More data in UBL class, make it a static class
- Make all `unified_bed_leveling` data/methods static - Move some UBL-related variables into the class - Replace `map_[xy]_index_to_bed_location` with `mesh_index_to_[xy]pos`
This commit is contained in:
452
Marlin/UBL.h
452
Marlin/UBL.h
@@ -39,7 +39,6 @@
|
||||
|
||||
enum MeshPointType { INVALID, REAL, SET_IN_BITMAP };
|
||||
|
||||
bool axis_unhomed_error(bool, bool, bool);
|
||||
void dump(char * const str, const float &f);
|
||||
bool ubl_lcd_clicked();
|
||||
void probe_entire_mesh(const float&, const float&, const bool, const bool, const bool);
|
||||
@@ -78,275 +77,273 @@
|
||||
|
||||
enum MBLStatus { MBL_STATUS_NONE = 0, MBL_STATUS_HAS_MESH_BIT = 0, MBL_STATUS_ACTIVE_BIT = 1 };
|
||||
|
||||
#define MESH_X_DIST ((float(UBL_MESH_MAX_X) - float(UBL_MESH_MIN_X)) / (float(UBL_MESH_NUM_X_POINTS) - 1.0))
|
||||
#define MESH_Y_DIST ((float(UBL_MESH_MAX_Y) - float(UBL_MESH_MIN_Y)) / (float(UBL_MESH_NUM_Y_POINTS) - 1.0))
|
||||
#define MESH_X_DIST (float(UBL_MESH_MAX_X - (UBL_MESH_MIN_X)) / float(UBL_MESH_NUM_X_POINTS - 1))
|
||||
#define MESH_Y_DIST (float(UBL_MESH_MAX_Y - (UBL_MESH_MIN_Y)) / float(UBL_MESH_NUM_Y_POINTS - 1))
|
||||
|
||||
extern float mesh_index_to_x_location[UBL_MESH_NUM_X_POINTS + 1]; // +1 just because of paranoia that we might end up on the
|
||||
extern float mesh_index_to_y_location[UBL_MESH_NUM_Y_POINTS + 1]; // the last Mesh Line and that is the start of a whole new cell
|
||||
typedef struct {
|
||||
bool active = false;
|
||||
float z_offset = 0.0;
|
||||
int8_t eeprom_storage_slot = -1,
|
||||
n_x = UBL_MESH_NUM_X_POINTS,
|
||||
n_y = UBL_MESH_NUM_Y_POINTS;
|
||||
|
||||
float mesh_x_min = UBL_MESH_MIN_X,
|
||||
mesh_y_min = UBL_MESH_MIN_Y,
|
||||
mesh_x_max = UBL_MESH_MAX_X,
|
||||
mesh_y_max = UBL_MESH_MAX_Y,
|
||||
mesh_x_dist = MESH_X_DIST,
|
||||
mesh_y_dist = MESH_Y_DIST;
|
||||
|
||||
#if ENABLED(ENABLE_LEVELING_FADE_HEIGHT)
|
||||
float g29_correction_fade_height = 10.0,
|
||||
g29_fade_height_multiplier = 1.0 / 10.0; // It's cheaper to do a floating point multiply than divide,
|
||||
// so keep this value and its reciprocal.
|
||||
#else
|
||||
const float g29_correction_fade_height = 10.0,
|
||||
g29_fade_height_multiplier = 1.0 / 10.0;
|
||||
#endif
|
||||
|
||||
// If you change this struct, adjust TOTAL_STRUCT_SIZE
|
||||
|
||||
#define TOTAL_STRUCT_SIZE 40 // Total size of the above fields
|
||||
|
||||
// padding provides space to add state variables without
|
||||
// changing the location of data structures in the EEPROM.
|
||||
// This is for compatibility with future versions to keep
|
||||
// users from having to regenerate their mesh data.
