remove unaligned write for #ifdef i386 (it was technically undefined

behavior and the compiler is now smart enough to turn the more elaborate
code into a single store)

test/uint.c

@@ -1,8 +1,8 @@
 #include <assert.h>
-#include <uint16.h>
-#include <uint32.h>
-#include <uint64.h>
-#include <byte.h>
+#include <libowfat/uint16.h>
+#include <libowfat/uint32.h>
+#include <libowfat/uint64.h>
+#include <libowfat/byte.h>
 
 int main() {
   char buf[8];
@@ -11,36 +11,38 @@ int main() {
     uint16 a;
 
     buf[0]=buf[1]=0;
-    uint16_pack_big(buf,0x1234);
-    assert(buf[0]==0x12 && buf[1]==0x34);
+    uint16_pack_big(buf,0xabcd);
+    assert((unsigned char)buf[0]==0xab && (unsigned char)buf[1]==0xcd);
     uint16_unpack_big(buf,&a);
-    assert(a==0x1234);
-    assert(uint16_read_big(buf)==0x1234);
+    assert(a==0xabcd);
+    assert(uint16_read_big(buf)==0xabcd);
 
     buf[0]=buf[1]=0;
-    uint16_pack(buf,0x1234);
-    assert(buf[0]==0x34 && buf[1]==0x12);
+    uint16_pack(buf,0xabcd);
+    assert((unsigned char)buf[0]==0xcd && (unsigned char)buf[1]==0xab);
     uint16_unpack(buf,&a);
-    assert(a==0x1234);
-    assert(uint16_read(buf)==0x1234);
+    assert(a==0xabcd);
+    assert(uint16_read(buf)==0xabcd);
   }
 
   {
     uint32 a;
 
     buf[0]=buf[1]=buf[2]=buf[3]=0;
-    uint32_pack_big(buf,0x12345678);
-    assert(buf[0]==0x12 && buf[1]==0x34 && buf[2]==0x56 && buf[3]==0x78);
+    uint32_pack_big(buf,0x89abcdef);
+    assert((unsigned char)buf[0]==0x89 && (unsigned char)buf[1]==0xab &&
+	   (unsigned char)buf[2]==0xcd && (unsigned char)buf[3]==0xef);
     uint32_unpack_big(buf,&a);
-    assert(a==0x12345678);
-    assert(uint32_read_big(buf)==0x12345678);
+    assert(a==0x89abcdef);
+    assert(uint32_read_big(buf)==0x89abcdef);
 
     buf[0]=buf[1]=buf[2]=buf[3]=0;
-    uint32_pack(buf,0x12345678);
-    assert(buf[0]==0x78 && buf[1]==0x56 && buf[2]==0x34 && buf[3]==0x12);
+    uint32_pack(buf,0x89abcdef);
+    assert((unsigned char)buf[3]==0x89 && (unsigned char)buf[2]==0xab &&
+	   (unsigned char)buf[1]==0xcd && (unsigned char)buf[0]==0xef);
     uint32_unpack(buf,&a);
-    assert(a==0x12345678);
-    assert(uint32_read(buf)==0x12345678);
+    assert(a==0x89abcdef);
+    assert(uint32_read(buf)==0x89abcdef);
   }
 
   {
@@ -48,18 +50,18 @@ int main() {
     unsigned int i;
 
     byte_zero(buf,sizeof(buf));
-    uint64_pack_big(buf,0x0102030405060708ull);
-    for (i=0; i<8; ++i) assert(buf[i]==i+1);
+    uint64_pack_big(buf,0x8182838485868788ull);
+    for (i=0; i<8; ++i) assert((unsigned char)buf[i]==i+0x81);
     uint64_unpack_big(buf,&a);
-    assert(a==0x0102030405060708ull);
-    assert(uint64_read_big(buf)==0x0102030405060708ull);
+    assert(a==0x8182838485868788ull);
+    assert(uint64_read_big(buf)==0x8182838485868788ull);
 
     byte_zero(buf,sizeof(buf));
-    uint64_pack(buf,0x0102030405060708ull);
-    for (i=0; i<8; ++i) assert(buf[7-i]==i+1);
+    uint64_pack(buf,0x8182838485868788ull);
+    for (i=0; i<8; ++i) assert((unsigned char)buf[7-i]==i+0x81);
     uint64_unpack(buf,&a);
-    assert(a==0x0102030405060708ull);
-    assert(uint64_read(buf)==0x0102030405060708ull);
+    assert(a==0x8182838485868788ull);
+    assert(uint64_read(buf)==0x8182838485868788ull);
   }
 
