X-Git-Url: https://git.xonotic.org/?p=xonotic%2Fgmqcc.git;a=blobdiff_plain;f=ast.h;h=16f980709d518f3eddf66b7120b90d2f71b928d6;hp=ec46c9f8b710ab6c20044f273d31a6f0de147e8b;hb=091173341eace2d290f6c0ebc4159c3afca7fbaa;hpb=69173876f11da6ea695f7d518d5036ddcf5179a6 diff --git a/ast.h b/ast.h index ec46c9f..16f9807 100644 --- a/ast.h +++ b/ast.h @@ -1,5 +1,5 @@ /* - * Copyright (C) 2012 + * Copyright (C) 2012 * Wolfgang Bumiller * * Permission is hereby granted, free of charge, to any person obtaining a copy of @@ -20,10 +20,8 @@ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ -#ifndef GMQCC_AST_H__ -#define GMQCC_AST_H__ - -#include "astir.h" +#ifndef GMQCC_AST_HDR +#define GMQCC_AST_HDR #include "ir.h" /* Note: I will not be using a _t suffix for the @@ -33,20 +31,54 @@ typedef union ast_node_u ast_node; typedef union ast_expression_u ast_expression; -typedef struct ast_value_s ast_value; -typedef struct ast_function_s ast_function; -typedef struct ast_block_s ast_block; -typedef struct ast_binary_s ast_binary; -typedef struct ast_store_s ast_store; +typedef struct ast_value_s ast_value; +typedef struct ast_function_s ast_function; +typedef struct ast_block_s ast_block; +typedef struct ast_binary_s ast_binary; +typedef struct ast_store_s ast_store; +typedef struct ast_binstore_s ast_binstore; +typedef struct ast_entfield_s ast_entfield; +typedef struct ast_ifthen_s ast_ifthen; +typedef struct ast_ternary_s ast_ternary; +typedef struct ast_loop_s ast_loop; +typedef struct ast_call_s ast_call; +typedef struct ast_unary_s ast_unary; +typedef struct ast_return_s ast_return; +typedef struct ast_member_s ast_member; +typedef struct ast_array_index_s ast_array_index; + +enum { + TYPE_ast_node, + TYPE_ast_expression, + TYPE_ast_value, + TYPE_ast_function, + TYPE_ast_block, + TYPE_ast_binary, + TYPE_ast_store, + TYPE_ast_binstore, + TYPE_ast_entfield, + TYPE_ast_ifthen, + TYPE_ast_ternary, + TYPE_ast_loop, + TYPE_ast_call, + TYPE_ast_unary, + TYPE_ast_return, + TYPE_ast_member, + TYPE_ast_array_index +}; + +#define ast_istype(x, t) ( ((ast_node_common*)x)->nodetype == (TYPE_##t) ) +#define ast_ctx(node) (((ast_node_common*)(node))->context) /* Node interface with common components */ typedef void ast_node_delete(ast_node*); typedef struct { - lex_ctx_t context; + lex_ctx context; /* I don't feel comfortable using keywords like 'delete' as names... */ ast_node_delete *destroy; + int nodetype; /* keep: if a node contains this node, 'keep' * prevents its dtor from destroying this node as well. */ @@ -68,11 +100,35 @@ typedef struct */ typedef bool ast_expression_codegen(ast_expression*, ast_function*, + bool lvalue, ir_value**); +/* TODO: the codegen function should take an output-type parameter + * indicating whether a variable, type, label etc. is expected, and + * an environment! + * Then later an ast_ident could have a codegen using this to figure + * out what to look for. + * eg. in code which uses a not-yet defined variable, the expression + * would take an ast_ident, and the codegen would be called with + * type `expression`, so the ast_ident's codegen would search for + * variables through the environment (or functions, constants...). + */ typedef struct { ast_node_common node; ast_expression_codegen *codegen; + int vtype; + ast_expression *next; + /* arrays get a member-count */ + size_t count; + ast_value* *params; + bool variadic; + /* The codegen functions should store their output values + * so we can call it multiple times without re-evaluating. + * Store lvalue and rvalue seperately though. So that + * ast_entfield for example can generate both if required. + */ + ir_value *outl; + ir_value *outr; } ast_expression_common; /* Value @@ -88,32 +144,51 @@ struct ast_value_s const char *name; + /* int vtype; ast_value *next; + */ bool isconst; union { double vfloat; int vint; - vector_t vvec; + vector vvec; const char *vstring; int ventity; ast_function *vfunc; } constval; + /* usecount for the parser */ + size_t uses; + ir_value *ir_v; + ir_value **ir_values; + size_t ir_value_count; - /* if vtype is qc_function, params contain parameters, and - * 'next' the return type. - */ - MEM_VECTOR_MAKE(ast_value*, params); + /* ONLY for arrays in progs version up to 6 */ + ast_value *setter; + ast_value *getter; }; -ast_value* ast_value_new(lex_ctx_t ctx, const char *name, int qctype, bool keep); + +ast_value* ast_value_new(lex_ctx ctx, const char *name, int qctype); +ast_value* ast_value_copy(const ast_value *self); +/* This will NOT delete an underlying ast_function */ void ast_value_delete(ast_value*); -void ast_value_set_name(ast_value*, const char *name); +bool ast_value_set_name(ast_value*, const char *name); + +bool ast_value_codegen(ast_value*, ast_function*, bool lvalue, ir_value**); +bool ast_local_codegen(ast_value *self, ir_function *func, bool isparam); +bool ast_global_codegen(ast_value *self, ir_builder *ir, bool isfield); -bool ast_value_codegen(ast_value*, ast_function*, ir_value**); +void ast_value_params_add(ast_value*, ast_value*); + +bool ast_compare_type(ast_expression *a, ast_expression *b); +ast_expression* ast_type_copy(lex_ctx ctx, const ast_expression *ex); +#define ast_type_adopt(a, b) ast_type_adopt_impl((ast_expression*)(a), (ast_expression*)(b)) +bool ast_type_adopt_impl(ast_expression *self, const ast_expression *other); +void ast_type_to_string(ast_expression *e, char *buf, size_t bufsize); /* Binary * @@ -123,20 +198,139 @@ struct ast_binary_s { ast_expression_common expression; - int op; - ast_value *left; - ast_value *right; + int op; + ast_expression *left; + ast_expression *right; }; -ast_binary* ast_binary_new(lex_ctx_t ctx, +ast_binary* ast_binary_new(lex_ctx ctx, int op, - ast_value *left, - ast_value *right); + ast_expression *left, + ast_expression *right); void ast_binary_delete(ast_binary*); -/* hmm, seperate functions? -bool ast_block_codegen(ast_block*, ast_function*, ir_value**); +bool ast_binary_codegen(ast_binary*, ast_function*, bool lvalue, ir_value**); + +/* Binstore + * + * An assignment including a binary expression with the source as left operand. + * Eg. a += b; is a binstore { INSTR_STORE, INSTR_ADD, a, b } + */ +struct ast_binstore_s +{ + ast_expression_common expression; + + int opstore; + int opbin; + ast_expression *dest; + ast_expression *source; +}; +ast_binstore* ast_binstore_new(lex_ctx ctx, + int storeop, + int op, + ast_expression *left, + ast_expression *right); +void ast_binstore_delete(ast_binstore*); + +bool ast_binstore_codegen(ast_binstore*, ast_function*, bool lvalue, ir_value**); + +/* Unary + * + * Regular unary expressions: not,neg + */ +struct ast_unary_s +{ + ast_expression_common expression; + + int op; + ast_expression *operand; +}; +ast_unary* ast_unary_new(lex_ctx ctx, + int op, + ast_expression *expr); +void ast_unary_delete(ast_unary*); + +bool ast_unary_codegen(ast_unary*, ast_function*, bool lvalue, ir_value**); + +/* Return + * + * Make sure 'return' only happens at the end of a block, otherwise the IR + * will refuse to create further instructions. + * This should be honored by the parser. + */ +struct ast_return_s +{ + ast_expression_common expression; + ast_expression *operand; +}; +ast_return* ast_return_new(lex_ctx ctx, + ast_expression *expr); +void ast_return_delete(ast_return*); + +bool ast_return_codegen(ast_return*, ast_function*, bool lvalue, ir_value**); + +/* Entity-field + * + * This must do 2 things: + * -) Provide a way to fetch an entity field value. (Rvalue) + * -) Provide a pointer to an entity field. (Lvalue) + * The problem: + * In original QC, there's only a STORE via pointer, but + * no LOAD via pointer. + * So we must know beforehand if we are going to read or assign + * the field. + * For this we will have to extend the codegen() functions with + * a flag saying whether or not we need an L or an R-value. + */ +struct ast_entfield_s +{ + ast_expression_common expression; + /* The entity can come from an expression of course. */ + ast_expression *entity; + /* As can the field, it just