+/* Floating point library in optimized assembly for 8051
+ * Copyright (c) 2004, Paul Stoffregen, paul@pjrc.com
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Library General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+
+
+#define SDCC_FLOAT_LIB
+#include <float.h>
+
+
+#ifdef FLOAT_ASM_MCS51
+
+// float __fsmul (float a, float b) __reentrant
+static void dummy(void) __naked
+{
+ __asm
+ .globl ___fsmul
+___fsmul:
+ // extract the two inputs, placing them into:
+ // sign exponent mantissa
+ // ---- -------- --------
+ // a: sign_a exp_a r4/r3/r2
+ // b: sign_b exp_b r7/r6/r5
+
+ lcall fsgetargs
+
+ // first check if either input is zero
+ cjne r4, #0, 00002$
+00001$:
+ ljmp fs_return_zero
+
+00002$:
+ mov a, r7
+ jz 00001$
+
+ // compute final sign bit
+ jnb sign_b, 00003$
+ cpl sign_a
+00003$:
+
+ // check if either input is infinity
+ mov a, exp_b
+ cjne a, #0xFF, 00004$
+ ljmp fs_return_inf
+00004$:
+ mov a, exp_a
+ cjne a, #0xFF, 00005$
+ ljmp fs_return_inf
+00005$:
+
+ // add the exponents
+ add a, exp_b
+ // if carry then no underflow
+ jc 00006$
+ add a, #130
+ jc 00007$
+ ljmp fs_return_zero
+
+00006$:
+ add a, #131
+ dec a
+ jnc 00007$
+ ljmp fs_return_inf
+
+00007$:
+ mov exp_a, a
+
+ // now we need to multipy r4/r3/r2 * r7/r6/r5
+ // ------------------------------------------
+ // r2 * r5 << 0
+ // r3 * r5 + r2 * r6 << 8
+ // r4 * r5 + r3 * r6 + r2 * r7 << 16
+ // r4 * r6 + r3 * r7 << 24
+ // r4 * r7 << 32
+ //
+ // This adds quite a bit of code, but it is a LOT faster
+ // than three calls to __mululong...
+
+ // output goes into r4/r3/r2/r1/r0/xx
+
+ mov a, r2
+ mov b, r5
+ mul ab // r2 * r5
+ // discard lowest 8 bits
+ mov r0, b
+ // range 0-FE
+
+ mov a, r2
+ mov b, r6
+ mul ab // r2 * r6
+ add a, r0
+ mov r0, a
+ clr a
+ addc a, b
+ mov r1, a
+ // range 0-FEFF
+
+ mov a, r3
+ mov b, r5
+ mul ab // r3 * r5
+ add a, r0
+ // discard lowest 8 bits
+ mov a, r1
+ addc a, b
+ mov r1, a
+ clr a
+ rlc a
+ xch a, r2
+ // range 0-1FD
+
+ mov b, r7
+ mul ab // r2 * r7
+ add a, r1
+ mov r1, a
+ mov a, r2
+ addc a, b
+ mov r2, a
+ // range 0-FFFE
+
+ mov a, r3
+ mov r0, a
+ mov b, r6
+ mul ab // r3 * r6
+ add a, r1
+ mov r1, a
+ mov a, r2
+ addc a, b
+ mov r2, a
+ clr a
+ rlc a
+ mov r3, a
+ // range 0-1FDFF
+
+ mov a, r4
+ mov b, r5
+ mul ab // r4 * r5
+ add a, r1
+ mov r1, a
+ mov a, r2
+ addc a, b
+ mov r2, a
+ clr a
+ addc a, r3
+ mov r3, a
+ // range 0-2FC00
+
+ mov a, r0 // r3
+ mov b, r7
+ mul ab // r3 * r7
+ add a, r2
+ mov r2, a
+ mov a, r3
+ addc a, b
+ mov r3, a
+ clr a
+ rlc a
+ xch a, r4
+ // range 0-100FD00
+
+ mov r5, a
+ mov b, r6
+ mul ab // r4 * r6
+ add a, r2
+ mov r2, a
+ mov a, r3
+ addc a, b
+ mov r3, a
+ clr a
+ addc a, r4
+ mov r4, a
+ // range 0-1FEFE00
+
+ mov a, r5 // r4
+ mov b, r7
+ mul ab // r4 * r7
+ add a, r3
+ mov r3, a
+ mov a, r4
+ addc a, b
+ mov r4, a
+ // range 40000000-FFFFFE00
+
+ jb acc.7, 00010$
+ lcall fs_normalize_a
+
+00010$:
+ ljmp fs_round_and_return
+ __endasm;
+}
+
+#else
+
/*
** libgcc support for software floating point.
** Copyright (C) 1991 by Pipeline Associates, Inc. All rights reserved.
** I would appreciate receiving any updates/patches/changes that anyone
** makes, and am willing to be the repository for said changes (am I
** making a big mistake?).
-
-Warning! Only single-precision is actually implemented. This file
-won't really be much use until double-precision is supported.