|
||||
unsigned char padding[64 - TOTAL_STRUCT_SIZE];
|
||||
|
||||
} ubl_state;
|
||||
|
||||
class unified_bed_leveling {
|
||||
private:
|
||||
|
||||
float last_specified_z,
|
||||
fade_scaling_factor_for_current_height;
|
||||
static float last_specified_z,
|
||||
fade_scaling_factor_for_current_height;
|
||||
|
||||
public:
|
||||
|
||||
float z_values[UBL_MESH_NUM_X_POINTS][UBL_MESH_NUM_Y_POINTS];
|
||||
static ubl_state state, pre_initialized;
|
||||
|
||||
bool g26_debug_flag = false,
|
||||
has_control_of_lcd_panel = false;
|
||||
static float z_values[UBL_MESH_NUM_X_POINTS][UBL_MESH_NUM_Y_POINTS],
|
||||
mesh_index_to_xpos[UBL_MESH_NUM_X_POINTS + 1], // +1 safety margin for now, until determinism prevails
|
||||
mesh_index_to_ypos[UBL_MESH_NUM_Y_POINTS + 1];
|
||||
|
||||
int8_t eeprom_start = -1;
|
||||
static bool g26_debug_flag,
|
||||
has_control_of_lcd_panel;
|
||||
|
||||
volatile int encoder_diff; // Volatile because it's changed at interrupt time.
|
||||
static int8_t eeprom_start;
|
||||
|
||||
struct ubl_state {
|
||||
bool active = false;
|
||||
float z_offset = 0.0;
|
||||
int8_t eeprom_storage_slot = -1,
|
||||
n_x = UBL_MESH_NUM_X_POINTS,
|
||||
n_y = UBL_MESH_NUM_Y_POINTS;
|
||||
static volatile int encoder_diff; // Volatile because it's changed at interrupt time.
|
||||
|
||||
float mesh_x_min = UBL_MESH_MIN_X,
|
||||
mesh_y_min = UBL_MESH_MIN_Y,
|
||||
mesh_x_max = UBL_MESH_MAX_X,
|
||||
mesh_y_max = UBL_MESH_MAX_Y,
|
||||
mesh_x_dist = MESH_X_DIST,
|
||||
mesh_y_dist = MESH_Y_DIST;
|
||||
unified_bed_leveling();
|
||||
|
||||
#if ENABLED(ENABLE_LEVELING_FADE_HEIGHT)
|
||||
float g29_correction_fade_height = 10.0,
|
||||
g29_fade_height_multiplier = 1.0 / 10.0; // It's cheaper to do a floating point multiply than divide,
|
||||
// so keep this value and its reciprocal.
|
||||
#else
|
||||
const float g29_correction_fade_height = 10.0,
|
||||
g29_fade_height_multiplier = 1.0 / 10.0;
|
||||
#endif
|
||||
static void display_map(const int);
|
||||
|
||||
// If you change this struct, adjust TOTAL_STRUCT_SIZE
|
||||
static void reset();
|
||||
static void invalidate();
|
||||
|
||||
#define TOTAL_STRUCT_SIZE 43 // Total size of the above fields
|
||||
static void store_state();
|
||||
static void load_state();
|
||||
static void store_mesh(const int16_t);
|
||||
static void load_mesh(const int16_t);
|
||||
|
||||
// padding provides space to add state variables without
|
||||
// changing the location of data structures in the EEPROM.
|
||||
// This is for compatibility with future versions to keep
|
||||
// users from having to regenerate their mesh data.
|
||||
unsigned char padding[64 - TOTAL_STRUCT_SIZE];
|
||||
static bool sanity_check();
|
||||
|
||||
} state, pre_initialized;
|
||||
static FORCE_INLINE void set_z(const int8_t px, const int8_t py, const float &z) { z_values[px][py] = z; }
|
||||
|
||||
unified_bed_leveling();
|
||||
static int8_t get_cell_index_x(const float &x) {
|
||||
const int8_t cx = (x - (UBL_MESH_MIN_X)) * (1.0 / (MESH_X_DIST));
|
||||
return constrain(cx, 0, (UBL_MESH_NUM_X_POINTS) - 1); // -1 is appropriate if we want all movement to the X_MAX
|
||||
} // position. But with this defined this way, it is possible
|
||||
// to extrapolate off of this point even further out. Probably
|
||||
// that is OK because something else should be keeping that from
|
||||
// happening and should not be worried about at this level.
|
||||
static int8_t get_cell_index_y(const float &y) {
|
||||
const int8_t cy = (y - (UBL_MESH_MIN_Y)) * (1.0 / (MESH_Y_DIST));
|
||||
return constrain(cy, 0, (UBL_MESH_NUM_Y_POINTS) - 1); // -1 is appropriate if we want all movement to the Y_MAX
|
||||
} // position. But with this defined this way, it is possible
|
||||
// to extrapolate off of this point even further out. Probably
|
||||
// that is OK because something else should be keeping that from
|
||||
// happening and should not be worried about at this level.
