[libcamera-devel] [PATCH v6 3/3] libcamera: Add exportFrameBuffers in HeapAllocator
Cheng-Hao Yang
chenghaoyang at chromium.org
Wed Aug 2 09:08:05 CEST 2023
Thanks Laurent,
On Tue, May 30, 2023 at 2:56 PM Laurent Pinchart <
laurent.pinchart at ideasonboard.com> wrote:
> On Mon, May 29, 2023 at 10:02:18AM +0200, Jacopo Mondi via libcamera-devel
> wrote:
> > Hi Harvey
> >
> > On Mon, May 22, 2023 at 08:35:08AM +0000, Harvey Yang via
> libcamera-devel wrote:
> > > From: Cheng-Hao Yang <chenghaoyang at chromium.org>
> >
> > This is the same email address with 2 different names. Unless this is
> > intentional could you change the authorship of the patch and use a
> > single identity (git commit --amend --autor="...")
> >
> > > Add a helper function exportFrameBuffers in HeapAllocator to make it
> > > easier to use.
> > >
> > > Signed-off-by: Harvey Yang <chenghaoyang at chromium.org>
> > > ---
> > > include/libcamera/internal/heap_allocator.h | 9 +++
> > > src/libcamera/heap_allocator.cpp | 62 +++++++++++++++++++++
> > > 2 files changed, 71 insertions(+)
> > >
> > > diff --git a/include/libcamera/internal/heap_allocator.h
> b/include/libcamera/internal/heap_allocator.h
> > > index 92d4488a..92beaaa4 100644
> > > --- a/include/libcamera/internal/heap_allocator.h
> > > +++ b/include/libcamera/internal/heap_allocator.h
> > > @@ -9,12 +9,16 @@
> > > #pragma once
> > >
> > > #include <stddef.h>
> > > +#include <vector>
> > >
> > > #include <libcamera/base/unique_fd.h>
> > >
> > > namespace libcamera {
> > >
> > > +class Camera;
> > > +class FrameBuffer;
> > > class Heap;
> > > +class Stream;
> > >
> > > class HeapAllocator
> > > {
> > > @@ -24,7 +28,12 @@ public:
> > > bool isValid() const;
> > > UniqueFD alloc(const char *name, std::size_t size);
> > >
> > > + int exportFrameBuffers(Camera *camera, Stream *stream,
> > > + std::vector<std::unique_ptr<FrameBuffer>>
> *buffers);
> > > +
> > > private:
> > > + std::unique_ptr<FrameBuffer> createBuffer(Stream *stream);
> > > +
> > > std::unique_ptr<Heap> heap_;
> > > };
> > >
> > > diff --git a/src/libcamera/heap_allocator.cpp
> b/src/libcamera/heap_allocator.cpp
> > > index 8f99f590..e99cdbe5 100644
> > > --- a/src/libcamera/heap_allocator.cpp
> > > +++ b/src/libcamera/heap_allocator.cpp
> > > @@ -22,6 +22,12 @@
> > >
> > > #include <libcamera/base/log.h>
> > >
> > > +#include <libcamera/camera.h>
> > > +#include <libcamera/framebuffer.h>
> > > +#include <libcamera/stream.h>
> > > +
> > > +#include "libcamera/internal/formats.h"
> > > +
> > > namespace libcamera {
> > >
> > > /*
> > > @@ -227,4 +233,60 @@ UniqueFD HeapAllocator::alloc(const char *name,
> std::size_t size)
> > > return heap_->alloc(name, size);
> > > }
> > >
> > > +int HeapAllocator::exportFrameBuffers([[maybe_unused]] Camera
> *camera, Stream *stream,
> > > +
> std::vector<std::unique_ptr<FrameBuffer>> *buffers)
>
> Why do you pass a camera pointer to this function if it's unused ?
>
>
Removed.
