adding hints for the compiler for optimization.
In case your custom build complains about "const", just remove the keyword. based on e82f38e277, but going further :-)
* "const" class functions : function does not modify any class attributes ( --> "this" becomes const)
* __attribute__((pure)) : function return value depends only on the parameters and/or global variables. The function does not modify any global or static variables.
* __attribute__((const)) : function only examines arguments (no globals), and has no effects except the return value. This slightly more strict than "pure"
* hot: tells the compiler "this functions is called very often"
* cold: the opposite of hot
* mode_2Dfloatingblobs() improvements for large panels
* Segment::fillCircle() speed optimizations
* HUB75 hack (disabled by default): skip first fill(BLACK) when using double buffering (as the buffer gets cleared after each frame)
* cache width, height, and a few more values that are normally re-calculated for each pixel
* make normal sPC a bit faster
* Segment::fade_out() optimization
--> only active in FASTPATH mode, to preserve flash on small devices (8266)
--> code still needs some polishing
--> up to 20% speedup with some 2D effects (esp32, flash qio 80Mhz, -O2)
* optimized fadeToBlackBy() - don't repaint unchanged pixels
* made MIN_HEAP_SIZE configurable by build_flags
* specific error message when not enough memory for LEDs buffer
* fix compiler warnings (uninitialized vars, ambiguous functions calls)
* restore some lost function prototypes
* avoid negative pixel indices
* only use "fast" color_add when there is no risk of "overshooting" results
* minor optimizations
If the segment is wider than 20 pixels, we optimize calculations due to symmetry - for smaller arcs the result looks better without optimization.
As a side-effect, we have enough computing power left to go through the complete circumference, avoiding holes.
as it turns out, the "_t" functions (from wled_math.cpp) are about 3 times (!!!) slower than the standard functions.
* mapping modes : Arc and Circle
* effects: 2D Drift, 2D Drift Rose
* use "float" math only - sinf(), cosf(), roundf()
* use fewer "rays" for medium-sized matrix (<=32 pixels wide/high)
* ray drawing optimized to use fixed point
up to 80% faster on esp32 and esp32-S3; -S2/-C3 should also see benefits, as these do not have floating point support in hardware.
* File.readbytesuntil does not terminate strings. So the string buffer needs to be filled with zero's before reading.
* fix crashes (mem corruption) when ledmap file has too many / too few entries.
* initialize unused map places with "identity" (same led) mapping
before fix:
> Reading LED map from /ledmap1.json
> ("width": 60edmap1.json) ("height": 90edmap1.json)
> resetSegments 1 60x90
> allocLeds (0,0 to 60,90), 16200 from 0
> allocLeds (0,0 to 60,90), 16200 from 6
> deserializeMap 60 x 90 customMappingTable alloc 5400 from 0
after:
> Reading LED map from /ledmap1.json
> ("width": 60) ("height": 9)
> resetSegments 1 60x9
> allocLeds (0,0 to 60,9), 1620 from 0
> allocLeds (0,0 to 60,9), 1620 from 6
> deserializeMap 60 x 9
> deserializemap customMappingTable alloc 540 from 0
* currentBri() is called for any setpixelColor(), so we can speed up everything (a bit) by allowing the compiler to inline which saves a few cycles of call/return overhead.
* aligned the function with upstream, and added another optimization by only calling progress() when a transition is active.
this avoids heap corruption, by double-checking that "use global leds" is not configured, before trying to free ledsrgb[].
It is still a mystery why Segment::_globalLeds == nullptr.
if random palette was used in the startup preset, all LEDs were orange until the first palette cycling happened.
This fix ensures that there is no all-orange palette initially.
I've found a code spellchecker, so this is what can be corrected easily. Changes are only affecting comments, readme and a few user-visible strings. So no functional impact expected.
index.js
- genPresets: Temporary for deterministic effects test: Set to 11/Raibow instead of 1/Random smooth palette
FX.cpp
- replace all millis() by strip.now
cleanup:
- version increase
- clean up debug messages
- minor fixes in platformio.ini
- fix compilation with -D WLED_DISABLE_2D
helping the compiler to optimize:
- a few more __attribute__((pure))
- fx.cpp marked a few local functions as "static"
Segments are created/deleted on-the-fly; it seems that "local leds" buffers did not follow properly.
* revise segment copy/move constructors
* de-optimize use of local leds (as they need to be re-allocated when segment size changes)
* minor stability improvements
it seems that SEGMENT.blur() is the main bottleneck for many 2D effects.
This change optimizes performance of the function:
* avoid to re-write unchanged pixels
* early exit when blur_amount == 0 (=nothing to do)
* use _fast_ types where possible
I've seen up to 20% speedup with this change.
power estimation results from estimateCurrentAndLimitBri() were too low (example: estimated 1.3Amp, measured 1.6Amp). This change corrects the power calculation. Due to the changed behavior of getPixelColor, powerSum must be used as-is, not scaled down again by brightness.
