QVLA itself is non-relocatable due to self references. (ptr pointing
to array[Prealloc] as long as capacity < Prealloc)
Seems we shot ourselves in the foot in multiple places with this.
Pick-to: 6.2 6.2.0
Fixes: QTBUG-96619
Change-Id: I57a2ce539b671326cd352dbe57a1f3d4c46a6456
Reviewed-by: Tobias Koenig <tobias.koenig@kdab.com>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
...when starting a render/compute pass.
This matches most other backends in fact, the Vulkan backend has
just certain historical differences, and is complicated due to the
fact that it has the option of using secondary command buffers for
passes that specify ExternalContents (to support the case of wanting
to issue direct Vulkan commands in a code block surrounded by calls
to beginExternal and endExternal).
Not resetting state such as the currently bound index buffer when
starting a pass quickly blows up when two consecutive render passes
use different settings, one targeting the primary while the other
the secondary command buffer. Instead of further complicating the
logic, just reset the relevant state in every begin(Compute)Pass.
Comes with an autotest that is crafted so that it manages to
downright crash when run with Vulkan without the fix to the backend.
Fixes: QTBUG-89765
Pick-to: 6.2
Change-Id: I8dc47bd179c17d45a0556ec31200dc90c4b67ca5
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Follow what has been done for QRhiShaderResourceBindings. Have a way
to retrieve an opaque blob (that just happens to be a list of integers)
so that a simple == comparison can be used to determine compatibility
even when the objects from which the blob was retrieved are no longer
alive.
The contract is the following:
bool a = rp1->isCompatible(rp2);
bool b = rp1->serializedFormat() == rp2->serializedFormat();
assert(a == b);
Pick-to: 6.2
Change-Id: I45e7d05eeb6dfa2b2de474da0a0644912aaf174a
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
Be idiomatic and return the output iterator one past the last element.
Otherwise passing in a plain pointer (as exercised by the autotest now)
fails to function because we write over the same 4 elements again and
again for each binding.
Pick-to: 6.2
Change-Id: If74463fa5140ffa2b1d5be97b71868848ad46614
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
...by the Qt Quick renderer, for example.
A typical Qt Quick material binding set serializes to 8 uints. This
would not demand a container like QVector. However, being implicitly
shared is essential here due to the intended usage (query the
serialized blob, put it into a cache key, hash it, compare it, all
without any copying and new allocs; we can afford an extra alloc
upon each srb construction, but don't want more afterwards in the
rendering engines)
Also make it clear in the pipeline docs that the optimization Qt Quick
is (soon going to be) doing is legal. (the srb ref in the pipeline can
be dead and dangling as long as every call to setShaderResources()
specifies a layout-compatible alternative)
Pick-to: 6.2
Change-Id: I97efbea1fa3516b10c9832adbab0a21b7bc0845d
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
One is a bad application or library in this case, but nonetheless
we should handle this more gracefully then just crashing due to
the QRhi already having been destroyed. Mainly because in Qt 5 one
could get away with the same: releasing OpenGL objects underneath,
for example, a QSGPlainTexture with no (or wrong) GL context did
not generate any user visible fatal errors. So we should not crash
in Qt 6 either with these code bases.
In debug builds or when QT_RHI_LEAK_CHECK is set, one will get the
unreleased resources warning printed in Qt 6, which is a step
forward compared to Qt 5. So there is still some indication that
something is badly designed, even if the application survives.
Task-number: QTBUG-95394
Pick-to: 6.2
Change-Id: I944f4f425ff126e7363a82aff926b280ccf1dfc3
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Supported on OpenGL (and ES) 3.0+ and everywhere else.
Can also be a render target, targeting a single slice at a time.
Can be mipmapped, cannot be multisample.
Reading back a given slice from a 3D texture is left as a future
exercise, for now it is documented to be not supported.
Upload is going to be limited to one slice in one upload entry,
just like we specify one face or one miplevel for cubemap and
mipmapped textures.
This also involves some welcome hardening of how texture subresources
are described internally: as we no longer can count on a layer index
between 0..5 (as is the case with cubemaps), simply arrays with
MAX_LAYER==6 are no longer sufficient. Switch to sufficiently dynamic
data structures where applicable.
On Vulkan rendering to a slice needs Vulkan 1.1 (and 1.1 enabled on the
VkInstance).
Task-number: QTBUG-89703
Change-Id: Ide6c20124ec9201d94ffc339dd479cd1ece777b0
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Normally we only allow creating wrappers for texture objects. These
can then be used with a QRhiTextureRenderTarget to allow rendering into
an externally created texture.
