TEXT back end
Table of Contents
- 1. TEXT back end
- 1.1. Template for group-related structures in text back end
- 1.2. Template for general structure in text back end
- 1.3. Initialize function (constant part)
- 1.4. Deinitialize function (templated part)
- 1.5. Template for text read struct
- 1.6. Template for text flush struct
- 1.7. Template for text free memory
- 1.8. Template for has/read/write the
$group_num$
attribute - 1.9. Template for has/read/write the
$group_dset$
dataset - 1.10. RDM struct (hard-coded)
1 TEXT back end
The "file" produced by the text back end is a directory with one file per group.
When the file is open, it is locked by the current process. No other
process can read/write the same file. This guarantees that the
representation in memory is consistent with the file and avoid
re-reading the file before writing.
To lock the file, we lock the .lock
file which is present in the
directory.
The file is written when closed, or when the flush function is called.
1.1 Template for group-related structures in text back end
typedef struct $group$_s { uint64_t $group_num$; $group_dset_dtype$* $group_dset$; uint32_t rank_$group_dset$; uint32_t to_flush; uint64_t dims_$group_dset$[16]; char file_name[TREXIO_MAX_FILENAME_LENGTH]; } $group$_t;
1.2 Template for general structure in text back end
typedef struct rdm_s { uint64_t dim_one_e; uint32_t to_flush; uint32_t padding; double* one_e; char file_name[TREXIO_MAX_FILENAME_LENGTH]; char two_e_file_name[TREXIO_MAX_FILENAME_LENGTH]; } rdm_t;
typedef struct trexio_text_s { trexio_t parent ; $group$_t* $group$; rdm_t* rdm; int lock_file; } trexio_text_t;
1.3 Initialize function (constant part)
trexio_exit_code trexio_text_init (trexio_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; trexio_text_t* const f = (trexio_text_t*) file; /* Put all pointers to NULL but leave parent untouched */ memset(&(f->parent)+1,0,sizeof(trexio_text_t)-sizeof(trexio_t)); /* If directory doesn't exist, create it in write mode */ struct stat st; if (stat(file->file_name, &st) == 0 && S_ISDIR(st.st_mode)) { /* Do nothing */ } else { if (file->mode == 'r') return TREXIO_READONLY; if (mkdir(file->file_name, 0777) != 0) { return TREXIO_FAILURE; } } /* Create the lock file in the directory */ const char* lock_file_name = "/.lock"; char file_name[TREXIO_MAX_FILENAME_LENGTH]; strncpy (file_name, file->file_name, TREXIO_MAX_FILENAME_LENGTH); strncat (file_name, lock_file_name, TREXIO_MAX_FILENAME_LENGTH-strlen(lock_file_name)); if (file_name[TREXIO_MAX_FILENAME_LENGTH-1] != '\0') { return TREXIO_LOCK_ERROR; } f->lock_file = open(file_name,O_WRONLY|O_CREAT|O_TRUNC, 0644); if (f->lock_file <= 0) { return TREXIO_FAILURE; } return TREXIO_SUCCESS; }
trexio_exit_code trexio_text_lock(trexio_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; trexio_text_t* const f = (trexio_text_t*) file; struct flock fl; fl.l_type = F_WRLCK; fl.l_whence = SEEK_SET; fl.l_start = 0; fl.l_len = 0; fl.l_pid = getpid(); int rc = fcntl(f->lock_file, F_SETLKW, &fl); if (rc == -1) return TREXIO_FAILURE; return TREXIO_SUCCESS; }
trexio_exit_code trexio_text_unlock (trexio_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; trexio_text_t* const f = (trexio_text_t*) file; struct flock fl; fl.l_type = F_UNLCK; fl.l_whence = SEEK_SET; fl.l_start = 0; fl.l_len = 0; fl.l_pid = getpid(); fcntl(f->lock_file, F_SETLK, &fl); close(f->lock_file); return TREXIO_SUCCESS; }
