -331,11 +333,13 @@ struct QGles2CommandBuffer : public QRhiCommandBuffer CompressedImage, CompressedSubImage, BlitFromRenderbuffer, + BlitFromTexture, GenMip, BindComputePipeline, Dispatch, BarriersForPass, - Barrier + Barrier, + InvalidateFramebuffer }; Cmd cmd; @@ -343,6 +347,12 @@ struct QGles2CommandBuffer : public QRhiCommandBuffer // QRhiTexture/Buffer/etc. pointers). union Args { struct { + GLuint timestampQuery; + } beginFrame; + struct { + GLuint timestampQuery; + } endFrame; + struct { float x, y, w, h; float d0, d1; } viewport; @@ -494,10 +504,24 @@ struct QGles2CommandBuffer : public QRhiCommandBuffer int w; int h; GLenum target; - GLuint texture; + GLuint dstTexture; int dstLevel; int dstLayer; - } blitFromRb; + bool isDepthStencil; + } blitFromRenderbuffer; + struct { + GLenum srcTarget; + GLuint srcTexture; + int srcLevel; + int srcLayer; + int w; + int h; + GLenum dstTarget; + GLuint dstTexture; + int dstLevel; + int dstLayer; + bool isDepthStencil; + } blitFromTexture; struct { GLenum target; GLuint texture; @@ -516,6 +540,10 @@ struct QGles2CommandBuffer : public QRhiCommandBuffer struct { GLbitfield barriers; } barrier; + struct { + int attCount; + GLenum att[3]; + } invalidateFramebuffer; } args; }; @@ -531,6 +559,7 @@ struct QGles2CommandBuffer : public QRhiCommandBuffer PassType recordingPass; bool passNeedsResourceTracking; + double lastGpuTime = 0; QRhiRenderTarget *currentTarget; QRhiGraphicsPipeline *currentGraphicsPipeline; QRhiComputePipeline *currentComputePipeline; @@ -626,6 +655,7 @@ struct QGles2CommandBuffer : public QRhiCommandBuffer void resetState() { recordingPass = NoPass; passNeedsResourceTracking = true; + // do not zero lastGpuTime currentTarget = nullptr; resetCommands(); resetCachedState(); @@ -687,6 +717,18 @@ inline bool operator!=(const QGles2CommandBuffer::GraphicsPassState::Blend &a, return !(a == b); } +struct QGles2SwapChainTimestamps +{ + static const int TIMESTAMP_PAIRS = 2; + + bool active[TIMESTAMP_PAIRS] = {}; + GLuint query[TIMESTAMP_PAIRS * 2] = {}; + + void prepare(QRhiGles2 *rhiD); + void destroy(QRhiGles2 *rhiD); + bool tryQueryTimestamps(int pairIndex, QRhiGles2 *rhiD, double *elapsedSec); +}; + struct QGles2SwapChain : public QRhiSwapChain { QGles2SwapChain(QRhiImplementation *rhi); @@ -712,6 +754,8 @@ struct QGles2SwapChain : public QRhiSwapChain QGles2SwapChainRenderTarget rtRight; QGles2CommandBuffer cb; int frameCount = 0; + QGles2SwapChainTimestamps timestamps; + int currentTimestampPairIndex = 0; }; class QRhiGles2 : public QRhiImplementation @@ -854,7 +898,6 @@ public: QGles2RenderTargetData *enqueueBindFramebuffer(QRhiRenderTarget *rt, QGles2CommandBuffer *cbD, bool *wantsColorClear = nullptr, bool *wantsDsClear = nullptr); void enqueueBarriersForPass(QGles2CommandBuffer *cbD); - int effectiveSampleCount(int sampleCount) const; QByteArray shaderSource(const QRhiShaderStage &shaderStage, QShaderVersion *shaderVersion); bool compileShader(GLuint program, const QRhiShaderStage &shaderStage, QShaderVersion *shaderVersion); bool linkProgram(GLuint program); @@ -909,7 +952,13 @@ public: GLsizei, const GLvoid *) = nullptr; void(QOPENGLF_APIENTRYP glFramebufferTexture1D)(GLenum, GLenum, GLenum, GLuint, GLint) = nullptr; - + void(QOPENGLF_APIENTRYP glFramebufferTextureMultiviewOVR)(GLenum, GLenum, GLuint, GLint, + GLint, GLsizei) = nullptr; + void (QOPENGLF_APIENTRYP glQueryCounter)(GLuint, GLenum) = nullptr; + void (QOPENGLF_APIENTRYP glGetQueryObjectui64v)(GLuint, GLenum, quint64 *) = nullptr; + void (QOPENGLF_APIENTRYP glObjectLabel)(GLenum, GLuint, GLsizei, const GLchar *) = nullptr; + void (QOPENGLF_APIENTRYP glFramebufferTexture2DMultisampleEXT)(GLenum, GLenum, GLenum, GLuint, GLint, GLsizei) = nullptr; + void (QOPENGLF_APIENTRYP glFramebufferTextureMultisampleMultiviewOVR)(GLenum, GLenum, GLuint, GLint, GLsizei, GLint, GLsizei) = nullptr; uint vao = 0; struct Caps { Caps() @@ -941,7 +990,7 @@ public: depthTexture(false), packedDepthStencil(false), needsDepthStencilCombinedAttach(false), - srgbCapableDefaultFramebuffer(false), + srgbWriteControl(false), coreProfile(false), uniformBuffers(false), elementIndexUint(false), @@ -962,7 +1011,13 @@ public: geometryShader(false), texture1D(false), hasDrawBuffersFunc(false), - halfAttributes(false) + halfAttributes(false), + multiView(false), + timestamps(false), + objectLabel(false), + glesMultisampleRenderToTexture(false), + glesMultiviewMultisampleRenderToTexture(false), + unpackRowLength(false) { } int ctxMajor; int ctxMinor; @@ -994,7 +1049,7 @@ public: uint depthTexture : 1; uint packedDepthStencil : 1; uint needsDepthStencilCombinedAttach : 1; - uint srgbCapableDefaultFramebuffer : 1; + uint srgbWriteControl : 1; uint coreProfile : 1; uint uniformBuffers : 1; uint elementIndexUint : 1; @@ -1016,6 +1071,12 @@ public: uint texture1D : 1; uint hasDrawBuffersFunc : 1; uint halfAttributes : 1; + uint multiView : 1; + uint timestamps : 1; + uint objectLabel : 1; + uint glesMultisampleRenderToTexture : 1; + uint glesMultiviewMultisampleRenderToTexture : 1; + uint unpackRowLength : 1; } caps; QGles2SwapChain *currentSwapChain = nullptr; QSet<GLint> supportedCompressedFormats; @@ -1049,6 +1110,7 @@ public: } renderbuffer; struct { GLuint framebuffer; + GLuint nonMsaaThrowawayDepthTexture; } textureRenderTarget; }; }; @@ -1058,6 +1120,7 @@ public: OffscreenFrame(QRhiImplementation *rhi) : cbWrapper(rhi) { } bool active = false; QGles2CommandBuffer cbWrapper; + GLuint tsQueries[2] = {}; } ofr; QHash<QRhiShaderStage, uint> m_shaderCache; diff --git a/src/gui/rhi/qrhimetal.mm b/src/gui/rhi/qrhimetal.mm index 7ef200649f..887d40c7a5 100644 --- a/src/gui/rhi/qrhimetal.mm +++ b/src/gui/rhi/qrhimetal.mm @@ -13,6 +13,7 @@ #include <QOperatingSystemVersion> #include <QtCore/private/qcore_mac_p.h> +#include <QtGui/private/qmetallayer_p.h> #ifdef Q_OS_MACOS #include <AppKit/AppKit.h> @@ -20,8 +21,9 @@ #include <UIKit/UIKit.h> #endif +#include <QuartzCore/CATransaction.h> + #include <Metal/Metal.h> -#include <QuartzCore/CAMetalLayer.h> QT_BEGIN_NAMESPACE @@ -40,12 +42,18 @@ QT_BEGIN_NAMESPACE #error ARC not supported #endif -// Note: we expect everything here pass the Metal API validation when running -// in Debug mode in XCode (or with METAL_DEVICE_WRAPPER_TYPE=1). An exception -// is the nextDrawable Called Early blah blah warning, which is plain and -// simply false. This may not be present with newer XCode. There may also be -// warnings about threading (e.g. about accessing view.layer), those are -// expected for now. +// Even though the macOS 13 MTLBinaryArchive problem (QTBUG-106703) seems +// to be solved in later 13.x releases, we have reports from old Intel hardware +// and older macOS versions where this causes problems (QTBUG-114338). +// Thus we no longer do OS version based differentiation, but rather have a +// single toggle that is currently on, and so QRhi::(set)pipelineCache() +// does nothing with Metal. +#define QRHI_METAL_DISABLE_BINARY_ARCHIVE + +// We should be able to operate with command buffers that do not automatically +// retain/release the resources used by them. (since we have logic that mirrors +// other backends such as the Vulkan one anyway) +#define QRHI_METAL_COMMAND_BUFFERS_WITH_UNRETAINED_REFERENCES /*! \class QRhiMetalInitParams @@ -53,7 +61,7 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief Metal specific initialization parameters. - \note This an RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. A Metal-based QRhi needs no special parameters for initialization. @@ -91,16 +99,21 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief Holds the Metal device used by the QRhi. - \note This an RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ /*! \variable QRhiMetalNativeHandles::dev + + Set to a valid MTLDevice to import an existing device. */ /*! \variable QRhiMetalNativeHandles::cmdQueue + + Set to a valid MTLCommandQueue when importing an existing command queue. + When \nullptr, QRhi will create a new command queue. */ /*! @@ -119,7 +132,7 @@ QT_BEGIN_NAMESPACE between \l{QRhiCommandBuffer::beginPass()} - \l{QRhiCommandBuffer::endPass()}. - \note This an RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -158,8 +171,8 @@ struct QRhiMetalData id<MTLDevice> dev = nil; id<MTLCommandQueue> cmdQueue = nil; API_AVAILABLE(macosx(11.0), ios(14.0)) id<MTLBinaryArchive> binArch = nil; - bool binArchWasEmpty = false; + id<MTLCommandBuffer> newCommandBuffer(); MTLRenderPassDescriptor *createDefaultRenderPass(bool hasDepthStencil, const QColor &colorClearValue, const QRhiDepthStencilClearValue &depthStencilClearValue, @@ -357,8 +370,11 @@ struct QMetalRenderTargetData struct { ColorAtt colorAtt[QMetalRenderPassDescriptor::MAX_COLOR_ATTACHMENTS]; id<MTLTexture> dsTex = nil; + id<MTLTexture> dsResolveTex = nil; bool hasStencil = false; bool depthNeedsStore = false; + bool preserveColor = false; + bool preserveDs = false; } fb; QRhiRenderTargetAttachmentTracker::ResIdList currentResIdList; @@ -377,6 +393,15 @@ struct QMetalGraphicsPipelineData float slopeScaledDepthBias; QMetalShader vs; QMetalShader fs; + struct ExtraBufferManager { + enum class WorkBufType { + DeviceLocal, + HostVisible + }; + QMetalBuffer *acquireWorkBuffer(QRhiMetal *rhiD, quint32 size, WorkBufType type = WorkBufType::DeviceLocal); + QVector<QMetalBuffer *> deviceLocalWorkBuffers; + QVector<QMetalBuffer *> hostVisibleWorkBuffers; + } extraBufMgr; struct Tessellation { QMetalGraphicsPipelineData *q = nullptr; bool enabled = false; @@ -410,13 +435,6 @@ struct QMetalGraphicsPipelineData id<MTLComputePipelineState> vsCompPipeline(QRhiMetal *rhiD, QShader::Variant vertexCompVariant); id<MTLComputePipelineState> tescCompPipeline(QRhiMetal *rhiD); id<MTLRenderPipelineState> teseFragRenderPipeline(QRhiMetal *rhiD, QMetalGraphicsPipeline *pipeline); - enum class WorkBufType { - DeviceLocal, - HostVisible - }; - QMetalBuffer *acquireWorkBuffer(QRhiMetal *rhiD, quint32 size, WorkBufType type = WorkBufType::DeviceLocal); - QVector<QMetalBuffer *> deviceLocalWorkBuffers; - QVector<QMetalBuffer *> hostVisibleWorkBuffers; } tess; void setupVertexInputDescriptor(MTLVertexDescriptor *desc); void setupStageInputDescriptor(MTLStageInputOutputDescriptor *desc); @@ -445,11 +463,6 @@ struct QMetalSwapChainData id<MTLTexture> msaaTex[QMTL_FRAMES_IN_FLIGHT]; QRhiTexture::Format rhiColorFormat; MTLPixelFormat colorFormat; -#ifdef Q_OS_MACOS - bool liveResizeObserverSet = false; - QMacNotificationObserver liveResizeStartObserver; - QMacNotificationObserver liveResizeEndObserver; -#endif }; QRhiMetal::QRhiMetal(QRhiMetalInitParams *params, QRhiMetalNativeHandles *importDevice) @@ -460,7 +473,7 @@ QRhiMetal::QRhiMetal(QRhiMetalInitParams *params, QRhiMetalNativeHandles *import importedDevice = importDevice != nullptr; if (importedDevice) { - if (d->dev) { + if (importDevice->dev) { d->dev = (id<MTLDevice>) importDevice->dev; importedCmdQueue = importDevice->cmdQueue != nullptr; if (importedCmdQueue) @@ -494,8 +507,24 @@ bool QRhiMetal::probe(QRhiMetalInitParams *params) return false; } +id<MTLCommandBuffer> QRhiMetalData::newCommandBuffer() +{ +#ifdef QRHI_METAL_COMMAND_BUFFERS_WITH_UNRETAINED_REFERENCES + // Do not let the command buffer mess with the refcount of objects. We do + // have a proper render loop and will manage lifetimes similarly to other + // backends (Vulkan). + return [cmdQueue commandBufferWithUnretainedReferences]; +#else + return [cmdQueue commandBuffer]; +#endif +} + bool QRhiMetalData::setupBinaryArchive(NSURL *sourceFileUrl) { +#ifdef QRHI_METAL_DISABLE_BINARY_ARCHIVE + return false; +#endif + if (@available(macOS 11.0, iOS 14.0, *)) { [binArch release]; MTLBinaryArchiveDescriptor *binArchDesc = [MTLBinaryArchiveDescriptor new]; @@ -508,7 +537,6 @@ bool QRhiMetalData::setupBinaryArchive(NSURL *sourceFileUrl) qWarning("newBinaryArchiveWithDescriptor failed: %s", qPrintable(msg)); return false; } - binArchWasEmpty = sourceFileUrl == nil; return true; } return false; @@ -536,9 +564,7 @@ bool QRhiMetal::create(QRhi::Flags flags) // suitable as deviceId because it does not seem stable on macOS and can // apparently change when the system is rebooted. -#ifdef Q_OS_IOS - driverInfoStruct.deviceType = QRhiDriverInfo::IntegratedDevice; -#else +#ifdef Q_OS_MACOS if (@available(macOS 10.15, *)) { const MTLDeviceLocation deviceLocation = [d->dev location]; switch (deviceLocation) { @@ -555,6 +581,8 @@ bool QRhiMetal::create(QRhi::Flags flags) break; } } +#else + driverInfoStruct.deviceType = QRhiDriverInfo::IntegratedDevice; #endif const QOperatingSystemVersion ver = QOperatingSystemVersion::current(); @@ -580,6 +608,7 @@ bool QRhiMetal::create(QRhi::Flags flags) if (@available(macOS 10.15, *)) caps.isAppleGPU = [d->dev supportsFamily:MTLGPUFamilyApple7]; caps.maxThreadGroupSize = 1024; + caps.multiView = true; #elif defined(Q_OS_TVOS) if ([d->dev supportsFeatureSet: MTLFeatureSet(30003)]) // MTLFeatureSet_tvOS_GPUFamily2_v1 caps.maxTextureSize = 16384; @@ -606,8 +635,10 @@ bool QRhiMetal::create(QRhi::Flags flags) } caps.isAppleGPU = true; if (@available(iOS 13, *)) { - if ([d->dev supportsFamily:MTLGPUFamilyApple4]) + if ([d->dev supportsFamily: MTLGPUFamilyApple4]) caps.maxThreadGroupSize = 1024; + if ([d->dev supportsFamily: MTLGPUFamilyApple5]) + caps.multiView = true; } #endif @@ -657,17 +688,6 @@ QVector<int> QRhiMetal::supportedSampleCounts() const return caps.supportedSampleCounts; } -int QRhiMetal::effectiveSampleCount(int sampleCount) const -{ - // Stay compatible with QSurfaceFormat and friends where samples == 0 means the same as 1. - const int s = qBound(1, sampleCount, 64); - if (!supportedSampleCounts().contains(s)) { - qWarning("Attempted to set unsupported sample count %d", sampleCount); - return 1; - } - return s; -} - QRhiSwapChain *QRhiMetal::createSwapChain() { return new QMetalSwapChain(this); @@ -833,6 +853,12 @@ bool QRhiMetal::isFeatureSupported(QRhi::Feature feature) const return false; case QRhi::ThreeDimensionalTextureMipmaps: return true; + case QRhi::MultiView: + return caps.multiView; + case QRhi::TextureViewFormat: + return false; + case QRhi::ResolveDepthStencil: + return true; default: Q_UNREACHABLE(); return false; @@ -1469,11 +1495,11 @@ void QRhiMetal::setGraphicsPipeline(QRhiCommandBuffer *cb, QRhiGraphicsPipeline psD->makeActiveForCurrentRenderPassEncoder(cbD); } else { // mark work buffers that can now be safely reused as reusable - for (QMetalBuffer *workBuf : psD->d->tess.deviceLocalWorkBuffers) { + for (QMetalBuffer *workBuf : psD->d->extraBufMgr.deviceLocalWorkBuffers) { if (workBuf && workBuf->lastActiveFrameSlot == currentFrameSlot) workBuf->lastActiveFrameSlot = -1; } - for (QMetalBuffer *workBuf : psD->d->tess.hostVisibleWorkBuffers) { + for (QMetalBuffer *workBuf : psD->d->extraBufMgr.hostVisibleWorkBuffers) { if (workBuf && workBuf->lastActiveFrameSlot == currentFrameSlot) workBuf->lastActiveFrameSlot = -1; } @@ -1965,6 +1991,7 @@ void QRhiMetal::tessellatedDraw(const TessDrawArgs &args) const quint32 vertexOrIndexCount = indexed ? args.drawIndexed.indexCount : args.draw.vertexCount; QMetalGraphicsPipelineData::Tessellation &tess(graphicsPipeline->d->tess); + QMetalGraphicsPipelineData::ExtraBufferManager &extraBufMgr(graphicsPipeline->d->extraBufMgr); const quint32 patchCount = tess.patchCountForDrawCall(vertexOrIndexCount, instanceCount); QMetalBuffer *vertOutBuf = nullptr; QMetalBuffer *tescOutBuf = nullptr; @@ -1998,7 +2025,7 @@ void QRhiMetal::tessellatedDraw(const TessDrawArgs &args) if (outputBufferBinding >= 0) { const quint32 workBufSize = tess.vsCompOutputBufferSize(vertexOrIndexCount, instanceCount); - vertOutBuf = tess.acquireWorkBuffer(this, workBufSize); + vertOutBuf = extraBufMgr.acquireWorkBuffer(this, workBufSize); if (!vertOutBuf) return; [computeEncoder setBuffer: vertOutBuf->d->buf[0] offset: 0 atIndex: outputBufferBinding]; @@ -2046,7 +2073,7 @@ void QRhiMetal::tessellatedDraw(const TessDrawArgs &args) if (outputBufferBinding >= 0) { const quint32 workBufSize = tess.tescCompOutputBufferSize(patchCount); - tescOutBuf = tess.acquireWorkBuffer(this, workBufSize); + tescOutBuf = extraBufMgr.acquireWorkBuffer(this, workBufSize); if (!tescOutBuf) return; [computeEncoder setBuffer: tescOutBuf->d->buf[0] offset: 0 atIndex: outputBufferBinding]; @@ -2054,14 +2081,14 @@ void QRhiMetal::tessellatedDraw(const TessDrawArgs &args) if (patchOutputBufferBinding >= 0) { const quint32 workBufSize = tess.tescCompPatchOutputBufferSize(patchCount); - tescPatchOutBuf = tess.acquireWorkBuffer(this, workBufSize); + tescPatchOutBuf = extraBufMgr.acquireWorkBuffer(this, workBufSize); if (!tescPatchOutBuf) return; [computeEncoder setBuffer: tescPatchOutBuf->d->buf[0] offset: 0 atIndex: patchOutputBufferBinding]; } if (tessFactorBufferBinding >= 0) { - tescFactorBuf = tess.acquireWorkBuffer(this, patchCount * sizeof(MTLQuadTessellationFactorsHalf)); + tescFactorBuf = extraBufMgr.acquireWorkBuffer(this, patchCount * sizeof(MTLQuadTessellationFactorsHalf)); [computeEncoder setBuffer: tescFactorBuf->d->buf[0] offset: 0 atIndex: tessFactorBufferBinding]; } @@ -2070,7 +2097,7 @@ void QRhiMetal::tessellatedDraw(const TessDrawArgs &args) quint32 inControlPointCount; quint32 patchCount; } params; - tescParamsBuf = tess.acquireWorkBuffer(this, sizeof(params), QMetalGraphicsPipelineData::Tessellation::WorkBufType::HostVisible); + tescParamsBuf = extraBufMgr.acquireWorkBuffer(this, sizeof(params), QMetalGraphicsPipelineData::ExtraBufferManager::WorkBufType::HostVisible); if (!tescParamsBuf) return; params.inControlPointCount = tess.inControlPointCount; @@ -2139,6 +2166,39 @@ void QRhiMetal::tessellatedDraw(const TessDrawArgs &args) } } +void QRhiMetal::adjustForMultiViewDraw(quint32 *instanceCount, QRhiCommandBuffer *cb) +{ + QMetalCommandBuffer *cbD = QRHI_RES(QMetalCommandBuffer, cb); + const int multiViewCount = cbD->currentGraphicsPipeline->m_multiViewCount; + if (multiViewCount <= 1) + return; + + const QMap<int, int> &ebb(cbD->currentGraphicsPipeline->d->vs.nativeShaderInfo.extraBufferBindings); + const int viewMaskBufBinding = ebb.value(QShaderPrivate::MslMultiViewMaskBufferBinding, -1); + if (viewMaskBufBinding == -1) { + qWarning("No extra buffer for multiview in the vertex shader; was it built with --view-count specified?"); + return; + } + struct { + quint32 viewOffset; + quint32 viewCount; + } multiViewInfo; + multiViewInfo.viewOffset = 0; + multiViewInfo.viewCount = quint32(multiViewCount); + QMetalBuffer *buf = cbD->currentGraphicsPipeline->d->extraBufMgr.acquireWorkBuffer(this, sizeof(multiViewInfo), + QMetalGraphicsPipelineData::ExtraBufferManager::WorkBufType::HostVisible); + if (buf) { + id<MTLBuffer> mtlbuf = buf->d->buf[0]; + char *p = reinterpret_cast<char *>([mtlbuf contents]); + memcpy(p, &multiViewInfo, sizeof(multiViewInfo)); + [cbD->d->currentRenderPassEncoder setVertexBuffer: mtlbuf offset: 0 atIndex: viewMaskBufBinding]; + // The instance count is adjusted for layered rendering. The vertex shader is expected to contain something like: + // uint gl_ViewIndex = spvViewMask[0] + (gl_InstanceIndex - gl_BaseInstance) % spvViewMask[1]; + // where spvViewMask is the buffer with multiViewInfo passed in above. + *instanceCount *= multiViewCount; + } +} + void QRhiMetal::draw(QRhiCommandBuffer *cb, quint32 vertexCount, quint32 instanceCount, quint32 firstVertex, quint32 firstInstance) { @@ -2157,6 +2217,8 @@ void QRhiMetal::draw(QRhiCommandBuffer *cb, quint32 vertexCount, return; } + adjustForMultiViewDraw(&instanceCount, cb); + if (caps.baseVertexAndInstance) { [cbD->d->currentRenderPassEncoder drawPrimitives: cbD->currentGraphicsPipeline->d->primitiveType vertexStart: firstVertex vertexCount: vertexCount instanceCount: instanceCount baseInstance: firstInstance]; @@ -2195,6 +2257,8 @@ void QRhiMetal::drawIndexed(QRhiCommandBuffer *cb, quint32 indexCount, return; } + adjustForMultiViewDraw(&instanceCount, cb); + if (caps.baseVertexAndInstance) { [cbD->d->currentRenderPassEncoder drawIndexedPrimitives: cbD->currentGraphicsPipeline->d->primitiveType indexCount: indexCount @@ -2297,10 +2361,7 @@ QRhi::FrameOpResult QRhiMetal::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginF [d->captureScope beginScope]; - // Do not let the command buffer mess with the refcount of objects. We do - // have a proper render loop and will manage lifetimes similarly to other - // backends (Vulkan). - swapChainD->cbWrapper.d->cb = [d->cmdQueue commandBufferWithUnretainedReferences]; + swapChainD->cbWrapper.d->cb = d->newCommandBuffer(); QMetalRenderTargetData::ColorAtt colorAtt; if (swapChainD->samples > 1) { @@ -2312,6 +2373,7 @@ QRhi::FrameOpResult QRhiMetal::beginFrame(QRhiSwapChain *swapChain, QRhi::BeginF swapChainD->rtWrapper.d->fb.colorAtt[0] = colorAtt; swapChainD->rtWrapper.d->fb.dsTex = swapChainD->ds ? swapChainD->ds->d->tex : nil; + swapChainD->rtWrapper.d->fb.dsResolveTex = nil; swapChainD->rtWrapper.d->fb.hasStencil = swapChainD->ds ? true : false; swapChainD->rtWrapper.d->fb.depthNeedsStore = false; @@ -2331,31 +2393,89 @@ QRhi::FrameOpResult QRhiMetal::endFrame(QRhiSwapChain *swapChain, QRhi::EndFrame QMetalSwapChain *swapChainD = QRHI_RES(QMetalSwapChain, swapChain); Q_ASSERT(currentSwapChain == swapChainD); + // Keep strong reference to command buffer + id<MTLCommandBuffer> commandBuffer = swapChainD->cbWrapper.d->cb; + __block int thisFrameSlot = currentFrameSlot; - [swapChainD->cbWrapper.d->cb addCompletedHandler: ^(id<MTLCommandBuffer> cb) { + [commandBuffer addCompletedHandler: ^(id<MTLCommandBuffer> cb) { swapChainD->d->lastGpuTime[thisFrameSlot] += cb.GPUEndTime - cb.GPUStartTime; dispatch_semaphore_signal(swapChainD->d->sem[thisFrameSlot]); }]; - const bool needsPresent = !flags.testFlag(QRhi::SkipPresent); - const bool presentsWithTransaction = swapChainD->d->layer.presentsWithTransaction; - if (!presentsWithTransaction && needsPresent) { - // beginFrame-endFrame without a render pass inbetween means there is no drawable. - if (id<CAMetalDrawable> drawable = swapChainD->d->curDrawable) - [swapChainD->cbWrapper.d->cb presentDrawable: drawable]; - } - - [swapChainD->cbWrapper.d->cb commit]; +#ifdef QRHI_METAL_COMMAND_BUFFERS_WITH_UNRETAINED_REFERENCES + // When Metal API validation diagnostics is enabled in Xcode the texture is + // released before the command buffer is done with it. Manually keep it alive + // to work around this. + id<MTLTexture> drawableTexture = [swapChainD->d->curDrawable.texture retain]; + [commandBuffer addCompletedHandler:^(id<MTLCommandBuffer>) { + [drawableTexture release]; + }]; +#endif - if (presentsWithTransaction && needsPresent) { - // beginFrame-endFrame without a render pass inbetween means there is no drawable. + if (flags.testFlag(QRhi::SkipPresent)) { + // Just need to commit, that's it + [commandBuffer commit]; + } else { if (id<CAMetalDrawable> drawable = swapChainD->d->curDrawable) { - // The layer has presentsWithTransaction set to true to avoid flicker on resizing, - // so here it is important to follow what the Metal docs say when it comes to the - // issuing the present. - [swapChainD->cbWrapper.d->cb waitUntilScheduled]; - [drawable present]; + // Got something to present + if (swapChainD->d->layer.presentsWithTransaction) { + [commandBuffer commit]; + // Keep strong reference to Metal layer + auto *metalLayer = swapChainD->d->layer; + auto presentWithTransaction = ^{ + [commandBuffer waitUntilScheduled]; + // If the layer has been resized while we waited to be scheduled we bail out, + // as the drawable is no longer valid for the layer, and we'll get a follow-up + // display with the right size. We know we are on the main thread here, which + // means we can access the layer directly. We also know that the layer is valid, + // since the block keeps a strong reference to it, compared to the QRhiSwapChain + // that can go away under our feet by the time we're scheduled. + const auto surfaceSize = QSizeF::fromCGSize(metalLayer.bounds.size) * metalLayer.contentsScale; + const auto textureSize = QSizeF(drawable.texture.width, drawable.texture.height); + if (textureSize == surfaceSize) { + [drawable present]; + } else { + qCDebug(QRHI_LOG_INFO) << "Skipping" << drawable << "due to texture size" + << textureSize << "not matching surface size" << surfaceSize; + } + }; + + if (NSThread.currentThread == NSThread.mainThread) { + presentWithTransaction(); + } else { + auto *qtMetalLayer = qt_objc_cast<QMetalLayer*>(swapChainD->d->layer); + Q_ASSERT(qtMetalLayer); + // Let the main thread present the drawable from displayLayer + qtMetalLayer.mainThreadPresentation = presentWithTransaction; + } + } else { + // Keep strong reference to Metal layer so it's valid in the block + auto *qtMetalLayer = qt_objc_cast<QMetalLayer*>(swapChainD->d->layer); + [commandBuffer addScheduledHandler:^(id<MTLCommandBuffer>) { + if (qtMetalLayer) { + // The schedule handler comes in on the com.Metal.CompletionQueueDispatch + // thread, which means we might be racing against a display cycle on the + // main thread. If the displayLayer is already in progress, we don't want + // to step on its toes. + if (qtMetalLayer.displayLock.tryLockForRead()) { + [drawable present]; + qtMetalLayer.displayLock.unlock(); + } else { + qCDebug(QRHI_LOG_INFO) << "Skipping" << drawable + << "due to" << qtMetalLayer << "needing display"; + } + } else { + [drawable present]; + } + }]; + [commandBuffer commit]; + } + } else { + // Still need to commit, even if we don't have a drawable + [commandBuffer commit]; } + + swapChainD->currentFrameSlot = (swapChainD->currentFrameSlot + 1) % QMTL_FRAMES_IN_FLIGHT; } // Must not hold on to the drawable, regardless of needsPresent @@ -2364,9 +2484,6 @@ QRhi::FrameOpResult QRhiMetal::endFrame(QRhiSwapChain *swapChain, QRhi::EndFrame [d->captureScope endScope]; - if (needsPresent) - swapChainD->currentFrameSlot = (swapChainD->currentFrameSlot + 1) % QMTL_FRAMES_IN_FLIGHT; - swapChainD->frameCount += 1; currentSwapChain = nullptr; return QRhi::FrameOpSuccess; @@ -2383,7 +2500,7 @@ QRhi::FrameOpResult QRhiMetal::beginOffscreenFrame(QRhiCommandBuffer **cb, QRhi: d->ofr.active = true; *cb = &d->ofr.cbWrapper; - d->ofr.cbWrapper.d->cb = [d->cmdQueue commandBufferWithUnretainedReferences]; + d->ofr.cbWrapper.d->cb = d->newCommandBuffer(); executeDeferredReleases(); d->ofr.cbWrapper.resetState(d->ofr.lastGpuTime); @@ -2448,10 +2565,10 @@ QRhi::FrameOpResult QRhiMetal::finish() if (inFrame) { if (d->ofr.active) { d->ofr.lastGpuTime += cb.GPUEndTime - cb.GPUStartTime; - d->ofr.cbWrapper.d->cb = [d->cmdQueue commandBufferWithUnretainedReferences]; + d->ofr.cbWrapper.d->cb = d->newCommandBuffer(); } else { swapChainD->d->lastGpuTime[currentFrameSlot] += cb.GPUEndTime - cb.GPUStartTime; - swapChainD->cbWrapper.d->cb = [d->cmdQueue commandBufferWithUnretainedReferences]; + swapChainD->cbWrapper.d->cb = d->newCommandBuffer(); } } @@ -2514,7 +2631,6 @@ void QRhiMetal::enqueueSubresUpload(QMetalTexture *texD, void *mp, void *blitEnc int w = img.width(); int h = img.height(); int bpl = img.bytesPerLine(); - int srcOffset = 0; if (!subresDesc.sourceSize().isEmpty() || !subresDesc.sourceTopLeft().isNull()) { const int sx = subresDesc.sourceTopLeft().x(); @@ -2523,10 +2639,12 @@ void QRhiMetal::enqueueSubresUpload(QMetalTexture *texD, void *mp, void *blitEnc w = subresDesc.sourceSize().width(); h = subresDesc.sourceSize().height(); } - if (img.depth() == 32) { - memcpy(reinterpret_cast<char *>(mp) + *curOfs, img.constBits(), size_t(fullImageSizeBytes)); - srcOffset = sy * bpl + sx * 4; - // bpl remains set to the original image's row stride + if (w == img.width()) { + const int bpc = qMax(1, img.depth() / 8); + Q_ASSERT(h * img.bytesPerLine() <= fullImageSizeBytes); + memcpy(reinterpret_cast<char *>(mp) + *curOfs, + img.constBits() + sy * img.bytesPerLine() + sx * bpc, + h * img.bytesPerLine()); } else { img = img.copy(sx, sy, w, h); bpl = img.bytesPerLine(); @@ -2538,7 +2656,7 @@ void QRhiMetal::enqueueSubresUpload(QMetalTexture *texD, void *mp, void *blitEnc } [blitEnc copyFromBuffer: texD->d->stagingBuf[currentFrameSlot] - sourceOffset: NSUInteger(*curOfs + srcOffset) + sourceOffset: NSUInteger(*curOfs) sourceBytesPerRow: NSUInteger(bpl) sourceBytesPerImage: 0 sourceSize: MTLSizeMake(NSUInteger(w), NSUInteger(h), 1) @@ -2630,6 +2748,15 @@ void QRhiMetal::enqueueResourceUpdates(QRhiCommandBuffer *cb, QRhiResourceUpdate QMetalCommandBuffer *cbD = QRHI_RES(QMetalCommandBuffer, cb); QRhiResourceUpdateBatchPrivate *ud = QRhiResourceUpdateBatchPrivate::get(resourceUpdates); + id<MTLBlitCommandEncoder> blitEnc = nil; + auto ensureBlit = [&blitEnc, cbD, this]() { + if (!blitEnc) { + blitEnc = [cbD->d->cb blitCommandEncoder]; + if (debugMarkers) + [blitEnc pushDebugGroup: @"Texture upload/copy"]; + } + }; + for (int opIdx = 0; opIdx < ud->activeBufferOpCount; ++opIdx) { const QRhiResourceUpdateBatchPrivate::BufferOp &u(ud->bufferOps[opIdx]); if (u.type == QRhiResourceUpdateBatchPrivate::BufferOp::DynamicUpdate) { @@ -2668,19 +2795,17 @@ void QRhiMetal::enqueueResourceUpdates(QRhiCommandBuffer *cb, QRhiResourceUpdate readback.readSize = u.readSize; readback.result = u.result; d->activeBufferReadbacks.append(readback); +#ifdef Q_OS_MACOS + if (bufD->d->managed) { + // On non-Apple Silicon, manually synchronize memory from GPU to CPU + ensureBlit(); + [blitEnc synchronizeResource:readback.buf]; + } +#endif } } } - id<MTLBlitCommandEncoder> blitEnc = nil; - auto ensureBlit = [&blitEnc, cbD, this] { - if (!blitEnc) { - blitEnc = [cbD->d->cb blitCommandEncoder]; - if (debugMarkers) - [blitEnc pushDebugGroup: @"Texture upload/copy"]; - } - }; - for (int opIdx = 0; opIdx < ud->activeTextureOpCount; ++opIdx) { const QRhiResourceUpdateBatchPrivate::TextureOp &u(ud->textureOps[opIdx]); if (u.type == QRhiResourceUpdateBatchPrivate::TextureOp::Upload) { @@ -2891,28 +3016,39 @@ void QRhiMetal::beginPass(QRhiCommandBuffer *cb, if (!QRhiRenderTargetAttachmentTracker::isUpToDate<QMetalTexture, QMetalRenderBuffer>(rtTex->description(), rtD->currentResIdList)) rtTex->create(); cbD->d->currentPassRpDesc = d->createDefaultRenderPass(rtD->dsAttCount, colorClearValue, depthStencilClearValue, rtD->colorAttCount); - if (rtTex->m_flags.testFlag(QRhiTextureRenderTarget::PreserveColorContents)) { + if (rtD->fb.preserveColor) { for (uint i = 0; i < uint(rtD->colorAttCount); ++i) cbD->d->currentPassRpDesc.colorAttachments[i].loadAction = MTLLoadActionLoad; } - if (rtD->dsAttCount && rtTex->m_flags.testFlag(QRhiTextureRenderTarget::PreserveDepthStencilContents)) { + if (rtD->dsAttCount && rtD->fb.preserveDs) { cbD->d->currentPassRpDesc.depthAttachment.loadAction = MTLLoadActionLoad; cbD->d->currentPassRpDesc.stencilAttachment.loadAction = MTLLoadActionLoad; } + int colorAttCount = 0; for (auto it = rtTex->m_desc.cbeginColorAttachments(), itEnd = rtTex->m_desc.cendColorAttachments(); it != itEnd; ++it) { - if (it->texture()) + colorAttCount += 1; + if (it->texture()) { QRHI_RES(QMetalTexture, it->texture())->lastActiveFrameSlot = currentFrameSlot; - else if (it->renderBuffer()) + if (it->multiViewCount() >= 2) + cbD->d->currentPassRpDesc.renderTargetArrayLength = NSUInteger(it->multiViewCount()); + } else if (it->renderBuffer()) { QRHI_RES(QMetalRenderBuffer, it->renderBuffer())->lastActiveFrameSlot = currentFrameSlot; + } if (it->resolveTexture()) QRHI_RES(QMetalTexture, it->resolveTexture())->lastActiveFrameSlot = currentFrameSlot; } if (rtTex->m_desc.depthStencilBuffer()) QRHI_RES(QMetalRenderBuffer, rtTex->m_desc.depthStencilBuffer())->lastActiveFrameSlot = currentFrameSlot; - if (rtTex->m_desc.depthTexture()) - QRHI_RES(QMetalTexture, rtTex->m_desc.depthTexture())->lastActiveFrameSlot = currentFrameSlot; + if (rtTex->m_desc.depthTexture()) { + QMetalTexture *depthTexture = QRHI_RES(QMetalTexture, rtTex->m_desc.depthTexture()); + depthTexture->lastActiveFrameSlot = currentFrameSlot; + if (colorAttCount == 0 && depthTexture->arraySize() >= 2) + cbD->d->currentPassRpDesc.renderTargetArrayLength = NSUInteger(depthTexture->arraySize()); + } + if (rtTex->m_desc.depthResolveTexture()) + QRHI_RES(QMetalTexture, rtTex->m_desc.depthResolveTexture())->lastActiveFrameSlot = currentFrameSlot; } break; default: @@ -2926,7 +3062,8 @@ void QRhiMetal::beginPass(QRhiCommandBuffer *cb, cbD->d->currentPassRpDesc.colorAttachments[i].depthPlane = NSUInteger(rtD->fb.colorAtt[i].slice); cbD->d->currentPassRpDesc.colorAttachments[i].level = NSUInteger(rtD->fb.colorAtt[i].level); if (rtD->fb.colorAtt[i].resolveTex) { - cbD->d->currentPassRpDesc.colorAttachments[i].storeAction = MTLStoreActionMultisampleResolve; + cbD->d->currentPassRpDesc.colorAttachments[i].storeAction = rtD->fb.preserveColor ? MTLStoreActionStoreAndMultisampleResolve + : MTLStoreActionMultisampleResolve; cbD->d->currentPassRpDesc.colorAttachments[i].resolveTexture = rtD->fb.colorAtt[i].resolveTex; cbD->d->currentPassRpDesc.colorAttachments[i].resolveSlice = NSUInteger(rtD->fb.colorAtt[i].resolveLayer); cbD->d->currentPassRpDesc.colorAttachments[i].resolveLevel = NSUInteger(rtD->fb.colorAtt[i].resolveLevel); @@ -2939,6 +3076,15 @@ void QRhiMetal::beginPass(QRhiCommandBuffer *cb, cbD->d->currentPassRpDesc.stencilAttachment.texture = rtD->fb.hasStencil ? rtD->fb.dsTex : nil; if (rtD->fb.depthNeedsStore) // Depth/Stencil is set to DontCare by default, override if needed cbD->d->currentPassRpDesc.depthAttachment.storeAction = MTLStoreActionStore; + if (rtD->fb.dsResolveTex) { + cbD->d->currentPassRpDesc.depthAttachment.storeAction = rtD->fb.depthNeedsStore ? MTLStoreActionStoreAndMultisampleResolve + : MTLStoreActionMultisampleResolve; + cbD->d->currentPassRpDesc.depthAttachment.resolveTexture = rtD->fb.dsResolveTex; + if (rtD->fb.hasStencil) { + cbD->d->currentPassRpDesc.stencilAttachment.resolveTexture = rtD->fb.dsResolveTex; + cbD->d->currentPassRpDesc.stencilAttachment.storeAction = cbD->d->currentPassRpDesc.depthAttachment.storeAction; + } + } } cbD->d->currentRenderPassEncoder = [cbD->d->cb renderCommandEncoderWithDescriptor: cbD->d->currentPassRpDesc]; @@ -3748,11 +3894,11 @@ bool QMetalTexture::create() desc.textureType = isArray ? MTLTextureType1DArray : MTLTextureType1D; } else if (isArray) { #ifdef Q_OS_IOS - if (samples > 1) { - // would be available on iOS 14.0+ but cannot test for that with a 13 SDK - qWarning("Multisample 2D texture array is not supported on iOS"); + if (@available(iOS 14, *)) { + desc.textureType = samples > 1 ? MTLTextureType2DMultisampleArray : MTLTextureType2DArray; + } else { + desc.textureType = MTLTextureType2DArray; } - desc.textureType = MTLTextureType2DArray; #else desc.textureType = samples > 1 ? MTLTextureType2DMultisampleArray : MTLTextureType2DArray; #endif @@ -4161,8 +4307,9 @@ bool QMetalTextureRenderTarget::create() if (m_desc.depthTexture()) { QMetalTexture *depthTexD = QRHI_RES(QMetalTexture, m_desc.depthTexture()); d->fb.dsTex = depthTexD->d->tex; - d->fb.hasStencil = false; - d->fb.depthNeedsStore = true; + d->fb.hasStencil = rhiD->isStencilSupportingFormat(depthTexD->format()); + d->fb.depthNeedsStore = !m_flags.testFlag(DoNotStoreDepthStencilContents) && !m_desc.depthResolveTexture(); + d->fb.preserveDs = m_flags.testFlag(QRhiTextureRenderTarget::PreserveDepthStencilContents); if (d->colorAttCount == 0) { d->pixelSize = depthTexD->pixelSize(); d->sampleCount = depthTexD->samples; @@ -4172,16 +4319,24 @@ bool QMetalTextureRenderTarget::create() d->fb.dsTex = depthRbD->d->tex; d->fb.hasStencil = true; d->fb.depthNeedsStore = false; + d->fb.preserveDs = false; if (d->colorAttCount == 0) { d->pixelSize = depthRbD->pixelSize(); d->sampleCount = depthRbD->samples; } } + if (m_desc.depthResolveTexture()) { + QMetalTexture *depthResolveTexD = QRHI_RES(QMetalTexture, m_desc.depthResolveTexture()); + d->fb.dsResolveTex = depthResolveTexD->d->tex; + } d->dsAttCount = 1; } else { d->dsAttCount = 0; } + if (d->colorAttCount > 0) + d->fb.preserveColor = m_flags.testFlag(QRhiTextureRenderTarget::PreserveColorContents); + QRhiRenderTargetAttachmentTracker::updateResIdList<QMetalTexture, QMetalRenderBuffer>(m_desc, &d->currentResIdList); rhiD->registerResource(this, false); @@ -4293,10 +4448,10 @@ void QMetalGraphicsPipeline::destroy() d->tess.compTesc.destroy(); d->tess.vertTese.destroy(); - qDeleteAll(d->tess.deviceLocalWorkBuffers); - d->tess.deviceLocalWorkBuffers.clear(); - qDeleteAll(d->tess.hostVisibleWorkBuffers); - d->tess.hostVisibleWorkBuffers.clear(); + qDeleteAll(d->extraBufMgr.deviceLocalWorkBuffers); + d->extraBufMgr.deviceLocalWorkBuffers.clear(); + qDeleteAll(d->extraBufMgr.hostVisibleWorkBuffers); + d->extraBufMgr.hostVisibleWorkBuffers.clear(); delete d->bufferSizeBuffer; d->bufferSizeBuffer = nullptr; @@ -4368,6 +4523,22 @@ static inline MTLVertexFormat toMetalAttributeFormat(QRhiVertexInputAttribute::F return MTLVertexFormatHalf2; case QRhiVertexInputAttribute::Half: return MTLVertexFormatHalf; + case QRhiVertexInputAttribute::UShort4: + return MTLVertexFormatUShort4; + case QRhiVertexInputAttribute::UShort3: + return MTLVertexFormatUShort3; + case QRhiVertexInputAttribute::UShort2: + return MTLVertexFormatUShort2; + case QRhiVertexInputAttribute::UShort: + return MTLVertexFormatUShort; + case QRhiVertexInputAttribute::SShort4: + return MTLVertexFormatShort4; + case QRhiVertexInputAttribute::SShort3: + return MTLVertexFormatShort3; + case QRhiVertexInputAttribute::SShort2: + return MTLVertexFormatShort2; + case QRhiVertexInputAttribute::SShort: + return MTLVertexFormatShort; default: Q_UNREACHABLE(); return MTLVertexFormatFloat4; @@ -4523,6 +4694,24 @@ static inline MTLPrimitiveType toMetalPrimitiveType(QRhiGraphicsPipeline::Topolo } } +static inline MTLPrimitiveTopologyClass toMetalPrimitiveTopologyClass(QRhiGraphicsPipeline::Topology t) +{ + switch (t) { + case QRhiGraphicsPipeline::Triangles: + case QRhiGraphicsPipeline::TriangleStrip: + case QRhiGraphicsPipeline::TriangleFan: + return MTLPrimitiveTopologyClassTriangle; + case QRhiGraphicsPipeline::Lines: + case QRhiGraphicsPipeline::LineStrip: + return MTLPrimitiveTopologyClassLine; + case QRhiGraphicsPipeline::Points: + return MTLPrimitiveTopologyClassPoint; + default: + Q_UNREACHABLE(); + return MTLPrimitiveTopologyClassTriangle; + } +} + static inline MTLCullMode toMetalCullMode(QRhiGraphicsPipeline::CullMode c) { switch (c) { @@ -4710,7 +4899,7 @@ void QMetalGraphicsPipeline::setupAttachmentsInMetalRenderPassDescriptor(void *m } QRHI_RES_RHI(QRhiMetal); - rpDesc.sampleCount = NSUInteger(rhiD->effectiveSampleCount(m_sampleCount)); + rpDesc.rasterSampleCount = NSUInteger(rhiD->effectiveSampleCount(m_sampleCount)); } void QMetalGraphicsPipeline::setupMetalDepthStencilDescriptor(void *metalDsDesc) @@ -4763,6 +4952,7 @@ void QMetalGraphicsPipelineData::setupVertexInputDescriptor(MTLVertexDescriptor desc.attributes[loc].bufferIndex = NSUInteger(firstVertexBinding + it->binding()); } int bindingIndex = 0; + const NSUInteger viewCount = qMax<NSUInteger>(1, q->multiViewCount()); for (auto it = vertexInputLayout.cbeginBindings(), itEnd = vertexInputLayout.cendBindings(); it != itEnd; ++it, ++bindingIndex) { @@ -4771,6 +4961,8 @@ void QMetalGraphicsPipelineData::setupVertexInputDescriptor(MTLVertexDescriptor it->classification() == QRhiVertexInputBinding::PerInstance ? MTLVertexStepFunctionPerInstance : MTLVertexStepFunctionPerVertex; desc.layouts[layoutIdx].stepRate = NSUInteger(it->instanceStepRate()); + if (desc.layouts[layoutIdx].stepFunction == MTLVertexStepFunctionPerInstance) + desc.layouts[layoutIdx].stepRate *= viewCount; desc.layouts[layoutIdx].stride = it->stride(); } } @@ -4819,33 +5011,10 @@ void QRhiMetalData::trySeedingRenderPipelineFromBinaryArchive(MTLRenderPipelineD } } -static bool canAddToBinaryArchive(QRhiMetalData *d) -{ - if (@available(macOS 11.0, iOS 14.0, *)) { - if (!d->binArch) - return false; - - // ### QTBUG-106703, QTBUG-108216, revisit after 13.0 - if (!d->binArchWasEmpty && d->q->osMajor >= 13) { - static bool logPrinted = false; - if (!logPrinted) { - logPrinted = true; - qCDebug(QRHI_LOG_INFO, "Skipping adding more pipelines to MTLBinaryArchive on this OS version (%d.%d) due to known issues.", - d->q->osMajor, d->q->osMinor); - } - return false; - } - - return true; - } else { - return false; - } -} - void QRhiMetalData::addRenderPipelineToBinaryArchive(MTLRenderPipelineDescriptor *rpDesc) { if (@available(macOS 11.0, iOS 14.0, *)) { - if (canAddToBinaryArchive(this)) { + if (binArch) { NSError *err = nil; if (![binArch addRenderPipelineFunctionsWithDescriptor: rpDesc error: &err]) { const QString msg = QString::fromNSString(err.localizedDescription); @@ -4947,6 +5116,9 @@ bool QMetalGraphicsPipeline::createVertexFragmentPipeline() QMetalRenderPassDescriptor *rpD = QRHI_RES(QMetalRenderPassDescriptor, m_renderPassDesc); setupAttachmentsInMetalRenderPassDescriptor(rpDesc, rpD); + if (m_multiViewCount >= 2) + rpDesc.inputPrimitiveTopology = toMetalPrimitiveTopologyClass(m_topology); + rhiD->d->trySeedingRenderPipelineFromBinaryArchive(rpDesc); if (rhiD->rhiFlags.testFlag(QRhi::EnablePipelineCacheDataSave)) @@ -5469,7 +5641,7 @@ id<MTLRenderPipelineState> QMetalGraphicsPipelineData::Tessellation::teseFragRen return ps; } -QMetalBuffer *QMetalGraphicsPipelineData::Tessellation::acquireWorkBuffer(QRhiMetal *rhiD, quint32 size, WorkBufType type) +QMetalBuffer *QMetalGraphicsPipelineData::ExtraBufferManager::acquireWorkBuffer(QRhiMetal *rhiD, quint32 size, WorkBufType type) { QVector<QMetalBuffer *> *workBuffers = type == WorkBufType::DeviceLocal ? &deviceLocalWorkBuffers : &hostVisibleWorkBuffers; @@ -5535,6 +5707,9 @@ bool QMetalGraphicsPipeline::createTessellationPipelines(const QShader &tessVert return false; } + if (m_multiViewCount >= 2) + qWarning("Multiview is not supported with tessellation"); + // Now the vertex shader is a compute shader. // It should have three dedicated *VertexAsComputeShader variants. // What the requested variant was (Standard or Batchable) plays no role here. @@ -5825,7 +6000,7 @@ void QRhiMetalData::trySeedingComputePipelineFromBinaryArchive(MTLComputePipelin void QRhiMetalData::addComputePipelineToBinaryArchive(MTLComputePipelineDescriptor *cpDesc) { if (@available(macOS 11.0, iOS 14.0, *)) { - if (canAddToBinaryArchive(this)) { + if (binArch) { NSError *err = nil; if (![binArch addComputePipelineFunctionsWithDescriptor: cpDesc error: &err]) { const QString msg = QString::fromNSString(err.localizedDescription); @@ -6035,13 +6210,8 @@ void QMetalSwapChain::destroy() d->msaaTex[i] = nil; } -#ifdef Q_OS_MACOS - d->liveResizeStartObserver.remove(); - d->liveResizeEndObserver.remove(); - d->liveResizeObserverSet = false; -#endif - d->layer = nullptr; + m_proxyData = {}; [d->curDrawable release]; d->curDrawable = nil; @@ -6105,9 +6275,17 @@ QSize QMetalSwapChain::surfacePixelSize() bool QMetalSwapChain::isFormatSupported(Format f) { -#ifdef Q_OS_MACOS - return f == SDR || f == HDRExtendedSrgbLinear; -#endif + if (f == HDRExtendedSrgbLinear) { + if (@available(macOS 10.11, iOS 16.0, *)) + return hdrInfo().limits.colorComponentValue.maxPotentialColorComponentValue > 1.0f; + else + return false; + } else if (f == HDRExtendedDisplayP3Linear) { + if (@available(macOS 11.0, iOS 14.0, *)) + return hdrInfo().limits.colorComponentValue.maxPotentialColorComponentValue > 1.0f; + else + return false; + } return f == SDR; } @@ -6142,7 +6320,7 @@ void QMetalSwapChain::chooseFormats() QRHI_RES_RHI(QRhiMetal); samples = rhiD->effectiveSampleCount(m_sampleCount); // pick a format that is allowed for CAMetalLayer.pixelFormat - if (m_format == HDRExtendedSrgbLinear) { + if (m_format == HDRExtendedSrgbLinear || m_format == HDRExtendedDisplayP3Linear) { d->colorFormat = MTLPixelFormatRGBA16Float; d->rhiColorFormat = QRhiTexture::RGBA16F; return; @@ -6189,13 +6367,18 @@ bool QMetalSwapChain::createOrResize() chooseFormats(); if (d->colorFormat != d->layer.pixelFormat) d->layer.pixelFormat = d->colorFormat; -#ifdef Q_OS_MACOS - // Can't enable this on iOS until wantsExtendedDynamicRangeContent is available + if (m_format == HDRExtendedSrgbLinear) { - d->layer.colorspace = CGColorSpaceCreateWithName(kCGColorSpaceExtendedLinearSRGB); - d->layer.wantsExtendedDynamicRangeContent = YES; + if (@available(macOS 10.11, iOS 16.0, *)) { + d->layer.colorspace = CGColorSpaceCreateWithName(kCGColorSpaceExtendedLinearSRGB); + d->layer.wantsExtendedDynamicRangeContent = YES; + } + } else if (m_format == HDRExtendedDisplayP3Linear) { + if (@available(macOS 11.0, iOS 16.0, *)) { + d->layer.colorspace = CGColorSpaceCreateWithName(kCGColorSpaceExtendedLinearDisplayP3); + d->layer.wantsExtendedDynamicRangeContent = YES; + } } -#endif if (m_flags.testFlag(UsedAsTransferSource)) d->layer.framebufferOnly = NO; @@ -6233,34 +6416,6 @@ bool QMetalSwapChain::createOrResize() [d->layer setDevice: rhiD->d->dev]; -#ifdef Q_OS_MACOS - // Can only use presentsWithTransaction (to get smooth resizing) when - // presenting from the main (gui) thread. We predict that based on the - // thread this function is called on since if the QRhiSwapChain is - // initialied on a given thread then that's almost certainly the thread on - // which the QRhi renders and presents. - const bool canUsePresentsWithTransaction = NSThread.isMainThread; - - // Have an env.var. just in case it turns out presentsWithTransaction is - // not desired in some specific case. - static bool allowPresentsWithTransaction = !qEnvironmentVariableIntValue("QT_MTL_NO_TRANSACTION"); - - if (allowPresentsWithTransaction && canUsePresentsWithTransaction && !d->liveResizeObserverSet) { - d->liveResizeObserverSet = true; - NSView *view = reinterpret_cast<NSView *>(window->winId()); - NSWindow *window = view.window; - if (window) { - qCDebug(QRHI_LOG_INFO, "will set presentsWithTransaction during live resize"); - d->liveResizeStartObserver = QMacNotificationObserver(window, NSWindowWillStartLiveResizeNotification, [this] { - d->layer.presentsWithTransaction = true; - }); - d->liveResizeEndObserver = QMacNotificationObserver(window, NSWindowDidEndLiveResizeNotification, [this] { - d->layer.presentsWithTransaction = false; - }); - } - } -#endif - [d->curDrawable release]; d->curDrawable = nil; @@ -6328,22 +6483,28 @@ QRhiSwapChainHdrInfo QMetalSwapChain::hdrInfo() { QRhiSwapChainHdrInfo info; info.limitsType = QRhiSwapChainHdrInfo::ColorComponentValue; - if (m_format == SDR) { - info.limits.colorComponentValue.maxColorComponentValue = 1; - return info; - } + info.limits.colorComponentValue.maxColorComponentValue = 1; + info.limits.colorComponentValue.maxPotentialColorComponentValue = 1; + info.luminanceBehavior = QRhiSwapChainHdrInfo::DisplayReferred; // 1.0 = SDR white + info.sdrWhiteLevel = 200; // typical value, but dummy (don't know the real one); won't matter due to being display-referred -#ifdef Q_OS_MACOS - info.isHardCodedDefaults = false; - // Must use m_window, not window, given this may be called before createOrResize(). - NSView *view = reinterpret_cast<NSView *>(m_window->winId()); - info.limits.colorComponentValue.maxColorComponentValue = view.window.screen.maximumExtendedDynamicRangeColorComponentValue; -#else - // ### Fixme: Maybe retrieve the brightness from the screen and if we're not at full brightness we might be able to do more. - // For now, assume 2, in line with iPhone 12 specs that claim 625 nits max brightness and 1200 nits max HDR brightness. - info.isHardCodedDefaults = true; - info.limits.colorComponentValue.maxColorComponentValue = 2; + if (m_window) { + // Must use m_window, not window, given this may be called before createOrResize(). +#if defined(Q_OS_MACOS) + NSView *view = reinterpret_cast<NSView *>(m_window->winId()); + NSScreen *screen = view.window.screen; + info.limits.colorComponentValue.maxColorComponentValue = screen.maximumExtendedDynamicRangeColorComponentValue; + info.limits.colorComponentValue.maxPotentialColorComponentValue = screen.maximumPotentialExtendedDynamicRangeColorComponentValue; +#elif defined(Q_OS_IOS) + if (@available(iOS 16.0, *)) { + UIView *view = reinterpret_cast<UIView *>(m_window->winId()); + UIScreen *screen = view.window.windowScene.screen; + info.limits.colorComponentValue.maxColorComponentValue = view.window.windowScene.screen.currentEDRHeadroom; + info.limits.colorComponentValue.maxPotentialColorComponentValue = screen.potentialEDRHeadroom; + } #endif + } + return info; } diff --git a/src/gui/rhi/qrhimetal_p.h b/src/gui/rhi/qrhimetal_p.h index 8fb2ce84b0..f539148b2c 100644 --- a/src/gui/rhi/qrhimetal_p.h +++ b/src/gui/rhi/qrhimetal_p.h @@ -454,7 +454,6 @@ public: const QRhiCommandBuffer::DynamicOffset *dynamicOffsets, bool offsetOnlyChange, const QShader::NativeResourceBindingMap *nativeResourceBindingMaps[SUPPORTED_STAGES]); - int effectiveSampleCount(int sampleCount) const; struct TessDrawArgs { QMetalCommandBuffer *cbD; enum { @@ -482,6 +481,7 @@ public: }; }; void tessellatedDraw(const TessDrawArgs &args); + void adjustForMultiViewDraw(quint32 *instanceCount, QRhiCommandBuffer *cb); QRhi::Flags rhiFlags; bool importedDevice = false; @@ -499,6 +499,7 @@ public: QVector<int> supportedSampleCounts; bool isAppleGPU = false; int maxThreadGroupSize = 512; + bool multiView = false; } caps; QRhiMetalData *d = nullptr; diff --git a/src/gui/rhi/qrhinull.cpp b/src/gui/rhi/qrhinull.cpp index 95106bfeb8..566b922c1b 100644 --- a/src/gui/rhi/qrhinull.cpp +++ b/src/gui/rhi/qrhinull.cpp @@ -13,7 +13,7 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief Null backend specific initialization parameters. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. A Null QRhi needs no special parameters for initialization. @@ -35,7 +35,7 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief Empty. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ diff --git a/src/gui/rhi/qrhivulkan.cpp b/src/gui/rhi/qrhivulkan.cpp index a78b459dd6..3dd3c57bd4 100644 --- a/src/gui/rhi/qrhivulkan.cpp +++ b/src/gui/rhi/qrhivulkan.cpp @@ -2,7 +2,6 @@ // SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only #include "qrhivulkan_p.h" -#include "qrhivulkanext_p.h" #include <qpa/qplatformvulkaninstance.h> #define VMA_IMPLEMENTATION @@ -64,7 +63,7 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief Vulkan specific initialization parameters. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. A Vulkan-based QRhi needs at minimum a valid QVulkanInstance. It is up to @@ -196,7 +195,7 @@ QT_BEGIN_NAMESPACE \note Ownership of the Vulkan objects is never transferred. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -258,7 +257,7 @@ QT_BEGIN_NAMESPACE \l{QRhi::beginOffscreenFrame()}{beginOffscreenFrame()} - \l{QRhi::endOffscreenFrame()}{endOffscreenFrame()} pair. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -274,7 +273,7 @@ QT_BEGIN_NAMESPACE \since 6.6 \brief Holds the Vulkan render pass object backing a QRhiRenderPassDescriptor. - \note This a RHI API with limited compatibility guarantees, see \l QRhi + \note This is a RHI API with limited compatibility guarantees, see \l QRhi for details. */ @@ -335,7 +334,9 @@ QByteArrayList QRhiVulkanInitParams::preferredExtensionsForImportedDevice() { return { QByteArrayLiteral("VK_KHR_swapchain"), - QByteArrayLiteral("VK_EXT_vertex_attribute_divisor") + QByteArrayLiteral("VK_EXT_vertex_attribute_divisor"), + QByteArrayLiteral("VK_KHR_create_renderpass2"), + QByteArrayLiteral("VK_KHR_depth_stencil_resolve") }; } @@ -429,6 +430,8 @@ bool QRhiVulkan::create(QRhi::Flags flags) for (const char *ext : inst->extensions()) qCDebug(QRHI_LOG_INFO, " %s", ext); } + + caps = {}; caps.debugUtils = inst->extensions().contains(QByteArrayLiteral("VK_EXT_debug_utils")); QList<VkQueueFamilyProperties> queueFamilyProps; @@ -530,7 +533,65 @@ bool QRhiVulkan::create(QRhi::Flags flags) driverInfoStruct.vendorId = physDevProperties.vendorID; driverInfoStruct.deviceType = toRhiDeviceType(physDevProperties.deviceType); - f->vkGetPhysicalDeviceFeatures(physDev, &physDevFeatures); + bool featuresQueried = false; +#ifdef VK_VERSION_1_1 + VkPhysicalDeviceFeatures2 physDevFeaturesChainable = {}; + physDevFeaturesChainable.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; +#endif + + // Vulkan >=1.2 headers at build time, >=1.2 implementation at run time +#ifdef VK_VERSION_1_2 + if (!featuresQueried) { + // Vulkan11Features, Vulkan12Features, etc. are only in Vulkan 1.2 and newer. + if (caps.apiVersion >= QVersionNumber(1, 2)) { + physDevFeatures11IfApi12OrNewer = {}; + physDevFeatures11IfApi12OrNewer.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES; + physDevFeatures12 = {}; + physDevFeatures12.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES; +#ifdef VK_VERSION_1_3 + physDevFeatures13 = {}; + physDevFeatures13.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_3_FEATURES; +#endif + physDevFeaturesChainable.pNext = &physDevFeatures11IfApi12OrNewer; + physDevFeatures11IfApi12OrNewer.pNext = &physDevFeatures12; +#ifdef VK_VERSION_1_3 + if (caps.apiVersion >= QVersionNumber(1, 3)) + physDevFeatures12.pNext = &physDevFeatures13; +#endif + f->vkGetPhysicalDeviceFeatures2(physDev, &physDevFeaturesChainable); + memcpy(&physDevFeatures, &physDevFeaturesChainable.features, sizeof(VkPhysicalDeviceFeatures)); + featuresQueried = true; + } + } +#endif // VK_VERSION_1_2 + + // Vulkan >=1.1 headers at build time, 1.1 implementation at run time +#ifdef VK_VERSION_1_1 + if (!featuresQueried) { + // Vulkan versioning nightmares: if the runtime API version is 1.1, + // there is no Vulkan11Features (introduced in 1.2+, the headers might + // have the types and structs, but the Vulkan implementation version at + // run time is what matters). But there are individual feature structs. + // For multiview, it is important to get this right since at the time of + // writing Quest 3 Android is a Vulkan 1.1 implementation at run time on + // the headset. + if (caps.apiVersion == QVersionNumber(1, 1)) { + multiviewFeaturesIfApi11 = {}; + multiviewFeaturesIfApi11.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES; + physDevFeaturesChainable.pNext = &multiviewFeaturesIfApi11; + f->vkGetPhysicalDeviceFeatures2(physDev, &physDevFeaturesChainable); + memcpy(&physDevFeatures, &physDevFeaturesChainable.features, sizeof(VkPhysicalDeviceFeatures)); + featuresQueried = true; + } + } +#endif + + if (!featuresQueried) { + // If the API version at run time is 1.0 (or we are building with + // ancient 1.0 headers), then do the Vulkan 1.0 query. + f->vkGetPhysicalDeviceFeatures(physDev, &physDevFeatures); + featuresQueried = true; + } // Choose queue and create device, unless the device was specified in importParams. if (!importedDevice) { @@ -602,13 +663,28 @@ bool QRhiVulkan::create(QRhi::Flags flags) } } - caps.vertexAttribDivisor = false; +#ifdef VK_EXT_vertex_attribute_divisor if (devExts.contains(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME)) { if (hasPhysDevProp2) { requestedDevExts.append(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME); caps.vertexAttribDivisor = true; } } +#endif + +#ifdef VK_KHR_create_renderpass2 + if (devExts.contains(VK_KHR_CREATE_RENDERPASS_2_EXTENSION_NAME)) { + requestedDevExts.append(VK_KHR_CREATE_RENDERPASS_2_EXTENSION_NAME); + caps.renderPass2KHR = true; + } +#endif + +#ifdef VK_KHR_depth_stencil_resolve + if (devExts.contains(VK_KHR_DEPTH_STENCIL_RESOLVE_EXTENSION_NAME)) { + requestedDevExts.append(VK_KHR_DEPTH_STENCIL_RESOLVE_EXTENSION_NAME); + caps.depthStencilResolveKHR = true; + } +#endif for (const QByteArray &ext : requestedDeviceExtensions) { if (!ext.isEmpty() && !requestedDevExts.contains(ext)) { @@ -661,42 +737,27 @@ bool QRhiVulkan::create(QRhi::Flags flags) // tessellationShader, geometryShader // textureCompressionETC2, textureCompressionASTC_LDR, textureCompressionBC -#ifdef VK_VERSION_1_2 // Vulkan11Features is only in Vulkan 1.2 - VkPhysicalDeviceFeatures2 physDevFeatures2 = {}; - physDevFeatures2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; - - VkPhysicalDeviceVulkan11Features features11 = {}; - features11.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES; - VkPhysicalDeviceVulkan12Features features12 = {}; - features12.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES; -#ifdef VK_VERSION_1_3 - VkPhysicalDeviceVulkan13Features features13 = {}; - features13.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_3_FEATURES; +#ifdef VK_VERSION_1_1 + physDevFeaturesChainable.features.robustBufferAccess = VK_FALSE; #endif - - if (caps.apiVersion >= QVersionNumber(1, 2)) { - physDevFeatures2.pNext = &features11; - features11.pNext = &features12; #ifdef VK_VERSION_1_3 - if (caps.apiVersion >= QVersionNumber(1, 3)) - features12.pNext = &features13; + physDevFeatures13.robustImageAccess = VK_FALSE; #endif - f->vkGetPhysicalDeviceFeatures2(physDev, &physDevFeatures2); - physDevFeatures2.features.robustBufferAccess = VK_FALSE; -#ifdef VK_VERSION_1_3 - features13.robustImageAccess = VK_FALSE; +#ifdef VK_VERSION_1_1 + if (caps.apiVersion >= QVersionNumber(1, 1)) { + // For a >=1.2 implementation at run time, this will enable all + // (1.0-1.3) features reported as supported, except the ones we turn + // off explicitly above. For a 1.1 implementation at run time, this + // only enables the 1.0 and multiview features reported as + // supported. We will not be bothering with the Vulkan 1.1 + // individual feature struct nonsense. + devInfo.pNext = &physDevFeaturesChainable; + } else #endif - - devInfo.pNext = &physDevFeatures2; - } -#endif // VK_VERSION_1_2 - - VkPhysicalDeviceFeatures features; - if (!devInfo.pNext) { - memcpy(&features, &physDevFeatures, sizeof(features)); - features.robustBufferAccess = VK_FALSE; - devInfo.pEnabledFeatures = &features; + { + physDevFeatures.robustBufferAccess = VK_FALSE; + devInfo.pEnabledFeatures = &physDevFeatures; } VkResult err = f->vkCreateDevice(physDev, &devInfo, nullptr, &dev); @@ -706,6 +767,13 @@ bool QRhiVulkan::create(QRhi::Flags flags) } } else { qCDebug(QRHI_LOG_INFO, "Using imported device %p", dev); + + // Here we have no way to tell if the extensions got enabled or not. + // Pretend it's all there and supported. If getProcAddress fails, we'll + // handle that gracefully. + caps.vertexAttribDivisor = true; + caps.renderPass2KHR = true; + caps.depthStencilResolveKHR = true; } vkGetPhysicalDeviceSurfaceCapabilitiesKHR = reinterpret_cast<PFN_vkGetPhysicalDeviceSurfaceCapabilitiesKHR>( @@ -753,6 +821,28 @@ bool QRhiVulkan::create(QRhi::Flags flags) caps.nonFillPolygonMode = physDevFeatures.fillModeNonSolid; +#ifdef VK_VERSION_1_2 + if (caps.apiVersion >= QVersionNumber(1, 2)) + caps.multiView = physDevFeatures11IfApi12OrNewer.multiview; +#endif + +#ifdef VK_VERSION_1_1 + if (caps.apiVersion == QVersionNumber(1, 1)) + caps.multiView = multiviewFeaturesIfApi11.multiview; +#endif + + // With Vulkan 1.2 renderpass2 and depth_stencil_resolve are core, but we + // have to support the case of 1.1 + extensions, in particular for the Quest + // 3 (Android, Vulkan 1.1 at the time of writing). Therefore, always rely on + // the KHR extension for now. +#ifdef VK_KHR_create_renderpass2 + if (caps.renderPass2KHR) { + vkCreateRenderPass2KHR = reinterpret_cast<PFN_vkCreateRenderPass2KHR>(f->vkGetDeviceProcAddr(dev, "vkCreateRenderPass2KHR")); + if (!vkCreateRenderPass2KHR) // handle it gracefully, the caps flag may be incorrect when using an imported VkDevice + caps.renderPass2KHR = false; + } +#endif + if (!importedAllocator) { VmaVulkanFunctions funcs = {}; funcs.vkGetInstanceProcAddr = wrap_vkGetInstanceProcAddr; @@ -766,12 +856,9 @@ bool QRhiVulkan::create(QRhi::Flags flags) allocatorInfo.device = dev; allocatorInfo.pVulkanFunctions = &funcs; allocatorInfo.instance = inst->vkInstance(); - const QVersionNumber apiVer = inst->apiVersion(); - if (!apiVer.isNull()) { - allocatorInfo.vulkanApiVersion = VK_MAKE_VERSION(apiVer.majorVersion(), - apiVer.minorVersion(), - apiVer.microVersion()); - } + allocatorInfo.vulkanApiVersion = VK_MAKE_VERSION(caps.apiVersion.majorVersion(), + caps.apiVersion.minorVersion(), + caps.apiVersion.microVersion()); VmaAllocator vmaallocator; VkResult err = vmaCreateAllocator(&allocatorInfo, &vmaallocator); if (err != VK_SUCCESS) { @@ -1280,6 +1367,8 @@ bool QRhiVulkan::createDefaultRenderPass(QVkRenderPassDescriptor *rpD, bool hasD rpD->colorRefs.append({ 0, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL }); rpD->hasDepthStencil = hasDepthStencil; + rpD->hasDepthStencilResolve = false; + rpD->multiViewCount = 0; if (hasDepthStencil) { // clear on load + no store + lazy alloc + transient image should play @@ -1349,28 +1438,190 @@ bool QRhiVulkan::createDefaultRenderPass(QVkRenderPassDescriptor *rpD, bool hasD return true; } +struct MultiViewRenderPassSetupHelper +{ + bool prepare(VkRenderPassCreateInfo *rpInfo, int multiViewCount, bool multiViewCap) + { + if (multiViewCount < 2) + return true; + if (!multiViewCap) { + qWarning("Cannot create multiview render pass without support for the Vulkan 1.1 multiview feature"); + return false; + } +#ifdef VK_VERSION_1_1 + uint32_t allViewsMask = 0; + for (uint32_t i = 0; i < uint32_t(multiViewCount); ++i) + allViewsMask |= (1 << i); + multiViewMask = allViewsMask; + multiViewCorrelationMask = allViewsMask; + multiViewInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO; + multiViewInfo.subpassCount = 1; + multiViewInfo.pViewMasks = &multiViewMask; + multiViewInfo.correlationMaskCount = 1; + multiViewInfo.pCorrelationMasks = &multiViewCorrelationMask; + rpInfo->pNext = &multiViewInfo; +#endif + return true; + } + +#ifdef VK_VERSION_1_1 + VkRenderPassMultiviewCreateInfo multiViewInfo = {}; + uint32_t multiViewMask = 0; + uint32_t multiViewCorrelationMask = 0; +#endif +}; + +#ifdef VK_KHR_create_renderpass2 +// Effectively converts a VkRenderPassCreateInfo into a VkRenderPassCreateInfo2, +// adding depth-stencil resolve support. Assumes a single subpass and no subpass +// dependencies. +struct RenderPass2SetupHelper +{ + bool prepare(VkRenderPassCreateInfo2 *rpInfo2, const VkRenderPassCreateInfo *rpInfo, const QVkRenderPassDescriptor *rpD, int multiViewCount) { + *rpInfo2 = {}; + + viewMask = 0; + if (multiViewCount >= 2) { + for (uint32_t i = 0; i < uint32_t(multiViewCount); ++i) + viewMask |= (1 << i); + } + + attDescs2.resize(rpInfo->attachmentCount); + for (qsizetype i = 0; i < attDescs2.count(); ++i) { + VkAttachmentDescription2KHR &att2(attDescs2[i]); + const VkAttachmentDescription &att(rpInfo->pAttachments[i]); + att2 = {}; + att2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2; + att2.flags = att.flags; + att2.format = att.format; + att2.samples = att.samples; + att2.loadOp = att.loadOp; + att2.storeOp = att.storeOp; + att2.stencilLoadOp = att.stencilLoadOp; + att2.stencilStoreOp = att.stencilStoreOp; + att2.initialLayout = att.initialLayout; + att2.finalLayout = att.finalLayout; + } + + attRefs2.clear(); + subpass2 = {}; + subpass2.