I wrote:
> The benchmarks clearly show the speedup, especially for the case of ASCII
> text (CPU time, measured on x86_64, with gcc 13, on an AMD Ryzen 7 CPU):
>
> mbiter mbiterf
> a 0.926 0.230
> b 0.765 0.233
> c 1.706 1.125
> d 1.324 0.847
> e 1.580 0.889
> f 14.134 13.095
> g 8.409 8.244
> h 8.882 8.553
> i 4.110 4.000
> j 4.430 4.212
Similarly with clang 16 (on the same machine):
gcc clang
mbiter mbiterf mbiter mbiterf
a 0.926 0.230 1.425 0.227
b 0.765 0.233 1.442 0.227
c 1.706 1.125 2.761 1.084
d 1.324 0.847 2.565 0.833
e 1.580 0.889 2.582 0.878
f 14.134 13.095 13.278 12.431
g 8.409 8.244 9.343 7.898
h 8.882 8.553 9.766 8.341
i 4.110 4.000 4.316 4.025
j 4.430 4.212 4.501 4.254
Here the speedup in the ASCII case is even 6-fold,
because clang apparently does not optimize the state-based code as well as gcc,
but optimizes the struct-as-return-value code as well as (or slightly better
than) gcc.
2023-07-18 Bruno Haible <[email protected]>
mbiterf: Fix compiler warnings.
* lib/mbiterf.h (mbiterf_next): Use C99 designated initializer syntax.
* tests/bench-mbiterf.c (do_test): Use a 'const char *' pointer.
diff --git a/lib/mbiterf.h b/lib/mbiterf.h
index 0059c97fbd..f5a817dbed 100644
--- a/lib/mbiterf.h
+++ b/lib/mbiterf.h
@@ -120,7 +120,7 @@ mbiterf_next (struct mbif_state *ps, const char *iter,
const char *endptr)
their char code. For the few other ones, this is the case as well,
in all locale encodings that are in use. The 32-bit wide character
code is the same as well. */
- return (mbchar_t) { ptr: iter, bytes: 1, wc_valid: true, wc: *iter };
+ return (mbchar_t) { .ptr = iter, .bytes = 1, .wc_valid = true, .wc =
*iter };
}
else
{
@@ -140,7 +140,7 @@ mbiterf_next (struct mbif_state *ps, const char *iter,
const char *endptr)
ps->in_shift = false;
#endif
mbszero (&ps->state);
- return (mbchar_t) { ptr: iter, bytes: 1, wc_valid: false };
+ return (mbchar_t) { .ptr = iter, .bytes = 1, .wc_valid = false };
}
else if (bytes == (size_t) -2)
{
@@ -150,7 +150,7 @@ mbiterf_next (struct mbif_state *ps, const char *iter,
const char *endptr)
#endif
/* Whether to reset ps->state or not is not important; the string end
is reached anyway. */
- return (mbchar_t) { ptr: iter, bytes: endptr - iter, wc_valid: false
};
+ return (mbchar_t) { .ptr = iter, .bytes = endptr - iter, .wc_valid =
false };
}
else
{
@@ -174,7 +174,7 @@ mbiterf_next (struct mbif_state *ps, const char *iter,
const char *endptr)
if (mbsinit (&ps->state))
ps->in_shift = false;
#endif
- return (mbchar_t) { ptr: iter, bytes: bytes, wc_valid: true, wc: wc
};
+ return (mbchar_t) { .ptr = iter, .bytes = bytes, .wc_valid = true,
.wc = wc };
}
}
}
diff --git a/tests/bench-mbiterf.c b/tests/bench-mbiterf.c
index 0949fb6752..bc3fbe4828 100644
--- a/tests/bench-mbiterf.c
+++ b/tests/bench-mbiterf.c
@@ -43,7 +43,7 @@ do_test (char test, int repeat, const char *locale_name,
const char *text)
unsigned long long sum = 0;
const char *text_end = text + text_len;
mbif_state_t state;
- char *iter;
+ const char *iter;
for (mbif_init (state), iter = text; mbif_avail (state, iter,
text_end); )
{
mbchar_t cur = mbif_next (state, iter, text_end);
