x86-host-specific performance improvement

If we're on an x86 host, we can do unaligned littleendian memory
references directly.  Just do'em.

nasmlib.c

@@ -361,6 +361,34 @@ int32_t seg_alloc(void)
     return (next_seg += 2) - 2;
 }
 
+#if X86_MEMORY
+
+void fwriteint16_t(int data, FILE * fp)
+{
+    uint16_t d = data;
+    fwrite(&d, 1, 2, fp);
+}
+
+void fwriteint32_t(int32_t data, FILE * fp)
+{
+    uint32_t d = data;
+    fwrite(&d, 1, 4, fp);
+}
+
+void fwriteint64_t(int64_t data, FILE * fp)
+{
+    uint64_t d = data;
+    fwrite(&d, 1, 8, fp);
+}
+
+void fwriteaddr(int64_t data, int size, FILE * fp)
+{
+    uint64_t d = data;
+    fwrite(&d, 1, size, fp);
+}
+
+#else /* !X86_MEMORY */
+
 void fwriteint16_t(int data, FILE * fp)
 {
     char buffer[2], *p = buffer;
@@ -389,6 +417,8 @@ void fwriteaddr(int64_t data, int size, FILE * fp)
     fwrite(buffer, 1, size, fp);
 }
 
+#endif
+
 void standard_extension(char *inname, char *outname, char *extension,
                         efunc error)
 {

nasmlib.h

@@ -178,6 +178,41 @@ void standard_extension(char *inname, char *outname, char *extension,
  * format in memory
  */
 
+#if X86_MEMORY
+
+#define WRITECHAR(p,v)				\
+    do {					\
+	*(uint8_t *)(p) = (v);			\
+	(p) += 1;				\
+    } while (0)
+
+#define WRITESHORT(p,v)				\
+    do {					\
+	*(uint16_t *)(p) = (v);			\
+	(p) += 2;				\
+    } while (0)
+
+#define WRITELONG(p,v)				\
+    do {					\
+	*(uint32_t *)(p) = (v);			\
+	(p) += 4;				\
+    } while (0)
+
+#define WRITEDLONG(p,v)				\
+    do {					\
+	*(uint64_t *)(p) = (v);			\
+	(p) += 8;				\
+    } while (0)
+
+#define WRITEADDR(p,v,s)			\
+    do {					\
+	uint64_t _v = (v);			\
+	memcpy((p), &_v, (s));			\
+	(p) += (s);				\
+    } while (0)
+
+#else /* !X86_MEMORY */
+
 #define WRITECHAR(p,v) \
   do { \
     *(p)++ = (v) & 0xFF; \
@@ -219,6 +254,8 @@ void standard_extension(char *inname, char *outname, char *extension,
 	}					\
     } while(0)
 
+#endif
+
 /*
  * and routines to do the same thing to a file
  */