LED-Mesh/libraries/FastLED/colorutils.cpp

1199 lines
33 KiB
C++
Raw Permalink Normal View History

2019-02-13 22:27:19 +01:00
#define FASTLED_INTERNAL
#define __PROG_TYPES_COMPAT__
#include <stdint.h>
#include <math.h>
#include "FastLED.h"
FASTLED_NAMESPACE_BEGIN
void fill_solid( struct CRGB * leds, int numToFill,
const struct CRGB& color)
{
for( int i = 0; i < numToFill; i++) {
leds[i] = color;
}
}
void fill_solid( struct CHSV * targetArray, int numToFill,
const struct CHSV& hsvColor)
{
for( int i = 0; i < numToFill; i++) {
targetArray[i] = hsvColor;
}
}
// void fill_solid( struct CRGB* targetArray, int numToFill,
// const struct CHSV& hsvColor)
// {
// fill_solid<CRGB>( targetArray, numToFill, (CRGB) hsvColor);
// }
void fill_rainbow( struct CRGB * pFirstLED, int numToFill,
uint8_t initialhue,
uint8_t deltahue )
{
CHSV hsv;
hsv.hue = initialhue;
hsv.val = 255;
hsv.sat = 240;
for( int i = 0; i < numToFill; i++) {
pFirstLED[i] = hsv;
hsv.hue += deltahue;
}
}
void fill_rainbow( struct CHSV * targetArray, int numToFill,
uint8_t initialhue,
uint8_t deltahue )
{
CHSV hsv;
hsv.hue = initialhue;
hsv.val = 255;
hsv.sat = 240;
for( int i = 0; i < numToFill; i++) {
targetArray[i] = hsv;
hsv.hue += deltahue;
}
}
void fill_gradient_RGB( CRGB* leds,
uint16_t startpos, CRGB startcolor,
uint16_t endpos, CRGB endcolor )
{
// if the points are in the wrong order, straighten them
if( endpos < startpos ) {
uint16_t t = endpos;
CRGB tc = endcolor;
endcolor = startcolor;
endpos = startpos;
startpos = t;
startcolor = tc;
}
saccum87 rdistance87;
saccum87 gdistance87;
saccum87 bdistance87;
rdistance87 = (endcolor.r - startcolor.r) << 7;
gdistance87 = (endcolor.g - startcolor.g) << 7;
bdistance87 = (endcolor.b - startcolor.b) << 7;
uint16_t pixeldistance = endpos - startpos;
int16_t divisor = pixeldistance ? pixeldistance : 1;
saccum87 rdelta87 = rdistance87 / divisor;
saccum87 gdelta87 = gdistance87 / divisor;
saccum87 bdelta87 = bdistance87 / divisor;
rdelta87 *= 2;
gdelta87 *= 2;
bdelta87 *= 2;
accum88 r88 = startcolor.r << 8;
accum88 g88 = startcolor.g << 8;
accum88 b88 = startcolor.b << 8;
for( uint16_t i = startpos; i <= endpos; i++) {
leds[i] = CRGB( r88 >> 8, g88 >> 8, b88 >> 8);
r88 += rdelta87;
g88 += gdelta87;
b88 += bdelta87;
}
}
#if 0
void fill_gradient( const CHSV& c1, const CHSV& c2)
{
fill_gradient( FastLED[0].leds(), FastLED[0].size(), c1, c2);
}
void fill_gradient( const CHSV& c1, const CHSV& c2, const CHSV& c3)
{
fill_gradient( FastLED[0].leds(), FastLED[0].size(), c1, c2, c3);
}
void fill_gradient( const CHSV& c1, const CHSV& c2, const CHSV& c3, const CHSV& c4)
{
fill_gradient( FastLED[0].leds(), FastLED[0].size(), c1, c2, c3, c4);
}
void fill_gradient_RGB( const CRGB& c1, const CRGB& c2)
{
fill_gradient_RGB( FastLED[0].leds(), FastLED[0].size(), c1, c2);
}
void fill_gradient_RGB( const CRGB& c1, const CRGB& c2, const CRGB& c3)
{
fill_gradient_RGB( FastLED[0].leds(), FastLED[0].size(), c1, c2, c3);
}
void fill_gradient_RGB( const CRGB& c1, const CRGB& c2, const CRGB& c3, const CRGB& c4)
{
fill_gradient_RGB( FastLED[0].leds(), FastLED[0].size(), c1, c2, c3, c4);
}
#endif
void fill_gradient_RGB( CRGB* leds, uint16_t numLeds, const CRGB& c1, const CRGB& c2)
{
uint16_t last = numLeds - 1;
fill_gradient_RGB( leds, 0, c1, last, c2);
}
void fill_gradient_RGB( CRGB* leds, uint16_t numLeds, const CRGB& c1, const CRGB& c2, const CRGB& c3)
{
uint16_t half = (numLeds / 2);
uint16_t last = numLeds - 1;
fill_gradient_RGB( leds, 0, c1, half, c2);
fill_gradient_RGB( leds, half, c2, last, c3);
}
