switched to fixed point arithmetic; saving 1046 byte and noticeable speedup

This commit is contained in:
Christian Kroll 2012-05-04 21:53:28 +00:00
parent 028dea2b87
commit af8a928123
1 changed files with 394 additions and 57 deletions

View File

@ -1,30 +1,277 @@
#include <math.h> // Floating point math is dog slow on AVR, but I don't care.
#include <string.h>
/**
* Routines for drawing patterns generated by fixed point math functions.
*/
#include <assert.h>
#include <stdint.h>
#include <string.h>
#include "../config.h"
#include "../pixel.h"
#include "../util.h"
#include "fpmath_patterns.h"
/**
* \defgroup fixedpoint Fixed-point based animated plasma patterns.
*/
/*@{*/
/**
* Double buffering helps in reducing the effect of visibly redrawing every
* frame. With this option turned on, a frame is rendered into an off-screen
* buffer first and then copied to the actual frame buffer in one piece.
* However, given the borg's graphics architecture, half painted frames may
* still occur, but they are barely noticeable with this option enabled.
*
* Turn this off (#undef DOUBLE_BUFFERING) if you prefer speed over beauty.
*/
#define DOUBLE_BUFFERING
#ifdef DOUBLE_BUFFERING
#define BUFFER pixmap_buffer
#ifdef LOW_PRECISION
#undef LOW_PRECISION
#endif
#if NUM_COLS <= 16 && NUM_ROWS <= 16
/**
* Low precision means that we use Q10.5 values and 16 bit types for almost
* every calculation (with multiplication and division as notable exceptions
* as they and their interim results utilize 32 bit).
*
* Use this precision mode with care as image quality will suffer
* noticeably. It produces leaner and faster code, though. This mode should
* not be used with resolutions higher than 16x16 as overflows are likely to
* occur in interim calculations.
*
* Normal precision (i.e. #undef LOW_PRECISION) conforms to Q7.8 with the
* ability to store every interim result as Q23.8. Most operations like
* square root, sine, cosine, multiplication etc. utilize 32 bit types.
*/
#define LOW_PRECISION
#endif
#ifdef LOW_PRECISION
/** This is the type we expect ordinary integers to be. */
typedef int16_t ordinary_int_t;
/** This is the type which we use for fixed point values. */
typedef int16_t fixp_t;
/** This type covers arguments of fixSin() and fixCos(). */
typedef int16_t fixp_trig_t;
/** This type covers interim results of fixed point operations. */
typedef uint32_t fixp_interim_t;
/** This type covers interim results of the fixed point sqrt() function. */
typedef uint16_t ufixp_interim_t;
/** Amount of bits the fixed point sqrt() function can handle. */
#define SQRT_BITS 16
// NOTE: If you change the following values, don't forget to adapt the sine
// lookup table as well!
/** Multiply a number by this factor to convert it to a fixed point value.*/
#define FIX 32
/** Amount of fractional bits of a value (i.e. ceil(log_2(FIX))). */
#define FIX_FRACBITS 5
/**
* The amount of temporal quantization steps of the sine lookup table. It
* must be a divisor of (FIX * 2 * pi) and this divisor must be divisable by
* 4 itself. Approximate this value as close as possible to keep rounding
* errors at a minimum.
*/
#define FIX_SIN_COUNT 200
/** The rounded down quotient of (FIX * 2 * pi) and FIX_SIN_COUNT */
#define FIX_SIN_DIVIDER 1
/** Type of the lookup table elements. */
typedef uint8_t lut_t;
/**
* Lookup table of fractional parts which model the first quarter of a
* sine period. The rest of that period is calculated by mirroring those
* values. These values are intended for Q5 types.
