|
 |
8ae002 |
Partial backports of:
|
|
|
8ae002 |
=====================
|
|
|
8ae002 |
|
|
|
8ae002 |
commit c5d5d574cbfa96d0f6c1db24d1e072c472627e41
|
|
|
8ae002 |
Author: Ondřej Bílka <neleai@seznam.cz>
|
|
|
8ae002 |
Date: Thu Oct 17 16:03:24 2013 +0200
|
|
|
8ae002 |
|
|
|
8ae002 |
Format floating routines.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit da08f647d58d674db08cdb3e61c8826c89470e2e
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Fri Dec 21 09:15:10 2012 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Move more constants into static variables
|
|
|
8ae002 |
|
|
|
8ae002 |
Code cleanup.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit f93a8d15699ee699282465dc1e03e033f3fabb52
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Wed Jan 16 16:06:48 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Consolidate constant defines into mpa.h
|
|
|
8ae002 |
|
|
|
8ae002 |
commit caa99d06e7f1403887294442af520b0f8c6f3de0
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Fri Jan 18 11:18:13 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Simplify calculation of 2^-m in __mpexp
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 107a5bf085f5c4ef8c28266a34d476724cfc3475
|
|
|
8ae002 |
Author: Joseph Myers <joseph@codesourcery.com>
|
|
|
8ae002 |
Date: Tue Nov 18 15:40:56 2014 +0000
|
|
|
8ae002 |
|
|
|
8ae002 |
Fix libm mpone, mptwo namespace (bug 17616).
|
|
|
8ae002 |
|
|
|
8ae002 |
To provided __mptwo for __inv.
|
|
|
8ae002 |
|
|
|
8ae002 |
Full backports of the following:
|
|
|
8ae002 |
================================
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 44e0d4c20ce5bf3825897e5d4b7caae94016214d
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Wed Jan 2 11:44:13 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Split mantissa calculation loop and add branch prediction
|
|
|
8ae002 |
|
|
|
8ae002 |
commit f8af25d218202ff2f5d167b8e44e4b79f91d147f
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Fri Jan 4 15:09:33 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Remove commented declarations
|
|
|
8ae002 |
|
|
|
8ae002 |
commit a9e48ab40e230c7fe34e4892bec8af4f3f975a20
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Fri Jan 4 15:42:09 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Clean up comment for MP_NO
|
|
|
8ae002 |
|
|
|
8ae002 |
commit fffb407f4668b40b3df1eb8ee3ae3bc64ee79e20
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Fri Jan 4 22:52:12 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Remove unused __cr and __cpymn
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 950c99ca9094e7dc6394e90395f51e12093393aa
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Wed Jan 9 19:07:15 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Update comments in mpa.c
|
|
|
8ae002 |
|
|
|
8ae002 |
Fixed comment style and clearer wording in some cases.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 1066a53440d2744566e97c59bcd0d422186b3e90
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Mon Jan 14 21:31:25 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Fix code formatting in mpa.c
|
|
|
8ae002 |
|
|
|
8ae002 |
This includes the overridden mpa.c in power4.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 2a91b5735ac1bc65ce5c2a3646d75ba7208e26e9
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Mon Jan 14 21:36:58 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Minor tweak to mp multiplication
|
|
|
8ae002 |
|
|
|
8ae002 |
Add a local variable to remove extra copies to/from memory in the Z
|
|
|
8ae002 |
array.
|
|
|
8ae002 |
|
|
|
8ae002 |
ommit 45f058844c33f670475bd02f266942746bcb332b
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Tue Feb 26 21:28:16 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Another tweak to the multiplication algorithm
|
|
|
8ae002 |
|
|
|
8ae002 |
Reduce the formula to calculate mantissa so that we reduce the net
|
|
|
8ae002 |
number of multiplications performed.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit bab8a695ee79a5a6e9b2b699938952b006fcbbec
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Thu Feb 21 14:29:18 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Fix whitespace differences between generic and powerpc mpa.c
|
|
|
8ae002 |
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 2d0e0f29f85036d1189231cb7c1f19f27ba14a89
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Fri Feb 15 23:56:20 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Fix determination of lower precision in __mul
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 909279a5cfa938c989c9b01c8f48a0276291ec45
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Wed Feb 13 14:16:23 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Optimized mp multiplication
|
|
|
8ae002 |
|
|
|
8ae002 |
Don't bother multiplying zeroes since that only wastes cycles.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit bdf028142eb77d6ae59500db875068fa5d7b059d
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Wed Feb 13 13:55:29 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Clean up add_magnitudes and sub_magnitudes
|
|
|
8ae002 |
|
|
|
8ae002 |
commit d6752ccd696c71d23cd3df8fb9cc60b61c32e65a
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Thu Feb 14 10:31:09 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
New __sqr function as a faster special case of __mul
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 4709fe7602b56e9f6ee1ab6afb4067409a784f29
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Sat Feb 16 00:09:29 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Use intermediate variable to compute exponent in __mul
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 20cd7fb3ae63795ae7c9a464abf5ed19b364ade0
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Wed Feb 20 18:56:20 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Copy comment about inner loop from powerpc mpa.c to the default one
|
|
|
8ae002 |
|
|
|
8ae002 |
commit e69804d14e43f14c3c65dc570afdbfb822c9838b
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Mon Feb 25 16:43:02 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Use long wherever possible in mpa.c
|
|
|
8ae002 |
|
|
|
8ae002 |
Using long throughout like powerpc does is beneficial since it reduces
|
|
|
8ae002 |
the need to switch to 32-bit instructions. It gives a very minor
|
|
|
8ae002 |
performance improvement.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 82a9811d29c00161c7c8ea7f70e2cc30988e192e
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Thu Mar 7 12:23:29 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Use generic mpa.c code for everything except __mul and __sqr
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 6f2e90e78f151bab153c2b38492505ae2012db06
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Tue Mar 26 19:28:50 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Make mantissa type of mp_no configurable
|
|
|
8ae002 |
|
|
|
8ae002 |
The mantissa of mp_no is intended to take only integral values. This
|
|
|
8ae002 |
is a relatively good choice for powerpc due to its 4 fpus, but not for
|
|
|
8ae002 |
other architectures, which suffer due to this choice. This change
|
|
|
8ae002 |
makes the default mantissa a long integer and allows powerpc to
|
|
|
8ae002 |
override it. Additionally, some operations have been optimized for
|
|
|
8ae002 |
integer manipulation, resulting in a significant improvement in
|
|
|
8ae002 |
performance.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 5739f705eed5cf58e7b439e5983542e06d7fc2da
|
|
|
8ae002 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
8ae002 |
Date: Tue Mar 26 20:24:04 2013 +0530
|
|
|
8ae002 |
|
|
|
8ae002 |
Use integral constants
|
|
|
8ae002 |
|
|
|
8ae002 |
The compiler is smart enough to convert those into double for powerpc,
|
|
|
8ae002 |
but if we put them as doubles, it adds overhead by performing those
|
|
|
8ae002 |
operations in floating point mode.
|
|
|
8ae002 |
|
|
|
8ae002 |
commit 89f3b6e18c6e7833438789746fcfc2e7189f7cac
|
|
|
8ae002 |
Author: Joseph Myers <joseph@codesourcery.com>
|
|
|
8ae002 |
Date: Thu May 21 23:05:45 2015 +0000
|
|
|
8ae002 |
|
|
|
8ae002 |
Fix sysdeps/ieee754/dbl-64/mpa.c for -Wuninitialized.
|
|
|
8ae002 |
|
|
|
8ae002 |
If you remove the "override CFLAGS += -Wno-uninitialized" in
|
|
|
8ae002 |
math/Makefile, one of the errors you get is:
|
|
|
8ae002 |
|
|
|
8ae002 |
../sysdeps/ieee754/dbl-64/mpa.c: In function '__mp_dbl.part.0':
|
|
|
8ae002 |
../sysdeps/ieee754/dbl-64/mpa.c:183:5: error: 'c' may be used uninitialized in this function [-Werror=maybe-uninitialized]
|
|
|
8ae002 |
c *= X[0];
|
|
|
8ae002 |
|
|
|
8ae002 |
The problem is that the p < 5 case initializes c if p is 1, 2, 3 or 4
|
|
|
8ae002 |
but not otherwise, and in fact p is positive for all calls to this
|
|
|
8ae002 |
function so the uninitialized case can't actually occur. This patch
|
|
|
8ae002 |
replaces the "if (p == 4)" last case with a comment so the compiler
|
|
|
8ae002 |
can see that all paths do initialize c.
|
|
|
8ae002 |
|
|
|
8ae002 |
Tested for x86_64.
|
|
|
8ae002 |
|
|
|
8ae002 |
* sysdeps/ieee754/dbl-64/mpa.c (norm): Remove if condition on
|
|
|
8ae002 |
(p == 4) case.
|
|
|
8ae002 |
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpa.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpa.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpa.c
|
|
|
8ae002 |
@@ -1,7 +1,7 @@
|
|
|
8ae002 |
/*
|
|
|
8ae002 |
* IBM Accurate Mathematical Library
|
|
|
8ae002 |
* written by International Business Machines Corp.
|
|
|
8ae002 |
- * Copyright (C) 2001, 2011 Free Software Foundation
|
|
|
8ae002 |
+ * Copyright (C) 2001-2017 Free Software Foundation, Inc.
|
|
|
8ae002 |
*
|
|
|
8ae002 |
* This program is free software; you can redistribute it and/or modify
|
|
|
8ae002 |
* it under the terms of the GNU Lesser General Public License as published by
|
|
|
8ae002 |
@@ -22,9 +22,7 @@
|
|
|
8ae002 |
/* FUNCTIONS: */
|
|
|
8ae002 |
/* mcr */
|
|
|
8ae002 |
/* acr */
|
|
|
8ae002 |
-/* cr */
|
|
|
8ae002 |
/* cpy */
|
|
|
8ae002 |
-/* cpymn */
|
|
|
8ae002 |
/* norm */
|
|
|
8ae002 |
/* denorm */
|
|
|
8ae002 |
/* mp_dbl */
|
|
|
8ae002 |
@@ -44,479 +42,868 @@
|
|
|
8ae002 |
|
|
|
8ae002 |
#include "endian.h"
|
|
|
8ae002 |
#include "mpa.h"
|
|
|
8ae002 |
-#include "mpa2.h"
|
|
|
8ae002 |
-#include <sys/param.h> /* For MIN() */
|
|
|
8ae002 |
+#include <sys/param.h>
|
|
|
8ae002 |
+#include <alloca.h>
|
|
|
8ae002 |
|
|
|
8ae002 |
#ifndef SECTION
|
|
|
8ae002 |
# define SECTION
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
+#ifndef NO__CONST
|
|
|
8ae002 |
+/* TODO: With only a partial backport of the constant cleanup from
|
|
|
8ae002 |
+ upstream we limit the definition of these two constants to
|
|
|
8ae002 |
+ this file. */
|
|
|
8ae002 |
+static const mp_no __mpone = { 1, { 1.0, 1.0 } };
|
|
|
8ae002 |
+static const mp_no __mptwo = { 1, { 1.0, 2.0 } };
|
|
|
8ae002 |
+#endif
|
|
|
8ae002 |
+
|
|
|
8ae002 |
#ifndef NO___ACR
|
|
|
8ae002 |
-/* mcr() compares the sizes of the mantissas of two multiple precision */
|
|
|
8ae002 |
-/* numbers. Mantissas are compared regardless of the signs of the */
|
|
|
8ae002 |
-/* numbers, even if x->d[0] or y->d[0] are zero. Exponents are also */
|
|
|
8ae002 |
-/* disregarded. */
|
|
|
8ae002 |
+/* Compare mantissa of two multiple precision numbers regardless of the sign
|
|
|
8ae002 |
+ and exponent of the numbers. */
|
|
|
8ae002 |
static int
|
|
|
8ae002 |
-mcr(const mp_no *x, const mp_no *y, int p) {
|
|
|
8ae002 |
- int i;
|
|
|
8ae002 |
- for (i=1; i<=p; i++) {
|
|
|
8ae002 |
- if (X[i] == Y[i]) continue;
|
|
|
8ae002 |
- else if (X[i] > Y[i]) return 1;
|
|
|
8ae002 |
- else return -1; }
|
|
|
8ae002 |
+mcr (const mp_no *x, const mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i;
|
|
|
8ae002 |
+ long p2 = p;
|
|
|
8ae002 |
+ for (i = 1; i <= p2; i++)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (X[i] == Y[i])
|
|
|
8ae002 |
+ continue;
|
|
|
8ae002 |
+ else if (X[i] > Y[i])
|
|
|
8ae002 |
+ return 1;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ return -1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
return 0;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* acr() compares the absolute values of two multiple precision numbers */
|
|
|
8ae002 |
+/* Compare the absolute values of two multiple precision numbers. */
|
|
|
8ae002 |
int
|
|
|
8ae002 |
-__acr(const mp_no *x, const mp_no *y, int p) {
|
|
|
8ae002 |
- int i;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] == ZERO) {
|
|
|
8ae002 |
- if (Y[0] == ZERO) i= 0;
|
|
|
8ae002 |
- else i=-1;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else if (Y[0] == ZERO) i= 1;
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if (EX > EY) i= 1;
|
|
|
8ae002 |
- else if (EX < EY) i=-1;
|
|
|
8ae002 |
- else i= mcr(x,y,p);
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- return i;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
-
|
|
|
8ae002 |
+__acr (const mp_no *x, const mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i;
|
|
|
8ae002 |
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
-/* cr() compares the values of two multiple precision numbers */
|
|
|
8ae002 |
-static int __cr(const mp_no *x, const mp_no *y, int p) {
|
|
|
8ae002 |
- int i;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] > Y[0]) i= 1;
|
|
|
8ae002 |
- else if (X[0] < Y[0]) i=-1;
|
|
|
8ae002 |
- else if (X[0] < ZERO ) i= __acr(y,x,p);
|
|
|
8ae002 |
- else i= __acr(x,y,p);
|
|
|
8ae002 |
+ if (X[0] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (Y[0] == 0)
|
|
|
8ae002 |
+ i = 0;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ i = -1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (Y[0] == 0)
|
|
|
8ae002 |
+ i = 1;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (EX > EY)
|
|
|
8ae002 |
+ i = 1;
|
|
|
8ae002 |
+ else if (EX < EY)
|
|
|
8ae002 |
+ i = -1;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ i = mcr (x, y, p);
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
return i;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
-
|
|
|
8ae002 |
#ifndef NO___CPY
|
|
|
8ae002 |
-/* Copy a multiple precision number. Set *y=*x. x=y is permissible. */
|
|
|
8ae002 |
-void __cpy(const mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
+/* Copy multiple precision number X into Y. They could be the same
|
|
|
8ae002 |
+ number. */
|
|
|
8ae002 |
+void
|
|
|
8ae002 |
+__cpy (const mp_no *x, mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
EY = EX;
|
|
|
8ae002 |
- for (int i=0; i <= p; i++) Y[i] = X[i];
|
|
|
8ae002 |
+ for (i = 0; i <= p; i++)
|
|
|
8ae002 |
+ Y[i] = X[i];
|
|
|
8ae002 |
}
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
+#ifndef NO___MP_DBL
|
|
|
8ae002 |
+/* Convert a multiple precision number *X into a double precision
|
|
|
8ae002 |
+ number *Y, normalized case (|x| >= 2**(-1022))). X has precision
|
|
|
8ae002 |
+ P, which is positive. */
|
|
|
8ae002 |
+static void
|
|
|
8ae002 |
+norm (const mp_no *x, double *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+# define R RADIXI
|
|
|
8ae002 |
+ long i;
|
|
|
8ae002 |
+ double c;
|
|
|
8ae002 |
+ mantissa_t a, u, v, z[5];
|
|
|
8ae002 |
+ if (p < 5)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (p == 1)
|
|
|
8ae002 |
+ c = X[1];
|
|
|
8ae002 |
+ else if (p == 2)
|
|
|
8ae002 |
+ c = X[1] + R * X[2];
|
|
|
8ae002 |
+ else if (p == 3)
|
|
|
8ae002 |
+ c = X[1] + R * (X[2] + R * X[3]);
|
|
|
8ae002 |
+ else /* p == 4. */
|
|
|
8ae002 |
+ c = (X[1] + R * X[2]) + R * R * (X[3] + R * X[4]);
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ for (a = 1, z[1] = X[1]; z[1] < TWO23; )
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ a *= 2;
|
|
|