void fill_gradient_RGB( CRGB* leds, uint16_t numLeds, const CRGB& c1, const CRGB& c2, const CRGB& c3, const CRGB& c4)
{
uint16_t onethird = (numLeds / 3);
uint16_t twothirds = ((numLeds * 2) / 3);
uint16_t last = numLeds - 1;
fill_gradient_RGB( leds, 0, c1, onethird, c2);
fill_gradient_RGB( leds, onethird, c2, twothirds, c3);
fill_gradient_RGB( leds, twothirds, c3, last, c4);
}
void nscale8_video( CRGB* leds, uint16_t num_leds, uint8_t scale)
{
for( uint16_t i = 0; i < num_leds; i++) {
leds[i].nscale8_video( scale);
}
}
void fade_video(CRGB* leds, uint16_t num_leds, uint8_t fadeBy)
{
nscale8_video( leds, num_leds, 255 - fadeBy);
}
void fadeLightBy(CRGB* leds, uint16_t num_leds, uint8_t fadeBy)
{
nscale8_video( leds, num_leds, 255 - fadeBy);
}
void fadeToBlackBy( CRGB* leds, uint16_t num_leds, uint8_t fadeBy)
{
nscale8( leds, num_leds, 255 - fadeBy);
}
void fade_raw( CRGB* leds, uint16_t num_leds, uint8_t fadeBy)
{
nscale8( leds, num_leds, 255 - fadeBy);
}
void nscale8_raw( CRGB* leds, uint16_t num_leds, uint8_t scale)
{
nscale8( leds, num_leds, scale);
}
void nscale8( CRGB* leds, uint16_t num_leds, uint8_t scale)
{
for( uint16_t i = 0; i < num_leds; i++) {
leds[i].nscale8( scale);
}
}
void fadeUsingColor( CRGB* leds, uint16_t numLeds, const CRGB& colormask)
{
uint8_t fr, fg, fb;
fr = colormask.r;
fg = colormask.g;
fb = colormask.b;
for( uint16_t i = 0; i < numLeds; i++) {
leds[i].r = scale8_LEAVING_R1_DIRTY( leds[i].r, fr);
leds[i].g = scale8_LEAVING_R1_DIRTY( leds[i].g, fg);
leds[i].b = scale8 ( leds[i].b, fb);
}
}
CRGB& nblend( CRGB& existing, const CRGB& overlay, fract8 amountOfOverlay )
{
if( amountOfOverlay == 0) {
return existing;
}
if( amountOfOverlay == 255) {
existing = overlay;
return existing;
}
#if 0
// Old blend method which unfortunately had some rounding errors
fract8 amountOfKeep = 255 - amountOfOverlay;
existing.red = scale8_LEAVING_R1_DIRTY( existing.red, amountOfKeep)
+ scale8_LEAVING_R1_DIRTY( overlay.red, amountOfOverlay);
existing.green = scale8_LEAVING_R1_DIRTY( existing.green, amountOfKeep)
+ scale8_LEAVING_R1_DIRTY( overlay.green, amountOfOverlay);
existing.blue = scale8_LEAVING_R1_DIRTY( existing.blue, amountOfKeep)
+ scale8_LEAVING_R1_DIRTY( overlay.blue, amountOfOverlay);
cleanup_R1();
#else
// Corrected blend method, with no loss-of-precision rounding errors
existing.red = blend8( existing.red, overlay.red, amountOfOverlay);
existing.green = blend8( existing.green, overlay.green, amountOfOverlay);
existing.blue = blend8( existing.blue, overlay.blue, amountOfOverlay);
#endif
return existing;
}
void nblend( CRGB* existing, CRGB* overlay, uint16_t count, fract8 amountOfOverlay)
{
for( uint16_t i = count; i; i--) {
nblend( *existing, *overlay, amountOfOverlay);
existing++;
overlay++;
}
}
CRGB blend( const CRGB& p1, const CRGB& p2, fract8 amountOfP2 )
{
CRGB nu(p1);
nblend( nu, p2, amountOfP2);
return nu;
}
CRGB* blend( const CRGB* src1, const CRGB* src2, CRGB* dest, uint16_t count, fract8 amountOfsrc2 )
{
for( uint16_t i = 0; i < count; i++) {
dest[i] = blend(src1[i], src2[i], amountOfsrc2);
}
return dest;
}
CHSV& nblend( CHSV& existing, const CHSV& overlay, fract8 amountOfOverlay, TGradientDirectionCode directionCode)
{
if( amountOfOverlay == 0) {
return existing;
}
if( amountOfOverlay == 255) {
existing = overlay;
return existing;
}
fract8 amountOfKeep = 255 - amountOfOverlay;
uint8_t huedelta8 = overlay.hue - existing.hue;
if( directionCode == SHORTEST_HUES ) {
directionCode = FORWARD_HUES;
if( huedelta8 > 127) {
directionCode = BACKWARD_HUES;
}
}
if( directionCode == LONGEST_HUES ) {
