linux/arch/mips/math-emu/dp_maddf.c
Douglas Leung 2cfa58259f MIPS: math-emu: <MADDF|MSUBF>.D: Fix accuracy (64-bit case)
Implement fused multiply-add with correct accuracy.

Fused multiply-add operation has better accuracy than respective
sequential execution of multiply and add operations applied on the
same inputs. This is because accuracy errors accumulate in latter
case.

This patch implements fused multiply-add with the same accuracy
as it is implemented in hardware, using 128-bit intermediate
calculations.

One test case example (raw bits) that this patch fixes:

MADDF.D fd,fs,ft:
  fd = 0x00000ca000000000
  fs = ft = 0x3f40624dd2f1a9fc

Fixes: e24c3bec3e ("MIPS: math-emu: Add support for the MIPS R6 MADDF FPU instruction")
Fixes: 83d43305a1 ("MIPS: math-emu: Add support for the MIPS R6 MSUBF FPU instruction")

Signed-off-by: Douglas Leung <douglas.leung@imgtec.com>
Signed-off-by: Miodrag Dinic <miodrag.dinic@imgtec.com>
Signed-off-by: Goran Ferenc <goran.ferenc@imgtec.com>
Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Cc: Douglas Leung <douglas.leung@imgtec.com>
Cc: Bo Hu <bohu@google.com>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: Jin Qian <jinqian@google.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Petar Jovanovic <petar.jovanovic@imgtec.com>
Cc: Raghu Gandham <raghu.gandham@imgtec.com>
Cc: <stable@vger.kernel.org> # 4.7+
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/16891/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
2017-08-29 15:21:56 +02:00

