crosstool-ng/patches/gmp/4.2.4/100-mpf_eq.patch
Yann E. MORIN" 20945a3465 Add patchset for MGP-4.2.4, from upstream.
/trunk/patches/gmp/4.2.4/120-perfpow.patch     |  149   149     0     0 ++++++++++++++++++++++
 /trunk/patches/gmp/4.2.4/110-mpf_set_str.patch |   32    32     0     0 +++++
 /trunk/patches/gmp/4.2.4/100-mpf_eq.patch      |  219   219     0     0 ++++++++++++++++++++++++++++++++
 3 files changed, 400 insertions(+)
2009-03-08 18:04:56 +00:00

220 lines
7.1 KiB
Diff

Original patch from: mpf_eq.diff
-= BEGIN original header =-
-= END original header =-
diff -durN gmp-4.2.4.orig/doc/gmp.texi gmp-4.2.4/doc/gmp.texi
--- gmp-4.2.4.orig/doc/gmp.texi 2008-09-18 17:36:14.000000000 +0200
+++ gmp-4.2.4/doc/gmp.texi 2009-03-08 18:36:16.000000000 +0100
@@ -4849,9 +4849,12 @@
equal, zero otherwise. I.e., test if @var{op1} and @var{op2} are approximately
equal.
-Caution: Currently only whole limbs are compared, and only in an exact
-fashion. In the future values like 1000 and 0111 may be considered the same
-to 3 bits (on the basis that their difference is that small).
+Caution 1: All version of GMP up to version 4.2.4 compared just whole limbs,
+meaning sometimes more than @var{op3} bits, sometimes fewer.
+
+Caution 2: This function will consider XXX11...111 and XX100...000 different,
+even if ... is replaced by a semi-infinite number of bits. Such numbers are
+really just one ulp off, and should be considered equal.
@end deftypefun
@deftypefun void mpf_reldiff (mpf_t @var{rop}, mpf_t @var{op1}, mpf_t @var{op2})
diff -durN gmp-4.2.4.orig/mpf/eq.c gmp-4.2.4/mpf/eq.c
--- gmp-4.2.4.orig/mpf/eq.c 2007-08-30 20:31:40.000000000 +0200
+++ gmp-4.2.4/mpf/eq.c 2009-03-08 18:36:16.000000000 +0100
@@ -1,6 +1,6 @@
/* mpf_eq -- Compare two floats up to a specified bit #.
-Copyright 1993, 1995, 1996, 2001, 2002 Free Software Foundation, Inc.
+Copyright 1993, 1995, 1996, 2001, 2002, 2008 Free Software Foundation, Inc.
This file is part of the GNU MP Library.
@@ -19,6 +19,7 @@
#include "gmp.h"
#include "gmp-impl.h"
+#include "longlong.h"
int
mpf_eq (mpf_srcptr u, mpf_srcptr v, unsigned long int n_bits)
@@ -26,6 +27,8 @@
mp_srcptr up, vp;
mp_size_t usize, vsize, size, i;
mp_exp_t uexp, vexp;
+ mp_limb_t diff;
+ int cnt;
uexp = u->_mp_exp;
vexp = v->_mp_exp;
@@ -53,10 +56,8 @@
/* U and V have the same sign and are both non-zero. */
/* 2. Are the exponents different? */
- if (uexp > vexp)
- return 0; /* ??? handle (uexp = vexp + 1) */
- if (vexp > uexp)
- return 0; /* ??? handle (vexp = uexp + 1) */
+ if (uexp != vexp)
+ return 0;
usize = ABS (usize);
vsize = ABS (vsize);
@@ -93,17 +94,26 @@
size = usize;
}
- if (size > (n_bits + GMP_NUMB_BITS - 1) / GMP_NUMB_BITS)
- size = (n_bits + GMP_NUMB_BITS - 1) / GMP_NUMB_BITS;
+ up += usize; /* point just above most significant limb */
+ vp += vsize; /* point just above most significant limb */
- up += usize - size;
- vp += vsize - size;
+ count_leading_zeros (cnt, up[-1]);
+ if ((vp[-1] >> (GMP_LIMB_BITS - 1 - cnt)) != 1)
+ return 0; /* msb positions different */
- for (i = size - 1; i >= 0; i--)
+ n_bits += cnt - GMP_NAIL_BITS;
+
+ size = MIN (size, (n_bits + GMP_NUMB_BITS - 1) / GMP_NUMB_BITS);
+
+ up -= size; /* point at least significant relevant limb */
+ vp -= size; /* point at least significant relevant limb */
+
+ for (i = size - 1; i > 0; i--)
{
if (up[i] != vp[i])
return 0;
}
- return 1;
+ diff = (up[0] ^ vp[0]) >> GMP_NUMB_BITS - 1 - (n_bits - 1) % GMP_NUMB_BITS;
+ return diff == 0;
}
diff -durN gmp-4.2.4.orig/tests/cxx/t-prec.cc gmp-4.2.4/tests/cxx/t-prec.cc
--- gmp-4.2.4.orig/tests/cxx/t-prec.cc 2007-09-01 12:09:03.000000000 +0200
+++ gmp-4.2.4/tests/cxx/t-prec.cc 2009-03-08 18:36:16.000000000 +0100
@@ -1,6 +1,6 @@
/* Test precision of mpf_class expressions.
