X-Git-Url: https://git.xonotic.org/?a=blobdiff_plain;f=code.c;h=ddfbb977f2b1dd6b451d4d4d3becc4426de1dea7;hb=b21e9675810c8743401e0bdab72c70e116ae1659;hp=8654ced3156163a611465d63a61e753d3c0ca20b;hpb=0b322a8944aac54e143cf251df5636fc3b77bcfc;p=xonotic%2Fgmqcc.git diff --git a/code.c b/code.c index 8654ced..ddfbb97 100644 --- a/code.c +++ b/code.c @@ -1,6 +1,7 @@ /* - * Copyright (C) 2012 - * Dale Weiler + * Copyright (C) 2012, 2013, 2014 + * Dale Weiler + * Wolfgang Bumiller * * Permission is hereby granted, free of charge, to any person obtaining a copy of * this software and associated documentation files (the "Software"), to deal in @@ -20,151 +21,441 @@ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ -#include -#include +#include #include "gmqcc.h" -typedef struct { - uint16_t opcode; - - /* operand 1 */ - union { - int16_t s1; /* signed */ - uint16_t u1; /* unsigned */ - }; - /* operand 2 */ - union { - int16_t s2; /* signed */ - uint16_t u2; /* unsigned */ - }; - /* operand 3 */ - union { - int16_t s3; /* signed */ - uint16_t u3; /* unsigned */ - }; - - /* - * This is the same as the structure in darkplaces - * { - * unsigned short op; - * short a,b,c; - * } - * But this one is more sane to work with, and the - * type sizes are guranteed. - */ -} prog_section_statement; - -typedef struct { - /* The type is (I assume) - * 0 = ev_void - * 1 = ev_string - * 2 = ev_float - * 3 = ev_vector - * 4 = ev_entity - * 5 = ev_field - * 6 = ev_function - * 7 = ev_pointer - * 8 = ev_bad (is this right for uint16_t type?) - */ - uint16_t type; - uint16_t offset; /* offset in file? (what about length) */ - uint32_t name; /* offset in string table? (confused :() */ -} prog_section_both; - /* - * var and field use the same structure. But lets not use the same - * name just for safety reasons? (still castable ...). + * We could use the old method of casting to uintptr_t then to void* + * or qcint_t; however, it's incredibly unsafe for two reasons. + * 1) The compilers aliasing optimization can legally make it unstable + * (it's undefined behaviour). + * + * 2) The cast itself depends on fresh storage (newly allocated in which + * ever function is using the cast macros), the contents of which are + * transferred in a way that the obligation to release storage is not + * propagated. */ -typedef prog_section_both prog_section_var; -typedef prog_section_both prog_section_field; - -typedef struct { - int32_t entry; /* in statement table for instructions */ - uint32_t args; /* What is this? */ - uint32_t locals; /* Total ints of params + locals */ - uint32_t profile; /* What is this? */ - uint32_t name; /* name of function in string table */ - uint32_t nargs; /* number of arguments */ - uint8_t argsize[8]; /* size of arguments (keep 8 always?) */ -} prog_section_function; - -typedef struct { - uint32_t offset; /* Offset in file of where data begins */ - uint32_t length; /* Length of section (how many of) */ -} prog_section; - -typedef struct { - uint32_t version; /* Program version (6) */ - uint32_t crc16; /* What is this? */ - prog_section statements; /* prog_section_statement */ - prog_section vars; /* prog_section_var */ - prog_section fields; /* prog_section_field */ - prog_section functions; /* prog_section_function */ - prog_section strings; /* What is this? */ - prog_section globals; /* What is this? */ - uint32_t entfield; /* Number of entity fields */ -} prog_header; +typedef union { + void *enter; + qcint_t leave; +} code_hash_entry_t; + +/* Some sanity macros */ +#define CODE_HASH_ENTER(ENTRY) ((ENTRY).enter) +#define CODE_HASH_LEAVE(ENTRY) ((ENTRY).leave) + +void code_push_statement(code_t *code, prog_section_statement_t *stmt_in, lex_ctx_t ctx) +{ + prog_section_statement_t stmt = *stmt_in; + + if (OPTS_FLAG(TYPELESS_STORES)) { + switch (stmt.opcode) { + case INSTR_LOAD_S: + case INSTR_LOAD_ENT: + case INSTR_LOAD_FLD: + case INSTR_LOAD_FNC: + stmt.opcode = INSTR_LOAD_F; + break; + case INSTR_STORE_S: + case INSTR_STORE_ENT: + case