directionCode = FORWARD_HUES;
if( huedelta8 < 128) {
directionCode = BACKWARD_HUES;
}
}
if( directionCode == FORWARD_HUES) {
existing.hue = existing.hue + scale8( huedelta8, amountOfOverlay);
}
else /* directionCode == BACKWARD_HUES */
{
huedelta8 = -huedelta8;
existing.hue = existing.hue - scale8( huedelta8, amountOfOverlay);
}
existing.sat = scale8_LEAVING_R1_DIRTY( existing.sat, amountOfKeep)
+ scale8_LEAVING_R1_DIRTY( overlay.sat, amountOfOverlay);
existing.val = scale8_LEAVING_R1_DIRTY( existing.val, amountOfKeep)
+ scale8_LEAVING_R1_DIRTY( overlay.val, amountOfOverlay);
cleanup_R1();
return existing;
}
void nblend( CHSV* existing, CHSV* overlay, uint16_t count, fract8 amountOfOverlay, TGradientDirectionCode directionCode )
{
if(existing == overlay) return;
for( uint16_t i = count; i; i--) {
nblend( *existing, *overlay, amountOfOverlay, directionCode);
existing++;
overlay++;
}
}
CHSV blend( const CHSV& p1, const CHSV& p2, fract8 amountOfP2, TGradientDirectionCode directionCode )
{
CHSV nu(p1);
nblend( nu, p2, amountOfP2, directionCode);
return nu;
}
CHSV* blend( const CHSV* src1, const CHSV* src2, CHSV* dest, uint16_t count, fract8 amountOfsrc2, TGradientDirectionCode directionCode )
{
for( uint16_t i = 0; i < count; i++) {
dest[i] = blend(src1[i], src2[i], amountOfsrc2, directionCode);
}
return dest;
}
// Forward declaration of the function "XY" which must be provided by
// the application for use in two-dimensional filter functions.
uint16_t XY( uint8_t, uint8_t);// __attribute__ ((weak));
// blur1d: one-dimensional blur filter. Spreads light to 2 line neighbors.
// blur2d: two-dimensional blur filter. Spreads light to 8 XY neighbors.
//
// 0 = no spread at all
// 64 = moderate spreading
// 172 = maximum smooth, even spreading
//
// 173..255 = wider spreading, but increasing flicker
//
// Total light is NOT entirely conserved, so many repeated
// calls to 'blur' will also result in the light fading,
// eventually all the way to black; this is by design so that
// it can be used to (slowly) clear the LEDs to black.
void blur1d( CRGB* leds, uint16_t numLeds, fract8 blur_amount)
{
uint8_t keep = 255 - blur_amount;
uint8_t seep = blur_amount >> 1;
CRGB carryover = CRGB::Black;
for( uint16_t i = 0; i < numLeds; i++) {
CRGB cur = leds[i];
CRGB part = cur;
part.nscale8( seep);
cur.nscale8( keep);
cur += carryover;
if( i) leds[i-1] += part;
leds[i] = cur;
carryover = part;
}
}
void blur2d( CRGB* leds, uint8_t width, uint8_t height, fract8 blur_amount)
{
blurRows(leds, width, height, blur_amount);
blurColumns(leds, width, height, blur_amount);
}
// blurRows: perform a blur1d on every row of a rectangular matrix
void blurRows( CRGB* leds, uint8_t width, uint8_t height, fract8 blur_amount)
{
for( uint8_t row = 0; row < height; row++) {
CRGB* rowbase = leds + (row * width);
blur1d( rowbase, width, blur_amount);
}
}
// blurColumns: perform a blur1d on each column of a rectangular matrix
void blurColumns(CRGB* leds, uint8_t width, uint8_t height, fract8 blur_amount)
{
// blur columns
uint8_t keep = 255 - blur_amount;
uint8_t seep = blur_amount >> 1;
for( uint8_t col = 0; col < width; col++) {
CRGB carryover = CRGB::Black;
for( uint8_t i = 0; i < height; i++) {
CRGB cur = leds[XY(col,i)];
CRGB part = cur;
part.nscale8( seep);
cur.nscale8( keep);
cur += carryover;
if( i) leds[XY(col,i-1)] += part;
leds[XY(col,i)] = cur;
carryover = part;
}
}
}
// CRGB HeatColor( uint8_t temperature)
//
// Approximates a 'black body radiation' spectrum for
// a given 'heat' level. This is useful for animations of 'fire'.