8ae002 |
+ z[1] *= 2;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
-/* Copy a multiple precision number x of precision m into a */
|
|
|
8ae002 |
-/* multiple precision number y of precision n. In case n>m, */
|
|
|
8ae002 |
-/* the digits of y beyond the m'th are set to zero. In case */
|
|
|
8ae002 |
-/* n
|
|
|
8ae002 |
-/* x=y is permissible. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-static void __cpymn(const mp_no *x, int m, mp_no *y, int n) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- int i,k;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- EY = EX; k=MIN(m,n);
|
|
|
8ae002 |
- for (i=0; i <= k; i++) Y[i] = X[i];
|
|
|
8ae002 |
- for ( ; i <= n; i++) Y[i] = ZERO;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
+ for (i = 2; i < 5; i++)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ mantissa_store_t d, r;
|
|
|
8ae002 |
+ d = X[i] * (mantissa_store_t) a;
|
|
|
8ae002 |
+ DIV_RADIX (d, r);
|
|
|
8ae002 |
+ z[i] = r;
|
|
|
8ae002 |
+ z[i - 1] += d;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
+ u = ALIGN_DOWN_TO (z[3], TWO19);
|
|
|
8ae002 |
+ v = z[3] - u;
|
|
|
8ae002 |
|
|
|
8ae002 |
-#ifndef NO___MP_DBL
|
|
|
8ae002 |
-/* Convert a multiple precision number *x into a double precision */
|
|
|
8ae002 |
-/* number *y, normalized case (|x| >= 2**(-1022))) */
|
|
|
8ae002 |
-static void norm(const mp_no *x, double *y, int p)
|
|
|
8ae002 |
-{
|
|
|
8ae002 |
- #define R radixi.d
|
|
|
8ae002 |
- int i;
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- int k;
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
- double a,c,u,v,z[5];
|
|
|
8ae002 |
- if (p<5) {
|
|
|
8ae002 |
- if (p==1) c = X[1];
|
|
|
8ae002 |
- else if (p==2) c = X[1] + R* X[2];
|
|
|
8ae002 |
- else if (p==3) c = X[1] + R*(X[2] + R* X[3]);
|
|
|
8ae002 |
- else if (p==4) c =(X[1] + R* X[2]) + R*R*(X[3] + R*X[4]);
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- for (a=ONE, z[1]=X[1]; z[1] < TWO23; )
|
|
|
8ae002 |
- {a *= TWO; z[1] *= TWO; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (i=2; i<5; i++) {
|
|
|
8ae002 |
- z[i] = X[i]*a;
|
|
|
8ae002 |
- u = (z[i] + CUTTER)-CUTTER;
|
|
|
8ae002 |
- if (u > z[i]) u -= RADIX;
|
|
|
8ae002 |
- z[i] -= u;
|
|
|
8ae002 |
- z[i-1] += u*RADIXI;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- u = (z[3] + TWO71) - TWO71;
|
|
|
8ae002 |
- if (u > z[3]) u -= TWO19;
|
|
|
8ae002 |
- v = z[3]-u;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (v == TWO18) {
|
|
|
8ae002 |
- if (z[4] == ZERO) {
|
|
|
8ae002 |
- for (i=5; i <= p; i++) {
|
|
|
8ae002 |
- if (X[i] == ZERO) continue;
|
|
|
8ae002 |
- else {z[3] += ONE; break; }
|
|
|
8ae002 |
+ if (v == TWO18)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (z[4] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ for (i = 5; i <= p; i++)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (X[i] == 0)
|
|
|
8ae002 |
+ continue;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[3] += 1;
|
|
|
8ae002 |
+ break;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ z[3] += 1;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else z[3] += ONE;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
|
|
|
8ae002 |
- c = (z[1] + R *(z[2] + R * z[3]))/a;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
+ c = (z[1] + R * (z[2] + R * z[3])) / a;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
c *= X[0];
|
|
|
8ae002 |
|
|
|
8ae002 |
- for (i=1; i
|
|
|
8ae002 |
- for (i=1; i>EX; i--) c *= RADIXI;
|
|
|
8ae002 |
+ for (i = 1; i < EX; i++)
|
|
|
8ae002 |
+ c *= RADIX;
|
|
|
8ae002 |
+ for (i = 1; i > EX; i--)
|
|
|
8ae002 |
+ c *= RADIXI;
|
|
|
8ae002 |
|
|
|
8ae002 |
*y = c;
|
|
|
8ae002 |
-#undef R
|
|
|
8ae002 |
+# undef R
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* Convert a multiple precision number *x into a double precision */
|
|
|
8ae002 |
-/* number *y, denormalized case (|x| < 2**(-1022))) */
|
|
|
8ae002 |
-static void denorm(const mp_no *x, double *y, int p)
|
|
|
8ae002 |
+/* Convert a multiple precision number *X into a double precision
|
|
|
8ae002 |
+ number *Y, Denormal case (|x| < 2**(-1022))). */
|
|
|
8ae002 |
+static void
|
|
|
8ae002 |
+denorm (const mp_no *x, double *y, int p)
|
|
|
8ae002 |
{
|
|
|
8ae002 |
- int i,k;
|
|
|
8ae002 |
- double c,u,z[5];
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- double a,v;
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
+ long i, k;
|
|
|
8ae002 |
+ long p2 = p;
|
|
|
8ae002 |
+ double c;
|
|
|
8ae002 |
+ mantissa_t u, z[5];
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+# define R RADIXI
|
|
|
8ae002 |
+ if (EX < -44 || (EX == -44 && X[1] < TWO5))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ *y = 0;
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-#define R radixi.d
|
|
|
8ae002 |
- if (EX<-44 || (EX==-44 && X[1]
|
|
|
8ae002 |
- { *y=ZERO; return; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (p==1) {
|
|
|
8ae002 |
- if (EX==-42) {z[1]=X[1]+TWO10; z[2]=ZERO; z[3]=ZERO; k=3;}
|
|
|
8ae002 |
- else if (EX==-43) {z[1]= TWO10; z[2]=X[1]; z[3]=ZERO; k=2;}
|
|
|
8ae002 |
- else {z[1]= TWO10; z[2]=ZERO; z[3]=X[1]; k=1;}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else if (p==2) {
|
|
|
8ae002 |
- if (EX==-42) {z[1]=X[1]+TWO10; z[2]=X[2]; z[3]=ZERO; k=3;}
|
|
|
8ae002 |
- else if (EX==-43) {z[1]= TWO10; z[2]=X[1]; z[3]=X[2]; k=2;}
|
|
|
8ae002 |
- else {z[1]= TWO10; z[2]=ZERO; z[3]=X[1]; k=1;}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if (EX==-42) {z[1]=X[1]+TWO10; z[2]=X[2]; k=3;}
|
|
|
8ae002 |
- else if (EX==-43) {z[1]= TWO10; z[2]=X[1]; k=2;}
|
|
|
8ae002 |
- else {z[1]= TWO10; z[2]=ZERO; k=1;}
|
|
|
8ae002 |
- z[3] = X[k];
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- u = (z[3] + TWO57) - TWO57;
|
|
|
8ae002 |
- if (u > z[3]) u -= TWO5;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (u==z[3]) {
|
|
|
8ae002 |
- for (i=k+1; i <= p; i++) {
|
|
|
8ae002 |
- if (X[i] == ZERO) continue;
|
|
|
8ae002 |
- else {z[3] += ONE; break; }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- c = X[0]*((z[1] + R*(z[2] + R*z[3])) - TWO10);
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- *y = c*TWOM1032;
|
|
|
8ae002 |
-#undef R
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Convert a multiple precision number *x into a double precision number *y. */
|
|
|
8ae002 |
-/* The result is correctly rounded to the nearest/even. *x is left unchanged */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __mp_dbl(const mp_no *x, double *y, int p) {
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- int i,k;
|
|
|
8ae002 |
- double a,c,u,v,z[5];
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
+ if (p2 == 1)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (EX == -42)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = X[1] + TWO10;
|
|
|
8ae002 |
+ z[2] = 0;
|
|
|
8ae002 |
+ z[3] = 0;
|
|
|
8ae002 |
+ k = 3;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (EX == -43)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = TWO10;
|
|
|
8ae002 |
+ z[2] = X[1];
|
|
|
8ae002 |
+ z[3] = 0;
|
|
|
8ae002 |
+ k = 2;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = TWO10;
|
|
|
8ae002 |
+ z[2] = 0;
|
|
|
8ae002 |
+ z[3] = X[1];
|
|
|
8ae002 |
+ k = 1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (p2 == 2)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (EX == -42)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = X[1] + TWO10;
|
|
|
8ae002 |
+ z[2] = X[2];
|
|
|
8ae002 |
+ z[3] = 0;
|
|
|
8ae002 |
+ k = 3;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (EX == -43)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = TWO10;
|
|
|
8ae002 |
+ z[2] = X[1];
|
|
|
8ae002 |
+ z[3] = X[2];
|
|
|
8ae002 |
+ k = 2;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = TWO10;
|
|
|
8ae002 |
+ z[2] = 0;
|
|
|
8ae002 |
+ z[3] = X[1];
|
|
|
8ae002 |
+ k = 1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (EX == -42)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = X[1] + TWO10;
|
|
|
8ae002 |
+ z[2] = X[2];
|
|
|
8ae002 |
+ k = 3;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (EX == -43)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = TWO10;
|
|
|
8ae002 |
+ z[2] = X[1];
|
|
|
8ae002 |
+ k = 2;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[1] = TWO10;
|
|
|
8ae002 |
+ z[2] = 0;
|
|
|
8ae002 |
+ k = 1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ z[3] = X[k];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] == ZERO) {*y = ZERO; return; }
|
|
|
8ae002 |
+ u = ALIGN_DOWN_TO (z[3], TWO5);
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (EX> -42) norm(x,y,p);
|
|
|
8ae002 |
- else if (EX==-42 && X[1]>=TWO10) norm(x,y,p);
|
|
|
8ae002 |
- else denorm(x,y,p);
|
|
|
8ae002 |
+ if (u == z[3])
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ for (i = k + 1; i <= p2; i++)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (X[i] == 0)
|
|
|
8ae002 |
+ continue;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ z[3] += 1;
|
|
|
8ae002 |
+ break;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ c = X[0] * ((z[1] + R * (z[2] + R * z[3])) - TWO10);
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ *y = c * TWOM1032;
|
|
|
8ae002 |
+# undef R
|
|
|
8ae002 |
}
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
+/* Convert multiple precision number *X into double precision number *Y. The
|
|
|
8ae002 |
+ result is correctly rounded to the nearest/even. */
|
|
|
8ae002 |
+void
|
|
|
8ae002 |
+__mp_dbl (const mp_no *x, double *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ if (X[0] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ *y = 0;
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* dbl_mp() converts a double precision number x into a multiple precision */
|
|
|
8ae002 |
-/* number *y. If the precision p is too small the result is truncated. x is */
|
|
|
8ae002 |
-/* left unchanged. */
|
|
|
8ae002 |
+ if (__glibc_likely (EX > -42 || (EX == -42 && X[1] >= TWO10)))
|
|
|
8ae002 |
+ norm (x, y, p);
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ denorm (x, y, p);
|
|
|
8ae002 |
+}
|
|
|
8ae002 |
+#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
+/* Get the multiple precision equivalent of X into *Y. If the precision is too
|
|
|
8ae002 |
+ small, the result is truncated. */
|
|
|
8ae002 |
void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
-__dbl_mp(double x, mp_no *y, int p) {
|
|
|
8ae002 |
+__dbl_mp (double x, mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i, n;
|
|
|
8ae002 |
+ long p2 = p;
|
|
|
8ae002 |
|
|
|
8ae002 |
- int i,n;
|
|
|
8ae002 |
- double u;
|
|
|
8ae002 |
+ /* Sign. */
|
|
|
8ae002 |
+ if (x == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Y[0] = 0;
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (x > 0)
|
|
|
8ae002 |
+ Y[0] = 1;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Y[0] = -1;
|
|
|
8ae002 |
+ x = -x;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- /* Sign */
|
|
|
8ae002 |
- if (x == ZERO) {Y[0] = ZERO; return; }
|
|
|
8ae002 |
- else if (x > ZERO) Y[0] = ONE;
|
|
|
8ae002 |
- else {Y[0] = MONE; x=-x; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Exponent */
|
|
|
8ae002 |
- for (EY=ONE; x >= RADIX; EY += ONE) x *= RADIXI;
|
|
|
8ae002 |
- for ( ; x < ONE; EY -= ONE) x *= RADIX;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Digits */
|
|
|
8ae002 |
- n=MIN(p,4);
|
|
|
8ae002 |
- for (i=1; i<=n; i++) {
|
|
|
8ae002 |
- u = (x + TWO52) - TWO52;
|
|
|
8ae002 |
- if (u>x) u -= ONE;
|
|
|
8ae002 |
- Y[i] = u; x -= u; x *= RADIX; }
|
|
|
8ae002 |
- for ( ; i<=p; i++) Y[i] = ZERO;
|
|
|
8ae002 |
+ /* Exponent. */
|
|
|
8ae002 |
+ for (EY = 1; x >= RADIX; EY += 1)
|
|
|
8ae002 |
+ x *= RADIXI;
|
|
|
8ae002 |
+ for (; x < 1; EY -= 1)
|
|
|
8ae002 |
+ x *= RADIX;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Digits. */
|
|
|
8ae002 |
+ n = MIN (p2, 4);
|
|
|
8ae002 |
+ for (i = 1; i <= n; i++)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ INTEGER_OF (x, Y[i]);
|
|
|
8ae002 |
+ x *= RADIX;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ for (; i <= p2; i++)
|
|
|
8ae002 |
+ Y[i] = 0;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* add_magnitudes() adds the magnitudes of *x & *y assuming that */
|
|
|
8ae002 |
-/* abs(*x) >= abs(*y) > 0. */
|
|
|
8ae002 |
-/* The sign of the sum *z is undefined. x&y may overlap but not x&z or y&z. */
|
|
|
8ae002 |
-/* No guard digit is used. The result equals the exact sum, truncated. */
|
|
|
8ae002 |
-/* *x & *y are left unchanged. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
+/* Add magnitudes of *X and *Y assuming that abs (*X) >= abs (*Y) > 0. The
|
|
|
8ae002 |
+ sign of the sum *Z is not changed. X and Y may overlap but not X and Z or
|
|
|
8ae002 |
+ Y and Z. No guard digit is used. The result equals the exact sum,
|
|
|
8ae002 |
+ truncated. */
|
|
|
8ae002 |
static void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
-add_magnitudes(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- int i,j,k;
|
|
|
8ae002 |
+add_magnitudes (const mp_no *x, const mp_no *y, mp_no *z, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i, j, k;
|
|
|
8ae002 |
+ long p2 = p;
|
|
|
8ae002 |
+ mantissa_t zk;
|
|
|
8ae002 |
|
|
|
8ae002 |
EZ = EX;
|
|
|
8ae002 |
|
|
|
8ae002 |
- i=p; j=p+ EY - EX; k=p+1;
|
|
|
8ae002 |
+ i = p2;
|
|
|
8ae002 |
+ j = p2 + EY - EX;
|
|
|
8ae002 |
+ k = p2 + 1;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ if (__glibc_unlikely (j < 1))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __cpy (x, z, p);
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (j<1)
|
|
|
8ae002 |
- {__cpy(x,z,p); return; }
|
|
|
8ae002 |
- else Z[k] = ZERO;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; j>0; i--,j--) {
|
|
|
8ae002 |
- Z[k] += X[i] + Y[j];
|
|
|
8ae002 |
- if (Z[k] >= RADIX) {
|
|
|
8ae002 |
- Z[k] -= RADIX;
|
|
|
8ae002 |
- Z[--k] = ONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; i>0; i--) {
|
|
|
8ae002 |
- Z[k] += X[i];
|
|
|
8ae002 |
- if (Z[k] >= RADIX) {
|
|
|
8ae002 |
- Z[k] -= RADIX;
|
|
|
8ae002 |
- Z[--k] = ONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (Z[1] == ZERO) {
|
|
|
8ae002 |
- for (i=1; i<=p; i++) Z[i] = Z[i+1]; }
|
|
|
8ae002 |
- else EZ += ONE;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
+ zk = 0;
|
|
|
8ae002 |
|
|
|
8ae002 |
+ for (; j > 0; i--, j--)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ zk += X[i] + Y[j];
|
|
|
8ae002 |
+ if (zk >= RADIX)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk - RADIX;
|
|
|
8ae002 |
+ zk = 1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk;
|
|
|
8ae002 |
+ zk = 0;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* sub_magnitudes() subtracts the magnitudes of *x & *y assuming that */
|
|
|
8ae002 |
-/* abs(*x) > abs(*y) > 0. */
|
|
|
8ae002 |
-/* The sign of the difference *z is undefined. x&y may overlap but not x&z */
|
|
|
8ae002 |
-/* or y&z. One guard digit is used. The error is less than one ulp. */
|
|
|
8ae002 |
-/* *x & *y are left unchanged. */
|
|
|
8ae002 |
+ for (; i > 0; i--)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ zk += X[i];
|
|
|
8ae002 |
+ if (zk >= RADIX)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk - RADIX;
|
|
|
8ae002 |
+ zk = 1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk;
|
|
|
8ae002 |
+ zk = 0;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
+ if (zk == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ for (i = 1; i <= p2; i++)
|
|
|
8ae002 |
+ Z[i] = Z[i + 1];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[1] = zk;
|
|
|
8ae002 |
+ EZ += 1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+}
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Subtract the magnitudes of *X and *Y assuming that abs (*x) > abs (*y) > 0.