INSTR_STORE_FLD: + case INSTR_STORE_FNC: + stmt.opcode = INSTR_STORE_F; + break; + case INSTR_STOREP_S: + case INSTR_STOREP_ENT: + case INSTR_STOREP_FLD: + case INSTR_STOREP_FNC: + stmt.opcode = INSTR_STOREP_F; + break; + } + } + + + if (OPTS_FLAG(SORT_OPERANDS)) { + uint16_t pair; + + switch (stmt.opcode) { + case INSTR_MUL_F: + case INSTR_MUL_V: + case INSTR_ADD_F: + case INSTR_EQ_F: + case INSTR_EQ_S: + case INSTR_EQ_E: + case INSTR_EQ_FNC: + case INSTR_NE_F: + case INSTR_NE_V: + case INSTR_NE_S: + case INSTR_NE_E: + case INSTR_NE_FNC: + case INSTR_AND: + case INSTR_OR: + case INSTR_BITAND: + case INSTR_BITOR: + if (stmt.o1.u1 < stmt.o2.u1) { + uint16_t a = stmt.o2.u1; + stmt.o1.u1 = stmt.o2.u1; + stmt.o2.u1 = a; + } + break; + + case INSTR_MUL_VF: pair = INSTR_MUL_FV; goto case_pair_gen; + case INSTR_MUL_FV: pair = INSTR_MUL_VF; goto case_pair_gen; + case INSTR_LT: pair = INSTR_GT; goto case_pair_gen; + case INSTR_GT: pair = INSTR_LT; goto case_pair_gen; + case INSTR_LE: pair = INSTR_GT; goto case_pair_gen; + case INSTR_GE: pair = INSTR_LE; + + case_pair_gen: + if (stmt.o1.u1 < stmt.o2.u1) { + uint16_t x = stmt.o1.u1; + stmt.o1.u1 = stmt.o2.u1; + stmt.o2.u1 = x; + stmt.opcode = pair; + } + break; + } + } + + vec_push(code->statements, stmt); + vec_push(code->linenums, (int)ctx.line); + vec_push(code->columnnums, (int)ctx.column); +} + +void code_pop_statement(code_t *code) +{ + vec_pop(code->statements); + vec_pop(code->linenums); + vec_pop(code->columnnums); +} + +code_t *code_init() { + static lex_ctx_t empty_ctx = {0, 0, 0}; + static prog_section_function_t empty_function = {0,0,0,0,0,0,0,{0,0,0,0,0,0,0,0}}; + static prog_section_statement_t empty_statement = {0,{0},{0},{0}}; + static prog_section_def_t empty_def = {0, 0, 0}; + + code_t *code = (code_t*)mem_a(sizeof(code_t)); + int i = 0; + + memset(code, 0, sizeof(code_t)); + code->entfields = 0; + code->string_cache = util_htnew(OPTS_OPTIMIZATION(OPTIM_OVERLAP_STRINGS) ? 0x100 : 1024); + + /* + * The way progs.dat is suppose to work is odd, there needs to be + * some null (empty) statements, functions, and 28 globals + */ + for(; i < 28; i++) + vec_push(code->globals, 0); + + vec_push(code->chars, '\0'); + vec_push(code->functions, empty_function); + + code_push_statement(code, &empty_statement, empty_ctx); + + vec_push(code->defs, empty_def); + vec_push(code->fields, empty_def); + + return code; +} + +void *code_util_str_htgeth(hash_table_t *ht, const char *key, size_t bin); + +uint32_t code_genstring(code_t *code, const char *str) { + size_t hash; + code_hash_entry_t existing; + + if (!str) + return 0; + + if (!*str) { + if (!code->string_cached_empty) { + code->string_cached_empty = vec_size(code->chars); + vec_push(code->chars, 0); + } + return code->string_cached_empty; + } + + if (OPTS_OPTIMIZATION(OPTIM_OVERLAP_STRINGS)) { + hash = ((unsigned char*)str)[strlen(str)-1]; + CODE_HASH_ENTER(existing) = code_util_str_htgeth(code->string_cache, str, hash); + } else { + hash = util_hthash(code->string_cache, str); + CODE_HASH_ENTER(existing) = util_htgeth(code->string_cache, str, hash); + } + + if (CODE_HASH_ENTER(existing)) + return CODE_HASH_LEAVE(existing); + + CODE_HASH_LEAVE(existing) = vec_size(code->chars); + vec_append(code->chars, strlen(str)+1, str); + + util_htseth(code->string_cache, str, hash, CODE_HASH_ENTER(existing)); + return CODE_HASH_LEAVE(existing); +} + +qcint_t code_alloc_field (code_t *code, size_t qcsize) +{ + qcint_t pos = (qcint_t)code->entfields; + code->entfields += qcsize; + return pos; +} + +static size_t code_size_generic(code_t *code, prog_header_t *code_header, bool lno) { + size_t size = 0; + if (lno) { + size += 4; /* LNOF */ + size += sizeof(uint32_t); /* version */ + size += sizeof(code_header->defs.length); + size += sizeof(code_header->globals.length); + size += sizeof(code_header->fields.length); + size += sizeof(code_header->statements.length); + size += sizeof(code->linenums[0]) * vec_size(code->linenums); + size += sizeof(code->columnnums[0]) * vec_size(code->columnnums); + } else { + size += sizeof(prog_header_t); + size += sizeof(prog_section_statement_t) * vec_size(code->statements); + size += sizeof(prog_section_def_t) * vec_size(code->defs); + size += sizeof(prog_section_field_t) * vec_size(code->fields); + size += sizeof(prog_section_function_t) * vec_size(code->functions); + size += sizeof(int32_t) * vec_size(code->globals); + size += 1 * vec_size(code->chars); + } + return size; +} + +#define code_size_binary(C, H) code_size_generic((C), (H), false) +#define code_size_debug(C, H) code_size_generic((C), (H), true) + +static void code_create_header(code_t *code, prog_header_t *code_header, const char *filename, const char *lnofile) { + size_t i; + + code_header->statements.offset = sizeof(prog_header_t); + code_header->statements.length = vec_size(code->statements); + code_header->defs.offset = code_header->statements.offset + (sizeof(prog_section_statement_t) * vec_size(code->statements)); + code_header->defs.length = vec_size(code->defs); + code_header->fields.offset = code_header->defs.offset + (sizeof(prog_section_def_t) * vec_size(code->defs)); + code_header->fields.length = vec_size(code->fields); + code_header->functions.offset = code_header->fields.offset + (sizeof(prog_section_field_t) * vec_size(code->fields)); + code_header->functions.length = vec_size(code->functions); + code_header->globals.offset = code_header->functions.offset + (sizeof(prog_section_function_t) * vec_size(code->functions)); + code_header->globals.length = vec_size(code->globals); + code_header->strings.offset = code_header->globals.offset + (sizeof(int32_t) * vec_size(code->globals)); + code_header->strings.length = vec_size(code->chars); + code_header->version = 6; + code_header->skip = 0; + + if (OPTS_OPTION_BOOL(OPTION_FORCECRC)) + code_header->crc16 = OPTS_OPTION_U16(OPTION_FORCED_CRC); + else + code_header->crc16 = code->crc; + code_header->entfield = code->entfields; + + if (OPTS_FLAG(DARKPLACES_STRING_TABLE_BUG)) { + /* >= + P */ + vec_push(code->chars, '\0'); /* > */ + vec_push(code->chars, '\0'); /* = */ + vec_push(code->chars, '\0'); /* P */ + } + + /* ensure all data is in LE format */ + util_swap_header(code_header); + + /* + * These are not part of the header but we ensure LE format here to save on duplicated + * code. + */ + + util_swap_statements (code->statements); + util_swap_defs_fields(code->defs); + util_swap_defs_fields(code->fields); + util_swap_functions (code->functions); + util_swap_globals (code->globals); + + if (!OPTS_OPTION_BOOL(OPTION_QUIET)) { + if (lnofile) + con_out("writing '%s' and '%s'...\n", filename, lnofile); + else + con_out("writing '%s'\n", filename); + } + + if (!OPTS_OPTION_BOOL(OPTION_QUIET) && + !OPTS_OPTION_BOOL(OPTION_PP_ONLY)) + { + char buffer[1024]; + con_out("\nOptimizations:\n"); + for (i = 0; i < COUNT_OPTIMIZATIONS; ++i) { + if (opts_optimizationcount[i]) { + util_optimizationtostr(opts_opt_list[i].name, buffer, sizeof(buffer)); + con_out( + " %s: %u\n", + buffer, + (unsigned int)opts_optimizationcount[i] + ); + } + } + } +} + +static void code_stats(const char *filename, const char *lnofile, code_t *code, prog_header_t *code_header) { + if (OPTS_OPTION_BOOL(OPTION_QUIET) || + OPTS_OPTION_BOOL(OPTION_PP_ONLY)) + return; + + con_out("\nFile statistics:\n"); + con_out(" dat:\n"); + con_out(" name: %s\n", filename); + con_out(" size: %u (bytes)\n", code_size_binary(code, code_header)); + con_out(" crc: 0x%04X\n", code->crc); + + if (lnofile) { + con_out(" lno:\n"); + con_out(" name: %s\n", lnofile); + con_out(" size: %u (bytes)\n", code_size_debug(code, code_header)); + } + + con_out("\n"); +} /* - * The macros below expand to a typesafe vector implementation, which - * can be viewed in gmqcc.h - * - * code_statements_data -- raw prog_section_statement array - * code_statements_elements -- number of elements - * code_statements_allocated -- size of the array allocated - * code_statements_add(T) -- add