Skip to content

Commit

Permalink
Remove inlines
Browse files Browse the repository at this point in the history
  • Loading branch information
@villekf
villekf committed Oct 29, 2020
1 parent 8d7cd14 commit ff3cfeb
Show file tree
Hide file tree
Showing 2 changed files with 44 additions and 44 deletions.
70 changes: 35 additions & 35 deletions source/general_opencl_functions.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,7 +43,7 @@
// This function was taken from: https://streamhpc.com/blog/2016-02-09/atomic-operations-for-floats-in-opencl-improved/
// Computes the atomic_add for floats
#ifndef ATOMIC
inline void atomicAdd_g_f(volatile __global float *addr, float val) {
void atomicAdd_g_f(volatile __global float *addr, float val) {
union {
unsigned int u32;
float f32;
Expand All @@ -69,7 +69,7 @@ typedef struct _RecMethodsOpenCL {

#ifndef MBSREM
// Denominator (forward projection)
inline void denominator(float local_ele, float* ax, uint local_ind, const uint d_N, const __global float* d_OSEM) {
void denominator(float local_ele, float* ax, uint local_ind, const uint d_N, const __global float* d_OSEM) {
#ifdef NREKOS1
ax[0] += (local_ele * d_OSEM[local_ind]);
#elif defined(NREKOS2)
Expand All @@ -85,7 +85,7 @@ inline void denominator(float local_ele, float* ax, uint local_ind, const uint d
}

// Nominator (backprojection) in MLEM
inline void nominator(__constant uchar* MethodList, float* ax, const float d_Sino, const float d_epsilon_mramla, const float d_epps,
void nominator(__constant uchar* MethodList, float* ax, const float d_Sino, const float d_epsilon_mramla, const float d_epps,
const float temp, const __global float* d_sc_ra, const uint idx) {
float local_rand = 0.f;
#ifdef RANDOMS
Expand Down Expand Up @@ -174,7 +174,7 @@ inline void nominator(__constant uchar* MethodList, float* ax, const float d_Sin
#endif
#ifdef MBSREM
// Nominator (backprojection), COSEM
inline void nominator_cosem(float* axCOSEM, const float local_sino, const float d_epps, const float temp, const __global float* d_sc_ra,
void nominator_cosem(float* axCOSEM, const float local_sino, const float d_epps, const float temp, const __global float* d_sc_ra,
const uint idx) {
*axCOSEM *= temp;
if (*axCOSEM < d_epps)
Expand All @@ -186,7 +186,7 @@ inline void nominator_cosem(float* axCOSEM, const float local_sino, const float
}
#endif
#ifndef MBSREM
inline void rhs(__constant uchar* MethodList, const float local_ele, const float* ax, const uint local_ind,
void rhs(__constant uchar* MethodList, const float local_ele, const float* ax, const uint local_ind,
const uint d_N, __global CAST* d_rhs_OSEM) {
#ifdef NREKOS1
#ifdef ATOMIC
Expand Down Expand Up @@ -220,7 +220,7 @@ inline void rhs(__constant uchar* MethodList, const float local_ele, const float
#else

// Nominator (backprojection), multi-GPU version
inline void nominator_multi(float* axOSEM, const float d_Sino, const float d_epps, const float temp, const __global float* d_sc_ra,
void nominator_multi(float* axOSEM, const float d_Sino, const float d_epps, const float temp, const __global float* d_sc_ra,
const uint idx) {
*axOSEM *= temp;
#ifdef BP
Expand All @@ -238,14 +238,14 @@ inline void nominator_multi(float* axOSEM, const float d_Sino, const float d_epp

#if defined(MBSREM) || !defined(AF)
// Denominator (forward projection), multi-GPU version
inline void denominator_multi(const float local_ele, float* axOSEM, const __global float* d_OSEM) {
void denominator_multi(const float local_ele, float* axOSEM, const __global float* d_OSEM) {
*axOSEM += (local_ele * *d_OSEM);
}
#endif