1.4 Deinitialize function (templated part)
trexio_exit_code trexio_text_deinit (trexio_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; trexio_exit_code rc; assert (trexio_text_free_$group$( (trexio_text_t*) file) == TREXIO_SUCCESS); rc = trexio_text_free_rdm( (trexio_text_t*) file); assert (rc == TREXIO_SUCCESS); return TREXIO_SUCCESS; }
1.5 Template for text read struct
$group$_t* trexio_text_read_$group$ (trexio_text_t* const file) { if (file == NULL) return NULL; /* If the data structure exists, return it */ if (file->$group$ != NULL) { return file->$group$; } /* Allocate the data structure */ $group$_t* $group$ = MALLOC($group$_t); if ($group$ == NULL) return NULL; memset($group$,0,sizeof($group$_t)); /* Build the file name */ const char* $group$_file_name = "/$group$.txt"; strncpy ($group$->file_name, file->parent.file_name, TREXIO_MAX_FILENAME_LENGTH); strncat ($group$->file_name, $group$_file_name, TREXIO_MAX_FILENAME_LENGTH-strlen($group$_file_name)); if ($group$->file_name[TREXIO_MAX_FILENAME_LENGTH-1] != '\0') { FREE($group$); return NULL; } /* If the file exists, read it */ FILE* f = fopen($group$->file_name,"r"); if (f != NULL) { /* Find size of file to allocate the max size of the string buffer */ fseek(f, 0L, SEEK_END); size_t sz = ftell(f); fseek(f, 0L, SEEK_SET); sz = (sz < 1024) ? (1024) : (sz); char* buffer = CALLOC(sz, char); if (buffer == NULL) { fclose(f); FREE($group$); return NULL; } /* Read the dimensioning variables */ int rc; // START REPEAT GROUP_DSET rc = fscanf(f, "%1023s", buffer); if ((rc != 1) || (strcmp(buffer, "rank_$group_dset$") != 0)) { FREE(buffer); fclose(f); FREE($group$); return NULL; } rc = fscanf(f, "%u", &($group$->rank_$group_dset$)); if (rc != 1) { FREE(buffer); fclose(f); FREE($group$); return NULL; } /* workaround for the case of missing blocks in the file */ uint64_t size_$group_dset$ = 0; if ($group$->rank_$group_dset$ != 0) size_$group_dset$ = 1; for (unsigned int i=0; i<$group$->rank_$group_dset$; ++i){ unsigned int j=0; rc = fscanf(f, "%1023s %u", buffer, &j); if ((rc != 2) || (strcmp(buffer, "dims_$group_dset$") != 0) || (j!=i)) { FREE(buffer); fclose(f); FREE($group$); return NULL; } rc = fscanf(f, "%" SCNu64 "\n", &($group$->dims_$group_dset$[i])); assert(!(rc != 1)); if (rc != 1) { FREE(buffer); fclose(f); FREE($group$); return NULL; } size_$group_dset$ *= $group$->dims_$group_dset$[i]; } // END REPEAT GROUP_DSET // START REPEAT GROUP_NUM /* Read data */ rc = fscanf(f, "%1023s", buffer); assert(!((rc != 1) || (strcmp(buffer, "$group_num$") != 0))); if ((rc != 1) || (strcmp(buffer, "$group_num$") != 0)) { FREE(buffer); fclose(f); FREE($group$); return NULL; } rc = fscanf(f, "%" SCNu64 "", &($group$->$group_num$)); assert(!(rc != 1)); if (rc != 1) { FREE(buffer); fclose(f); FREE($group$); return NULL; } // END REPEAT GROUP_NUM // START REPEAT GROUP_DSET /* Allocate arrays */ $group$->$group_dset$ = CALLOC(size_$group_dset$, $group_dset_dtype$); assert (!($group$->$group_dset$ == NULL)); if ($group$->$group_dset$ == NULL) { FREE(buffer); fclose(f); FREE($group$); return NULL; } rc = fscanf(f, "%1023s", buffer); assert(!((rc != 1) || (strcmp(buffer, "$group_dset$") != 0))); if ((rc != 1) || (strcmp(buffer, "$group_dset$") != 0)) { FREE(buffer); fclose(f); FREE($group$->$group_dset$); FREE($group$); return NULL; } for (uint64_t i=0 ; i<size_$group_dset$ ; ++i) { rc = fscanf(f, "%$group_dset_std_dtype_in$", &($group$->$group_dset$[i])); assert(!(rc != 1)); if (rc != 1) { FREE(buffer); fclose(f); FREE($group$->$group_dset$); FREE($group$); return NULL; } } // END REPEAT GROUP_DSET FREE(buffer); fclose(f); f = NULL; } file->$group$ = $group$; return $group$; }