With OpenGL (ES), there are additional, special cases, especially on
embedded. Consider EGLImages for example. An EGLImageKHR can be bound to
a renderbuffer object (glEGLImageTargetRenderbufferStorageOES), which
can then be associated with a framebuffer object to allow rendering into
the external buffer represented by the EGLImage. To implement the same
via QRhi one needs a way to create a wrapping QRhiRenderBuffer for the
native OpenGL renderbuffer object.
Here we add a createFrom() to QRhiRenderBuffer, while providing a dummy,
default implementation. The only real implementation is in the OpenGL
backend, which simply takes a renderbuffer id, without taking ownership.
Task-number: QTBUG-92116
Change-Id: I4e68e665fb35a7d7803b7780db901c8bed5740e2
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Following patterns from the other backends is insufficient with OpenGL
because we do not use real uniform buffers. There is currently a
possibility that a shader program will be bound without following it
with setting uniforms. Correct this by having a second level of tracking
of the associated srb object in the pipelines.
Pick-to: 6.0 6.1
Fixes: QTBUG-91630
Change-Id: I74a012daade826dd22c436bde06381c1233bad11
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Reviewed-by: Eirik Aavitsland <eirik.aavitsland@qt.io>
Add QRhi APIs to retrieve and reload the contents of the "pipeline
cache".
The only API where there is a true pipeline cache is object is Vulkan
(VkPipelineCache). For OpenGL, the other backend where we support this,
it is simulated with program binaries. The Qt 5 style OpenGL program
binary disk cache continues to work like before, but one has now the
option to do things in a more modern, graphics API agnostic way, that
leads to generating a single blob instead of a large set of files in
some system location, allowing easier "pre-baking" of the cache content.
It is expected that Qt Quick exposes the two new functions in form
if QSG_RHI_ environment variables, thus allowing easy testing and
cache file generation.
As an example for the performance improvements this can give, consider
Vulkan, where we do not have any existing persistent caching mechanism
in place:
Running BenchmarkDemoQt6.exe --scene flythrough --mode demo creates 18
QRhiGraphicsPipeline objects from Qt Quick and Qt Quick 3D.
The total time spent in QRhiGraphicsPipeline::create() during application
startup for these 18 pipelines is 35-40 ms on a given Windows (NVIDIA)
system.
When exporting the pipeline cache contents to a file, and then, in a
subsequent run, reloading the cache contents, this is reduced to 5-7 ms
on the same system, meaning we get a 6-7x improvement.
The generated data is always specific to a given Qt version, RHI
backend, graphics device, and driver version. Much of the implementation
consists of adding and verifying the appropriate header to the blobs
retrieved from the driver, to allow gracefully ignoring data that was
generated with a device or driver that differs from the one used at
run time. This should provide robustness, even if the Vulkan or OpenGL
implementation is for some reason not prepared to identity and reject
incompatible cache/program blobs.
Fixes: QTBUG-90398
Change-Id: I67b197f393562434f372c7b7377f638abab85cb3
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Instead, have a static function in QRhiVulkanInitParams then Qt Quick
and anyone else who creates a QVulkanInstance that is then used in
combination with QRhi can query.
Change-Id: I046e0d84541fc00f5487a7527c97be262221527f
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
...to the extent it is sensible. We have to make compromises still,
meaning some fields will only be applicable with certain APIs.
Most of this is already shown upon QRhi::create() as info debug
prints, when enabled. Now expose it all through the QRhi API as
well.
This is useful for printing in qtdiag, and, while it should be
avoided as much as possible, to make decisions about disabling
3D rendering features depending on the driver and GPU in use.
Change-Id: Iebe1e192965c928b82a094d1c7c50ddf4b38b9a2
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Complete search and replace of QtTest and QtTest/QtTest with QTest, as
QtTest includes the whole module. Replace all such instances with
correct header includes. See Jira task for more discussion.
Fixes: QTBUG-88831
Change-Id: I981cfae18a1cabcabcabee376016b086d9d01f44
Pick-to: 6.0
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
This patch disables four failing unit tests when executed with the
offscreen backend.
Change-Id: Ie67341b886984e6de19cd8dd8a8a237a620a1b7a
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
The original restriction to UniformBuffer was due to the GL backend
where there is no GL buffer object for QRhiBuffers with usage
UniformBuffer. However, we can still implement this for cases when
there is a true GL buffer object underneath. With other backends it
should all work as-is already.
This becomes useful when one has buffers with usage Vertex that need
full updates every frame. (f.ex. instance data)
Unfortunately this involves renaming the function. But while at it, add
an autotest case as well.