*/
static lut_t const fix_sine_lut[FIX_SIN_COUNT / 4] =
{ 0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15,
15, 16, 17, 18, 19, 20, 20, 21,
22, 22, 23, 24, 24, 25, 26, 26,
27, 27, 28, 28, 29, 29, 29, 30,
30, 30, 31, 31, 31, 31, 31, 31,
31, 31};
#else
#define BUFFER pixmap
/** This is the type we expect ordinary integers to be. */
typedef int16_t ordinary_int_t;
/** This is the type which we use for fixed point values. */
typedef int16_t fixp_t;
/** This type covers arguments of fixSin() and fixCos(). */
typedef int32_t fixp_trig_t;
/** This type covers interim results of fixed point operations. */
typedef int32_t fixp_interim_t;
/** This type covers interim results of the fixed point sqrt() function. */
typedef uint32_t ufixp_interim_t;
/** Amount of bits the fixed point sqrt() function can handle. */
#define SQRT_BITS 32
// NOTE: If you change the following values, don't forget to adapt the sine
// lookup table as well!
/** Multiply a number by this factor to convert it to a fixed point value.*/
#define FIX 256
/** Amount of fractional bits of a value (i.e. ceil(log_2(FIX))). */
#define FIX_FRACBITS 8
/**
* The amount of temporal quantization steps of the sine lookup table. It
* must be a divisor of (FIX * 2 * pi) and this divisor must be divisable by
* 4 itself. Approximate this value as close as possible to keep rounding
* errors at a minimum.
*/
#define FIX_SIN_COUNT 200
/** The rounded down quotient of (FIX * 2 * pi) and FIX_SIN_COUNT */
#define FIX_SIN_DIVIDER 8
/** Type of the lookup table elements. */
typedef uint8_t lut_t;
/**
* Lookup table of fractional parts which model the first quarter of a
* sine period. The rest of that period is calculated by mirroring those
* values. These values are intended for Q8 types.
*/
static lut_t const fix_sine_lut[FIX_SIN_COUNT / 4] =
{ 0, 9, 17, 24, 32, 40, 48, 56,
64, 72, 79, 87, 94, 102, 109, 116,
123, 130, 137, 144, 150, 157, 163, 169,
175, 181, 186, 192, 197, 202, 207, 211,
216, 220, 224, 228, 231, 235, 238, 240,
243, 245, 247, 249, 251, 252, 253, 254,
255, 255};
#endif
/** The ordinary pi constant. */
#define PI 3.14159265358979323846
/** Fixed point version of (pi / 2). */
#define FIX_PI_2 ((fixp_t)(PI * FIX / 2))
/** Fixed point version of pi. */
#define FIX_PI ((fixp_t)(PI * FIX))
/** Fixed point version of (2 * pi). */
#define FIX_2PI ((fixp_t)(2 * PI * FIX))
/**
* Pointer to a function which return a value depending on two-dimensional
* coordinates and a "step" value.
* @param x x-coordinate
* @param y y-coordinate
* @param t a step value which changes for each frame, allowing for animations
* @return
* Scales an ordinary integer up to its fixed point format.
* @param a an ordinary integer to be scaled up
* @return The given value in fixed point format.
*/
typedef unsigned char (*fpmath_pattern_func_t)(unsigned char const x,
unsigned char const y,
double t);
inline static fixp_t fixScaleUp(ordinary_int_t a)
{
return (fixp_t)a * FIX;
}
/**
* Scales a fixed point value down to an ordinary integer (omitting the
* fractional part).
* @param a fixed point value to be scaled down
* @return The given value in fixed point format.
*/
inline static ordinary_int_t fixScaleDown(fixp_t const a)
{
return a / FIX;
}
/**
* Multiplies two fixed point values.
* @param a operand a
* @param b operand b
* @return Product of a and b.
*/
inline static fixp_interim_t fixMul(fixp_t const a, fixp_t const b)
{
return ((fixp_interim_t)a * (fixp_interim_t)b) / FIX;
}
/**
* Divides two fixed point values.
* @param a operand a
* @param b operand b
* @return Quotient of a and b.
*/
inline static fixp_t fixDiv(fixp_interim_t const a, fixp_interim_t const b)
{
return (a * FIX) / b;
}
/**
* Fixed point variant of the sine function which receives a fixed point angle
* (radian). It uses a lookup table which models the first quarter of a full
* sine period and calculates the rest from that quarter.