#if defined(RAW) && !defined(N_RAYS)
// Get the detector coordinates for the current (raw) measurement
inline void get_detector_coordinates_raw(const __global float *d_x, const __global float *d_y, const __global float *d_zdet, const __global ushort* d_L,
void get_detector_coordinates_raw(const __global float *d_x, const __global float *d_y, const __global float *d_zdet, const __global ushort* d_L,
const uint d_det_per_ring, const uint idx, __constant uint *d_pseudos, const uint d_pRows, float *xs, float* xd, float* ys, float* yd, float* zs,
float* zd) {
// Get the current detector numbers
Expand Down Expand Up @@ -275,7 +275,7 @@ inline void get_detector_coordinates_raw(const __global float *d_x, const __glob

#if defined(N_RAYS) && defined(RAW)
// Get the detector coordinates for the current (raw) measurement
inline void get_detector_coordinates_raw_multiray(const __global float* d_x, const __global float* d_y, const __global float* d_zdet, const __global ushort* d_L,
void get_detector_coordinates_raw_multiray(const __global float* d_x, const __global float* d_y, const __global float* d_zdet, const __global ushort* d_L,
const uint d_det_per_ring, const uint idx, __constant uint* d_pseudos, const uint d_pRows, float* xs, float* xd, float* ys, float* yd, float* zs, float* zd,
const ushort lor, const float cr_pz) {
uint ps = 0;
Expand Down Expand Up @@ -330,7 +330,7 @@ inline void get_detector_coordinates_raw_multiray(const __global float* d_x, con

#if !defined(N_RAYS) && !defined(RAW)
// Get the detector coordinates for the current sinogram bin
inline void get_detector_coordinates(const __global uint *d_xyindex, const __global ushort *d_zindex, const uint d_size_x, const uint idx,
void get_detector_coordinates(const __global uint *d_xyindex, const __global ushort *d_zindex, const uint d_size_x, const uint idx,
const ushort d_TotSinos, float *xs, float* xd, float* ys, float* yd, float* zs, float* zd, const __global float *d_x, const __global float *d_y,
const __global float *d_zdet) {

Expand Down Expand Up @@ -359,7 +359,7 @@ inline void get_detector_coordinates(const __global uint *d_xyindex, const __glo

#if defined(N_RAYS) && !defined(RAW)
// Get the detector coordinates for the current sinogram bin
inline void get_detector_coordinates_multiray(const __global uint* d_xyindex, const __global ushort* d_zindex, const uint d_size_x, const uint idx,
void get_detector_coordinates_multiray(const __global uint* d_xyindex, const __global ushort* d_zindex, const uint d_size_x, const uint idx,
const ushort d_TotSinos, float* xs, float* xd, float* ys, float* yd, float* zs, float* zd, const __global float* d_x, const __global float* d_y,
const __global float* d_zdet, const ushort lor, const float cr_pz) {

Expand Down Expand Up @@ -424,7 +424,7 @@ inline void get_detector_coordinates_multiray(const __global uint* d_xyindex, co

#ifdef FIND_LORS
// Get the detector coordinates for the current measurement (precomputation phase)
inline void get_detector_coordinates_precomp(const uint d_size_x, const uint idx, const ushort d_TotSinos, float *xs, float* xd, float* ys, float* yd, float* zs,
void get_detector_coordinates_precomp(const uint d_size_x, const uint idx, const ushort d_TotSinos, float *xs, float* xd, float* ys, float* yd, float* zs,
float* zd, const __global float *d_x, const __global float *d_y, const __global float *d_zdet) {

const uint id = idx % d_size_x;
Expand All @@ -439,7 +439,7 @@ inline void get_detector_coordinates_precomp(const uint d_size_x, const uint idx
#endif

// Compute the voxel index where the current perpendicular measurement starts
inline int perpendicular_start(const float d_b, const float d, const float d_d, const uint d_N) {
int perpendicular_start(const float d_b, const float d, const float d_d, const uint d_N) {
int tempi = 0;
float start = d_b - d + d_d;
for (uint ii = 0u; ii < d_N; ii++) {
Expand All @@ -453,7 +453,7 @@ inline int perpendicular_start(const float d_b, const float d, const float d_d,
}