// Heat is specified as an arbitrary scale from 0 (cool) to 255 (hot).
// This is NOT a chromatically correct 'black body radiation'
// spectrum, but it's surprisingly close, and it's fast and small.
//
// On AVR/Arduino, this typically takes around 70 bytes of program memory,
// versus 768 bytes for a full 256-entry RGB lookup table.
CRGB HeatColor( uint8_t temperature)
{
CRGB heatcolor;
// Scale 'heat' down from 0-255 to 0-191,
// which can then be easily divided into three
// equal 'thirds' of 64 units each.
uint8_t t192 = scale8_video( temperature, 191);
// calculate a value that ramps up from
// zero to 255 in each 'third' of the scale.
uint8_t heatramp = t192 & 0x3F; // 0..63
heatramp <<= 2; // scale up to 0..252
// now figure out which third of the spectrum we're in:
if( t192 & 0x80) {
// we're in the hottest third
heatcolor.r = 255; // full red
heatcolor.g = 255; // full green
heatcolor.b = heatramp; // ramp up blue
} else if( t192 & 0x40 ) {
// we're in the middle third
heatcolor.r = 255; // full red
heatcolor.g = heatramp; // ramp up green
heatcolor.b = 0; // no blue
} else {
// we're in the coolest third
heatcolor.r = heatramp; // ramp up red
heatcolor.g = 0; // no green
heatcolor.b = 0; // no blue
}
return heatcolor;
}
// lsrX4: helper function to divide a number by 16, aka four LSR's.
// On avr-gcc, "u8 >> 4" generates a loop, which is big, and slow.
// merely forcing it to be four /=2's causes avr-gcc to emit
// a SWAP instruction followed by an AND 0x0F, which is faster, and smaller.
inline uint8_t lsrX4( uint8_t dividend) __attribute__((always_inline));
inline uint8_t lsrX4( uint8_t dividend)
{
#if defined(__AVR__)
dividend /= 2;
dividend /= 2;
dividend /= 2;
dividend /= 2;
#else
dividend >>= 4;
#endif
return dividend;
}
CRGB ColorFromPalette( const CRGBPalette16& pal, uint8_t index, uint8_t brightness, TBlendType blendType)
{
// hi4 = index >> 4;
uint8_t hi4 = lsrX4(index);
uint8_t lo4 = index & 0x0F;
// const CRGB* entry = &(pal[0]) + hi4;
// since hi4 is always 0..15, hi4 * sizeof(CRGB) can be a single-byte value,
// instead of the two byte 'int' that avr-gcc defaults to.
// So, we multiply hi4 X sizeof(CRGB), giving hi4XsizeofCRGB;
uint8_t hi4XsizeofCRGB = hi4 * sizeof(CRGB);
// We then add that to a base array pointer.
const CRGB* entry = (CRGB*)( (uint8_t*)(&(pal[0])) + hi4XsizeofCRGB);
uint8_t blend = lo4 && (blendType != NOBLEND);
uint8_t red1 = entry->red;
uint8_t green1 = entry->green;
uint8_t blue1 = entry->blue;
if( blend ) {
if( hi4 == 15 ) {
entry = &(pal[0]);
} else {
entry++;
}
uint8_t f2 = lo4 << 4;
uint8_t f1 = 255 - f2;
// rgb1.nscale8(f1);
uint8_t red2 = entry->red;
red1 = scale8_LEAVING_R1_DIRTY( red1, f1);
red2 = scale8_LEAVING_R1_DIRTY( red2, f2);
red1 += red2;
uint8_t green2 = entry->green;
green1 = scale8_LEAVING_R1_DIRTY( green1, f1);
green2 = scale8_LEAVING_R1_DIRTY( green2, f2);
green1 += green2;
uint8_t blue2 = entry->blue;
blue1 = scale8_LEAVING_R1_DIRTY( blue1, f1);
blue2 = scale8_LEAVING_R1_DIRTY( blue2, f2);
blue1 += blue2;
cleanup_R1();
}
if( brightness != 255) {
if( brightness ) {
brightness++; // adjust for rounding
// Now, since brightness is nonzero, we don't need the full scale8_video logic;
// we can just to scale8 and then add one (unless scale8 fixed) to all nonzero inputs.
