X-Git-Url: http://shamusworld.gotdns.org/cgi-bin/gitweb.cgi?p=rmac;a=blobdiff_plain;f=expr.c;h=f31e9e59ddf67c1be317af23a1428c22da4ee534;hp=44fd892bab2d0594170dd834741b411dc9498284;hb=HEAD;hpb=03dd34951a331e0b8971195ccef1600fffaea2e6 diff --git a/expr.c b/expr.c index 44fd892..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, 2011-2017 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 // @@ -23,20 +23,20 @@ // N.B.: The size of tokenClass should be identical to the largest value of // a token; we're assuming 256 but not 100% sure! static char tokenClass[256]; // Generated table of token classes -static VALUE evstk[EVSTACKSIZE]; // Evaluator value stack +static uint64_t evstk[EVSTACKSIZE]; // Evaluator value stack static WORD evattr[EVSTACKSIZE]; // Evaluator attribute stack // Token-class initialization list char itokcl[] = { 0, // END - CONST, SYMBOL, 0, // ID + CONST, FCONST, SYMBOL, 0, // ID '(', '[', '{', 0, // OPAR ')', ']', '}', 0, // CPAR CR_DEFINED, CR_REFERENCED, // SUNARY (special unary) CR_STREQ, CR_MACDEF, CR_DATE, CR_TIME, - CR_ABSCOUNT, 0, - '!', '~', UNMINUS, 0, // UNARY + CR_ABSCOUNT, CR_FILESIZE, 0, + '!', '~', UNMINUS, UNLT, UNGT, 0, // UNARY '*', '/', '%', 0, // MULT '+', '-', 0, // ADD SHL, SHR, 0, // SHIFT @@ -49,21 +49,21 @@ char itokcl[] = { const char missym_error[] = "missing symbol"; const char str_error[] = "missing symbol or string"; +const char noflt_error[] = "operator not usable with float"; // Convert expression to postfix -static TOKEN * evalTokenBuffer; // Deposit tokens here (this is really a - // pointer to exprbuf from direct.c) - // (Can also be from others, like - // riscasm.c) -static int symbolNum; // Pointer to the entry in symbolPtr[] - +static PTR evalTokenBuffer; // Deposit tokens here (this is really a + // pointer to exprbuf from direct.c) + // (Can also be from others, like + // riscasm.c) +static int symbolNum; // Pointer to the entry in symbolPtr[] // // Obtain a string value // -static VALUE str_value(char * p) +static uint32_t str_value(char * p) { - VALUE v; + uint32_t v; for(v=0; *p; p++) v = (v << 8) | (*p & 0xFF); @@ -71,7 +71,6 @@ static VALUE str_value(char * p) return v; } - // // Initialize expression analyzer // @@ -94,84 +93,201 @@ 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) { - TOKEN t; - - if (expr1() != OK) - return ERROR; - - while (tokenClass[*tok] >= MULT) + if ( correctMathRules == 0 ) { - t = *tok++; - if (expr1() != OK) return ERROR; - *evalTokenBuffer++ = 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 '-') -// ggn: If expression starts with a plus then also eat it up. -// For some reason the parser gets confused when this happens and -// emits a "bad expression". +// ggn: If expression starts with a plus then also eat it up. For some reason +// the parser gets confused when this happens and emits a "bad +// expression". // int expr1(void) { - int class; - TOKEN t; - SYM * sy; - char * p, * p2; + char * p; WORD w; - int j; - class = tokenClass[*tok]; + int class = tokenClass[*tok]; - if (*tok == '-' || *tok == '+' || class == UNARY) + if (*tok == '-' || *tok == '+' || *tok == '<' || *tok == '>' || class == UNARY) { - t = *tok++; + TOKEN t = *tok++; if (expr2() != OK) return ERROR; if (t == '-') t = UNMINUS; - - *evalTokenBuffer++ = t; + else if (t == '<') + t = UNLT; + else if (t == '>') + t = UNGT; + + // With leading + we don't have to deposit anything to the buffer + // because there's no unary '+' nor we have to do anything about it + if (t != '+') + *evalTokenBuffer.u32++ = t; } else if (class == SUNARY) { - switch ((int)*tok++) + switch (*tok++) { case CR_ABSCOUNT: - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = (LONG)sect[ABS].sloc; + 