Optimized intkorr() and intstore(functions) for intel 64 bits.

(Benchmarked, about 30 % faster and 50 % smaller than original)

include/byte_order_generic_x86_64.h

@@ -16,6 +16,7 @@
 /*
   Optimized function-like macros for the x86 architecture (_WIN32 included).
 */
+
 #define sint2korr(A)	(int16) (*((int16 *) (A)))
 #define sint3korr(A)	((int32) ((((uchar) (A)[2]) & 128) ? \
 				  (((uint32) 255L << 24) | \
@@ -39,17 +40,21 @@
 #define uint3korr(A)	(uint32) (*((unsigned int *) (A)) & 0xFFFFFF)
 #endif
 #define uint4korr(A)	(uint32) (*((uint32 *) (A)))
-#define uint5korr(A)	((ulonglong)(((uint32) ((uchar) (A)[0])) +\
+
-				    (((uint32) ((uchar) (A)[1])) << 8) +\
+
-				    (((uint32) ((uchar) (A)[2])) << 16) +\
+static inline ulonglong uint5korr(const void *p)
-				    (((uint32) ((uchar) (A)[3])) << 24)) +\
+{
-				    (((ulonglong) ((uchar) (A)[4])) << 32))
+  ulonglong a= *(uint32 *) p;
-#define uint6korr(A)	((ulonglong)(((uint32)    ((uchar) (A)[0]))          + \
+  ulonglong b= *(4 + (uchar *) p);
-                                     (((uint32)    ((uchar) (A)[1])) << 8)   + \
+  return a | (b << 32);
-                                     (((uint32)    ((uchar) (A)[2])) << 16)  + \
+}
-                                     (((uint32)    ((uchar) (A)[3])) << 24)) + \
+static inline ulonglong uint6korr(const void *p)
-                         (((ulonglong) ((uchar) (A)[4])) << 32) +       \
+{
-                         (((ulonglong) ((uchar) (A)[5])) << 40))
+  ulonglong a= *(uint32 *) p;
+  ulonglong b= *(uint16 *) (4 + (char *) p);
+  return a | (b << 32);
+}
+
 #define uint8korr(A)	(ulonglong) (*((ulonglong *) (A)))
 #define sint8korr(A)	(longlong) (*((longlong *) (A)))
 
@@ -61,23 +66,67 @@
                             *(T+1)=(uchar) (((uint) (A) >> 8));\
                             *(T+2)=(uchar) (((A) >> 16));\
                         } while (0)
+
 #define int4store(T,A)	do { uchar *pT= (uchar*)(T);\
                              *((uint32 *) (pT))= (uint32) (A); \
                         } while (0)
 
-#define int5store(T,A)  do { *(T)= (uchar)((A));\
+#define int5store(T,A)  do { uchar *pT= (uchar*)(T);\
-                             *((T)+1)=(uchar) (((A) >> 8));\
+                             *((uint32 *) (pT))= (uint32) (A); \
-                             *((T)+2)=(uchar) (((A) >> 16));\
+                             *((pT)+4)=(uchar) (((A) >> 32));\
-                             *((T)+3)=(uchar) (((A) >> 24));\
+                        } while (0)
-                             *((T)+4)=(uchar) (((A) >> 32));\
+
-                        } while(0)
+#define int6store(T,A)  do { uchar *pT= (uchar*)(T);\
-#define int6store(T,A)  do { *(T)=    (uchar)((A));          \
+                             *((uint32 *) (pT))= (uint32) (A); \
-                             *((T)+1)=(uchar) (((A) >> 8));  \
+                             *((uint16*)(pT+4))= (uint16) (A >> 32);\
-                             *((T)+2)=(uchar) (((A) >> 16)); \
+                        } while (0)
-                             *((T)+3)=(uchar) (((A) >> 24)); \
+
-                             *((T)+4)=(uchar) (((A) >> 32)); \
-                             *((T)+5)=(uchar) (((A) >> 40)); \
-                        } while(0)
 #define int8store(T,A)	do { uchar *pT= (uchar*)(T);\
                              *((ulonglong *) (pT))= (ulonglong) (A);\
                         } while(0)
+
+#if defined(__GNUC__)
+
+#define HAVE_mi_uint5korr
+#define HAVE_mi_uint6korr
+#define HAVE_mi_uint7korr
+#define HAVE_mi_uint78orr
+
+/* Read numbers stored in high-bytes-first order */
+
+static inline ulonglong mi_uint5korr(const void *p)
+{
+  ulonglong a= *(uint32 *) p;
+  ulonglong b= *(4 + (uchar *) p);
+  ulonglong v= (a | (b << 32)) << 24;
+  asm ("bswapq %0" : "=r" (v) : "0" (v));
+  return v;
+}
+
+static inline ulonglong mi_uint6korr(const void *p)
+{
+  ulonglong a= *(uint32 *) p;
+  ulonglong b= *(uint16 *) (4 + (char *) p);
+  ulonglong v= (a | (b << 32)) << 16;
+  asm ("bswapq %0" : "=r" (v) : "0" (v));
+  return v;
+}
+
+static inline ulonglong mi_uint7korr(const void *p)
+{
+  ulonglong a= *(uint32 *) p;
+  ulonglong b= *(uint16 *) (4 + (char *) p);
+  ulonglong c= *(6 + (uchar *) p);
+  ulonglong v= (a | (b << 32) | (c << 48)) << 8;
+  asm ("bswapq %0" : "=r" (v) : "0" (v));
+  return v;
+}
+
+static inline ulonglong mi_uint8korr(const void *p)
+{
+  ulonglong v= *(ulonglong *) p;
+  asm ("bswapq %0" : "=r" (v) : "0" (v));
+  return v;
+}
+
+#endif

