aboutsummaryrefslogtreecommitdiffstats
path: root/zahl
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--zahl.h1
-rw-r--r--zahl/inlines.h10
-rw-r--r--zahl/internals.h61
-rw-r--r--zahl/memory.h133
4 files changed, 143 insertions, 62 deletions
diff --git a/zahl.h b/zahl.h
index e2a77c7..cfd85a3 100644
--- a/zahl.h
+++ b/zahl.h
@@ -11,6 +11,7 @@
#include <setjmp.h>
#include <stddef.h>
#include <stdint.h>
+#include <unistd.h>
#include "zahl/internals.h"
diff --git a/zahl/inlines.h b/zahl/inlines.h
index 9884812..c6ddc84 100644
--- a/zahl/inlines.h
+++ b/zahl/inlines.h
@@ -259,12 +259,16 @@ zsave(z_t a, void *buffer)
{
if (ZAHL_LIKELY(buffer)) {
char *buf = buffer;
- *((int *)buf) = a->sign, buf += sizeof(int);
+ *((long *)buf) = a->sign, buf += sizeof(long); /* Use `long` for alignment. */
*((size_t *)buf) = a->used, buf += sizeof(size_t);
- if (ZAHL_LIKELY(!zzero(a)))
+ if (ZAHL_LIKELY(!zzero(a))) {
+ a->chars[a->used + 2] = 0;
+ a->chars[a->used + 1] = 0;
+ a->chars[a->used + 0] = 0;
libzahl_memcpy((zahl_char_t *)buf, a->chars, a->used);
+ }
}
- return sizeof(int) + sizeof(size_t) + (zzero(a) ? 0 : a->used * sizeof(zahl_char_t));
+ return sizeof(long) + sizeof(size_t) + (zzero(a) ? 0 : ((a->used + 3) & ~3) * sizeof(zahl_char_t));
}
diff --git a/zahl/internals.h b/zahl/internals.h
index 672dfcd..30e2765 100644
--- a/zahl/internals.h
+++ b/zahl/internals.h
@@ -51,6 +51,7 @@
#define ZAHL_BITS_PER_CHAR 64
#define ZAHL_LB_BITS_PER_CHAR 6
#define ZAHL_CHAR_MAX UINT64_MAX
+#define ZAHL_FLUFF 4
/* Note: These cannot be changed willy-nilly, some code depends
* on them, be cause being flexible would just be too painful. */
@@ -113,62 +114,4 @@ extern struct zahl libzahl_tmp_mod[1];
void libzahl_realloc(struct zahl *, size_t);
-ZAHL_INLINE void
-libzahl_memcpy(register zahl_char_t *d, register const zahl_char_t *s, size_t n)
-{
- size_t i;
- if (n <= 4) {
- if (n >= 1)
- d[0] = s[0];
- if (n >= 2)
- d[1] = s[1];
- if (n >= 3)
- d[2] = s[2];
- if (n >= 4)
- d[3] = s[3];
- } else {
- for (i = 0; (i += 4) <= n;) {
- d[i - 4] = s[i - 4];
- d[i - 3] = s[i - 3];
- d[i - 2] = s[i - 2];
- d[i - 1] = s[i - 1];
- }
- if (i > n) {
- i -= 4;
- if (i < n)
- d[i] = s[i], i++;
- if (i < n)
- d[i] = s[i], i++;
- if (i < n)
- d[i] = s[i], i++;
- if (i < n)
- d[i] = s[i];
- }
- }
-}
-
-ZAHL_INLINE void
-libzahl_memset(register zahl_char_t *a, register zahl_char_t v, size_t n)
-{
- size_t i;
- if (n <= 4) {
- if (n >= 1)
- a[0] = v;
- if (n >= 2)
- a[1] = v;
- if (n >= 3)
- a[2] = v;
- if (n >= 4)
- a[3] = v;
- } else {
- for (i = 0; (i += 4) <= n;) {
- a[i - 1] = v;
- a[i - 2] = v;
- a[i - 3] = v;
- a[i - 4] = v;
- }
- if (i > n)
- for (i -= 4; i < n; i++)
- a[i] = v;
- }
-}
+#include "memory.h"
diff --git a/zahl/memory.h b/zahl/memory.h
new file mode 100644
index 0000000..2eb2b68