 

uint16.h

@@ -14,42 +14,42 @@ typedef int16_t int16;
 
 #if (defined(__i386__) || defined(__x86_64__)) && !defined(NO_UINT16_MACROS)
 
+/* On x86 and x64 unaligned access are OK and the compiler knows it and
+ * will collapse this long-looking code into a single load or store
+ * instruction, so turn it into inline functions */
+
 static inline void uint16_pack(char* out,uint16 in) {
-  *(uint16*)out=in;
+  out[0]=in;
+  out[1]=in>>8;
 }
 
 static inline void uint16_unpack(const char *in,uint16* out) {
-  *out=*(const uint16*)in;
+  const unsigned char* i = (const unsigned char*)in;
+  *out=i[0] | (i[1]<<8);
 }
 
 static inline uint16 uint16_read(const char* in) {
-  return *(const uint16*)in;
+  const unsigned char* i = (const unsigned char*)in;
+  return i[0] | (i[1]<<8);
 }
 
-#if defined(__x86_64__) && defined(__GNUC__)
-
 static inline void uint16_pack_big(char* out,uint16 in) {
-  *(uint16*)out=__builtin_bswap16(in);
+  out[0]=in>>8;
+  out[1]=in;
 }
 
-static inline void uint16_unpack_big(const char in[2],uint16* out) {
-  *out=__builtin_bswap16(*(const uint16*)in);
+static inline void uint16_unpack_big(const char *in,uint16* out) {
+  const unsigned char* i = (const unsigned char*)in;
+  *out=(i[0]<<8) | i[1];
 }
 
-static inline uint16 uint16_read_big(const char in[2]) {
-  return __builtin_bswap16(*(const uint16*)in);
+static inline uint16 uint16_read_big(const char* in) {
+  const unsigned char* i = (const unsigned char*)in;
+  return (i[0]<<8) | i[1];
 }
 
 #else
 
-void uint16_pack_big(char *out,uint16 in);
-void uint16_unpack_big(const char *in,uint16* out);
-uint16 uint16_read_big(const char *in);
-
-#endif
-
-#else
-
 void uint16_pack(char *out,uint16 in);
 void uint16_pack_big(char *out,uint16 in);
 void uint16_unpack(const char *in,uint16* out);

uint32.h

@@ -14,42 +14,46 @@ typedef int32_t int32;
 
 #if (defined(__i386__) || defined(__x86_64__)) && !defined(NO_UINT32_MACROS)
 
+/* On x86 and x64 unaligned access are OK and the compiler knows it and
+ * will collapse this long-looking code into a single load or store
+ * instruction, so turn it into inline functions */
+
 static inline void uint32_pack(char* out,uint32 in) {
-  *(uint32*)out=in;
+  out[0]=in;
+  out[1]=in>>8;
+  out[2]=in>>16;
+  out[3]=in>>24;
 }
 
 static inline void uint32_unpack(const char *in,uint32* out) {
-  *out=*(const uint32*)in;
+  const unsigned char* i = (const unsigned char*)in;
+  *out=i[0] | (i[1]<<8) | (i[2]<<16) | (i[3]<<24);
 }
 
 static inline uint32 uint32_read(const char* in) {
-  return *(const uint32*)in;
+  const unsigned char* i = (const unsigned char*)in;
+  return i[0] | (i[1]<<8) | (i[2]<<16) | (i[3]<<24);
 }
 
