This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
A small performance improvement for standalone cpp
- To: gcc-patches at gcc dot gnu dot org
- Subject: A small performance improvement for standalone cpp
- From: Zack Weinberg <zack at codesourcery dot com>
- Date: Sat, 6 Oct 2001 16:11:00 -0700
This patch gives numbers (and, obscurely, comments in -C mode) their
own spelling category. That lets us avoid a bunch of mispredicted
branches in cpp_output_token. I get about 1% performance improvement.
This change exposed a longstanding bug in cppexp.c which was causing
confusing error messages about ill-formed #if statements.
bootstrapped i686-linux, no regressions, although the test suite is in
kinda bad shape right now.
zw
* cpplex.c (enum spell_type): Add SPELL_NUMBER.
(cpp_token_len, cpp_spell_token, cpp_output_token,
_cpp_equiv_tokens): Handle it.
* cpplib.h (TTYPE_TABLE): Use SPELL_NUMBER for CPP_NUMBER and
CPP_COMMENT, SPELL_NONE for CPP_PADDING and CPP_EOF.
* cppexp.c (_cpp_parse_expr): Use the correct operator code in
error messages.
===================================================================
Index: cppexp.c
--- cppexp.c 2001/09/24 22:53:04 1.96
+++ cppexp.c 2001/10/06 23:02:29
@@ -819,13 +819,13 @@ _cpp_parse_expr (pfile)
{
if (top->flags & HAVE_VALUE)
SYNTAX_ERROR2 ("missing binary operator before '%s'",
- op_as_text (pfile, top->op));
+ op_as_text (pfile, op.op));
}
else
{
if (!(top->flags & HAVE_VALUE))
SYNTAX_ERROR2 ("operator '%s' has no left operand",
- op_as_text (pfile, top->op));
+ op_as_text (pfile, op.op));
}
/* Check for and handle stack overflow. */
===================================================================
Index: cpplex.c
--- cpplex.c 2001/09/30 10:03:09 1.172
+++ cpplex.c 2001/10/06 23:02:30
@@ -58,6 +58,7 @@ enum spell_type
SPELL_OPERATOR = 0,
SPELL_CHAR,
SPELL_IDENT,
+ SPELL_NUMBER,
SPELL_STRING,
SPELL_NONE
};
@@ -1412,10 +1413,11 @@ cpp_token_len (token)
switch (TOKEN_SPELL (token))
{
default: len = 0; break;
+ case SPELL_NUMBER:
case SPELL_STRING: len = token->val.str.len; break;
case SPELL_IDENT: len = NODE_LEN (token->val.node); break;
}
- /* 1 for whitespace, 4 for comment delimeters. */
+ /* 1 for whitespace, 4 for comment delimiters. */
return len + 5;
}
@@ -1449,12 +1451,21 @@ cpp_spell_token (pfile, token, buffer)
}
break;
+ case SPELL_CHAR:
+ *buffer++ = token->val.c;
+ break;
+
+ spell_ident:
case SPELL_IDENT:
- spell_ident:
memcpy (buffer, NODE_NAME (token->val.node), NODE_LEN (token->val.node));
buffer += NODE_LEN (token->val.node);
break;
+ case SPELL_NUMBER:
+ memcpy (buffer, token->val.str.text, token->val.str.len);
+ buffer += token->val.str.len;
+ break;
+
case SPELL_STRING:
{
int left, right, tag;
@@ -1465,20 +1476,18 @@ cpp_spell_token (pfile, token, buffer)
case CPP_CHAR: left = '\''; right = '\''; tag = '\0'; break;
case CPP_WCHAR: left = '\''; right = '\''; tag = 'L'; break;
case CPP_HEADER_NAME: left = '<'; right = '>'; tag = '\0'; break;
