There's no reason to release GstMemory manually at all.
If we do release GstMemory, corresponding GstBuffer will be
discarded by GstBufferPool baseclass because the size is changed
to zero.
Actual cause of heavy CPU usage in case of fixed-size pool
(i.e., decoder output buffer pool) and if we remove GstMemory from
GstBuffer is that GstBufferPool baseclass is doing busy wait in acquire_buffer()
for some reason. That needs to be investigated though, discarding
and re-alloc every GstBuffer is not ideal already.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4935>
The current way of using gst_video_info_set_format() will change all
fields of the GstVideoInfo. We only need to change its format, stride
and offset fields.
In order to keep the consistency with th common drm API, we rename the
gst_va_video_info_from_dma_info() into gst_va_dma_drm_info_to_video_info().
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4883>
As VK_FORMAT_FEATURE_2_xxx are defined as static const variable, the
vscoce C compiler prevents the initialization of the vk_usage_map
structure with error "C2099: initializer is not a constant".
Init the structure separately.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4904>
Adding GST_CUDA_CRITICAL_ERRORS env variable so that program can be
terminated on unrecoverable error.
Example)
GST_CUDA_CRITICAL_ERRORS=2,700 gst-launch-1.0 ...
In this example, CUDA_ERROR_OUT_OF_MEMORY(2) and
CUDA_ERROR_ILLEGAL_ADDRESS(700) are registered as critical error
and program will be aborted on those errors
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4729>
Change the internal GstVideoInfo structure in the GstVaDmabufAllocator to
GstVideoInfoDmaDrm in order to keep track of the exported DRM format by the
driver, and thus removing the DRMModifier quark attached as qdata in the
GstMemory. Though, the exposed API isn't updated yet; that has to go in a
second iteration.
Also this patch clean up some code (remove an unused buffer size assignation)
and fix some typos in documentation.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4821>
Some surface formats such as GST_VIDEO_FORMAT_Y42B and GST_VIDEO_FORMAT_RGB
can be created but can not be exported as DMA buffer. You can not say that
this is a driver bug because the driver may never want to share this kind of
surface out of libva.
And this function will be used to detect modifiers later, so the error message
will be annoying.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4821>
Decoder bounded CUDA memory is allocated by driver and the pool size
is fixed. Since we don't know how many buffers would be held by
downstream non-CUDA element, we should download such CUDA memory
and release it back to decoder.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4810>
New vulkan formats don't match the number of planes with the number of memories
attached to the buffer. This patch changes the pattern of using planes for
traverse the memories with the number of attached memories.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4351>
It's a generalization of the original gst_vulkan_get_or_create_image_view().
The reason for passing the whole VkImageViewCreateInfo structure rather than
just the missing fields, is because VkImageSubresourceRange and
VkComponentMapping can be different and those are most of VkImageViewCreateInfo.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4351>
This is going to be used when the pool is used by a video decoder for
VK_IMAGE_USAGE_VIDEO_DECODE_DST_BIT_KHR, since the frame allocation needs the
VkVideoProfileInfoKHR, and for that here GstCaps is used to wire it.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4351>
The specification says:
VUID-vkAllocateMemory-pAllocateInfo-01713
must pAllocateInfo->allocationSize be less than or equal to
VkPhysicalDeviceMemoryProperties::memoryHeaps[memindex].size where memindex =
VkPhysicalDeviceMemoryProperties::memoryTypes[pAllocateInfo->memoryTypeIndex].heapIndex
as returned by vkGetPhysicalDeviceMemoryProperties for the VkPhysicalDevice that
device was created from.
Though this can be catch by the validation layer, the requested frame size
depends on the use case so it's better to check this restriction by our code.
This patch also makes use of this new function to find memory type index,
and removes the unused function to find memory type index, which, as GstVulkan is
considered unstable, we can do it.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4351>
The purpose of this function is to get more info about the mapped Vulkan format
from the GStreamer format, since they can be multiple Vulkan formats for one
GStreamer format.
Also a Vulkan format may have certain usage and aspects that must be verified.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4351>
Originally the opened device only created one queue of one family queue, to say
graphics one. This approach felt short when other queue family is required not
shared with the graphics queue family, for example video decoding.