|
||||
|
||||
void display_map(const int);
|
||||
|
||||
void reset();
|
||||
void invalidate();
|
||||
|
||||
void store_state();
|
||||
void load_state();
|
||||
void store_mesh(const int16_t);
|
||||
void load_mesh(const int16_t);
|
||||
|
||||
bool sanity_check();
|
||||
|
||||
FORCE_INLINE static float map_x_index_to_bed_location(const int8_t i) { return ((float) UBL_MESH_MIN_X) + (((float) MESH_X_DIST) * (float) i); };
|
||||
FORCE_INLINE static float map_y_index_to_bed_location(const int8_t i) { return ((float) UBL_MESH_MIN_Y) + (((float) MESH_Y_DIST) * (float) i); };
|
||||
|
||||
FORCE_INLINE void set_z(const int8_t px, const int8_t py, const float &z) { z_values[px][py] = z; }
|
||||
|
||||
static int8_t get_cell_index_x(const float &x) {
|
||||
const int8_t cx = (x - (UBL_MESH_MIN_X)) * (1.0 / (MESH_X_DIST));
|
||||
return constrain(cx, 0, (UBL_MESH_NUM_X_POINTS) - 1); // -1 is appropriate if we want all movement to the X_MAX
|
||||
} // position. But with this defined this way, it is possible
|
||||
// to extrapolate off of this point even further out. Probably
|
||||
// that is OK because something else should be keeping that from
|
||||
// happening and should not be worried about at this level.
|
||||
static int8_t get_cell_index_y(const float &y) {
|
||||
const int8_t cy = (y - (UBL_MESH_MIN_Y)) * (1.0 / (MESH_Y_DIST));
|
||||
return constrain(cy, 0, (UBL_MESH_NUM_Y_POINTS) - 1); // -1 is appropriate if we want all movement to the Y_MAX
|
||||
} // position. But with this defined this way, it is possible
|
||||
// to extrapolate off of this point even further out. Probably
|
||||
// that is OK because something else should be keeping that from
|
||||
// happening and should not be worried about at this level.
|
||||
|
||||
static int8_t find_closest_x_index(const float &x) {
|
||||
const int8_t px = (x - (UBL_MESH_MIN_X) + (MESH_X_DIST) * 0.5) * (1.0 / (MESH_X_DIST));
|
||||
return (px >= 0 && px < (UBL_MESH_NUM_X_POINTS)) ? px : -1;
|
||||
}
|
||||
|
||||
static int8_t find_closest_y_index(const float &y) {
|
||||
const int8_t py = (y - (UBL_MESH_MIN_Y) + (MESH_Y_DIST) * 0.5) * (1.0 / (MESH_Y_DIST));
|
||||
return (py >= 0 && py < (UBL_MESH_NUM_Y_POINTS)) ? py : -1;
|
||||
}
|
||||
|
||||
/**
|
||||
* z2 --|
|
||||
* z0 | |
|
||||
* | | + (z2-z1)
|
||||
* z1 | | |
|
||||
* ---+-------------+--------+-- --|
|
||||
* a1 a0 a2
|
||||
* |<---delta_a---------->|
|
||||
*
|
||||
* calc_z0 is the basis for all the Mesh Based correction. It is used to
|
||||
* find the expected Z Height at a position between two known Z-Height locations.
|
||||
*
|
||||
* It is fairly expensive with its 4 floating point additions and 2 floating point
|
||||
* multiplications.
|
||||
*/
|
||||
static FORCE_INLINE float calc_z0(const float &a0, const float &a1, const float &z1, const float &a2, const float &z2) {
|
||||
const float delta_z = (z2 - z1),
|
||||
delta_a = (a0 - a1) / (a2 - a1);
|
||||
return z1 + delta_a * delta_z;
|
||||
}
|
||||
|
||||
/**
|
||||
* get_z_correction_at_Y_intercept(float x0, int x1_i, int yi) only takes
|
||||
* three parameters. It assumes the x0 point is on a Mesh line denoted by yi. In theory
|
||||
* we could use get_cell_index_x(float x) to obtain the 2nd parameter x1_i but any code calling
|
||||
* the get_z_correction_along_vertical_mesh_line_at_specific_X routine will already have
|
||||
* the X index of the x0 intersection available and we don't want to perform any extra floating
|
||||
* point operations.
