#include #include #include #include #include #include #include #include #include #include "CCubes.h" #include "real.h" #include "cl_setup.h" #include "clccubes.h" #include "config.h" #include "logging.h" // $ export BITS_PER_WORD=32 in the Terminal, before running R SEXP CCubes(SEXP tt) { // simulate the R command: // system.file("ccubes.cl", package = "IEEE") const char *kernel_file = NULL; SEXP args = PROTECT(allocVector(VECSXP, 2)); SET_VECTOR_ELT(args, 0, mkString("ccubes.cl")); SET_VECTOR_ELT(args, 1, mkString("IEEE")); SEXP names = PROTECT(allocVector(STRSXP, 2)); SET_STRING_ELT(names, 1, mkChar("package")); Rf_setAttrib(args, R_NamesSymbol, names); SEXP r_result = PROTECT(R_tryEval( Rf_lang3(install("do.call"), mkString("system.file"), args), R_GlobalEnv, NULL )); Rprintf("Kernel file: '%s'\n", CHAR(STRING_ELT(r_result, 0))); kernel_file = CHAR(STRING_ELT(r_result, 0)); UNPROTECT(3); // 32 bits per word, in bit shifting representation char *bits_per_word = getenv("BITS_PER_WORD"); // Read from the PATH int BITS_PER_WORD = bits_per_word ? atoi(bits_per_word) : 32; if (BITS_PER_WORD != 8 && BITS_PER_WORD != 16 && BITS_PER_WORD != 32 && BITS_PER_WORD != 64) { BITS_PER_WORD = 32; // Default to 32 } // $ export PRINT_INFO=true in the Terminal, before running R char *print_info = getenv("PRINT_INFO"); // Read from the PATH Rboolean PRINT_INFO = print_info && ( strcmp(print_info, "1") == 0 || strcmp(print_info, "TRUE") == 0 || strcmp(print_info, "true") == 0 || strcmp(print_info, "T") == 0 || strcmp(print_info, "t") == 0 ); int multiplier = 0; struct timeval start, end; double elapsed_time; config_set_int("log", LOG_LEVEL_WARN); config_set_int("log:clccubes", LOG_LEVEL_DEBUG); config_set_int("log:ccubes", LOG_LEVEL_DEBUG); config_set_int("log:cl", LOG_LEVEL_DEBUG); char log[100] = {0}; sprintf(log, "log-ccubes"); config_set_string("out", log); struct ccubes_context *ctx = ccubes_create(kernel_file); if (ctx == NULL) { log_error("ccubes", "ccubes_do_tasks failed"); } if (PRINT_INFO) { Rprintf("--- START ---\n"); gettimeofday(&start, NULL); // Start time } int *p_tt = INTEGER(tt); int ttrows = nrows(tt); // number of rows in the data matrix int ninputs = ncols(tt) - 1; // number of inputs (columns - 1, the last one is the outcome) // calculate the number of positive output rows (the ON set) int posrows = 0; for (int r = 0; r < ttrows; r++) { posrows += p_tt[ninputs * ttrows + r]; } // calculate the number of negative output rows (the OFF set) int negrows = ttrows - posrows; if (negrows == 0) { // if there are no negative output rows, no PIs can be found // all inputs will be completely minimized return(R_NilValue); } // split the minterms in the ON and OFF set matrices int ON_set[posrows * ninputs]; int OFF_set[ninputs * negrows]; int rowpos = 0, rowneg = 0; int max_value = 0; for (int r = 0; r < ttrows; r++) { if (p_tt[ninputs * ttrows + r] == 1) { // positive for (int c = 0; c < ninputs; c++) { int value = p_tt[c * ttrows + r]; ON_set[c * posrows + rowpos] = value; if (value > max_value) { max_value = value; } } rowpos += 1; } else { // negative for (int c = 0; c < ninputs; c++) { int value = p_tt[c * ttrows + r]; OFF_set[c * negrows + rowneg] = value; if (value > max_value) { max_value = value; } } rowneg += 1; } } int bits_needed = ceil(log2(max_value)); // Compute the necessary bits int value_bit_width = 1; while (value_bit_width < bits_needed) { value_bit_width *= 2; // Round up to the next power of 2 } if (value_bit_width > BITS_PER_WORD) { BITS_PER_WORD = value_bit_width; // Adjust the bits per word } int value_bit_mask = (1U << value_bit_width) - 1U; // calculate the number of values (biggest number) for each input int nofvalues[ninputs]; int nofpi[ninputs]; for (int i = 0; i < ninputs; i++) { nofvalues[i] = 0; // initialize nofpi[i] = 0; // initialize for (int r = 0; r < ttrows; r++) { if (nofvalues[i] < p_tt[i * ttrows + r]) { nofvalues[i] = p_tt[i * ttrows + r]; } } // add 1 because if the biggest number is 2 then it has three levels: 0, 1 and 2 nofvalues[i] += 1; if (nofvalues[i] == 1) { // no input ever has less than two values nofvalues[i] = 2; } } // preallocating for an estimated large number of 10000 found PIs // this number will be iteratively increased when the found PIs reach the upper limit int estimPI = 250000; // the index of the PIs, in descending order of their number of covered ON-set minterms int *p_covered = (int *) R_Calloc(estimPI, int); // many PIs will have the same coverage, but they don't necessarily cover the same minterms // to employ row dominance when solving the coverage matrix, we need to compare the coverage of the // current PI with the coverage of the previous PIs. If this PI survives the comparison, its // coverage has to be added in the p_covered vector, and its order in the p_covered // vector, at the last index of the PI coverage with the same number of minterms int last_index[posrows]; // descending order // p_pichart = malloc(estimPI * posrows * sizeof(int)); // memset(p_pichart, false, estimPI * posrows * sizeof(int)); int *p_pichart = (int *) R_Calloc(estimPI * posrows, int); // prefixing (int *) before R_Calloc() prefills all values with 0s int pichart_words = (posrows + BITS_PER_WORD - 1) / BITS_PER_WORD; // Words needed per coverage matrix columns unsigned int *p_pichart_pos = (unsigned int *) R_Calloc(estimPI * pichart_words, unsigned int); int implicant_words = (ninputs + BITS_PER_WORD - 1) / BITS_PER_WORD; // Words needed per PI representation unsigned int *p_implicants_pos = (unsigned int *) R_Calloc(estimPI * implicant_words, unsigned int); unsigned int *p_implicants_val = (unsigned int *) R_Calloc(estimPI * implicant_words, unsigned int); int prevfoundPI = 0; // the number of previously found PIs int foundPI = 0; int prevsolmin = 0; // the previous (level k - 1), minimum number of PIs to solve the coverage matrix int solmin = 0; // the positions of the PIs solving the coverage matrix // this vector can never be lengthier than the number of ON minterms (posrows) int previndices[posrows]; int indices[posrows]; for (int i = 0; i < posrows; i++) { previndices[i] = 0; indices[i] = 0; last_index[i] = 0; } Rboolean ON_set_covered = false; if (PRINT_INFO) { Rprintf("ON-set minterms: %d\n", posrows); } int stop_counter = 0; // to stop if two consecutive levels of complexity yield no PIs int k; for (k = 1; k <= ninputs; k++) { if (PRINT_INFO) { Rprintf("---k: %d\n", k); } log_debug("ccubes", "---k: %d\n", k); int n_tasks = nchoosek(ninputs, k); if (n_tasks == 0) { // overflow, too many tasks return (R_NilValue); } int n_tasks_batch = 512; for (int task = 0; task < n_tasks; task+=n_tasks_batch) { /* adjust if batch size is larger than total job size */ int current_batch = n_tasks_batch; /* single batch */ if (n_tasks < n_tasks_batch) current_batch = n_tasks; /* last batch */ if (n_tasks - task < n_tasks_batch) current_batch = n_tasks - task; log_debug("ccubes", "Tasks %d - %d out of %d", task, task + current_batch - 1, n_tasks); bool *redundant; bool *coverage; int *taskpis; unsigned int *fixed_bits; unsigned int *value_bits; unsigned int *pichart_values; ccubes_do_tasks(ctx, current_batch, task, k, prevfoundPI, ninputs, posrows, negrows, implicant_words, value_bit_width, value_bit_mask, pichart_words, estimPI, nofvalues, ON_set, OFF_set, p_implicants_pos, p_implicants_val, last_index, p_covered, p_pichart_pos, redundant, coverage, taskpis, fixed_bits, value_bits, pichart_values ); for (int current_task = 0; current_task < current_batch; current_task++) { int foundpis = ctx->h_taskpis[current_task]; log_debug("ccubes", "Task %d, found: %d", current_task, foundpis); if (foundpis > 5) { log_debug("ccubes", "Something is wrong\n"); foundpis = 0; } for (int f = 0; f < foundpis; f++) { // if (!ctx->h_redundant[current_task + f]) { /* LOG TASK */ log_debug_raw("ccubes", "coverage[%d][%d]:", current_task, f); for (int j = 0; j < posrows; j++) { log_debug_raw("ccubes", " %d", ctx->h_coverage[current_task * posrows * posrows + f * posrows + j]); } log_debug_raw("ccubes", "\n"); log_debug_raw("ccubes", "fixed_bits[%d][%d]:", current_task, f); for (int j = 0; j < implicant_words; j++) { log_debug_raw("ccubes", " %d", ctx->h_fixed_bits[current_task * posrows * implicant_words + f * posrows + j]); } log_debug_raw("ccubes", "\n"); log_debug_raw("ccubes", "value_bits[%d][%d]:", current_task, f); for (int j = 0; j < implicant_words; j++) { log_debug_raw("ccubes", " %d", ctx->h_value_bits[current_task * posrows * implicant_words + f * posrows + j]); } log_debug_raw("ccubes", "\n"); log_debug_raw("ccubes", "pichart_values[%d][%d]:", current_task, f); for (int j = 0; j < pichart_words; j++) { log_debug_raw("ccubes", " %d", ctx->h_pichart_values[current_task * posrows * pichart_words + f * posrows + j]); } log_debug_raw("ccubes", "\n"); int covsum = 0; for (int i = 0; i < posrows; i++) { covsum += ctx->h_coverage[current_task * posrows * posrows + f * posrows + i]; } // push the PI information to the global arrays for (int i = foundPI; i > last_index[covsum - 1]; i--) { p_covered[i] = p_covered[i - 1]; } p_covered[last_index[covsum - 1]] = foundPI; for (int l = 1; l < covsum; l++) { last_index[l - 1] += 1; } for (int w = 0; w < implicant_words; w++) { p_implicants_pos[implicant_words * foundPI + w] = ctx->h_fixed_bits[current_task * posrows * implicant_words + f * implicant_words + w]; p_implicants_val[implicant_words * foundPI + w] = ctx->h_value_bits[current_task * posrows * implicant_words + f * implicant_words + w]; } // populate the coverage matrix for (int r = 0; r < posrows; r++) { for (int w = 0; w < pichart_words; w++) { p_pichart_pos[foundPI * pichart_words + w] = ctx->h_pichart_values[current_task * posrows * pichart_words + f * pichart_words + w]; } p_pichart[posrows * foundPI + r] = ctx->h_coverage[current_task * posrows * posrows + f * posrows + r]; } ++foundPI; // when needed, increase allocated memory if (foundPI / estimPI > 0.9) { estimPI += 100000; p_pichart = R_Realloc(p_pichart, posrows * estimPI, int); p_pichart_pos = R_Realloc(p_pichart_pos, pichart_words * estimPI, unsigned int); p_implicants_val = R_Realloc(p_implicants_val, implicant_words * estimPI, unsigned int); p_implicants_pos = R_Realloc(p_implicants_pos, implicant_words * estimPI, unsigned int); p_covered = R_Realloc(p_covered, estimPI, int); for (unsigned int i = foundPI; i < posrows * estimPI; i++) { p_pichart[i] = 0; } for (unsigned int i = foundPI; i < pichart_words * estimPI; i++) { p_pichart_pos[i] = 0U; } for (unsigned int i = foundPI; i < implicant_words * estimPI; i++) { p_implicants_val[i] = 0U; p_implicants_pos[i] = 0U; } if (PRINT_INFO) { multiplier++; Rprintf("%dx ", multiplier); } } // } } } /* change to something less aggresive for reuse */ ccubes_clean_up(ctx); } // TODO: remove this after fixing the problems above // return(R_NilValue); nofpi[k - 1] = foundPI; if (foundPI > 0 && !ON_set_covered) { Rboolean test_coverage = true; int r = 0; while (r < posrows && test_coverage) { Rboolean minterm_covered = false; int c = 0; while (c < foundPI && !minterm_covered) { minterm_covered = p_pichart[c * posrows + r]; c++; } test_coverage = minterm_covered; r++; } ON_set_covered = test_coverage; } if (ON_set_covered) { // Rprintf("posrows: %d; foundPI: %d\n", posrows, foundPI); if (PRINT_INFO) { gettimeofday(&end, NULL); // End time elapsed_time = (end.tv_sec - start.tv_sec) + (end.tv_usec - start.tv_usec) / 1e6; Rprintf("Time taken finding %d PIs: %f sec.