Jan/WLED_MM_Infinity
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

bus.setPicelColor optimization: optimizing some CPU cycles out of the hot path in sPC and gPC

Browse Source 
* optimize loops that scan through all busses
* small speedups for  Bus::autoWhiteCalc()
* small speedups for ColorOrderMap::getPixelColorOrder()

thanks to github Copilot for giving me the right ideas for this optimization
This commit is contained in:
Frank
2025-10-20 21:58:23 +02:00
parent 6e8ab94ee9
commit 541b201855
2 changed files with 67 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

107
wled00/bus_manager.cpp
View File

@@ -114,31 +114,40 @@ void ColorOrderMap::add(uint16_t start, uint16_t len, uint8_t colorOrder) {
_count++;
}
uint8_t IRAM_ATTR ColorOrderMap::getPixelColorOrder(uint16_t pix, uint8_t defaultColorOrder) const {
uint8_t __attribute__((hot)) ColorOrderMap::getPixelColorOrder(uint16_t pix, uint8_t defaultColorOrder) const {
if (_count == 0) return defaultColorOrder;
// upper nibble contains W swap information
uint8_t swapW = defaultColorOrder >> 4;
for (uint8_t i = 0; i < _count; i++) {
if (pix >= _mappings[i].start && pix < (_mappings[i].start + _mappings[i].len)) {
return _mappings[i].colorOrder | (swapW << 4);
// upper nibble contains W swap information // WLEDMM optimization: avoid shifting >>4 and later undo by <<4
uint8_t swapW = defaultColorOrder & 0xF0;
// Scan mappings, using unsigned range test: pix in [start, start+len)
for (uint_fast8_t i = 0, n = _count; i < n; i++) { // WLEDMM small speedup, by avoiding repeated class member access
const auto &m = _mappings[i]; // WLEDMM help the compiler to optimize
if ((uint16_t)(pix - m.start) < m.len) { // True iff m.len > 0 and pix >= m.start and pix < m.start + m.len
return (m.colorOrder & 0x0F) | swapW; // add W swap information
}
}
return defaultColorOrder;
}
uint32_t Bus::autoWhiteCalc(uint32_t c) const {
uint8_t aWM = _autoWhiteMode;
if (_gAWM != AW_GLOBAL_DISABLED) aWM = _gAWM;
uint32_t __attribute__((hot)) Bus::autoWhiteCalc(uint32_t c) const {
uint8_t aWM = (_gAWM != AW_GLOBAL_DISABLED) ? _gAWM : _autoWhiteMode;
if (aWM == RGBW_MODE_MANUAL_ONLY) return c;
uint8_t w = W(c);
uint_fast8_t w = W(c);
//ignore auto-white calculation if w>0 and mode DUAL (DUAL behaves as BRIGHTER if w==0)
if (w > 0 && aWM == RGBW_MODE_DUAL) return c;
uint8_t r = R(c);
uint8_t g = G(c);
uint8_t b = B(c);
if (aWM == RGBW_MODE_MAX) return RGBW32(r, g, b, r > g ? (r > b ? r : b) : (g > b ? g : b)); // brightest RGB channel
w = r < g ? (r < b ? r : b) : (g < b ? g : b);
uint_fast8_t r = R(c);
uint_fast8_t g = G(c);
uint_fast8_t b = B(c);
// brightest RGB channel
if (aWM == RGBW_MODE_MAX) { // WLEDMM use max() instead of several nested conditions
w = max(r, g);
w = max(w, b);
return RGBW32(r, g, b, w);
}
// Other modes: smallest RGB channel // WLEDMM use min() instead of several nested conditions
w = min(r, g);
w = min(w, b);
if (aWM == RGBW_MODE_AUTO_ACCURATE) { r -= w; g -= w; b -= w; } //subtract w in ACCURATE mode
return RGBW32(r, g, b, w);
}
@@ -1231,7 +1240,7 @@ uint32_t BusManager::memUsage(BusConfig &bc) {
int BusManager::add(BusConfig &bc) {
if (getNumBusses() - getNumVirtualBusses() >= WLED_MAX_BUSSES) return -1;
// WLEDMM clear cached Bus info first
lastend = 0;
lastlen = 0;
laststart = 0;
lastBus = nullptr;
slowMode = false;
@@ -1272,7 +1281,7 @@ void BusManager::removeAll() {
// WLEDMM clear cached Bus info
lastBus = nullptr;
laststart = 0;
lastend = 0;
lastlen = 0;
slowMode = false;
}
@@ -1293,24 +1302,28 @@ void BusManager::setStatusPixel(uint32_t c) {
}
}
void IRAM_ATTR __attribute__((hot)) BusManager::setPixelColor(uint16_t pix, uint32_t c, int16_t cct) {
if (!slowMode && (pix >= laststart) && (pix < lastend ) && lastBus->isOk()) {
// WLEDMM same bus as last time - no need to search again
void IRAM_ATTR __attribute__((hot)) BusManager::setPixelColor(uint16_t pix, uint32_t c) {
// Fast path: check cached bus first (with proper nullptr check)
// optimization: below is True iff lastlen > 0 and pix >= laststart and pix < laststart + lastlen
if (!slowMode && lastBus && ((uint_fast16_t)(pix - laststart) < lastlen) && lastBus->isOk()) { // WLEDMM saves us a few cycles for each pixel