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2_KHR; + const VkSubpassDescription &subpassDesc(rpInfo->pSubpasses[0]); + subpass2.flags = subpassDesc.flags; + subpass2.pipelineBindPoint = subpassDesc.pipelineBindPoint; + if (multiViewCount >= 2) + subpass2.viewMask = viewMask; + + // color attachment refs + qsizetype startIndex = attRefs2.count(); + for (uint32_t j = 0; j < subpassDesc.colorAttachmentCount; ++j) { + attRefs2.append({}); + VkAttachmentReference2KHR &attref2(attRefs2.last()); + const VkAttachmentReference &attref(subpassDesc.pColorAttachments[j]); + attref2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; + attref2.attachment = attref.attachment; + attref2.layout = attref.layout; + attref2.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + } + subpass2.colorAttachmentCount = subpassDesc.colorAttachmentCount; + subpass2.pColorAttachments = attRefs2.constData() + startIndex; + + // color resolve refs + if (subpassDesc.pResolveAttachments) { + startIndex = attRefs2.count(); + for (uint32_t j = 0; j < subpassDesc.colorAttachmentCount; ++j) { + attRefs2.append({}); + VkAttachmentReference2KHR &attref2(attRefs2.last()); + const VkAttachmentReference &attref(subpassDesc.pResolveAttachments[j]); + attref2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; + attref2.attachment = attref.attachment; + attref2.layout = attref.layout; + attref2.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + } + subpass2.pResolveAttachments = attRefs2.constData() + startIndex; + } + + // depth-stencil ref + if (subpassDesc.pDepthStencilAttachment) { + startIndex = attRefs2.count(); + attRefs2.append({}); + VkAttachmentReference2KHR &attref2(attRefs2.last()); + const VkAttachmentReference &attref(*subpassDesc.pDepthStencilAttachment); + attref2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; + attref2.attachment = attref.attachment; + attref2.layout = attref.layout; + attref2.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT; + subpass2.pDepthStencilAttachment = attRefs2.constData() + startIndex; + } + + // depth-stencil resolve ref +#ifdef VK_KHR_depth_stencil_resolve + dsResolveDesc = {}; + if (rpD->hasDepthStencilResolve) { + startIndex = attRefs2.count(); + attRefs2.append({}); + VkAttachmentReference2KHR &attref2(attRefs2.last()); + attref2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2_KHR; + attref2.attachment = rpD->dsResolveRef.attachment; + attref2.layout = rpD->dsResolveRef.layout; + attref2.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT; + dsResolveDesc.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE_KHR; + dsResolveDesc.depthResolveMode = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT; + dsResolveDesc.stencilResolveMode = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT; + dsResolveDesc.pDepthStencilResolveAttachment = attRefs2.constData() + startIndex; + subpass2.pNext = &dsResolveDesc; + } +#endif + + rpInfo2->sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR; + rpInfo2->pNext = nullptr; // the 1.1 VkRenderPassMultiviewCreateInfo is part of the '2' structs + rpInfo2->flags = rpInfo->flags; + rpInfo2->attachmentCount = rpInfo->attachmentCount; + rpInfo2->pAttachments = attDescs2.constData(); + rpInfo2->subpassCount = 1; + rpInfo2->pSubpasses = &subpass2; + if (multiViewCount >= 2) { + rpInfo2->correlatedViewMaskCount = 1; + rpInfo2->pCorrelatedViewMasks = &viewMask; + } + return true; + } + + QVarLengthArray<VkAttachmentDescription2KHR, 8> attDescs2; + QVarLengthArray<VkAttachmentReference2KHR, 8> attRefs2; + VkSubpassDescription2KHR subpass2; +#ifdef VK_KHR_depth_stencil_resolve + VkSubpassDescriptionDepthStencilResolveKHR dsResolveDesc; +#endif + uint32_t viewMask; +}; +#endif // VK_KHR_create_renderpass2 + bool QRhiVulkan::createOffscreenRenderPass(QVkRenderPassDescriptor *rpD, - const QRhiColorAttachment *firstColorAttachment, - const QRhiColorAttachment *lastColorAttachment, + const QRhiColorAttachment *colorAttachmentsBegin, + const QRhiColorAttachment *colorAttachmentsEnd, bool preserveColor, bool preserveDs, + bool storeDs, QRhiRenderBuffer *depthStencilBuffer, - QRhiTexture *depthTexture) + QRhiTexture *depthTexture, + QRhiTexture *depthResolveTexture) { - // attachment list layout is color (0-8), ds (0-1), resolve (0-8) + // attachment list layout is color (0-8), ds (0-1), resolve (0-8), ds resolve (0-1) - for (auto it = firstColorAttachment; it != lastColorAttachment; ++it) { + int multiViewCount = 0; + for (auto it = colorAttachmentsBegin; it != colorAttachmentsEnd; ++it) { QVkTexture *texD = QRHI_RES(QVkTexture, it->texture()); QVkRenderBuffer *rbD = QRHI_RES(QVkRenderBuffer, it->renderBuffer()); Q_ASSERT(texD || rbD); - const VkFormat vkformat = texD ? texD->vkformat : rbD->vkformat; + const VkFormat vkformat = texD ? texD->viewFormat : rbD->vkformat; const VkSampleCountFlagBits samples = texD ? texD->samples : rbD->samples; VkAttachmentDescription attDesc = {}; attDesc.format = vkformat; attDesc.samples = samples; attDesc.loadOp = preserveColor ? VK_ATTACHMENT_LOAD_OP_LOAD : VK_ATTACHMENT_LOAD_OP_CLEAR; - attDesc.storeOp = it->resolveTexture() ? VK_ATTACHMENT_STORE_OP_DONT_CARE : VK_ATTACHMENT_STORE_OP_STORE; + attDesc.storeOp = (it->resolveTexture() && !preserveColor) ? VK_ATTACHMENT_STORE_OP_DONT_CARE : VK_ATTACHMENT_STORE_OP_STORE; attDesc.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; attDesc.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; // this has to interact correctly with activateTextureRenderTarget(), hence leaving in COLOR_ATT @@ -1380,16 +1631,27 @@ bool QRhiVulkan::createOffscreenRenderPass(QVkRenderPassDescriptor *rpD, const VkAttachmentReference ref = { uint32_t(rpD->attDescs.size() - 1), VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL }; rpD->colorRefs.append(ref); + + if (it->multiViewCount() >= 2) { + if (multiViewCount > 0 && multiViewCount != it->multiViewCount()) + qWarning("Inconsistent multiViewCount in color attachment set"); + else + multiViewCount = it->multiViewCount(); + } else if (multiViewCount > 0) { + qWarning("Mixing non-multiview color attachments within a multiview render pass"); + } } + Q_ASSERT(multiViewCount == 0 || multiViewCount >= 2); + rpD->multiViewCount = uint32_t(multiViewCount); rpD->hasDepthStencil = depthStencilBuffer || depthTexture; if (rpD->hasDepthStencil) { - const VkFormat dsFormat = depthTexture ? QRHI_RES(QVkTexture, depthTexture)->vkformat + const VkFormat dsFormat = depthTexture ? QRHI_RES(QVkTexture, depthTexture)->viewFormat : QRHI_RES(QVkRenderBuffer, depthStencilBuffer)->vkformat; const VkSampleCountFlagBits samples = depthTexture ? QRHI_RES(QVkTexture, depthTexture)->samples : QRHI_RES(QVkRenderBuffer, depthStencilBuffer)->samples; const VkAttachmentLoadOp loadOp = preserveDs ? VK_ATTACHMENT_LOAD_OP_LOAD : VK_ATTACHMENT_LOAD_OP_CLEAR; - const VkAttachmentStoreOp storeOp = depthTexture ? VK_ATTACHMENT_STORE_OP_STORE : VK_ATTACHMENT_STORE_OP_DONT_CARE; + const VkAttachmentStoreOp storeOp = storeDs ? VK_ATTACHMENT_STORE_OP_STORE : VK_ATTACHMENT_STORE_OP_DONT_CARE; VkAttachmentDescription attDesc = {}; attDesc.format = dsFormat; attDesc.samples = samples; @@ -1397,13 +1659,17 @@ bool QRhiVulkan::createOffscreenRenderPass(QVkRenderPassDescriptor *rpD, attDesc.storeOp = storeOp; attDesc.stencilLoadOp = loadOp; attDesc.stencilStoreOp = storeOp; - attDesc.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; + attDesc.initialLayout = preserveDs ? VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL : VK_IMAGE_LAYOUT_UNDEFINED; attDesc.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; rpD->attDescs.append(attDesc); + if (depthTexture && depthTexture->arraySize() >= 2 && colorAttachmentsBegin == colorAttachmentsEnd) { + multiViewCount = depthTexture->arraySize(); + rpD->multiViewCount = multiViewCount; + } } rpD->dsRef = { uint32_t(rpD->attDescs.size() - 1), VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL }; - for (auto it = firstColorAttachment; it != lastColorAttachment; ++it) { + for (auto it = colorAttachmentsBegin; it != colorAttachmentsEnd; ++it) { if (it->resolveTexture()) { QVkTexture *rtexD = QRHI_RES(QVkTexture, it->resolveTexture()); const VkFormat dstFormat = rtexD->vkformat; @@ -1422,7 +1688,7 @@ bool QRhiVulkan::createOffscreenRenderPass(QVkRenderPassDescriptor *rpD, } VkAttachmentDescription attDesc = {}; - attDesc.format = dstFormat; + attDesc.format = rtexD->viewFormat; attDesc.samples = VK_SAMPLE_COUNT_1_BIT; attDesc.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; // ignored attDesc.storeOp = VK_ATTACHMENT_STORE_OP_STORE; @@ -1441,6 +1707,31 @@ bool QRhiVulkan::createOffscreenRenderPass(QVkRenderPassDescriptor *rpD, } Q_ASSERT(rpD->colorRefs.size() == rpD->resolveRefs.size()); + rpD->hasDepthStencilResolve = rpD->hasDepthStencil && depthResolveTexture; + if (rpD->hasDepthStencilResolve) { + QVkTexture *rtexD = QRHI_RES(QVkTexture, depthResolveTexture); + if (rtexD->samples > VK_SAMPLE_COUNT_1_BIT) + qWarning("Resolving into a multisample depth texture is not supported"); + + QVkTexture *texD = QRHI_RES(QVkTexture, depthResolveTexture); + if (texD->vkformat != rtexD->vkformat) { + qWarning("Multisample resolve between different depth-stencil formats (%d and %d) is not supported.", + int(texD->vkformat), int(rtexD->vkformat)); + } + + VkAttachmentDescription attDesc = {}; + attDesc.format = rtexD->viewFormat; + attDesc.samples = VK_SAMPLE_COUNT_1_BIT; + attDesc.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; // ignored + attDesc.storeOp = VK_ATTACHMENT_STORE_OP_STORE; + attDesc.stencilLoadOp = attDesc.loadOp; + attDesc.stencilStoreOp = attDesc.storeOp; + attDesc.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; + attDesc.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; + rpD->attDescs.append(attDesc); + } + rpD->dsResolveRef = { uint32_t(rpD->attDescs.size() - 1), VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL }; + // rpD->subpassDeps stays empty: don't yet know the correct initial/final // access and stage stuff for the implicit deps at this point, so leave it // to the resource tracking and activateTextureRenderTarget() to generate @@ -1450,10 +1741,35 @@ bool QRhiVulkan::createOffscreenRenderPass(QVkRenderPassDescriptor *rpD, VkSubpassDescription subpassDesc; fillRenderPassCreateInfo(&rpInfo, &subpassDesc, rpD); - VkResult err = df->vkCreateRenderPass(dev, &rpInfo, nullptr, &rpD->rp); - if (err != VK_SUCCESS) { - qWarning("Failed to create renderpass: %d", err); + MultiViewRenderPassSetupHelper multiViewHelper; + if (!multiViewHelper.prepare(&rpInfo, multiViewCount, caps.multiView)) return false; + +#ifdef VK_KHR_create_renderpass2 + if (rpD->hasDepthStencilResolve && caps.renderPass2KHR) { + // Use the KHR extension, not the 1.2 core API, in order to support Vulkan 1.1. + VkRenderPassCreateInfo2KHR rpInfo2; + RenderPass2SetupHelper rp2Helper; + if (!rp2Helper.prepare(&rpInfo2, &rpInfo, rpD, multiViewCount)) + return false; + + VkResult err = vkCreateRenderPass2KHR(dev, &rpInfo2, nullptr, &rpD->rp); + if (err != VK_SUCCESS) { + qWarning("Failed to create renderpass (using VkRenderPassCreateInfo2KHR): %d", err); + return false; + } + } else +#endif + { + if (rpD->hasDepthStencilResolve) { + qWarning("Resolving multisample depth-stencil buffers is not supported without " + "VK_KHR_depth_stencil_resolve and VK_KHR_create_renderpass2"); + } + VkResult err = df->vkCreateRenderPass(dev, &rpInfo, nullptr, &rpD->rp); + if (err != VK_SUCCESS) { + qWarning("Failed to create renderpass: %d", err); + return false; + } } return true; @@ -1540,9 +1856,16 @@ bool QRhiVulkan::recreateSwapChain(QRhiSwapChain *swapChain) if (swapChainD->supportsReadback && swapChainD->m_flags.testFlag(QRhiSwapChain::UsedAsTransferSource)) usage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT; + const bool stereo = bool(swapChainD->m_window) && (swapChainD->m_window->format().stereo()) + && surfaceCaps.maxImageArrayLayers > 1; + swapChainD->stereo = stereo; + VkPresentModeKHR presentMode = VK_PRESENT_MODE_FIFO_KHR; if (swapChainD->m_flags.testFlag(QRhiSwapChain::NoVSync)) { - if (swapChainD->supportedPresentationModes.contains(VK_PRESENT_MODE_MAILBOX_KHR)) + // Stereo has a weird bug, when using VK_PRESENT_MODE_MAILBOX_KHR, + // black screen is shown, but there is no validation error. + // Detected on Windows, with NVidia RTX A series (at least 4000 and 6000) driver 535.98 + if (swapChainD->supportedPresentationModes.contains(VK_PRESENT_MODE_MAILBOX_KHR) && !stereo) presentMode = VK_PRESENT_MODE_MAILBOX_KHR; else if (swapChainD->supportedPresentationModes.contains(VK_PRESENT_MODE_IMMEDIATE_KHR)) presentMode = VK_PRESENT_MODE_IMMEDIATE_KHR; @@ -1566,7 +1889,7 @@ bool QRhiVulkan::recreateSwapChain(QRhiSwapChain *swapChain) swapChainInfo.imageFormat = swapChainD->colorFormat; swapChainInfo.imageColorSpace = swapChainD->colorSpace; swapChainInfo.imageExtent = VkExtent2D { uint32_t(swapChainD->pixelSize.width()), uint32_t(swapChainD->pixelSize.height()) }; - swapChainInfo.imageArrayLayers = 1; + swapChainInfo.imageArrayLayers = stereo ? 