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; + // @@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++ = CONST; - *evalTokenBuffer++ = dos_time(); + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = dos_time(); break; case CR_DATE: - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = dos_date(); + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = dos_date(); break; - case CR_MACDEF: // ^^macdef + case CR_MACDEF: // ^^macdef if (*tok++ != SYMBOL) return error(missym_error); p = string[*tok++]; w = (lookup(p, MACRO, 0) == NULL ? 0 : 1); - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = (TOKEN)w; + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = (uint64_t)w; break; case CR_DEFINED: w = DEFINED; @@ -183,10 +299,11 @@ getsym: return error(missym_error); p = string[*tok++]; - j = (*p == '.' ? curenv : 0); - w = ((sy = lookup(p, LABEL, j)) != NULL && (sy->sattr & w) ? 1 : 0); - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = (TOKEN)w; + int j = (*p == '.' ? curenv : 0); + SYM * sy = lookup(p, LABEL, j); + w = ((sy != NULL) && (sy->sattr & w ? 1 : 0)); + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = (uint64_t)w; break; case CR_STREQ: if (*tok != SYMBOL && *tok != STRING) @@ -201,12 +318,12 @@ getsym: if (*tok != SYMBOL && *tok != STRING) return error(str_error); - p2 = string[tok[1]]; + char * p2 = string[tok[1]]; tok += 2; w = (WORD)(!strcmp(p, p2)); - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = (TOKEN)w; + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = (uint64_t)w; break; } } @@ -216,55 +333,52 @@ getsym: return OK; } - // // Terminals (CONSTs) and parenthesis grouping // int expr2(void) { - char * p; - SYM * sy; - int j; + PTR ptk; - switch ((int)*tok++) + switch (*tok++) { case CONST: - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = *tok++; + ptk.u32 = tok; + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = *ptk.u64++; + tok = ptk.u32; + break; + case FCONST: + ptk.u32 = tok; + *evalTokenBuffer.u32++ = FCONST; + *evalTokenBuffer.u64++ = *ptk.u64++; + tok = ptk.u32; break; case SYMBOL: - p = string[*tok++]; - j = (*p == '.' ? curenv : 0); - sy = lookup(p, LABEL, j); + { + char * p = string[*tok++]; + int j = (*p == '.' ? curenv : 0); + SYM * sy = lookup(p, LABEL, j); 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) - warns("equated symbol \'%s\' cannot be used in register bank 0", sy->sname); - - if ((regbank == BANK_1) && (sy->sattre & BANK_0) && !altbankok) - warns("equated symbol \'%s\' cannot be used in register bank 1", sy->sname); - } - - *evalTokenBuffer++ = SYMBOL; - *evalTokenBuffer++ = symbolNum; + *evalTokenBuffer.u32++ = SYMBOL; + *evalTokenBuffer.u32++ = symbolNum; symbolPtr[symbolNum] = sy; symbolNum++; break; + } case STRING: - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = str_value(string[*tok++]); + *evalTokenBuffer.u32++ = CONST; + *evalTokenBuffer.u64++ = str_value(string[*tok++]); break; case '(': if (expr0() != OK) return ERROR; if (*tok++ != ')') - return error("missing close parenthesis ')'"); + return error("missing closing parenthesis ')'"); break; case '[': @@ -272,35 +386,36 @@ int expr2(void) return ERROR; if (*tok++ != ']') - return error("missing close parenthesis ']'"); + return error("missing closing bracket ']'"); - break; - case '$': - *evalTokenBuffer++ = ACONST; // Attributed const - *evalTokenBuffer++ = sloc; // Current location - *evalTokenBuffer++ = cursect | DEFINED; // Store attribs - break; - case '*': - *evalTokenBuffer++ = ACONST; // Attributed const - - // pcloc == location at start of line - *evalTokenBuffer++ = (orgactive ? orgaddr : pcloc); - // '*' takes