1.6 Template for text flush struct
trexio_exit_code trexio_text_flush_$group$ (trexio_text_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (file->parent.mode == 'r') return TREXIO_READONLY; $group$_t* $group$ = file->$group$; if ($group$ == NULL) return TREXIO_SUCCESS; if ($group$->to_flush == 0) return TREXIO_SUCCESS; assert (file->parent.mode == 'w'); FILE* f = fopen($group$->file_name, "w"); if (f == NULL) return TREXIO_INVALID_ARG_1; /* Write the dimensioning variables */ // START REPEAT GROUP_DSET fprintf(f, "rank_$group_dset$ %u\n", $group$->rank_$group_dset$); // workaround for the case of missing blocks in the file uint64_t size_$group_dset$ = 0; if ($group$->rank_$group_dset$ != 0) size_$group_dset$ = 1; for (unsigned int i=0; i<$group$->rank_$group_dset$; ++i){ fprintf(f, "dims_$group_dset$ %u %" PRIu64 "\n", i, $group$->dims_$group_dset$[i]); size_$group_dset$ *= $group$->dims_$group_dset$[i]; } // END REPEAT GROUP_DSET // START REPEAT GROUP_NUM fprintf(f, "$group_num$ %" PRIu64 "\n", $group$->$group_num$); // END REPEAT GROUP_NUM /* Write arrays */ // START REPEAT GROUP_DSET fprintf(f, "$group_dset$\n"); for (uint64_t i=0 ; i<size_$group_dset$ ; ++i) { fprintf(f, "%$group_dset_std_dtype_out$\n", $group$->$group_dset$[i]); } // END REPEAT GROUP_DSET fclose(f); $group$->to_flush = 0; return TREXIO_SUCCESS; }
1.7 Template for text free memory
Memory is allocated when reading. The following function frees memory.
trexio_exit_code trexio_text_free_$group$ (trexio_text_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (file->parent.mode != 'r') { trexio_exit_code rc = trexio_text_flush_$group$(file); if (rc != TREXIO_SUCCESS) return TREXIO_FAILURE; } $group$_t* $group$ = file->$group$; if ($group$ == NULL) return TREXIO_SUCCESS; // START REPEAT GROUP_DSET if ($group$->$group_dset$ != NULL) { FREE ($group$->$group_dset$); } // END REPEAT GROUP_DSET FREE ($group$); return TREXIO_SUCCESS; }
1.8 Template for has/read/write the $group_num$
attribute
trexio_exit_code trexio_text_read_$group_num$ (trexio_t* const file, uint64_t* const num) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (num == NULL) return TREXIO_INVALID_ARG_2; $group$_t* $group$ = trexio_text_read_$group$((trexio_text_t*) file); if ($group$ == NULL) return TREXIO_FAILURE; *num = $group$->$group_num$; return TREXIO_SUCCESS; }
trexio_exit_code trexio_text_write_$group_num$ (trexio_t* const file, const uint64_t num) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (file->mode == 'r') return TREXIO_READONLY; $group$_t* $group$ = trexio_text_read_$group$((trexio_text_t*) file); if ($group$ == NULL) return TREXIO_FAILURE; $group$->$group_num$ = num; $group$->to_flush = 1; return TREXIO_SUCCESS; }
trexio_exit_code trexio_text_has_$group_num$ (trexio_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; $group$_t* $group$ = trexio_text_read_$group$((trexio_text_t*) file); if ($group$ == NULL) return TREXIO_FAILURE; if ($group$->$group_num$ > 0L){ return TREXIO_SUCCESS; } else { return TREXIO_HAS_NOT; } }
1.9 Template for has/read/write the $group_dset$
dataset
The group_dset
array is assumed allocated with the appropriate size.