-
-However, once that is done, this file might eventually become a
-replacement for libgcc1.c. It might also make possible
-cross-compilation for an IEEE target machine from a non-IEEE
-host such as a VAX.
-
-If you'd like to work on completing this, please talk to rms@gnu.ai.mit.edu.
-
-
**
** Pat Wood
** Pipeline Associates, Inc.
** pipeline!phw@motown.com or
** sun!pipeline!phw or
** uunet!motown!pipeline!phw
-**
-** 05/01/91 -- V1.0 -- first release to gcc mailing lists
-** 05/04/91 -- V1.1 -- added float and double prototypes and return values
-** -- fixed problems with adding and subtracting zero
-** -- fixed rounding in truncdfsf2
-** -- fixed SWAP define and tested on 386
*/
-/*
-** The following are routines that replace the libgcc soft floating point
-** routines that are called automatically when -msoft-float is selected.
-** The support single and double precision IEEE format, with provisions
-** for byte-swapped machines (tested on 386). Some of the double-precision
-** routines work at full precision, but most of the hard ones simply punt
-** and call the single precision routines, producing a loss of accuracy.
-** long long support is not assumed or included.
-** Overall accuracy is close to IEEE (actually 68882) for single-precision
-** arithmetic. I think there may still be a 1 in 1000 chance of a bit
-** being rounded the wrong way during a multiply. I'm not fussy enough to
-** bother with it, but if anyone is, knock yourself out.
-**
-** Efficiency has only been addressed where it was obvious that something
-** would make a big difference. Anyone who wants to do this right for
-** best speed should go in and rewrite in assembler.
-**
-** I have tested this only on a 68030 workstation and 386/ix integrated
-** in with -msoft-float.
-*/
-
-/* the following deal with IEEE single-precision numbers */
-#define EXCESS 126
-#define SIGNBIT ((unsigned long)0x80000000)
-#define HIDDEN (unsigned long)(1 << 23)
-#define SIGN(fp) ((fp >> (8*sizeof(fp)-1)) & 1)
-#define EXP(fp) (((fp) >> 23) & (unsigned int) 0x00FF)
-#define MANT(fp) (((fp) & (unsigned long)0x007FFFFF) | HIDDEN)
-#define PACK(s,e,m) ((s) | ((e) << 23) | (m))
-
-/* the following deal with IEEE double-precision numbers */
-#define EXCESSD 1022
-#define HIDDEND (1 << 20)
-#define EXPD(fp) (((fp.l.upper) >> 20) & 0x7FF)
-#define SIGND(fp) ((fp.l.upper) & SIGNBIT)
-#define MANTD(fp) (((((fp.l.upper) & 0xFFFFF) | HIDDEND) << 10) | \
- (fp.l.lower >> 22))
+/* (c)2000/2001: hacked a little by johan.knol@iduna.nl for sdcc */
union float_long
{
float f;
- long l;
+ unsigned long l;
};
/* multiply two floats */
-float
-__fsmul (float a1, float a2)
-{
+float __fsmul (float a1, float a2) {
volatile union float_long fl1, fl2;
volatile unsigned long result;
volatile int exp;
- short sign;
+ char sign;
fl1.f = a1;
fl2.f = a2;
result += ((fl1.l & (unsigned long) 0xFF) * (fl2.l >> 8)) >> 8;
result += ((fl2.l & (unsigned long) 0xFF) * (fl1.l >> 8)) >> 8;
- if (result & (unsigned long)0x80000000)
+ if (result & SIGNBIT)
{
/* round */
result += 0x80;
result &= ~HIDDEN;
/* pack up and go home */
- fl1.l = PACK (sign ? ((unsigned long) 0x80000000) : 0 , (unsigned long)exp, result);
+ if (exp >= 0x100)
+ fl1.l = (sign ? SIGNBIT : 0) | __INFINITY;
+ else if (exp < 0)
+ fl1.l = 0;
+ else
+ fl1.l = PACK (sign ? SIGNBIT : 0 , exp, result);
return (fl1.f);
}
+
+#endif
projects / fw / sdcc / blobdiff