Change-Id: Iff59e4509a8bae06654cc92fe8428bd79eb012fb
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Reports on the Web have it that there's nothing guaranteeing a
driver does any actual freeing of resources then doing
vkFreeCommandBuffer for a command buffer from a command pool that
does not have VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, thus
leading to continuously growing resource usage with our current
allocate/free pattern. It could be that this is the source of out
of memory problems we are seeing on some Android devices.
Instead of just going straight for said command pool flag and doing
ResetCommandBuffer for the command buffers individually, take one
step further and use per-slot (slot being 0 or 1 if QVK_FRAMES_IN_FLIGHT
is 2) command pools. The current pool is reset in each
beginFrame/beginOffscreenFrame, moving all allocated command buffers
to the initial state (while other command buffers with the other pool
are not affected).
This may be (while impossible to tell from just guessing based on the
spec) our best approach to command buffer allocation since a Vulkan
implementation can likely just use some simple per pool allocator,
knowing that we never want to free or reset individual command buffers,
but we rather only reset the whole pool at once.
The option of importing an existing VkCommandPool when creating the
QRhi instance is now gone, but there was probably no point in offering
that in the first place.
When it comes to VK_COMMAND_POOL_RESET_RELEASE_RESOURCES_BIT it will
not be set unless releaseCachedResources() (in Qt Quick this is hooked
into QQuickWindow::releaseResources()) was called. What this does in
practice is unknown, but have an option to set it now and then if the
application really wants.
While we are at it, rename secondaryCbs to activeSecondaryCbStack to
indicate what it really is. (it's a stack as each call to
beginExternal() pushes a new one, while each endExternal() pops)
Change-Id: I2e5c1fad26d794e1f56c778e38f750998d706d84
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
ExternalContentsInPass becomes a per-pass flag now. Why is this
beneficial? Because while Qt Quick has no choice for its render
pass, not being able to guess if the application wants to do some
native rendering in there, Quick 3D's render passes, all the ones
that are under Quick3D's control, do not have native rendering
from the application in them, and so using secondary command
buffers with Vulkan is not necessary.
Introduce something similar for compute and OpenGL. By knowing that
none of the resources used in a pass are used with a compute pass
(e.g. because we know that there are no compute passes at all) a small
amount of time can be saved by skipping tracking buffers and textures
because the only purpose of said tracking is to generate barriers that
are relevant only to compute.
Change-Id: I0eceb4774d87803c73a39db527f5707a9f4d75c1
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
It is a valid use case to have a vertex shader with no
vertex attributes. Using gl_VertexID or gl_InstanceID
is enough to generate vertex data out of thin air.
Change-Id: If7689914624a84723923fd2d7aef355bda592e24
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
In this case the srb represents the layout only, and can still be used
to create a pipeline. For setShaderResources() one will then need to use
another, layout compatible, srb that references valid resources.
Change-Id: I3ea5b63df3be8847540ca4c0c40fbd29dbed8fb7
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Required by OpenXR. A VkPhysicalDevice or an adapter LUID + feature
level pair should be adoptable while leaving the rest (device, queue,
etc. setup) to QRhi as normal.
Change-Id: Iada0972671b037b4efb03e7831b7c9b8c5f2393d
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Macros and the await helper function from qfunctions_winrt(_p).h are
needed in other Qt modules which use UWP APIs on desktop windows.
Task-number: QTBUG-84434
Change-Id: Ice09c11436ad151c17bdccd2c7defadd08c13925
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
For historical reasons we use build and release instead of create and
destroy. This becomes confusing now that more modules in Qt start taking
QRhi into use. Migrate to the more familiar naming, so those who have
used QWindow or QOpenGLContext before will find it natural.
Change-Id: I05eb2243ce274c59b03a5f8bcbb2792a4f37120f
Reviewed-by: Eirik Aavitsland <eirik.aavitsland@qt.io>
When storing a void* pointer to the texture handle, we had
to ensure that the variable would exist until the build phase,
which is error prone and caused errors in QQuickWidget because
we copied the texture ID from the FBO into a local variable
before passing it into QQuickWindow::setRenderTarget().
The reason for using a void* was that we cannot know the width
of the handles in the different backends, but we do know that
they are 64-bit at maximum, so instead of storing potentially
dangling pointers, we just make it a 64-bit integer and cast
it back and forth in the backends.
Task-number: QTBUG-78638
Change-Id: I7951e24351ddb209045ab6197d81eb1290b4da67
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
The native command buffer handle was not updated, so the subsequent
finish() call attempted to record an invalid VkCommandBuffer. The
problem was not present with offscreen frames, only when finish() is
called with a swapchain-based frame active.
