From 2c238039c2efda1788ea72c9fb41ff354acc8e97 Mon Sep 17 00:00:00 2001 From: Thomas White Date: Tue, 15 Nov 2011 16:04:46 +0100 Subject: Move the "indexed reflection array" thing to where it can't do any harm --- src/diffraction.c | 464 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 464 insertions(+) create mode 100644 src/diffraction.c (limited to 'src/diffraction.c') diff --git a/src/diffraction.c b/src/diffraction.c new file mode 100644 index 00000000..de994133 --- /dev/null +++ b/src/diffraction.c @@ -0,0 +1,464 @@ +/* + * diffraction.c + * + * Calculate diffraction patterns by Fourier methods + * + * (c) 2006-2011 Thomas White + * + * Part of CrystFEL - crystallography with a FEL + * + */ + + +#include +#include +#include +#include +#include +#include +#include + +#include "image.h" +#include "utils.h" +#include "cell.h" +#include "diffraction.h" +#include "beam-parameters.h" +#include "symmetry.h" +#include "pattern_sim.h" + + +#define SAMPLING (4) +#define BWSAMPLING (10) +#define DIVSAMPLING (1) +#define SINC_LUT_ELEMENTS (4096) + + +static double *get_sinc_lut(int n) +{ + int i; + double *lut; + + lut = malloc(SINC_LUT_ELEMENTS*sizeof(double)); + lut[0] = n; + if ( n == 1 ) { + for ( i=1; i= 0.0); + + val1 = sym_lookup_intensity(ref, flags, sym, h, k, l); + val2 = sym_lookup_intensity(ref, flags, sym, h+1, k, l); + + val1 = val1; + val2 = val2; + + return (1.0-f)*val1 + f*val2; +} + + +static double interpolate_bilinear(const double *ref, + const unsigned char *flags, + const SymOpList *sym, + float hd, float kd, signed int l) +{ + signed int k; + double val1, val2; + float f; + + k = (signed int)kd; + if ( kd < 0.0 ) k -= 1; + f = kd - (float)k; + assert(f >= 0.0); + + val1 = interpolate_linear(ref, flags, sym, hd, k, l); + val2 = interpolate_linear(ref, flags, sym, hd, k+1, l); + + return (1.0-f)*val1 + f*val2; +} + + +static double interpolate_intensity(const double *ref, + const unsigned char *flags, + const SymOpList *sym, + float hd, float kd, float ld) +{ + signed int l; + double val1, val2; + float f; + + l = (signed int)ld; + if ( ld < 0.0 ) l -= 1; + f = ld - (float)l; + assert(f >= 0.0); + + val1 = interpolate_bilinear(ref, flags, sym, hd, kd, l); + val2 = interpolate_bilinear(ref, flags, sym, hd, kd, l+1); + + return (1.0-f)*val1 + f*val2; +} + + +static double complex interpolate_phased_linear(const double *ref, + const double *phases, + const unsigned char *flags, + const SymOpList *sym, + float hd, + signed int k, signed int l) +{ + signed int h; + double val1, val2; + float f; + double ph1, ph2; + double re1, re2, im1, im2; + double re, im; + + h = (signed int)hd; + if ( hd < 0.0 ) h -= 1; + f = hd - (float)h; + assert(f >= 0.0); + + val1 = sym_lookup_intensity(ref, flags, sym, h, k, l); + val2 = sym_lookup_intensity(ref, flags, sym, h+1, k, l); + ph1 = sym_lookup_phase(phases, flags, sym, h, k, l); + ph2 = sym_lookup_phase(phases, flags, sym, h+1, k, l); + + val1 = val1; + val2 = val2; + + /* Calculate real and imaginary parts */ + re1 = val1 * cos(ph1); + im1 = val1 * sin(ph1); + re2 = val2 * cos(ph2); + im2 = val2 * sin(ph2); + + re = (1.0-f)*re1 + f*re2; + im = (1.0-f)*im1 + f*im2; + + return re + im*I; +} + + +static double complex interpolate_phased_bilinear(const double *ref, + const double *phases, + const unsigned char *flags, + const SymOpList *sym, + float hd, float kd, + signed int l) +{ + signed int k; + double complex val1, val2; + float f; + + k = (signed int)kd; + if ( kd < 0.0 ) k -= 1; + f = kd - (float)k; + assert(f >= 0.0); + + val1 = interpolate_phased_linear(ref, phases, flags, sym, hd, k, l); + val2 = interpolate_phased_linear(ref, phases, flags, sym, hd, k+1, l); + + return (1.0-f)*val1 + f*val2; +} + + +static double interpolate_phased_intensity(const double *ref, + const double *phases, + const unsigned char *flags, + const SymOpList *sym, + float hd, float kd, float ld) +{ + signed int l; + double complex val1, val2; + float f; + + l = (signed int)ld; + if ( ld < 0.0 ) l -= 1; + f = ld - (float)l; + assert(f >= 0.0); + + val1 = interpolate_phased_bilinear(ref, phases, flags, sym, + hd, kd, l); + val2 = interpolate_phased_bilinear(ref, phases, flags, sym, + hd, kd, l+1); + + return cabs((1.0-f)*val1 + f*val2); +} + + +/* Look up the structure factor for the nearest Bragg condition */ +static double molecule_factor(const double *intensities, const double *phases, + const unsigned char *flags, struct rvec q, + double ax, double ay, double az, + double bx, double by, double bz, + double cx, double cy, double cz, + GradientMethod m, const SymOpList *sym) +{ + float hd, kd, ld; + signed int h, k, l; + double r; + + hd = q.u * ax + q.v * ay + q.w * az; + kd = q.u * bx + q.v * by + q.w * bz; + ld = q.u * cx + q.v * cy + q.w * cz; + + /* No flags -> flat intensity distribution */ + if ( flags == NULL ) return 1.0e5; + + switch ( m ) { + case GRADIENT_MOSAIC : + fesetround(1); /* Round to nearest */ + h = (signed int)rint(hd); + k = (signed int)rint(kd); + l = (signed int)rint(ld); + if ( abs(h) > INDMAX ) r = 0.0; + else if ( abs(k) > INDMAX ) r = 0.0; + else if ( abs(l) > INDMAX ) r = 0.0; + else r = sym_lookup_intensity(intensities, flags, sym, h, k, l); + break; + case GRADIENT_INTERPOLATE : + r = interpolate_intensity(intensities, flags, sym, hd, kd, ld); + break; + case GRADIENT_PHASED : + r = interpolate_phased_intensity(intensities, phases, flags, + sym, hd, kd, ld); + break; + default: + ERROR("This gradient method not implemented yet.\n"); + exit(1); + } + + return r; +} + + +void get_diffraction(struct image *image, int na, int nb, int nc, + const double *intensities, const double *phases, + const unsigned char *flags, UnitCell *cell, + GradientMethod m, const SymOpList *sym) +{ + unsigned int fs, ss; + double ax, ay, az; + double bx, by, bz; + double cx, cy, cz; + float klow, khigh, bwstep; + double *lut_a; + double *lut_b; + double *lut_c; + double divxlow, divylow, divxstep, divystep; + + cell_get_cartesian(cell, &ax, &ay, &az, &bx, &by, &bz, &cx, &cy, &cz); + + /* Allocate (and zero) the "diffraction array" */ + image->data = calloc(image->width * image->height, sizeof(float)); + + /* Needed later for Lorentz calculation */ + image->twotheta = malloc(image->width * image->height * sizeof(double)); + + klow = 1.0/(image->lambda*(1.0 + image->beam->bandwidth/2.0)); + khigh = 1.0/(image->lambda*(1.0 - image->beam->bandwidth/2.0)); + bwstep = (khigh-klow) / BWSAMPLING; + + divxlow = -image->beam->divergence/2.0; + divylow = -image->beam->divergence/2.0; + divxstep = image->beam->divergence / DIVSAMPLING; + divystep = image->beam->divergence / DIVSAMPLING; + + lut_a = get_sinc_lut(na); + lut_b = get_sinc_lut(nb); + lut_c = get_sinc_lut(nc); + + for ( fs=0; fswidth; fs++ ) { + for ( ss=0; ssheight; ss++ ) { + + int fs_step, ss_step, kstep; + int divxval, divyval; + int idx = fs + image->width*ss; + + for ( fs_step=0; fs_stepdata[idx] += intensity; + + if ( fs_step + ss_step + kstep == 0 ) { + image->twotheta[idx] = twotheta; + } + + } + } + } + } + } + + image->data[idx] /= (SAMPLING*SAMPLING*BWSAMPLING + *DIVSAMPLING*DIVSAMPLING); + + + } + progress_bar(fs, image->width-1, "Calculating diffraction"); + } + + free(lut_a); + free(lut_b); + free(lut_c); +} -- cgit v1.2.3