if( red1 ) {
red1 = scale8_LEAVING_R1_DIRTY( red1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
red1++;
#endif
}
if( green1 ) {
green1 = scale8_LEAVING_R1_DIRTY( green1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
green1++;
#endif
}
if( blue1 ) {
blue1 = scale8_LEAVING_R1_DIRTY( blue1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
blue1++;
#endif
}
cleanup_R1();
} else {
red1 = 0;
green1 = 0;
blue1 = 0;
}
}
return CRGB( red1, green1, blue1);
}
CRGB ColorFromPalette( const TProgmemRGBPalette16& pal, uint8_t index, uint8_t brightness, TBlendType blendType)
{
// hi4 = index >> 4;
uint8_t hi4 = lsrX4(index);
uint8_t lo4 = index & 0x0F;
CRGB entry = FL_PGM_READ_DWORD_NEAR( &(pal[0]) + hi4 );
uint8_t red1 = entry.red;
uint8_t green1 = entry.green;
uint8_t blue1 = entry.blue;
uint8_t blend = lo4 && (blendType != NOBLEND);
if( blend ) {
if( hi4 == 15 ) {
entry = FL_PGM_READ_DWORD_NEAR( &(pal[0]) );
} else {
entry = FL_PGM_READ_DWORD_NEAR( &(pal[1]) + hi4 );
}
uint8_t f2 = lo4 << 4;
uint8_t f1 = 255 - f2;
uint8_t red2 = entry.red;
red1 = scale8_LEAVING_R1_DIRTY( red1, f1);
red2 = scale8_LEAVING_R1_DIRTY( red2, f2);
red1 += red2;
uint8_t green2 = entry.green;
green1 = scale8_LEAVING_R1_DIRTY( green1, f1);
green2 = scale8_LEAVING_R1_DIRTY( green2, f2);
green1 += green2;
uint8_t blue2 = entry.blue;
blue1 = scale8_LEAVING_R1_DIRTY( blue1, f1);
blue2 = scale8_LEAVING_R1_DIRTY( blue2, f2);
blue1 += blue2;
cleanup_R1();
}
if( brightness != 255) {
if( brightness ) {
brightness++; // adjust for rounding
// Now, since brightness is nonzero, we don't need the full scale8_video logic;
// we can just to scale8 and then add one (unless scale8 fixed) to all nonzero inputs.
if( red1 ) {
red1 = scale8_LEAVING_R1_DIRTY( red1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
red1++;
#endif
}
if( green1 ) {
green1 = scale8_LEAVING_R1_DIRTY( green1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
green1++;
#endif
}
if( blue1 ) {
blue1 = scale8_LEAVING_R1_DIRTY( blue1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
blue1++;
#endif
}
cleanup_R1();
} else {
red1 = 0;
green1 = 0;
blue1 = 0;
}
}
return CRGB( red1, green1, blue1);
}
CRGB ColorFromPalette( const CRGBPalette32& pal, uint8_t index, uint8_t brightness, TBlendType blendType)
{
uint8_t hi5 = index;
#if defined(__AVR__)
hi5 /= 2;
hi5 /= 2;
hi5 /= 2;
#else
hi5 >>= 3;
#endif
uint8_t lo3 = index & 0x07;
// const CRGB* entry = &(pal[0]) + hi5;
// since hi5 is always 0..31, hi4 * sizeof(CRGB) can be a single-byte value,
// instead of the two byte 'int' that avr-gcc defaults to.
// So, we multiply hi5 X sizeof(CRGB), giving hi5XsizeofCRGB;
uint8_t hi5XsizeofCRGB = hi5 * sizeof(CRGB);
// We then add that to a base array pointer.
const CRGB* entry = (CRGB*)( (uint8_t*)(&(pal[0])) + hi5XsizeofCRGB);
uint8_t red1 = entry->red;
uint8_t green1 = entry->green;
uint8_t blue1 = entry->blue;
uint8_t blend = lo3 && (blendType != NOBLEND);
if( blend ) {
if( hi5 == 31 ) {
entry = &(pal[0]);
} else {
entry++;
}
uint8_t f2 = lo3 << 5;
uint8_t f1 = 255 - f2;
uint8_t red2 = entry->red;
red1 = scale8_LEAVING_R1_DIRTY( red1, f1);
red2 = scale8_LEAVING_R1_DIRTY( red2, f2);
red1 += red2;
uint8_t green2 = entry->green;
green1 = scale8_LEAVING_R1_DIRTY( green1, f1);
green2 = scale8_LEAVING_R1_DIRTY( green2, f2);
green1 += green2;
uint8_t blue2 = entry->blue;
blue1 = scale8_LEAVING_R1_DIRTY( blue1, f1);
blue2 = scale8_LEAVING_R1_DIRTY( blue2, f2);
blue1 += blue2;
cleanup_R1();
}
if( brightness != 255) {
if( brightness ) {
brightness++; // adjust for rounding
// Now, since brightness is nonzero, we don't need the full scale8_video logic;
// we can just to scale8 and then add one (unless scale8 fixed) to all nonzero inputs.