include/myisampack.h

@@ -52,11 +52,16 @@
                                    (((uint32) (((const uchar*) (A))[2])) << 8) +\
                                    (((uint32) (((const uchar*) (A))[1])) << 16) +\
                                    (((uint32) (((const uchar*) (A))[0])) << 24)))
+
+#ifndef HAVE_mi_uint5korr
 #define mi_uint5korr(A) ((ulonglong)(((uint32) (((const uchar*) (A))[4])) +\
                                     (((uint32) (((const uchar*) (A))[3])) << 8) +\
                                     (((uint32) (((const uchar*) (A))[2])) << 16) +\
                                     (((uint32) (((const uchar*) (A))[1])) << 24)) +\
                                     (((ulonglong) (((const uchar*) (A))[0])) << 32))
+#endif /* HAVE_mi_uint5korr */
+
+#ifndef HAVE_mi_uint6korr
 #define mi_uint6korr(A) ((ulonglong)(((uint32) (((const uchar*) (A))[5])) +\
                                     (((uint32) (((const uchar*) (A))[4])) << 8) +\
                                     (((uint32) (((const uchar*) (A))[3])) << 16) +\
@@ -64,6 +69,9 @@
                         (((ulonglong) (((uint32) (((const uchar*) (A))[1])) +\
                                     (((uint32) (((const uchar*) (A))[0]) << 8)))) <<\
                                      32))
+#endif /* HAVE_mi_uint6korr */
+
+#ifndef HAVE_mi_uint7korr
 #define mi_uint7korr(A) ((ulonglong)(((uint32) (((const uchar*) (A))[6])) +\
                                     (((uint32) (((const uchar*) (A))[5])) << 8) +\
                                     (((uint32) (((const uchar*) (A))[4])) << 16) +\
@@ -72,6 +80,9 @@
                                     (((uint32) (((const uchar*) (A))[1])) << 8) +\
                                     (((uint32) (((const uchar*) (A))[0])) << 16))) <<\
                                      32))
+#endif /* HAVE_mi_uint7korr */
+
+#ifndef HAVE_mi_uint8korr
 #define mi_uint8korr(A) ((ulonglong)(((uint32) (((const uchar*) (A))[7])) +\
                                     (((uint32) (((const uchar*) (A))[6])) << 8) +\
                                     (((uint32) (((const uchar*) (A))[5])) << 16) +\
@@ -81,6 +92,7 @@
                                     (((uint32) (((const uchar*) (A))[1])) << 16) +\
                                     (((uint32) (((const uchar*) (A))[0])) << 24))) <<\
                                     32))
+#endif /* HAVE_mi_uint8korr */
 
 /* This one is for uniformity */
 #define mi_int1store(T,A) *((uchar*)(T))= (uchar) (A)