* @param angle fixed point radian value
* @return Result of the sine function normalized to a range from -FIX to FIX.
*/
static fixp_t fixSin(fixp_t const fAngle)
{
// convert given angle to its corresponding lookup table quantization step
ordinary_int_t nNormAng = fAngle / FIX_SIN_DIVIDER;
// trim that value so that it fits into a range between [0, FIX_SIN_COUNT]
nNormAng = (nNormAng - (nNormAng / FIX_SIN_COUNT * FIX_SIN_COUNT) +
FIX_SIN_COUNT) % FIX_SIN_COUNT;
uint8_t nIndex = nNormAng % (FIX_SIN_COUNT / 2);
if (nIndex >= (FIX_SIN_COUNT / 4))
{
nIndex = (FIX_SIN_COUNT / 2 - 1) - nIndex;
}
assert(nIndex < (FIX_SIN_COUNT / 4));
return ((fixp_t)fix_sine_lut[nIndex]) *
(nNormAng < (FIX_SIN_COUNT / 2) ? 1 : -1);
}
/**
* Fixed point variant of the cosine function which takes a fixed point angle
* (radian). It adds FIX_PI_2 to the given angle and consults the fixSin()
* function for the final result.
* @param angle fixed point radian value
* @return Result of the cosine function normalized to a range from -FIX to FIX.
*/
static fixp_t fixCos(fixp_t const angle)
{
return fixSin(angle + FIX_PI_2);
}
/**
* Fixed point square root algorithm as proposed by Ken Turkowski:
* http://www.realitypixels.com/turk/computergraphics/FixedSqrt.pdf
* @param radicant we want the square root of
* @return The square root of the given value.
*/
static fixp_t fixSqrt(ufixp_interim_t const a)
{
ufixp_interim_t nRoot, nRemainingHigh, nRemainingLow, nTestDiv, nCount;
nRoot = 0; // clear root
nRemainingHigh = 0; // clear high part of partial remainder
nRemainingLow = a; // get argument into low part of partial remainder
nCount = (SQRT_BITS / 2 - 1) + (FIX_FRACBITS >> 1); // load loop counter
do
{
nRemainingHigh =
(nRemainingHigh << 2) | (nRemainingLow >> (SQRT_BITS - 2));
nRemainingLow <<= 2; // get 2 bits of the argument
nRoot <<= 1; // get ready for the next bit in the root
nTestDiv = (nRoot << 1) + 1; // test radical
if (nRemainingHigh >= nTestDiv)
{
nRemainingHigh -= nTestDiv;
nRoot++;
}
} while (nCount-- != 0);
return (nRoot);
}
/**
@ -33,41 +280,69 @@ typedef unsigned char (*fpmath_pattern_func_t)(unsigned char const x,
* @param y1 y coordinate of the first point
* @param x2 x coordinate of the second point
* @param y2 y coordinate of the second point
* @return distance between the points
* @return The distance between the given coordinates.
*/
static double dist(double x1, double y1, double x2, double y2)
static fixp_t fixDist(fixp_t const x1,
fixp_t const y1,
fixp_t const x2,
fixp_t const y2)
{
return sqrt(((x1 - x2) * (x1 - x2)) + ((y1 - y2) * (y1 - y2)));
return fixSqrt(fixMul((x1 - x2), (x1 - x2)) + fixMul((y1 - y2), (y1 - y2)));
}
/**
* Draws an animated two dimensional graph for a given function f(x,y,t).
* @param t_start start value for the function's time variable
* @param t_stop stop value for the function's time variable
* @param t_delta value by which the function's timing variable gets incremented
* @param frame_delay frame delay in ms
* @param fpPattern function which generates a pattern depending on x, y, t
* This pointer type covers functions which return a brightness value for the
* given coordinates and a "step" value. This actually results in a more or less
* "beautiful" pattern.
* @param x x-coordinate
* @param y y-coordinate
* @param t step value which changes for each frame, allowing for animations
* @param r pointer to persistent data required by the pattern function
* @return The brightness value (0 < n <= NUM_PLANES) of the given coordinate.
