1.4 KiB
1.4 KiB
Estimating fillrate
Here are some results of simple tests on filling DMA buffer with data. Filling DMA buffer requires lots of memory operations on a bit level rather than doing simple byte/word wide store and copy. And it looks like it's quite a task both for esp32 core and compiler. I've done this while optimizing loops and bit logic along with testing compiler results.
So the testbed is:
- Matrix modules: 4 x FM6126A based 64x64 modules chained in 256x64
A testpatterns sketch:
- allocating single DMA buffs for 256x64
- allocating (NUM_LEDS*3) bytes for CRGB buffer
- measuring microseconds for the following calls:
- clearScreen() - full blanking
- fillScreenRGB888() with monochrome/gray colors
- fill screen using drawPixel()
- filling some gradient into CRGB buff
- painting CRGB buff into DMA buff with looped drawPixelRGB888()
- drawing lines
| clearScreen() | drawPixelRGB888(), ticks | fillScreen() | fillScreen with a drawPixel() | fillRect() over Matrix | V-line with drawPixel | fast-V-line | H-line with drawPixel | fast-H-line | |
|---|---|---|---|---|---|---|---|---|---|
| v1.2.4 | 1503113 ticks | 9244 non-cached, 675 cached | 1719 us, 412272 t | 47149 us, 11315418 ticks | - | 24505 us, 5880209 ticks | - | 24200 us | - |
| FastLines | 1503113 ticks | 1350 non-cached, 405 cached | 1677 us, 401198 t | 28511 us, 6841440 ticks | 10395 us | 14462 us, 3469605 ticks | 10391 us, 2492743 ticks | 14575 us | 5180 us, 1242041 ticks |
to be continued...