|
|
|
8ae002 |
+ The sign of the difference *Z is not changed. X and Y may overlap but not X
|
|
|
8ae002 |
+ and Z or Y and Z. One guard digit is used. The error is less than one
|
|
|
8ae002 |
+ ULP. */
|
|
|
8ae002 |
static void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
-sub_magnitudes(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- int i,j,k;
|
|
|
8ae002 |
+sub_magnitudes (const mp_no *x, const mp_no *y, mp_no *z, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i, j, k;
|
|
|
8ae002 |
+ long p2 = p;
|
|
|
8ae002 |
+ mantissa_t zk;
|
|
|
8ae002 |
|
|
|
8ae002 |
EZ = EX;
|
|
|
8ae002 |
+ i = p2;
|
|
|
8ae002 |
+ j = p2 + EY - EX;
|
|
|
8ae002 |
+ k = p2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Y is too small compared to X, copy X over to the result. */
|
|
|
8ae002 |
+ if (__glibc_unlikely (j < 1))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __cpy (x, z, p);
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* The relevant least significant digit in Y is non-zero, so we factor it in
|
|
|
8ae002 |
+ to enhance accuracy. */
|
|
|
8ae002 |
+ if (j < p2 && Y[j + 1] > 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k + 1] = RADIX - Y[j + 1];
|
|
|
8ae002 |
+ zk = -1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ zk = Z[k + 1] = 0;
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (EX == EY) {
|
|
|
8ae002 |
- i=j=k=p;
|
|
|
8ae002 |
- Z[k] = Z[k+1] = ZERO; }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- j= EX - EY;
|
|
|
8ae002 |
- if (j > p) {__cpy(x,z,p); return; }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- i=p; j=p+1-j; k=p;
|
|
|
8ae002 |
- if (Y[j] > ZERO) {
|
|
|
8ae002 |
- Z[k+1] = RADIX - Y[j--];
|
|
|
8ae002 |
- Z[k] = MONE; }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- Z[k+1] = ZERO;
|
|
|
8ae002 |
- Z[k] = ZERO; j--;}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; j>0; i--,j--) {
|
|
|
8ae002 |
- Z[k] += (X[i] - Y[j]);
|
|
|
8ae002 |
- if (Z[k] < ZERO) {
|
|
|
8ae002 |
- Z[k] += RADIX;
|
|
|
8ae002 |
- Z[--k] = MONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; i>0; i--) {
|
|
|
8ae002 |
- Z[k] += X[i];
|
|
|
8ae002 |
- if (Z[k] < ZERO) {
|
|
|
8ae002 |
- Z[k] += RADIX;
|
|
|
8ae002 |
- Z[--k] = MONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
+ /* Subtract and borrow. */
|
|
|
8ae002 |
+ for (; j > 0; i--, j--)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ zk += (X[i] - Y[j]);
|
|
|
8ae002 |
+ if (zk < 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk + RADIX;
|
|
|
8ae002 |
+ zk = -1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk;
|
|
|
8ae002 |
+ zk = 0;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- for (i=1; Z[i] == ZERO; i++) ;
|
|
|
8ae002 |
+ /* We're done with digits from Y, so it's just digits in X. */
|
|
|
8ae002 |
+ for (; i > 0; i--)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ zk += X[i];
|
|
|
8ae002 |
+ if (zk < 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk + RADIX;
|
|
|
8ae002 |
+ zk = -1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[k--] = zk;
|
|
|
8ae002 |
+ zk = 0;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Normalize. */
|
|
|
8ae002 |
+ for (i = 1; Z[i] == 0; i++)
|
|
|
8ae002 |
+ ;
|
|
|
8ae002 |
EZ = EZ - i + 1;
|
|
|
8ae002 |
- for (k=1; i <= p+1; )
|
|
|
8ae002 |
+ for (k = 1; i <= p2 + 1; )
|
|
|
8ae002 |
Z[k++] = Z[i++];
|
|
|
8ae002 |
- for (; k <= p; )
|
|
|
8ae002 |
- Z[k++] = ZERO;
|
|
|
8ae002 |
+ for (; k <= p2; )
|
|
|
8ae002 |
+ Z[k++] = 0;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Add two multiple precision numbers. Set *z = *x + *y. x&y may overlap */
|
|
|
8ae002 |
-/* but not x&z or y&z. One guard digit is used. The error is less than */
|
|
|
8ae002 |
-/* one ulp. *x & *y are left unchanged. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
+/* Add *X and *Y and store the result in *Z. X and Y may overlap, but not X
|
|
|
8ae002 |
+ and Z or Y and Z. One guard digit is used. The error is less than one
|
|
|
8ae002 |
+ ULP. */
|
|
|
8ae002 |
void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
-__add(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
+__add (const mp_no *x, const mp_no *y, mp_no *z, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
int n;
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] == ZERO) {__cpy(y,z,p); return; }
|
|
|
8ae002 |
- else if (Y[0] == ZERO) {__cpy(x,z,p); return; }
|
|
|
8ae002 |
+ if (X[0] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __cpy (y, z, p);
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (Y[0] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __cpy (x, z, p);
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] == Y[0]) {
|
|
|
8ae002 |
- if (__acr(x,y,p) > 0) {add_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else {add_magnitudes(y,x,z,p); Z[0] = Y[0]; }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if ((n=__acr(x,y,p)) == 1) {sub_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else if (n == -1) {sub_magnitudes(y,x,z,p); Z[0] = Y[0]; }
|
|
|
8ae002 |
- else Z[0] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
+ if (X[0] == Y[0])
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (__acr (x, y, p) > 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ add_magnitudes (x, y, z, p);
|
|
|
8ae002 |
+ Z[0] = X[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ add_magnitudes (y, x, z, p);
|
|
|
8ae002 |
+ Z[0] = Y[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if ((n = __acr (x, y, p)) == 1)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ sub_magnitudes (x, y, z, p);
|
|
|
8ae002 |
+ Z[0] = X[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (n == -1)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ sub_magnitudes (y, x, z, p);
|
|
|
8ae002 |
+ Z[0] = Y[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ Z[0] = 0;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
+/* Subtract *Y from *X and return the result in *Z. X and Y may overlap but
|
|
|
8ae002 |
+ not X and Z or Y and Z. One guard digit is used. The error is less than
|
|
|
8ae002 |
+ one ULP. */
|
|
|
8ae002 |
+void
|
|
|
8ae002 |
+SECTION
|
|
|
8ae002 |
+__sub (const mp_no *x, const mp_no *y, mp_no *z, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ int n;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ if (X[0] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __cpy (y, z, p);
|
|
|
8ae002 |
+ Z[0] = -Z[0];
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (Y[0] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __cpy (x, z, p);
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* Subtract two multiple precision numbers. *z is set to *x - *y. x&y may */
|
|
|
8ae002 |
-/* overlap but not x&z or y&z. One guard digit is used. The error is */
|
|
|
8ae002 |
-/* less than one ulp. *x & *y are left unchanged. */
|
|
|
8ae002 |
+ if (X[0] != Y[0])
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (__acr (x, y, p) > 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ add_magnitudes (x, y, z, p);
|
|
|
8ae002 |
+ Z[0] = X[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ add_magnitudes (y, x, z, p);
|
|
|
8ae002 |
+ Z[0] = -Y[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if ((n = __acr (x, y, p)) == 1)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ sub_magnitudes (x, y, z, p);
|
|
|
8ae002 |
+ Z[0] = X[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else if (n == -1)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ sub_magnitudes (y, x, z, p);
|
|
|
8ae002 |
+ Z[0] = -Y[0];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ Z[0] = 0;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+}
|
|
|
8ae002 |
|
|
|
8ae002 |
+#ifndef NO__MUL
|
|
|
8ae002 |
+/* Multiply *X and *Y and store result in *Z. X and Y may overlap but not X
|
|
|
8ae002 |
+ and Z or Y and Z. For P in [1, 2, 3], the exact result is truncated to P
|
|
|
8ae002 |
+ digits. In case P > 3 the error is bounded by 1.001 ULP. */
|
|
|
8ae002 |
void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
-__sub(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
+__mul (const mp_no *x, const mp_no *y, mp_no *z, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i, j, k, ip, ip2;
|
|
|
8ae002 |
+ long p2 = p;
|
|
|
8ae002 |
+ mantissa_store_t zk;
|
|
|
8ae002 |
+ const mp_no *a;
|
|
|
8ae002 |
+ mantissa_store_t *diag;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Is z=0? */
|
|
|
8ae002 |
+ if (__glibc_unlikely (X[0] * Y[0] == 0))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[0] = 0;
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- int n;
|
|
|
8ae002 |
+ /* We need not iterate through all X's and Y's since it's pointless to
|
|
|
8ae002 |
+ multiply zeroes. Here, both are zero... */
|
|
|
8ae002 |
+ for (ip2 = p2; ip2 > 0; ip2--)
|
|
|
8ae002 |
+ if (X[ip2] != 0 || Y[ip2] != 0)
|
|
|
8ae002 |
+ break;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ a = X[ip2] != 0 ? y : x;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* ... and here, at least one of them is still zero. */
|
|
|
8ae002 |
+ for (ip = ip2; ip > 0; ip--)
|
|
|
8ae002 |
+ if (a->d[ip] != 0)
|
|
|
8ae002 |
+ break;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* The product looks like this for p = 3 (as an example):
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ a1 a2 a3
|
|
|
8ae002 |
+ x b1 b2 b3
|
|
|
8ae002 |
+ -----------------------------
|
|
|
8ae002 |
+ a1*b3 a2*b3 a3*b3
|
|
|
8ae002 |
+ a1*b2 a2*b2 a3*b2
|
|
|
8ae002 |
+ a1*b1 a2*b1 a3*b1
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ So our K needs to ideally be P*2, but we're limiting ourselves to P + 3
|
|
|
8ae002 |
+ for P >= 3. We compute the above digits in two parts; the last P-1
|
|
|
8ae002 |
+ digits and then the first P digits. The last P-1 digits are a sum of
|
|
|
8ae002 |
+ products of the input digits from P to P-k where K is 0 for the least
|
|
|
8ae002 |
+ significant digit and increases as we go towards the left. The product
|
|
|
8ae002 |
+ term is of the form X[k]*X[P-k] as can be seen in the above example.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ The first P digits are also a sum of products with the same product term,
|
|
|
8ae002 |
+ except that the sum is from 1 to k. This is also evident from the above
|
|
|
8ae002 |
+ example.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ Another thing that becomes evident is that only the most significant
|
|
|
8ae002 |
+ ip+ip2 digits of the result are non-zero, where ip and ip2 are the
|
|
|
8ae002 |
+ 'internal precision' of the input numbers, i.e. digits after ip and ip2
|
|
|
8ae002 |
+ are all 0. */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ k = (__glibc_unlikely (p2 < 3)) ? p2 + p2 : p2 + 3;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ while (k > ip + ip2 + 1)
|
|
|
8ae002 |
+ Z[k--] = 0;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ zk = 0;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Precompute sums of diagonal elements so that we can directly use them
|
|
|
8ae002 |
+ later. See the next comment to know we why need them. */
|
|
|
8ae002 |
+ diag = alloca (k * sizeof (mantissa_store_t));
|
|
|
8ae002 |
+ mantissa_store_t d = 0;
|
|
|
8ae002 |
+ for (i = 1; i <= ip; i++)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ d += X[i] * (mantissa_store_t) Y[i];
|
|
|
8ae002 |
+ diag[i] = d;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ while (i < k)
|
|
|
8ae002 |
+ diag[i++] = d;
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] == ZERO) {__cpy(y,z,p); Z[0] = -Z[0]; return; }
|
|
|
8ae002 |
- else if (Y[0] == ZERO) {__cpy(x,z,p); return; }
|
|
|
8ae002 |
+ while (k > p2)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ long lim = k / 2;
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] != Y[0]) {
|
|
|
8ae002 |
- if (__acr(x,y,p) > 0) {add_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else {add_magnitudes(y,x,z,p); Z[0] = -Y[0]; }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if ((n=__acr(x,y,p)) == 1) {sub_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else if (n == -1) {sub_magnitudes(y,x,z,p); Z[0] = -Y[0]; }
|
|
|
8ae002 |
- else Z[0] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
+ if (k % 2 == 0)
|
|
|
8ae002 |
+ /* We want to add this only once, but since we subtract it in the sum
|
|
|
8ae002 |
+ of products above, we add twice. */
|
|
|
8ae002 |
+ zk += 2 * X[lim] * (mantissa_store_t) Y[lim];
|
|
|
8ae002 |
|
|
|
8ae002 |
+ for (i = k - p2, j = p2; i < j; i++, j--)
|
|
|
8ae002 |
+ zk += (X[i] + X[j]) * (mantissa_store_t) (Y[i] + Y[j]);
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* Multiply two multiple precision numbers. *z is set to *x * *y. x&y */
|
|
|
8ae002 |
-/* may overlap but not x&z or y&z. In case p=1,2,3 the exact result is */
|
|
|
8ae002 |
-/* truncated to p digits. In case p>3 the error is bounded by 1.001 ulp. */
|
|
|
8ae002 |
-/* *x & *y are left unchanged. */
|
|
|
8ae002 |
+ zk -= diag[k - 1];
|
|
|
8ae002 |
|
|
|
8ae002 |
-void
|
|
|
8ae002 |
-SECTION
|
|
|
8ae002 |
-__mul(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
+ DIV_RADIX (zk, Z[k]);
|
|
|
8ae002 |
+ k--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- int i, i1, i2, j, k, k2;
|
|
|
8ae002 |
- double u;
|
|
|
8ae002 |
+ /* The real deal. Mantissa digit Z[k] is the sum of all X[i] * Y[j] where i
|
|
|
8ae002 |
+ goes from 1 -> k - 1 and j goes the same range in reverse. To reduce the
|
|
|
8ae002 |
+ number of multiplications, we halve the range and if k is an even number,
|
|
|
8ae002 |
+ add the diagonal element X[k/2]Y[k/2]. Through the half range, we compute
|
|
|
8ae002 |
+ X[i] * Y[j] as (X[i] + X[j]) * (Y[i] + Y[j]) - X[i] * Y[i] - X[j] * Y[j].
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ This reduction tells us that we're summing two things, the first term
|
|
|
8ae002 |
+ through the half range and the negative of the sum of the product of all
|
|
|
8ae002 |
+ terms of X and Y in the full range. i.e.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ SUM(X[i] * Y[i]) for k terms. This is precalculated above for each k in
|
|
|
8ae002 |
+ a single loop so that it completes in O(n) time and can hence be directly
|
|
|
8ae002 |
+ used in the loop below. */
|
|
|
8ae002 |
+ while (k > 1)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ long lim = k / 2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ if (k % 2 == 0)
|
|
|
8ae002 |
+ /* We want to add this only once, but since we subtract it in the sum
|
|
|
8ae002 |
+ of products above, we add twice. */
|
|
|
8ae002 |
+ zk += 2 * X[lim] * (mantissa_store_t) Y[lim];
|
|
|
8ae002 |
|
|
|
8ae002 |
- /* Is z=0? */
|
|
|
8ae002 |
- if (X[0]*Y[0]==ZERO)
|
|
|
8ae002 |
- { Z[0]=ZERO; return; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Multiply, add and carry */
|
|
|
8ae002 |
- k2 = (p<3) ? p+p : p+3;
|
|
|
8ae002 |
- Z[k2]=ZERO;
|
|
|
8ae002 |
- for (k=k2; k>1; ) {
|
|
|
8ae002 |
- if (k > p) {i1=k-p; i2=p+1; }
|
|
|
8ae002 |
- else {i1=1; i2=k; }
|
|
|
8ae002 |
- for (i=i1,j=i2-1; i
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- u = (Z[k] + CUTTER)-CUTTER;
|
|
|
8ae002 |
- if (u > Z[k]) u -= RADIX;
|
|
|
8ae002 |
- Z[k] -= u;
|
|
|
8ae002 |
- Z[--k] = u*RADIXI;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Is there a carry beyond the most significant digit? */
|
|
|
8ae002 |
- if (Z[1] == ZERO) {
|
|
|
8ae002 |
- for (i=1; i<=p; i++) Z[i]=Z[i+1];
|
|
|
8ae002 |
- EZ = EX + EY - 1; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- EZ = EX + EY;
|
|
|
8ae002 |
+ for (i = 1, j = k - 1; i < j; i++, j--)
|
|
|
8ae002 |
+ zk += (X[i] + X[j]) * (mantissa_store_t) (Y[i] + Y[j]);
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ zk -= diag[k - 1];
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ DIV_RADIX (zk, Z[k]);
|
|
|
8ae002 |
+ k--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ Z[k] = zk;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Get the exponent sum into an intermediate variable. This is a subtle
|
|
|
8ae002 |
+ optimization, where given enough registers, all operations on the exponent
|
|
|
8ae002 |
+ happen in registers and the result is written out only once into EZ. */
|
|
|
8ae002 |
+ int e = EX + EY;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Is there a carry beyond the most significant digit? */
|
|
|
8ae002 |
+ if (__glibc_unlikely (Z[1] == 0))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ for (i = 1; i <= p2; i++)
|
|
|
8ae002 |
+ Z[i] = Z[i + 1];
|
|
|
8ae002 |
+ e--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
+ EZ = e;
|
|
|
8ae002 |
Z[0] = X[0] * Y[0];
|
|
|
8ae002 |
}
|
|
|
8ae002 |
+#endif
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#ifndef NO__SQR
|
|
|
8ae002 |
+/* Square *X and store result in *Y. X and Y may not overlap. For P in
|
|
|
8ae002 |
+ [1, 2, 3], the exact result is truncated to P digits. In case P > 3 the
|
|
|
8ae002 |
+ error is bounded by 1.001 ULP. This is a faster special case of
|
|
|
8ae002 |
+ multiplication. */
|
|
|
8ae002 |
+void
|
|
|
8ae002 |
+SECTION
|
|
|
8ae002 |
+__sqr (const mp_no *x, mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i, j, k, ip;
|
|
|
8ae002 |
+ mantissa_store_t yk;
|
|
|
8ae002 |
|
|
|
8ae002 |
+ /* Is z=0? */
|
|
|
8ae002 |
+ if (__glibc_unlikely (X[0] == 0))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Y[0] = 0;
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* Invert a multiple precision number. Set *y = 1 / *x. */
|
|
|
8ae002 |
-/* Relative error bound = 1.001*r**(1-p) for p=2, 1.063*r**(1-p) for p=3, */
|
|
|
8ae002 |
-/* 2.001*r**(1-p) for p>3. */
|
|
|
8ae002 |
-/* *x=0 is not permissible. *x is left unchanged. */
|
|
|
8ae002 |
+ /* We need not iterate through all X's since it's pointless to
|
|
|
8ae002 |
+ multiply zeroes. */
|
|
|
8ae002 |
+ for (ip = p; ip > 0; ip--)
|
|
|
8ae002 |
+ if (X[ip] != 0)
|
|
|
8ae002 |
+ break;
|
|
|
8ae002 |
|
|
|
8ae002 |
-static
|
|
|
8ae002 |
-SECTION
|
|
|
8ae002 |
-void __inv(const mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
- int i;
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- int l;
|
|
|
8ae002 |
+ k = (__glibc_unlikely (p < 3)) ? p + p : p + 3;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ while (k > 2 * ip + 1)
|
|
|
8ae002 |
+ Y[k--] = 0;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ yk = 0;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ while (k > p)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ mantissa_store_t yk2 = 0;
|
|
|
8ae002 |
+ long lim = k / 2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ if (k % 2 == 0)
|
|
|
8ae002 |
+ yk += X[lim] * (mantissa_store_t) X[lim];
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* In __mul, this loop (and the one within the next while loop) run
|
|
|
8ae002 |
+ between a range to calculate the mantissa as follows:
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ Z[k] = X[k] * Y[n] + X[k+1] * Y[n-1] ... + X[n-1] * Y[k+1]
|
|
|
8ae002 |
+ + X[n] * Y[k]
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ For X == Y, we can get away with summing halfway and doubling the
|
|
|
8ae002 |
+ result. For cases where the range size is even, the mid-point needs
|
|
|
8ae002 |
+ to be added separately (above). */
|
|
|
8ae002 |
+ for (i = k - p, j = p; i < j; i++, j--)
|
|
|
8ae002 |
+ yk2 += X[i] * (mantissa_store_t) X[j];
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ yk += 2 * yk2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ DIV_RADIX (yk, Y[k]);
|
|
|
8ae002 |
+ k--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ while (k > 1)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ mantissa_store_t yk2 = 0;
|
|
|
8ae002 |
+ long lim = k / 2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ if (k % 2 == 0)
|
|
|
8ae002 |
+ yk += X[lim] * (mantissa_store_t) X[lim];
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Likewise for this loop. */
|
|
|
8ae002 |
+ for (i = 1, j = k - 1; i < j; i++, j--)
|
|
|
8ae002 |
+ yk2 += X[i] * (mantissa_store_t) X[j];
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ yk += 2 * yk2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ DIV_RADIX (yk, Y[k]);
|
|
|
8ae002 |
+ k--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ Y[k] = yk;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Squares are always positive. */
|
|
|
8ae002 |
+ Y[0] = 1;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Get the exponent sum into an intermediate variable. This is a subtle
|
|
|
8ae002 |
+ optimization, where given enough registers, all operations on the exponent
|
|
|
8ae002 |
+ happen in registers and the result is written out only once into EZ. */
|
|
|
8ae002 |
+ int e = EX * 2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Is there a carry beyond the most significant digit? */
|
|
|
8ae002 |
+ if (__glibc_unlikely (Y[1] == 0))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ for (i = 1; i <= p; i++)
|
|
|
8ae002 |
+ Y[i] = Y[i + 1];
|
|
|
8ae002 |
+ e--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ EY = e;
|
|
|
8ae002 |
+}
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Invert *X and store in *Y. Relative error bound:
|
|
|
8ae002 |
+ - For P = 2: 1.001 * R ^ (1 - P)
|
|
|
8ae002 |
+ - For P = 3: 1.063 * R ^ (1 - P)
|
|
|
8ae002 |
+ - For P > 3: 2.001 * R ^ (1 - P)
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ *X = 0 is not permissible. */
|
|
|
8ae002 |
+static void
|
|
|
8ae002 |
+SECTION
|
|
|
8ae002 |
+__inv (const mp_no *x, mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i;
|
|
|
8ae002 |
double t;
|
|
|
8ae002 |
- mp_no z,w;
|
|
|
8ae002 |
- static const int np1[] = {0,0,0,0,1,2,2,2,2,3,3,3,3,3,3,3,3,3,
|
|
|
8ae002 |
- 4,4,4,4,4,4,4,4,4,4,4,4,4,4,4};
|
|
|
8ae002 |
- const mp_no mptwo = {1,{1.0,2.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
- 0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
- 0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
- 0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}};
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- __cpy(x,&z,p); z.e=0; __mp_dbl(&z,&t,p);
|
|
|
8ae002 |
- t=ONE/t; __dbl_mp(t,y,p); EY -= EX;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (i=0; i
|
|
|
8ae002 |
- __cpy(y,&w,p);
|
|
|
8ae002 |
- __mul(x,&w,y,p);
|
|
|
8ae002 |
- __sub(&mptwo,y,&z,p);
|
|
|
8ae002 |
- __mul(&w,&z,y,p);
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
+ mp_no z, w;
|
|
|
8ae002 |
+ static const int np1[] =
|
|
|
8ae002 |
+ { 0, 0, 0, 0, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3,
|
|
|
8ae002 |
+ 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4
|
|
|
8ae002 |
+ };
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ __cpy (x, &z, p);
|
|
|
8ae002 |
+ z.e = 0;
|
|
|
8ae002 |
+ __mp_dbl (&z, &t, p);
|
|
|
8ae002 |
+ t = 1 / t;
|
|
|
8ae002 |
+ __dbl_mp (t, y, p);
|
|
|
8ae002 |
+ EY -= EX;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ for (i = 0; i < np1[p]; i++)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __cpy (y, &w, p);
|
|
|
8ae002 |
+ __mul (x, &w, y, p);
|
|
|
8ae002 |
+ __sub (&__mptwo, y, &z, p);
|
|
|
8ae002 |
+ __mul (&w, &z, y, p);
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
+/* Divide *X by *Y and store result in *Z. X and Y may overlap but not X and Z
|
|
|
8ae002 |
+ or Y and Z. Relative error bound:
|
|
|
8ae002 |
+ - For P = 2: 2.001 * R ^ (1 - P)
|
|
|
8ae002 |
+ - For P = 3: 2.063 * R ^ (1 - P)
|
|
|
8ae002 |
+ - For P > 3: 3.001 * R ^ (1 - P)
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* Divide one multiple precision number by another.Set *z = *x / *y. *x & *y */
|
|
|
8ae002 |
-/* are left unchanged. x&y may overlap but not x&z or y&z. */
|
|
|
8ae002 |
-/* Relative error bound = 2.001*r**(1-p) for p=2, 2.063*r**(1-p) for p=3 */
|
|
|
8ae002 |
-/* and 3.001*r**(1-p) for p>3. *y=0 is not permissible. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
+ *X = 0 is not permissible. */
|
|
|
8ae002 |
void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
-__dvd(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
+__dvd (const mp_no *x, const mp_no *y, mp_no *z, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
mp_no w;
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] == ZERO) Z[0] = ZERO;
|
|
|
8ae002 |
- else {__inv(y,&w,p); __mul(x,&w,z,p);}
|
|
|
8ae002 |
+ if (X[0] == 0)
|
|
|
8ae002 |
+ Z[0] = 0;
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ __inv (y, &w, p);
|
|
|
8ae002 |
+ __mul (x, &w, z, p);
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
}
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpa.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpa.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpa.h
|
|
|
8ae002 |
@@ -1,7 +1,7 @@
|
|
|
8ae002 |
/*
|
|
|
8ae002 |
* IBM Accurate Mathematical Library
|
|
|
8ae002 |
* Written by International Business Machines Corp.
