Add ceil implementation for 64-bit machines.

On 64-bit machines we should not split doubles into two 32 bit
integer and handle the words separately.  We have wide registers.
This patch implements a 64-bit ceil version.  Ideally all other
functions will be converted over time.

diff --git a/ChangeLog b/ChangeLog
index c29660d..b2d98e4 100644 (file)
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,12 @@
 2009-08-24  Ulrich Drepper  <drepper@redhat.com>
 
+       * math/math_private.h (ieee_double_shape_type): Add uint64_t word to
+       union.
+       (EXTRACT_WORDS64, INSERT_WORDS64): Define.
+       * sysdeps/x86_64/fpu/math_private.h (EXTRACT_WORDS64, INSERT_WORDS64):
+       Redefine.
+       * sysdeps/ieee754/dbl-64/wordsize-64/s_ceil.c: New file.
+
        * sysdeps/x86_64/fpu/math_private.h: Add specialized GET_FLOAT_WORD
        and SET_FLOAT_WORD definitions.
 

diff --git a/math/math_private.h b/math/math_private.h
index 129646f..fade7e1 100644 (file)
--- a/math/math_private.h
+++ b/math/math_private.h
@@ -17,6 +17,7 @@
 #define _MATH_PRIVATE_H_
 
 #include <endian.h>
+#include <stdint.h>
 #include <sys/types.h>
 
 /* The original fdlibm code used statements like:
@@ -43,6 +44,7 @@ typedef union
     u_int32_t msw;
     u_int32_t lsw;
   } parts;
+  uint64_t word;
 } ieee_double_shape_type;
 
 #endif
@@ -57,6 +59,7 @@ typedef union
     u_int32_t lsw;
     u_int32_t msw;
   } parts;
+  uint64_t word;
 } ieee_double_shape_type;
 
 #endif
@@ -89,6 +92,14 @@ do {                                                         \
   (i) = gl_u.parts.lsw;                                                \
 } while (0)
 
+/* Get all in one, efficient on 64-bit machines.  */
+#define EXTRACT_WORDS64(i,d)                                   \
+do {                                                           \
+  ieee_double_shape_type gh_u;                                 \
+  gh_u.value = (d);                                            \
+  (i) = gh_u.word;                                             \
+} while (0)
+
 /* Set a double from two 32 bit ints.  */
 
 #define INSERT_WORDS(d,ix0,ix1)                                        \
@@ -99,6 +110,14 @@ do {                                                                \
   (d) = iw_u.value;                                            \
 } while (0)
 
+/* Get all in one, efficient on 64-bit machines.  */
+#define INSERT_WORDS64(i,d)                                    \
+do {                                                           \
+  ieee_double_shape_type iw_u;                                 \
+  iw_u.word = (i);                                             \
+  (d) = iw_u.value;                                            \
+} while (0)
+
 /* Set the more significant 32 bits of a double from an int.  */
 
 #define SET_HIGH_WORD(d,v)                                     \

diff --git a/sysdeps/ieee754/dbl-64/wordsize-64/s_ceil.c b/sysdeps/ieee754/dbl-64/wordsize-64/s_ceil.c
new file mode 100644 (file)
index 0000000..9123fdc
--- /dev/null
+++ b/sysdeps/ieee754/dbl-64/wordsize-64/s_ceil.c
@@ -0,0 +1,67 @@
+/* @(#)s_ceil.c 5.1 93/09/24 */
+/*
+ * ====================================================
+ * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
+ *
+ * Developed at SunPro, a Sun Microsystems, Inc. business.
+ * Permission to use, copy, modify, and distribute this
+ * software is freely granted, provided that this notice
+ * is preserved.
+ * ====================================================
+ */
+
+/*
+ * ceil(x)
+ * Return x rounded toward -inf to integral value
+ * Method:
+ *     Bit twiddling.
+ * Exception:
+ *     Inexact flag raised if x not equal to ceil(x).
+ */
+
+#include "math.h"
+#include "math_private.h"
+
+#ifdef __STDC__
+static const double huge = 1.0e300;
+#else
+static double huge = 1.0e300;
+#endif
+
+#ifdef __STDC__
+       double __ceil(double x)
+#else
+       double __ceil(x)
+       double x;
+#endif
+{
+       int64_t i0,i;
+       int32_t j0;
+       EXTRACT_WORDS64(i0,x);
+       j0 = ((i0>>52)&0x7ff)-0x3ff;
+       if(j0<=51) {
+           if(j0<0) {  /* raise inexact if x != 0 */
+               if(huge+x>0.0) {/* return 0*sign(x) if |x|<1 */
+                   if(i0<0) {i0=INT64_C(0x8000000000000000);}
+                   else if(i0!=0) { i0=INT64_C(0x3ff0000000000000);}
+               }
+           } else {
+               i = INT64_C(0x000fffffffffffff)>>j0;
+               if((i0&i)==0) return x; /* x is integral */
+               if(huge+x>0.0) {        /* raise inexact flag */
+                   if(i0>0) i0 += UINT64_C(0x0010000000000000)>>j0;
+                   i0 &= (~i);
+               }
+           }
+       } else {
+           if(j0==0x400) return x+x;   /* inf or NaN */
+           else return x;              /* x is integral */
+       }
+       INSERT_WORDS64(x,i0);
+       return x;
+}
+weak_alias (__ceil, ceil)
+#ifdef NO_LONG_DOUBLE
+strong_alias (__ceil, __ceill)
+weak_alias (__ceil, ceill)
+#endif

diff --git a/sysdeps/x86_64/fpu/math_private.h b/sysdeps/x86_64/fpu/math_private.h
index 8f4b792..4be7536 100644 (file)
--- a/sysdeps/x86_64/fpu/math_private.h
+++ b/sysdeps/x86_64/fpu/math_private.h
@@ -22,6 +22,23 @@ while (0)
 /* We can do a few things better on x86-64.  */
 
 /* Direct movement of float into integer register.  */
+#undef EXTRACT_WORDS64
+#define EXTRACT_WORDS64(i,d)                                   \
+do {                                                           \
+  long int i_;                                                 \
+  asm ("movd %1, %0" : "=rm" (i_) : "x" (d));                  \
+  (i) = i_;                                                    \
+} while (0)
+
+/* And the reverse.  */
+#undef INSERT_WORDS64
+#define INSERT_WORDS64(d,i) \
+do {                                                           \
+  long int i_ = i;                                             \
+  asm ("movd %1, %0" : "=x" (d) : "rm" (i_));                  \
+} while (0)
+
+/* Direct movement of float into integer register.  */
 #undef GET_FLOAT_WORD
 #define GET_FLOAT_WORD(i,d) \
 do {                                                           \