Maker Chris Greening has been investigating the right way to get an Espressif ESP32 microcontroller to play again JPEG recordsdata sooner — being pointed to a intelligent method which makes use of single-instruction multiple-data (SIMD) extensions within the processor to dramatically increase decoding efficiency.

“A barely cryptic [GitHub] challenge appeared on the esp32-tv undertaking,” Greening explains in a put up delivered to our consideration by Adafruit. “JPEG decoding with SIMD,” GitHub person “modi12jin” wrote within the challenge in query, “at present the entire body, can’t be partial [frame], there will probably be extra sooner or later.”

Connected to the difficulty was a code snippet which used a unique method to decoding JPEG-format photographs, used within the esp32-tv undertaking to stream video from microSD card and show it on a coloration display. “That is fairly attention-grabbing,” Greening notes of the code provided. “SIMD (Single Instruction A number of Knowledge) can doubtlessly actually pace up decoding of JPEG recordsdata.”

The thought is easy, although it took just a few many years for laptop processors to realize the flexibility: if in case you have a complete bunch of information, just like the blocks which make up a JPEG picture, and you should run the identical instruction throughout all of it, SIMD gives an actual acceleration — within the case of the pattern code offered dropping the time required to decode a 272×233 JPEG picture on an Espressif ESP32-S3 rom 109ms with the JPEGDecoder library and a best-case 32ms with the JPEGDEC library to only 20ms.

Issues weren’t fairly that straightforward, sadly: Greening’s preliminary checks revolved round merely decoding the picture, however including the step of displaying it to the display noticed it beating JPEGDEC by a mere 1ms — 37ms to 38ms. The rationale: that warning from the code’s creator about it engaged on complete frames solely, stopping the microcontroller from working the direct reminiscence entry (DMA) show operations in parallel with the CPU decoding operations. As an alternative, the CPU has to decode the complete picture then push the entire thing over DMA.

“Initially this might sound a bit bit disappointing, however it’s really not the tip of the world — notably for out TV undertaking the place we’re streaming frames,” Greening explains. “We will decode a body utilizing the CPU and kick off the DMA switch to show the pixels. After which we are able to instantly kick off decoding the following body. This overlapped decode/show course of can provide us a very quick body fee.”

In testing, Greening discovered that switching to the SIMD-based JPEG decoder offered a lift to 40 frames per second on the identical {hardware} as earlier than — making the esp32-tv undertaking significantly smoother with no change to the microcontroller.

Greening’s full write-up is out there on his web site.