if( red1 ) {
red1 = scale8_LEAVING_R1_DIRTY( red1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
red1++;
#endif
}
if( green1 ) {
green1 = scale8_LEAVING_R1_DIRTY( green1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
green1++;
#endif
}
if( blue1 ) {
blue1 = scale8_LEAVING_R1_DIRTY( blue1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
blue1++;
#endif
}
cleanup_R1();
} else {
red1 = 0;
green1 = 0;
blue1 = 0;
}
}
return CRGB( red1, green1, blue1);
}
CRGB ColorFromPalette( const TProgmemRGBPalette32& pal, uint8_t index, uint8_t brightness, TBlendType blendType)
{
uint8_t hi5 = index;
#if defined(__AVR__)
hi5 /= 2;
hi5 /= 2;
hi5 /= 2;
#else
hi5 >>= 3;
#endif
uint8_t lo3 = index & 0x07;
CRGB entry = FL_PGM_READ_DWORD_NEAR( &(pal[0]) + hi5);
uint8_t red1 = entry.red;
uint8_t green1 = entry.green;
uint8_t blue1 = entry.blue;
uint8_t blend = lo3 && (blendType != NOBLEND);
if( blend ) {
if( hi5 == 31 ) {
entry = FL_PGM_READ_DWORD_NEAR( &(pal[0]) );
} else {
entry = FL_PGM_READ_DWORD_NEAR( &(pal[1]) + hi5 );
}
uint8_t f2 = lo3 << 5;
uint8_t f1 = 255 - f2;
uint8_t red2 = entry.red;
red1 = scale8_LEAVING_R1_DIRTY( red1, f1);
red2 = scale8_LEAVING_R1_DIRTY( red2, f2);
red1 += red2;
uint8_t green2 = entry.green;
green1 = scale8_LEAVING_R1_DIRTY( green1, f1);
green2 = scale8_LEAVING_R1_DIRTY( green2, f2);
green1 += green2;
uint8_t blue2 = entry.blue;
blue1 = scale8_LEAVING_R1_DIRTY( blue1, f1);
blue2 = scale8_LEAVING_R1_DIRTY( blue2, f2);
blue1 += blue2;
cleanup_R1();
}
if( brightness != 255) {
if( brightness ) {
brightness++; // adjust for rounding
// Now, since brightness is nonzero, we don't need the full scale8_video logic;
// we can just to scale8 and then add one (unless scale8 fixed) to all nonzero inputs.
if( red1 ) {
red1 = scale8_LEAVING_R1_DIRTY( red1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
red1++;
#endif
}
if( green1 ) {
green1 = scale8_LEAVING_R1_DIRTY( green1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
green1++;
#endif
}
if( blue1 ) {
blue1 = scale8_LEAVING_R1_DIRTY( blue1, brightness);
#if !(FASTLED_SCALE8_FIXED==1)
blue1++;
#endif
}
cleanup_R1();
} else {
red1 = 0;
green1 = 0;
blue1 = 0;
}
}
return CRGB( red1, green1, blue1);
}
CRGB ColorFromPalette( const CRGBPalette256& pal, uint8_t index, uint8_t brightness, TBlendType)
{
const CRGB* entry = &(pal[0]) + index;
uint8_t red = entry->red;
uint8_t green = entry->green;
uint8_t blue = entry->blue;
if( brightness != 255) {
brightness++; // adjust for rounding
red = scale8_video_LEAVING_R1_DIRTY( red, brightness);
green = scale8_video_LEAVING_R1_DIRTY( green, brightness);
blue = scale8_video_LEAVING_R1_DIRTY( blue, brightness);
cleanup_R1();
}
return CRGB( red, green, blue);
}
CHSV ColorFromPalette( const struct CHSVPalette16& pal, uint8_t index, uint8_t brightness, TBlendType blendType)
{
// hi4 = index >> 4;
uint8_t hi4 = lsrX4(index);
uint8_t lo4 = index & 0x0F;
// CRGB rgb1 = pal[ hi4];
const CHSV* entry = &(pal[0]) + hi4;
uint8_t hue1 = entry->hue;
uint8_t sat1 = entry->sat;
uint8_t val1 = entry->val;
uint8_t blend = lo4 && (blendType != NOBLEND);
if( blend ) {
if( hi4 == 15 ) {
entry = &(pal[0]);
} else {
entry++;
}
uint8_t f2 = lo4 << 4;
uint8_t f1 = 255 - f2;
uint8_t hue2 = entry->hue;
uint8_t sat2 = entry->sat;
uint8_t val2 = entry->val;
// Now some special casing for blending to or from
// either black or white. Black and white don't have
// proper 'hue' of their own, so when ramping from
// something else to/from black/white, we set the 'hue'
// of the black/white color to be the same as the hue
// of the other color, so that you get the expected
// brightness or saturation ramp, with hue staying
// constant:
// If we are starting from white (sat=0)
// or black (val=0), adopt the target hue.