trexio_exit_code trexio_text_read_$group_dset$ (trexio_t* const file, $group_dset_dtype$* const $group_dset$, const uint32_t rank, const uint64_t* dims) { if (file == NULL) return TREXIO_INVALID_ARG_1; if ($group_dset$ == NULL) return TREXIO_INVALID_ARG_2; $group$_t* const $group$ = trexio_text_read_$group$((trexio_text_t*) file); if ($group$ == NULL) return TREXIO_FAILURE; if (rank != $group$->rank_$group_dset$) return TREXIO_INVALID_ARG_3; uint64_t dim_size = 1; for (unsigned int i=0; i<rank; ++i){ if (dims[i] != $group$->dims_$group_dset$[i]) return TREXIO_INVALID_ARG_4; dim_size *= dims[i]; } for (uint64_t i=0 ; i<dim_size ; ++i) { $group_dset$[i] = $group$->$group_dset$[i]; } return TREXIO_SUCCESS; }
trexio_exit_code trexio_text_write_$group_dset$ (trexio_t* const file, const $group_dset_dtype$* $group_dset$, const uint32_t rank, const uint64_t* dims) { if (file == NULL) return TREXIO_INVALID_ARG_1; if ($group_dset$ == NULL) return TREXIO_INVALID_ARG_2; if (file->mode == 'r') return TREXIO_READONLY; $group$_t* const $group$ = trexio_text_read_$group$((trexio_text_t*) file); if ($group$ == NULL) return TREXIO_FAILURE; if ($group$->$group_dset$ != NULL) { FREE($group$->$group_dset$); } $group$->rank_$group_dset$ = rank; uint64_t dim_size = 1; for (unsigned int i=0; i<$group$->rank_$group_dset$; ++i){ $group$->dims_$group_dset$[i] = dims[i]; dim_size *= dims[i]; } $group$->$group_dset$ = CALLOC(dim_size, $group_dset_dtype$); for (uint64_t i=0 ; i<dim_size ; ++i) { $group$->$group_dset$[i] = $group_dset$[i]; } $group$->to_flush = 1; return TREXIO_SUCCESS; }
trexio_exit_code trexio_text_has_$group_dset$ (trexio_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; $group$_t* const $group$ = trexio_text_read_$group$((trexio_text_t*) file); if ($group$ == NULL) return TREXIO_FAILURE; if ($group$->rank_$group_dset$ > 0){ return TREXIO_SUCCESS; } else { return TREXIO_HAS_NOT; } }
1.10 RDM struct (hard-coded)
1.10.1 Read the complete struct
rdm_t* trexio_text_read_rdm(trexio_text_t* const file);
rdm_t* trexio_text_read_rdm(trexio_text_t* const file) { if (file == NULL) return NULL; if (file->rdm != NULL) return file->rdm; /* Allocate the data structure */ rdm_t* rdm = MALLOC(rdm_t); assert (rdm != NULL); rdm->one_e = NULL; rdm->two_e_file_name[0] = '\0'; rdm->to_flush = 0; /* Try to open the file. If the file does not exist, return */ const char* rdm_file_name = "/rdm.txt"; strncpy (rdm->file_name, file->parent.file_name, TREXIO_MAX_FILENAME_LENGTH); strncat (rdm->file_name, rdm_file_name, TREXIO_MAX_FILENAME_LENGTH-strlen(rdm_file_name)); if (rdm->file_name[TREXIO_MAX_FILENAME_LENGTH-1] != '\0') { FREE(rdm); return NULL; } /* If