\n", foundPI, elapsed_time); gettimeofday(&start, NULL); } SEXP pic = PROTECT(allocMatrix(INTSXP, posrows, foundPI)); for (long long unsigned int i = 0; i < posrows * foundPI; i++) { INTEGER(pic)[i] = p_pichart[i]; } SEXP PIlayers = PROTECT(allocVector(INTSXP, ninputs)); for (int i = 0; i < ninputs; i++) { INTEGER(PIlayers)[i] = nofpi[i]; } setAttrib(pic, install("PIlayers"), PIlayers); // if this file is run directly using SHLIB, the following line is needed // R_ParseEvalString("library(IEEE)", R_GlobalEnv); SEXP pkg_env = PROTECT(R_FindNamespace(mkString("IEEE"))); SEXP solvechart = PROTECT(Rf_findVarInFrame(pkg_env, Rf_install("solvechart"))); SEXP evalinR = PROTECT(R_tryEval(Rf_lang2(solvechart, pic), pkg_env, NULL)); solmin = length(evalinR); for (int i = 0; i < solmin; i++) { indices[i] = INTEGER(evalinR)[i] - 1; // R is 1-based } UNPROTECT(5); // Rprintf("solution minima: %d\n", solmin); if (PRINT_INFO) { gettimeofday(&end, NULL); elapsed_time = (end.tv_sec - start.tv_sec) + (end.tv_usec - start.tv_usec) / 1e6; Rprintf("Time spent solving the coverage matrix: %f sec.\n", elapsed_time); gettimeofday(&start, NULL); } if (solmin == prevsolmin) { // the minimum number of PIs did not change in the current level of complexity // we can safely retain the less complex PIs from the previous level for (int i = 0; i < solmin; i++) { indices[i] = previndices[i]; } stop_counter += 1; } else { // this means solmin is in fact smaller than the previously found solmin // or it is the very first time a solmin was found // only here it makes sense to overwrite prevsolmin and previndices, // otherwise they are just as good as the ones from the previous level prevsolmin = solmin; for (int i = 0; i < solmin; i++) { previndices[i] = indices[i]; } stop_counter = 0; } } prevfoundPI = foundPI; // printf("stop_counter: %d\n", stop_counter); // One level of complexity up, and the solution minima does not change if (stop_counter > 0) { // the search can stop break; } } // Rprintf("solmin: %d\n", solmin); if (PRINT_INFO) { Rprintf("--- END ---\n"); } SEXP sol = PROTECT(allocMatrix(INTSXP, solmin, ninputs)); int *p_sol = INTEGER(sol); for (int c = 0; c < solmin; c++) { for (int r = 0; r < ninputs; r++) { unsigned int value = 0; int word_index = r / (BITS_PER_WORD / value_bit_width); // Word index within the implicant int bit_index = (r % (BITS_PER_WORD / value_bit_width)) * value_bit_width; // Bit position within the word // (1U << bit_index) is just as good if (p_implicants_pos[indices[c] * implicant_words + word_index] & (value_bit_mask << bit_index)) { value = (p_implicants_val[indices[c] * implicant_words + word_index] >> bit_index) & value_bit_mask; value++; // 0 indicates a minimization, so we increment the value } p_sol[r * solmin + c] = value; // transposed } } ccubes_destroy(ctx); R_Free(p_pichart); R_Free(p_implicants_val); R_Free(p_implicants_pos); R_Free(p_pichart_pos); R_Free(p_covered); UNPROTECT(1); return (sol); } long long unsigned int nchoosek( int n, int k ) { if (k > n) return 0; if (k == 0 || k == n) return 1; long long unsigned int result = 1; if (k > n - k) { k = n - k; } for (int i = 0; i < k; i++) { // result = result * (n - i) / (i + 1); // Check for potential overflow before multiplication if (result > ULLONG_MAX / (n - i)) { return 0; // Indicate overflow } result *= (n - i); // Check for potential overflow before division if (result % (i + 1) != 0) { return 0; // Indicate overflow } result /= (i + 1); } return result; }