|
||||
*/
|
||||
inline float get_z_correction_along_horizontal_mesh_line_at_specific_X(const float &x0, const int x1_i, const int yi) {
|
||||
if (x1_i < 0 || yi < 0 || x1_i >= UBL_MESH_NUM_X_POINTS || yi >= UBL_MESH_NUM_Y_POINTS) {
|
||||
SERIAL_ECHOPAIR("? in get_z_correction_along_horizontal_mesh_line_at_specific_X(x0=", x0);
|
||||
SERIAL_ECHOPAIR(",x1_i=", x1_i);
|
||||
SERIAL_ECHOPAIR(",yi=", yi);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_EOL;
|
||||
return NAN;
|
||||
static int8_t find_closest_x_index(const float &x) {
|
||||
const int8_t px = (x - (UBL_MESH_MIN_X) + (MESH_X_DIST) * 0.5) * (1.0 / (MESH_X_DIST));
|
||||
return (px >= 0 && px < (UBL_MESH_NUM_X_POINTS)) ? px : -1;
|
||||
}
|
||||
|
||||
const float xratio = (RAW_X_POSITION(x0) - mesh_index_to_x_location[x1_i]) * (1.0 / (MESH_X_DIST)),
|
||||
z1 = z_values[x1_i][yi],
|
||||
z2 = z_values[x1_i + 1][yi],
|
||||
dz = (z2 - z1);
|
||||
|
||||
return z1 + xratio * dz;
|
||||
}
|
||||
|
||||
//
|
||||
// See comments above for get_z_correction_along_horizontal_mesh_line_at_specific_X
|
||||
//
|
||||
inline float get_z_correction_along_vertical_mesh_line_at_specific_Y(const float &y0, const int xi, const int y1_i) {
|
||||
if (xi < 0 || y1_i < 0 || xi >= UBL_MESH_NUM_X_POINTS || y1_i >= UBL_MESH_NUM_Y_POINTS) {
|
||||
SERIAL_ECHOPAIR("? in get_z_correction_along_vertical_mesh_line_at_specific_X(y0=", y0);
|
||||
SERIAL_ECHOPAIR(", x1_i=", xi);
|
||||
SERIAL_ECHOPAIR(", yi=", y1_i);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_EOL;
|
||||
return NAN;
|
||||
static int8_t find_closest_y_index(const float &y) {
|
||||
const int8_t py = (y - (UBL_MESH_MIN_Y) + (MESH_Y_DIST) * 0.5) * (1.0 / (MESH_Y_DIST));
|
||||
return (py >= 0 && py < (UBL_MESH_NUM_Y_POINTS)) ? py : -1;
|
||||
}
|
||||
|
||||
const float yratio = (RAW_Y_POSITION(y0) - mesh_index_to_y_location[y1_i]) * (1.0 / (MESH_Y_DIST)),
|
||||
z1 = z_values[xi][y1_i],
|
||||
z2 = z_values[xi][y1_i + 1],
|
||||
dz = (z2 - z1);
|
||||
|
||||
return z1 + yratio * dz;
|
||||
}
|
||||
|
||||
/**
|
||||
* This is the generic Z-Correction. It works anywhere within a Mesh Cell. It first
|
||||
* does a linear interpolation along both of the bounding X-Mesh-Lines to find the
|
||||
* Z-Height at both ends. Then it does a linear interpolation of these heights based
|
||||
* on the Y position within the cell.
|
||||
*/
|
||||
float get_z_correction(const float &x0, const float &y0) const {
|
||||
const int8_t cx = get_cell_index_x(RAW_X_POSITION(x0)),
|
||||
cy = get_cell_index_y(RAW_Y_POSITION(y0));
|
||||
|
||||
if (cx < 0 || cy < 0 || cx >= UBL_MESH_NUM_X_POINTS || cy >= UBL_MESH_NUM_Y_POINTS) {
|
||||
|
||||
SERIAL_ECHOPAIR("? in get_z_correction(x0=", x0);
|
||||
SERIAL_ECHOPAIR(", y0=", y0);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_EOL;
|
||||
|
||||
#if ENABLED(ULTRA_LCD)
|
||||
strcpy(lcd_status_message, "get_z_correction() indexes out of range.");
|
||||
lcd_quick_feedback();
|
||||
#endif
|
||||
return 0.0; // this used to return state.z_offset
|
||||
/**
|
||||
* z2 --|
|
||||
* z0 | |
|
||||
* | | + (z2-z1)
|
||||
* z1 | | |
|
||||
* ---+-------------+--------+-- --|
|
||||
* a1 a0 a2
|
||||
* |<---delta_a---------->|
|
||||
*
|
||||
* calc_z0 is the basis for all the Mesh Based correction. It is used to
|
||||
* find the expected Z Height at a position between two known Z-Height locations.