if( sat1 == 0 || val1 == 0) {
hue1 = hue2;
}
// If we are ending at white (sat=0)
// or black (val=0), adopt the starting hue.
if( sat2 == 0 || val2 == 0) {
hue2 = hue1;
}
sat1 = scale8_LEAVING_R1_DIRTY( sat1, f1);
val1 = scale8_LEAVING_R1_DIRTY( val1, f1);
sat2 = scale8_LEAVING_R1_DIRTY( sat2, f2);
val2 = scale8_LEAVING_R1_DIRTY( val2, f2);
// cleanup_R1();
// These sums can't overflow, so no qadd8 needed.
sat1 += sat2;
val1 += val2;
uint8_t deltaHue = (uint8_t)(hue2 - hue1);
if( deltaHue & 0x80 ) {
// go backwards
hue1 -= scale8( 256 - deltaHue, f2);
} else {
// go forwards
hue1 += scale8( deltaHue, f2);
}
cleanup_R1();
}
if( brightness != 255) {
val1 = scale8_video( val1, brightness);
}
return CHSV( hue1, sat1, val1);
}
CHSV ColorFromPalette( const struct CHSVPalette32& pal, uint8_t index, uint8_t brightness, TBlendType blendType)
{
uint8_t hi5 = index;
#if defined(__AVR__)
hi5 /= 2;
hi5 /= 2;
hi5 /= 2;
#else
hi5 >>= 3;
#endif
uint8_t lo3 = index & 0x07;
uint8_t hi5XsizeofCHSV = hi5 * sizeof(CHSV);
const CHSV* entry = (CHSV*)( (uint8_t*)(&(pal[0])) + hi5XsizeofCHSV);
uint8_t hue1 = entry->hue;
uint8_t sat1 = entry->sat;
uint8_t val1 = entry->val;
uint8_t blend = lo3 && (blendType != NOBLEND);
if( blend ) {
if( hi5 == 31 ) {
entry = &(pal[0]);
} else {
entry++;
}
uint8_t f2 = lo3 << 5;
uint8_t f1 = 255 - f2;
uint8_t hue2 = entry->hue;
uint8_t sat2 = entry->sat;
uint8_t val2 = entry->val;
// Now some special casing for blending to or from
// either black or white. Black and white don't have
// proper 'hue' of their own, so when ramping from
// something else to/from black/white, we set the 'hue'
// of the black/white color to be the same as the hue
// of the other color, so that you get the expected
// brightness or saturation ramp, with hue staying
// constant:
// If we are starting from white (sat=0)
// or black (val=0), adopt the target hue.
if( sat1 == 0 || val1 == 0) {
hue1 = hue2;
}
// If we are ending at white (sat=0)
// or black (val=0), adopt the starting hue.
if( sat2 == 0 || val2 == 0) {
hue2 = hue1;
}
sat1 = scale8_LEAVING_R1_DIRTY( sat1, f1);
val1 = scale8_LEAVING_R1_DIRTY( val1, f1);
sat2 = scale8_LEAVING_R1_DIRTY( sat2, f2);
val2 = scale8_LEAVING_R1_DIRTY( val2, f2);
// cleanup_R1();
// These sums can't overflow, so no qadd8 needed.
