Author: kjs
Date: Wed Mar 28 01:19:45 2007
New Revision: 17809
Modified:
trunk/compilers/pirc/src/pirparser.c
Log:
compilers/pirc:
* add support for x = global $S0
* add suggestions for future optimizations
* add support for forgotten .pcc_sub directive
Modified: trunk/compilers/pirc/src/pirparser.c
==============================================================================
--- trunk/compilers/pirc/src/pirparser.c (original)
+++ trunk/compilers/pirc/src/pirparser.c Wed Mar 28 01:19:45 2007
@@ -640,9 +640,9 @@
break;
}
break;
- case T_GLOBAL: /* x = global 'i' */
+ case T_GLOBAL: /* x = global string_value */
next(p);
- stringconstant(p);
+ string_value(p);
break;
case T_NULL:
next(p);
@@ -653,10 +653,10 @@
expression(p);
while (p->curtoken == T_COMMA) {
next(p);
- expression(p);
+ expression(p);
}
- }
- break;
+ }
+ break;
case T_HEREDOC_ID: { /* parse heredoc string */
char *heredocid = clone_string(get_current_token(p->lexer));
/* read_heredoc() returns a special token */
@@ -1352,6 +1352,22 @@
/*
+=item string_value()
+
+ string_value -> SREG | PASM_SREG | stringconstant
+
+=cut
+
+*/
+static void
+string_value(parser_state *p) {
+ /* also allow string registers */
+ if (p->curtoken == T_SREG || p->curtoken == T_PASM_SREG) next(p);
+ else stringconstant(p);
+}
+
+/*
+
=item global_assignment()
Only PMCs can be stored as globals, so only PMC registers and identifiers
@@ -1365,7 +1381,7 @@
static void
global_assignment(parser_state *p) {
match(p, T_GLOBAL);
- stringconstant(p);
+ string_value(p);
match(p, T_ASSIGN);
switch (p->curtoken) {
@@ -1719,7 +1735,10 @@
*/
static void
sub_definition(parser_state *p) {
- match(p, T_SUB);
+ /* either '.sub' or '.pcc_sub'. This kind of optimization (just skipping)
+ * can be done more often, if we're sure the token has already been
checked for.
+ */
+ next(p);
if (p->curtoken == T_IDENTIFIER) match(p, T_IDENTIFIER);
else stringconstant(p);
@@ -2045,6 +2064,32 @@
=back
+=head1 POSSIBLE FUTURE OPTIMIZATIONS
+
+=over 4
+
+=item *
+
+Use #define's to inline short functions that are called often (such as next(),
match())
+
+=item *
+
+Remove functions for short statements, and put in the matching into the
switch() for compilation_unit.
+
+=item *
+
+Skip first token of statements if it's sure that the token has already been
checked by the caller. No need
+to match 'goto' again, if it was already checked in compilation_unit().
+
+=item *
+
+Add a 'token category'. For instance, when parsing a binary operator, you're
just interested if it actually is
+a binary op. The lexer can set the token category to "binop", and set token to
T_PLUS, etc. This prevents
+big switch statements. Also, we still have access to the actual binary
operator (through token), so semantic
+analysis can continue.
+
+=back
+
=cut
*/