--- /dev/null
+++ b/zahl/memory.h
@@ -0,0 +1,133 @@
+/* See LICENSE file for copyright and license details. */
+
+#define LIBZAHL_MEM_CASES \
+ LIBZAHL_X(20); \
+ LIBZAHL_X(19); \
+ LIBZAHL_X(18); \
+ LIBZAHL_X(17); \
+ LIBZAHL_X(16); \
+ LIBZAHL_X(15); \
+ LIBZAHL_X(14); \
+ LIBZAHL_X(13); \
+ LIBZAHL_X(12); \
+ LIBZAHL_X(11); \
+ LIBZAHL_X(10); \
+ LIBZAHL_X( 9); \
+ LIBZAHL_X( 8); \
+ LIBZAHL_X( 7); \
+ LIBZAHL_X( 6); \
+ LIBZAHL_X( 5); \
+ LIBZAHL_X( 4); \
+ LIBZAHL_X( 3); \
+ LIBZAHL_X( 2); \
+ LIBZAHL_X( 1); \
+ case 0: break;
+
+
+#if defined(LIBZAHL_ISA_MISSING_INDIRECT_JUMP)
+# define LIBZAHL_SMALL_INPUT_BEGIN(n)
+# define LIBZAHL_SMALL_INPUT_END
+#else
+# define LIBZAHL_SMALL_INPUT_BEGIN(n) switch (n) { LIBZAHL_MEM_CASES default:
+# define LIBZAHL_SMALL_INPUT_END break; }
+#endif
+
+
+ZAHL_INLINE void
+libzahl_memcpy(register zahl_char_t *restrict d, register const zahl_char_t *restrict s, size_t n)
+{
+ size_t i;
+#define LIBZAHL_X(I) case I: d[I - 1] = s[I - 1];
+ LIBZAHL_SMALL_INPUT_BEGIN(n);
+ for (i = 0; i < n; i += 4) {
+ d[i + 0] = s[i + 0];
+ d[i + 1] = s[i + 1];
+ d[i + 2] = s[i + 2];
+ d[i + 3] = s[i + 3];
+ }
+ LIBZAHL_SMALL_INPUT_END;
+#undef LIBZAHL_X
+}
+
+
+ZAHL_INLINE void
+libzahl_memset(register zahl_char_t *a, register zahl_char_t v, size_t n)
+{
+ size_t i;
+ for (i = 0; i < n; i += 4) {
+ a[i + 0] = v;
+ a[i + 1] = v;
+ a[i + 2] = v;
+ a[i + 3] = v;
+ }
+}
+
+ZAHL_INLINE void
+libzahl_memset_precise(register zahl_char_t *a, register zahl_char_t v, size_t n)
+{
+ size_t i;
+ if (n <= 4) {
+ if (n >= 1)
+ a[0] = v;
+ if (n >= 2)
+ a[1] = v;
+ if (n >= 3)
+ a[2] = v;
+ if (n >= 4)
+ a[3] = v;
+ } else {
+ for (i = 0; (i += 4) <= n;) {
+ a[i - 1] = v;
+ a[i - 2] = v;
+ a[i - 3] = v;
+ a[i - 4] = v;
+ }
+ if (i > n)
+ for (i -= 4; i < n; i++)
+ a[i] = v;
+ }
+}
+
+
+ZAHL_INLINE void
+libzahl_memmovef(register zahl_char_t *d, register const zahl_char_t *s, size_t n)
+{
+ if (n && n < 4) {
+ d[0] = s[0];
+ d[1] = s[1];
+ d[2] = s[2];
+ } else {
+ size_t i;
+ for (i = 0; i < n; i += 4) {
+ d[i + 0] = s[i + 0];
+ d[i + 1] = s[i + 1];
+ d[i + 2] = s[i + 2];
+ d[i + 3] = s[i + 3];
+ }
+ }
+}
+
+ZAHL_INLINE void
+libzahl_memmoveb(register zahl_char_t *d, register const zahl_char_t *s, size_t n)
+{
+ ssize_t i;
+#define LIBZAHL_X(I) case I: d[I - 1] = s[I - 1];
+ LIBZAHL_SMALL_INPUT_BEGIN(n);
+ for (i = ((ssize_t)n + 3) & ~3; (i -= 4) >= 0;) {
+ d[i + 3] = s[i + 3];
+ d[i + 2] = s[i + 2];
+ d[i + 1] = s[i + 1];
+ d[i + 0] = s[i + 0];
+ }
+ LIBZAHL_SMALL_INPUT_END;
+#undef LIBZAHL_X
+}
+
+ZAHL_INLINE void
+libzahl_memmove(register zahl_char_t *d, register const zahl_char_t *s, size_t n)
+{
+ if (d < s)
+ libzahl_memmovef(d, s, n);
+ else
+ libzahl_memmoveb(d, s, n);
+}
generated by cgit v1.2.3-70-g09d2 (git 2.51.0) at 2026-02-18 22:11:54 +0000