X-Git-Url: http://shamusworld.gotdns.org/cgi-bin/gitweb.cgi?p=rmac;a=blobdiff_plain;f=expr.c;h=dc0e22bf90b88485ee2872d3dccdc7d35ac4313d;hp=6db7435dd205546e0afc4d25c3e018fcb1077301;hb=HEAD;hpb=30a208654896284b50e7b362e97d3e63ec717b96 diff --git a/expr.c b/expr.c index 6db7435..e597257 100644 --- a/expr.c +++ b/expr.c @@ -1,7 +1,7 @@ // -// RMAC - Reboot's Macro Assembler for all Atari computers +// RMAC - Renamed Macro Assembler for all Atari computers // EXPR.C - Expression Analyzer -// Copyright (C) 199x Landon Dyer, 2011-2018 Reboot and Friends +// Copyright (C) 199x Landon Dyer, 2011-2021 Reboot and Friends // RMAC derived from MADMAC v1.07 Written by Landon Dyer, 1986 // Source utilised with the kind permission of Landon Dyer // @@ -35,7 +35,7 @@ char itokcl[] = { CR_DEFINED, CR_REFERENCED, // SUNARY (special unary) CR_STREQ, CR_MACDEF, CR_DATE, CR_TIME, - CR_ABSCOUNT, 0, + CR_ABSCOUNT, CR_FILESIZE, 0, '!', '~', UNMINUS, UNLT, UNGT, 0, // UNARY '*', '/', '%', 0, // MULT '+', '-', 0, // ADD @@ -58,7 +58,6 @@ static PTR evalTokenBuffer; // Deposit tokens here (this is really a // riscasm.c) static int symbolNum; // Pointer to the entry in symbolPtr[] - // // Obtain a string value // @@ -72,7 +71,6 @@ static uint32_t str_value(char * p) return v; } - // // Initialize expression analyzer // @@ -95,28 +93,99 @@ void InitExpression(void) symbolNum = 0; } +extern int correctMathRules; +int xor(void); +int and(void); +int rel(void); +int shift(void); +int sum(void); +int product(void); // -// Binary operators (all the same precedence) +// Binary operators (all the same precedence, +// except if -4 is passed to the command line) // +#define precedence(HIERARCHY_HIGHER, HIERARCHY_CURRENT) \ +do \ +{ \ + if (HIERARCHY_HIGHER() != OK) \ + return ERROR; \ + while (tokenClass[*tok] == HIERARCHY_CURRENT) \ + { \ + TOKEN t = *tok++; \ + if (HIERARCHY_HIGHER() != OK) \ + return ERROR; \ + *evalTokenBuffer.u32++ = t; \ + } \ +}while (0) + int expr0(void) { - if (expr1() != OK) - return ERROR; - - while (tokenClass[*tok] >= MULT) + if ( correctMathRules == 0 ) { - TOKEN t = *tok++; - if (expr1() != OK) return ERROR; - *evalTokenBuffer.u32++ = t; + while (tokenClass[*tok] >= MULT) + { + TOKEN t = *tok++; + + if (expr1() != OK) + return ERROR; + + *evalTokenBuffer.u32++ = t; + } } + else + { + // The order of C precedence (lower to higher): + // bitwise XOR ^ + // bitwise OR | + // bitwise AND & + // relational = < <= >= > != + // shifts << >> + // sum + - + // product * / + precedence(xor, OR); + } + return OK; +} + +int xor(void) +{ + precedence(and, XOR); + return OK; +} + +int and(void) +{ + precedence(rel, AND); + return OK; +} +int rel(void) +{ + precedence(shift, REL); return OK; } +int shift(void) +{ + precedence(sum, SHIFT); + return OK; +} + +int sum(void) +{ + precedence(product, ADD); + return OK; +} + +int product(void) +{ + precedence(expr1, MULT); + return OK; +} // // Unary operators (detect unary '-') @@ -155,8 +224,53 @@ int expr1(void) switch (*tok++) { case CR_ABSCOUNT: + if (cursect != ABS) + { + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = sect[ABS].sloc; + } + else + { + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = sloc; + } + break; + case CR_FILESIZE: + if (*tok++ != STRING) + return error("^^FILESIZE expects filename inside string"); *evalTokenBuffer.u32++ = CONST; - *evalTokenBuffer.u64++ = (uint64_t)sect[ABS].sloc; + // @@copypasted from d_incbin, maybe factor this out somehow? + // Attempt to open the include file in the current directory, then (if that + // failed) try list of include files passed in the enviroment string or by + // the "-d" option. + int fd, i; + char buf1[256]; + + if ((fd = open(string[*tok], _OPEN_INC)) < 0) + { + for(i=0; nthpath("RMACPATH", i, buf1)!