- default: left = '\0'; right = '\0'; tag = '\0'; break;
+ default:
+ cpp_ice (pfile, "unknown string token %s\n", TOKEN_NAME (token));
+ return buffer;
}
if (tag) *buffer++ = tag;
- if (left) *buffer++ = left;
+ *buffer++ = left;
memcpy (buffer, token->val.str.text, token->val.str.len);
buffer += token->val.str.len;
- if (right) *buffer++ = right;
+ *buffer++ = right;
}
break;
- case SPELL_CHAR:
- *buffer++ = token->val.c;
- break;
-
case SPELL_NONE:
cpp_ice (pfile, "Unspellable token %s", TOKEN_NAME (token));
break;
@@ -1541,11 +1550,19 @@ cpp_output_token (token, fp)
}
break;
+ case SPELL_CHAR:
+ putc (token->val.c, fp);
+ break;
+
spell_ident:
case SPELL_IDENT:
fwrite (NODE_NAME (token->val.node), 1, NODE_LEN (token->val.node), fp);
break;
+ case SPELL_NUMBER:
+ fwrite (token->val.str.text, 1, token->val.str.len, fp);
+ break;
+
case SPELL_STRING:
{
int left, right, tag;
@@ -1556,19 +1573,17 @@ cpp_output_token (token, fp)
case CPP_CHAR: left = '\''; right = '\''; tag = '\0'; break;
case CPP_WCHAR: left = '\''; right = '\''; tag = 'L'; break;
case CPP_HEADER_NAME: left = '<'; right = '>'; tag = '\0'; break;
- default: left = '\0'; right = '\0'; tag = '\0'; break;
+ default:
+ fprintf (stderr, "impossible STRING token %s\n", TOKEN_NAME (token));
+ return;
}
if (tag) putc (tag, fp);
- if (left) putc (left, fp);
+ putc (left, fp);
fwrite (token->val.str.text, 1, token->val.str.len, fp);
- if (right) putc (right, fp);
+ putc (right, fp);
}
break;
- case SPELL_CHAR:
- putc (token->val.c, fp);
- break;
-
case SPELL_NONE:
/* An error, most probably. */
break;
@@ -1592,6 +1607,7 @@ _cpp_equiv_tokens (a, b)
return (a->type != CPP_MACRO_ARG || a->val.arg_no == b->val.arg_no);
case SPELL_IDENT:
return a->val.node == b->val.node;
+ case SPELL_NUMBER:
case SPELL_STRING:
return (a->val.str.len == b->val.str.len
&& !memcmp (a->val.str.text, b->val.str.text,
===================================================================
Index: cpplib.h
--- cpplib.h 2001/10/02 12:57:23 1.194
+++ cpplib.h 2001/10/06 23:02:30
@@ -122,7 +122,7 @@ struct file_name_map_list;
OP(CPP_ATSIGN, "@") /* used in Objective C */ \
\
TK(CPP_NAME, SPELL_IDENT) /* word */ \
- TK(CPP_NUMBER, SPELL_STRING) /* 34_be+ta */ \
+ TK(CPP_NUMBER, SPELL_NUMBER) /* 34_be+ta */ \
\
TK(CPP_CHAR, SPELL_STRING) /* 'char' */ \
TK(CPP_WCHAR, SPELL_STRING) /* L'char' */ \
@@ -132,10 +132,11 @@ struct file_name_map_list;
TK(CPP_WSTRING, SPELL_STRING) /* L"string" */ \
TK(CPP_HEADER_NAME, SPELL_STRING) /* <stdio.h> in #include */ \
\
- TK(CPP_COMMENT, SPELL_STRING) /* Only if output comments. */ \
+ TK(CPP_COMMENT, SPELL_NUMBER) /* Only if output comments. */ \
+ /* SPELL_NUMBER happens to DTRT. */ \
TK(CPP_MACRO_ARG, SPELL_NONE) /* Macro argument. */ \
- OP(CPP_PADDING, "") /* Whitespace for cpp0. */ \
- OP(CPP_EOF, "EOL") /* End of line or file. */
+ TK(CPP_PADDING, SPELL_NONE) /* Whitespace for cpp0. */ \
+ TK(CPP_EOF, SPELL_NONE) /* End of line or file. */
#define OP(e, s) e,
#define TK(e, s) e,