|
|
|
8ae002 |
- * Copyright (C) 2001, 2011 Free Software Foundation, Inc.
|
|
|
8ae002 |
+ * Copyright (C) 2001-2017 Free Software Foundation, Inc.
|
|
|
8ae002 |
*
|
|
|
8ae002 |
* This program is free software; you can redistribute it and/or modify
|
|
|
8ae002 |
* it under the terms of the GNU Lesser General Public License as published by
|
|
|
8ae002 |
@@ -23,36 +23,58 @@
|
|
|
8ae002 |
/* FUNCTIONS: */
|
|
|
8ae002 |
/* mcr */
|
|
|
8ae002 |
/* acr */
|
|
|
8ae002 |
-/* cr */
|
|
|
8ae002 |
/* cpy */
|
|
|
8ae002 |
-/* cpymn */
|
|
|
8ae002 |
/* mp_dbl */
|
|
|
8ae002 |
/* dbl_mp */
|
|
|
8ae002 |
/* add */
|
|
|
8ae002 |
/* sub */
|
|
|
8ae002 |
/* mul */
|
|
|
8ae002 |
-/* inv */
|
|
|
8ae002 |
/* dvd */
|
|
|
8ae002 |
/* */
|
|
|
8ae002 |
/* Arithmetic functions for multiple precision numbers. */
|
|
|
8ae002 |
/* Common types and definition */
|
|
|
8ae002 |
/************************************************************************/
|
|
|
8ae002 |
|
|
|
8ae002 |
+#include <mpa-arch.h>
|
|
|
8ae002 |
|
|
|
8ae002 |
-typedef struct {/* This structure holds the details of a multi-precision */
|
|
|
8ae002 |
- int e; /* floating point number, x: d[0] holds its sign (-1,0 or 1) */
|
|
|
8ae002 |
- double d[40]; /* e holds its exponent (...,-2,-1,0,1,2,...) and */
|
|
|
8ae002 |
-} mp_no; /* d[1]...d[p] hold its mantissa digits. The value of x is, */
|
|
|
8ae002 |
- /* x = d[1]*r**(e-1) + d[2]*r**(e-2) + ... + d[p]*r**(e-p). */
|
|
|
8ae002 |
- /* Here r = 2**24, 0 <= d[i] < r and 1 <= p <= 32. */
|
|
|
8ae002 |
- /* p is a global variable. A multi-precision number is */
|
|
|
8ae002 |
- /* always normalized. Namely, d[1] > 0. An exception is */
|
|
|
8ae002 |
- /* a zero which is characterized by d[0] = 0. The terms */
|
|
|
8ae002 |
- /* d[p+1], d[p+2], ... of a none zero number have no */
|
|
|
8ae002 |
- /* significance and so are the terms e, d[1],d[2],... */
|
|
|
8ae002 |
- /* of a zero. */
|
|
|
8ae002 |
+/* The mp_no structure holds the details of a multi-precision floating point
|
|
|
8ae002 |
+ number.
|
|
|
8ae002 |
|
|
|
8ae002 |
-typedef union { int i[2]; double d; } number;
|
|
|
8ae002 |
+ - The radix of the number (R) is 2 ^ 24.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ - E: The exponent of the number.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ - D[0]: The sign (-1, 1) or 0 if the value is 0. In the latter case, the
|
|
|
8ae002 |
+ values of the remaining members of the structure are ignored.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ - D[1] - D[p]: The mantissa of the number where:
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ 0 <= D[i] < R and
|
|
|
8ae002 |
+ P is the precision of the number and 1 <= p <= 32
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ D[p+1] ... D[39] have no significance.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ - The value of the number is:
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ D[1] * R ^ (E - 1) + D[2] * R ^ (E - 2) ... D[p] * R ^ (E - p)
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ */
|
|
|
8ae002 |
+typedef struct
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ int e;
|
|
|
8ae002 |
+ mantissa_t d[40];
|
|
|
8ae002 |
+} mp_no;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+typedef union
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ int i[2];
|
|
|
8ae002 |
+ double d;
|
|
|
8ae002 |
+} number;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* TODO: With only a partial backport of the constant cleanup we don't
|
|
|
8ae002 |
+ define __mpone or __mptwo here for other code to use. */
|
|
|
8ae002 |
+/* extern const mp_no __mpone;
|
|
|
8ae002 |
+extern const mp_no __mptwo; */
|
|
|
8ae002 |
|
|
|
8ae002 |
#define X x->d
|
|
|
8ae002 |
#define Y y->d
|
|
|
8ae002 |
@@ -63,21 +85,73 @@ typedef union { int i[2]; double d; } nu
|
|
|
8ae002 |
|
|
|
8ae002 |
#define ABS(x) ((x) < 0 ? -(x) : (x))
|
|
|
8ae002 |
|
|
|
8ae002 |
-int __acr(const mp_no *, const mp_no *, int);
|
|
|
8ae002 |
-// int __cr(const mp_no *, const mp_no *, int);
|
|
|
8ae002 |
-void __cpy(const mp_no *, mp_no *, int);
|
|
|
8ae002 |
-// void __cpymn(const mp_no *, int, mp_no *, int);
|
|
|
8ae002 |
-void __mp_dbl(const mp_no *, double *, int);
|
|
|
8ae002 |
-void __dbl_mp(double, mp_no *, int);
|
|
|
8ae002 |
-void __add(const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
-void __sub(const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
-void __mul(const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
-// void __inv(const mp_no *, mp_no *, int);
|
|
|
8ae002 |
-void __dvd(const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
+#ifndef RADIXI
|
|
|
8ae002 |
+# define RADIXI 0x1.0p-24 /* 2^-24 */
|
|
|
8ae002 |
+#endif
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#ifndef TWO52
|
|
|
8ae002 |
+# define TWO52 0x1.0p52 /* 2^52 */
|
|
|
8ae002 |
+#endif
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#define TWO5 TWOPOW (5) /* 2^5 */
|
|
|
8ae002 |
+#define TWO8 TWOPOW (8) /* 2^52 */
|
|
|
8ae002 |
+#define TWO10 TWOPOW (10) /* 2^10 */
|
|
|
8ae002 |
+#define TWO18 TWOPOW (18) /* 2^18 */
|
|
|
8ae002 |
+#define TWO19 TWOPOW (19) /* 2^19 */
|
|
|
8ae002 |
+#define TWO23 TWOPOW (23) /* 2^23 */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#define TWO57 0x1.0p57 /* 2^57 */
|
|
|
8ae002 |
+#define TWO71 0x1.0p71 /* 2^71 */
|
|
|
8ae002 |
+#define TWOM1032 0x1.0p-1032 /* 2^-1032 */
|
|
|
8ae002 |
+#define TWOM1022 0x1.0p-1022 /* 2^-1022 */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#define HALF 0x1.0p-1 /* 1/2 */
|
|
|
8ae002 |
+#define MHALF -0x1.0p-1 /* -1/2 */
|
|
|
8ae002 |
+#define HALFRAD 0x1.0p23 /* 2^23 */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+int __acr (const mp_no *, const mp_no *, int);
|
|
|
8ae002 |
+void __cpy (const mp_no *, mp_no *, int);
|
|
|
8ae002 |
+void __mp_dbl (const mp_no *, double *, int);
|
|
|
8ae002 |
+void __dbl_mp (double, mp_no *, int);
|
|
|
8ae002 |
+void __add (const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
+void __sub (const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
+void __mul (const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
+void __sqr (const mp_no *, mp_no *, int);
|
|
|
8ae002 |
+void __dvd (const mp_no *, const mp_no *, mp_no *, int);
|
|
|
8ae002 |
|
|
|
8ae002 |
extern void __mpatan (mp_no *, mp_no *, int);
|
|
|
8ae002 |
extern void __mpatan2 (mp_no *, mp_no *, mp_no *, int);
|
|
|
8ae002 |
extern void __mpsqrt (mp_no *, mp_no *, int);
|
|
|
8ae002 |
-extern void __mpexp (mp_no *, mp_no *__y, int);
|
|
|
8ae002 |
+extern void __mpexp (mp_no *, mp_no *, int);
|
|
|
8ae002 |
extern void __c32 (mp_no *, mp_no *, mp_no *, int);
|
|
|
8ae002 |
extern int __mpranred (double, mp_no *, int);
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Given a power POW, build a multiprecision number 2^POW. */
|
|
|
8ae002 |
+static inline void
|
|
|
8ae002 |
+__pow_mp (int pow, mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ int i, rem;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* The exponent is E such that E is a factor of 2^24. The remainder (of the
|
|
|
8ae002 |
+ form 2^x) goes entirely into the first digit of the mantissa as it is
|
|
|
8ae002 |
+ always less than 2^24. */
|
|
|
8ae002 |
+ EY = pow / 24;
|
|
|
8ae002 |
+ rem = pow - EY * 24;
|
|
|
8ae002 |
+ EY++;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* If the remainder is negative, it means that POW was negative since
|
|
|
8ae002 |
+ |EY * 24| <= |pow|. Adjust so that REM is positive and still less than
|
|
|
8ae002 |
+ 24 because of which, the mantissa digit is less than 2^24. */
|
|
|
8ae002 |
+ if (rem < 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ EY--;
|
|
|
8ae002 |
+ rem += 24;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ /* The sign of any 2^x is always positive. */
|
|
|
8ae002 |
+ Y[0] = 1;
|
|
|
8ae002 |
+ Y[1] = 1 << rem;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Everything else is 0. */
|
|
|
8ae002 |
+ for (i = 2; i <= p; i++)
|
|
|
8ae002 |
+ Y[i] = 0;
|
|
|
8ae002 |
+}
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpexp.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpexp.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpexp.c
|
|
|
8ae002 |
@@ -69,13 +69,13 @@ __mpexp(mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
for (i=0; i
|
|
|
8ae002 |
for ( ; i>EX; i--) a *= RADIX;
|
|
|
8ae002 |
b = X[1]*RADIXI; m2 = 24*EX;
|
|
|
8ae002 |
- for (; b
|
|
|
8ae002 |
+ for (; b
|
|
|
8ae002 |
if (b == HALF) {
|
|
|
8ae002 |
- for (i=2; i<=p; i++) { if (X[i]!=ZERO) break; }
|
|
|
8ae002 |
- if (i==p+1) { m2--; a *= TWO; }
|
|
|
8ae002 |
+ for (i=2; i<=p; i++) { if (X[i]!=0) break; }
|
|
|
8ae002 |
+ if (i==p+1) { m2--; a *= 2; }
|
|
|
8ae002 |
}
|
|
|
8ae002 |
if ((m=m1+m2) <= 0) {
|
|
|
8ae002 |
- m=0; a=ONE;
|
|
|
8ae002 |
+ m=0; a=1;
|
|
|
8ae002 |
for (i=n-1; i>0; i--,n--) { if (m1np[i][p]+m2>0) break; }
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -84,8 +84,8 @@ __mpexp(mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
__mul(x,&mpt1,&mps,p);
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Evaluate the polynomial. Put result in mpt2 */
|
|
|
8ae002 |
- mpone.e=1; mpone.d[0]=ONE; mpone.d[1]=ONE;
|
|
|
8ae002 |
- mpk.e = 1; mpk.d[0] = ONE; mpk.d[1]=__mpexp_nn[n].d;
|
|
|
8ae002 |
+ mpone.e=1; mpone.d[0]=1; mpone.d[1]=1;
|
|
|
8ae002 |
+ mpk.e = 1; mpk.d[0] = 1; mpk.d[1]=__mpexp_nn[n].d;
|
|
|
8ae002 |
__dvd(&mps,&mpk,&mpt1,p);
|
|
|
8ae002 |
__add(&mpone,&mpt1,&mpak,p);
|
|
|
8ae002 |
for (k=n-1; k>1; k--) {
|
|
|
8ae002 |
@@ -99,9 +99,9 @@ __mpexp(mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Raise polynomial value to the power of 2**m. Put result in y */
|
|
|
8ae002 |
for (k=0,j=0; k
|
|
|
8ae002 |
- __mul(&mpt2,&mpt2,&mpt1,p); k++;
|
|
|
8ae002 |
+ __sqr (&mpt2, &mpt1, p); k++;
|
|
|
8ae002 |
if (k==m) { j=1; break; }
|
|
|
8ae002 |
- __mul(&mpt1,&mpt1,&mpt2,p); k++;
|
|
|
8ae002 |
+ __sqr (&mpt1, &mpt2, p); k++;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
if (j) __cpy(&mpt1,y,p);
|
|
|
8ae002 |
else __cpy(&mpt2,y,p);
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/x86_64/fpu/multiarch/mpa-avx.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/x86_64/fpu/multiarch/mpa-avx.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/x86_64/fpu/multiarch/mpa-avx.c
|
|
|
8ae002 |
@@ -1,5 +1,6 @@
|
|
|
8ae002 |
#define __add __add_avx
|
|
|
8ae002 |
#define __mul __mul_avx
|
|
|
8ae002 |
+#define __sqr __sqr_avx
|
|
|
8ae002 |
#define __sub __sub_avx
|
|
|
8ae002 |
#define __dbl_mp __dbl_mp_avx
|
|
|
8ae002 |
#define __dvd __dvd_avx
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/x86_64/fpu/multiarch/mpa-fma4.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/x86_64/fpu/multiarch/mpa-fma4.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/x86_64/fpu/multiarch/mpa-fma4.c
|
|
|
8ae002 |
@@ -1,5 +1,6 @@
|
|
|
8ae002 |
#define __add __add_fma4
|
|
|
8ae002 |
#define __mul __mul_fma4
|
|
|
8ae002 |
+#define __sqr __sqr_fma4
|
|
|
8ae002 |
#define __sub __sub_fma4
|
|
|
8ae002 |
#define __dbl_mp __dbl_mp_fma4
|
|
|
8ae002 |
#define __dvd __dvd_fma4
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/powerpc/power4/fpu/mpa.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/powerpc/power4/fpu/mpa.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/power4/fpu/mpa.c
|
|
|
8ae002 |
@@ -2,7 +2,7 @@
|
|
|
8ae002 |
/*
|
|
|
8ae002 |
* IBM Accurate Mathematical Library
|
|
|
8ae002 |
* written by International Business Machines Corp.
|
|
|
8ae002 |
- * Copyright (C) 2001, 2006 Free Software Foundation
|
|
|
8ae002 |
+ * Copyright (C) 2001-2017 Free Software Foundation, Inc.
|
|
|
8ae002 |
*
|
|
|
8ae002 |
* This program is free software; you can redistribute it and/or modify
|
|
|
8ae002 |
* it under the terms of the GNU Lesser General Public License as published by
|
|
|
8ae002 |
@@ -17,532 +17,198 @@
|
|
|
8ae002 |
* You should have received a copy of the GNU Lesser General Public License
|
|
|
8ae002 |
* along with this program; if not, see <http://www.gnu.org/licenses/>.