must result in a value of TYPE_FIELD */ + ast_expression *field; +}; +ast_entfield* ast_entfield_new(lex_ctx ctx, ast_expression *entity, ast_expression *field); +ast_entfield* ast_entfield_new_force(lex_ctx ctx, ast_expression *entity, ast_expression *field, const ast_expression *outtype); +void ast_entfield_delete(ast_entfield*); + +bool ast_entfield_codegen(ast_entfield*, ast_function*, bool lvalue, ir_value**); + +/* Member access: + * + * For now used for vectors. If we get structs or unions + * we can have them handled here as well. + */ +struct ast_member_s +{ + ast_expression_common expression; + ast_expression *owner; + unsigned int field; +}; +ast_member* ast_member_new(lex_ctx ctx, ast_expression *owner, unsigned int field); +void ast_member_delete(ast_member*); + +bool ast_member_codegen(ast_member*, ast_function*, bool lvalue, ir_value**); + +/* Array index access: + * + * QC forces us to take special action on arrays: + * an ast_store on an ast_array_index must not codegen the index, + * but call its setter - unless we have an instruction set which supports + * what we need. + * Any other array index access will be codegened to a call to the getter. + * In any case, accessing an element via a compiletime-constant index will + * result in quick access to that variable. */ -bool ast_binary_codegen(ast_binary*, ast_function*, ir_value**); +struct ast_array_index_s +{ + ast_expression_common expression; + ast_expression *array; + ast_expression *index; +}; +ast_array_index* ast_array_index_new(lex_ctx ctx, ast_expression *array, ast_expression *index); +void ast_array_index_delete(ast_array_index*); + +bool ast_array_index_codegen(ast_array_index*, ast_function*, bool lvalue, ir_value**); /* Store * @@ -146,15 +340,132 @@ bool ast_binary_codegen(ast_binary*, ast_function*, ir_value**); struct ast_store_s { ast_expression_common expression; - int op; - ast_value *dest; - ast_value *source; + int op; + ast_expression *dest; + ast_expression *source; }; -ast_store* ast_store_new(lex_ctx_t ctx, int op, - ast_value *d, ast_value *s); +ast_store* ast_store_new(lex_ctx ctx, int op, + ast_expression *d, ast_expression *s); void ast_store_delete(ast_store*); -bool ast_store_codegen(ast_store*, ast_function*, ir_value**); +bool ast_store_codegen(ast_store*, ast_function*, bool lvalue, ir_value**); + +/* If + * + * A general 'if then else' statement, either side can be NULL and will + * thus be omitted. It is an error for *both* cases to be NULL at once. + * + * During its 'codegen' it'll be changing the ast_function's block. + * + * An if is also an "expression". Its codegen will put NULL into the + * output field though. For ternary expressions an ast_ternary will be + * added. + */ +struct ast_ifthen_s +{ + ast_expression_common expression; + ast_expression *cond; + /* It's all just 'expressions', since an ast_block is one too. */ + ast_expression *on_true; + ast_expression *on_false; +}; +ast_ifthen* ast_ifthen_new(lex_ctx ctx, ast_expression *cond, ast_expression *ontrue, ast_expression *onfalse); +void ast_ifthen_delete(ast_ifthen*); + +bool ast_ifthen_codegen(ast_ifthen*, ast_function*, bool lvalue, ir_value**); + +/* Ternary expressions... + * + * Contrary to 'if-then-else' nodes, ternary expressions actually + * return a value, otherwise they behave the very same way. + * The difference in 'codegen' is that it'll return the value of + * a PHI node. + * + * The other difference is that in an ast_ternary, NEITHER side + * must be NULL, there's ALWAYS an else branch. + * + * This is the only ast_node beside ast_value which contains + * an ir_value. Theoretically we don't need to remember it though. + */ +struct ast_ternary_s +{ + ast_expression_common expression; + ast_expression *cond; + /* It's all just 'expressions', since an ast_block is one too. */ + ast_expression *on_true; + ast_expression *on_false; + /* After a ternary expression we find ourselves in a new IR block + * and start with a PHI node */ + ir_value *phi_out; +}; +ast_ternary* ast_ternary_new(lex_ctx