Task-number: QTBUG-84066
Change-Id: I9c4cb701c3dbbc28f237d6ae1cbf65aafd1fa95f
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
Also extend autotesting, both for rendering into a given mip level
and for rendering into a given cubemap face.
Change-Id: Ida94b71150477ceb50a3b5616d8b7be13174558b
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Introduces a new QRhiShaderResourceBinding function that takes an array
of texture-sampler pairs. The existing function is also available and is
equivalent to calling the array-based version with array size 1.
It is important to note that for Metal one needs MSL 2.0 for array of
textures, so qsb needs --msl 20 instead of --msl 12 for such shaders.
Comes with an autotest, and also updates all .qsb files for said test
with the latest shadertools.
Task-number: QTBUG-82624
Change-Id: Ibc1973aae826836f16d842c41d6c8403fd7ff876
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
Modeled after QRhiTexture's NativeTexture query.
This becomes valuable in advanced cases of integrating external native
rendering code with Qt Quick(3D), because it allows using (typically
vertex and index) buffers created by Quick(3D) in the custom renderer as
well, without having to duplicate the content by manually creating native
buffers with the same vertex and index data.
Change-Id: I659193345fa1dfe6221b898043f0b75ba649d296
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Add a new queriable resource limit value MaxAsyncReadbackFrames. Change
the autotest to rely on this instead of relying on the unspecified,
works-by-accident relation between readbacks and FramesInFlight. This
way even if the behavior diverges in some backend in the future, clients
(well written ones, that is), will continue to function correctly.
Also clarify the docs for FramesInFlight, and change d3d and gl to return
the correct value (which is 1 from QRhi perspective; the expanded docs now
explain a bit what this really means and what it does not).
Change-Id: I0486715570a9e6fc5d3dc431694d1712875dfe01
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
The new version takes/returns a value that can be unpacked and passed to
other functions without knowing which backend is in use.
The old API will be removed in a later change when dependent modules have
been updated
Task-number: QTBUG-78570
Change-Id: I18d928ceef3cb617c0c509ecccb345551a7990af
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
For Metal and Vulkan this needs actual work because that's where
the concept of renderpass descriptors is relevant. GL and D3D can
just return true always.
The big benefit of this is that Qt Quick can now compare renderpass
descriptors via isCompatible() for its pipeline cache (similarly to
how it is already using isLayoutCompatible() for srbs), and so
renderpass descriptors for layers (Item.layer, ShaderEffect) will
typically be compatible and so can pick up pipelines created by other
layers from the cache.
Also add autotests for shader resource binding and renderpass descriptor
compatibility.
Task-number: QTBUG-80318
Change-Id: I0008bc51c4ee13b0113d2c8caf799e1257f18a18
Reviewed-by: Paul Olav Tvete <paul.tvete@qt.io>
While we are at it, remove the Border and MirrorOnce wrap modes that have
not been supported on OpenGL, because they are unsupported with Metal+iOS
as well.
Task-number: QTBUG-78580
Change-Id: I0db94b9d3a6125b3bb5d7b1db5d02a42cd94d2c2
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Call adjustedFormat() as advised by the docs: "Applications are advised
to set this format on their QWindow in order to avoid potential BAD_MATCH
failures."
Task-number: QTBUG-79659
Change-Id: Ibf415fb0ee64bdd3f01d4ba744244bce811c0d27
Reviewed-by: Paul Olav Tvete <paul.tvete@qt.io>
Also improve (docs and runtime checks) and test the minimum set
of required data to create a graphics pipeline.
Task-number: QTBUG-78971
Change-Id: If5c14f1ab1ff3cf70f168fde585f05fc9d28ec91
Reviewed-by: Paul Olav Tvete <paul.tvete@qt.io>
The joys of "level - Specifies the mipmap level of the texture
image to be attached, which must be 0." for glFramebufferTexture2D
in OpenGL ES 2.0.
Change-Id: Iaf19502f48d7ba73b26abb72535bfa6696a1e182
Reviewed-by: Paul Olav Tvete <paul.tvete@qt.io>
...and make the Null backend able to deal with these, for RGBA8 textures
at least. Naturally it is all QImage and QPainter under the hood.
Also fix a bug in the OpenGL backend, as discovered by the autotest:
the size from the readback did not reflect the mip level.
Task-number: QTBUG-78971
Change-Id: Ie424b268bf5feb09021099b67068f4418a9b583e
Reviewed-by: Paul Olav Tvete <paul.tvete@qt.io>