|
|
|
8ae002 |
*/
|
|
|
8ae002 |
-/************************************************************************/
|
|
|
8ae002 |
-/* MODULE_NAME: mpa.c */
|
|
|
8ae002 |
-/* */
|
|
|
8ae002 |
-/* FUNCTIONS: */
|
|
|
8ae002 |
-/* mcr */
|
|
|
8ae002 |
-/* acr */
|
|
|
8ae002 |
-/* cr */
|
|
|
8ae002 |
-/* cpy */
|
|
|
8ae002 |
-/* cpymn */
|
|
|
8ae002 |
-/* norm */
|
|
|
8ae002 |
-/* denorm */
|
|
|
8ae002 |
-/* mp_dbl */
|
|
|
8ae002 |
-/* dbl_mp */
|
|
|
8ae002 |
-/* add_magnitudes */
|
|
|
8ae002 |
-/* sub_magnitudes */
|
|
|
8ae002 |
-/* add */
|
|
|
8ae002 |
-/* sub */
|
|
|
8ae002 |
-/* mul */
|
|
|
8ae002 |
-/* inv */
|
|
|
8ae002 |
-/* dvd */
|
|
|
8ae002 |
-/* */
|
|
|
8ae002 |
-/* Arithmetic functions for multiple precision numbers. */
|
|
|
8ae002 |
-/* Relative errors are bounded */
|
|
|
8ae002 |
-/************************************************************************/
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#include "endian.h"
|
|
|
8ae002 |
-#include "mpa.h"
|
|
|
8ae002 |
-#include "mpa2.h"
|
|
|
8ae002 |
-#include <sys/param.h> /* For MIN() */
|
|
|
8ae002 |
-/* mcr() compares the sizes of the mantissas of two multiple precision */
|
|
|
8ae002 |
-/* numbers. Mantissas are compared regardless of the signs of the */
|
|
|
8ae002 |
-/* numbers, even if x->d[0] or y->d[0] are zero. Exponents are also */
|
|
|
8ae002 |
-/* disregarded. */
|
|
|
8ae002 |
-static int mcr(const mp_no *x, const mp_no *y, int p) {
|
|
|
8ae002 |
- long i;
|
|
|
8ae002 |
- long p2 = p;
|
|
|
8ae002 |
- for (i=1; i<=p2; i++) {
|
|
|
8ae002 |
- if (X[i] == Y[i]) continue;
|
|
|
8ae002 |
- else if (X[i] > Y[i]) return 1;
|
|
|
8ae002 |
- else return -1; }
|
|
|
8ae002 |
- return 0;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* acr() compares the absolute values of two multiple precision numbers */
|
|
|
8ae002 |
-int __acr(const mp_no *x, const mp_no *y, int p) {
|
|
|
8ae002 |
- long i;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] == ZERO) {
|
|
|
8ae002 |
- if (Y[0] == ZERO) i= 0;
|
|
|
8ae002 |
- else i=-1;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else if (Y[0] == ZERO) i= 1;
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if (EX > EY) i= 1;
|
|
|
8ae002 |
- else if (EX < EY) i=-1;
|
|
|
8ae002 |
- else i= mcr(x,y,p);
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- return i;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* cr90 compares the values of two multiple precision numbers */
|
|
|
8ae002 |
-int __cr(const mp_no *x, const mp_no *y, int p) {
|
|
|
8ae002 |
- int i;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] > Y[0]) i= 1;
|
|
|
8ae002 |
- else if (X[0] < Y[0]) i=-1;
|
|
|
8ae002 |
- else if (X[0] < ZERO ) i= __acr(y,x,p);
|
|
|
8ae002 |
- else i= __acr(x,y,p);
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- return i;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Copy a multiple precision number. Set *y=*x. x=y is permissible. */
|
|
|
8ae002 |
-void __cpy(const mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
- long i;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- EY = EX;
|
|
|
8ae002 |
- for (i=0; i <= p; i++) Y[i] = X[i];
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Copy a multiple precision number x of precision m into a */
|
|
|
8ae002 |
-/* multiple precision number y of precision n. In case n>m, */
|
|
|
8ae002 |
-/* the digits of y beyond the m'th are set to zero. In case */
|
|
|
8ae002 |
-/* n
|
|
|
8ae002 |
-/* x=y is permissible. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __cpymn(const mp_no *x, int m, mp_no *y, int n) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- long i,k;
|
|
|
8ae002 |
- long n2 = n;
|
|
|
8ae002 |
- long m2 = m;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- EY = EX; k=MIN(m2,n2);
|
|
|
8ae002 |
- for (i=0; i <= k; i++) Y[i] = X[i];
|
|
|
8ae002 |
- for ( ; i <= n2; i++) Y[i] = ZERO;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Convert a multiple precision number *x into a double precision */
|
|
|
8ae002 |
-/* number *y, normalized case (|x| >= 2**(-1022))) */
|
|
|
8ae002 |
-static void norm(const mp_no *x, double *y, int p)
|
|
|
8ae002 |
-{
|
|
|
8ae002 |
- #define R radixi.d
|
|
|
8ae002 |
- long i;
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- int k;
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
- double a,c,u,v,z[5];
|
|
|
8ae002 |
- if (p<5) {
|
|
|
8ae002 |
- if (p==1) c = X[1];
|
|
|
8ae002 |
- else if (p==2) c = X[1] + R* X[2];
|
|
|
8ae002 |
- else if (p==3) c = X[1] + R*(X[2] + R* X[3]);
|
|
|
8ae002 |
- else if (p==4) c =(X[1] + R* X[2]) + R*R*(X[3] + R*X[4]);
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- for (a=ONE, z[1]=X[1]; z[1] < TWO23; )
|
|
|
8ae002 |
- {a *= TWO; z[1] *= TWO; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (i=2; i<5; i++) {
|
|
|
8ae002 |
- z[i] = X[i]*a;
|
|
|
8ae002 |
- u = (z[i] + CUTTER)-CUTTER;
|
|
|
8ae002 |
- if (u > z[i]) u -= RADIX;
|
|
|
8ae002 |
- z[i] -= u;
|
|
|
8ae002 |
- z[i-1] += u*RADIXI;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- u = (z[3] + TWO71) - TWO71;
|
|
|
8ae002 |
- if (u > z[3]) u -= TWO19;
|
|
|
8ae002 |
- v = z[3]-u;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (v == TWO18) {
|
|
|
8ae002 |
- if (z[4] == ZERO) {
|
|
|
8ae002 |
- for (i=5; i <= p; i++) {
|
|
|
8ae002 |
- if (X[i] == ZERO) continue;
|
|
|
8ae002 |
- else {z[3] += ONE; break; }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else z[3] += ONE;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- c = (z[1] + R *(z[2] + R * z[3]))/a;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
|
|
|
8ae002 |
- c *= X[0];
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (i=1; i
|
|
|
8ae002 |
- for (i=1; i>EX; i--) c *= RADIXI;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- *y = c;
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-#undef R
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Convert a multiple precision number *x into a double precision */
|
|
|
8ae002 |
-/* number *y, denormalized case (|x| < 2**(-1022))) */
|
|
|
8ae002 |
-static void denorm(const mp_no *x, double *y, int p)
|
|
|
8ae002 |
+/* Define __mul and __sqr and use the rest from generic code. */
|
|
|
8ae002 |
+#define NO__MUL
|
|
|
8ae002 |
+#define NO__SQR
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#include <sysdeps/ieee754/dbl-64/mpa.c>
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Multiply *X and *Y and store result in *Z. X and Y may overlap but not X
|
|
|
8ae002 |
+ and Z or Y and Z. For P in [1, 2, 3], the exact result is truncated to P
|
|
|
8ae002 |
+ digits. In case P > 3 the error is bounded by 1.001 ULP. */
|
|
|
8ae002 |
+void
|
|
|
8ae002 |
+__mul (const mp_no *x, const mp_no *y, mp_no *z, int p)
|
|
|
8ae002 |
{
|
|
|
8ae002 |
- long i,k;
|
|
|
8ae002 |
+ long i, i1, i2, j, k, k2;
|
|
|
8ae002 |
long p2 = p;
|
|
|
8ae002 |
- double c,u,z[5];
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- double a,v;
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
+ double u, zk, zk2;
|
|
|
8ae002 |
|
|
|
8ae002 |
-#define R radixi.d
|
|
|
8ae002 |
- if (EX<-44 || (EX==-44 && X[1]
|
|
|
8ae002 |
- { *y=ZERO; return; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (p2==1) {
|
|
|
8ae002 |
- if (EX==-42) {z[1]=X[1]+TWO10; z[2]=ZERO; z[3]=ZERO; k=3;}
|
|
|
8ae002 |
- else if (EX==-43) {z[1]= TWO10; z[2]=X[1]; z[3]=ZERO; k=2;}
|
|
|
8ae002 |
- else {z[1]= TWO10; z[2]=ZERO; z[3]=X[1]; k=1;}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else if (p2==2) {
|
|
|
8ae002 |
- if (EX==-42) {z[1]=X[1]+TWO10; z[2]=X[2]; z[3]=ZERO; k=3;}
|
|
|
8ae002 |
- else if (EX==-43) {z[1]= TWO10; z[2]=X[1]; z[3]=X[2]; k=2;}
|
|
|
8ae002 |
- else {z[1]= TWO10; z[2]=ZERO; z[3]=X[1]; k=1;}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if (EX==-42) {z[1]=X[1]+TWO10; z[2]=X[2]; k=3;}
|
|
|
8ae002 |
- else if (EX==-43) {z[1]= TWO10; z[2]=X[1]; k=2;}
|
|
|
8ae002 |
- else {z[1]= TWO10; z[2]=ZERO; k=1;}
|
|
|
8ae002 |
- z[3] = X[k];
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- u = (z[3] + TWO57) - TWO57;
|
|
|
8ae002 |
- if (u > z[3]) u -= TWO5;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (u==z[3]) {
|
|
|
8ae002 |
- for (i=k+1; i <= p2; i++) {
|
|
|
8ae002 |
- if (X[i] == ZERO) continue;
|
|
|
8ae002 |
- else {z[3] += ONE; break; }
|
|
|
8ae002 |
+ /* Is z=0? */
|
|
|
8ae002 |
+ if (__glibc_unlikely (X[0] * Y[0] == 0))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Z[0] = 0;
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- c = X[0]*((z[1] + R*(z[2] + R*z[3])) - TWO10);
|
|
|
8ae002 |
|
|
|
8ae002 |
- *y = c*TWOM1032;
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#undef R
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Convert a multiple precision number *x into a double precision number *y. */
|
|
|
8ae002 |
-/* The result is correctly rounded to the nearest/even. *x is left unchanged */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __mp_dbl(const mp_no *x, double *y, int p) {
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- int i,k;
|
|
|
8ae002 |
- double a,c,u,v,z[5];
|
|
|
8ae002 |
+ /* Multiply, add and carry */
|
|
|
8ae002 |
+ k2 = (p2 < 3) ? p2 + p2 : p2 + 3;
|
|
|
8ae002 |
+ zk = Z[k2] = 0;
|
|
|
8ae002 |
+ for (k = k2; k > 1;)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ if (k > p2)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ i1 = k - p2;
|
|
|
8ae002 |
+ i2 = p2 + 1;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ i1 = 1;
|
|
|
8ae002 |
+ i2 = k;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+#if 1
|
|
|
8ae002 |
+ /* Rearrange this inner loop to allow the fmadd instructions to be
|
|
|
8ae002 |
+ independent and execute in parallel on processors that have
|
|
|
8ae002 |
+ dual symmetrical FP pipelines. */
|
|
|
8ae002 |
+ if (i1 < (i2 - 1))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ /* Make sure we have at least 2 iterations. */
|
|
|
8ae002 |
+ if (((i2 - i1) & 1L) == 1L)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ /* Handle the odd iterations case. */
|
|
|
8ae002 |
+ zk2 = x->d[i2 - 1] * y->d[i1];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ zk2 = 0.0;
|
|
|
8ae002 |
+ /* Do two multiply/adds per loop iteration, using independent
|
|
|
8ae002 |
+ accumulators; zk and zk2. */
|
|
|
8ae002 |
+ for (i = i1, j = i2 - 1; i < i2 - 1; i += 2, j -= 2)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ zk += x->d[i] * y->d[j];
|
|
|
8ae002 |
+ zk2 += x->d[i + 1] * y->d[j - 1];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ zk += zk2; /* Final sum. */
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ else
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ /* Special case when iterations is 1. */
|
|
|
8ae002 |
+ zk += x->d[i1] * y->d[i1];
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+#else
|
|
|
8ae002 |
+ /* The original code. */
|
|
|
8ae002 |
+ for (i = i1, j = i2 - 1; i < i2; i++, j--)
|
|
|
8ae002 |
+ zk += X[i] * Y[j];
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] == ZERO) {*y = ZERO; return; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (EX> -42) norm(x,y,p);
|
|
|
8ae002 |
- else if (EX==-42 && X[1]>=TWO10) norm(x,y,p);
|
|
|
8ae002 |
- else denorm(x,y,p);
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* dbl_mp() converts a double precision number x into a multiple precision */
|
|
|
8ae002 |
-/* number *y. If the precision p is too small the result is truncated. x is */
|
|
|
8ae002 |
-/* left unchanged. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __dbl_mp(double x, mp_no *y, int p) {
|
|
|
8ae002 |
+ u = (zk + CUTTER) - CUTTER;
|
|
|
8ae002 |
+ if (u > zk)
|
|
|
8ae002 |
+ u -= RADIX;
|
|
|
8ae002 |
+ Z[k] = zk - u;
|
|
|
8ae002 |
+ zk = u * RADIXI;
|
|
|
8ae002 |
+ --k;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ Z[k] = zk;
|
|
|
8ae002 |
|
|
|
8ae002 |
- long i,n;
|
|
|
8ae002 |
- long p2 = p;
|
|
|
8ae002 |
- double u;
|
|
|
8ae002 |
+ int e = EX + EY;
|
|
|
8ae002 |
+ /* Is there a carry beyond the most significant digit? */
|
|
|
8ae002 |
+ if (Z[1] == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ for (i = 1; i <= p2; i++)
|
|
|
8ae002 |
+ Z[i] = Z[i + 1];
|
|
|
8ae002 |
+ e--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
- /* Sign */
|
|
|
8ae002 |
- if (x == ZERO) {Y[0] = ZERO; return; }
|
|
|
8ae002 |
- else if (x > ZERO) Y[0] = ONE;
|
|
|
8ae002 |
- else {Y[0] = MONE; x=-x; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Exponent */
|
|
|
8ae002 |
- for (EY=ONE; x >= RADIX; EY += ONE) x *= RADIXI;
|
|
|
8ae002 |
- for ( ; x < ONE; EY -= ONE) x *= RADIX;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Digits */
|
|
|
8ae002 |
- n=MIN(p2,4);
|
|
|
8ae002 |
- for (i=1; i<=n; i++) {
|
|
|
8ae002 |
- u = (x + TWO52) - TWO52;
|
|
|
8ae002 |
- if (u>x) u -= ONE;
|
|
|
8ae002 |
- Y[i] = u; x -= u; x *= RADIX; }
|
|
|
8ae002 |
- for ( ; i<=p2; i++) Y[i] = ZERO;
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
+ EZ = e;
|
|
|
8ae002 |
+ Z[0] = X[0] * Y[0];
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
+/* Square *X and store result in *Y. X and Y may not overlap. For P in
|
|
|
8ae002 |
+ [1, 2, 3], the exact result is truncated to P digits. In case P > 3 the
|
|
|
8ae002 |
+ error is bounded by 1.001 ULP. This is a faster special case of
|
|
|
8ae002 |
+ multiplication. */
|
|
|
8ae002 |
+void
|
|
|
8ae002 |
+__sqr (const mp_no *x, mp_no *y, int p)
|
|
|
8ae002 |
+{
|
|
|
8ae002 |
+ long i, j, k, ip;
|
|
|
8ae002 |
+ double u, yk;
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* add_magnitudes() adds the magnitudes of *x & *y assuming that */
|
|
|
8ae002 |
-/* abs(*x) >= abs(*y) > 0. */
|
|
|
8ae002 |
-/* The sign of the sum *z is undefined. x&y may overlap but not x&z or y&z. */
|
|
|
8ae002 |
-/* No guard digit is used. The result equals the exact sum, truncated. */
|
|
|
8ae002 |
-/* *x & *y are left unchanged. */
|
|
|
8ae002 |
+ /* Is z=0? */
|
|
|
8ae002 |
+ if (__glibc_unlikely (X[0] == 0))
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ Y[0] = 0;
|
|
|
8ae002 |
+ return;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-static void add_magnitudes(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
+ /* We need not iterate through all X's since it's pointless to
|
|
|
8ae002 |
+ multiply zeroes. */
|
|
|
8ae002 |
+ for (ip = p; ip > 0; ip--)
|
|
|
8ae002 |
+ if (X[ip] != 0)
|
|
|
8ae002 |
+ break;
|
|
|
8ae002 |
|
|
|
8ae002 |
- long i,j,k;
|
|
|
8ae002 |
- long p2 = p;
|
|
|
8ae002 |
+ k = (__glibc_unlikely (p < 3)) ? p + p : p + 3;
|
|
|
8ae002 |
|
|
|
8ae002 |
- EZ = EX;
|
|
|
8ae002 |
+ while (k > 2 * ip + 1)
|
|
|
8ae002 |
+ Y[k--] = 0;
|
|
|
8ae002 |
|
|
|
8ae002 |
- i=p2; j=p2+ EY - EX; k=p2+1;
|
|
|
8ae002 |
+ yk = 0;
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (j<1)
|
|
|
8ae002 |
- {__cpy(x,z,p); return; }
|
|
|
8ae002 |
- else Z[k] = ZERO;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; j>0; i--,j--) {
|
|
|
8ae002 |
- Z[k] += X[i] + Y[j];
|
|
|
8ae002 |
- if (Z[k] >= RADIX) {
|
|
|
8ae002 |
- Z[k] -= RADIX;
|
|
|
8ae002 |
- Z[--k] = ONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; i>0; i--) {
|
|
|
8ae002 |
- Z[k] += X[i];
|
|
|
8ae002 |
- if (Z[k] >= RADIX) {
|
|
|
8ae002 |
- Z[k] -= RADIX;
|
|
|
8ae002 |
- Z[--k] = ONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (Z[1] == ZERO) {
|
|
|
8ae002 |
- for (i=1; i<=p2; i++) Z[i] = Z[i+1]; }
|
|
|
8ae002 |
- else EZ += ONE;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
+ while (k > p)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ double yk2 = 0.0;
|
|
|
8ae002 |
+ long lim = k / 2;
|
|
|
8ae002 |
|
|
|
8ae002 |
+ if (k % 2 == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ yk += X[lim] * X[lim];
|
|
|
8ae002 |
+ lim--;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
|
|
|
8ae002 |
-/* sub_magnitudes() subtracts the magnitudes of *x & *y assuming that */
|
|
|
8ae002 |
-/* abs(*x) > abs(*y) > 0. */
|
|
|
8ae002 |
-/* The sign of the difference *z is undefined. x&y may overlap but not x&z */
|
|
|
8ae002 |
-/* or y&z. One guard digit is used. The error is less than one ulp. */
|
|
|
8ae002 |
-/* *x & *y are left unchanged. */
|
|
|
8ae002 |
+ /* In __mul, this loop (and the one within the next while loop) run
|
|
|
8ae002 |
+ between a range to calculate the mantissa as follows:
|
|
|
8ae002 |
|
|
|
8ae002 |
-static void sub_magnitudes(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
+ Z[k] = X[k] * Y[n] + X[k+1] * Y[n-1] ... + X[n-1] * Y[k+1]
|
|
|
8ae002 |
+ + X[n] * Y[k]
|
|
|
8ae002 |
|
|
|
8ae002 |
- long i,j,k;
|
|
|
8ae002 |
- long p2 = p;
|
|
|
8ae002 |
+ For X == Y, we can get away with summing halfway and doubling the
|
|
|
8ae002 |
+ result. For cases where the range size is even, the mid-point needs
|
|
|
8ae002 |
+ to be added separately (above). */
|
|
|
8ae002 |
+ for (i = k - p, j = p; i <= lim; i++, j--)
|
|
|
8ae002 |
+ yk2 += X[i] * X[j];
|
|
|
8ae002 |
|
|
|
8ae002 |
- EZ = EX;
|
|
|
8ae002 |
+ yk += 2.0 * yk2;
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (EX == EY) {
|
|
|
8ae002 |
- i=j=k=p2;
|
|
|
8ae002 |
- Z[k] = Z[k+1] = ZERO; }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- j= EX - EY;
|
|
|
8ae002 |
- if (j > p2) {__cpy(x,z,p); return; }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- i=p2; j=p2+1-j; k=p2;
|
|
|
8ae002 |
- if (Y[j] > ZERO) {
|
|
|
8ae002 |
- Z[k+1] = RADIX - Y[j--];
|
|
|
8ae002 |
- Z[k] = MONE; }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- Z[k+1] = ZERO;
|
|
|
8ae002 |
- Z[k] = ZERO; j--;}
|
|
|
8ae002 |
+ u = (yk + CUTTER) - CUTTER;
|
|
|
8ae002 |
+ if (u > yk)
|
|
|
8ae002 |
+ u -= RADIX;
|
|
|
8ae002 |
+ Y[k--] = yk - u;
|
|
|
8ae002 |
+ yk = u * RADIXI;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; j>0; i--,j--) {
|
|
|
8ae002 |
- Z[k] += (X[i] - Y[j]);
|
|
|
8ae002 |
- if (Z[k] < ZERO) {
|
|
|
8ae002 |
- Z[k] += RADIX;
|
|
|
8ae002 |
- Z[--k] = MONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (; i>0; i--) {
|
|
|
8ae002 |
- Z[k] += X[i];
|
|
|
8ae002 |
- if (Z[k] < ZERO) {
|
|
|
8ae002 |
- Z[k] += RADIX;
|
|
|
8ae002 |
- Z[--k] = MONE; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- Z[--k] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (i=1; Z[i] == ZERO; i++) ;
|
|
|
8ae002 |
- EZ = EZ - i + 1;
|
|
|
8ae002 |
- for (k=1; i <= p2+1; )
|
|
|
8ae002 |
- Z[k++] = Z[i++];
|
|
|
8ae002 |
- for (; k <= p2; )
|
|
|
8ae002 |
- Z[k++] = ZERO;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Add two multiple precision numbers. Set *z = *x + *y. x&y may overlap */
|
|
|
8ae002 |
-/* but not x&z or y&z. One guard digit is used. The error is less than */
|
|
|
8ae002 |
