This patch speeds up m4 by 20% in its typical use by autoconf. It was
tested with the configure script of libtool 1.9b. Before the patch we
have the following top 5 functions:
% cumulative self self total
time seconds seconds calls s/call s/call name
50.33 26.54 26.54 833006410 0.00 0.00 next_char
43.97 49.73 23.19 12563045 0.00 0.00 next_token
2.09 50.83 1.10 413305 0.00 0.00 expand_token
0.86 51.28 0.46 19684516 0.00 0.00 peek_input
0.75 51.68 0.39 251578 0.00 0.00 expand_user_macro
After the patch we have this profile:
% cumulative self self total
time seconds seconds calls s/call s/call name
91.77 33.86 33.86 12563045 0.00 0.00 next_token
3.01 34.97 1.11 413305 0.00 0.00 expand_token
1.25 35.43 0.46 19684516 0.00 0.00 peek_input
1.15 35.86 0.43 251578 0.00 0.00 expand_user_macro
[...]
0.03 36.79 0.01 290882 0.00 0.00 next_char_1
As you can see, 99.9% of all calls to next_char are reading from a string,
so we want that case to be fast. In the test case the total runtime
decreased from 56 to 44 seconds.
Andreas.
2004-09-05 Andreas Schwab <[EMAIL PROTECTED]>
* src/input.c (next_char_1): Renamed from next_char.
(next_char): New macro.
--- src/input.c.~1.1.1.1.~ 2000-02-17 04:03:19.000000000 +0100
+++ src/input.c 2004-09-05 17:44:00.622015557 +0200
@@ -424,10 +424,17 @@ peek_input (void)
| they do not get wrong, due to lookahead. The token consisting of a |
| newline alone is taken as belonging to the line it ends, and the current |
| line number is not incremented until the next character is read. |
+| 99.9% of all calls will read from a string, so factor that out into a |
+| macro for speed. |
`-------------------------------------------------------------------------*/
+#define next_char() \
+ (isp && isp->type == INPUT_STRING && isp->u.u_s.string[0] \
+ ? *isp->u.u_s.string++ \
+ : next_char_1 ())
+
static int
-next_char (void)
+next_char_1 (void)
{
register int ch;
--
Andreas Schwab, SuSE Labs, [EMAIL PROTECTED]
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756 01D3 44D5 214B 8276 4ED5
"And now for something completely different."
_______________________________________________
Bug-m4 mailing list
[EMAIL PROTECTED]
http://lists.gnu.org/mailman/listinfo/bug-m4