// Compute the probability for the perpendicular elements
inline void perpendicular_elements(const float d_b, const float d_d1, const uint d_N1, const float d, const float d_d2, const uint d_N2,
void perpendicular_elements(const float d_b, const float d_d1, const uint d_N1, const float d, const float d_d2, const uint d_N2,
const __global float* d_atten, float* templ_ijk, uint* tempk, const uint z_loop, const uint d_N, const uint d_NN,
const __global float* d_norm, const uint idx, const float global_factor, const __global float* d_scat) {
int apu = perpendicular_start(d_b, d, d_d1, d_N1);
Expand All @@ -480,7 +480,7 @@ inline void perpendicular_elements(const float d_b, const float d_d1, const uint

#ifdef N_RAYS
// Compute the probability for the perpendicular elements
inline float perpendicular_elements_multiray(const float d_b, const float d_d1, const uint d_N1, const float d, const float d_d2, const uint d_N2,
float perpendicular_elements_multiray(const float d_b, const float d_d1, const uint d_N1, const float d, const float d_d2, const uint d_N2,
const __global float* d_atten, uint* tempk, const uint z_loop, const uint d_N, const uint d_NN, float* jelppi) {
int apu = perpendicular_start(d_b, d, d_d1, d_N1);

Expand All @@ -491,7 +491,7 @@ inline float perpendicular_elements_multiray(const float d_b, const float d_d1,
#endif

// Compute functions (9) and (29) (detector larger than source)
inline void d_g_s(const float tmin, const float t_min, uint* v_min, float* t_0, int* v_u, const float diff, const float b, const float d, const float s) {
void d_g_s(const float tmin, const float t_min, uint* v_min, float* t_0, int* v_u, const float diff, const float b, const float d, const float s) {

if (tmin == t_min)
// (11)
Expand All @@ -509,7 +509,7 @@ inline void d_g_s(const float tmin, const float t_min, uint* v_min, float* t_0,
}

// Compute functions (9) and (29) (source larger than detector)
inline void s_g_d(const float tmin, const float t_min, uint* v_max, float* t_0, int* v_u, const float diff, const float b, const float d, const float s,
void s_g_d(const float tmin, const float t_min, uint* v_max, float* t_0, int* v_u, const float diff, const float b, const float d, const float s,
const uint N) {

if (tmin == t_min)
Expand All @@ -528,7 +528,7 @@ inline void s_g_d(const float tmin, const float t_min, uint* v_max, float* t_0,
}

// same as above, but for precomputation phase
inline void d_g_s_precomp(const float tmin, const float t_min, const float tmax, const float t_max, uint* v_min, uint* v_max, float* t_0, int* v_u,
void d_g_s_precomp(const float tmin, const float t_min, const float tmax, const float t_max, uint* v_min, uint* v_max, float* t_0, int* v_u,
const float diff, const float b, const float d, const float s, const uint N) {

if (tmin == t_min)
Expand Down Expand Up @@ -556,7 +556,7 @@ inline void d_g_s_precomp(const float tmin, const float t_min, const float tmax,
}

// same as above, but for precomputation phase
inline void s_g_d_precomp(const float tmin, const float t_min, const float tmax, const float t_max, uint* v_min, uint* v_max, float* t_0, int* v_u,
void s_g_d_precomp(const float tmin, const float t_min, const float tmax, const float t_max, uint* v_min, uint* v_max, float* t_0, int* v_u,
const float diff, const float b, const float d, const float s, const uint N) {

if (tmin == t_min)
Expand Down Expand Up @@ -584,7 +584,7 @@ inline void s_g_d_precomp(const float tmin, const float t_min, const float tmax,
}

// Compute the index of the current voxel
inline uint compute_ind(const int tempj, const int tempi, const int tempk, const uint d_N1, const uint d_N2, const uint d_N, const uint d_Nx,
uint compute_ind(const int tempj, const int tempi, const int tempk, const uint d_N1, const uint d_N2, const uint d_N, const uint d_Nx,
const uint d_Nyx) {
uint local_ind = convert_uint_sat(tempj) * d_Nx + convert_uint_sat(tempi) + convert_uint_sat(tempk) * d_Nyx;
#ifndef PRECOMPUTE
Expand All @@ -601,11 +601,11 @@ inline uint compute_ind(const int tempj, const int tempi, const int tempk, const
}

#ifdef ATN
inline float compute_matrix_element(const float t0, const float tc, const float L) {
float compute_matrix_element(const float t0, const float tc, const float L) {
return (t0 - tc) * L;
}

inline void compute_attenuation(float* tc, float* jelppi, const float LL, const float t0, const int tempi, const int tempj, const int tempk, const uint Nx,
void compute_attenuation(float* tc, float* jelppi, const float LL, const float t0, const int tempi, const int tempj, const int tempk, const uint Nx,
const uint Nyx, const __global float* d_atten) {
*jelppi += (compute_matrix_element(t0, *tc, LL) * -d_atten[tempi + tempj * Nx + Nyx * tempk]);
*tc = t0;
Expand All @@ -614,7 +614,7 @@ inline void compute_attenuation(float* tc, float* jelppi, const float LL, const

#ifdef SIDDON
// compute the distance that the ray traverses in the current voxel
inline float compute_element(float* t0, float* tc, const float L, const float tu, const int u, int* temp_ijk, float* temp) {
float compute_element(float* t0, float* tc, const float L, const float tu, const int u, int* temp_ijk, float* temp) {
float local_ele = (*t0 - *tc) * L;
*temp_ijk += u;
*tc = *t0;
Expand All @@ -624,7 +624,7 @@ inline float compute_element(float* t0, float* tc, const float L, const float tu
}

// compute the probability of emission in the current voxel
inline float compute_element_2nd(float* t0, float* tc, const float L, const float tu, const int u, int* temp_ijk, const float temp) {
float compute_element_2nd(float* t0, float* tc, const float L, const float tu, const int u, int* temp_ijk, const float temp) {
float local_ele = (*t0 - *tc) * L * temp;
*temp_ijk += u;
*tc = *t0;
Expand All @@ -634,11 +634,11 @@ inline float compute_element_2nd(float* t0, float* tc, const float L, const floa
#endif

// compute the initial voxel index (beginning of the ray)
inline int voxel_index(const float pt, const float diff, const float d, const float apu) {
int voxel_index(const float pt, const float diff, const float d, const float apu) {
return convert_int_rtz((pt * diff - apu) / d);
}

inline bool siddon_pre_loop_2D(const float b1, const float b2, const float diff1, const float diff2, const float max1, const float max2,
bool siddon_pre_loop_2D(const float b1, const float b2, const float diff1, const float diff2, const float max1, const float max2,
const float d1, const float d2, const uint N1, const uint N2, int* temp1, int* temp2, float* t1u, float* t2u, uint* Np,
const int TYYPPI, const float ys, const float xs, const float yd, const float xd, float* tc, int* u1, int* u2, float* t10, float* t20) {
// If neither x- nor y-directions are perpendicular
Expand Down Expand Up @@ -720,7 +720,7 @@ inline bool siddon_pre_loop_2D(const float b1, const float b2, const float diff1
return false;
}

inline bool siddon_pre_loop_3D(const float bx, const float by, const float bz, const float x_diff, const float y_diff, const float z_diff,
bool siddon_pre_loop_3D(const float bx, const float by, const float bz, const float x_diff, const float y_diff, const float z_diff,
const float maxxx, const float maxyy, const float bzb, const float dx, const float dy, const float dz,
const uint Nx, const uint Ny, const uint Nz, int* tempi, int* tempj, int* tempk, float* tyu, float* txu, float* tzu,
uint* Np, const int TYYPPI, const float ys, const float xs, const float yd, const float xd, const float zs, const float zd, float* tc,
Expand Down Expand Up @@ -808,27 +808,27 @@ inline bool siddon_pre_loop_3D(const float bx, const float by, const float bz, c
#ifdef TOF
#define _2PI 0.3989423f