-/* one ulp. *x & *y are left unchanged. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __add(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- int n;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] == ZERO) {__cpy(y,z,p); return; }
|
|
|
8ae002 |
- else if (Y[0] == ZERO) {__cpy(x,z,p); return; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] == Y[0]) {
|
|
|
8ae002 |
- if (__acr(x,y,p) > 0) {add_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else {add_magnitudes(y,x,z,p); Z[0] = Y[0]; }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if ((n=__acr(x,y,p)) == 1) {sub_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else if (n == -1) {sub_magnitudes(y,x,z,p); Z[0] = Y[0]; }
|
|
|
8ae002 |
- else Z[0] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Subtract two multiple precision numbers. *z is set to *x - *y. x&y may */
|
|
|
8ae002 |
-/* overlap but not x&z or y&z. One guard digit is used. The error is */
|
|
|
8ae002 |
-/* less than one ulp. *x & *y are left unchanged. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __sub(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- int n;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] == ZERO) {__cpy(y,z,p); Z[0] = -Z[0]; return; }
|
|
|
8ae002 |
- else if (Y[0] == ZERO) {__cpy(x,z,p); return; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] != Y[0]) {
|
|
|
8ae002 |
- if (__acr(x,y,p) > 0) {add_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else {add_magnitudes(y,x,z,p); Z[0] = -Y[0]; }
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else {
|
|
|
8ae002 |
- if ((n=__acr(x,y,p)) == 1) {sub_magnitudes(x,y,z,p); Z[0] = X[0]; }
|
|
|
8ae002 |
- else if (n == -1) {sub_magnitudes(y,x,z,p); Z[0] = -Y[0]; }
|
|
|
8ae002 |
- else Z[0] = ZERO;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Multiply two multiple precision numbers. *z is set to *x * *y. x&y */
|
|
|
8ae002 |
-/* may overlap but not x&z or y&z. In case p=1,2,3 the exact result is */
|
|
|
8ae002 |
-/* truncated to p digits. In case p>3 the error is bounded by 1.001 ulp. */
|
|
|
8ae002 |
-/* *x & *y are left unchanged. */
|
|
|
8ae002 |
|
|
|
8ae002 |
-void __mul(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- long i, i1, i2, j, k, k2;
|
|
|
8ae002 |
- long p2 = p;
|
|
|
8ae002 |
- double u, zk, zk2;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Is z=0? */
|
|
|
8ae002 |
- if (X[0]*Y[0]==ZERO)
|
|
|
8ae002 |
- { Z[0]=ZERO; return; }
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Multiply, add and carry */
|
|
|
8ae002 |
- k2 = (p2<3) ? p2+p2 : p2+3;
|
|
|
8ae002 |
- zk = Z[k2]=ZERO;
|
|
|
8ae002 |
- for (k=k2; k>1; ) {
|
|
|
8ae002 |
- if (k > p2) {i1=k-p2; i2=p2+1; }
|
|
|
8ae002 |
- else {i1=1; i2=k; }
|
|
|
8ae002 |
-#if 1
|
|
|
8ae002 |
- /* rearange this inner loop to allow the fmadd instructions to be
|
|
|
8ae002 |
- independent and execute in parallel on processors that have
|
|
|
8ae002 |
- dual symetrical FP pipelines. */
|
|
|
8ae002 |
- if (i1 < (i2-1))
|
|
|
8ae002 |
+ while (k > 1)
|
|
|
8ae002 |
{
|
|
|
8ae002 |
- /* make sure we have at least 2 iterations */
|
|
|
8ae002 |
- if (((i2 - i1) & 1L) == 1L)
|
|
|
8ae002 |
- {
|
|
|
8ae002 |
- /* Handle the odd iterations case. */
|
|
|
8ae002 |
- zk2 = x->d[i2-1]*y->d[i1];
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- zk2 = zero.d;
|
|
|
8ae002 |
- /* Do two multiply/adds per loop iteration, using independent
|
|
|
8ae002 |
- accumulators; zk and zk2. */
|
|
|
8ae002 |
- for (i=i1,j=i2-1; i
|
|
|
8ae002 |
- {
|
|
|
8ae002 |
- zk += x->d[i]*y->d[j];
|
|
|
8ae002 |
- zk2 += x->d[i+1]*y->d[j-1];
|
|
|
8ae002 |
+ double yk2 = 0.0;
|
|
|
8ae002 |
+ long lim = k / 2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ if (k % 2 == 0)
|
|
|
8ae002 |
+ {
|
|
|
8ae002 |
+ yk += X[lim] * X[lim];
|
|
|
8ae002 |
+ lim--;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
- zk += zk2; /* final sum. */
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Likewise for this loop. */
|
|
|
8ae002 |
+ for (i = 1, j = k - 1; i <= lim; i++, j--)
|
|
|
8ae002 |
+ yk2 += X[i] * X[j];
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ yk += 2.0 * yk2;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ u = (yk + CUTTER) - CUTTER;
|
|
|
8ae002 |
+ if (u > yk)
|
|
|
8ae002 |
+ u -= RADIX;
|
|
|
8ae002 |
+ Y[k--] = yk - u;
|
|
|
8ae002 |
+ yk = u * RADIXI;
|
|
|
8ae002 |
+ }
|
|
|
8ae002 |
+ Y[k] = yk;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ /* Squares are always positive. */
|
|
|
8ae002 |
+ Y[0] = 1.0;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ int e = EX * 2;
|
|
|
8ae002 |
+ /* Is there a carry beyond the most significant digit? */
|
|
|
8ae002 |
+ if (__glibc_unlikely (Y[1] == 0))
|
|
|
8ae002 |
{
|
|
|
8ae002 |
- /* Special case when iterations is 1. */
|
|
|
8ae002 |
- zk += x->d[i1]*y->d[i1];
|
|
|
8ae002 |
+ for (i = 1; i <= p; i++)
|
|
|
8ae002 |
+ Y[i] = Y[i + 1];
|
|
|
8ae002 |
+ e--;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
-#else
|
|
|
8ae002 |
- /* The orginal code. */
|
|
|
8ae002 |
- for (i=i1,j=i2-1; i
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- u = (zk + CUTTER)-CUTTER;
|
|
|
8ae002 |
- if (u > zk) u -= RADIX;
|
|
|
8ae002 |
- Z[k] = zk - u;
|
|
|
8ae002 |
- zk = u*RADIXI;
|
|
|
8ae002 |
- --k;
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- Z[k] = zk;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- /* Is there a carry beyond the most significant digit? */
|
|
|
8ae002 |
- if (Z[1] == ZERO) {
|
|
|
8ae002 |
- for (i=1; i<=p2; i++) Z[i]=Z[i+1];
|
|
|
8ae002 |
- EZ = EX + EY - 1; }
|
|
|
8ae002 |
- else
|
|
|
8ae002 |
- EZ = EX + EY;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- Z[0] = X[0] * Y[0];
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Invert a multiple precision number. Set *y = 1 / *x. */
|
|
|
8ae002 |
-/* Relative error bound = 1.001*r**(1-p) for p=2, 1.063*r**(1-p) for p=3, */
|
|
|
8ae002 |
-/* 2.001*r**(1-p) for p>3. */
|
|
|
8ae002 |
-/* *x=0 is not permissible. *x is left unchanged. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __inv(const mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
- long i;
|
|
|
8ae002 |
-#if 0
|
|
|
8ae002 |
- int l;
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
- double t;
|
|
|
8ae002 |
- mp_no z,w;
|
|
|
8ae002 |
- static const int np1[] = {0,0,0,0,1,2,2,2,2,3,3,3,3,3,3,3,3,3,
|
|
|
8ae002 |
- 4,4,4,4,4,4,4,4,4,4,4,4,4,4,4};
|
|
|
8ae002 |
- const mp_no mptwo = {1,{1.0,2.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
- 0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
- 0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
- 0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}};
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- __cpy(x,&z,p); z.e=0; __mp_dbl(&z,&t,p);
|
|
|
8ae002 |
- t=ONE/t; __dbl_mp(t,y,p); EY -= EX;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- for (i=0; i
|
|
|
8ae002 |
- __cpy(y,&w,p);
|
|
|
8ae002 |
- __mul(x,&w,y,p);
|
|
|
8ae002 |
- __sub(&mptwo,y,&z,p);
|
|
|
8ae002 |
- __mul(&w,&z,y,p);
|
|
|
8ae002 |
- }
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
-}
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/* Divide one multiple precision number by another.Set *z = *x / *y. *x & *y */
|
|
|
8ae002 |
-/* are left unchanged. x&y may overlap but not x&z or y&z. */
|
|
|
8ae002 |
-/* Relative error bound = 2.001*r**(1-p) for p=2, 2.063*r**(1-p) for p=3 */
|
|
|
8ae002 |
-/* and 3.001*r**(1-p) for p>3. *y=0 is not permissible. */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-void __dvd(const mp_no *x, const mp_no *y, mp_no *z, int p) {
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- mp_no w;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
- if (X[0] == ZERO) Z[0] = ZERO;
|
|
|
8ae002 |
- else {__inv(y,&w,p); __mul(x,&w,z,p);}
|
|
|
8ae002 |
- return;
|
|
|
8ae002 |
+ EY = e;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/atnat.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/atnat.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/atnat.h
|
|
|
8ae002 |
@@ -138,8 +138,6 @@
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
-#define ZERO zero.d
|
|
|
8ae002 |
-#define ONE one.d
|
|
|
8ae002 |
#define A a.d
|
|
|
8ae002 |
#define B b.d
|
|
|
8ae002 |
#define C c.d
|
|
|
8ae002 |
@@ -160,7 +158,5 @@
|
|
|
8ae002 |
#define U6 u6.d
|
|
|
8ae002 |
#define U7 u7.d
|
|
|
8ae002 |
#define U8 u8.d
|
|
|
8ae002 |
-#define TWO8 two8.d
|
|
|
8ae002 |
-#define TWO52 two52.d
|
|
|
8ae002 |
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/atnat2.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/atnat2.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/atnat2.h
|
|
|
8ae002 |
@@ -174,11 +174,4 @@
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
-#define ZERO zero.d
|
|
|
8ae002 |
-#define MZERO mzero.d
|
|
|
8ae002 |
-#define ONE one.d
|
|
|
8ae002 |
-#define TWO8 two8.d
|
|
|
8ae002 |
-#define TWO52 two52.d
|
|
|
8ae002 |
-#define TWOM1022 twom1022.d
|
|
|
8ae002 |
-
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpa2.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpa2.h
|
|
|
8ae002 |
+++ /dev/null
|
|
|
8ae002 |
@@ -1,94 +0,0 @@
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/*
|
|
|
8ae002 |
- * IBM Accurate Mathematical Library
|
|
|
8ae002 |
- * Written by International Business Machines Corp.
|
|
|
8ae002 |
- * Copyright (C) 2001 Free Software Foundation, Inc.
|
|
|
8ae002 |
- *
|
|
|
8ae002 |
- * This program is free software; you can redistribute it and/or modify
|
|
|
8ae002 |
- * it under the terms of the GNU Lesser General Public License as published by
|
|
|
8ae002 |
- * the Free Software Foundation; either version 2.1 of the License, or
|
|
|
8ae002 |
- * (at your option) any later version.
|
|
|
8ae002 |
- *
|
|
|
8ae002 |
- * This program is distributed in the hope that it will be useful,
|
|
|
8ae002 |
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
8ae002 |
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
8ae002 |
- * GNU Lesser General Public License for more details.
|
|
|
8ae002 |
- *
|
|
|
8ae002 |
- * You should have received a copy of the GNU Lesser General Public License
|
|
|
8ae002 |
- * along with this program; if not, see <http://www.gnu.org/licenses/>.
|
|
|
8ae002 |
- */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-/**************************************************************************/
|
|
|
8ae002 |
-/* */
|
|
|
8ae002 |
-/* MODULE_NAME:mpa2.h */
|
|
|
8ae002 |
-/* */
|
|
|
8ae002 |
-/* */
|
|
|
8ae002 |
-/* variables prototype and definition according to type of processor */
|
|
|
8ae002 |
-/* types definition */
|
|
|
8ae002 |
-/**************************************************************************/
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#ifndef MPA2_H
|
|
|
8ae002 |
-#define MPA2_H
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#ifdef BIG_ENDI
|
|
|
8ae002 |
-static const number
|
|
|
8ae002 |
-/**/ radix = {{0x41700000, 0x00000000} }, /* 2**24 */
|
|
|
8ae002 |
-/**/ radixi = {{0x3e700000, 0x00000000} }, /* 2**-24 */
|
|
|
8ae002 |
-/**/ cutter = {{0x44b00000, 0x00000000} }, /* 2**76 */
|
|
|
8ae002 |
-/**/ zero = {{0x00000000, 0x00000000} }, /* 0 */
|
|
|
8ae002 |
-/**/ one = {{0x3ff00000, 0x00000000} }, /* 1 */
|
|
|
8ae002 |
-/**/ mone = {{0xbff00000, 0x00000000} }, /* -1 */
|
|
|
8ae002 |
-/**/ two = {{0x40000000, 0x00000000} }, /* 2 */
|
|
|
8ae002 |
-/**/ two5 = {{0x40400000, 0x00000000} }, /* 2**5 */
|
|
|
8ae002 |
-/**/ two10 = {{0x40900000, 0x00000000} }, /* 2**10 */
|
|
|
8ae002 |
-/**/ two18 = {{0x41100000, 0x00000000} }, /* 2**18 */
|
|
|
8ae002 |
-/**/ two19 = {{0x41200000, 0x00000000} }, /* 2**19 */
|
|
|
8ae002 |
-/**/ two23 = {{0x41600000, 0x00000000} }, /* 2**23 */
|
|
|
8ae002 |
-/**/ two52 = {{0x43300000, 0x00000000} }, /* 2**52 */
|
|
|
8ae002 |
-/**/ two57 = {{0x43800000, 0x00000000} }, /* 2**57 */
|
|
|
8ae002 |
-/**/ two71 = {{0x44600000, 0x00000000} }, /* 2**71 */
|
|
|
8ae002 |
-/**/ twom1032 = {{0x00000400, 0x00000000} }; /* 2**-1032 */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#else
|
|
|
8ae002 |
-#ifdef LITTLE_ENDI
|
|
|
8ae002 |
-static const number
|
|
|
8ae002 |
-/**/ radix = {{0x00000000, 0x41700000} }, /* 2**24 */
|
|
|
8ae002 |
-/**/ radixi = {{0x00000000, 0x3e700000} }, /* 2**-24 */
|
|
|
8ae002 |
-/**/ cutter = {{0x00000000, 0x44b00000} }, /* 2**76 */
|
|
|
8ae002 |
-/**/ zero = {{0x00000000, 0x00000000} }, /* 0 */
|
|
|
8ae002 |
-/**/ one = {{0x00000000, 0x3ff00000} }, /* 1 */
|
|
|
8ae002 |
-/**/ mone = {{0x00000000, 0xbff00000} }, /* -1 */
|
|
|
8ae002 |
-/**/ two = {{0x00000000, 0x40000000} }, /* 2 */
|
|
|
8ae002 |
-/**/ two5 = {{0x00000000, 0x40400000} }, /* 2**5 */
|
|
|
8ae002 |
-/**/ two10 = {{0x00000000, 0x40900000} }, /* 2**10 */
|
|
|
8ae002 |
-/**/ two18 = {{0x00000000, 0x41100000} }, /* 2**18 */
|
|
|
8ae002 |
-/**/ two19 = {{0x00000000, 0x41200000} }, /* 2**19 */
|
|
|
8ae002 |
-/**/ two23 = {{0x00000000, 0x41600000} }, /* 2**23 */
|
|
|
8ae002 |
-/**/ two52 = {{0x00000000, 0x43300000} }, /* 2**52 */
|
|
|
8ae002 |
-/**/ two57 = {{0x00000000, 0x43800000} }, /* 2**57 */
|
|
|
8ae002 |
-/**/ two71 = {{0x00000000, 0x44600000} }, /* 2**71 */
|
|
|
8ae002 |
-/**/ twom1032 = {{0x00000000, 0x00000400} }; /* 2**-1032 */
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#define RADIX radix.d
|
|
|
8ae002 |
-#define RADIXI radixi.d
|
|
|
8ae002 |
-#define CUTTER cutter.d
|
|
|
8ae002 |
-#define ZERO zero.d
|
|
|
8ae002 |
-#define ONE one.d
|
|
|
8ae002 |
-#define MONE mone.d
|
|
|
8ae002 |
-#define TWO two.d
|
|
|
8ae002 |
-#define TWO5 two5.d
|
|
|
8ae002 |
-#define TWO10 two10.d
|
|
|
8ae002 |
-#define TWO18 two18.d
|
|
|
8ae002 |
-#define TWO19 two19.d
|
|
|
8ae002 |
-#define TWO23 two23.d
|
|
|
8ae002 |
-#define TWO52 two52.d
|
|
|
8ae002 |
-#define TWO57 two57.d
|
|
|
8ae002 |
-#define TWO71 two71.d
|
|
|
8ae002 |
-#define TWOM1032 twom1032.d
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#endif
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpatan.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpatan.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpatan.h
|
|
|
8ae002 |
@@ -177,6 +177,3 @@ __atan_twonm1[33] = {
|
|
|
8ae002 |
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#define ONE __atan_one.d
|
|
|
8ae002 |
-#define TWO __atan_two.d
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpatan2.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpatan2.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpatan2.c
|
|
|
8ae002 |
@@ -49,18 +49,16 @@ void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
__mpatan2(mp_no *y, mp_no *x, mp_no *z, int p) {
|
|
|
8ae002 |
|
|
|
8ae002 |
- static const double ZERO = 0.0, ONE = 1.0;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
mp_no mpone = {0,{0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
|
|
|
8ae002 |
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}};
|
|
|
8ae002 |
mp_no mpt1,mpt2,mpt3;
|
|
|
8ae002 |
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (X[0] <= ZERO) {
|
|
|
8ae002 |
- mpone.e = 1; mpone.d[0] = mpone.d[1] = ONE;
|
|
|
8ae002 |
+ if (X[0] <= 0) {
|
|
|
8ae002 |
+ mpone.e = 1; mpone.d[0] = mpone.d[1] = 1;
|
|
|
8ae002 |
__dvd(x,y,&mpt1,p); __mul(&mpt1,&mpt1,&mpt2,p);
|
|
|
8ae002 |
- if (mpt1.d[0] != ZERO) mpt1.d[0] = ONE;
|
|
|
8ae002 |
+ if (mpt1.d[0] != 0) mpt1.d[0] = 1;
|
|
|
8ae002 |
__add(&mpt2,&mpone,&mpt3,p); __mpsqrt(&mpt3,&mpt2,p);
|
|
|
8ae002 |
__add(&mpt1,&mpt2,&mpt3,p); mpt3.d[0]=Y[0];
|
|
|
8ae002 |
__mpatan(&mpt3,&mpt1,p); __add(&mpt1,&mpt1,z,p);
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpexp.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpexp.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpexp.h
|
|
|
8ae002 |
@@ -159,11 +159,4 @@ extern const number __mpexp_half attribu
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
-#define RADIX __mpexp_radix.d
|
|
|
8ae002 |
-#define RADIXI __mpexp_radixi.d
|
|
|
8ae002 |
-#define ZERO __mpexp_zero.d
|
|
|
8ae002 |
-#define ONE __mpexp_one.d
|
|
|
8ae002 |
-#define TWO __mpexp_two.d
|
|
|
8ae002 |
-#define HALF __mpexp_half.d
|
|
|
8ae002 |
-
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpsqrt.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpsqrt.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpsqrt.h
|
|
|
8ae002 |
@@ -51,7 +51,4 @@ extern const int __mpsqrt_mp[33] attribu
|
|
|
8ae002 |
4,4,4,4,4,4,4,4,4};
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
-#define ONE __mpsqrt_one.d
|
|
|
8ae002 |
-#define HALFRAD __mpsqrt_halfrad.d
|
|
|
8ae002 |
-
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mptan.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mptan.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mptan.c
|
|
|
8ae002 |
@@ -47,8 +47,6 @@ void
|
|
|
8ae002 |
SECTION
|
|
|
8ae002 |
__mptan(double x, mp_no *mpy, int p) {
|
|
|
8ae002 |
|
|
|
8ae002 |
- static const double MONE = -1.0;
|
|
|
8ae002 |
-
|
|
|
8ae002 |
int n;
|
|
|
8ae002 |
mp_no mpw, mpc, mps;
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -56,7 +54,7 @@ __mptan(double x, mp_no *mpy, int p) {
|
|
|
8ae002 |
__c32(&mpw, &mpc, &mps, p); /* computing sin(x) and cos(x) */
|
|
|
8ae002 |
if (n) /* second or fourth quarter of unit circle */
|
|
|
8ae002 |
{ __dvd(&mpc,&mps,mpy,p);
|
|
|
8ae002 |
- mpy->d[0] *= MONE;
|
|
|
8ae002 |
+ mpy->d[0] *= -1;
|
|
|
8ae002 |
} /* tan is negative in this area */
|
|
|
8ae002 |
else __dvd(&mps,&mpc,mpy,p);
|
|
|
8ae002 |
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/ulog.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/ulog.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/ulog.h
|
|
|
8ae002 |
@@ -181,10 +181,6 @@
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
-#define ZERO zero.d
|
|
|
8ae002 |
-#define ONE one.d
|
|
|
8ae002 |
-#define HALF half.d
|
|
|
8ae002 |
-#define MHALF mhalf.d
|
|
|
8ae002 |
#define SQRT_2 sqrt_2.d
|
|
|
8ae002 |
#define DEL_U delu.d
|
|
|
8ae002 |
#define DEL_V delv.d
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/utan.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/utan.h
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/utan.h
|
|
|
8ae002 |
@@ -270,10 +270,4 @@
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
|
|
|
8ae002 |
-
|
|
|
8ae002 |
-#define ZERO zero.d
|
|
|
8ae002 |
-#define ONE one.d
|
|
|
8ae002 |
-#define MONE mone.d
|
|
|
8ae002 |
-#define TWO8 two8.d
|
|
|
8ae002 |
-
|
|
|
8ae002 |
#endif
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpa-arch.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- /dev/null
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpa-arch.h
|
|
|
8ae002 |
@@ -0,0 +1,47 @@
|
|
|
8ae002 |
+/* Overridable constants and operations.