attributes of current section, not ABS! - *evalTokenBuffer++ = cursect | DEFINED; break; case '{': - if (expr0() != OK) // Eat up first parameter (register or immediate) + if (expr0() != OK) // Eat up first parameter (register or immediate) return ERROR; - if (*tok++ != ':') // Demand a ':' there + if (*tok++ != ':') // Demand a ':' there return error("missing colon ':'"); - if (expr0() != OK) // Eat up second parameter (register or immediate) + if (expr0() != OK) // Eat up second parameter (register or immediate) return ERROR; if (*tok++ != '}') - return error("missing close bracket '}'"); + return error("missing closing brace '}'"); + break; + case '$': + *evalTokenBuffer.u32++ = ACONST; // Attributed const + *evalTokenBuffer.u32++ = sloc; // Current location + *evalTokenBuffer.u32++ = DEFINED | ((orgactive | org68k_active) ? 0 : cursect); // Store attribs + break; + case '*': + *evalTokenBuffer.u32++ = ACONST; // Attributed const + + // pcloc == location at start of line + *evalTokenBuffer.u32++ = (orgactive ? orgaddr : pcloc); + // '*' takes attributes of current section, not ABS! + // Also, if we're ORG'd, the symbol is absolute + *evalTokenBuffer.u32++ = DEFINED | ((orgactive | org68k_active) ? 0 : cursect); break; default: return error("bad expression"); @@ -309,11 +424,10 @@ int expr2(void) return OK; } - // // Recursive-descent expression analyzer (with some simple speed hacks) // -int expr(TOKEN * otk, VALUE * a_value, WORD * a_attr, SYM ** a_esym) +int expr(TOKEN * otk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) { // Passed in values (once derefenced, that is) can all be zero. They are // there so that the expression analyzer can fill them in as needed. The @@ -322,56 +436,72 @@ int expr(TOKEN * otk, VALUE * a_value, WORD * a_attr, SYM ** a_esym) SYM * symbol; char * p; int j; + PTR ptk; - evalTokenBuffer = otk; // Set token pointer to 'exprbuf' (direct.c) - // Also set in various other places too (riscasm.c, e.g.) + evalTokenBuffer.u32 = otk; // Set token pointer to 'exprbuf' (direct.c) + // 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 ((tok[1] == EOL) - if ((tok[1] == EOL && (tok[0] != CONST && tokenClass[tok[0]] != SUNARY)) - || (((*tok == CONST || *tok == SYMBOL) || (*tok >= KW_R0 && *tok <= KW_R31)) - && (tokenClass[tok[2]] < UNARY))) + // 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) + && (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 naively checking tok[1] for an EOL. O_o { - if (*tok >= KW_R0 && *tok <= KW_R31) + if (*tok == CONST) { - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = *a_value = (*tok - KW_R0); + ptk.u32 = tok; + *evalTokenBuffer.u32++ = *ptk.u32++; + *evalTokenBuffer.u64++ = *a_value = *ptk.u64++; *a_attr = ABS | DEFINED; + tok = ptk.u32; if (a_esym != NULL) *a_esym = NULL; - tok++; +//printf("Quick eval in expr(): CONST = %i, tokenClass[tok[2]] = %i\n", *a_value, tokenClass[*tok]); } - else if (*tok == CONST) +// Not sure that removing float constant here is going to break anything and/or +// make things significantly slower, but having this here seems to cause the +// complexity of the check to get to this part of the parse to go through the +// roof, and dammit, I just don't feel like fighting that fight ATM. :-P +#if 0 + else if (*tok == FCONST) { - *evalTokenBuffer++ = CONST; - *evalTokenBuffer++ = *a_value = tok[1]; - *a_attr = ABS | DEFINED; + *evalTokenBuffer.u32++ = *tok++; + *evalTokenBuffer.u64++ = *a_value = *tok.u64++; + *a_attr = ABS | DEFINED | FLOAT; if (a_esym != NULL) *a_esym = NULL; - tok += 2; //printf("Quick eval in expr(): CONST = %i, tokenClass[tok[2]] = %i\n", *a_value, tokenClass[*tok]); } +#endif else if (*tok == '*') { - *evalTokenBuffer++ = CONST; + *evalTokenBuffer.u32++ = CONST; - if (orgactive) - *evalTokenBuffer++ = *a_value = orgaddr; + 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++ = *a_value = pcloc; + { + *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; @@ -383,11 +513,6 @@ int expr(TOKEN * otk, VALUE * 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); @@ -400,22 +525,10 @@ if (symbol) // means it will be fixed up later, and thus, not an error. if ((symbol->sattre & UNDEF_EQUR) && !riscImmTokenSeen) { - errors("undefined register equate '%s'", symbol->sname); -//if we return right away, it returns some spurious errors... -// return ERROR; + error("undefined register equate '%s'", symbol->sname); } - // Check register bank usage - if (symbol->sattre & EQUATEDREG) - { - if ((regbank == BANK_0) && (symbol->sattre & BANK_1) && !altbankok) - warns("equated symbol '%s' cannot be used in register bank 0", symbol->sname); - - if ((regbank == BANK_1) && (symbol->sattre & BANK_0) && !altbankok) - warns("equated symbol '%s' cannot be used in register bank 1", symbol->sname); - } - - *evalTokenBuffer++ = SYMBOL; + *evalTokenBuffer.u32++ = SYMBOL; #if 0 *evalTokenBuffer++ = (TOKEN)symbol; #else @@ -424,24 +537,19 @@ While this approach works, it's wasteful. It would be better to use something that's already available, like the symbol "order defined" table (which needs to be converted from a linked list into an array). */ - *evalTokenBuffer++ = symbolNum; + *evalTokenBuffer.u32++ = symbolNum; symbolPtr[symbolNum] = symbol; 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) @@ -449,49 +557,54 @@ thrown away right here. What the hell is it for? tok += 2; } + // Holy hell... This is likely due to the fact that LSR is mistakenly set as a SUNARY type... Need to fix this... !!! FIX !!! + else if (m6502) + { + *evalTokenBuffer.u32++ = *tok++; + } else { // Unknown type here... Alert the user!, - error("undefined RISC register in expression"); + error("undefined RISC register in expression [token=$%X]", *tok); // Prevent spurious error reporting... tok++; return ERROR; } - *evalTokenBuffer++ = ENDEXPR; + *evalTokenBuffer.u32++ = ENDEXPR; return OK; } if (expr0() != OK) return ERROR; - *evalTokenBuffer++ = ENDEXPR; + *evalTokenBuffer.u32++ = ENDEXPR; 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 -// `a_esym' is set to the external symbol. +// UNDEFINED, but its value includes everything but the symbol value, and +// 'a_esym' is set to the external symbol. // -int evexpr(TOKEN * tk, VALUE * a_value, WORD * a_attr, SYM ** a_esym) +int evexpr(TOKEN * _tk, uint64_t * a_value, WORD * a_attr, SYM ** a_esym) { WORD attr; SYM * sy; - VALUE * sval = evstk; // (Empty) initial stack + uint64_t * sval = evstk; // (Empty) initial stack WORD * sattr = evattr; SYM * esym = NULL; // No external symbol involved WORD sym_seg = 0; + PTR tk; + tk.u32 = _tk; - while (*tk != ENDEXPR) + while (*tk.u32 != ENDEXPR) { - switch ((int)*tk++) + switch ((int)*tk.u32++) { case SYMBOL: -//printf("evexpr(): SYMBOL\n"); - sy = symbolPtr[*tk++]; + sy = symbolPtr[*tk.u32++]; sy->sattr |= REFERENCED; // Set "referenced" bit if (!