ctx, ast_expression *cond, ast_expression *ontrue, ast_expression *onfalse); +void ast_ternary_delete(ast_ternary*); + +bool ast_ternary_codegen(ast_ternary*, ast_function*, bool lvalue, ir_value**); + +/* A general loop node + * + * For convenience it contains 4 parts: + * -) (ini) = initializing expression + * -) (pre) = pre-loop condition + * -) (pst) = post-loop condition + * -) (inc) = "increment" expression + * The following is a psudo-representation of this loop + * note that '=>' bears the logical meaning of "implies". + * (a => b) equals (!a || b) + +{ini}; +while (has_pre => {pre}) +{ + {body}; + +continue: // a 'continue' will jump here + if (has_pst => {pst}) + break; + + {inc}; +} + */ +struct ast_loop_s +{ + ast_expression_common expression; + ast_expression *initexpr; + ast_expression *precond; + ast_expression *postcond; + ast_expression *increment; + ast_expression *body; +}; +ast_loop* ast_loop_new(lex_ctx ctx, + ast_expression *initexpr, + ast_expression *precond, + ast_expression *postcond, + ast_expression *increment, + ast_expression *body); +void ast_loop_delete(ast_loop*); + +bool ast_loop_codegen(ast_loop*, ast_function*, bool lvalue, ir_value**); + +/* CALL node + * + * Contains an ast_expression as target, rather than an ast_function/value. + * Since it's how QC works, every ast_function has an ast_value + * associated anyway - in other words, the VM contains function + * pointers for every function anyway. Thus, this node will call + * expression. + * Additionally it contains a list of ast_expressions as parameters. + * Since calls can return values, an ast_call is also an ast_expression. + */ +struct ast_call_s +{ + ast_expression_common expression; + ast_expression *func; + ast_expression* *params; +}; +ast_call* ast_call_new(lex_ctx ctx, + ast_expression *funcexpr); +void ast_call_delete(ast_call*); +bool ast_call_codegen(ast_call*, ast_function*, bool lvalue, ir_value**); +bool ast_call_check_types(ast_call*); /* Blocks * @@ -163,16 +474,16 @@ struct ast_block_s { ast_expression_common expression; - MEM_VECTOR_MAKE(ast_value*, locals); - MEM_VECTOR_MAKE(ast_expression*, exprs); + ast_value* *locals; + ast_expression* *exprs; + ast_expression* *collect; }; -ast_block* ast_block_new(lex_ctx_t ctx); +ast_block* ast_block_new(lex_ctx ctx); void ast_block_delete(ast_block*); +bool ast_block_set_type(ast_block*, ast_expression *from); -MEM_VECTOR_PROTO(ast_block, ast_value*, locals); -MEM_VECTOR_PROTO(ast_block, ast_expression*, exprs); - -bool ast_block_codegen(ast_block*, ast_function*, ir_value**); +bool ast_block_codegen(ast_block*, ast_function*, bool lvalue, ir_value**); +void ast_block_collect(ast_block*, ast_expression*); /* Function * @@ -191,12 +502,29 @@ struct ast_function_s ast_value *vtype; const char *name; - MEM_VECTOR_MAKE(ast_block*, blocks); + int builtin; + + ir_function *ir_func; + ir_block *curblock; + ir_block *breakblock; + ir_block *continueblock; + + size_t labelcount; + /* in order for thread safety - for the optional + * channel abesed multithreading... keeping a buffer + * here to use in ast_function_label. + */ + char labelbuf[64]; + + ast_block* *blocks; }; -ast_function* ast_function_new(lex_ctx_t ctx, const char *name, ast_value *vtype); +ast_function* ast_function_new(lex_ctx ctx, const char *name, ast_value *vtype); +/* This will NOT delete the underlying ast_value */ void ast_function_delete(ast_function*); - -MEM_VECTOR_PROTO(ast_function, ast_block*, blocks); +/* For "optimized" builds this can just keep returning "foo"... + * or whatever... + */ +const char* ast_function_label(ast_function*, const char *prefix); bool ast_function_codegen(ast_function *self, ir_builder *builder); @@ -206,8 +534,13 @@ union ast_expression_u { ast_expression_common expression; - ast_binary binary; - ast_block block; + ast_value value; + ast_binary binary; + ast_block block; + ast_ternary ternary; + ast_ifthen ifthen; + ast_store store; + ast_entfield entfield; }; /* Node union