the file exists, read it */ FILE* f = fopen(rdm->file_name,"r"); if (f != NULL) { /* Find size of file to allocate the max size of the string buffer */ fseek(f, 0L, SEEK_END); size_t sz = ftell(f); fseek(f, 0L, SEEK_SET); sz = (sz < 1024) ? (1024) : (sz); char* buffer = CALLOC(sz, char); /* Read the dimensioning variables */ int rc; rc = fscanf(f, "%1023s", buffer); assert (rc == 1); assert (strcmp(buffer, "dim_one_e") == 0); rc = fscanf(f, "%" SCNu64 "", &(rdm->dim_one_e)); assert (rc == 1); /* Allocate arrays */ rdm->one_e = CALLOC(rdm->dim_one_e, double); assert (rdm->one_e != NULL); /* Read one_e */ rc = fscanf(f, "%1023s", buffer); assert (rc == 1); assert (strcmp(buffer, "one_e") == 0); for (uint64_t i=0 ; i<rdm->dim_one_e; ++i) { rc = fscanf(f, "%lf", &(rdm->one_e[i])); assert (rc == 1); } /* Read two_e */ rc = fscanf(f, "%1023s", buffer); assert (rc == 1); assert (strcmp(buffer, "two_e_file_name") == 0); rc = fscanf(f, "%1023s", buffer); assert (rc == 1); strncpy(rdm->two_e_file_name, buffer, 1024); if (rdm->two_e_file_name[TREXIO_MAX_FILENAME_LENGTH-1] != '\0') { FREE(buffer); FREE(rdm->one_e); FREE(rdm); fclose(f); return NULL; } FREE(buffer); fclose(f); f = NULL; } file->rdm = rdm ; return rdm; }
1.10.2 Flush the complete struct
trexio_exit_code trexio_text_flush_rdm(trexio_text_t* const file);
trexio_exit_code trexio_text_flush_rdm(trexio_text_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (file->parent.mode == 'r') return TREXIO_READONLY; rdm_t* const rdm = file->rdm; if (rdm == NULL) return TREXIO_SUCCESS; if (rdm->to_flush == 0) return TREXIO_SUCCESS; FILE* f = fopen(rdm->file_name,"w"); assert (f != NULL); /* Write the dimensioning variables */ fprintf(f, "num %" PRIu64 "\n", rdm->dim_one_e); /* Write arrays */ fprintf(f, "one_e\n"); for (uint64_t i=0 ; i< rdm->dim_one_e; ++i) { fprintf(f, "%lf\n", rdm->one_e[i]); } fprintf(f, "two_e_file_name\n"); fprintf(f, "%s\n", rdm->two_e_file_name); fclose(f); rdm->to_flush = 0; return TREXIO_SUCCESS; }
1.10.3 Free memory
Memory is allocated when reading. The followig function frees memory.
trexio_exit_code trexio_text_free_rdm(trexio_text_t* const file);
trexio_exit_code trexio_text_free_rdm(trexio_text_t* const file) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (file->parent.mode != 'r') { trexio_exit_code rc = trexio_text_flush_rdm(file); if (rc != TREXIO_SUCCESS) return TREXIO_FAILURE; } rdm_t* const rdm = file->rdm; if (rdm == NULL) return TREXIO_SUCCESS; if (rdm->one_e != NULL) { FREE (rdm->one_e); } free (rdm); file->rdm = NULL; return TREXIO_SUCCESS; }
1.10.4 Read/Write the onee attribute
The one_e
array is assumed allocated with the appropriate size.