(sy->sattr & DEFINED)) @@ -511,26 +624,29 @@ int evexpr(TOKEN * tk, VALUE * a_value, WORD * a_attr, SYM ** a_esym) } if (sy->sattr & DEFINED) - { *++sval = sy->svalue; // Push symbol's value - } else - { *++sval = 0; // 0 for undefined symbols - } *++sattr = (WORD)(sy->sattr & ~GLOBAL); // Push attribs sym_seg = (WORD)(sy->sattr & TDB); break; + case CONST: -//printf("evexpr(): CONST = %i\n", *tk); - *++sval = *tk++; // Push value + *++sval = *tk.u64++; *++sattr = ABS | DEFINED; // Push simple attribs break; + + case FCONST: + // 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++; + *++sattr = ABS | DEFINED | FLOAT; // Push simple attribs + break; + case ACONST: -//printf("evexpr(): ACONST = %i\n", *tk); - *++sval = *tk++; // Push value - *++sattr = (WORD)*tk++; // Push attribs + *++sval = *tk.u32++; // Push value + *++sattr = (WORD)*tk.u32++; // Push attribs break; // Binary "+" and "-" matrix: @@ -544,32 +660,62 @@ int evexpr(TOKEN * tk, VALUE * a_value, WORD * a_attr, SYM ** a_esym) // // [1] + : Error // - : ABS + case '+': -//printf("evexpr(): +\n"); --sval; // Pop value --sattr; // Pop attrib -//printf("--> N+N: %i + %i = ", *sval, sval[1]); - *sval += sval[1]; // Compute value -//printf("%i\n", *sval); + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Since adding an int to a fp value promotes it to a fp value, we + // don't care whether it's first or second; we cast to to a double + // regardless. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *(double *)sval = fpval1 + fpval2; + } + else + { + *sval += sval[1]; // Compute value + } if (!(*sattr & TDB)) - *sattr = sattr[1]; + *sattr = sattr[1] | attr; else if (sattr[1] & TDB) return error(seg_error); break; + case '-': -//printf("evexpr(): -\n"); --sval; // Pop value --sattr; // Pop attrib -//printf("--> N-N: %i - %i = ", *sval, sval[1]); - *sval -= sval[1]; // Compute value -//printf("%i\n", *sval); + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + // Since subtracting an int to a fp value promotes it to a fp + // value, we don't care whether it's first or second; we cast to to + // a double regardless. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *(double *)sval = fpval1 - fpval2; + } + else + { + *sval -= sval[1]; + } + + *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]; @@ -578,168 +724,383 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); *sattr &= ~TDB; break; + // Unary operators only work on ABS items case UNMINUS: -//printf("evexpr(): UNMINUS\n"); if (*sattr & TDB) - error(seg_error); + return error(seg_error); + + if (*sattr & FLOAT) + { + double * dst = (double *)sval; + *dst = -*dst; + *sattr = ABS | DEFINED | FLOAT; // Expr becomes absolute + } + else + { + *sval = -(int64_t)*sval; + *sattr = ABS | DEFINED; // Expr becomes absolute + } + + break; + + case UNLT: // Unary < (get the low byte of a word) + if (*sattr & TDB) + return error(seg_error); + + if (*sattr & FLOAT) + return error(noflt_error); + + *sval = (int64_t)((*sval) & 0x00FF); + *sattr = ABS | DEFINED; // Expr becomes absolute + break; + + case UNGT: // Unary > (get the high byte of a word) + if (*sattr & TDB) + return error(seg_error); - *sval = -(int)*sval; + if (*sattr & FLOAT) + return error(noflt_error); + + *sval = (int64_t)(((*sval) >> 8) & 0x00FF); *sattr = ABS | DEFINED; // Expr becomes absolute break; + case '!': -//printf("evexpr(): !\n"); if (*sattr & TDB) - error(seg_error); + return error(seg_error); + + if (*sattr & FLOAT) + return error("floating point numbers not allowed with operator '!'."); *sval = !*sval; *sattr = ABS | DEFINED; // Expr becomes absolute break; + case '~': -//printf("evexpr(): ~\n"); if (*sattr & TDB) - error(seg_error); + return error(seg_error); + + if (*sattr & FLOAT) + return error("floating point numbers not allowed with operator '~'."); *sval = ~*sval; *sattr = ABS | DEFINED; // Expr becomes absolute break; + // 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--; if ((*sattr & TDB) != (sattr[1] & TDB)) - error(seg_error); + return error(seg_error); + + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Cast any ints in the comparison to double, if there's at least + // one double in the comparison. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *sval = (fpval1 <= fpval2); + } + else + { + *sval = (*sval <= sval[1]); + } *sattr = ABS | DEFINED; - *sval = *sval <= sval[1]; break; + case GE: -//printf("evexpr(): GE\n"); sattr--; sval--; if ((*sattr & TDB) != (sattr[1] & TDB)) - error(seg_error); + return error(seg_error); + + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Cast any ints in the comparison to double, if there's at least + // one double in the comparison. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *sval = (fpval1 >= fpval2); + } + else + { + *sval = (*sval >= sval[1]); + } *sattr = ABS | DEFINED; - *sval = *sval >= sval[1]; break; + case '>': -//printf("evexpr(): >\n"); sattr--; sval--; if ((*sattr & TDB) != (sattr[1] & TDB)) - error(seg_error); + return error(seg_error); + + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Cast any ints in the comparison to double, if there's at least + // one double in the comparison. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *sval = (fpval1 > fpval2); + } + else + { + *sval = (*sval > sval[1]); + } *sattr = ABS | DEFINED; - *sval = *sval > sval[1]; break; + case '<': -//printf("evexpr(): <\n"); sattr--; sval--; if ((*sattr & TDB) != (sattr[1] & TDB)) - error(seg_error); + return error(seg_error); - *sattr = ABS | DEFINED; - *sval = *sval < sval[1]; + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Cast any ints in the comparison to double, if there's at least + // one double in the comparison. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *sval = (fpval1 < fpval2); + } + else + { + *sval = (*sval < sval[1]); + } + + *sattr = ABS | DEFINED; // Expr forced to ABS break; + case NE: -//printf("evexpr(): NE\n"); sattr--; sval--; if ((*sattr & TDB) != (sattr[1] & TDB)) - error(seg_error); + return error(seg_error); - *sattr = ABS | DEFINED; - *sval = *sval != sval[1]; + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Cast any ints in the comparison to double, if there's at least + // one double in the comparison. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *sval = (fpval1 != fpval2); + } + else + { + *sval = (*sval != sval[1]); + } + + *sattr = ABS | DEFINED; // Expr forced to ABS break; + case '=': -//printf("evexpr(): =\n"); sattr--; sval--; if ((*sattr & TDB) != (sattr[1] & TDB)) - error(seg_error); + return error(seg_error); + + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Cast any ints in the comparison to double, if there's at least + // one double in the comparison. + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *sval = (fpval1 == fpval2); + } + else + { + *sval = (*sval == sval[1]); + } + + *sattr = ABS | DEFINED; // Expr forced to ABS - *sattr = ABS | DEFINED; - *sval = *sval == sval[1]; break; - // All other binary operators must have two ABS items - // to work with. They all produce an ABS value. + + // All other binary operators must have two ABS items to work with. + // They all produce an ABS value. + // 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"); - // GH - Removed for v1.0.15 as part of the fix for indexed loads. - //if ((*sattr & (TEXT|DATA|BSS)) || (*--sattr & (TEXT|DATA|BSS))) - //error(seg_error); - *sattr = ABS | DEFINED; // Expr becomes absolute - switch ((int)tk[-1]) + switch ((int)tk.u32[-1]) { case '*': sval--; - sattr--; // Pop attrib -//printf("--> NxN: %i x %i = ", *sval, sval[1]); - *sval *= sval[1]; -//printf("%i\n", *sval); + sattr--; + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + // Since multiplying an int to a fp value promotes it to a fp + // value, we don't care whether it's first or second; it will + // be cast to a double regardless. +/* +An open question here is do we