2u : 1u; swapChainInfo.imageUsage = usage; swapChainInfo.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE; swapChainInfo.preTransform = preTransform; @@ -1628,7 +1951,9 @@ bool QRhiVulkan::recreateSwapChain(QRhiSwapChain *swapChain) fenceInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; fenceInfo.flags = VK_FENCE_CREATE_SIGNALED_BIT; - swapChainD->imageRes.resize(swapChainD->bufferCount); + // Double up for stereo + swapChainD->imageRes.resize(swapChainD->bufferCount * (stereo ? 2u : 1u)); + for (int i = 0; i < swapChainD->bufferCount; ++i) { QVkSwapChain::ImageResources &image(swapChainD->imageRes[i]); image.image = swapChainImages[i]; @@ -1656,6 +1981,36 @@ bool QRhiVulkan::recreateSwapChain(QRhiSwapChain *swapChain) image.lastUse = QVkSwapChain::ImageResources::ScImageUseNone; } + if (stereo) { + for (int i = 0; i < swapChainD->bufferCount; ++i) { + QVkSwapChain::ImageResources &image(swapChainD->imageRes[i + swapChainD->bufferCount]); + image.image = swapChainImages[i]; + if (swapChainD->samples > VK_SAMPLE_COUNT_1_BIT) { + image.msaaImage = msaaImages[i]; + image.msaaImageView = msaaViews[i]; + } + + VkImageViewCreateInfo imgViewInfo = {}; + imgViewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + imgViewInfo.image = swapChainImages[i]; + imgViewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D; + imgViewInfo.format = swapChainD->colorFormat; + imgViewInfo.components.r = VK_COMPONENT_SWIZZLE_R; + imgViewInfo.components.g = VK_COMPONENT_SWIZZLE_G; + imgViewInfo.components.b = VK_COMPONENT_SWIZZLE_B; + imgViewInfo.components.a = VK_COMPONENT_SWIZZLE_A; + imgViewInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + imgViewInfo.subresourceRange.baseArrayLayer = 1; + imgViewInfo.subresourceRange.levelCount = imgViewInfo.subresourceRange.layerCount = 1; + err = df->vkCreateImageView(dev, &imgViewInfo, nullptr, &image.imageView); + if (err != VK_SUCCESS) { + qWarning("Failed to create swapchain image view %d: %d", i, err); + return false; + } + + image.lastUse = QVkSwapChain::ImageResources::ScImageUseNone; + } + } swapChainD->currentImageIndex = 0; @@ -1723,7 +2078,7 @@ void QRhiVulkan::releaseSwapChainResources(QRhiSwapChain *swapChain) } } - for (int i = 0; i < swapChainD->bufferCount; ++i) { + for (int i = 0; i < swapChainD->bufferCount * (swapChainD->stereo ? 2 : 1); ++i) { QVkSwapChain::ImageResources &image(swapChainD->imageRes[i]); if (image.fb) { df->vkDestroyFramebuffer(dev, image.fb, nullptr); @@ -1855,6 +2210,12 @@ QRhi::FrameOpResult QRhiVulkan::beginFrame(QRhiSwapChain *swapChain, QRhi::Begin QVkSwapChain::ImageResources &image(swapChainD->imageRes[swapChainD->currentImageIndex]); swapChainD->rtWrapper.d.fb = image.fb; + if (swapChainD->stereo) { + QVkSwapChain::ImageResources &image( + swapChainD->imageRes[swapChainD->currentImageIndex + swapChainD->bufferCount]); + swapChainD->rtWrapperRight.d.fb = image.fb; + } + prepareNewFrame(&swapChainD->cbWrapper); // Read the timestamps for the previous frame for this slot. @@ -2340,6 +2701,13 @@ void QRhiVulkan::activateTextureRenderTarget(QVkCommandBuffer *cbD, QVkTextureRe QRhiPassResourceTracker::TexDepthOutputStage); depthTexD->lastActiveFrameSlot = currentFrameSlot; } + if (rtD->m_desc.depthResolveTexture()) { + QVkTexture *depthResolveTexD = QRHI_RES(QVkTexture, rtD->m_desc.depthResolveTexture()); + trackedRegisterTexture(&passResTracker, depthResolveTexD, + QRhiPassResourceTracker::TexDepthOutput, + QRhiPassResourceTracker::TexDepthOutputStage); + depthResolveTexD->lastActiveFrameSlot = currentFrameSlot; + } } void QRhiVulkan::resourceUpdate(QRhiCommandBuffer *cb, QRhiResourceUpdateBatch *resourceUpdates) @@ -2481,6 +2849,11 @@ void QRhiVulkan::beginPass(QRhiCommandBuffer *cb, float(colorClearValue.alphaF()) } }; cvs.append(cv); } + for (int i = 0; i < rtD->dsResolveAttCount; ++i) { + VkClearValue cv; + cv.depthStencil = { depthStencilClearValue.depthClearValue(), depthStencilClearValue.stencilClearValue() }; + cvs.append(cv); + } rpBeginInfo.clearValueCount = uint32_t(cvs.size()); QVkCommandBuffer::Command &cmd(cbD->commands.get()); @@ -2874,7 +3247,7 @@ void QRhiVulkan::updateShaderResourceBindings(QRhiShaderResourceBindings *srb, i case QRhiShaderResourceBinding::ImageLoadStore: { QVkTexture *texD = QRHI_RES(QVkTexture, b->u.simage.tex); - VkImageView view = texD->imageViewForLevel(b->u.simage.level); + VkImageView view = texD->perLevelImageViewForLoadStore(b->u.simage.level); if (view) { writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; bd.simage.id = texD->m_id; @@ -3128,12 +3501,12 @@ void QRhiVulkan::prepareUploadSubres(QVkTexture *texD, int layer, int level, const int sy = subresDesc.sourceTopLeft().y(); if (!subresDesc.sourceSize().isEmpty()) size = subresDesc.sourceSize(); - if (image.depth() == 32) { - // The staging buffer will get the full image - // regardless, just adjust the vk - // buffer-to-image copy start offset. - copyInfo.bufferOffset += VkDeviceSize(sy * image.bytesPerLine() + sx * 4); - // bufferRowLength remains set to the original image's width + + if (size.width() == image.width()) { + // No need to make a QImage copy here, can copy from the source + // QImage into staging directly. + src = image.constBits() + sy * image.bytesPerLine() + sx * bpc; + copySizeBytes = size.height() * image.bytesPerLine(); } else { image = image.copy(sx, sy, size.width(), size.height()); src = image.constBits(); @@ -3196,6 +3569,12 @@ void QRhiVulkan::prepareUploadSubres(QVkTexture *texD, int layer, int level, } } +void QRhiVulkan::printExtraErrorInfo(VkResult err) +{ + if (err == VK_ERROR_OUT_OF_DEVICE_MEMORY) + qWarning() << "Out of device memory, current allocator statistics are" << statistics(); +} + void QRhiVulkan::enqueueResourceUpdates(QVkCommandBuffer *cbD, QRhiResourceUpdateBatch *resourceUpdates) { QRhiResourceUpdateBatchPrivate *ud = QRhiResourceUpdateBatchPrivate::get(resourceUpdates); @@ -3233,6 +3612,7 @@ void QRhiVulkan::enqueueResourceUpdates(QVkCommandBuffer *cbD, QRhiResourceUpdat bufD->stagingAllocations[currentFrameSlot] = allocation; } else { qWarning("Failed to create staging buffer of size %u: %d", bufD->m_size, err); + printExtraErrorInfo(err); continue; } } @@ -3321,6 +3701,7 @@ void QRhiVulkan::enqueueResourceUpdates(QVkCommandBuffer *cbD, QRhiResourceUpdat readback.stagingAlloc = allocation; } else { qWarning("Failed to create readback buffer of size %u: %d", readback.byteSize, err); + printExtraErrorInfo(err); continue; } @@ -3370,6 +3751,7 @@ void QRhiVulkan::enqueueResourceUpdates(QVkCommandBuffer *cbD, QRhiResourceUpdat &utexD->stagingBuffers[currentFrameSlot], &allocation, nullptr); if (err != VK_SUCCESS) { qWarning("Failed to create image staging buffer of size %d: %d", int(stagingSize), err); + printExtraErrorInfo(err); continue; } utexD->stagingAllocations[currentFrameSlot] = allocation; @@ -3526,6 +3908,7 @@ void QRhiVulkan::enqueueResourceUpdates(QVkCommandBuffer *cbD, QRhiResourceUpdat readback.stagingAlloc = allocation; } else { qWarning("Failed to create readback buffer of size %u: %d", readback.byteSize, err); + printExtraErrorInfo(err); continue; } @@ -3773,6 +4156,8 @@ void QRhiVulkan::executeDeferredReleases(bool forced) df->vkDestroyImageView(dev, e.textureRenderTarget.rtv[att], nullptr); df->vkDestroyImageView(dev, e.textureRenderTarget.resrtv[att], nullptr); } + df->vkDestroyImageView(dev, e.textureRenderTarget.dsv, nullptr); + df->vkDestroyImageView(dev, e.textureRenderTarget.resdsv, nullptr); break; case QRhiVulkan::DeferredReleaseEntry::RenderPass: df->vkDestroyRenderPass(dev, e.renderPass.rp, nullptr); @@ -3882,18 +4267,12 @@ QList<int> QRhiVulkan::supportedSampleCounts() const return result; } -VkSampleCountFlagBits QRhiVulkan::effectiveSampleCount(int sampleCount) +VkSampleCountFlagBits QRhiVulkan::effectiveSampleCountBits(int sampleCount) { - // Stay compatible with QSurfaceFormat and friends where samples == 0 means the same as 1. - sampleCount = qBound(1, sampleCount, 64); - - if (!supportedSampleCounts().contains(sampleCount)) { - qWarning("Attempted to set unsupported sample count %d", sampleCount); - return VK_SAMPLE_COUNT_1_BIT; - } + const int s = effectiveSampleCount(sampleCount); for (const auto &qvk_sampleCount : qvk_sampleCounts) { - if (qvk_sampleCount.count == sampleCount) + if (qvk_sampleCount.count == s) return qvk_sampleCount.mask; } @@ -4436,6 +4815,12 @@ bool QRhiVulkan::isFeatureSupported(QRhi::Feature feature) const return true; case QRhi::ThreeDimensionalTextureMipmaps: return true; + case QRhi::MultiView: + return caps.multiView; + case QRhi::TextureViewFormat: + return true; + case QRhi::ResolveDepthStencil: + return caps.renderPass2KHR && caps.depthStencilResolveKHR; default: Q_UNREACHABLE_RETURN(false); } @@ -4568,6 +4953,7 @@ QByteArray QRhiVulkan::pipelineCacheData() header.deviceId = physDevProperties.deviceID; header.dataSize = quint32(dataSize); header.uuidSize = VK_UUID_SIZE; + header.reserved = 0; memcpy(data.data(), &header, headerSize); memcpy(data.data() + headerSize, physDevProperties.pipelineCacheUUID, VK_UUID_SIZE); @@ -5464,6 +5850,22 @@ static inline VkFormat toVkAttributeFormat(QRhiVertexInputAttribute::Format form return VK_FORMAT_R16G16_SFLOAT; case QRhiVertexInputAttribute::Half: return VK_FORMAT_R16_SFLOAT; + case QRhiVertexInputAttribute::UShort4: + return VK_FORMAT_R16G16B16A16_UINT; + case QRhiVertexInputAttribute::UShort3: + return VK_FORMAT_R16G16B16_UINT; + case QRhiVertexInputAttribute::UShort2: + return VK_FORMAT_R16G16_UINT; + case QRhiVertexInputAttribute::UShort: + return VK_FORMAT_R16_UINT; + case QRhiVertexInputAttribute::SShort4: + return VK_FORMAT_R16G16B16A16_SINT; + case QRhiVertexInputAttribute::SShort3: + return VK_FORMAT_R16G16B16_SINT; + case QRhiVertexInputAttribute::SShort2: + return VK_FORMAT_R16G16_SINT; + case QRhiVertexInputAttribute::SShort: + return VK_FORMAT_R16_SINT; default: Q_UNREACHABLE_RETURN(VK_FORMAT_R32G32B32A32_SFLOAT); } @@ -5833,7 +6235,8 @@ bool QVkBuffer::create() } if (err != VK_SUCCESS) { - qWarning("Failed to create buffer: %d", err); + qWarning("Failed to create buffer of size %u: %d", nonZeroSize, err); + rhiD->printExtraErrorInfo(err); return false; } @@ -5941,7 +6344,7 @@ bool QVkRenderBuffer::create() return false; QRHI_RES_RHI(QRhiVulkan); - samples = rhiD->effectiveSampleCount(m_sampleCount); + samples = rhiD->effectiveSampleCountBits(m_sampleCount); switch (m_type) { case QRhiRenderBuffer::Color: @@ -6059,6 +6462,15 @@ bool QVkTexture::prepareCreate(QSize *adjustedSize) QRHI_RES_RHI(QRhiVulkan); vkformat = toVkTextureFormat(m_format, m_flags); + if (m_writeViewFormat.format != UnknownFormat) + viewFormat = toVkTextureFormat(m_writeViewFormat.format, m_writeViewFormat.srgb ? sRGB : Flags()); + else + viewFormat = vkformat; + if (m_readViewFormat.format != UnknownFormat) + viewFormatForSampling = toVkTextureFormat(m_readViewFormat.format, m_readViewFormat.srgb ? sRGB : Flags()); + else + viewFormatForSampling = vkformat; + VkFormatProperties props; rhiD->f->vkGetPhysicalDeviceFormatProperties(rhiD->physDev, vkformat, &props); const bool canSampleOptimal = (props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT); @@ -6082,7 +6494,7 @@ bool QVkTexture::prepareCreate(QSize *adjustedSize) qWarning("Too many mip levels (%d, max is %d), truncating mip chain", mipLevelCount, maxLevels); mipLevelCount = maxLevels; } - samples = rhiD->effectiveSampleCount(m_sampleCount); + samples = rhiD->effectiveSampleCountBits(m_sampleCount); if (samples > VK_SAMPLE_COUNT_1_BIT) { if (isCube) { qWarning("Cubemap texture cannot be multisample"); @@ -6154,7 +6566,7 @@ bool QVkTexture::finishCreate() : (is3D ? VK_IMAGE_VIEW_TYPE_3D : (is1D ? (isArray ? VK_IMAGE_VIEW_TYPE_1D_ARRAY : VK_IMAGE_VIEW_TYPE_1D) : (isArray ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D))); - viewInfo.format = vkformat; + viewInfo.format = viewFormatForSampling; viewInfo.components.r = VK_COMPONENT_SWIZZLE_R; viewInfo.components.g = VK_COMPONENT_SWIZZLE_G; viewInfo.components.b = VK_COMPONENT_SWIZZLE_B; @@ -6246,7 +6658,14 @@ bool QVkTexture::create() VmaAllocation allocation; VkResult err = vmaCreateImage(toVmaAllocator(rhiD->allocator), &imageInfo, &allocInfo, &image, &allocation, nullptr); if (err != VK_SUCCESS) { - qWarning("Failed to create image: %d", err); + qWarning("Failed to create image (with VkImageCreateInfo %ux%u depth %u vkformat 0x%X mips %u layers %u vksamples 0x%X): %d", + imageInfo.extent.width, imageInfo.extent.height, imageInfo.extent.depth, + int(imageInfo.format), + imageInfo.mipLevels, + imageInfo.arrayLayers, + int(imageInfo.samples), + err); + rhiD->printExtraErrorInfo(err); return false; } imageAlloc = allocation; @@ -6293,7 +6712,7 @@ void QVkTexture::setNativeLayout(int layout) usageState.layout = VkImageLayout(layout); } -VkImageView QVkTexture::imageViewForLevel(int level) +VkImageView QVkTexture::perLevelImageViewForLoadStore(int level) { Q_ASSERT(level >= 0 && level < int(mipLevelCount)); if (perLevelImageViews[level] != VK_NULL_HANDLE) @@ -6313,7 +6732,7 @@ VkImageView QVkTexture::imageViewForLevel(int level) : (is3D ? VK_IMAGE_VIEW_TYPE_3D : (is1D ? (isArray ? VK_IMAGE_VIEW_TYPE_1D_ARRAY : VK_IMAGE_VIEW_TYPE_1D) : (isArray ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D))); - viewInfo.format = vkformat; + viewInfo.format = viewFormat; // this is writeViewFormat, regardless of Load, Store, or LoadStore; intentional viewInfo.components.r = VK_COMPONENT_SWIZZLE_R; viewInfo.components.g = VK_COMPONENT_SWIZZLE_G; viewInfo.components.b = VK_COMPONENT_SWIZZLE_B; @@ -6400,7 +6819,7 @@ bool QVkSampler::create() QVkRenderPassDescriptor::QVkRenderPassDescriptor(QRhiImplementation *rhi) : QRhiRenderPassDescriptor(rhi) { - serializedFormatData.reserve(32); + serializedFormatData.reserve(64); } QVkRenderPassDescriptor::~QVkRenderPassDescriptor() @@ -6463,6 +6882,10 @@ bool QVkRenderPassDescriptor::isCompatible(const QRhiRenderPassDescriptor *other return false; if (hasDepthStencil != o->hasDepthStencil) return false; + if (hasDepthStencilResolve != o->hasDepthStencilResolve) + return false; + if (multiViewCount != o->multiViewCount) + return false; for (int i = 0, ie = colorRefs.size(); i != ie; ++i) { const uint32_t attIdx = colorRefs[i].attachment; @@ -6488,6 +6911,14 @@ bool QVkRenderPassDescriptor::isCompatible(const QRhiRenderPassDescriptor *other return false; } + if (hasDepthStencilResolve) { + const uint32_t attIdx = dsResolveRef.attachment; + if (attIdx != o->dsResolveRef.attachment) + return false; + if (attIdx != VK_ATTACHMENT_UNUSED && !attachmentDescriptionEquals(attDescs[attIdx], o->attDescs[attIdx])) + return false; + } + // subpassDeps is not included return true; @@ -6502,6 +6933,8 @@ void QVkRenderPassDescriptor::updateSerializedFormat() *p++ = colorRefs.size(); *p++ = resolveRefs.size(); *p++ = hasDepthStencil; + *p++ = hasDepthStencilResolve; + *p++ = multiViewCount; auto serializeAttachmentData = [this, &p](uint32_t attIdx) { const bool used = attIdx != VK_ATTACHMENT_UNUSED; @@ -6533,6 +6966,12 @@ void QVkRenderPassDescriptor::updateSerializedFormat() *p++ = attIdx; serializeAttachmentData(attIdx); } + + if (hasDepthStencilResolve) { + const uint32_t attIdx = dsResolveRef.attachment; + *p++ = attIdx; + serializeAttachmentData(attIdx); + } } QRhiRenderPassDescriptor *QVkRenderPassDescriptor::newCompatibleRenderPassDescriptor() const @@ -6545,13 +6984,22 @@ QRhiRenderPassDescriptor *QVkRenderPassDescriptor::newCompatibleRenderPassDescri rpD->resolveRefs = resolveRefs; rpD->subpassDeps = subpassDeps; rpD->hasDepthStencil = hasDepthStencil; + rpD->hasDepthStencilResolve = hasDepthStencilResolve; + rpD->multiViewCount = multiViewCount; rpD->dsRef = dsRef; + rpD->dsResolveRef = dsResolveRef; VkRenderPassCreateInfo rpInfo; VkSubpassDescription subpassDesc; fillRenderPassCreateInfo(&rpInfo, &subpassDesc, rpD); QRHI_RES_RHI(QRhiVulkan); + MultiViewRenderPassSetupHelper multiViewHelper; + if (!multiViewHelper.prepare(&rpInfo, multiViewCount, rhiD->caps.multiView)) { + delete rpD; + return nullptr; + } + VkResult err = rhiD->df->vkCreateRenderPass(rhiD->dev, &rpInfo, nullptr, &rpD->rp); if (err != VK_SUCCESS) { qWarning("Failed to create renderpass: %d", err); @@ -6640,6 +7088,11 @@ void QVkTextureRenderTarget::destroy() resrtv[att] = VK_NULL_HANDLE; } + e.textureRenderTarget.dsv = dsv; + dsv = VK_NULL_HANDLE; + e.textureRenderTarget.resdsv = resdsv; + resdsv = VK_NULL_HANDLE; + QRHI_RES_RHI(QRhiVulkan); if (rhiD) { rhiD->releaseQueue.append(e); @@ -6658,8 +7111,10 @@ QRhiRenderPassDescriptor *QVkTextureRenderTarget::newCompatibleRenderPassDescrip m_desc.cendColorAttachments(), m_flags.testFlag(QRhiTextureRenderTarget::PreserveColorContents), m_flags.testFlag(QRhiTextureRenderTarget::PreserveDepthStencilContents), + m_desc.depthTexture() && !m_flags.testFlag(DoNotStoreDepthStencilContents) && !m_desc.depthResolveTexture(), m_desc.depthStencilBuffer(), - m_desc.depthTexture())) + m_desc.depthTexture(), + m_desc.depthResolveTexture())) { delete rp; return nullptr; @@ -6682,6 +7137,7 @@ bool QVkTextureRenderTarget::create() QRHI_RES_RHI(QRhiVulkan); QVarLengthArray<VkImageView, 8> views; + d.multiViewCount = 0; d.colorAttCount = 0; int attIndex = 0; @@ -6692,13 +7148,17 @@ bool QVkTextureRenderTarget::create() Q_ASSERT(texD || rbD); if (texD) { Q_ASSERT(texD->flags().testFlag(QRhiTexture::RenderTarget)); + const bool is1D = texD->flags().testFlag(QRhiTexture::OneDimensional); + const bool isMultiView = it->multiViewCount() >= 2; + if (isMultiView && d.multiViewCount == 0) + d.multiViewCount = it->multiViewCount(); VkImageViewCreateInfo viewInfo = {}; viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; viewInfo.image = texD->image; - viewInfo.viewType = texD->flags().testFlag(QRhiTexture::OneDimensional) - ? VK_IMAGE_VIEW_TYPE_1D - : VK_IMAGE_VIEW_TYPE_2D; - viewInfo.format = texD->vkformat; + viewInfo.viewType = is1D ? VK_IMAGE_VIEW_TYPE_1D + : (isMultiView ? VK_IMAGE_VIEW_TYPE_2D_ARRAY + : VK_IMAGE_VIEW_TYPE_2D); + viewInfo.format = texD->viewFormat; viewInfo.components.r = VK_COMPONENT_SWIZZLE_R; viewInfo.components.g = VK_COMPONENT_SWIZZLE_G; viewInfo.components.b = VK_COMPONENT_SWIZZLE_B; @@ -6707,7 +7167,7 @@ bool QVkTextureRenderTarget::create() viewInfo.subresourceRange.baseMipLevel = uint32_t(it->level()); viewInfo.subresourceRange.levelCount = 1; viewInfo.subresourceRange.baseArrayLayer = uint32_t(it->layer()); - viewInfo.subresourceRange.layerCount = 1; + viewInfo.subresourceRange.layerCount = uint32_t(isMultiView ? it->multiViewCount() : 1); VkResult err = rhiD->df->vkCreateImageView(rhiD->dev, &viewInfo, nullptr, &rtv[attIndex]); if (err != VK_SUCCESS) { qWarning("Failed to create render target image view: %d", err); @@ -6732,7 +7192,25 @@ bool QVkTextureRenderTarget::create() if (hasDepthStencil) { if (m_desc.depthTexture()) { QVkTexture *depthTexD = QRHI_RES(QVkTexture, m_desc.depthTexture()); - views.append(depthTexD->imageView); + // need a dedicated view just because viewFormat may differ from vkformat + VkImageViewCreateInfo viewInfo = {}; + viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + viewInfo.image = depthTexD->image; + viewInfo.viewType = d.multiViewCount > 1 ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D; + viewInfo.format = depthTexD->viewFormat; + viewInfo.components.r = VK_COMPONENT_SWIZZLE_R; + viewInfo.components.g = VK_COMPONENT_SWIZZLE_G; + viewInfo.components.b = VK_COMPONENT_SWIZZLE_B; + viewInfo.components.a = VK_COMPONENT_SWIZZLE_A; + viewInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT; + viewInfo.subresourceRange.levelCount = 1; + viewInfo.subresourceRange.layerCount = qMax<uint32_t>(1, d.multiViewCount); + VkResult err = rhiD->df->vkCreateImageView(rhiD->dev, &viewInfo, nullptr, &dsv); + if (err != VK_SUCCESS) { + qWarning("Failed to create depth-stencil image view for rt: %d", err); + return false; + } + views.append(dsv); if (d.colorAttCount == 0) { d.pixelSize = depthTexD->pixelSize(); d.sampleCount = depthTexD->samples; @@ -6752,6 +7230,7 @@ bool QVkTextureRenderTarget::create() d.resolveAttCount = 0; attIndex = 0; + Q_ASSERT(d.multiViewCount == 0 || d.multiViewCount >= 2); for (auto it = m_desc.cbeginColorAttachments(), itEnd = m_desc.cendColorAttachments(); it != itEnd; ++it, ++attIndex) { if (it->resolveTexture()) { QVkTexture *resTexD = QRHI_RES(QVkTexture, it->resolveTexture()); @@ -6761,8 +7240,9 @@ bool QVkTextureRenderTarget::create() VkImageViewCreateInfo viewInfo = {}; viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; viewInfo.image = resTexD->image; - viewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D; - viewInfo.format = resTexD->vkformat; + viewInfo.viewType = d.multiViewCount ? VK_IMAGE_VIEW_TYPE_2D_ARRAY + : VK_IMAGE_VIEW_TYPE_2D; + viewInfo.format = resTexD->viewFormat; viewInfo.components.r = VK_COMPONENT_SWIZZLE_R; viewInfo.components.g = VK_COMPONENT_SWIZZLE_G; viewInfo.components.b = VK_COMPONENT_SWIZZLE_B; @@ -6771,7 +7251,7 @@ bool QVkTextureRenderTarget::create() viewInfo.subresourceRange.baseMipLevel = uint32_t(it->resolveLevel()); viewInfo.subresourceRange.levelCount = 1; viewInfo.subresourceRange.baseArrayLayer = uint32_t(it->resolveLayer()); - viewInfo.subresourceRange.layerCount = 1; + viewInfo.subresourceRange.layerCount = qMax<uint32_t>(1, d.multiViewCount); VkResult err = rhiD->df->vkCreateImageView(rhiD->dev, &viewInfo, nullptr, &resrtv[attIndex]); if (err != VK_SUCCESS) { qWarning("Failed to create render target resolve image view: %d", err); @@ -6781,6 +7261,36 @@ bool QVkTextureRenderTarget::create() } } + if (m_desc.depthResolveTexture()) { + QVkTexture *resTexD = QRHI_RES(QVkTexture, m_desc.depthResolveTexture()); + Q_ASSERT(resTexD->flags().testFlag(QRhiTexture::RenderTarget)); + + VkImageViewCreateInfo viewInfo = {}; + viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + viewInfo.image = resTexD->image; + viewInfo.viewType = d.multiViewCount ? VK_IMAGE_VIEW_TYPE_2D_ARRAY + : VK_IMAGE_VIEW_TYPE_2D; + viewInfo.format = resTexD->viewFormat; + viewInfo.components.r = VK_COMPONENT_SWIZZLE_R; + viewInfo.components.g = VK_COMPONENT_SWIZZLE_G; + viewInfo.components.b = VK_COMPONENT_SWIZZLE_B; + viewInfo.components.a = VK_COMPONENT_SWIZZLE_A; + viewInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT; + viewInfo.subresourceRange.baseMipLevel = 0; + viewInfo.subresourceRange.levelCount = 1; + viewInfo.subresourceRange.baseArrayLayer = 0; + viewInfo.subresourceRange.layerCount = qMax<uint32_t>(1, d.multiViewCount); + VkResult err = rhiD->df->vkCreateImageView(rhiD->dev, &viewInfo, nullptr, &resdsv); + if (err != VK_SUCCESS) { + qWarning("Failed to create render target depth resolve image view: %d", err); + return false; + } + views.append(resdsv); + d.dsResolveAttCount = 1; + } else { + d.dsResolveAttCount = 0; + } + if (!m_renderPassDesc) qWarning("QVkTextureRenderTarget: No renderpass descriptor set. See newCompatibleRenderPassDescriptor() and setRenderPassDescriptor()."); @@ -6790,7 +7300,7 @@ bool QVkTextureRenderTarget::create() VkFramebufferCreateInfo fbInfo = {}; fbInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO; fbInfo.renderPass = d.rp->rp; - fbInfo.attachmentCount = uint32_t(d.colorAttCount + d.dsAttCount + d.resolveAttCount); + fbInfo.attachmentCount = uint32_t(d.colorAttCount + d.dsAttCount + d.resolveAttCount + d.dsResolveAttCount); fbInfo.pAttachments = views.constData(); fbInfo.width = uint32_t(d.pixelSize.width()); fbInfo.height = uint32_t(d.pixelSize.height()); @@ -7049,7 +7559,9 @@ bool QVkGraphicsPipeline::create() pipelineInfo.pStages = shaderStageCreateInfos.constData(); QVarLengthArray<VkVertexInputBindingDescription, 4> vertexBindings; +#ifdef VK_EXT_vertex_attribute_divisor QVarLengthArray<VkVertexInputBindingDivisorDescriptionEXT> nonOneStepRates; +#endif int bindingIndex = 0; for (auto it = m_vertexInputLayout.cbeginBindings(), itEnd = m_vertexInputLayout.cendBindings(); it != itEnd; ++it, ++bindingIndex) @@ -7061,9 +7573,12 @@ bool QVkGraphicsPipeline::create() ? VK_VERTEX_INPUT_RATE_VERTEX : VK_VERTEX_INPUT_RATE_INSTANCE }; if (it->classification() == QRhiVertexInputBinding::PerInstance && it->instanceStepRate() != 1) { +#ifdef VK_EXT_vertex_attribute_divisor if (rhiD->caps.vertexAttribDivisor) { nonOneStepRates.append({ uint32_t(bindingIndex), it->instanceStepRate() }); - } else { + } else +#endif + { qWarning("QRhiVulkan: Instance step rates other than 1 not supported without " "VK_EXT_vertex_attribute_divisor on the device and " "VK_KHR_get_physical_device_properties2 on the instance"); @@ -7089,13 +7604,15 @@ bool QVkGraphicsPipeline::create() vertexInputInfo.pVertexBindingDescriptions = vertexBindings.constData(); vertexInputInfo.vertexAttributeDescriptionCount = uint32_t(vertexAttributes.size()); vertexInputInfo.pVertexAttributeDescriptions = vertexAttributes.constData(); +#ifdef VK_EXT_vertex_attribute_divisor VkPipelineVertexInputDivisorStateCreateInfoEXT divisorInfo = {}; if (!nonOneStepRates.isEmpty()) { - divisorInfo.sType = VkStructureType(1000190001); // VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_DIVISOR_STATE_CREATE_INFO_EXT + divisorInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_DIVISOR_STATE_CREATE_INFO_EXT; divisorInfo.vertexBindingDivisorCount = uint32_t(nonOneStepRates.size()); divisorInfo.pVertexBindingDivisors = nonOneStepRates.constData(); vertexInputInfo.pNext = &divisorInfo; } +#endif pipelineInfo.pVertexInputState = &vertexInputInfo; QVarLengthArray<VkDynamicState, 8> dynEnable; @@ -7166,7 +7683,7 @@ bool QVkGraphicsPipeline::create() VkPipelineMultisampleStateCreateInfo msInfo = {}; msInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO; - msInfo.rasterizationSamples = rhiD->effectiveSampleCount(m_sampleCount); + msInfo.rasterizationSamples = rhiD->effectiveSampleCountBits(m_sampleCount); pipelineInfo.pMultisampleState = &msInfo; VkPipelineDepthStencilStateCreateInfo dsInfo = {}; @@ -7365,6 +7882,7 @@ const QRhiNativeHandles *QVkCommandBuffer::nativeHandles() QVkSwapChain::QVkSwapChain(QRhiImplementation *rhi) : QRhiSwapChain(rhi), rtWrapper(rhi, this), + rtWrapperRight(rhi, this), cbWrapper(rhi) { } @@ -7407,6 +7925,11 @@ QRhiRenderTarget *QVkSwapChain::currentFrameRenderTarget() return &rtWrapper; } +QRhiRenderTarget *QVkSwapChain::currentFrameRenderTarget(StereoTargetBuffer targetBuffer) +{ + return !stereo || targetBuffer == StereoTargetBuffer::LeftBuffer ? &rtWrapper : &rtWrapperRight; +} + QSize QVkSwapChain::surfacePixelSize() { if (!ensureSurface()) @@ -7434,6 +7957,9 @@ static inline bool hdrFormatMatchesVkSurfaceFormat(QRhiSwapChain::Format f, cons case QRhiSwapChain::HDR10: return (s.format == VK_FORMAT_A2B10G10R10_UNORM_PACK32 || s.format == VK_FORMAT_A2R10G10B10_UNORM_PACK32) && s.colorSpace == VK_COLOR_SPACE_HDR10_ST2084_EXT; + case QRhiSwapChain::HDRExtendedDisplayP3Linear: + return s.format == VK_FORMAT_R16G16B16A16_SFLOAT + && s.colorSpace == VK_COLOR_SPACE_DISPLAY_P3_LINEAR_EXT; default: break; } @@ -7554,7 +8080,7 @@ bool QVkSwapChain::ensureSurface() } } - samples = rhiD->effectiveSampleCount(m_sampleCount); + samples = rhiD->effectiveSampleCountBits(m_sampleCount); quint32 presModeCount = 0; rhiD->vkGetPhysicalDeviceSurfacePresentModesKHR(rhiD->physDev, surface, &presModeCount, nullptr); @@ -7625,6 +8151,7 @@ bool QVkSwapChain::createOrResize() rtWrapper.d.dsAttCount = 0; ds = nullptr; } + rtWrapper.d.dsResolveAttCount = 0; if (samples > VK_SAMPLE_COUNT_1_BIT) rtWrapper.d.resolveAttCount = 1; else @@ -7641,7 +8168,7 @@ bool QVkSwapChain::createOrResize() VkFramebufferCreateInfo fbInfo = {}; fbInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO; fbInfo.renderPass = rtWrapper.d.rp->rp; - fbInfo.attachmentCount = uint32_t(rtWrapper.d.colorAttCount + rtWrapper.d.dsAttCount + rtWrapper.d.resolveAttCount); + fbInfo.attachmentCount = uint32_t(rtWrapper.d.colorAttCount + rtWrapper.d.dsAttCount + rtWrapper.d.resolveAttCount + rtWrapper.d.dsResolveAttCount); fbInfo.pAttachments = views; fbInfo.width = uint32_t(pixelSize.width()); fbInfo.height = uint32_t(pixelSize.height()); @@ -7654,6 +8181,56 @@ bool QVkSwapChain::createOrResize() } } + if (stereo) { + rtWrapperRight.setRenderPassDescriptor( + m_renderPassDesc); // for the public getter in QRhiRenderTarget + rtWrapperRight.d.rp = QRHI_RES(QVkRenderPassDescriptor, m_renderPassDesc); + Q_ASSERT(rtWrapperRight.d.rp && rtWrapperRight.d.rp->rp); + + rtWrapperRight.d.pixelSize = pixelSize; + rtWrapperRight.d.dpr = float(window->devicePixelRatio()); + rtWrapperRight.d.sampleCount = samples; + rtWrapperRight.d.colorAttCount = 1; + if (m_depthStencil) { + rtWrapperRight.d.dsAttCount = 1; + ds = QRHI_RES(QVkRenderBuffer, m_depthStencil); + } else { + rtWrapperRight.d.dsAttCount = 0; + ds = nullptr; + } + rtWrapperRight.d.dsResolveAttCount = 0; + if (samples > VK_SAMPLE_COUNT_1_BIT) + rtWrapperRight.d.resolveAttCount = 1; + else + rtWrapperRight.d.resolveAttCount = 0; + + for (int i = 0; i < bufferCount; ++i) { + QVkSwapChain::ImageResources &image(imageRes[i + bufferCount]); + VkImageView views[3] = { + // color, ds, resolve + samples > VK_SAMPLE_COUNT_1_BIT ? image.msaaImageView : image.imageView, + ds ? ds->imageView : VK_NULL_HANDLE, + samples > VK_SAMPLE_COUNT_1_BIT ? image.imageView : VK_NULL_HANDLE + }; + + VkFramebufferCreateInfo fbInfo = {}; + fbInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO; + fbInfo.renderPass = rtWrapperRight.d.rp->rp; + fbInfo.attachmentCount = uint32_t(rtWrapperRight.d.colorAttCount + rtWrapperRight.d.dsAttCount + + rtWrapperRight.d.resolveAttCount + rtWrapperRight.d.dsResolveAttCount); + fbInfo.pAttachments = views; + fbInfo.width = uint32_t(pixelSize.width()); + fbInfo.height = uint32_t(pixelSize.height()); + fbInfo.layers = 1; + + VkResult err = rhiD->df->vkCreateFramebuffer(rhiD->dev, &fbInfo, nullptr, &image.fb); + if (err != VK_SUCCESS) { + qWarning("Failed to create framebuffer: %d", err); + return false; + } + } + } + frameCount = 0; if (needsRegistration) diff --git a/src/gui/rhi/qrhivulkan_p.h b/src/gui/rhi/qrhivulkan_p.h index 7ba1b8c89b..f23d8550f0 100644 --- a/src/gui/rhi/qrhivulkan_p.h +++ b/src/gui/rhi/qrhivulkan_p.h @@ -103,7 +103,7 @@ struct