trexio_exit_code trexio_text_read_rdm_one_e(trexio_t* const file, double* const one_e, const uint64_t dim_one_e); trexio_exit_code trexio_text_write_rdm_one_e(trexio_t* const file, const double* one_e, const uint64_t dim_one_e);
trexio_exit_code trexio_text_read_rdm_one_e(trexio_t* const file, double* const one_e, const uint64_t dim_one_e) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (one_e == NULL) return TREXIO_INVALID_ARG_2; rdm_t* const rdm = trexio_text_read_rdm((trexio_text_t*) file); if (rdm == NULL) return TREXIO_FAILURE; if (dim_one_e != rdm->dim_one_e) return TREXIO_INVALID_ARG_3; for (uint64_t i=0 ; i<dim_one_e ; ++i) { one_e[i] = rdm->one_e[i]; } return TREXIO_SUCCESS; } trexio_exit_code trexio_text_write_rdm_one_e(trexio_t* const file, const double* one_e, const uint64_t dim_one_e) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (one_e == NULL) return TREXIO_INVALID_ARG_2; if (file->mode != 'r') return TREXIO_READONLY; rdm_t* const rdm = trexio_text_read_rdm((trexio_text_t*) file); if (rdm == NULL) return TREXIO_FAILURE; rdm->dim_one_e = dim_one_e; for (uint64_t i=0 ; i<dim_one_e ; ++i) { rdm->one_e[i] = one_e[i]; } rdm->to_flush = 1; return TREXIO_SUCCESS; }
1.10.5 Read/Write the twoe attribute
two_e
is a sparse data structure, which can be too large to fit
in memory. So we provide functions to read and write it by
chunks.
In the text back end, the easiest way to do it is to create a
file for each sparse float structure.
trexio_exit_code trexio_text_buffered_read_rdm_two_e(trexio_t* const file, const uint64_t offset, const uint64_t size, int64_t* const index, double* const value); trexio_exit_code trexio_text_buffered_write_rdm_two_e(trexio_t* const file, const uint64_t offset, const uint64_t size, const int64_t* index, const double* value);
trexio_exit_code trexio_text_buffered_read_rdm_two_e(trexio_t* const file, const uint64_t offset, const uint64_t size, int64_t* const index, double* const value) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (index == NULL) return TREXIO_INVALID_ARG_4; if (value == NULL) return TREXIO_INVALID_ARG_5; rdm_t* const rdm = trexio_text_read_rdm((trexio_text_t*) file); if (rdm == NULL) return TREXIO_FAILURE; FILE* f = fopen(rdm->two_e_file_name, "r"); if (f == NULL) return TREXIO_END; const uint64_t line_length = 64L; fseek(f, (long) offset * line_length, SEEK_SET); for (uint64_t i=0 ; i<size ; ++i) { int rc = fscanf(f, "%9" SCNd64 " %9" SCNd64 " %9" SCNd64 " %9" SCNd64 " %24le\n", &index[4*i], &index[4*i+1], &index[4*i+2], &index[4*i+3], &value[i]); if (rc == 5) { /* Do nothing */ } else if (rc == EOF) { return TREXIO_END; } } return TREXIO_SUCCESS; } trexio_exit_code trexio_text_buffered_write_rdm_two_e(trexio_t* const file, const uint64_t offset, const uint64_t size, const int64_t* index, const double* value) { if (file == NULL) return TREXIO_INVALID_ARG_1; if (index == NULL) return TREXIO_INVALID_ARG_4; if (value == NULL) return TREXIO_INVALID_ARG_5; if (file->mode != 'r') return TREXIO_READONLY; rdm_t* const rdm = trexio_text_read_rdm((trexio_text_t*) file); if (rdm == NULL) return TREXIO_FAILURE; FILE* f = fopen(rdm->two_e_file_name, "w"); if (f == NULL) return TREXIO_FAILURE; const uint64_t line_length = 64L; fseek(f, (long) offset * line_length, SEEK_SET); for (uint64_t i=0 ; i<size ; ++i) { int rc = fprintf(f, "%9" PRId64 " %9" PRId64 " %9" PRId64 " %9" PRId64 " %24le\n", index[4*i], index[4*i+1], index[4*i+2], index[4*i+3], value[i]); if (rc != 5) return TREXIO_FAILURE; } return TREXIO_SUCCESS; }