> > > +{
> > > + unsigned int count = stream->configuration().bufferCount;
>
> You're creating a tight dependency between the allocator and the stream
> here, as you require the stream to have already been configured, and
> rely on the current configuration. Passing a stream configuration
> explicitly would be better to make the allocator more generic.
>
>
Good point. Using `const StreamConfiguration&` directly.
> > > +
> > > + /** \todo Support multiplanar allocations */
>
> I'd like to see this being addressed already, as it's a key requirement
> that will show whether the API design is good or not.
>
>
IIUC, this means the same thing as `a dedicated fd per plane`, right?
Let me know if I misunderstood.
> > > +
> > > + for (unsigned i = 0; i < count; ++i) {
> > > + std::unique_ptr<FrameBuffer> buffer = createBuffer(stream);
> > > + if (!buffer) {
> > > + LOG(HeapAllocator, Error) << "Unable to create
> buffer";
> > > +
> > > + buffers->clear();
> > > + return -EINVAL;
> > > + }
> > > +
> > > + buffers->push_back(std::move(buffer));
> > > + }
> > > +
> > > + return count;
> > > +}
> > > +
> > > +std::unique_ptr<FrameBuffer> HeapAllocator::createBuffer(Stream
> *stream)
> > > +{
> > > + std::vector<FrameBuffer::Plane> planes;
> > > +
> > > + unsigned int frameSize = stream->configuration().frameSize;
> > > + int pageSize = getpagesize();
> > > + /** Allocation size need to be a direct multiple of |pageSize|. */
> >
> > No need for /**, just use /*
> >
> > > + unsigned int numPage = (frameSize + pageSize - 1) / pageSize;
>
> What are the pros and cons for rounding it up to the page size here
> compared to the alloc() function ? Does the kernel require a rounded
> size, of will it round internally ?
>
>
Moved to `HeapAllocator::alloc` to prevent misuse in other components.
Yeah, the kernel requires a rounded size or it fails (in my environment with
udmabuf).
> > +
> > > + UniqueFD fd = alloc("FrameBuffer", numPage * pageSize);
> > > + if (!fd.isValid())
> > > + return nullptr;
> > > +
> > > + auto info =
> PixelFormatInfo::info(stream->configuration().pixelFormat);
> > > + SharedFD shared_fd(std::move(fd));
> >
> > s/shared_fd/sharedFd/
> >
> > > + unsigned int offset = 0;
> > > + for (size_t i = 0; i < info.planes.size(); ++i) {
> > > + unsigned int planeSize =
> info.planeSize(stream->configuration().size, i);
>
> This doesn't take line stride into account.
>
>
IIUC, `PixelFormatInfo::planeSize` considers stride in the
implementation...?
https://git.libcamera.org/libcamera/libcamera.git/tree/src/libcamera/formats.cpp#n1022
> > > + if (planeSize == 0)
> > > + continue;
> > > +
> > > + /** \todo We don't support allocating buffers with a
> dedicated fd per plane */
>
> This then makes the buffer incompatible with V4L2, that seems to be a
> bad issue for a generic allocator in libcamera.
>
>
Updated. Please check if it makes sense.
> > > + FrameBuffer::Plane plane;
> > > + plane.fd = shared_fd;
> > > + plane.offset = offset;
> > > + plane.length = planeSize;
> > > + planes.push_back(std::move(plane));
> > > +
> > > + offset += planeSize;
> > > + }
> > > +
> > > + return std::make_unique<FrameBuffer>(planes);
> > > +}
> > > +
> >
> > Only minors, the patch looks good. The series looks good as well,
> > please add documentation and fix all the minors and I think we can
> > collect it :)
>
> I'm afraid I disagree. See my comment in patch 2/3, in addition to the
> above.
>
> > > } /* namespace libcamera */
>
> --
> Regards,
>
> Laurent Pinchart
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.libcamera.org/pipermail/libcamera-devel/attachments/20230802/a8e8692e/attachment.htm>
More information about the libcamera-devel
mailing list