|
||||
*
|
||||
* It is fairly expensive with its 4 floating point additions and 2 floating point
|
||||
* multiplications.
|
||||
*/
|
||||
static FORCE_INLINE float calc_z0(const float &a0, const float &a1, const float &z1, const float &a2, const float &z2) {
|
||||
const float delta_z = (z2 - z1),
|
||||
delta_a = (a0 - a1) / (a2 - a1);
|
||||
return z1 + delta_a * delta_z;
|
||||
}
|
||||
|
||||
const float z1 = calc_z0(RAW_X_POSITION(x0),
|
||||
map_x_index_to_bed_location(cx), z_values[cx][cy],
|
||||
map_x_index_to_bed_location(cx + 1), z_values[cx + 1][cy]),
|
||||
z2 = calc_z0(RAW_X_POSITION(x0),
|
||||
map_x_index_to_bed_location(cx), z_values[cx][cy + 1],
|
||||
map_x_index_to_bed_location(cx + 1), z_values[cx + 1][cy + 1]);
|
||||
float z0 = calc_z0(RAW_Y_POSITION(y0),
|
||||
map_y_index_to_bed_location(cy), z1,
|
||||
map_y_index_to_bed_location(cy + 1), z2);
|
||||
|
||||
#if ENABLED(DEBUG_LEVELING_FEATURE)
|
||||
if (DEBUGGING(MESH_ADJUST)) {
|
||||
SERIAL_ECHOPAIR(" raw get_z_correction(", x0);
|
||||
SERIAL_CHAR(',')
|
||||
SERIAL_ECHO(y0);
|
||||
SERIAL_ECHOPGM(") = ");
|
||||
SERIAL_ECHO_F(z0, 6);
|
||||
}
|
||||
#endif
|
||||
|
||||
#if ENABLED(DEBUG_LEVELING_FEATURE)
|
||||
if (DEBUGGING(MESH_ADJUST)) {
|
||||
SERIAL_ECHOPGM(" >>>---> ");
|
||||
SERIAL_ECHO_F(z0, 6);
|
||||
/**
|
||||
* get_z_correction_at_Y_intercept(float x0, int x1_i, int yi) only takes
|
||||
* three parameters. It assumes the x0 point is on a Mesh line denoted by yi. In theory
|
||||
* we could use get_cell_index_x(float x) to obtain the 2nd parameter x1_i but any code calling
|
||||
* the get_z_correction_along_vertical_mesh_line_at_specific_X routine will already have
|
||||
* the X index of the x0 intersection available and we don't want to perform any extra floating
|
||||
* point operations.
|
||||
*/
|
||||
static inline float get_z_correction_along_horizontal_mesh_line_at_specific_X(const float &x0, const int x1_i, const int yi) {
|
||||
if (x1_i < 0 || yi < 0 || x1_i >= UBL_MESH_NUM_X_POINTS || yi >= UBL_MESH_NUM_Y_POINTS) {
|
||||
SERIAL_ECHOPAIR("? in get_z_correction_along_horizontal_mesh_line_at_specific_X(x0=", x0);
|
||||
SERIAL_ECHOPAIR(",x1_i=", x1_i);
|
||||
SERIAL_ECHOPAIR(",yi=", yi);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_EOL;
|
||||
return NAN;
|
||||
}
|
||||
#endif
|
||||
|
||||
if (isnan(z0)) { // if part of the Mesh is undefined, it will show up as NAN
|
||||
z0 = 0.0; // in ubl.z_values[][] and propagate through the
|
||||
// calculations. If our correction is NAN, we throw it out
|
||||
// because part of the Mesh is undefined and we don't have the
|
||||
// information we need to complete the height correction.