=0; i++) + { + fd = strlen(buf1); + + // Append path char if necessary + if ((fd > 0) && (buf1[fd - 1] != SLASHCHAR)) + strcat(buf1, SLASHSTRING); + + strcat(buf1, string[*tok]); + + if ((fd = open(buf1, _OPEN_INC)) >= 0) + goto allright; + } + + return error("cannot open: \"%s\"", string[*tok]); + } + +allright: + *evalTokenBuffer.u64++ = (uint64_t)lseek(fd, 0L, SEEK_END); + close(fd); + + // Advance tok because of consumed string token + tok++; break; case CR_TIME: *evalTokenBuffer.u32++ = CONST; @@ -219,7 +333,6 @@ getsym: return OK; } - // // Terminals (CONSTs) and parenthesis grouping // @@ -250,16 +363,6 @@ int expr2(void) if (sy == NULL) sy = NewSymbol(p, LABEL, j); - // Check register bank usage - if (sy->sattre & EQUATEDREG) - { - if ((regbank == BANK_0) && (sy->sattre & BANK_1) && !altbankok) - warn("equated symbol \'%s\' cannot be used in register bank 0", sy->sname); - - if ((regbank == BANK_1) && (sy->sattre & BANK_0) && !altbankok) - warn("equated symbol \'%s\' cannot be used in register bank 1", sy->sname); - } - *evalTokenBuffer.u32++ = SYMBOL; *evalTokenBuffer.u32++ = symbolNum; symbolPtr[symbolNum] = sy; @@ -303,7 +406,7 @@ int expr2(void) case '$': *evalTokenBuffer.u32++ = ACONST; // Attributed const *evalTokenBuffer.u32++ = sloc; // Current location - *evalTokenBuffer.u32++ = cursect | DEFINED; // Store attribs + *evalTokenBuffer.u32++ = DEFINED | ((orgactive | org68k_active) ? 0 : cursect); // Store attribs break; case '*': *evalTokenBuffer.u32++ = ACONST; // Attributed const @@ -311,7 +414,8 @@ int expr2(void) // pcloc == location at start of line *evalTokenBuffer.u32++ = (orgactive ? orgaddr : pcloc); // '*' takes attributes of current section, not ABS! - *evalTokenBuffer.u32++ = cursect | DEFINED; + // Also, if we're ORG'd, the symbol is absolute + *evalTokenBuffer.u32++ = DEFINED | ((orgactive | org68k_active) ? 0 : cursect); break; default: return error("bad expression"); @@ -320,7 +424,6 @@ int expr2(void) return OK; } - // // Recursive-descent expression analyzer (with some simple speed hacks) // @@ -336,49 +439,24 @@ int expr(TOKEN * otk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) PTR ptk; evalTokenBuffer.u32 = otk; // Set token pointer to 'exprbuf' (direct.c) - // Also set in various other places too (riscasm.c, - // e.g.) + // Also set in various other places too (riscasm.c, + // e.g.) -//printf("expr(): tokens 0-2: %i %i %i (%c %c %c); tc[2] = %i\n", tok[0], tok[1], tok[2], tok[0], tok[1], tok[2], tokenClass[tok[2]]); // Optimize for single constant or single symbol. - // Shamus: Subtle bug here. EOL token is 101; if you have a constant token - // followed by the value 101, it will trigger a bad evaluation here. - // This is probably a really bad assumption to be making here...! - // (assuming tok[1] == EOL is a single token that is) - // Seems that even other tokens (SUNARY type) can fuck this up too. -#if 0 -// if ((tok[1] == EOL) - if ((tok[1] == EOL && ((tok[0] != CONST || tok[0] != FCONST) && tokenClass[tok[0]] != SUNARY)) -// || (((*tok == CONST || *tok == FCONST || *tok == SYMBOL) || (*tok >= KW_R0 && *tok <= KW_R31)) -// && (tokenClass[tok[2]] < UNARY))) - || (((tok[0] == SYMBOL) || (tok[0] >= KW_R0 && tok[0] <= KW_R31)) - && (tokenClass[tok[2]] < UNARY)) - || ((tok[0] == CONST || tok[0] == FCONST) && (tokenClass[tok[3]] < UNARY)) - ) -#else -// Shamus: Seems to me that this could be greatly simplified by 1st checking if the first token is a multibyte token, *then* checking if there's an EOL after it depending on the actual length of the token (multiple