promote ints to floats as signed or unsigned? It makes a difference if, say, the int is put in as -1 but is promoted to a double as $FFFFFFFFFFFFFFFF--you get very different results that way! For now, we promote as signed until proven detrimental otherwise. +*/ + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + *(double *)sval = fpval1 * fpval2; + } + else + { + *sval *= sval[1]; + } + break; + case '/': sval--; - sattr--; // Pop attrib + sattr--; + // Get FLOAT attribute, if any + attr = (sattr[0] | sattr[1]) & FLOAT; + + if (attr == FLOAT) + { + PTR p; + p.u64 = sval; + double fpval1 = (sattr[0] & FLOAT ? *p.dp : (double)*p.i64); + p.u64++; + double fpval2 = (sattr[1] & FLOAT ? *p.dp : (double)*p.i64); + + if (fpval2 == 0) + return error("divide by zero"); + + *(double *)sval = fpval1 / fpval2; + } + else + { + if (sval[1] == 0) + return error("divide by zero"); + + // Compiler is picky here: Without casting these, it + // discards the sign if dividing a negative # by a + // positive one, creating a bad result. :-/ + // Definitely a side effect of using uint32_ts intead of + // ints. + *sval = (int32_t)sval[0] / (int32_t)sval[1]; + } - 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 positive one, - // creating a bad result. :-/ - // Probably a side effect of using VALUE intead of ints. - *sval = (int)sval[0] / (int)sval[1]; -//printf("%i\n", *sval); break; + case '%': sval--; - sattr--; // Pop attrib + sattr--; + + if ((*sattr | sattr[1]) & FLOAT) + return error("floating point numbers not allowed with operator '%'."); if (sval[1] == 0) return error("mod (%) by zero"); *sval %= sval[1]; break; + case SHL: sval--; sattr--; // Pop attrib + + if ((*sattr | sattr[1]) & FLOAT) + return error("floating point numbers not allowed with operator '<<'."); + *sval <<= sval[1]; break; + case SHR: sval--; sattr--; // Pop attrib + + if ((*sattr | sattr[1]) & FLOAT) + return error("floating point numbers not allowed with operator '>>'."); + *sval >>= sval[1]; break; + case '&': sval--; sattr--; // Pop attrib + + if ((*sattr | sattr[1]) & FLOAT) + return error("floating point numbers not allowed with operator '&'."); + *sval &= sval[1]; break; + case '^': sval--; sattr--; // Pop attrib + + if ((*sattr | sattr[1]) & FLOAT) + return error("floating point numbers not allowed with operator '^'."); + *sval ^= sval[1]; break; + case '|': sval--; - sattr--; // Pop attrib + sattr--; + + if ((*sattr | sattr[1]) & FLOAT) + return error("floating point numbers not allowed with operator '|'."); + *sval |= sval[1]; break; + default: - interror(5); // Bad operator in expression stream + // Bad operator in expression stream (this should never happen!) + interror(5); } } } @@ -750,14 +1111,32 @@ printf("EVEXPR (-): sym1 = %X, sym2 = %X\n", attr, sattr[1]); if (a_esym != NULL) *a_esym = esym; - // sym_seg added in 1.0.16 to solve a problem with forward symbols in - // expressions where absolute values also existed. The absolutes were - // overiding the symbol segments and not being included :( - //*a_attr = *sattr | sym_seg; // Copy value + attrib - - *a_attr = *sattr; // Copy value + attrib + // Copy value + attrib into return variables *a_value = *sval; + *a_attr = *sattr; return OK; } +// +// Count the # of tokens in the passed in expression +// N.B.: 64-bit constants count as two tokens each +// +uint16_t ExpressionLength(TOKEN * tk) +{ + uint16_t length; + + for(length=0; tk[length]!=ENDEXPR; length++) + { + // Add one to length for 2X tokens, two for 3X tokens + if (tk[length] == SYMBOL) + length++; + else if ((tk[length] == CONST) || (tk[length] == FCONST)) + length += 2; + } + + // Add 1 for ENDEXPR + length++; + + return length; +}