Commit b3b8e1eb authored by Douglas Leung's avatar Douglas Leung Committed by Ralf Baechle

MIPS: math-emu: <MADDF|MSUBF>.S: Fix accuracy (32-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 64-bit intermediate
calculations.

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

MADDF.S fd,fs,ft:
  fd = 0x22575225
  fs = ft = 0x3727c5ac

Fixes: e24c3bec ("MIPS: math-emu: Add support for the MIPS R6 MADDF FPU instruction")
Fixes: 83d43305 ("MIPS: math-emu: Add support for the MIPS R6 MSUBF FPU instruction")
Signed-off-by: default avatarDouglas Leung <douglas.leung@imgtec.com>
Signed-off-by: default avatarMiodrag Dinic <miodrag.dinic@imgtec.com>
Signed-off-by: default avatarGoran Ferenc <goran.ferenc@imgtec.com>
Signed-off-by: default avatarAleksandar 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/16890/Signed-off-by: default avatarRalf Baechle <ralf@linux-mips.org>
parent ae11c061
...@@ -45,6 +45,10 @@ static inline int ieee754sp_finite(union ieee754sp x) ...@@ -45,6 +45,10 @@ static inline int ieee754sp_finite(union ieee754sp x)
return SPBEXP(x) != SP_EMAX + 1 + SP_EBIAS; return SPBEXP(x) != SP_EMAX + 1 + SP_EBIAS;
} }
/* 64 bit right shift with rounding */
#define XSPSRS64(v, rs) \
(((rs) >= 64) ? ((v) != 0) : ((v) >> (rs)) | ((v) << (64-(rs)) != 0))
/* 3bit extended single precision sticky right shift */ /* 3bit extended single precision sticky right shift */
#define XSPSRS(v, rs) \ #define XSPSRS(v, rs) \
((rs > (SP_FBITS+3))?1:((v) >> (rs)) | ((v) << (32-(rs)) != 0)) ((rs > (SP_FBITS+3))?1:((v) >> (rs)) | ((v) << (32-(rs)) != 0))
......
...@@ -21,14 +21,8 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, ...@@ -21,14 +21,8 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x,
int re; int re;
int rs; int rs;
unsigned rm; unsigned rm;
unsigned short lxm; uint64_t rm64;
unsigned short hxm; uint64_t zm64;
unsigned short lym;
unsigned short hym;
unsigned lrm;
unsigned hrm;
unsigned t;
unsigned at;
int s; int s;
COMPXSP; COMPXSP;
...@@ -170,108 +164,90 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, ...@@ -170,108 +164,90 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x,
if (flags & MADDF_NEGATE_PRODUCT) if (flags & MADDF_NEGATE_PRODUCT)
rs ^= 1; rs ^= 1;
/* shunt to top of word */ /* Multiple 24 bit xm and ym to give 48 bit results */
xm <<= 32 - (SP_FBITS + 1); rm64 = (uint64_t)xm * ym;
ym <<= 32 - (SP_FBITS + 1);
/* /* Shunt to top of word */
* Multiply 32 bits xm, ym to give high 32 bits rm with stickness. rm64 = rm64 << 16;
*/
lxm = xm & 0xffff;
hxm = xm >> 16;
lym = ym & 0xffff;
hym = ym >> 16;
lrm = lxm * lym; /* 16 * 16 => 32 */
hrm = hxm * hym; /* 16 * 16 => 32 */
t = lxm * hym; /* 16 * 16 => 32 */
at = lrm + (t << 16);
hrm += at < lrm;
lrm = at;
hrm = hrm + (t >> 16);
t = hxm * lym; /* 16 * 16 => 32 */
at = lrm + (t << 16);
hrm += at < lrm;
lrm = at;
hrm = hrm + (t >> 16);
rm = hrm | (lrm != 0);
/* /* Put explicit bit at bit 62 if necessary */
* Sticky shift down to normal rounding precision. if ((int64_t) rm64 < 0) {
*/ rm64 = rm64 >> 1;
if ((int) rm < 0) {
rm = (rm >> (32 - (SP_FBITS + 1 + 3))) |
((rm << (SP_FBITS + 1 + 3)) != 0);
re++; re++;
} else {
rm = (rm >> (32 - (SP_FBITS + 1 + 3 + 1))) |
((rm << (SP_FBITS + 1 + 3 + 1)) != 0);
} }
assert(rm & (SP_HIDDEN_BIT << 3));
if (zc == IEEE754_CLASS_ZERO)
return ieee754sp_format(rs, re, rm);
/* And now the addition */ assert(rm64 & (1 << 62));
assert(zm & SP_HIDDEN_BIT);
if (zc == IEEE754_CLASS_ZERO) {
/* /*
* Provide guard,round and stick bit space. * Move explicit bit from bit 62 to bit 26 since the
* ieee754sp_format code expects the mantissa to be
* 27 bits wide (24 + 3 rounding bits).
*/ */
zm <<= 3; rm = XSPSRS64(rm64, (62 - 26));
return ieee754sp_format(rs, re, rm);
}
/* Move explicit bit from bit 23 to bit 62 */
zm64 = (uint64_t)zm << (62 - 23);
assert(zm64 & (1 << 62));
/* Make the exponents the same */
if (ze > re) { if (ze > re) {
/* /*
* Have to shift r fraction right to align. * Have to shift r fraction right to align.
*/ */
s = ze - re; s = ze - re;
rm = XSPSRS(rm, s); rm64 = XSPSRS64(rm64, s);
re += s; re += s;
} else if (re > ze) { } else if (re > ze) {
/* /*
* Have to shift z fraction right to align. * Have to shift z fraction right to align.
*/ */
s = re - ze; s = re - ze;
zm = XSPSRS(zm, s); zm64 = XSPSRS64(zm64, s);
ze += s; ze += s;
} }
assert(ze == re); assert(ze == re);
assert(ze <= SP_EMAX); assert(ze <= SP_EMAX);
/* Do the addition */
if (zs == rs) { if (zs == rs) {
/* /*
* Generate 28 bit result of adding two 27 bit numbers * Generate 64 bit result by adding two 63 bit numbers
* leaving result in zm, zs and ze. * leaving result in zm64, zs and ze.
*/ */
zm = zm + rm; zm64 = zm64 + rm64;
if ((int64_t)zm64 < 0) { /* carry out */
if (zm >> (SP_FBITS + 1 + 3)) { /* carry out */ zm64 = XSPSRS1(zm64);
zm = XSPSRS1(zm);
ze++; ze++;
} }
} else { } else {
if (zm >= rm) { if (zm64 >= rm64) {
zm = zm - rm; zm64 = zm64 - rm64;
} else { } else {
zm = rm - zm; zm64 = rm64 - zm64;
zs = rs; zs = rs;
} }
if (zm == 0) if (zm64 == 0)
return ieee754sp_zero(ieee754_csr.rm == FPU_CSR_RD); return ieee754sp_zero(ieee754_csr.rm == FPU_CSR_RD);
/* /*
* Normalize in extended single precision * Put explicit bit at bit 62 if necessary.
*/ */
while ((zm >> (SP_MBITS + 3)) == 0) { while ((zm64 >> 62) == 0) {
zm <<= 1; zm64 <<= 1;
ze--; ze--;
} }
} }
/*
* Move explicit bit from bit 62 to bit 26 since the
* ieee754sp_format code expects the mantissa to be
* 27 bits wide (24 + 3 rounding bits).
*/
zm = XSPSRS64(zm64, (62 - 26));
return ieee754sp_format(zs, ze, zm); return ieee754sp_format(zs, ze, zm);
} }
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment