/**
* Function to read the input shape file.
*/
unsigned int NumericFormFactor::read_shapes_file_dat(const char* filename, real_vec_t &shape_def) {
std::ifstream f(filename);
if(!f.is_open()) {
std::cout << "Cannot open file " << filename << std::endl;
return 1;
} // if
real_t s = 0.0, cx = 0.0, cy = 0.0, cz = 0.0, nx = 0.0, ny = 0.0, nz = 0.0;
while(true) {
f >> s;
if(f.eof() || !f.good()) break;
f >> nx; f >> ny; f >> nz;
f >> cx; f >> cy; f >> cz;
shape_def.push_back(s);
shape_def.push_back(nx);
shape_def.push_back(ny);
shape_def.push_back(nz);
shape_def.push_back(cx);
shape_def.push_back(cy);
shape_def.push_back(cz);
} // while
f.close();
return shape_def.size() / 7;
} // NumericFormFactor::read_shapes_file_dat()
/**
* Function to read the shape definition input file in HDF5 format.
*/
unsigned int NumericFormFactor::read_shapes_file(const char* filename,
// #ifndef __SSE3__
real_vec_t &shape_def
// #else
// #ifdef USE_GPU
// real_vec_t &shape_def
// #else
// real_t* &shape_def
// #endif
// #endif
) {
unsigned int num_triangles = 0;
double* temp_shape_def = NULL;
// TODO: shape definition is already in HigInput ...
// utilize ...
ShapeFileType type = get_shapes_file_format(filename);
if(type == shape_file_data) {
RawShapeReader temp(filename, temp_shape_def, num_triangles);
} else if(type == shape_file_object) {
ObjectShapeReader temp(filename, temp_shape_def, num_triangles);
} else if(type == shape_file_hdf5) {
#ifdef USE_PARALLEL_HDF5
h5_shape_reader(filename, &temp_shape_def, &num_triangles);
#else
std::cerr << "error: use of parallel hdf5 format has not been enabled in your installation. "
<< "Please reinstall with the support enabled." << std::endl;
return false;
#endif
} else if(type == shape_file_null) {
std::cerr << "error: shape definition file extension is null" << std::endl;
return 0;
} else if(type == shape_file_error) {
std::cerr << "error: shape definition file format unknown" << std::endl;
return 0;
} else {
std::cerr << "error: shape definition file format unknown" << std::endl;
return 0;
} // if-else
#ifdef FF_NUM_GPU
#ifndef KERNEL2
for(unsigned int i = 0; i < num_triangles * 7; ++ i)
shape_def.push_back((real_t)temp_shape_def[i]);
#else // KERNEL2
for(unsigned int i = 0, j = 0; i < num_triangles * T_PROP_SIZE_; ++ i) {
if((i + 1) % T_PROP_SIZE_ == 0) shape_def.push_back((real_t) 0.0); // padding
else { shape_def.push_back((real_t)temp_shape_def[j]); ++ j; }
} // for
#endif // KERNEL2
//#elif defined USE_MIC // using MIC
// for(unsigned int i = 0; i < num_triangles * 7; ++ i)
// shape_def.push_back((real_t)temp_shape_def[i]);
#else // using CPU or MIC
// #ifndef __SSE3__
for(unsigned int i = 0, j = 0; i < num_triangles * CPU_T_PROP_SIZE_; ++ i) {
if((i + 1) % CPU_T_PROP_SIZE_ == 0) shape_def.push_back((real_t) 0.0); // padding
else { shape_def.push_back((real_t)temp_shape_def[j]); ++ j; }
} // for
/* #else // using SSE3, so store data differently: FOR CPU AND MIC (vectorization)
#ifndef USE_MIC // generic cpu version with SSE3 or AVX
#ifdef INTEL_SB_AVX // CPU version with AVX
// group all 's', 'nx', 'ny', 'nz', 'x', 'y', 'z' together
// for alignment at 32 bytes, make sure each of the 7 groups is padded
// compute amount of padding
// 32 bytes = 8 floats or 4 doubles. FIXME: assuming float only for now ...
unsigned int padding = (8 - (num_triangles & 7)) & 7;
unsigned int shape_size = (num_triangles + padding) * 7;
shape_def = (real_t*) _mm_malloc(shape_size * sizeof(real_t), 32);
if(shape_def == NULL) {
std::cerr << "error: failed to allocate aligned memory for shape_def"
<< std::endl;
return 0;
} // if
memset(shape_def, 0, shape_size * sizeof(real_t));
for(int i = 0; i < num_triangles; ++ i) {
for(int j = 0; j < 7; ++ j) {
shape_def[(num_triangles + padding) * j + i] = temp_shape_def[7 * i + j];
} // for
} // for
#else // CPU version with SSE3
// group all 's', 'nx', 'ny', 'nz', 'x', 'y', 'z' together
// for alignment at 16 bytes, make sure each of the 7 groups is padded
// compute amount of padding
// 16 bytes = 4 floats or 2 doubles. FIXME: assuming float only for now ...
unsigned int padding = (4 - (num_triangles & 3)) & 3;
unsigned int shape_size = (num_triangles + padding) * 7;
shape_def = (real_t*) _mm_malloc(shape_size * sizeof(real_t), 16);
if(shape_def == NULL) {
std::cerr << "error: failed to allocate aligned memory for shape_def"
<< std::endl;
return 0;
} // if
memset(shape_def, 0, shape_size * sizeof(real_t));
for(int i = 0; i < num_triangles; ++ i) {
for(int j = 0; j < 7; ++ j) {
shape_def[(num_triangles + padding) * j + i] = temp_shape_def[7 * i + j];
} // for
} // for
#endif
#else // optimized for MIC only: AVX2, 64 byte alignments (512-bit vector registers)
// FIXME: float only for now: 16 floats in one vector!
unsigned int padding = (16 - (num_triangles & 15)) & 15;
unsigned int shape_size = (num_triangles + padding) * 7;
shape_def = (real_t*) _mm_malloc(shape_size * sizeof(real_t), 64);
if(shape_def == NULL) {
std::cerr << "error: failed to allocate aligned memory for shape_def"
<< std::endl;
return 0;
} // if
memset(shape_def, 0, shape_size * sizeof(real_t));
for(int i = 0; i < num_triangles; ++ i) {
for(int j = 0; j < 7; ++ j) {
shape_def[(num_triangles + padding) * j + i] = temp_shape_def[7 * i + j];
} // for
} // for
// TODO: try grouping 16 triangles together ...
// that will give completely sequential memory access!
#endif
#endif // __SSE3__ */
#endif // FF_NUM_GPU
return num_triangles;
} // NumericFormFactor::read_shapes_file()