*/
static void fpmath_pattern(double const t_start,
double const t_stop,
double const t_delta,
unsigned int const frame_delay,
fpmath_pattern_func_t fpPattern)
typedef unsigned char (*fpmath_pattern_func_t)(unsigned char const x,
unsigned char const y,
fixp_t const t,
void *const r);
#ifdef DOUBLE_BUFFERING
# define BUFFER pixmap_buffer
#else
# define BUFFER pixmap
#endif
/**
* Draws an animated two dimensional graph for a given function f(x, y, t).
* @param t_start start value for the function's step variable
* @param t_stop stop value for the function's step variable
* @param t_delta value by which the function's step variable gets incremented
* @param frame_delay frame delay in milliseconds
* @param fpPattern function which generates a pattern depending on x, y and t
* @param r pointer to persistent data required by the fpPattern function
*/
static void fixPattern(fixp_t const t_start,
fixp_t const t_stop,
fixp_t const t_delta,
int const frame_delay,
fpmath_pattern_func_t fpPattern,
void *r)
{
#ifdef DOUBLE_BUFFERING
// double buffering to reduce half painted pictures
unsigned char pixmap_buffer[NUMPLANE][NUM_ROWS][LINEBYTES];
#endif
for (double t = t_start; t < t_stop; t += t_delta)
for (fixp_t t = t_start; t < t_stop; t += t_delta)
{
for (unsigned char y = 0; y < NUM_ROWS; ++y)
{
unsigned char nChunk[NUMPLANE + 1][LINEBYTES] = {{0}};
for (unsigned char x = 0; x < (LINEBYTES * 8); ++x)
{
nChunk[fpPattern(x, y, t) - 1][x / 8u] |= shl_table[x % 8u];
assert (y < 16);
nChunk[fpPattern(x, y, t, r) - 1][x / 8u] |= shl_table[x % 8u];
}
for (unsigned char p = NUMPLANE; p--;)
{
@ -89,71 +364,133 @@ static void fpmath_pattern(double const t_start,
#ifdef ANIMATION_PLASMA
#define PLASMA_X (1.0 / (NUM_COLS / (2.0 * M_PI)))
/**
* Draws a simple plasma like pattern.
* This type maintains values relevant for the Plasma animation which need to be
* persistent over consecutive invocations.
*/
static unsigned char fpmath_plasma(unsigned char x, unsigned char y, double t)
typedef struct fixp_plasma_s
{
fixp_t fFunc1[NUM_COLS]; /**< Result of 1st trig. func. depending on x. */
fixp_t fFunc2CosArg; /**< Arg. of 2st trig. func. depending on the frame. */
fixp_t fFunc2SinArg; /**< Arg. of 2st trig. func. depending on the frame. */
} fixp_plasma_t;
/**
* Draws a plasma like pattern (sort of... four shades of grey are pretty
* scarce for a neat plasma animation).
* @param x x-coordinate
* @param y y-coordinate
* @param t step value which changes for each frame, allowing for animations
* @param r pointer to persistent interim results
* @return The brightness value (0 < n <= NUM_PLANES) of the given coordinate.