|
|
|
8ae002 |
+ Copyright (C) 2013 Free Software Foundation, Inc.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ This program is free software; you can redistribute it and/or modify
|
|
|
8ae002 |
+ it under the terms of the GNU Lesser General Public License as published by
|
|
|
8ae002 |
+ the Free Software Foundation; either version 2.1 of the License, or
|
|
|
8ae002 |
+ (at your option) any later version.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ This program is distributed in the hope that it will be useful,
|
|
|
8ae002 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
8ae002 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
8ae002 |
+ GNU Lesser General Public License for more details.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ You should have received a copy of the GNU Lesser General Public License
|
|
|
8ae002 |
+ along with this program; if not, see <http://www.gnu.org/licenses/>. */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#include <stdint.h>
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+typedef long mantissa_t;
|
|
|
8ae002 |
+typedef int64_t mantissa_store_t;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#define TWOPOW(i) (1L << i)
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#define RADIX_EXP 24
|
|
|
8ae002 |
+#define RADIX TWOPOW (RADIX_EXP) /* 2^24 */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Divide D by RADIX and put the remainder in R. D must be a non-negative
|
|
|
8ae002 |
+ integral value. */
|
|
|
8ae002 |
+#define DIV_RADIX(d, r) \
|
|
|
8ae002 |
+ ({ \
|
|
|
8ae002 |
+ r = d & (RADIX - 1); \
|
|
|
8ae002 |
+ d >>= RADIX_EXP; \
|
|
|
8ae002 |
+ })
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Put the integer component of a double X in R and retain the fraction in
|
|
|
8ae002 |
+ X. This is used in extracting mantissa digits for MP_NO by using the
|
|
|
8ae002 |
+ integer portion of the current value of the number as the current mantissa
|
|
|
8ae002 |
+ digit and then scaling by RADIX to get the next mantissa digit in the same
|
|
|
8ae002 |
+ manner. */
|
|
|
8ae002 |
+#define INTEGER_OF(x, i) \
|
|
|
8ae002 |
+ ({ \
|
|
|
8ae002 |
+ i = (mantissa_t) x; \
|
|
|
8ae002 |
+ x -= i; \
|
|
|
8ae002 |
+ })
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Align IN down to F. The code assumes that F is a power of two. */
|
|
|
8ae002 |
+#define ALIGN_DOWN_TO(in, f) ((in) & -(f))
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/powerpc/power4/fpu/mpa-arch.h
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- /dev/null
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/power4/fpu/mpa-arch.h
|
|
|
8ae002 |
@@ -0,0 +1,56 @@
|
|
|
8ae002 |
+/* Overridable constants and operations.
|
|
|
8ae002 |
+ Copyright (C) 2013-2017 Free Software Foundation, Inc.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ This program is free software; you can redistribute it and/or modify
|
|
|
8ae002 |
+ it under the terms of the GNU Lesser General Public License as published by
|
|
|
8ae002 |
+ the Free Software Foundation; either version 2.1 of the License, or
|
|
|
8ae002 |
+ (at your option) any later version.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ This program is distributed in the hope that it will be useful,
|
|
|
8ae002 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
8ae002 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
8ae002 |
+ GNU Lesser General Public License for more details.
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+ You should have received a copy of the GNU Lesser General Public License
|
|
|
8ae002 |
+ along with this program; if not, see <http://www.gnu.org/licenses/>. */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+typedef double mantissa_t;
|
|
|
8ae002 |
+typedef double mantissa_store_t;
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#define TWOPOW(i) (0x1.0p##i)
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+#define RADIX TWOPOW (24) /* 2^24 */
|
|
|
8ae002 |
+#define CUTTER TWOPOW (76) /* 2^76 */
|
|
|
8ae002 |
+#define RADIXI 0x1.0p-24 /* 2^-24 */
|
|
|
8ae002 |
+#define TWO52 TWOPOW (52) /* 2^52 */
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Divide D by RADIX and put the remainder in R. */
|
|
|
8ae002 |
+#define DIV_RADIX(d,r) \
|
|
|
8ae002 |
+ ({ \
|
|
|
8ae002 |
+ double u = ((d) + CUTTER) - CUTTER; \
|
|
|
8ae002 |
+ if (u > (d)) \
|
|
|
8ae002 |
+ u -= RADIX; \
|
|
|
8ae002 |
+ r = (d) - u; \
|
|
|
8ae002 |
+ (d) = u * RADIXI; \
|
|
|
8ae002 |
+ })
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Put the integer component of a double X in R and retain the fraction in
|
|
|
8ae002 |
+ X. */
|
|
|
8ae002 |
+#define INTEGER_OF(x, r) \
|
|
|
8ae002 |
+ ({ \
|
|
|
8ae002 |
+ double u = ((x) + TWO52) - TWO52; \
|
|
|
8ae002 |
+ if (u > (x)) \
|
|
|
8ae002 |
+ u -= 1; \
|
|
|
8ae002 |
+ (r) = u; \
|
|
|
8ae002 |
+ (x) -= u; \
|
|
|
8ae002 |
+ })
|
|
|
8ae002 |
+
|
|
|
8ae002 |
+/* Align IN down to a multiple of F, where F is a power of two. */
|
|
|
8ae002 |
+#define ALIGN_DOWN_TO(in, f) \
|
|
|
8ae002 |
+ ({ \
|
|
|
8ae002 |
+ double factor = f * TWO52; \
|
|
|
8ae002 |
+ double u = (in + factor) - factor; \
|
|
|
8ae002 |
+ if (u > in) \
|
|
|
8ae002 |
+ u -= f; \
|
|
|
8ae002 |
+ u; \
|
|
|
8ae002 |
+ })
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/e_atan2.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/e_atan2.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/e_atan2.c
|
|
|
8ae002 |
@@ -98,15 +98,15 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
/* y=+-0 */
|
|
|
8ae002 |
if (uy==0x00000000) {
|
|
|
8ae002 |
if (dy==0x00000000) {
|
|
|
8ae002 |
- if ((ux&0x80000000)==0x00000000) return ZERO;
|
|
|
8ae002 |
+ if ((ux&0x80000000)==0x00000000) return 0;
|
|
|
8ae002 |
else return opi.d; } }
|
|
|
8ae002 |
else if (uy==0x80000000) {
|
|
|
8ae002 |
if (dy==0x00000000) {
|
|
|
8ae002 |
- if ((ux&0x80000000)==0x00000000) return MZERO;
|
|
|
8ae002 |
+ if ((ux&0x80000000)==0x00000000) return -0.0;
|
|
|
8ae002 |
else return mopi.d;} }
|
|
|
8ae002 |
|
|
|
8ae002 |
/* x=+-0 */
|
|
|
8ae002 |
- if (x==ZERO) {
|
|
|
8ae002 |
+ if (x==0) {
|
|
|
8ae002 |
if ((uy&0x80000000)==0x00000000) return hpi.d;
|
|
|
8ae002 |
else return mhpi.d; }
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -118,8 +118,8 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
else if (uy==0xfff00000) {
|
|
|
8ae002 |
if (dy==0x00000000) return mqpi.d; }
|
|
|
8ae002 |
else {
|
|
|
8ae002 |
- if ((uy&0x80000000)==0x00000000) return ZERO;
|
|
|
8ae002 |
- else return MZERO; }
|
|
|
8ae002 |
+ if ((uy&0x80000000)==0x00000000) return 0;
|
|
|
8ae002 |
+ else return -0.0; }
|
|
|
8ae002 |
}
|
|
|
8ae002 |
}
|
|
|
8ae002 |
else if (ux==0xfff00000) {
|
|
|
8ae002 |
@@ -141,14 +141,14 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
if (dy==0x00000000) return mhpi.d; }
|
|
|
8ae002 |
|
|
|
8ae002 |
/* either x/y or y/x is very close to zero */
|
|
|
8ae002 |
- ax = (x
|
|
|
8ae002 |
+ ax = (x<0) ? -x : x; ay = (y<0) ? -y : y;
|
|
|
8ae002 |
de = (uy & 0x7ff00000) - (ux & 0x7ff00000);
|
|
|
8ae002 |
- if (de>=ep) { return ((y>ZERO) ? hpi.d : mhpi.d); }
|
|
|
8ae002 |
+ if (de>=ep) { return ((y>0) ? hpi.d : mhpi.d); }
|
|
|
8ae002 |
else if (de<=em) {
|
|
|
8ae002 |
- if (x>ZERO) {
|
|
|
8ae002 |
+ if (x>0) {
|
|
|
8ae002 |
if ((z=ay/ax)
|
|
|
8ae002 |
else return signArctan2(y,z); }
|
|
|
8ae002 |
- else { return ((y>ZERO) ? opi.d : mopi.d); } }
|
|
|
8ae002 |
+ else { return ((y>0) ? opi.d : mopi.d); } }
|
|
|
8ae002 |
|
|
|
8ae002 |
/* if either x or y is extremely close to zero, scale abs(x), abs(y). */
|
|
|
8ae002 |
if (ax
|
|
|
8ae002 |
@@ -170,7 +170,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
EMULV(ay,u,v,vv,t1,t2,t3,t4,t5)
|
|
|
8ae002 |
du=((ax-v)-vv)/ay; }
|
|
|
8ae002 |
|
|
|
8ae002 |
- if (x>ZERO) {
|
|
|
8ae002 |
+ if (x>0) {
|
|
|
8ae002 |
|
|
|
8ae002 |
/* (i) x>0, abs(y)< abs(x): atan(ay/ax) */
|
|
|
8ae002 |
if (ay
|
|
|
8ae002 |
@@ -180,7 +180,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
MUL2(u,du,u,du,v,vv,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
s1=v*(f11.d+v*(f13.d+v*(f15.d+v*(f17.d+v*f19.d))));
|
|
|
8ae002 |
- ADD2(f9.d,ff9.d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(f9.d,ff9.d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(f7.d,ff7.d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -212,7 +212,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
EADD(t1,du,v,vv)
|
|
|
8ae002 |
s1=v*(hij[i][11].d+v*(hij[i][12].d+v*(hij[i][13].d+
|
|
|
8ae002 |
v*(hij[i][14].d+v* hij[i][15].d))));
|
|
|
8ae002 |
- ADD2(hij[i][9].d,hij[i][10].d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(hij[i][9].d,hij[i][10].d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][7].d,hij[i][8].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -237,7 +237,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
MUL2(u,du,u,du,v,vv,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
s1=v*(f11.d+v*(f13.d+v*(f15.d+v*(f17.d+v*f19.d))));
|
|
|
8ae002 |
- ADD2(f9.d,ff9.d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(f9.d,ff9.d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(f7.d,ff7.d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -265,7 +265,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
EADD(t1,du,v,vv)
|
|
|
8ae002 |
s1=v*(hij[i][11].d+v*(hij[i][12].d+v*(hij[i][13].d+
|
|
|
8ae002 |
v*(hij[i][14].d+v* hij[i][15].d))));
|
|
|
8ae002 |
- ADD2(hij[i][9].d,hij[i][10].d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(hij[i][9].d,hij[i][10].d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][7].d,hij[i][8].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -293,7 +293,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
MUL2(u,du,u,du,v,vv,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
s1=v*(f11.d+v*(f13.d+v*(f15.d+v*(f17.d+v*f19.d))));
|
|
|
8ae002 |
- ADD2(f9.d,ff9.d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(f9.d,ff9.d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(f7.d,ff7.d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -321,7 +321,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
EADD(t1,du,v,vv)
|
|
|
8ae002 |
s1=v*(hij[i][11].d+v*(hij[i][12].d+v*(hij[i][13].d+
|
|
|
8ae002 |
v*(hij[i][14].d+v* hij[i][15].d))));
|
|
|
8ae002 |
- ADD2(hij[i][9].d,hij[i][10].d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(hij[i][9].d,hij[i][10].d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][7].d,hij[i][8].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -347,7 +347,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
MUL2(u,du,u,du,v,vv,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
s1=v*(f11.d+v*(f13.d+v*(f15.d+v*(f17.d+v*f19.d))));
|
|
|
8ae002 |
- ADD2(f9.d,ff9.d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(f9.d,ff9.d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(f7.d,ff7.d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -375,7 +375,7 @@ __ieee754_atan2(double y,double x) {
|
|
|
8ae002 |
EADD(t1,du,v,vv)
|
|
|
8ae002 |
s1=v*(hij[i][11].d+v*(hij[i][12].d+v*(hij[i][13].d+
|
|
|
8ae002 |
v*(hij[i][14].d+v* hij[i][15].d))));
|
|
|
8ae002 |
- ADD2(hij[i][9].d,hij[i][10].d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(hij[i][9].d,hij[i][10].d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][7].d,hij[i][8].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/e_log.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/e_log.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/e_log.c
|
|
|
8ae002 |
@@ -78,9 +78,9 @@ __ieee754_log(double x) {
|
|
|
8ae002 |
n=0;
|
|
|
8ae002 |
if (__builtin_expect(ux < 0x00100000, 0)) {
|
|
|
8ae002 |
if (__builtin_expect(((ux & 0x7fffffff) | dx) == 0, 0))
|
|
|
8ae002 |
- return MHALF/ZERO; /* return -INF */
|
|
|
8ae002 |
+ return MHALF/0; /* return -INF */
|
|
|
8ae002 |
if (__builtin_expect(ux < 0, 0))
|
|
|
8ae002 |
- return (x-x)/ZERO; /* return NaN */
|
|
|
8ae002 |
+ return (x-x)/0; /* return NaN */
|
|
|
8ae002 |
n -= 54; x *= two54.d; /* scale x */
|
|
|
8ae002 |
num.d = x;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
@@ -89,7 +89,7 @@ __ieee754_log(double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Regular values of x */
|
|
|
8ae002 |
|
|
|
8ae002 |
- w = x-ONE;
|
|
|
8ae002 |
+ w = x-1;
|
|
|
8ae002 |
if (__builtin_expect(ABS(w) > U03, 1)) { goto case_03; }
|
|
|
8ae002 |
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -113,25 +113,25 @@ __ieee754_log(double x) {
|
|
|
8ae002 |
w*(d17.d+w*(d18.d+w*(d19.d+w*d20.d))))))));
|
|
|
8ae002 |
EMULV(w,a,s2,ss2,t1,t2,t3,t4,t5)
|
|
|
8ae002 |
ADD2(d10.d,dd10.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d9.d,dd9.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d8.d,dd8.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d7.d,dd7.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d6.d,dd6.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d5.d,dd5.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d4.d,dd4.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d3.d,dd3.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(d2.d,dd2.d,s2,ss2,s3,ss3,t1,t2)
|
|
|
8ae002 |
- MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
- MUL2(w,ZERO,s2,ss2,s3,ss3,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
- ADD2(w,ZERO, s3,ss3, b, bb,t1,t2)
|
|
|
8ae002 |