This new approach proposes to create those queues with supported families. For
now, only video decoding and encoder are created, if they are available.
In order to hold multiple queues opened, an array of VkDeviceQueueCreateInfo is
held along the live the device object, because it's used to traverse or get the
opened queues.
The algorithm to choose which queues create (or open) is to look for the queue
with more family bits, which also supports the one we are requesting, thus
minimizing the number of global queues of a certain family to create.
Nonetheless, the number of queues to open per family is set to be all of them,
widening the possibility of parallelism.
Also, this commit do a cosmetic refactor the assigning the physical device
nearer where it's used.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4351>
Also adds a meson option to enable them.
The symbol GST_VULKAN_HAVE_VIDEO_EXTENSIONS is an alias of
defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 238
if the option is allowed.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4351>
when play rtsp stream with playbin3 enabled, there are some critical logs:
g_object_get_is_valid_property: object class 'GstPlayBin3' has no property named 'n-video'
g_object_get_is_valid_property: object class 'GstPlayBin3' has no property named 'n-audio'
g_object_get_is_valid_property: object class 'GstPlayBin3' has no property named 'n-text'
self->collection could be NULL when READY->PAUSED if the pipeline
is live, then it will fallback to query playbin2's property,
we can call gst_play_streams_info_create_from_collection
directly, it will check self->collection internal.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4460>
This is no longer needed since the introduction of `gst_macos_main()` in 1.22.
Before that existed, we had a patch for GLib in Cerbero, which did work but made it
impossible to update GLib at all. The code being removed was a fail-safe in case of
running without said patch being applied. It's no longer needed, since for macOS
we just wrap our GStreamer with an NSApplication using `gst_macos_main()`.
Warnings will be displayed if no NSApp/NSRunLoop is found wherever needed,
pointing the user towards using the new API.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4366>
The addresses we get from `resolve_host_finish()` (via
`resolve_host_async()`, `resolve_host_main_cb()`, `on_resolve_host()`,
`g_resolver_lookup_by_name_finish()`) must be freed. Otherwise we leak
memory.
Leak found and confirmed fixed with GCC AddressSanitizer.
Change-Id: If32d24452d626234f01b253b77a7d6d16eac1cee
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4469>
Running element_vkcolorconver test with Vulkan validation layer this error is
raised:
Code 0 : Validation Error: [ VUID-VkMappedMemoryRange-size-01390 ] Object 0:
handle = 0x100000000010, type = VK_OBJECT_TYPE_DEVICE_MEMORY;
| MessageID = 0xdd4e6d8b
| vkFlushMappedMemoryRanges: Size in pMemRanges[0] is 0x4, which is not a
multiple of VkPhysicalDeviceLimits::nonCoherentAtomSize (0x40) and offset +
size (0x0 + 0x4 = 0x4) not equal to the memory size (0xb). The Vulkan spec
states: If size is not equal to VK_WHOLE_SIZE, size must either be a multiple of
VkPhysicalDeviceLimits::nonCoherentAtomSize, or offset plus size must equal the
size of memory
The reason of is that the image size used in the test doesn't comply hardware
restrictions. In order to avoid juggling with image size and hardware
restrictions, this patch proposes to use VK_WHOLE_SIZE macro.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4296>
While using the validation layer with this pipeline:
gst-launch-1.0 videotestsrc num-buffers=10 ! vulkanupload ! vulkancolorconvert ! vulkansink
The validation layer throws this message:
Code 0 : Validation Error: [ VUID-VkAttachmentDescription-format-06699 ]
Object 0: handle = 0x5555562e9610, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x52b3229e |
vkCreateRenderPass: pCreateInfo->pAttachments[0] format is
VK_FORMAT_B8G8R8A8_UNORM and loadOp is VK_ATTACHMENT_LOAD_OP_LOAD, but
initialLayout is VK_IMAGE_LAYOUT_UNDEFINED.