|
||||
const float xratio = (RAW_X_POSITION(x0) - mesh_index_to_xpos[x1_i]) * (1.0 / (MESH_X_DIST)),
|
||||
z1 = z_values[x1_i][yi],
|
||||
z2 = z_values[x1_i + 1][yi],
|
||||
dz = (z2 - z1);
|
||||
|
||||
return z1 + xratio * dz;
|
||||
}
|
||||
|
||||
//
|
||||
// See comments above for get_z_correction_along_horizontal_mesh_line_at_specific_X
|
||||
//
|
||||
static inline float get_z_correction_along_vertical_mesh_line_at_specific_Y(const float &y0, const int xi, const int y1_i) {
|
||||
if (xi < 0 || y1_i < 0 || xi >= UBL_MESH_NUM_X_POINTS || y1_i >= UBL_MESH_NUM_Y_POINTS) {
|
||||
SERIAL_ECHOPAIR("? in get_z_correction_along_vertical_mesh_line_at_specific_X(y0=", y0);
|
||||
SERIAL_ECHOPAIR(", x1_i=", xi);
|
||||
SERIAL_ECHOPAIR(", yi=", y1_i);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_EOL;
|
||||
return NAN;
|
||||
}
|
||||
|
||||
const float yratio = (RAW_Y_POSITION(y0) - mesh_index_to_ypos[y1_i]) * (1.0 / (MESH_Y_DIST)),
|
||||
z1 = z_values[xi][y1_i],
|
||||
z2 = z_values[xi][y1_i + 1],
|
||||
dz = (z2 - z1);
|
||||
|
||||
return z1 + yratio * dz;
|
||||
}
|
||||
|
||||
/**
|
||||
* This is the generic Z-Correction. It works anywhere within a Mesh Cell. It first
|
||||
* does a linear interpolation along both of the bounding X-Mesh-Lines to find the
|
||||
* Z-Height at both ends. Then it does a linear interpolation of these heights based
|
||||
* on the Y position within the cell.
|
||||
*/
|
||||
static float get_z_correction(const float &x0, const float &y0) {
|
||||
const int8_t cx = get_cell_index_x(RAW_X_POSITION(x0)),
|
||||
cy = get_cell_index_y(RAW_Y_POSITION(y0));
|
||||
|
||||
if (cx < 0 || cy < 0 || cx >= UBL_MESH_NUM_X_POINTS || cy >= UBL_MESH_NUM_Y_POINTS) {
|
||||
|
||||
SERIAL_ECHOPAIR("? in get_z_correction(x0=", x0);
|
||||
SERIAL_ECHOPAIR(", y0=", y0);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_EOL;
|
||||
|
||||
#if ENABLED(ULTRA_LCD)
|
||||
strcpy(lcd_status_message, "get_z_correction() indexes out of range.");
|
||||
lcd_quick_feedback();
|
||||
#endif
|
||||
return 0.0; // this used to return state.z_offset
|
||||
}
|
||||
|
||||
const float z1 = calc_z0(RAW_X_POSITION(x0),
|
||||
mesh_index_to_xpos[cx], z_values[cx][cy],
|
||||
mesh_index_to_xpos[cx + 1], z_values[cx + 1][cy]),
|
||||
z2 = calc_z0(RAW_X_POSITION(x0),
|
||||
mesh_index_to_xpos[cx], z_values[cx][cy + 1],
|
||||
mesh_index_to_xpos[cx + 1], z_values[cx + 1][cy + 1]);
|
||||
float z0 = calc_z0(RAW_Y_POSITION(y0),
|
||||
mesh_index_to_ypos[cy], z1,
|
||||
mesh_index_to_ypos[cy + 1], z2);
|
||||
|
||||
#if ENABLED(DEBUG_LEVELING_FEATURE)
|
||||
if (DEBUGGING(MESH_ADJUST)) {
|
||||
SERIAL_ECHOPAIR("??? Yikes! NAN in get_z_correction(", x0);
|
||||
SERIAL_CHAR(',');
|
||||
SERIAL_ECHOPAIR(" raw get_z_correction(", x0);
|
||||
SERIAL_CHAR(',')
|
||||
SERIAL_ECHO(y0);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_ECHOPGM(") = ");
|
||||
SERIAL_ECHO_F(z0, 6);
|
||||
}
|
||||
#endif
|
||||
|
||||
#if ENABLED(DEBUG_LEVELING_FEATURE)
|
||||
if (DEBUGGING(MESH_ADJUST)) {
|
||||
SERIAL_ECHOPGM(" >>>---> ");
|
||||
SERIAL_ECHO_F(z0, 6);
|
||||
SERIAL_EOL;
|
||||
}
|
||||
#endif
|
||||
}
|
||||
return z0; // there used to be a +state.z_offset on this line
|
||||
}
|
||||
|
||||
/**
|
||||
* This routine is used to scale the Z correction depending upon the current nozzle height. It is
|
||||
* optimized for speed. It avoids floating point operations by checking if the requested scaling
|
||||
* factor is going to be the same as the last time the function calculated a value. If so, it just
|
||||
* returns it.