-Copyright 2001, 2002, 2003 Free Software Foundation, Inc.
+Copyright 2001, 2002, 2003, 2008 Free Software Foundation, Inc.
This file is part of the GNU MP Library.
@@ -61,7 +61,7 @@
g = 1 / f;
ASSERT_ALWAYS_PREC
(g, "0.11111 11111 11111 11111 11111 11111 11111 11111 11111 11111"
- " 11111 11111 11111 11111 11111 11", very_large_prec);
+ " 11111 11111 11111 11111 11111 111", very_large_prec);
}
{
mpf_class f(15.0, large_prec);
@@ -69,7 +69,7 @@
g = 1 / f;
ASSERT_ALWAYS_PREC
(g, "0.06666 66666 66666 66666 66666 66666 66666 66666 66666 66666"
- " 66666 66666 66666 66666 66666 67", very_large_prec);
+ " 66666 66666 66666 66666 66666 667", very_large_prec);
}
// compound expressions
@@ -94,14 +94,14 @@
i = f / g + h;
ASSERT_ALWAYS_PREC
(i, "15.33333 33333 33333 33333 33333 33333 33333 33333 33333 33333"
- " 33333 33333 33333 333", very_large_prec);
+ " 33333 33333 33333 33333 33333 3", very_large_prec);
}
{
mpf_class f(3.0, small_prec);
mpf_class g(-(1 + f) / 3, very_large_prec);
ASSERT_ALWAYS_PREC
(g, "-1.33333 33333 33333 33333 33333 33333 33333 33333 33333 33333"
- " 33333 33333 33333 333", very_large_prec);
+ " 33333 33333 33333 33333 33333 33", very_large_prec);
}
{
mpf_class f(9.0, medium_prec);
@@ -117,7 +117,7 @@
g = hypot(1 + 5 / f, 1.0);
ASSERT_ALWAYS_PREC
(g, "1.66666 66666 66666 66666 66666 66666 66666 66666 66666 66666"
- " 66666 66666 66666 667", very_large_prec);
+ " 66666 66666 66666 66666 66666 67", very_large_prec);
}
// compound assignments
@@ -142,7 +142,7 @@
mpf_class g(0.0, very_large_prec);
g = mpf_class(1 / f);
ASSERT_ALWAYS_PREC
- (g, "0.11111 11111 11111 11111 11111 11111 11111 111", medium_prec);
+ (g, "0.11111 11111 11111 11111 11111 11111 11111 1111", medium_prec);
}
{
mpf_class f(15.0, large_prec);
@@ -150,7 +150,7 @@
g = mpf_class(1 / f);
ASSERT_ALWAYS_PREC
(g, "0.06666 66666 66666 66666 66666 66666 66666 66666 66666 66666"
- " 66666 667", large_prec);
+ " 66666 6667", large_prec);
}
{
@@ -158,7 +158,8 @@
mpf_class h(0.0, very_large_prec);
h = mpf_class(f / g + 1, large_prec);
ASSERT_ALWAYS_PREC
- (h, "1.33333 33333 33333 33333 33333 33333 33333 33333 33333 3333",
+ (h, "1.33333 33333 33333 33333 33333 33333 33333 33333 33333 33333"
+ " 33333 333",
large_prec);
}
@@ -170,7 +171,7 @@
g = f - q;
ASSERT_ALWAYS_PREC
(g, "2.66666 66666 66666 66666 66666 66666 66666 66666 66666 66666"
- " 66666 66666 66666 667", very_large_prec);
+ " 66666 66666 66666 66666 66666 67", very_large_prec);
}
{
@@ -179,7 +180,8 @@
mpf_class g(0.0, very_large_prec);
g = mpf_class(f - q, large_prec);
ASSERT_ALWAYS_PREC
- (g, "2.66666 66666 66666 66666 66666 66666 66666 66666 66666 6667",
+ (g, "2.66666 66666 66666 66666 66666 66666 66666 66666 66666 66666"
+ " 66666 667",
large_prec);
}
{
@@ -188,7 +190,7 @@
mpf_class g(0.0, very_large_prec);
g = mpf_class(f - q);
ASSERT_ALWAYS_PREC
- (g, "2.66666 66666 66666 66666 66666 6667", medium_prec);
+ (g, "2.66666 66666 66666 66666 66666 66666 66666 667", medium_prec);
}
{
mpf_class f(15.0, large_prec);
@@ -196,7 +198,8 @@
mpf_class g(0.0, very_large_prec);
g = mpf_class(f + q);
ASSERT_ALWAYS_PREC
- (g, "15.33333 33333 33333 33333 33333 33333 33333 33333 33333 3333",
+ (g, "15.33333 33333 33333 33333 33333 33333 33333 33333 33333 33333"
+ " 33333 33",
large_prec);
}
}