inline float normPDF(const float x, const float mu, const float sigma) {
float normPDF(const float x, const float mu, const float sigma) {

const float a = (x - mu) / sigma;

return _2PI / sigma * native_exp(-0.5f * a * a);
}

inline void TOFDis(const float x_diff, const float y_diff, const float z_diff, const float tc, const float LL, float* D, float* DD) {
void TOFDis(const float x_diff, const float y_diff, const float z_diff, const float tc, const float LL, float* D, float* DD) {
const float xI = x_diff * tc;
const float yI = y_diff * tc;
const float zI = z_diff * tc;
*D = native_sqrt(xI * xI + yI * yI + zI * zI) - LL / 2.f;
*DD = *D;
}

inline float TOFWeight(const float element, const float sigma_x, const float D, const float DD, const float TOFCenter, const float epps) {
float TOFWeight(const float element, const float sigma_x, const float D, const float DD, const float TOFCenter, const float epps) {
return (element * (normPDF(D, TOFCenter, sigma_x) + normPDF(D - element * sign(DD), TOFCenter, sigma_x)) / 2.f) + epps;
}


inline float TOFLoop(const float DD, const float element, __private float* TOFVal, __constant float* TOFCenter,
float TOFLoop(const float DD, const float element, __private float* TOFVal, __constant float* TOFCenter,
const float sigma_x, float* D, const uint tid, const float epps) {
float TOFSum = 0.f;
#ifdef DEC
Expand Down Expand Up @@ -858,7 +858,7 @@ inline float TOFLoop(const float DD, const float element, __private float* TOFVa
}


inline void denominatorTOF(float* ax, const float element, const __global float* d_OSEM, uint local_ind, const float TOFSum, __private float* TOFVal,
void denominatorTOF(float* ax, const float element, const __global float* d_OSEM, uint local_ind, const float TOFSum, __private float* TOFVal,
const float DD, __constant float* TOFCenter, const float sigma_x, float* D, const uint tid, const float epps, const uint d_N) {
#if defined(AF) && !defined(MBSREM)
uint ll = NBINS;
Expand Down Expand Up @@ -890,7 +890,7 @@ inline void denominatorTOF(float* ax, const float element, const __global float*


// Nominator (y for backprojection)
inline void nominatorTOF(__constant uchar* MethodList, float* ax, const __global float* d_Sino, const float d_epsilon_mramla, const float d_epps,
void nominatorTOF(__constant uchar* MethodList, float* ax, const __global float* d_Sino, const float d_epsilon_mramla, const float d_epps,
const float temp, const __global float* d_sc_ra, const uint idx, const long TOFSize, const float local_sino) {
float local_rand = 0.f;
#ifdef RANDOMS
Expand Down Expand Up @@ -933,7 +933,7 @@ inline void nominatorTOF(__constant uchar* MethodList, float* ax, const __global
}


inline void backprojectTOF(const uint local_ind, const float local_ele, const uint tid, const __private float* TOFVal, const float* yax,
void backprojectTOF(const uint local_ind, const float local_ele, const uint tid, const __private float* TOFVal, const float* yax,
__global CAST* d_Summ
#ifndef DEC
, const float temp, const float sigma_x, float* D, const float DD, __constant float* TOFCenter, const float epps, const float TOFSum
Expand Down Expand Up @@ -1026,7 +1026,7 @@ inline void backprojectTOF(const uint local_ind, const float local_ele, const ui
}


inline void sensTOF(const uint local_ind, const float local_ele, const uint tid, const __private float* TOFVal, __global CAST * d_Summ,
void sensTOF(const uint local_ind, const float local_ele, const uint tid, const __private float* TOFVal, __global CAST * d_Summ,
#ifndef DEC
const float temp, const float sigma_x, float* D, const float DD, __constant float* TOFCenter, const float epps, const float TOFSum,
#endif
Expand Down
Loading

0 comments on commit ff3cfeb

Please sign in to comment.