lastBus->setPixelColor(pix - laststart, c);
return;
}
for (uint_fast8_t i = 0; i < numBusses; i++) { // WLEDMM use fast native types
Bus* b = busses[i];
if (b->isOk() == false) continue; // WLEDMM ignore invalid (=not ready) busses
// Slow path: search through all buses
uint_fast8_t count = numBusses; // Cache to avoid repeated member access
for (uint_fast8_t i = 0; i < count; i++) {
Bus* const b = busses[i]; // Use const pointer for optimization hint
if ((!b) || (b->isOk() == false)) continue; // WLEDMM ignore invalid (=not ready) busses
uint_fast16_t bstart = b->getStart();
if (pix < bstart || pix >= bstart + b->getLength()) continue;
else {
uint_fast16_t blen = b->getLength();
if ((uint_fast16_t)(pix - bstart) < blen) { // Unsigned arithmetic trick for fast range check
if (!slowMode) {
// WLEDMM remember last Bus we took
// Cache bus info for next call
lastBus = b;
laststart = bstart;
lastend = bstart + b->getLength();
lastlen = blen;
}
b->setPixelColor(pix - bstart, c);
if (!slowMode) break; // WLEDMM found the right Bus -> so we can stop searching - unless we have busses that overlap
@@ -1334,46 +1347,52 @@ void __attribute__((cold)) BusManager::setSegmentCCT(int16_t cct, bool allowWBCo
}
uint32_t IRAM_ATTR __attribute__((hot)) BusManager::getPixelColor(uint_fast16_t pix) { // WLEDMM use fast native types, IRAM_ATTR
if ((pix >= laststart) && (pix < lastend ) && (lastBus != nullptr) && lastBus->isOk()) {
// Fast path: check cached bus first (with proper null check, and unsigned arithmetic trick for faster range check)
if (lastBus && ((uint_fast16_t)(pix - laststart) < lastlen) && lastBus->isOk()) {
// WLEDMM same bus as last time - no need to search again
return lastBus->getPixelColor(pix - laststart);
}
for (uint_fast8_t i = 0; i < numBusses; i++) {
Bus* b = busses[i];
if (b->isOk() == false) continue; // WLEDMM ignore invalid (=not ready) busses
uint_fast8_t count = numBusses; // Cache to avoid repeated member access
for (uint_fast8_t i = 0; i < count; i++) {
Bus* const b = busses[i]; // Use const pointer for optimization hint
if ((!b) || (b->isOk() == false)) continue; // WLEDMM ignore invalid (=not ready) busses
uint_fast16_t bstart = b->getStart();
if (pix < bstart || pix >= bstart + b->getLength()) continue;
else {
uint_fast16_t blen = b->getLength();
if ((uint_fast16_t)(pix - bstart) < blen) { // Unsigned arithmetic trick for fast range check
if (!slowMode) {
// WLEDMM remember last Bus we took
// Cache bus info for next call
lastBus = b;
laststart = bstart;
lastend = bstart + b->getLength();
lastlen = blen;
}
return b->getPixelColor(pix - bstart);
return b->getPixelColor(pix - bstart); // done - found one
}
}
return 0;
}
uint32_t IRAM_ATTR __attribute__((hot)) BusManager::getPixelColorRestored(uint_fast16_t pix) { // WLEDMM uses bus::getPixelColorRestored()
if ((pix >= laststart) && (pix < lastend ) && (lastBus != nullptr) && lastBus->isOk()) {
// Fast path: check cached bus first (with proper null check, and unsigned arithmetic trick for faster range check)
if (lastBus && ((uint_fast16_t)(pix - laststart) < lastlen) && lastBus->isOk()) {
// WLEDMM same bus as last time - no need to search again
return lastBus->getPixelColorRestored(pix - laststart);
}
for (uint_fast8_t i = 0; i < numBusses; i++) {
Bus* b = busses[i];
if (b->isOk() == false) continue; // WLEDMM ignore invalid (=not ready) busses
uint_fast8_t count = numBusses; // Cache to avoid repeated member access
for (uint_fast8_t i = 0; i < count; i++) {
Bus* const b = busses[i]; // Use const pointer for optimization hint
if ((!b) || (b->isOk() == false)) continue; // WLEDMM ignore invalid (=not ready) busses
uint_fast16_t bstart = b->getStart();
if (pix < bstart || pix >= bstart + b->getLength()) continue;
else {
uint_fast16_t blen = b->getLength();
if ((uint_fast16_t)(pix - bstart) < blen) { // Unsigned arithmetic trick for range check
if (!slowMode) {
// WLEDMM remember last Bus we took
// Cache bus info for next call
lastBus = b;
laststart = bstart;
lastend = bstart + b->getLength();
lastlen = blen;
}
return b->getPixelColorRestored(pix - bstart);
}