]> wimlib.net Git - wimlib/blobdiff - include/wimlib/unaligned.h
Stop force-inlining everything marked 'inline'
[wimlib] / include / wimlib / unaligned.h
index 7c2ade22031894debbf1f449295c5499d32faf96..ead4629583d9eec3b270eb495aefb121ac248766 100644 (file)
 /*
- * unaligned.h
+ * unaligned.h - inline functions for unaligned memory accesses
  *
- * Inline functions for unaligned memory accesses.
+ * The following copying information applies to this specific source code file:
  *
- * The author dedicates this file to the public domain.
- * You can do whatever you want with this file.
+ * Written in 2014-2015 by Eric Biggers <ebiggers3@gmail.com>
+ *
+ * To the extent possible under law, the author(s) have dedicated all copyright
+ * and related and neighboring rights to this software to the public domain
+ * worldwide via the Creative Commons Zero 1.0 Universal Public Domain
+ * Dedication (the "CC0").
+ *
+ * This software is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE. See the CC0 for more details.
+ *
+ * You should have received a copy of the CC0 along with this software; if not
+ * see <http://creativecommons.org/publicdomain/zero/1.0/>.
  */
 
 #ifndef _WIMLIB_UNALIGNED_H
 #define _WIMLIB_UNALIGNED_H
 
-#include "compiler.h"
-#include "endianness.h"
-#include "types.h"
+#include "wimlib/compiler.h"
+#include "wimlib/endianness.h"
+#include "wimlib/types.h"
 
 #define DEFINE_UNALIGNED_TYPE(type)                            \
 struct type##_unaligned {                                      \
        type v;                                                 \
 } _packed_attribute;                                           \
                                                                \
-static inline type                                             \
+static forceinline type                                                \
 load_##type##_unaligned(const void *p)                         \
 {                                                              \
        return ((const struct type##_unaligned *)p)->v;         \
 }                                                              \
                                                                \
-static inline void                                             \
+static forceinline void                                                \
 store_##type##_unaligned(type val, void *p)                    \
 {                                                              \
        ((struct type##_unaligned *)p)->v = val;                \
 }
 
+DEFINE_UNALIGNED_TYPE(u16);
+DEFINE_UNALIGNED_TYPE(u32);
+DEFINE_UNALIGNED_TYPE(u64);
 DEFINE_UNALIGNED_TYPE(le16);
 DEFINE_UNALIGNED_TYPE(le32);
 DEFINE_UNALIGNED_TYPE(le64);
+DEFINE_UNALIGNED_TYPE(be16);
 DEFINE_UNALIGNED_TYPE(be32);
+DEFINE_UNALIGNED_TYPE(be64);
 DEFINE_UNALIGNED_TYPE(size_t);
+DEFINE_UNALIGNED_TYPE(machine_word_t);
+
+#define load_word_unaligned    load_machine_word_t_unaligned
+#define store_word_unaligned   store_machine_word_t_unaligned
+
+static forceinline u16
+get_unaligned_le16(const u8 *p)
+{
+       if (UNALIGNED_ACCESS_IS_FAST)
+               return le16_to_cpu(load_le16_unaligned(p));
+       else
+               return ((u16)p[1] << 8) | p[0];
+}
+
+static forceinline u32
+get_unaligned_le32(const u8 *p)
+{
+       if (UNALIGNED_ACCESS_IS_FAST)
+               return le32_to_cpu(load_le32_unaligned(p));
+       else
+               return ((u32)p[3] << 24) | ((u32)p[2] << 16) |
+                       ((u32)p[1] << 8) | p[0];
+}
+
+static forceinline void
+put_unaligned_le16(u16 v, u8 *p)
+{
+       if (UNALIGNED_ACCESS_IS_FAST) {
+               store_le16_unaligned(cpu_to_le16(v), p);
+       } else {
+               p[0] = (u8)(v >> 0);
+               p[1] = (u8)(v >> 8);
+       }
+}
+
+static forceinline void
+put_unaligned_le32(u32 v, u8 *p)
+{
+       if (UNALIGNED_ACCESS_IS_FAST) {
+               store_le32_unaligned(cpu_to_le32(v), p);
+       } else {
+               p[0] = (u8)(v >> 0);
+               p[1] = (u8)(v >> 8);
+               p[2] = (u8)(v >> 16);
+               p[3] = (u8)(v >> 24);
+       }
+}
+
+/*
+ * Given a 32-bit value that was loaded with the platform's native endianness,
+ * return a 32-bit value whose high-order 8 bits are 0 and whose low-order 24
+ * bits contain the first 3 bytes, arranged in octets in a platform-dependent
+ * order, at the memory location from which the input 32-bit value was loaded.
+ */
+static forceinline u32
+loaded_u32_to_u24(u32 v)
+{
+       if (CPU_IS_LITTLE_ENDIAN)
+               return v & 0xFFFFFF;
+       else
+               return v >> 8;
+}
+
+/*
+ * Load the next 3 bytes from the memory location @p into the 24 low-order bits
+ * of a 32-bit value.  The order in which the 3 bytes will be arranged as octets
+ * in the 24 bits is platform-dependent.  At least LOAD_U24_REQUIRED_NBYTES
+ * bytes must be available at @p; note that this may be more than 3.
+ */
+static forceinline u32
+load_u24_unaligned(const u8 *p)
+{
+#if UNALIGNED_ACCESS_IS_FAST
+#  define LOAD_U24_REQUIRED_NBYTES 4
+       return loaded_u32_to_u24(load_u32_unaligned(p));
+#else
+#  define LOAD_U24_REQUIRED_NBYTES 3
+#  if CPU_IS_BIG_ENDIAN
+       return ((u32)p[2] << 0) | ((u32)p[1] << 8) | ((u32)p[0] << 16);
+#  else
+       return ((u32)p[0] << 0) | ((u32)p[1] << 8) | ((u32)p[2] << 16);
+#  endif
+#endif
+}
+
 
 #endif /* _WIMLIB_UNALIGNED_H */