The Vulkan spec states: If format includes a color or depth aspect and loadOp is
VK_ATTACHMENT_LOAD_OP_LOAD, then initialLayout must not be VK_IMAGE_LAYOUT_UNDEFINED
When creating the render pass the loadOp can be either
`VK_ATTACHMENT_LOAD_OP_CLEAR` or `VK_ATTACHMENT_LOAD_OP_LOAD` depending on
`enable_clear`. While `enable_clear` is FALSE by default (which means
`VK_ATTACHMENT_LOAD_OP_LOAD`). Nonetheless, its value is explicitly changed by
`vkoverlaycompositor` to FALSE too!
This behavior was introduced in merge request #2470 where
`VK_ATTACHMENT_LOAD_OP_CLEAR` was a fixed value for loadOp. Thus, the bug
consists in a missing initialization of `enable_clear` to TRUE from that merge
request.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4296>
Specification says:
"""
engineVersion is an unsigned integer variable containing the developer-supplied
version number of the engine used to create the application.
"""
Assuming the engine is GStreamer, it would be expected to set its version as
engine version.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4243>
In the same spirit of libva-win32 elements this patch shows the driver of each
element in gst-inspect, giving more information to the user. This driver
description is parsed from vaQueryVendorString from mesa and intel drivers,
while copied as is for others. Also appends the render node for multi gpu
systems.
Fixes#2349
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4204>
The appropriate return value for incomplete NAL header should be
GST_H264_PARSER_NO_NAL_END. This tells the parser element to
gather more data. Previously, it would assume the NAL is corrupted
and would drop the data, potentially causing stream corruption.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3234>
Previously, reassigning loop index l in nicestream.c
could cause a segfault if l->data was null, as it could
reassign l to a null variable, triggering the loop
postassignment l->next, which then segfaults due to
l now being null. It is instead moved into the loop.
_delete_transport already performs the reassignment
inline.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4192>
This allow simplifying the GstVideoInfo handling in the sinks. Instead
of having to update a video info for the import, the sink can simply pass the
video info associated with the caps and rely on the VideoMeta in the GstBuffer
to obtain the appropriate offset and stride.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3801>
gstcudaloader.cpp defines GST_DEBUG_CATEGORY (gst_cudaloader_debug);
but it wasn't initializing it anywhere.
This caused the following error to be logged by gst-plugin-scanner when
libcuda.so.1/nvcuda.dll couldn't be loaded, e.g. in systems without
CUDA:
(gst-plugin-scanner:39618): GStreamer-CRITICAL **: 14:40:22.346:
gst_debug_log_full_valist: assertion 'category != NULL' failed
This patch fixes the bug by initializing the category in
gst_cuda_load_library_once_func() before any logging occurs.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4154>
GstBufferPool implementation was referenced for this GstD3D11PoolAllocator,
for example GstAtomicQueue, various atomic operations, and GstPoll ones.
However, such combination seems to be almost pointless
since gst_poll_{read,write}_control() takes mutex and also
GstPoll uses Win32 event handle internally.
Use simple SRWLOCK and CONDITION_VARIABLE instead, and don't make things
complicated/inefficient.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2887>
The `add_candidate` vfunc of the GstWebRTCICE interface gained a GstPromise
argument, which is an ABI break. We're not aware of any external user of this
interface yet so we think it's OK.
This change is useful in cases where the application needs to bubble up errors
from the underlying ICE agent, for instance when the agent was given an invalid
ICE candidate.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3960>
The av1decoder class does not implement the ->parse() virtual function,
and we always need to add the av1parse element before it. So we should
set_needs_format of the decoder to TRUE, then if no parse before it, it
can fail with a "not-negotiated" error early, rather than go on and
generate unexpected error.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4064>
The vp9decoder class does not implement the ->parse() virtual function,
and we always need to add the vp9parse element before it. So we should
set_needs_format of the decoder to TRUE, then if no parse before it, it
can fail with a "not-negotiated" error early, rather than go on and
generate unexpected error.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4064>
The vp8decoder class does not implement the ->parse() virtual function,
it can only accepts frame aligned data. If some element such as filesrc
feed it with unaligned data, the behaviour is undecided. So we should
set_needs_format of the decoder to TRUE, then it can fail with a
"not-negotiated" error early, rather than go on and generate unexpected
error.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4064>
The mpeg2decoder class does not implement the ->parse() virtual function,
and we always need to add the mpegvideoparse element before it. So we should
set_needs_format of the decoder to TRUE, then if no parse before it, it
can fail with a "not-negotiated" error early, rather than go on and
generate unexpected error.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4064>
The h264decoder class does not implement the ->parse() virtual function,
and we always need to add the h264parse element before it. So we should
set_needs_format of the decoder to TRUE, then if no parse before it, it
can fail with a "not-negotiated" error early, rather than go on and
generate unexpected error.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4064>
The h265decoder class does not implement the ->parse() virtual function,
and we always need to add the h265parse element before it. So we should
set_needs_format of the decoder to TRUE, then if no parse before it, it
can fail with a "not-negotiated" error early, rather than go on and
generate unexpected error.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4064>
This will be used for CUDA stream sharing.