+ MUL2(w,0,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(w,0,s2,ss2,s3,ss3,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ ADD2(w,0, s3,ss3, b, bb,t1,t2)
|
|
|
8ae002 |
|
|
|
8ae002 |
/* End stage II, case abs(x-1) < 0.03 */
|
|
|
8ae002 |
if ((y=b+(bb+b*E4)) == b+(bb-b*E4)) return y;
|
|
|
8ae002 |
@@ -155,7 +155,7 @@ __ieee754_log(double x) {
|
|
|
8ae002 |
j = (num.i[HIGH_HALF] & 0x000fffff) >> 4;
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Compute w=(u-ui*vj)/(ui*vj) */
|
|
|
8ae002 |
- p0=(ONE+(i-75)*DEL_U)*(ONE+(j-180)*DEL_V);
|
|
|
8ae002 |
+ p0=(1+(i-75)*DEL_U)*(1+(j-180)*DEL_V);
|
|
|
8ae002 |
q=u-p0; r0=Iu[i].d*Iv[j].d; w=q*r0;
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Evaluate polynomial I */
|
|
|
8ae002 |
@@ -178,11 +178,11 @@ __ieee754_log(double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Improve the accuracy of r0 */
|
|
|
8ae002 |
EMULV(p0,r0,sa,sb,t1,t2,t3,t4,t5)
|
|
|
8ae002 |
- t=r0*((ONE-sa)-sb);
|
|
|
8ae002 |
+ t=r0*((1-sa)-sb);
|
|
|
8ae002 |
EADD(r0,t,ra,rb)
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Compute w */
|
|
|
8ae002 |
- MUL2(q,ZERO,ra,rb,w,ww,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(q,0,ra,rb,w,ww,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
|
|
|
8ae002 |
EADD(A,B0,a0,aa0)
|
|
|
8ae002 |
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/s_atan.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/s_atan.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/s_atan.c
|
|
|
8ae002 |
@@ -82,7 +82,7 @@ double atan(double x) {
|
|
|
8ae002 |
return x+x;
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Regular values of x, including denormals +-0 and +-INF */
|
|
|
8ae002 |
- u = (x
|
|
|
8ae002 |
+ u = (x<0) ? -x : x;
|
|
|
8ae002 |
if (u
|
|
|
8ae002 |
if (u
|
|
|
8ae002 |
if (u
|
|
|
8ae002 |
@@ -93,7 +93,7 @@ double atan(double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
EMULV(x,x,v,vv,t1,t2,t3,t4,t5) /* v+vv=x^2 */
|
|
|
8ae002 |
s1=v*(f11.d+v*(f13.d+v*(f15.d+v*(f17.d+v*f19.d))));
|
|
|
8ae002 |
- ADD2(f9.d,ff9.d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(f9.d,ff9.d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(f7.d,ff7.d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -101,8 +101,8 @@ double atan(double x) {
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(f3.d,ff3.d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
- MUL2(x,ZERO,s1,ss1,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
- ADD2(x,ZERO,s2,ss2,s1,ss1,t1,t2)
|
|
|
8ae002 |
+ MUL2(x,0,s1,ss1,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ ADD2(x,0,s2,ss2,s1,ss1,t1,t2)
|
|
|
8ae002 |
if ((y=s1+(ss1-U5*s1)) == s1+(ss1+U5*s1)) return y;
|
|
|
8ae002 |
|
|
|
8ae002 |
return atanMp(x,pr);
|
|
|
8ae002 |
@@ -124,14 +124,14 @@ double atan(double x) {
|
|
|
8ae002 |
z=u-hij[i][0].d;
|
|
|
8ae002 |
s1=z*(hij[i][11].d+z*(hij[i][12].d+z*(hij[i][13].d+
|
|
|
8ae002 |
z*(hij[i][14].d+z* hij[i][15].d))));
|
|
|
8ae002 |
- ADD2(hij[i][9].d,hij[i][10].d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
- MUL2(z,ZERO,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ ADD2(hij[i][9].d,hij[i][10].d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ MUL2(z,0,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][7].d,hij[i][8].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
- MUL2(z,ZERO,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(z,0,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][5].d,hij[i][6].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
- MUL2(z,ZERO,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(z,0,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][3].d,hij[i][4].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
- MUL2(z,ZERO,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
+ MUL2(z,0,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][1].d,hij[i][2].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
if ((y=s2+(ss2-U6*s2)) == s2+(ss2+U6*s2)) return __signArctan(x,y);
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -140,9 +140,9 @@ double atan(double x) {
|
|
|
8ae002 |
}
|
|
|
8ae002 |
else {
|
|
|
8ae002 |
if (u
|
|
|
8ae002 |
- w=ONE/u;
|
|
|
8ae002 |
+ w=1/u;
|
|
|
8ae002 |
EMULV(w,u,t1,t2,t3,t4,t5,t6,t7)
|
|
|
8ae002 |
- ww=w*((ONE-t1)-t2);
|
|
|
8ae002 |
+ ww=w*((1-t1)-t2);
|
|
|
8ae002 |
i=(TWO52+TWO8*w)-TWO52; i-=16;
|
|
|
8ae002 |
z=(w-cij[i][0].d)+ww;
|
|
|
8ae002 |
yy=HPI1-z*(cij[i][2].d+z*(cij[i][3].d+z*(cij[i][4].d+
|
|
|
8ae002 |
@@ -152,12 +152,12 @@ double atan(double x) {
|
|
|
8ae002 |
else u3=U32; /* w >= 1/2 */
|
|
|
8ae002 |
if ((y=t1+(yy-u3)) == t1+(yy+u3)) return __signArctan(x,y);
|
|
|
8ae002 |
|
|
|
8ae002 |
- DIV2(ONE,ZERO,u,ZERO,w,ww,t1,t2,t3,t4,t5,t6,t7,t8,t9,t10)
|
|
|
8ae002 |
+ DIV2(1,0,u,0,w,ww,t1,t2,t3,t4,t5,t6,t7,t8,t9,t10)
|
|
|
8ae002 |
t1=w-hij[i][0].d;
|
|
|
8ae002 |
EADD(t1,ww,z,zz)
|
|
|
8ae002 |
s1=z*(hij[i][11].d+z*(hij[i][12].d+z*(hij[i][13].d+
|
|
|
8ae002 |
z*(hij[i][14].d+z* hij[i][15].d))));
|
|
|
8ae002 |
- ADD2(hij[i][9].d,hij[i][10].d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(hij[i][9].d,hij[i][10].d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(z,zz,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(hij[i][7].d,hij[i][8].d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(z,zz,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
@@ -173,18 +173,18 @@ double atan(double x) {
|
|
|
8ae002 |
}
|
|
|
8ae002 |
else {
|
|
|
8ae002 |
if (u
|
|
|
8ae002 |
- w=ONE/u; v=w*w;
|
|
|
8ae002 |
+ w=1/u; v=w*w;
|
|
|
8ae002 |
EMULV(w,u,t1,t2,t3,t4,t5,t6,t7)
|
|
|
8ae002 |
yy=w*v*(d3.d+v*(d5.d+v*(d7.d+v*(d9.d+v*(d11.d+v*d13.d)))));
|
|
|
8ae002 |
- ww=w*((ONE-t1)-t2);
|
|
|
8ae002 |
+ ww=w*((1-t1)-t2);
|
|
|
8ae002 |
ESUB(HPI,w,t3,cor)
|
|
|
8ae002 |
yy=((HPI1+cor)-ww)-yy;
|
|
|
8ae002 |
if ((y=t3+(yy-U4)) == t3+(yy+U4)) return __signArctan(x,y);
|
|
|
8ae002 |
|
|
|
8ae002 |
- DIV2(ONE,ZERO,u,ZERO,w,ww,t1,t2,t3,t4,t5,t6,t7,t8,t9,t10)
|
|
|
8ae002 |
+ DIV2(1,0,u,0,w,ww,t1,t2,t3,t4,t5,t6,t7,t8,t9,t10)
|
|
|
8ae002 |
MUL2(w,ww,w,ww,v,vv,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
s1=v*(f11.d+v*(f13.d+v*(f15.d+v*(f17.d+v*f19.d))));
|
|
|
8ae002 |
- ADD2(f9.d,ff9.d,s1,ZERO,s2,ss2,t1,t2)
|
|
|
8ae002 |
+ ADD2(f9.d,ff9.d,s1,0,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
ADD2(f7.d,ff7.d,s1,ss1,s2,ss2,t1,t2)
|
|
|
8ae002 |
MUL2(v,vv,s2,ss2,s1,ss1,t1,t2,t3,t4,t5,t6,t7,t8)
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/s_tan.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/s_tan.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/s_tan.c
|
|
|
8ae002 |
@@ -84,7 +84,7 @@ tan(double x) {
|
|
|
8ae002 |
goto ret;
|
|
|
8ae002 |
}
|
|
|
8ae002 |
|
|
|
8ae002 |
- w=(x
|
|
|
8ae002 |
+ w=(x<0) ? -x : x;
|
|
|
8ae002 |
|
|
|
8ae002 |
/* (I) The case abs(x) <= 1.259e-8 */
|
|
|
8ae002 |
if (w<=g1.d) { retval = x; goto ret; }
|
|
|
8ae002 |
@@ -125,11 +125,11 @@ tan(double x) {
|
|
|
8ae002 |
|
|
|
8ae002 |
/* First stage */
|
|
|
8ae002 |
i = ((int) (mfftnhf.d+TWO8*w));
|
|
|
8ae002 |
- z = w-xfg[i][0].d; z2 = z*z; s = (x
|
|
|
8ae002 |
+ z = w-xfg[i][0].d; z2 = z*z; s = (x<0) ? -1 : 1;
|
|
|
8ae002 |
pz = z+z*z2*(e0.d+z2*e1.d);
|
|
|
8ae002 |
fi = xfg[i][1].d; gi = xfg[i][2].d; t2 = pz*(gi+fi)/(gi-pz);
|
|
|
8ae002 |
if ((y=fi+(t2-fi*u3.d))==fi+(t2+fi*u3.d)) { retval = (s*y); goto ret; }
|
|
|
8ae002 |
- t3 = (t2
|
|
|
8ae002 |
+ t3 = (t2<0) ? -t2 : t2;
|
|
|
8ae002 |
t4 = fi*ua3.d+t3*ub3.d;
|
|
|
8ae002 |
if ((y=fi+(t2-t4))==fi+(t2+t4)) { retval = (s*y); goto ret; }
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -165,8 +165,8 @@ tan(double x) {
|
|
|
8ae002 |
da = xn*mp3.d;
|
|
|
8ae002 |
a=t1-da;
|
|
|
8ae002 |
da = (t1-a)-da;
|
|
|
8ae002 |
- if (a
|
|
|
8ae002 |
- else {ya= a; yya= da; sy= ONE;}
|
|
|
8ae002 |
+ if (a<0) {ya=-a; yya=-da; sy=-1;}
|
|
|
8ae002 |
+ else {ya= a; yya= da; sy= 1;}
|
|
|
8ae002 |
|
|
|
8ae002 |
/* (IV),(V) The case 0.787 < abs(x) <= 25, abs(y) <= 1e-7 */
|
|
|
8ae002 |
if (ya<=gy1.d) { retval = tanMp(x); goto ret; }
|
|
|
8ae002 |
@@ -240,14 +240,14 @@ tan(double x) {
|
|
|
8ae002 |
/* -cot */
|
|
|
8ae002 |
t2 = pz*(fi+gi)/(fi+pz);
|
|
|
8ae002 |
if ((y=gi-(t2-gi*u10.d))==gi-(t2+gi*u10.d)) { retval = (-sy*y); goto ret; }
|
|
|
8ae002 |
- t3 = (t2
|
|
|
8ae002 |
+ t3 = (t2<0) ? -t2 : t2;
|
|
|
8ae002 |
t4 = gi*ua10.d+t3*ub10.d;
|
|
|
8ae002 |
if ((y=gi-(t2-t4))==gi-(t2+t4)) { retval = (-sy*y); goto ret; } }
|
|
|
8ae002 |
else {
|
|
|
8ae002 |
/* tan */
|
|
|
8ae002 |
t2 = pz*(gi+fi)/(gi-pz);
|
|
|
8ae002 |
if ((y=fi+(t2-fi*u9.d))==fi+(t2+fi*u9.d)) { retval = (sy*y); goto ret; }
|
|
|
8ae002 |
- t3 = (t2
|
|
|
8ae002 |
+ t3 = (t2<0) ? -t2 : t2;
|
|
|
8ae002 |
t4 = fi*ua9.d+t3*ub9.d;
|
|
|
8ae002 |
if ((y=fi+(t2-t4))==fi+(t2+t4)) { retval = (sy*y); goto ret; } }
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -295,8 +295,8 @@ tan(double x) {
|
|
|
8ae002 |
a = t - t1;
|
|
|
8ae002 |
da = ((t-a)-t1)+da;
|
|
|
8ae002 |
EADD(a,da,t1,t2) a=t1; da=t2;
|
|
|
8ae002 |
- if (a
|
|
|
8ae002 |
- else {ya= a; yya= da; sy= ONE;}
|
|
|
8ae002 |
+ if (a<0) {ya=-a; yya=-da; sy=-1;}
|
|
|
8ae002 |
+ else {ya= a; yya= da; sy= 1;}
|
|
|
8ae002 |
|
|
|
8ae002 |
/* (+++) The case 25 < abs(x) <= 1e8, abs(y) <= 1e-7 */
|
|
|
8ae002 |
if (ya<=gy1.d) { retval = tanMp(x); goto ret; }
|
|
|
8ae002 |
@@ -355,14 +355,14 @@ tan(double x) {
|
|
|
8ae002 |
/* -cot */
|
|
|
8ae002 |
t2 = pz*(fi+gi)/(fi+pz);
|
|
|
8ae002 |
if ((y=gi-(t2-gi*u18.d))==gi-(t2+gi*u18.d)) { retval = (-sy*y); goto ret; }
|
|
|
8ae002 |
- t3 = (t2
|
|
|
8ae002 |
+ t3 = (t2<0) ? -t2 : t2;
|
|
|
8ae002 |
t4 = gi*ua18.d+t3*ub18.d;
|
|
|
8ae002 |
if ((y=gi-(t2-t4))==gi-(t2+t4)) { retval = (-sy*y); goto ret; } }
|
|
|
8ae002 |
else {
|
|
|
8ae002 |
/* tan */
|
|
|
8ae002 |
t2 = pz*(gi+fi)/(gi-pz);
|
|
|
8ae002 |
if ((y=fi+(t2-fi*u17.d))==fi+(t2+fi*u17.d)) { retval = (sy*y); goto ret; }
|
|
|
8ae002 |
- t3 = (t2
|
|
|
8ae002 |
+ t3 = (t2<0) ? -t2 : t2;
|
|
|
8ae002 |
t4 = fi*ua17.d+t3*ub17.d;
|
|
|
8ae002 |
if ((y=fi+(t2-t4))==fi+(t2+t4)) { retval = (sy*y); goto ret; } }
|
|
|
8ae002 |
|
|
|
8ae002 |
@@ -398,8 +398,8 @@ tan(double x) {
|
|
|
8ae002 |
/* Range reduction by algorithm iii */
|
|
|
8ae002 |
n = (__branred(x,&a,&da)) & 0x00000001;
|
|
|
8ae002 |
EADD(a,da,t1,t2) a=t1; da=t2;
|
|
|
8ae002 |
- if (a
|
|
|
8ae002 |
- else {ya= a; yya= da; sy= ONE;}
|
|
|
8ae002 |
+ if (a<0) {ya=-a; yya=-da; sy=-1;}
|
|
|
8ae002 |
+ else {ya= a; yya= da; sy= 1;}
|
|
|
8ae002 |
|
|
|
8ae002 |
/* (+++) The case 1e8 < abs(x) < 2**1024, abs(y) <= 1e-7 */
|
|
|
8ae002 |
if (ya<=gy1.d) { retval = tanMp(x); goto ret; }
|
|
|
8ae002 |
@@ -463,14 +463,14 @@ tan(double x) {
|
|
|
8ae002 |
/* -cot */
|
|
|
8ae002 |
t2 = pz*(fi+gi)/(fi+pz);
|
|
|
8ae002 |
if ((y=gi-(t2-gi*u26.d))==gi-(t2+gi*u26.d)) { retval = (-sy*y); goto ret; }
|
|
|
8ae002 |
- t3 = (t2
|
|
|
8ae002 |
+ t3 = (t2<0) ? -t2 : t2;
|
|
|
8ae002 |
t4 = gi*ua26.d+t3*ub26.d;
|
|
|
8ae002 |
if ((y=gi-(t2-t4))==gi-(t2+t4)) { retval = (-sy*y); goto ret; } }
|
|
|
8ae002 |
else {
|
|
|
8ae002 |
/* tan */
|
|
|
8ae002 |
t2 = pz*(gi+fi)/(gi-pz);
|
|
|
8ae002 |
if ((y=fi+(t2-fi*u25.d))==fi+(t2+fi*u25.d)) { retval = (sy*y); goto ret; }
|
|
|
8ae002 |
- t3 = (t2
|
|
|
8ae002 |
+ t3 = (t2<0) ? -t2 : t2;
|
|
|
8ae002 |
t4 = fi*ua25.d+t3*ub25.d;
|
|
|
8ae002 |
if ((y=fi+(t2-t4))==fi+(t2+t4)) { retval = (sy*y); goto ret; } }
|
|
|
8ae002 |
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpatan.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpatan.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpatan.c
|
|
|
8ae002 |
@@ -69,8 +69,8 @@ __mpatan(mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
{if (dx>__atan_xm[m].d) break;}
|
|
|
8ae002 |
}
|
|
|
8ae002 |
mpone.e = mptwo.e = mptwoim1.e = 1;
|
|
|
8ae002 |
- mpone.d[0] = mpone.d[1] = mptwo.d[0] = mptwoim1.d[0] = ONE;
|
|
|
8ae002 |
- mptwo.d[1] = TWO;
|
|
|
8ae002 |
+ mpone.d[0] = mpone.d[1] = mptwo.d[0] = mptwoim1.d[0] = 1;
|
|
|
8ae002 |
+ mptwo.d[1] = 2;
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Reduce x m times */
|
|
|
8ae002 |
__mul(x,x,&mpsm,p);
|
|
|
8ae002 |
@@ -92,7 +92,7 @@ __mpatan(mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
n=__atan_np[p]; mptwoim1.d[1] = __atan_twonm1[p].d;
|
|
|
8ae002 |
__dvd(&mpsm,&mptwoim1,&mpt,p);
|
|
|
8ae002 |
for (i=n-1; i>1; i--) {
|
|
|
8ae002 |
- mptwoim1.d[1] -= TWO;
|
|
|
8ae002 |
+ mptwoim1.d[1] -= 2;
|
|
|
8ae002 |
__dvd(&mpsm,&mptwoim1,&mpt1,p);
|
|
|
8ae002 |
__mul(&mpsm,&mpt,&mpt2,p);
|
|
|
8ae002 |
__sub(&mpt1,&mpt2,&mpt,p);
|
|
|
8ae002 |
Index: glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpsqrt.c
|
|
|
8ae002 |
===================================================================
|
|
|
8ae002 |
--- glibc-2.17-c758a686.orig/sysdeps/ieee754/dbl-64/mpsqrt.c
|
|
|
8ae002 |
+++ glibc-2.17-c758a686/sysdeps/ieee754/dbl-64/mpsqrt.c
|
|
|
8ae002 |
@@ -62,8 +62,8 @@ __mpsqrt(mp_no *x, mp_no *y, int p) {
|
|
|
8ae002 |
mp_no mpxn,mpz,mpu,mpt1,mpt2;
|
|
|
8ae002 |
|
|
|
8ae002 |
/* Prepare multi-precision 1/2 and 3/2 */
|
|
|
8ae002 |
- mphalf.e =0; mphalf.d[0] =ONE; mphalf.d[1] =HALFRAD;
|
|
|
8ae002 |
- mp3halfs.e=1; mp3halfs.d[0]=ONE; mp3halfs.d[1]=ONE; mp3halfs.d[2]=HALFRAD;
|
|
|
8ae002 |
+ mphalf.e =0; mphalf.d[0] =1; mphalf.d[1] =HALFRAD;
|
|
|
8ae002 |
+ mp3halfs.e=1; mp3halfs.d[0]=1; mp3halfs.d[1]=1; mp3halfs.d[2]=HALFRAD;
|
|
|
8ae002 |
|
|
|
8ae002 |
ey=EX/2; __cpy(x,&mpxn,p); mpxn.e -= (ey+ey);
|
|
|
8ae002 |
__mp_dbl(&mpxn,&dx,p); dy=fastiroot(dx); __dbl_mp(dy,&mpu,p);
|