Skip to content

use fast path in blendSegment to bump up FPS #5464

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
DedeHai wants to merge 1 commit into
base: main
Choose a base branch
from
+61 −6
Open

use fast path in blendSegment to bump up FPS #5464

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 61 additions & 6 deletions wled00/FX_fcn.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1367,6 +1367,7 @@ static uint8_t _dummy (uint8_t a, uint8_t b) { return a; } // dummy (same as
void WS2812FX::blendSegment(const Segment &topSegment) const {
typedef uint8_t(*FuncType)(uint8_t, uint8_t);
// function pointer array: fill with _dummy if using special case: avoid OOB access and always provide a valid path
// note: making the function array static const uses more ram and comes at no significant speed gain
FuncType funcs[] = {
_dummy, _dummy, _dummy, _subtract,
_difference, _average, _dummy, _divide,
Expand Down Expand Up @@ -1398,14 +1399,69 @@ void WS2812FX::blendSegment(const Segment &topSegment) const {
const size_t matrixSize = Segment::maxWidth * Segment::maxHeight;
const size_t startIndx = XY(topSegment.start, topSegment.startY);
const size_t stopIndx = startIndx + length;
const unsigned progress = topSegment.progress();
const unsigned progInv = 0xFFFFU - progress;
uint8_t opacity = topSegment.currentBri(); // returns transitioned opacity for style FADE
uint8_t cct = topSegment.currentCCT();
if (gammaCorrectCol) opacity = gamma8inv(opacity); // use inverse gamma on brightness for correct color scaling after gamma correction (see #5343 for details)

Segment::setClippingRect(0, 0); // disable clipping by default
const Segment *segO = topSegment.getOldSegment();
const bool hasGrouping = topSegment.groupLength() != 1;

// fast path: handle the default case - no transitions, no grouping/spacing, no mirroring, no CCT
if (!segO && blendingStyle == TRANSITION_FADE && !hasGrouping && !topSegment.mirror && !topSegment.mirror_y) {
if (isMatrix && width * height <= (int)matrixSize && !_pixelCCT) {
#ifndef WLED_DISABLE_2D
// Calculate pointer steps to avoid 'if' and 'XY()' inside loops
int x_inc = 1;
int y_inc = Segment::maxWidth;
int start_offset = XY(topSegment.start, topSegment.startY);

// adjust starting position and steps based on Reverse/Transpose
// note: transpose is handled in seperate loop so it is still fast and no branching is needed in default path
if (!topSegment.transpose) {
if (topSegment.reverse) { start_offset += (width - 1); x_inc = -1; }
if (topSegment.reverse_y) { start_offset += (height - 1) * Segment::maxWidth; y_inc = -Segment::maxWidth; }

for (int y = 0; y < height; y++) {
uint32_t* pRow = &_pixels[start_offset + y * y_inc];
for (int x = 0; x < width; x++) {
uint32_t* p = pRow + x * x_inc;
uint32_t c_a = topSegment.getPixelColorRaw(x + y * width);
*p = color_blend(*p, segblend(c_a, *p), opacity);
}
}
} else { // transposed
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
int px = topSegment.reverse ? (width - x - 1) : x;
int py = topSegment.reverse_y ? (height - y - 1) : y;
size_t idx = XY(topSegment.start + py, topSegment.startY + px); // Swapped X/Y
_pixels[idx] = color_blend(_pixels[idx], segblend(topSegment.getPixelColorRaw(x + y * width), _pixels[idx]), opacity);
}
}
}
return;
#endif
} else if (!isMatrix) {
// 1D fast path, include CCT as it is more common on 1D setups
uint32_t* strip = _pixels;
int start = topSegment.start;
int off = topSegment.offset;
for (int i = 0; i < length; i++) {
uint32_t c_a = topSegment.getPixelColorRaw(i);
int p = topSegment.reverse ? (length - i - 1) : i;
int idx = start + p + off;
if (idx >= topSegment.stop) idx -= length;
strip[idx] = color_blend(strip[idx], segblend(c_a, strip[idx]), opacity);
if (_pixelCCT) _pixelCCT[idx] = cct;
}
return;
}
}

// slow path: handle transitions, grouping/spacing, segments with clipping and CCT pixels
Segment::setClippingRect(0, 0); // disable clipping by default
const unsigned progress = topSegment.progress();
const unsigned progInv = 0xFFFFU - progress;
const unsigned dw = (blendingStyle==TRANSITION_OUTSIDE_IN ? progInv : progress) * width / 0xFFFFU + 1;
const unsigned dh = (blendingStyle==TRANSITION_OUTSIDE_IN ? progInv : progress) * height / 0xFFFFU + 1;
const unsigned orgBS = blendingStyle;
Expand Down Expand Up @@ -1466,7 +1522,6 @@ void WS2812FX::blendSegment(const Segment &topSegment) const {
#ifndef WLED_DISABLE_2D
const int nCols = topSegment.virtualWidth();
const int nRows = topSegment.virtualHeight();
const Segment *segO = topSegment.getOldSegment();
const int oCols = segO ? segO->virtualWidth() : nCols;
const int oRows = segO ? segO->virtualHeight() : nRows;

Expand Down Expand Up @@ -1562,8 +1617,8 @@ void WS2812FX::blendSegment(const Segment &topSegment) const {
}
#endif
} else {
// 1D Slow Path
const int nLen = topSegment.virtualLength();
const Segment *segO = topSegment.getOldSegment();
const int oLen = segO ? segO->virtualLength() : nLen;

const auto setMirroredPixel = [&](int i, uint32_t c, uint8_t o) {
Expand Down Expand Up @@ -1643,7 +1698,7 @@ void WS2812FX::show() {

if (realtimeMode == REALTIME_MODE_INACTIVE || useMainSegmentOnly || realtimeOverride > REALTIME_OVERRIDE_NONE) {
// clear frame buffer
for (size_t i = 0; i < totalLen; i++) _pixels[i] = BLACK; // memset(_pixels, 0, sizeof(uint32_t) * getLengthTotal());
memset(_pixels, 0, sizeof(uint32_t) * totalLen);
// blend all segments into (cleared) buffer
for (Segment &seg : _segments) if (seg.isActive() && (seg.on || seg.isInTransition())) {
blendSegment(seg); // blend segment's buffer into frame buffer
Expand Down
Loading