element (returns -1 on error) - * - * code_vars_data -- raw prog_section_var array - * code_vars_elements -- number of elements - * code_vars_allocated -- size of the array allocated - * code_vars_add(T) -- add element (returns -1 on error) - * - * code_fields_data -- raw prog_section_field array - * code_fields_elements -- number of elements - * code_fields_allocated -- size of the array allocated - * code_fields_add(T) -- add element (returns -1 on error) - * - * code_functions_data -- raw prog_section_function array - * code_functions_elements -- number of elements - * code_functions_allocated -- size of the array allocated - * code_functions_add(T) -- add element (returns -1 on error) - * - * code_globals_data -- raw prog_section_var array - * code_globals_elements -- number of elements - * code_globals_allocated -- size of the array allocated - * code_globals_add(T) -- add element (returns -1 on error) - * - * code_strings_data -- raw char* array - * code_strings_elements -- number of elements - * code_strings_allocated -- size of the array allocated - * code_strings_add(T) -- add element (returns -1 on error) + * Same principle except this one allocates memory and writes the lno(optional) and the dat file + * directly out to allocated memory. Which is actually very useful for the future library support + * we're going to add. */ -VECTOR_MAKE(prog_section_statement, code_statements); -VECTOR_MAKE(prog_section_var, code_vars ); -VECTOR_MAKE(prog_section_field, code_fields ); -VECTOR_MAKE(prog_section_function, code_functions ); -VECTOR_MAKE(prog_section_var, code_globals ); -VECTOR_MAKE(char*, code_strings ); - -/* program header */ -prog_header code_header; -void code_write() { - code_header.version = 6; - code_header.crc16 = 0; /* TODO: */ - code_header.statements = (prog_section){sizeof(prog_header), code_statements_elements }; - code_header.vars = (prog_section){sizeof(prog_header)+sizeof(prog_section_statement)*code_statements_elements, code_vars_elements }; - code_header.fields = (prog_section){sizeof(prog_header)+sizeof(prog_section_var) *code_vars_elements, code_fields_elements }; - code_header.functions = (prog_section){sizeof(prog_header)+sizeof(prog_section_field) *code_fields_elements, code_functions_elements }; - code_header.globals = (prog_section){sizeof(prog_header)+sizeof(prog_section_function) *code_functions_elements, code_globals_elements }; - /* how, I think I don't have strings figured out yet :| */ - code_header.entfield = 0; /* TODO: */ - - #if 0 /* is this right? */ - fwrite(&code_header, 1, sizeof(prog_header), fp); - fwrite(code_statements_data, 1, sizeof(prog_section_statement)*code_statements_elements, fp); - fwrite(code_vars_data, 1, sizeof(prog_section_var) *code_vars_elements, fp); - fwrite(code_fields_data, 1, sizeof(prog_section_field) *code_fields_elements, fp); - fwrite(code_functions_data, 1, sizeof(prog_section_function) *code_functions_elements, fp); - fwrite(code_globals_data, 1, sizeof(prog_section_var) *code_globals_elements, fp); - #endif +#if 0 +static bool code_write_memory(code_t *code, uint8_t **datmem, size_t *sizedat, uint8_t **lnomem, size_t *sizelno) GMQCC_UNUSED { + prog_header_t code_header; + uint32_t offset = 0; + + if (!datmem) + return false; + + code_create_header(code, &code_header, "<>", "<>"); + + #define WRITE_CHUNK(C,X,S) \ + do { \ + memcpy((void*)(&(*C)[offset]), (const void*)(X), (S)); \ + offset += (S); \ + } while (0) + + /* Calculate size required to store entire file out to memory */ + if (lnomem) { + uint32_t version = 1; + + *sizelno = code_size_debug(code, &code_header); + *lnomem = (uint8_t*)mem_a(*sizelno); + + WRITE_CHUNK(lnomem, "LNOF", 4); + WRITE_CHUNK(lnomem, &version, sizeof(version)); + WRITE_CHUNK(lnomem, &code_header.defs.length, sizeof(code_header.defs.length)); + WRITE_CHUNK(lnomem, &code_header.globals.length, sizeof(code_header.globals.length)); + WRITE_CHUNK(lnomem, &code_header.fields.length, sizeof(code_header.fields.length)); + WRITE_CHUNK(lnomem, &code_header.statements.length, sizeof(code_header.statements.length)); + + /* something