vs. single). Otherwise, we have the horror show that is the following: + // Shamus: Seems to me that this could be greatly simplified by 1st + // checking if the first token is a multibyte token, *then* + // checking if there's an EOL after it depending on the actual + // length of the token (multiple vs. single). Otherwise, we have + // the horror show that is the following: if ((tok[1] == EOL && (tok[0] != CONST && tokenClass[tok[0]] != SUNARY)) - || (((tok[0] == SYMBOL) - || (tok[0] >= KW_R0 && tok[0] <= KW_R31)) + || ((tok[0] == SYMBOL) && (tokenClass[tok[2]] < UNARY)) || ((tok[0] == CONST) && (tokenClass[tok[3]] < UNARY)) ) -// Shamus: Yes, you can parse that out and make some kind of sense of it, but damn, it takes a while to get it and understand the subtle bugs that result from not being careful about what you're checking; especially vis-a-vis niavely checking tok[1] for an EOL. O_o -#endif +// Shamus: Yes, you can parse that out and make some kind of sense of it, but damn, it takes a while to get it and understand the subtle bugs that result from not being careful about what you're checking; especially vis-a-vis naively checking tok[1] for an EOL. O_o { - if (*tok >= KW_R0 && *tok <= KW_R31) - { - *evalTokenBuffer.u32++ = CONST; - *evalTokenBuffer.u64++ = *a_value = (*tok - KW_R0); - *a_attr = ABS | DEFINED; - - if (a_esym != NULL) - *a_esym = NULL; - - tok++; - } - else if (*tok == CONST) + if (*tok == CONST) { ptk.u32 = tok; *evalTokenBuffer.u32++ = *ptk.u32++; @@ -412,13 +490,18 @@ int expr(TOKEN * otk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) { *evalTokenBuffer.u32++ = CONST; - if (orgactive) + if (orgactive | org68k_active) + { *evalTokenBuffer.u64++ = *a_value = orgaddr; + *a_attr = DEFINED; // We have ORG active, it doesn't belong in a section! + } else + { *evalTokenBuffer.u64++ = *a_value = pcloc; + // '*' takes attributes of current section, not ABS! + *a_attr = cursect | DEFINED; + } - // '*' takes attributes of current section, not ABS! - *a_attr = cursect | DEFINED; if (a_esym != NULL) *a_esym = NULL; @@ -430,11 +513,6 @@ int expr(TOKEN * otk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) p = string[tok[1]]; j = (*p == '.' ? curenv : 0); symbol = lookup(p, LABEL, j); -#if 0 -printf("eval: Looking up symbol (%s) [=%08X]\n", p, symbol); -if (symbol) - printf(" attr=%04X, attre=%08X, val=%i, name=%s\n", symbol->sattr, symbol->sattre, symbol->svalue, symbol->sname); -#endif if (symbol == NULL) symbol = NewSymbol(p, LABEL, j); @@ -448,18 +526,6 @@ if (symbol) if ((symbol->sattre & UNDEF_EQUR) && !riscImmTokenSeen) { error("undefined register equate '%s'", symbol->sname); -//if we return right away, it returns some spurious errors... -// return ERROR; - } - - // Check register bank usage - if (symbol->sattre & EQUATEDREG) - { - if ((regbank == BANK_0) && (symbol->sattre & BANK_1) && !altbankok) - warn("equated symbol '%s' cannot be used in register bank 0", symbol->sname); - - if ((regbank == BANK_1) && (symbol->sattre & BANK_0) && !altbankok) - warn("equated symbol '%s' cannot be used in register bank 1", symbol->sname); } *evalTokenBuffer.u32++ = SYMBOL; @@ -476,19 +542,14 @@ be converted from a linked list into an array). symbolNum++; #endif - if (symbol->sattr & DEFINED) - *a_value = symbol->svalue; - else - *a_value = 0; + *a_value = (symbol->sattr & DEFINED ? symbol->svalue : 0); + *a_attr = (WORD)(symbol->sattr & ~GLOBAL); -/* -All that extra crap that was put into the svalue when doing the equr stuff is -thrown away right here. What the hell is it for? -*/ if (symbol->sattre & EQUATEDREG) - *a_value &= 0x1F; - - *a_attr = (WORD)(symbol->sattr & ~GLOBAL); + { + *a_attr |= RISCREG; // Mark it as a register, 'cause it is + *a_esym = symbol; + } if ((symbol->sattr & (GLOBAL | DEFINED)) == GLOBAL && a_esym != NULL) @@ -521,11 +582,10 @@ thrown away right here. What the hell is it for? return evexpr(otk, a_value, a_attr, a_esym); } - // // Evaluate expression. // If the expression involves only ONE external symbol, the expression is -// UNDEFINED, but it's value includes everything but the symbol value, and +// UNDEFINED, but its value includes everything but the symbol value, and // 'a_esym' is set to the external symbol. // int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) @@ -544,7 +604,6 @@ int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) switch ((int)*tk.u32++) { case SYMBOL: -//printf("evexpr(): SYMBOL\n"); sy = symbolPtr[*tk.u32++]; sy->sattr |= REFERENCED; // Set "referenced" bit @@ -575,12 +634,10 @@ int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) case CONST: *++sval = *tk.u64++; -//printf("evexpr(): CONST = %lX\n", *sval); *++sattr = ABS | DEFINED; // Push simple attribs break; case FCONST: -//printf("evexpr(): FCONST = %lf\n", *tk.dp); // Even though it's a double, we can treat it like a uint64_t since // we're just moving the bits around. *++sval = *tk.u64++; @@ -588,7 +645,6 @@ int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) break; case ACONST: -//printf("evexpr(): ACONST = %i\n", *tk.u32); *++sval = *tk.u32++; // Push value *++sattr = (WORD)*tk.u32++; // Push attribs break; @@ -606,10 +662,8 @@ int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) // - : ABS case '+': -//printf("evexpr(): +\n"); --sval; // Pop value --sattr; // Pop attrib -//printf("--> N+N: %i + %i = ", *sval, sval[1]); // Get FLOAT attribute, if any attr = (sattr[0] | sattr[1]) & FLOAT; @@ -629,7 +683,6 @@ int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) { *sval += sval[1]; // Compute value } -//printf("%i\n", *sval); if (!(*sattr & TDB)) *sattr = sattr[1] | attr; @@ -639,10 +692,8 @@ int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) break; case '-': -//printf("evexpr(): -\n"); --sval; // Pop value --sattr; // Pop attrib -//printf("--> N-N: %i - %i = ", *sval, sval[1]); // Get FLOAT attribute, if any attr = (sattr[0] | sattr[1]) & FLOAT; @@ -662,13 +713,9 @@ int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) { *sval -= sval[1]; } -//printf("%i\n", *sval); *sattr |= attr; // Inherit FLOAT attribute attr = (WORD)(*sattr & TDB); -#if 0 -printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); -#endif // If symbol1 is ABS, take attributes from symbol2 if (!attr) *sattr = sattr[1]; @@ -680,7 +727,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); // Unary operators only work on ABS items case UNMINUS: -//printf("evexpr(): UNMINUS\n"); if (*sattr & TDB) return error(seg_error); @@ -699,7 +745,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case UNLT: // Unary < (get the low byte of a word) -//printf("evexpr(): UNLT\n"); if (*sattr & TDB) return error(seg_error); @@ -711,7 +756,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case UNGT: // Unary > (get the high byte of a word) -//printf("evexpr(): UNGT\n"); if (*sattr & TDB) return error(seg_error); @@ -723,7 +767,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case '!': -//printf("evexpr(): !