*/
static unsigned char fixAnimPlasma(unsigned char const x,
unsigned char const y,
fixp_t const t,
void *const r)
{
assert(x < NUM_COLS);
assert(y < NUM_ROWS);
static double fFunc1[NUM_COLS];
static double fFunc2CosArg;
static double fFunc2SinArg;
// scaling factor
static fixp_t const fPlasmaX = (2 * PI * FIX) / NUM_COLS;
// reentrant data
fixp_plasma_t *const p = (fixp_plasma_t *)r;
if (x == 0 && y == 0)
{
fFunc2CosArg = NUM_ROWS * cos(t) + NUM_ROWS;
fFunc2SinArg = NUM_COLS * sin(t) + NUM_COLS;
p->fFunc2CosArg = NUM_ROWS * fixCos(t) + fixScaleUp(NUM_ROWS);
p->fFunc2SinArg = NUM_COLS * fixSin(t) + fixScaleUp(NUM_COLS);
}
if (y == 0)
{
fFunc1[x] = sin(x * PLASMA_X + t);
p->fFunc1[x] = fixSin(fixMul(fixScaleUp(x), fPlasmaX) + t);
}
return (fFunc1[x] + sin(dist(x, y, fFunc2SinArg, fFunc2CosArg) * PLASMA_X)
+ 2) * (NUMPLANE - 1) / 2;
fixp_t const fFunc2 = fixSin(fixMul(fixDist(fixScaleUp(x), fixScaleUp(y),
p->fFunc2SinArg, p->fFunc2CosArg), fPlasmaX));
uint8_t const nRes = fixScaleDown(fixDiv(fixMul(p->fFunc1[x] + fFunc2 +
fixScaleUp(2), fixScaleUp(NUMPLANE - 1)), fixScaleUp(2)));
assert (nRes <= 3);
return nRes;
}
void plasma(void)
{
fixp_plasma_t r;
#ifndef __AVR__
fpmath_pattern(0.0, 75, 0.1, 80, fpmath_plasma);
fixPattern(0, fixScaleUp(75), 0.1 * FIX, 80, fixAnimPlasma, &r);
#else
fpmath_pattern(0.0, 60.0, 0.1, 1, fpmath_plasma);
#endif
fixPattern(0, fixScaleUp(60), 0.1 * FIX, 1, fixAnimPlasma, &r);
#endif /* __AVR__ */
}
#endif
#endif /* ANIMATION_PLASMA */
#ifdef ANIMATION_PSYCHEDELIC
/**
* This type maintains values relevant for the Psychedelic animation which need
* to be persistent over consecutive invocations.
*/
typedef struct fixp_psychedelic_s
{
fixp_t fCos; /** column factor for the circle calculation */
fixp_t fSin; /** row factor for the circle calculation */
fixp_interim_t ft10; /** value involved in rotating the animation's center*/
} fixp_psychedelic_t;
/**
* Draws flowing circular waves with a rotating center.
* @param x x-coordinate
* @param y y-coordinate
* @param t step value which changes for each frame, allowing for animations
* @param r pointer to persistent interim results
* @return The brightness value (0 < n <= NUM_PLANES) of the given coordinate.
*/
static unsigned char fpmath_psycho(unsigned char x, unsigned char y, double t)
static unsigned char fixAnimPsychedelic(unsigned char const x,
unsigned char const y,
fixp_t const t,
void *const r)
{
assert(x < NUM_COLS);
assert(y < NUM_ROWS);
fixp_psychedelic_t *p = (fixp_psychedelic_t *)r;
static double fCosinus;
static double fSinus;
static double t10;
if (x == 0 && y == 0)
{
fCosinus = NUM_COLS * cos(t);
fSinus = NUM_ROWS * sin(t);
t10 = t * 10;
p->fCos = NUM_COLS/2 * fixCos(t);
p->fSin = NUM_ROWS/2 * fixSin(t);
p->ft10 = fixMul(t, fixScaleUp(10));
}
return (sin(dist(x, y, fCosinus, fSinus) - t10) + 1) * (NUMPLANE - 1);
uint8_t const nResult =
fixScaleDown(fixMul(fixSin((fixp_interim_t)fixDist(fixScaleUp(x),
fixScaleUp(y), p->fCos, p->fSin) - p->ft10) + fixScaleUp(1),
fixScaleUp(NUMPLANE - 1)));
assert(nResult <= NUMPLANE);
return nResult;
}
void psychedelic(void)
{
fixp_psychedelic_t r;
#ifndef __AVR__
fpmath_pattern(0.0, 75, 0.1, 80, fpmath_psycho);
fixPattern(0, fixScaleUp(75), 0.1 * FIX, 80, fixAnimPsychedelic, &r);
#else
fpmath_pattern(0.0, 60.0, 0.1, 1, fpmath_psycho);
#endif
fixPattern(0, fixScaleUp(60), 0.1 * FIX, 15, fixAnimPsychedelic, &r);
#endif /* __AVR__ */
}
#endif
#endif /* ANIMATION_PSYCHEDELIC */
/*@}*/