//------------------------------------------------------------------------------------ // DSt_cube constructor. Example creation: DSt_cube *integer_cube = new DSt_cube(nwZ, nwY, nwX); // // Description: Creates cube with dimensions (nwZ, nwY, nwX). Can use ->data[#][#][#], or datac(#,#,#) to access spots. // // Parameters: // nwZ, nwY, nwX - dimensions of cube to create // max_block_size - max contiguous bytes allowed, useful for large cubes that could run into fragmentation issues //------------------------------------------------------------------------------------ template DSt_cube::DSt_cube(unsigned short nwZ, unsigned short nwY, unsigned short nwX, mem_block max_block_size_bytes/* = DST_AUTO_SIZE_BLOCK*/) { // Fill in sample size information wZ = nwZ; wY = nwY; wX = nwX; dimensions.from_zyx(wZ, wY, wX); max_block_size = max_block_size_bytes; cache_state = CACHE_DISABLED; // Allocate cube allocateCube(nwZ, nwY, nwX); }; //------------------------------------------------------------------------------------ // DSt_cube using file cache constructor. // // Description: Creates cube with dimensions (nwZ, nwY, nwX), but uses disc to store most of the data. // A cache in RAM of size (cache_wZ, cache_wY, cache_wX) will be automatically used and kept up to date. // MUST use datac(#,#,#) to access spots. Slower, but allows much larger cubes. // // NOTE: There are three levels of performance: // 1) cache_wY and cache_wX are size of nwY and nwX. Fastest in file access, but uses most memory. // 2) cache_wX is same size as nwX. Middle ground for performance. // 3) None of the sizes match cache, this can result in any size cache cube, but slow disc access. // Cache type should be based on algorithm to be used. If cube will be traveled in a linear fashion, option (1) is // probably best. If cube is traveled in random directions, or equally in all, then (3) is probably best. // // Parameters: // nwZ, nwY, nwX - dimensions of cube to create // max_block_size - max contiguous bytes allowed, useful for large cubes that could run into fragmentation issues //------------------------------------------------------------------------------------ template DSt_cube::DSt_cube( unsigned short nwZ, unsigned short nwY, unsigned short nwX, cache_type cache_config, UINT64 desired_cache_size_bytes /*= 100000000*/, unsigned char sub_row_size_2pow/*=7*/ ) { // Fill in sample size and cache size information wZ = nwZ; wY = nwY; wX = nwX; dimensions.from_zyx(wZ, wY, wX); data = NULL; // Not used, set NULL to prevent deallocation at destruction // Disable multiple blocks for simplicity (might want to change this in the future) max_block_size = DST_SINGLE_BLOCK; cache_is_full = false; cache_sub_row_shift_val = sub_row_size_2pow; // Hard code for now at 128 size subrows #if CACHING_IS_ENABLED if ( ((UINT64)nwZ*nwY*nwX) < (desired_cache_size_bytes/sizeof(T)) ) { cache_state = CACHE_DISABLED; } else { cache_state = cache_config; // If subrow size is bigger than wX, might as well ust FULL_ROW mode if ( (cache_state == CACHE_SUB_ROW) && (wX < (1 << cache_sub_row_shift_val)) ) { cache_state = CACHE_FULL_ROW; } } #else cache_state = CACHE_DISABLED; #endif // Determine what type of cache to use, depending on how well sample sizes match // requested cache size. switch ( cache_state ) { case CACHE_DISABLED: max_block_size = DST_AUTO_SIZE_BLOCK; // Allocate cube normally and RETURN allocateCube(nwZ, nwY, nwX); return; case CACHE_FULL_SLICE: cache_validity_table_dim.from_zyx( wZ, 1, 1 ); cache_entry_size = wX*wY; cache_entry_dimensions.from_zyx( 1, wY, wX ); if ( allocateCacheMemory(desired_cache_size_bytes) != DST_NO_ERROR ) { write_to_log("DSt_cube::DSt_cube - FATAL ERROR - Out of memory while creating cache validity matrix CACHE_FULL_SLICE: %d, %d, %d.", wZ, 1, 1); exit(1); } break; case CACHE_FULL_ROW: cache_validity_table_dim.from_zyx( wZ, wY, 1 ); cache_entry_size = wX; cache_entry_dimensions.from_zyx( 1, 1, wX ); if ( allocateCacheMemory(desired_cache_size_bytes) != DST_NO_ERROR ) { write_to_log("DSt_cube::DSt_cube - FATAL ERROR - Out of memory while creating cache validity matrix CACHE_FULL_ROW: %d, %d, %d.", wZ, wY, 1); exit(1); } break; case CACHE_SUB_ROW: cache_sub_row_last_seg_indx = (unsigned short)(wX >> cache_sub_row_shift_val); cache_validity_table_dim.from_zyx( wZ, wY, cache_sub_row_last_seg_indx + 1 ); cache_entry_size = (unsigned short)(1 << cache_sub_row_shift_val); cache_entry_dimensions.from_zyx( 1, 1, cache_entry_size ); cache_sub_row_last_entry_size = (unsigned short)( (~(0xFFFF << cache_sub_row_shift_val)) & wX ); if ( allocateCacheMemory(desired_cache_size_bytes) != DST_NO_ERROR ) { write_to_log("DSt_cube::DSt_cube - FATAL ERROR - Out of memory while creating cache validity matrix CACHE_SUB_ROW: %d, %d, %d.", wZ, wY, cache_validity_table_dim.x); exit(1); } break; default: write_to_log("DSt_cube::DSt_cube - FATAL ERROR - Invalid Cache type requested %d.", cache_state); exit(1); } // Create and open the cache file ++dst_cache_file_num; sprintf( cache_file_name, "poromedia_cache_file_%d.cache", dst_cache_file_num ); // Create the file with temporary and random access flags, system will delete on close cache_file = CreateFile( cache_file_name, GENERIC_READ | GENERIC_WRITE, FILE_SHARE_READ, NULL, CREATE_NEW, /*FILE_FLAG_SEQUENTIAL_SCAN |*/ FILE_FLAG_DELETE_ON_CLOSE/* | FILE_FLAG_RANDOM_ACCESS*/, NULL ); if( cache_file == INVALID_HANDLE_VALUE ) { write_to_log("DSt_cube::DSt_cube - FATAL ERROR - Cannot create a cache file, %s", cache_file_name); exit(1); // Fatal error } // Allocate entire file size in advance LARGE_INTEGER file_size; DWORD bytes_written; file_size.QuadPart = (LONGLONG) wZ*wY*wX*sizeof(T); SetFilePointerEx( cache_file, file_size, NULL, FILE_BEGIN ); T single_item = (T) 0; WriteFile( cache_file, &single_item, sizeof(T), &bytes_written, NULL ); FlushFileBuffers( cache_file ); }; //------------------------------------------------------------------------------------ // allocateCube // // Description: Allocates memory for 'data' cube of desired size //------------------------------------------------------------------------------------ template dst_status DSt_cube::allocateCube(unsigned short nwZ, unsigned short nwY, unsigned short nwX) { register unsigned short int y, z; // Check if we need to break up into multiple smaller block sizes. If auto_size is specified, and total cube // size is less than 512 MB, then we will not break up into multiple blocks. if ( (max_block_size == DST_AUTO_SIZE_BLOCK) && (((unsigned int)nwZ*nwY*nwX*sizeof(T)) < 512000000) ) { memory_num_of_blocks = 1; slices_per_block = nwZ; slices_per_block_last = nwZ; } else { // Determine how many slices can fit within block memory limit (max block size divided by memory required per slice) unsigned int block_size_in_bytes = (max_block_size == DST_AUTO_SIZE_BLOCK) ? (DST_SMALL_BLOCK) : (max_block_size); slices_per_block = (unsigned short)(block_size_in_bytes / (sizeof(T)*nwX*nwY) ); // Last block needs to be initialized to same size as original, in case of perfect fit slices_per_block_last = slices_per_block; // Must be at least one slice per block if ( slices_per_block == 0 ) { slices_per_block = 1; } // Number of blocks needed will be Ceiling of total number of slices divided by slices per block memory_num_of_blocks = nwZ/slices_per_block; if ( (nwZ%slices_per_block) > 0 ) { memory_num_of_blocks = memory_num_of_blocks + 1; } if ( memory_num_of_blocks == 1 ) { slices_per_block = wZ; } } // Create array of pointers for start of each block, set to null memory_block_ptrs = new T * [memory_num_of_blocks]; memset( memory_block_ptrs, 0, memory_num_of_blocks*sizeof(T) ); data = new T ** [nwZ]; rows = new T * [nwZ*nwY]; // Allocate memory for each block for ( z = 0, y = 0; z < memory_num_of_blocks; z++) { if ( (y + slices_per_block) > nwZ ) { // Allocate fewer slices if we are at last block of imperfect block divisions slices_per_block_last = nwZ - y; memory_block_ptrs[z] = new T [slices_per_block_last * nwX * nwY]; break; } else { memory_block_ptrs[z] = new T [slices_per_block * nwX * nwY]; y = y + slices_per_block; } } // Make sure memory allocation is successful if (!data || !rows || !memory_block_ptrs) { write_to_log("DSt_cube::allocateCube - Out of memory while creating cube of dimensions %dx%dx%d.", nwX, nwY, nwZ); data = NULL; return DST_ERROR_MEMORY_ALLOCATION; } unsigned short cur_block = 0; unsigned short cur_block_slice = 0; // Make pointers point into correct places for (z = 0; z < nwZ; z++) { for (y = 0; y < nwY; y++) { // Link row point into memory rows rows[z*nwY + y] = &memory_block_ptrs[cur_block][cur_block_slice*nwY*nwX + y*nwX]; } data[z] = &rows[z*nwY]; // Link final data pointers into rows ++cur_block_slice; // Check to see if we should move on to the next block if ( cur_block_slice == slices_per_block ) { ++cur_block; cur_block_slice = 0; } } return DST_NO_ERROR; }; //------------------------------------------------------------------------------------ // allocateCube // // Description: Allocates all memory needed for cache'ing //------------------------------------------------------------------------------------ template dst_status DSt_cube::allocateCacheMemory( UINT64 desired_size ) { // Allocate row validity 2d matrix. cache_validity_table_mem = new cache_entry_info* [cache_validity_table_dim.z * cache_validity_table_dim.y * cache_validity_table_dim.x]; cache_validity_table_rows = new cache_entry_info** [cache_validity_table_dim.z*cache_validity_table_dim.y]; cache_validity_table = new cache_entry_info*** [cache_validity_table_dim.z]; // Make sure memory allocation is successful if ( (cache_validity_table_mem == NULL) || (cache_validity_table == NULL) ) { return DST_ERROR_MEMORY_ALLOCATION; } // Make pointers point into correct places for (int z = 0; z < cache_validity_table_dim.z; z++) { for (int y = 0; y < cache_validity_table_dim.y; y++) { // Link row point into memory rows cache_validity_table_rows[z*cache_validity_table_dim.y + y] = &cache_validity_table_mem[z*cache_validity_table_dim.y*cache_validity_table_dim.x + y*cache_validity_table_dim.x]; } cache_validity_table[z] = &cache_validity_table_rows[ z*cache_validity_table_dim.y ]; } // Now allocate buffer memory return allocateBufferMemory( desired_size ); } //------------------------------------------------------------------------------------ // allocateCube // // Description: Allocates actual buffer memory needed for cache'ing //------------------------------------------------------------------------------------ template dst_status DSt_cube::allocateBufferMemory( UINT64 desired_size ) { // Need to make sure validity table is completely invalid if we are messing with buffer memset( cache_validity_table_mem, 0, sizeof(cache_entry_info*) * cache_validity_table_dim.z * cache_validity_table_dim.y * cache_validity_table_dim.x ); // Determine number of cache entries we can hold at once cache_number_of_entries = (int)( desired_size / (sizeof(T) * cache_entry_size) ); if ( cache_number_of_entries == 0 ) { cache_number_of_entries = 1; } // --- Allocate memory for each entry --- // memory_num_of_blocks = 1; memory_block_ptrs = new T* [memory_num_of_blocks]; memory_total_used = cache_number_of_entries * cache_entry_size; memory_block_ptrs[0] = new T [memory_total_used]; rows = new T * [cache_number_of_entries*cache_entry_dimensions.y]; // Make sure memory allocation is successful if ( (memory_block_ptrs == NULL) || (memory_block_ptrs == NULL) || (rows == NULL) ) { return DST_ERROR_MEMORY_ALLOCATION; } #if CACHING_USE_LINKED_LIST cache_available_indx = 0; cache_age_queue = new LinkedList( cache_number_of_entries ); // Make sure memory allocation is successful if ( cache_age_queue == NULL ) { return DST_ERROR_MEMORY_ALLOCATION; } // Link up the queue for (int i = 0; i < cache_number_of_entries; i++) { oldest_seg_ptr = cache_age_queue->getNewNode(); oldest_seg_ptr->item.needs_sync = false; oldest_seg_ptr->item.queue_node_ptr = oldest_seg_ptr; for ( int j = 0; j < cache_entry_dimensions.y; j++ ) { // Link row point into memory rows rows[ i*cache_entry_dimensions.y + j ] = &memory_block_ptrs[0][i*cache_entry_size + j*cache_entry_dimensions.x]; } oldest_seg_ptr->item.data = &( rows[i * cache_entry_dimensions.y] ); cache_age_queue->insertAtStart( oldest_seg_ptr ); } #else // Now allocate and initialize the circular array cache_oldest_indx = 0; cache_available_indx = 0; cache_age_list = new cache_entry_info [cache_number_of_entries]; // Make sure memory allocation is successful if ( cache_age_list == NULL ) { return DST_ERROR_MEMORY_ALLOCATION; } for (int i = 0; i < cache_number_of_entries; i++) { cache_age_list[i].needs_sync = false; for ( int j = 0; j < cache_entry_dimensions.y; j++ ) { // Link row point into memory rows rows[ i*cache_entry_dimensions.y + j ] = &memory_block_ptrs[0][i*cache_entry_size + j*cache_entry_dimensions.x]; } cache_age_list[i].data = &( rows[i * cache_entry_dimensions.y] ); } #endif return DST_NO_ERROR; } //------------------------------------------------------------------------------------ // DSt_cube::~DSt_cube // deletes all memory belonging to object //------------------------------------------------------------------------------------ template DSt_cube::~DSt_cube(void) { if (data) delete[] data; if (rows) delete[] rows; if (memory_block_ptrs) { // Delete each individual block for (unsigned short i = 0; i < memory_num_of_blocks; i++) { if ( memory_block_ptrs[i] ) delete[] memory_block_ptrs[i]; } // Delete array delete[] memory_block_ptrs; } // Close and delete cache related stuff if used if ( cache_state != CACHE_DISABLED ) { // Close the cache file, system should delete it automatically CloseHandle( cache_file ); // Cache memory delete[] cache_validity_table; delete[] cache_validity_table_rows; delete[] cache_validity_table_mem; #if CACHING_USE_LINKED_LIST delete cache_age_queue; #else delete[] cache_age_list; #endif } };