sat1 += sat2;
val1 += val2;
uint8_t deltaHue = (uint8_t)(hue2 - hue1);
if( deltaHue & 0x80 ) {
// go backwards
hue1 -= scale8( 256 - deltaHue, f2);
} else {
// go forwards
hue1 += scale8( deltaHue, f2);
}
cleanup_R1();
}
if( brightness != 255) {
val1 = scale8_video( val1, brightness);
}
return CHSV( hue1, sat1, val1);
}
CHSV ColorFromPalette( const struct CHSVPalette256& pal, uint8_t index, uint8_t brightness, TBlendType)
{
CHSV hsv = *( &(pal[0]) + index );
if( brightness != 255) {
hsv.value = scale8_video( hsv.value, brightness);
}
return hsv;
}
void UpscalePalette(const struct CRGBPalette16& srcpal16, struct CRGBPalette256& destpal256)
{
for( int i = 0; i < 256; i++) {
destpal256[(uint8_t)(i)] = ColorFromPalette( srcpal16, i);
}
}
void UpscalePalette(const struct CHSVPalette16& srcpal16, struct CHSVPalette256& destpal256)
{
for( int i = 0; i < 256; i++) {
destpal256[(uint8_t)(i)] = ColorFromPalette( srcpal16, i);
}
}
void UpscalePalette(const struct CRGBPalette16& srcpal16, struct CRGBPalette32& destpal32)
{
for( uint8_t i = 0; i < 16; i++) {
uint8_t j = i * 2;
destpal32[j+0] = srcpal16[i];
destpal32[j+1] = srcpal16[i];
}
}
void UpscalePalette(const struct CHSVPalette16& srcpal16, struct CHSVPalette32& destpal32)
{
for( uint8_t i = 0; i < 16; i++) {
uint8_t j = i * 2;
destpal32[j+0] = srcpal16[i];
destpal32[j+1] = srcpal16[i];
}
}
void UpscalePalette(const struct CRGBPalette32& srcpal32, struct CRGBPalette256& destpal256)
{
for( int i = 0; i < 256; i++) {
destpal256[(uint8_t)(i)] = ColorFromPalette( srcpal32, i);
}
}
void UpscalePalette(const struct CHSVPalette32& srcpal32, struct CHSVPalette256& destpal256)
{
for( int i = 0; i < 256; i++) {
destpal256[(uint8_t)(i)] = ColorFromPalette( srcpal32, i);
}
}
#if 0
// replaced by PartyColors_p
void SetupPartyColors(CRGBPalette16& pal)
{
fill_gradient( pal, 0, CHSV( HUE_PURPLE,255,255), 7, CHSV(HUE_YELLOW - 18,255,255), FORWARD_HUES);
fill_gradient( pal, 8, CHSV( HUE_ORANGE,255,255), 15, CHSV(HUE_BLUE + 18,255,255), BACKWARD_HUES);
}
#endif
void nblendPaletteTowardPalette( CRGBPalette16& current, CRGBPalette16& target, uint8_t maxChanges)
{
uint8_t* p1;
uint8_t* p2;
uint8_t changes = 0;
p1 = (uint8_t*)current.entries;
p2 = (uint8_t*)target.entries;
const uint8_t totalChannels = sizeof(CRGBPalette16);
for( uint8_t i = 0; i < totalChannels; i++) {
// if the values are equal, no changes are needed
if( p1[i] == p2[i] ) { continue; }
// if the current value is less than the target, increase it by one
if( p1[i] < p2[i] ) { p1[i]++; changes++; }
// if the current value is greater than the target,
// increase it by one (or two if it's still greater).
if( p1[i] > p2[i] ) {
p1[i]--; changes++;
if( p1[i] > p2[i] ) { p1[i]--; }
}
// if we've hit the maximum number of changes, exit
if( changes >= maxChanges) { break; }
}
}
uint8_t applyGamma_video( uint8_t brightness, float gamma)
{
float orig;
float adj;
orig = (float)(brightness) / (255.0);
adj = pow( orig, gamma) * (255.0);
uint8_t result = (uint8_t)(adj);
if( (brightness > 0) && (result == 0)) {
result = 1; // never gamma-adjust a positive number down to zero
}
return result;
}
CRGB applyGamma_video( const CRGB& orig, float gamma)
{
CRGB adj;
adj.r = applyGamma_video( orig.r, gamma);
adj.g = applyGamma_video( orig.g, gamma);
adj.b = applyGamma_video( orig.b, gamma);
return adj;
}
CRGB applyGamma_video( const CRGB& orig, float gammaR, float gammaG, float gammaB)
{
CRGB adj;
adj.r = applyGamma_video( orig.r, gammaR);
adj.g = applyGamma_video( orig.g, gammaG);
adj.b = applyGamma_video( orig.b, gammaB);
return adj;
}
CRGB& napplyGamma_video( CRGB& rgb, float gamma)
{
rgb = applyGamma_video( rgb, gamma);
return rgb;
}
CRGB& napplyGamma_video( CRGB& rgb, float gammaR, float gammaG, float gammaB)
{
rgb = applyGamma_video( rgb, gammaR, gammaG, gammaB);
return rgb;
}
void napplyGamma_video( CRGB* rgbarray, uint16_t count, float gamma)
{
for( uint16_t i = 0; i < count; i++) {
rgbarray[i] = applyGamma_video( rgbarray[i], gamma);
}
}
void napplyGamma_video( CRGB* rgbarray, uint16_t count, float gammaR, float gammaG, float gammaB)
{
for( uint16_t i = 0; i < count; i++) {
rgbarray[i] = applyGamma_video( rgbarray[i], gammaR, gammaG, gammaB);
}
}
FASTLED_NAMESPACE_END