Hi Eric,
> > | Effects on "autoconf" in gettext/gettext-tools:
> > |
> > | m4-1.4.10 50.6 s real + 1.0 sec system
> > | m4-1.4.10b 42.1 s real + 0.9 sec system
>
> And with --disable-assert, I get:
>
> 41.2 s real + 0.9 sec system
With the patch below, I get:
36.9 s real + 0.9 sec system
i.e. an additional 10% speedup.
The speed-relevant code is only the case for INPUT_CHAIN. I added the one
for INPUT_STRING because it may be useful it situations I don't know of.
2008-02-28 Bruno Haible <[EMAIL PROTECTED]>
Optimize a code path frequently exercised by autoconf.
Memory impact: none.
Speed impact: 10% speedup.
* src/input.c (next_token): Add an optimized code path for the
frequent case of rescanning chains or strings.
*** src/input.c.bak 2008-02-23 18:13:02.000000000 +0100
--- src/input.c 2008-02-28 14:52:02.000000000 +0100
***************
*** 1632,1637 ****
--- 1632,1741 ----
quote_level = 1;
while (1)
{
+ if (curr_quote.len1 == 1 && curr_quote.len2 == 1 && !input_change
+ && isp)
+ {
+ /* The optimized case. It heavily inlines the MATCH macro and
+ the next_char and next_char_1 functions, to the point that
+ the scan is a loop over a region of memory followed by a
+ simple memory copy operation.
+ The case with INPUT_CHAIN alone can speed up GNU autoconf
+ runs by 10%. */
+ if (isp->type == INPUT_CHAIN)
+ {
+ token_chain *chain = isp->u.u_c.chain;
+
+ if (chain)
+ {
+ if (obs != NULL && current_quote_age
+ && chain->quote_age == current_quote_age)
+ {
+ /* next_char () would return CHAR_QUOTE here. */
+ append_quote_token (obs, td);
+ }
+ else if (chain->type == CHAIN_STR && chain->u.u_s.len > 0)
+ {
+ unsigned char curr_quote_1 =
+ to_uchar (curr_quote.str1[0]);
+ unsigned char curr_quote_2 =
+ to_uchar (curr_quote.str2[0]);
+ const char *p;
+ size_t n;
+ size_t count;
+
+ /* Partial consumption invalidates quote age. */
+ chain->quote_age = 0;
+
+ p = chain->u.u_s.str;
+ n = chain->u.u_s.len;
+ do
+ {
+ unsigned char ch = to_uchar (*p);
+ if (ch == curr_quote_2)
+ {
+ if (--quote_level == 0)
+ break;
+ }
+ else
+ quote_level += (ch == curr_quote_1);
+ p++;
+ n--;
+ }
+ while (n > 0);
+
+ count = p - chain->u.u_s.str;
+ if (count > 0)
+ obstack_grow (obs_td, chain->u.u_s.str, count);
+ count += (quote_level == 0);
+ chain->u.u_s.str += count;
+ chain->u.u_s.len -= count;
+ if (quote_level == 0)
+ break;
+ continue;
+ }
+ }
+ }
+ else if (isp->type == INPUT_STRING)
+ {
+ if (isp->u.u_s.len > 0)
+ {
+ unsigned char curr_quote_1 =
+ to_uchar (curr_quote.str1[0]);
+ unsigned char curr_quote_2 =
+ to_uchar (curr_quote.str2[0]);
+ const char *p = isp->u.u_s.str;
+ size_t n = isp->u.u_s.len;
+ size_t count;
+
+ do
+ {
+ unsigned char ch = to_uchar (*p);
+ if (ch == curr_quote_2)
+ {
+ if (--quote_level == 0)
+ break;
+ }
+ else
+ quote_level += (ch == curr_quote_1);
+ p++;
+ n--;
+ }
+ while (n > 0);
+
+ count = p - isp->u.u_s.str;
+ if (count > 0)
+ obstack_grow (obs_td, isp->u.u_s.str, count);
+ count += (quote_level == 0);
+ isp->u.u_s.str += count;
+ isp->u.u_s.len -= count;
+ if (quote_level == 0)
+ break;
+ continue;
+ }
+ }
+ }
+
+ /* The general case. Proceed character by character. */
ch = next_char (obs != NULL && current_quote_age);
if (ch == CHAR_EOF)
/* Current_file changed to "" if we see CHAR_EOF, use
_______________________________________________
M4-discuss mailing list
M4-discuss@gnu.org
http://lists.gnu.org/mailman/listinfo/m4-discuss