|
||||
*
|
||||
* It returns a scaling factor of 1.0 if UBL is inactive.
|
||||
* It returns a scaling factor of 0.0 if Z is past the specified 'Fade Height'
|
||||
*/
|
||||
#if ENABLED(ENABLE_LEVELING_FADE_HEIGHT)
|
||||
if (isnan(z0)) { // if part of the Mesh is undefined, it will show up as NAN
|
||||
z0 = 0.0; // in ubl.z_values[][] and propagate through the
|
||||
// calculations. If our correction is NAN, we throw it out
|
||||
// because part of the Mesh is undefined and we don't have the
|
||||
// information we need to complete the height correction.
|
||||
|
||||
FORCE_INLINE float fade_scaling_factor_for_z(const float &lz) {
|
||||
const float rz = RAW_Z_POSITION(lz);
|
||||
if (last_specified_z != rz) {
|
||||
last_specified_z = rz;
|
||||
fade_scaling_factor_for_current_height =
|
||||
state.active && rz < state.g29_correction_fade_height
|
||||
? 1.0 - (rz * state.g29_fade_height_multiplier)
|
||||
: 0.0;
|
||||
#if ENABLED(DEBUG_LEVELING_FEATURE)
|
||||
if (DEBUGGING(MESH_ADJUST)) {
|
||||
SERIAL_ECHOPAIR("??? Yikes! NAN in get_z_correction(", x0);
|
||||
SERIAL_CHAR(',');
|
||||
SERIAL_ECHO(y0);
|
||||
SERIAL_CHAR(')');
|
||||
SERIAL_EOL;
|
||||
}
|
||||
#endif
|
||||
}
|
||||
return fade_scaling_factor_for_current_height;
|
||||
return z0; // there used to be a +state.z_offset on this line
|
||||
}
|
||||
|
||||
#else
|
||||
/**
|
||||
* This routine is used to scale the Z correction depending upon the current nozzle height. It is
|
||||
* optimized for speed. It avoids floating point operations by checking if the requested scaling
|
||||
* factor is going to be the same as the last time the function calculated a value. If so, it just
|
||||
* returns it.
|
||||
*
|
||||
* It returns a scaling factor of 1.0 if UBL is inactive.
|
||||
* It returns a scaling factor of 0.0 if Z is past the specified 'Fade Height'
|
||||
*/
|
||||
#if ENABLED(ENABLE_LEVELING_FADE_HEIGHT)
|
||||
|
||||
static constexpr float fade_scaling_factor_for_z(const float &lz) { UNUSED(lz); return 1.0; }
|
||||
FORCE_INLINE float fade_scaling_factor_for_z(const float &lz) {
|
||||
const float rz = RAW_Z_POSITION(lz);
|
||||
if (last_specified_z != rz) {
|
||||
last_specified_z = rz;
|
||||
fade_scaling_factor_for_current_height =
|
||||
state.active && rz < state.g29_correction_fade_height
|
||||
? 1.0 - (rz * state.g29_fade_height_multiplier)
|
||||
: 0.0;
|
||||
}
|
||||
return fade_scaling_factor_for_current_height;
|
||||
}
|
||||
|
||||
#endif
|
||||
#else
|
||||
|
||||
static constexpr float fade_scaling_factor_for_z(const float &lz) { UNUSED(lz); return 1.0; }
|
||||
|
||||
#endif
|
||||
|
||||
}; // class unified_bed_leveling
|
||||
|
||||
@@ -355,5 +352,4 @@
|
||||
#define UBL_LAST_EEPROM_INDEX (E2END - sizeof(unified_bed_leveling::state))
|
||||
|
||||
#endif // AUTO_BED_LEVELING_UBL
|
||||
|
||||
#endif // UNIFIED_BED_LEVELING_H
|
||||
|
||||
Reference in New Issue
Block a user