348 lines
7.8 KiB
C

/*
* IEEE754 floating point arithmetic
* double precision: MADDF.f (Fused Multiply Add)
* MADDF.fmt: FPR[fd] = FPR[fd] + (FPR[fs] x FPR[ft])
*
* MIPS floating point support
* Copyright (C) 2015 Imagination Technologies, Ltd.
* Author: Markos Chandras <markos.chandras@imgtec.com>
*
* This program is free software; you can distribute it and/or modify it
* under the terms of the GNU General Public License as published by the
* Free Software Foundation; version 2 of the License.
*/
#include "ieee754dp.h"
/* 128 bits shift right logical with rounding. */
void srl128(u64 *hptr, u64 *lptr, int count)
{
u64 low;
if (count >= 128) {
*lptr = *hptr != 0 || *lptr != 0;
*hptr = 0;
} else if (count >= 64) {
if (count == 64) {
*lptr = *hptr | (*lptr != 0);
} else {
low = *lptr;
*lptr = *hptr >> (count - 64);
*lptr |= (*hptr << (128 - count)) != 0 || low != 0;
}
*hptr = 0;
} else {
low = *lptr;
*lptr = low >> count | *hptr << (64 - count);
*lptr |= (low << (64 - count)) != 0;
*hptr = *hptr >> count;
}
}
static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x,
union ieee754dp y, enum maddf_flags flags)
{
int re;
int rs;
unsigned lxm;
unsigned hxm;
unsigned lym;
unsigned hym;
u64 lrm;
u64 hrm;
u64 lzm;
u64 hzm;
u64 t;
u64 at;
int s;
COMPXDP;
COMPYDP;
COMPZDP;
EXPLODEXDP;
EXPLODEYDP;
EXPLODEZDP;
FLUSHXDP;
FLUSHYDP;
FLUSHZDP;
ieee754_clearcx();
/*
* Handle the cases when at least one of x, y or z is a NaN.
* Order of precedence is sNaN, qNaN and z, x, y.
*/
if (zc == IEEE754_CLASS_SNAN)
return ieee754dp_nanxcpt(z);
if (xc == IEEE754_CLASS_SNAN)
return ieee754dp_nanxcpt(x);
if (yc == IEEE754_CLASS_SNAN)
return ieee754dp_nanxcpt(y);
if (zc == IEEE754_CLASS_QNAN)
return z;
if (xc == IEEE754_CLASS_QNAN)
return x;
if (yc == IEEE754_CLASS_QNAN)
return y;
if (zc == IEEE754_CLASS_DNORM)
DPDNORMZ;
/* ZERO z cases are handled separately below */
switch (CLPAIR(xc, yc)) {
/*
* Infinity handling
*/
case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
ieee754_setcx(IEEE754_INVALID_OPERATION);
return ieee754dp_indef();
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
if ((zc == IEEE754_CLASS_INF) &&
((!(flags & MADDF_NEGATE_PRODUCT) && (zs != (xs ^ ys))) ||
((flags & MADDF_NEGATE_PRODUCT) && (zs == (xs ^ ys))))) {
/*
* Cases of addition of infinities with opposite signs
* or subtraction of infinities with same signs.
*/
ieee754_setcx(IEEE754_INVALID_OPERATION);
return ieee754dp_indef();
}
/*
* z is here either not an infinity, or an infinity having the
* same sign as product (x*y) (in case of MADDF.D instruction)
* or product -(x*y) (in MSUBF.D case). The result must be an
* infinity, and its sign is determined only by the value of
* (flags & MADDF_NEGATE_PRODUCT) and the signs of x and y.
*/
if (flags & MADDF_NEGATE_PRODUCT)
return ieee754dp_inf(1 ^ (xs ^ ys));
else
return ieee754dp_inf(xs ^ ys);
case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
if (zc == IEEE754_CLASS_INF)
return ieee754dp_inf(zs);
if (zc == IEEE754_CLASS_ZERO) {
/* Handle cases +0 + (-0) and similar ones. */
if ((!(flags & MADDF_NEGATE_PRODUCT)
&& (zs == (xs ^ ys))) ||
((flags & MADDF_NEGATE_PRODUCT)
&& (zs != (xs ^ ys))))
/*
* Cases of addition of zeros of equal signs
* or subtraction of zeroes of opposite signs.
* The sign of the resulting zero is in any
* such case determined only by the sign of z.
*/
return z;
return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);
}
/* x*y is here 0, and z is not 0, so just return z */
return z;
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
DPDNORMX;
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
if (zc == IEEE754_CLASS_INF)
return ieee754dp_inf(zs);
DPDNORMY;
break;
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
if (zc == IEEE754_CLASS_INF)
return ieee754dp_inf(zs);
DPDNORMX;
break;
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
if (zc == IEEE754_CLASS_INF)
return ieee754dp_inf(zs);
/* fall through to real computations */
}
/* Finally get to do some computation */
/*
* Do the multiplication bit first
*
* rm = xm * ym, re = xe + ye basically
*
* At this point xm and ym should have been normalized.
*/
assert(xm & DP_HIDDEN_BIT);
assert(ym & DP_HIDDEN_BIT);
re = xe + ye;
rs = xs ^ ys;
if (flags & MADDF_NEGATE_PRODUCT)
rs ^= 1;
/* shunt to top of word */
xm <<= 64 - (DP_FBITS + 1);
ym <<= 64 - (DP_FBITS + 1);
/*
* Multiply 64 bits xm and ym to give 128 bits result in hrm:lrm.
*/
/* 32 * 32 => 64 */
#define DPXMULT(x, y) ((u64)(x) * (u64)y)
lxm = xm;
hxm = xm >> 32;
lym = ym;
hym = ym >> 32;
lrm = DPXMULT(lxm, lym);
hrm = DPXMULT(hxm, hym);
t = DPXMULT(lxm, hym);
at = lrm + (t << 32);
hrm += at < lrm;
lrm = at;
hrm = hrm + (t >> 32);
t = DPXMULT(hxm, lym);
at = lrm + (t << 32);
hrm += at < lrm;
lrm = at;
hrm = hrm + (t >> 32);
/* Put explicit bit at bit 126 if necessary */
if ((int64_t)hrm < 0) {
lrm = (hrm << 63) | (lrm >> 1);
hrm = hrm >> 1;
re++;
}
assert(hrm & (1 << 62));
if (zc == IEEE754_CLASS_ZERO) {
/*
* Move explicit bit from bit 126 to bit 55 since the
* ieee754dp_format code expects the mantissa to be
* 56 bits wide (53 + 3 rounding bits).
*/
srl128(&hrm, &lrm, (126 - 55));
return ieee754dp_format(rs, re, lrm);
}
/* Move explicit bit from bit 52 to bit 126 */
lzm = 0;
hzm = zm << 10;
assert(hzm & (1 << 62));
/* Make the exponents the same */
if (ze > re) {
/*
* Have to shift y fraction right to align.
*/
s = ze - re;
srl128(&hrm, &lrm, s);
re += s;
} else if (re > ze) {
/*
* Have to shift x fraction right to align.
*/
s = re - ze;
srl128(&hzm, &lzm, s);
ze += s;
}
assert(ze == re);
assert(ze <= DP_EMAX);
/* Do the addition */
if (zs == rs) {
/*
* Generate 128 bit result by adding two 127 bit numbers
* leaving result in hzm:lzm, zs and ze.
*/
hzm = hzm + hrm + (lzm > (lzm + lrm));
lzm = lzm + lrm;
if ((int64_t)hzm < 0) { /* carry out */
srl128(&hzm, &lzm, 1);
ze++;
}
} else {
if (hzm > hrm || (hzm == hrm && lzm >= lrm)) {
hzm = hzm - hrm - (lzm < lrm);
lzm = lzm - lrm;
} else {
hzm = hrm - hzm - (lrm < lzm);
lzm = lrm - lzm;
zs = rs;
}
if (lzm == 0 && hzm == 0)
return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);
/*
* Put explicit bit at bit 126 if necessary.
*/
if (hzm == 0) {
/* left shift by 63 or 64 bits */
if ((int64_t)lzm < 0) {
/* MSB of lzm is the explicit bit */
hzm = lzm >> 1;
lzm = lzm << 63;
ze -= 63;
} else {
hzm = lzm;
lzm = 0;
ze -= 64;
}
}
t = 0;
while ((hzm >> (62 - t)) == 0)
t++;
assert(t <= 62);
if (t) {
hzm = hzm << t | lzm >> (64 - t);
lzm = lzm << t;
ze -= t;
}
}
/*
* Move explicit bit from bit 126 to bit 55 since the
* ieee754dp_format code expects the mantissa to be
* 56 bits wide (53 + 3 rounding bits).
*/
srl128(&hzm, &lzm, (126 - 55));
return ieee754dp_format(zs, ze, lzm);
}
union ieee754dp ieee754dp_maddf(union ieee754dp z, union ieee754dp x,
union ieee754dp y)
{
return _dp_maddf(z, x, y, 0);
}
union ieee754dp ieee754dp_msubf(union ieee754dp z, union ieee754dp x,
union ieee754dp y)
{
return _dp_maddf(z, x, y, MADDF_NEGATE_PRODUCT);
}