-#if defined(__x86_64__) && defined(__GNUC__)
-
 static inline void uint32_pack_big(char *out,uint32 in) {
-  *(uint32*)out=__builtin_bswap32(in);
+  out[0]=in>>24;
+  out[1]=in>>16;
+  out[2]=in>>8;
+  out[3]=in;
 }
 
 static inline void uint32_unpack_big(const char *in,uint32* out) {
-  *out=__builtin_bswap32(*(const uint32*)in);
+  const unsigned char* i = (const unsigned char*)in;
+  *out=(i[0]<<24) | (i[1]<<16) | (i[2]<<8) | i[3];
 }
 
 static inline uint32 uint32_read_big(const char *in) {
-  return __builtin_bswap32(*(const uint32*)in);
+  const unsigned char* i = (const unsigned char*)in;
+  return (i[0]<<24) | (i[1]<<16) | (i[2]<<8) | i[3];
 }
 
 #else
 
-void uint32_pack_big(char *out,uint32 in);
-void uint32_unpack_big(const char *in,uint32* out);
-uint32 uint32_read_big(const char *in);
-
-#endif
-
-#else
-
 void uint32_pack(char *out,uint32 in);
 void uint32_pack_big(char *out,uint32 in);
 void uint32_unpack(const char *in,uint32* out);

uint64.h

@@ -13,12 +13,61 @@ typedef uint64_t uint64;
 typedef int64_t int64;
 
 #if (defined(__i386__) || defined(__x86_64__)) && !defined(NO_UINT64_MACROS)
-#define uint64_pack(out,in) (*(uint64*)(out)=(in))
-#define uint64_unpack(in,out) (*(out)=*(uint64*)(in))
-#define uint64_read(in) (*(uint64*)(in))
-void uint64_pack_big(char *out,uint64 in);
-void uint64_unpack_big(const char *in,uint64* out);
-uint64 uint64_read_big(const char *in);
+
+/* On x86 and x64 unaligned access are OK and the compiler knows it and
+ * will collapse this long-looking code into a single load or store
+ * instruction, so turn it into inline functions */
+
+static inline void uint64_pack(char* out,uint64 in) {
+  out[0]=in;
+  out[1]=in>>8;
+  out[2]=in>>16;
+  out[3]=in>>24;
+  out[4]=in>>32;
+  out[5]=in>>40;
+  out[6]=in>>48;
+  out[7]=in>>56;
+}
+
+static inline void uint64_unpack(const char *in,uint64* out) {
+  const unsigned char* i = (const unsigned char*)in;
+  *out=i[0] | (i[1]<<8) | (i[2]<<16) | ((uint64)i[3]<<24) |
+    ((uint64)i[4]<<32) | ((uint64)i[5]<<40) |
+    ((uint64)i[6]<<48) | ((uint64)i[7]<<56);
+}
+
+static inline uint64 uint64_read(const char *in) {
+  const unsigned char* i = (const unsigned char*)in;
+  return i[0] | (i[1]<<8) | (i[2]<<16) | ((uint64)i[3]<<24) |
+    ((uint64)i[4]<<32) | ((uint64)i[5]<<40) |
+    ((uint64)i[6]<<48) | ((uint64)i[7]<<56);
+}
+
+static inline void uint64_pack_big(char *out,uint64 in) {
+  out[0]=in>>56;
+  out[1]=in>>48;
+  out[2]=in>>40;
+  out[3]=in>>32;
+  out[4]=in>>24;
+  out[5]=in>>16;
+  out[6]=in>>8;
+  out[7]=in;
+}
+
+static inline void uint64_unpack_big(const char *in,uint64* out) {
+  const unsigned char* i = (const unsigned char*)in;
+  *out=(i[7] | (i[6]<<8) | (i[5]<<16) | ((uint64)(i[4])<<24)) |
+    ((uint64)(i[3])<<32) | ((uint64)(i[2])<<40) |
+    ((uint64)(i[1])<<48) | ((uint64)(i[0])<<56);
+}
+
+static inline uint64 uint64_read_big(const char *in) {
+  const unsigned char* i = (const unsigned char*)in;
+  return (i[7] | (i[6]<<8) | (i[5]<<16) | ((uint64)(i[4])<<24)) |
+    ((uint64)(i[3])<<32) | ((uint64)(i[2])<<40) |
+    ((uint64)(i[1])<<48) | ((uint64)(i[0])<<56);
+}
+
 #else
 
 void uint64_pack(char *out,uint64 in);