[v2,3/4] debayer_egl: Implement dmabuf import for input buffers

26770 diff mbox series

In many cases we can import the GPU-ISP input buffers, dmabufs from v4l2,
directly into EGL instead of mapping and uploading - i.e. copying - them.

Doing so can have positive effects in multiple areas, including reducing
memory bandwidth and CPU usage, as well as avoiding expensive dmabuf syncs
and syscalls.

The main reason direct imports may not work are the more demanding stride
alignment requirements many GPUs have - often 128 or 256 bytes - compared
to ISPs - apparently often closer to 32 bytes.

Thus we first try to import buffers directly and - if that fails - fall back
to the previous upload path. Failing imports should come at low cost as
drivers know the limitations and can bail out early, without causing
additional IO or context switches.

In the future we might be able to request buffers with a matching stride
from v4l2 drivers in many cases, making direct import the norm instead
of a hit-or-miss. An optional kernel API for that exists, but doesn't
seem to be implemented by any driver tested so far.

While on it, add some minor code adjustments to make it easier to
follow.

Signed-off-by: Robert Mader <robert.mader@collabora.com>
---
 src/libcamera/software_isp/debayer_egl.cpp | 32 ++++++++++++----------
 src/libcamera/software_isp/debayer_egl.h   |  2 +-
 2 files changed, 19 insertions(+), 15 deletions(-)

On 18/05/2026 21:15, Robert Mader wrote:
> +	if (egl_.createInputDMABufTexture2D(*eglImageBayerIn_, input->planes()[0].fd.get()) != 0) {
> +		LOG(Debayer, Debug) << "Importing input buffer with DMABuf import failed, falling back to upload";
> +
> +		inputBufferDmaSyncer->emplace(input->planes()[0].fd, DmaSyncer::SyncType::Read);
> +		MappedFrameBuffer in(input, MappedFrameBuffer::MapFlag::Read);
> +		if (!in.isValid()) {
> +			LOG(Debayer, Error) << "mmap-ing buffer(s) failed";
> +			return -ENODEV;
> +		}
> +		egl_.createTexture2D(*eglImageBayerIn_, in.planes()[0].data());
> +	}

So for this set a flag in the DebayerEGL class to flag the behavior - 
try once and then keep to the detected behaviour.

---
bod

On 19.05.26 17:58, Bryan O'Donoghue wrote:
> On 18/05/2026 21:15, Robert Mader wrote:
>> +	if (egl_.createInputDMABufTexture2D(*eglImageBayerIn_, input->planes()[0].fd.get()) != 0) {
>> +		LOG(Debayer, Debug) << "Importing input buffer with DMABuf import failed, falling back to upload";
>> +
>> +		inputBufferDmaSyncer->emplace(input->planes()[0].fd, DmaSyncer::SyncType::Read);
>> +		MappedFrameBuffer in(input, MappedFrameBuffer::MapFlag::Read);
>> +		if (!in.isValid()) {
>> +			LOG(Debayer, Error) << "mmap-ing buffer(s) failed";
>> +			return -ENODEV;
>> +		}
>> +		egl_.createTexture2D(*eglImageBayerIn_, in.planes()[0].data());
>> +	}
> So for this set a flag in the DebayerEGL class to flag the behavior -
> try once and then keep to the detected behaviour.

Did that in v3.

FTR.: I *think* the overhead of trying the import every time is 
negligible - that was the reason I didn't add such a flag in the first 
place.

2026. 05. 19. 17:58 keltezéssel, Bryan O'Donoghue írta:
> On 18/05/2026 21:15, Robert Mader wrote:
>> +	if (egl_.createInputDMABufTexture2D(*eglImageBayerIn_, input->planes()[0].fd.get()) != 0) {
>> +		LOG(Debayer, Debug) << "Importing input buffer with DMABuf import failed, falling back to upload";
>> +
>> +		inputBufferDmaSyncer->emplace(input->planes()[0].fd, DmaSyncer::SyncType::Read);
>> +		MappedFrameBuffer in(input, MappedFrameBuffer::MapFlag::Read);
>> +		if (!in.isValid()) {
>> +			LOG(Debayer, Error) << "mmap-ing buffer(s) failed";
>> +			return -ENODEV;
>> +		}
>> +		egl_.createTexture2D(*eglImageBayerIn_, in.planes()[0].data());
>> +	}
> 
> So for this set a flag in the DebayerEGL class to flag the behavior -
> try once and then keep to the detected behaviour.

I think that opens an interesting question. At the moment, in libcamera, applications
are not constrained in any way what buffers they might use. Meaning that some might be
appropriate, some might not. There is no "buffer import" step that would provide
information about the set of possible buffers. So disabling it on the first failure
might be a pessimization. But of course they will generally cycle through a single
set of buffers, so it's most likely fine to have a flag and stop trying after
the first failure.


> 
> ---
> bod
>