\n"); if (*sattr & TDB) return error(seg_error); @@ -735,7 +778,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case '~': -//printf("evexpr(): ~\n"); if (*sattr & TDB) return error(seg_error); @@ -749,7 +791,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); // Comparison operators must have two values that // are in the same segment, but that's the only requirement. case LE: -//printf("evexpr(): LE\n"); sattr--; sval--; @@ -779,7 +820,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case GE: -//printf("evexpr(): GE\n"); sattr--; sval--; @@ -809,7 +849,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case '>': -//printf("evexpr(): >\n"); sattr--; sval--; @@ -839,7 +878,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case '<': -//printf("evexpr(): <\n"); sattr--; sval--; @@ -869,7 +907,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case NE: -//printf("evexpr(): NE\n"); sattr--; sval--; @@ -899,7 +936,6 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); break; case '=': -//printf("evexpr(): =\n"); sattr--; sval--; @@ -934,14 +970,12 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); // Shamus: Is this true? There's at least one counterexample of legit // code where this assumption fails to produce correct code. default: -//printf("evexpr(): default\n"); switch ((int)tk.u32[-1]) { case '*': sval--; sattr--; -//printf("--> NxN: %i x %i = ", *sval, sval[1]); // Get FLOAT attribute, if any attr = (sattr[0] | sattr[1]) & FLOAT; @@ -964,15 +998,12 @@ An open question here is do we promote ints to floats as signed or unsigned? It { *sval *= sval[1]; } -//printf("%i\n", *sval); -//no *sattr = ABS | DEFINED | attr; // Expr becomes absolute break; case '/': sval--; sattr--; -//printf("--> N/N: %i / %i = ", sval[0], sval[1]); // Get FLOAT attribute, if any attr = (sattr[0] | sattr[1]) & FLOAT; @@ -993,7 +1024,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It { if (sval[1] == 0) return error("divide by zero"); -//printf("--> N/N: %i / %i = ", sval[0], sval[1]); // Compiler is picky here: Without casting these, it // discards the sign if dividing a negative # by a @@ -1002,9 +1032,7 @@ An open question here is do we promote ints to floats as signed or unsigned? It // ints. *sval = (int32_t)sval[0] / (int32_t)sval[1]; } -//printf("%i\n", *sval); -//no *sattr = ABS | DEFINED | attr; // Expr becomes absolute break; case '%': @@ -1018,7 +1046,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It return error("mod (%) by zero"); *sval %= sval[1]; -//no *sattr = ABS | DEFINED; // Expr becomes absolute break; case SHL: @@ -1029,7 +1056,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It return error("floating point numbers not allowed with operator '<<'."); *sval <<= sval[1]; -//no *sattr = ABS | DEFINED; // Expr becomes absolute break; case SHR: @@ -1040,7 +1066,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It return error("floating point numbers not allowed with operator '>>'."); *sval >>= sval[1]; -//no *sattr = ABS | DEFINED; // Expr becomes absolute break; case '&': @@ -1051,7 +1076,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It return error("floating point numbers not allowed with operator '&'."); *sval &= sval[1]; -//no *sattr = ABS | DEFINED; // Expr becomes absolute break; case '^': @@ -1062,7 +1086,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It return error("floating point numbers not allowed with operator '^'."); *sval ^= sval[1]; -//no *sattr = ABS | DEFINED; // Expr becomes absolute break; case '|': @@ -1073,7 +1096,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It return error("floating point numbers not allowed with operator '|'."); *sval |= sval[1]; -//no *sattr = ABS | DEFINED; // Expr becomes absolute break; default: @@ -1096,7 +1118,6 @@ An open question here is do we promote ints to floats as signed or unsigned? It return OK; } - // // Count the # of tokens in the passed in expression // N.B.: 64-bit constants count as two tokens each @@ -1119,4 +1140,3 @@ uint16_t ExpressionLength(TOKEN * tk) return length; } -