* Adding GstCudaPoolAllocator object. The pool allocator will
control synchronization of allocated memory objects.
* Modify gst_cuda_allocator_alloc() API so that caller can specify/set
GstCudaStream object for the newly allocated memory.
* GST_CUDA_MEMORY_TRANSFER_NEED_SYNC flag is added in addition to
existing GST_CUDA_MEMORY_TRANSFER_NEED_{UPLOAD,DOWNLOAD}.
The flag indicates that any GPU command queued in the CUDA stream
may not be finished yet, and caller should take care of the
synchronization.
The flag is controlled by GstCudaMemory object if the memory holds
GstCudaStream. (Otherwise, GstCudaMemory will do synchronization
as before this commit). Specifically, GstCudaMemory object will set
the new flag automatically when memory is mapped with
(GST_MAP_CUDA | GST_MAP_WRITE) flags. Caller will need to unset
the flag via GST_MEMORY_FLAG_UNSET() if it's already synchronized
by client code.
* gst_cuda_memory_sync() helper function is added to perform synchronization
* Why not use CUevent object to keep track of synchronization status?
CUDA provides fence-like interface already via CUevent object,
but cuEventRecord/cuEventQuery APIs are not zero-cost operations.
Instead, in this version, the status is tracked by using map and
object flags.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3629>
Spec 7.1.3:
If a memory object does not have the VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
property, then vkFlushMappedMemoryRanges must be called in order to guarantee
that writes to the memory object from the host are made available to the host
domain, where they can be further made available to the device domain via a
domain operation. Similarly, vkInvalidateMappedMemoryRanges must be called to
guarantee that writes which are available to the host domain are made visible to
host operations.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3723>
_alloca CRT function is deprecated. Moreover, stack allocation
for string is not a good idea. We can use _malloca inline
function instead, but all use of _alloca in d3d11 library/plugin
are not performance critical path at all.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3652>
Systems like musl libc don't support ISO 6937 in iconv. This ensures
that the MPEG-TS plugin can cope with that. There is existing support
in the plugin for other methods, so it seems to have been the original
intent anyway.
Fixes: #1314
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3245>
Currently the element calls abort when failed to prepare reference
picture set. This can happent when the input stream is somehow
corrupted, like a rtsp strem with lost packets. Now it will only
return with GST_FLOW_ERROR instead of terminating whole process.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3505>
Because of the asynchronous resolving of mDNS ICE candidates it is
possible that GstWebRTCICE outlives webrtcbin. This in turn prolongs
the lifetime of the GstWebRTCNiceStream objects via refs in
nice_stream_map. Thus the GstWebRTCICETransport objects held in
GstWebRTCNiceStream may be invalid at the time they are accessed by
the _on_candidate_gathering_done() callback since GstWebRTCNiceStream
doesn't take a reference to them. Doing so would create a circular
reference, so instead this commit introduces weak references to the
transport objects and then we can check if the objects are valid before
accessing them.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3502>
In theory, input caps can be updated anytime at non-keyframe or
sequence boundary, such as HDR10 metadata, framerate, aspect-ratio
or so. Those information update might not trigger ::new_sequence()
or subclass may ignore the changes.
By this commit, input state change will be tracked by baseclass
and subclass will be able to know the non-decoding-essential
update by checking the codec specific picture struct
on ::output_picture()
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3328>
