FBI is hardware accelerated?

[matteoraggi]

FBI is hardware accelerated for the images like omxplayer is hardware accelerated for videos?

[jamesh]

No, it's all done in software I believe. The HW JPEG decoder isn't actually any faster than the faster ARM's on recent Pi's at decoding JPEG, once you take in the data transfers time to and from the GPU.

[Mortimer]

Even if it isn't actually any faster, presumably it is still beneficial to use it if doing lots of conversions so the ARM remains free to do other things that the GPU can't?

[jamesh]

Even if it isn't actually any faster, presumably it is still beneficial to use it if doing lots of conversions so the ARM remains free to do other things that the GPU can't?

There's quite an overhead in getting the data to and from the GPU (a chunk of SDRAM bandwidth), so its going to be a pretty specific use case where, with 4 fast ARM cores, it's better to use the GPU. AUIU, the latest ARMs are quite a bit faster.

[6by9]

Even if it isn't actually any faster, presumably it is still beneficial to use it if doing lots of conversions so the ARM remains free to do other things that the GPU can't?

There's quite an overhead in getting the data to and from the GPU (a chunk of SDRAM bandwidth), so its going to be a pretty specific use case where, with 4 fast ARM cores, it's better to use the GPU. AUIU, the latest ARMs are quite a bit faster.

IIRC On a previous chip with dual 1.2GHz ARM A9's, the threshold was at around XGA.
It also depends what you want to then do with the image. If you can pass it directly to the display without it leaving the GPU then you can save some more performance as the ARM doesn't have to resize the image to fit the frame buffer. Horses for courses though as that requires a fair amount of Pi specfiic code to be written.

[matteoraggi]

I just want to do a slideshow of 10-150 .png images 1920x1080or 1080x1920 with a blend transition of half second. So the blend transition risk to be scattering. Not smooth cause of the not accelerated hardware?

[tvjon]

...The HW JPEG decoder isn't actually any faster than the faster ARM's on recent Pi's at decoding JPEG, once you take in the data transfers time to and from the GPU.

That has not been my experience...

Viewing the contents of a 30000+ folder of not too big photo's, Broadcom's jpeg decoder is significantly faster. Holding down the relevant "next" key is like watching a video.

Once upon a time what bits of code I wrote were designed to be as general purpose as practical. For the last few years it's been *only* able to run on ARM®, & more specifically some bits just on Broadcom's VC4.

[6by9]

I just want to do a slideshow of 10-150 .png images 1920x1080or 1080x1920 with a blend transition of half second. So the blend transition risk to be scattering. Not smooth cause of the not accelerated hardware?

Png decode is not hardware accelerated even in the gpu. In places it uses the gpu's vector processor (equivalent in neon) , but not dedicated hardware.

[matteoraggi]

Sowhich image format can be hardware accelerated??

[6by9]

Sowhich image format can be hardware accelerated??

JPEG.

[Gavinmc42]

h.264 was the only video hardware acceleration, but I think mjpeg might be too now?
Jpeg is the only still image acceleration.
These are done with the VC4.

But I think they are explaining that the new 3B+ quad core can do it faster in software.
Is there any particular library that is faster than others?

[jamesh]

...The HW JPEG decoder isn't actually any faster than the faster ARM's on recent Pi's at decoding JPEG, once you take in the data transfers time to and from the GPU.

That has not been my experience...

Viewing the contents of a 30000+ folder of not too big photo's, Broadcom's jpeg decoder is significantly faster. Holding down the relevant "next" key is like watching a video.

Really? I would expect a NEON accelerated JPEG decode to be faster than the HW decoder. prerhaps just for larger images? What library were you using for SW decode?

[6by9]

h.264 was the only video hardware acceleration, but I think mjpeg might be too now?

For video formats, H264, MPEG4, H263, MPEG2, and VC1 all have hardware acceleration. They are pretty much extensions of each other (H264 being the most complex), so the operations are similar.
MJPEG is a concatenation of JPEGs, therefore reuses the JPEG hardware block.
VP6 and VP8 decode have vector implementations but no dedicated hardware blocks.

Jpeg is the only still image acceleration.

JPEG is the only format using dedicated hardware blocks.
Again PNG, GIF, BMP, TGA, TIF, and PPM all have some level of vector acceleration.