went terribly wrong */ + if (offset != *sizelno) { + mem_d(*lnomem); + *sizelno = 0; + return false; + } + offset = 0; + } + + /* Write out the dat */ + *sizedat = code_size_binary(code, &code_header); + *datmem = (uint8_t*)mem_a(*sizedat); + + WRITE_CHUNK(datmem, &code_header, sizeof(prog_header_t)); + WRITE_CHUNK(datmem, code->statements, sizeof(prog_section_statement_t) * vec_size(code->statements)); + WRITE_CHUNK(datmem, code->defs, sizeof(prog_section_def_t) * vec_size(code->defs)); + WRITE_CHUNK(datmem, code->fields, sizeof(prog_section_field_t) * vec_size(code->fields)); + WRITE_CHUNK(datmem, code->functions, sizeof(prog_section_function_t) * vec_size(code->functions)); + WRITE_CHUNK(datmem, code->globals, sizeof(int32_t) * vec_size(code->globals)); + WRITE_CHUNK(datmem, code->chars, 1 * vec_size(code->chars)); + + vec_free(code->statements); + vec_free(code->linenums); + vec_free(code->columnnums); + vec_free(code->defs); + vec_free(code->fields); + vec_free(code->functions); + vec_free(code->globals); + vec_free(code->chars); + + util_htdel(code->string_cache); + mem_d(code); + code_stats("<>", (lnomem) ? "<>" : NULL, code, &code_header); + return true; +} +#endif /*!#if 0 reenable when ready to be used */ +#undef WRITE_CHUNK + +bool code_write(code_t *code, const char *filename, const char *lnofile) { + prog_header_t code_header; + fs_file_t *fp = NULL; + + code_create_header(code, &code_header, filename, lnofile); + + if (lnofile) { + uint32_t version = 1; + + fp = fs_file_open(lnofile, "wb"); + if (!fp) + return false; + + util_endianswap(&version, 1, sizeof(version)); + util_endianswap(code->linenums, vec_size(code->linenums), sizeof(code->linenums[0])); + util_endianswap(code->columnnums, vec_size(code->columnnums), sizeof(code->columnnums[0])); + + if (fs_file_write("LNOF", 4, 1, fp) != 1 || + fs_file_write(&version, sizeof(version), 1, fp) != 1 || + fs_file_write(&code_header.defs.length, sizeof(code_header.defs.length), 1, fp) != 1 || + fs_file_write(&code_header.globals.length, sizeof(code_header.globals.length), 1, fp) != 1 || + fs_file_write(&code_header.fields.length, sizeof(code_header.fields.length), 1, fp) != 1 || + fs_file_write(&code_header.statements.length, sizeof(code_header.statements.length), 1, fp) != 1 || + fs_file_write(code->linenums, sizeof(code->linenums[0]), vec_size(code->linenums), fp) != vec_size(code->linenums) || + fs_file_write(code->columnnums, sizeof(code->columnnums[0]), vec_size(code->columnnums), fp) != vec_size(code->columnnums)) + { + con_err("failed to write lno file\n"); + } + + fs_file_close(fp); + fp = NULL; + } + + fp = fs_file_open(filename, "wb"); + if (!fp) + return false; + + if (1 != fs_file_write(&code_header, sizeof(prog_header_t) , 1 , fp) || + vec_size(code->statements) != fs_file_write(code->statements, sizeof(prog_section_statement_t), vec_size(code->statements), fp) || + vec_size(code->defs) != fs_file_write(code->defs, sizeof(prog_section_def_t) , vec_size(code->defs) , fp) || + vec_size(code->fields) != fs_file_write(code->fields, sizeof(prog_section_field_t) , vec_size(code->fields) , fp) || + vec_size(code->functions) != fs_file_write(code->functions, sizeof(prog_section_function_t) , vec_size(code->functions) , fp) || + vec_size(code->globals) != fs_file_write(code->globals, sizeof(int32_t) , vec_size(code->globals) , fp) || + vec_size(code->chars) != fs_file_write(code->chars, 1 , vec_size(code->chars) , fp)) + { + fs_file_close(fp); + return false; + } + + fs_file_close(fp); + code_stats(filename, lnofile, code, &code_header); + return true; +} + +void code_cleanup(code_t *code) { + vec_free(code->statements); + vec_free(code->linenums); + vec_free(code->columnnums); + vec_free(code->defs); + vec_free(code->fields); + vec_free(code->functions); + vec_free(code->globals); + vec_free(code->chars); + + util_htdel(code->string_cache); + + mem_d(code); }