QVkTexture : public QRhiTexture bool prepareCreate(QSize *adjustedSize = nullptr); bool finishCreate(); - VkImageView imageViewForLevel(int level); + VkImageView perLevelImageViewForLoadStore(int level); VkImage image = VK_NULL_HANDLE; VkImageView imageView = VK_NULL_HANDLE; @@ -124,6 +124,8 @@ struct QVkTexture : public QRhiTexture VkFormat vkformat; uint mipLevelCount = 0; VkSampleCountFlagBits samples; + VkFormat viewFormat; + VkFormat viewFormatForSampling; int lastActiveFrameSlot = -1; uint generation = 0; friend class QRhiVulkan; @@ -162,7 +164,10 @@ struct QVkRenderPassDescriptor : public QRhiRenderPassDescriptor QVarLengthArray<VkAttachmentReference, 8> resolveRefs; QVarLengthArray<VkSubpassDependency, 2> subpassDeps; bool hasDepthStencil = false; + bool hasDepthStencilResolve = false; + uint32_t multiViewCount = 0; VkAttachmentReference dsRef; + VkAttachmentReference dsResolveRef; QVector<quint32> serializedFormatData; QRhiVulkanRenderPassNativeHandles nativeHandlesStruct; int lastActiveFrameSlot = -1; @@ -178,6 +183,8 @@ struct QVkRenderTargetData int colorAttCount = 0; int dsAttCount = 0; int resolveAttCount = 0; + int dsResolveAttCount = 0; + int multiViewCount = 0; QRhiRenderTargetAttachmentTracker::ResIdList currentResIdList; static const int MAX_COLOR_ATTACHMENTS = 8; }; @@ -210,7 +217,9 @@ struct QVkTextureRenderTarget : public QRhiTextureRenderTarget QVkRenderTargetData d; VkImageView rtv[QVkRenderTargetData::MAX_COLOR_ATTACHMENTS]; + VkImageView dsv = VK_NULL_HANDLE; VkImageView resrtv[QVkRenderTargetData::MAX_COLOR_ATTACHMENTS]; + VkImageView resdsv = VK_NULL_HANDLE; int lastActiveFrameSlot = -1; friend class QRhiVulkan; }; @@ -570,6 +579,7 @@ struct QVkSwapChain : public QRhiSwapChain QRhiCommandBuffer *currentFrameCommandBuffer() override; QRhiRenderTarget *currentFrameRenderTarget() override; + QRhiRenderTarget *currentFrameRenderTarget(StereoTargetBuffer targetBuffer) override; QSize surfacePixelSize() override; bool isFormatSupported(Format f) override; @@ -584,6 +594,7 @@ struct QVkSwapChain : public QRhiSwapChain QWindow *window = nullptr; QSize pixelSize; bool supportsReadback = false; + bool stereo = false; VkSwapchainKHR sc = VK_NULL_HANDLE; int bufferCount = 0; VkSurfaceKHR surface = VK_NULL_HANDLE; @@ -595,6 +606,7 @@ struct QVkSwapChain : public QRhiSwapChain QVarLengthArray<VkPresentModeKHR, 8> supportedPresentationModes; VkDeviceMemory msaaImageMem = VK_NULL_HANDLE; QVkSwapChainRenderTarget rtWrapper; + QVkSwapChainRenderTarget rtWrapperRight; QVkCommandBuffer cbWrapper; struct ImageResources { @@ -755,18 +767,20 @@ public: void releaseSwapChainResources(QRhiSwapChain *swapChain); VkFormat optimalDepthStencilFormat(); - VkSampleCountFlagBits effectiveSampleCount(int sampleCount); + VkSampleCountFlagBits effectiveSampleCountBits(int sampleCount); bool createDefaultRenderPass(QVkRenderPassDescriptor *rpD, bool hasDepthStencil, VkSampleCountFlagBits samples, VkFormat colorFormat); bool createOffscreenRenderPass(QVkRenderPassDescriptor *rpD, - const QRhiColorAttachment *firstColorAttachment, - const QRhiColorAttachment *lastColorAttachment, + const QRhiColorAttachment *colorAttachmentsBegin, + const QRhiColorAttachment *colorAttachmentsEnd, bool preserveColor, bool preserveDs, + bool storeDs, QRhiRenderBuffer *depthStencilBuffer, - QRhiTexture *depthTexture); + QRhiTexture *depthTexture, + QRhiTexture *depthResolveTexture); bool ensurePipelineCache(const void *initialData = nullptr, size_t initialDataSize = 0); VkShaderModule createShader(const QByteArray &spirv); @@ -816,6 +830,7 @@ public: void updateShaderResourceBindings(QRhiShaderResourceBindings *srb, int descSetIdx = -1); void ensureCommandPoolForNewFrame(); double elapsedSecondsFromTimestamp(quint64 timestamp[2], bool *ok); + void printExtraErrorInfo(VkResult err); QVulkanInstance *inst = nullptr; QWindow *maybeWindow = nullptr; @@ -834,6 +849,16 @@ public: QVulkanDeviceFunctions *df = nullptr; QRhi::Flags rhiFlags; VkPhysicalDeviceFeatures physDevFeatures; +#ifdef VK_VERSION_1_1 + VkPhysicalDeviceMultiviewFeatures multiviewFeaturesIfApi11; +#endif +#ifdef VK_VERSION_1_2 + VkPhysicalDeviceVulkan11Features physDevFeatures11IfApi12OrNewer; + VkPhysicalDeviceVulkan12Features physDevFeatures12; +#endif +#ifdef VK_VERSION_1_3 + VkPhysicalDeviceVulkan13Features physDevFeatures13; +#endif VkPhysicalDeviceProperties physDevProperties; VkDeviceSize ubufAlign; VkDeviceSize texbufAlign; @@ -856,6 +881,10 @@ public: PFN_vkGetPhysicalDeviceSurfaceFormatsKHR vkGetPhysicalDeviceSurfaceFormatsKHR; PFN_vkGetPhysicalDeviceSurfacePresentModesKHR vkGetPhysicalDeviceSurfacePresentModesKHR; +#ifdef VK_KHR_create_renderpass2 + PFN_vkCreateRenderPass2KHR vkCreateRenderPass2KHR = nullptr; +#endif + struct { bool compute = false; bool wideLines = false; @@ -865,6 +894,9 @@ public: bool tessellation = false; bool geometryShader = false; bool nonFillPolygonMode = false; + bool multiView = false; + bool renderPass2KHR = false; + bool depthStencilResolveKHR = false; QVersionNumber apiVersion; } caps; @@ -979,6 +1011,8 @@ public: VkFramebuffer fb; VkImageView rtv[QVkRenderTargetData::MAX_COLOR_ATTACHMENTS]; VkImageView resrtv[QVkRenderTargetData::MAX_COLOR_ATTACHMENTS]; + VkImageView dsv; + VkImageView resdsv; } textureRenderTarget; struct { VkRenderPass rp; diff --git a/src/gui/rhi/qrhivulkanext_p.h b/src/gui/rhi/qrhivulkanext_p.h deleted file mode 100644 index 02b346948b..0000000000 --- a/src/gui/rhi/qrhivulkanext_p.h +++ /dev/null @@ -1,48 +0,0 @@ -// Copyright (C) 2018 The Qt Company Ltd. -// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only - -#ifndef QRHIVULKANEXT_P_H -#define QRHIVULKANEXT_P_H - -// -// W A R N I N G -// ------------- -// -// This file is not part of the Qt API. It exists purely as an -// implementation detail. This header file may change from version to -// version without notice, or even be removed. -// -// We mean it. -// - -#include "qrhivulkan_p.h" - -QT_BEGIN_NAMESPACE - -#ifndef VK_EXT_vertex_attribute_divisor -#define VK_EXT_vertex_attribute_divisor 1 -#define VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_SPEC_VERSION 2 -#define VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME "VK_EXT_vertex_attribute_divisor" - -typedef struct VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT { - VkStructureType sType; - void* pNext; - uint32_t maxVertexAttribDivisor; -} VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT; - -typedef struct VkVertexInputBindingDivisorDescriptionEXT { - uint32_t binding; - uint32_t divisor; -} VkVertexInputBindingDivisorDescriptionEXT; - -typedef struct VkPipelineVertexInputDivisorStateCreateInfoEXT { - VkStructureType sType; - const void* pNext; - uint32_t vertexBindingDivisorCount; - const VkVertexInputBindingDivisorDescriptionEXT* pVertexBindingDivisors; -} VkPipelineVertexInputDivisorStateCreateInfoEXT; -#endif // VK_EXT_vertex_attribute_divisor - -QT_END_NAMESPACE - -#endif diff --git a/src/gui/rhi/qshader.cpp b/src/gui/rhi/qshader.cpp index af2275f088..d5fb53e7e6 100644 --- a/src/gui/rhi/qshader.cpp +++ b/src/gui/rhi/qshader.cpp @@ -141,7 +141,7 @@ QT_BEGIN_NAMESPACE A default constructed QShaderVersion contains a version of 100 and no flags set. - \note This a RHI API with limited compatibility guarantees, see \l QShader + \note This is a RHI API with limited compatibility guarantees, see \l QShader for details. */ @@ -163,7 +163,7 @@ QT_BEGIN_NAMESPACE A default constructed QShaderKey has source set to SpirvShader and sourceVersion set to 100. sourceVariant defaults to StandardShader. - \note This a RHI API with limited compatibility guarantees, see \l QShader + \note This is a RHI API with limited compatibility guarantees, see \l QShader for details. */ @@ -214,6 +214,28 @@ QT_BEGIN_NAMESPACE */ /*! + \enum QShader::SerializedFormatVersion + Describes the desired output format when serializing the QShader. + + The default value for the \c version argument of serialized() is \c Latest. + This is sufficient in the vast majority of cases. Specifying another value + is needed only when the intention is to generate serialized data that can + be loaded by earlier Qt versions. For example, the \c qsb tool uses these + enum values when the \c{--qsbversion} command-line argument is given. + + \note Targeting earlier versions will make certain features disfunctional + with the generated asset. This is not an issue when using the asset with + the specified, older Qt version, given that that Qt version does not have + the newer features in newer Qt versions that rely on additional data + generated in the QShader and the serialized data stream, but may become a + problem if the generated asset is then used with a newer Qt version. + + \value Latest The current Qt version + \value Qt_6_5 Qt 6.5 + \value Qt_6_4 Qt 6.4 + */ + +/*! \class QShaderCode \inmodule QtGui \since 6.6 @@ -223,7 +245,7 @@ QT_BEGIN_NAMESPACE When shader() is empty after retrieving a QShaderCode instance from QShader, it indicates no shader code was found for the requested key. - \note This a RHI API with limited compatibility guarantees, see \l QShader + \note This is a RHI API with limited compatibility guarantees, see \l QShader for details. */ @@ -277,6 +299,28 @@ QShader &QShader::operator=(const QShader &other) } /*! + \fn QShader::QShader(QShader &&other) noexcept + \since 6.7 + + Move-constructs a new QShader from \a other. + + \note The moved-from object \a other is placed in a + partially-formed state, in which the only valid operations are + destruction and assignment of a new value. +*/ + +/*! + \fn QShader &QShader::operator=(QShader &&other) + \since 6.7 + + Move-assigns \a other to this QShader instance. + + \note The moved-from object \a other is placed in a + partially-formed state, in which the only valid operations are + destruction and assignment of a new value. +*/ + +/*! Destructor. */ QShader::~QShader() @@ -286,6 +330,14 @@ QShader::~QShader() } /*! + \fn void QShader::swap(QShader &other) + \since 6.7 + + Swaps shader \a other with this shader. This operation is very fast and + never fails. +*/ + +/*! \return true if the QShader contains at least one shader version. */ bool QShader::isValid() const @@ -387,7 +439,11 @@ static void writeShaderKey(QDataStream *ds, const QShaderKey &k) QShader, suitable for writing to files or other I/O devices. By default the latest serialization format is used. Use \a version - parameter to serialize for a compatibility Qt version. + parameter to serialize for a compatibility Qt version. Only when it is + known that the generated data stream must be made compatible with an older + Qt version at the expense of making it incompatible with features + introduced since that Qt version, should another value (for example, + \l{SerializedFormatVersion}{Qt_6_5} for Qt 6.5) be used. \sa fromSerialized() */ @@ -476,6 +532,9 @@ static void readShaderKey(QDataStream *ds, QShaderKey *k) /*! Creates a new QShader instance from the given \a data. + If \a data cannot be deserialized successfully, the result is a default + constructed QShader for which isValid() returns \c false. + \sa serialized() */ QShader QShader::fromSerialized(const QByteArray &data) @@ -748,7 +807,7 @@ size_t qHash(const QShader &s, size_t seed) noexcept seed = hash(seed, s.stage()); if (!s.d->shaders.isEmpty()) { seed = hash(seed, s.d->shaders.firstKey()); - seed = hash(seed, s.d->shaders.first()); + seed = hash(seed, std::as_const(s.d->shaders).first()); } } return seed; @@ -1021,7 +1080,7 @@ void QShader::removeResourceBindingMap(const QShaderKey &key) \c{_54} which corresponds to two separate resource bindings (\c 1 and \c 2) in the original shader. - \note This a RHI API with limited compatibility guarantees, see \l QShader + \note This is a RHI API with limited compatibility guarantees, see \l QShader for details. */ @@ -1103,7 +1162,7 @@ void QShader::removeSeparateToCombinedImageSamplerMappingList(const QShaderKey & that the shader code relies on such a buffer present can be indicated by the data in this struct. - \note This a RHI API with limited compatibility guarantees, see \l QShader + \note This is a RHI API with limited compatibility guarantees, see \l QShader for details. */ diff --git a/src/gui/rhi/qshader.h b/src/gui/rhi/qshader.h index 0b52022596..2465081366 100644 --- a/src/gui/rhi/qshader.h +++ b/src/gui/rhi/qshader.h @@ -117,7 +117,11 @@ public: QShader(); QShader(const QShader &other); QShader &operator=(const QShader &other); + QShader(QShader &&other) noexcept : d(std::exchange(other.d, nullptr)) {} + QT_MOVE_ASSIGNMENT_OPERATOR_IMPL_VIA_PURE_SWAP(QShader) ~QShader(); + + void swap(QShader &other) noexcept { qt_ptr_swap(d, other.d); } void detach(); bool isValid() const; diff --git a/src/gui/rhi/qshader_p.h b/src/gui/rhi/qshader_p.h index a2df33d96b..f77bcb1259 100644 --- a/src/gui/rhi/qshader_p.h +++ b/src/gui/rhi/qshader_p.h @@ -41,7 +41,8 @@ struct Q_GUI_EXPORT QShaderPrivate MslTessTescPatchOutputBufferBinding, MslTessTescParamsBufferBinding, MslTessTescInputBufferBinding, - MslBufferSizeBufferBinding + MslBufferSizeBufferBinding, + MslMultiViewMaskBufferBinding }; QShaderPrivate() diff --git a/src/gui/rhi/qshaderdescription.cpp b/src/gui/rhi/qshaderdescription.cpp index 97b8798034..f64daf02ef 100644 --- a/src/gui/rhi/qshaderdescription.cpp +++ b/src/gui/rhi/qshaderdescription.cpp @@ -287,7 +287,7 @@ QT_BEGIN_NAMESPACE \brief Describes an input or output variable in the shader. - \note This a RHI API with limited compatibility guarantees, see \l QShaderDescription + \note This is a RHI API with limited compatibility guarantees, see \l QShaderDescription for details. */ @@ -338,7 +338,7 @@ QT_BEGIN_NAMESPACE \brief Describes a member of a uniform or push constant block. - \note This a RHI API with limited compatibility guarantees, see \l QShaderDescription + \note This is a RHI API with limited compatibility guarantees, see \l QShaderDescription for details. */ @@ -390,7 +390,7 @@ QT_BEGIN_NAMESPACE uniforms in a struct. The name of the struct, and so the prefix for the uniforms generated from the block members, is given by structName. - \note This a RHI API with limited compatibility guarantees, see \l QShaderDescription + \note This is a RHI API with limited compatibility guarantees, see \l QShaderDescription for details. */ @@ -425,7 +425,7 @@ QT_BEGIN_NAMESPACE \brief Describes a push constant block. - \note This a RHI API with limited compatibility guarantees, see \l QShaderDescription + \note This is a RHI API with limited compatibility guarantees, see \l QShaderDescription for details. */ @@ -448,7 +448,7 @@ QT_BEGIN_NAMESPACE \brief Describes a shader storage block. - \note This a RHI API with limited compatibility guarantees, see \l QShaderDescription + \note This is a RHI API with limited compatibility guarantees, see \l QShaderDescription for details. */ @@ -491,7 +491,7 @@ QT_BEGIN_NAMESPACE \brief Describes a built-in variable. - \note This a RHI API with limited compatibility guarantees, see \l QShaderDescription + \note This is a RHI API with limited compatibility guarantees, see \l QShaderDescription for details. */ |