diff options
author | Peter Maydell <peter.maydell@linaro.org> | 2014-02-20 10:35:50 +0000 |
---|---|---|
committer | Peter Maydell <peter.maydell@linaro.org> | 2014-02-20 10:35:50 +0000 |
commit | 67d43538aee10b6cfe8f3606c69187a3e142a2ba (patch) | |
tree | 4eb0637464a3011cc1ecc91b4e4ae9e6867d8445 /fpu/softfloat.c | |
parent | bc242f9bb6324a50e7572c0997904b66b630f73a (diff) |
softfloat: Support halving the result of muladd operation
The ARMv8 instruction set includes a fused floating point
reciprocal square root step instruction which demands an
"(x * y + z) / 2" fused operation. Support this by adding
a flag to the softfloat muladd operations which requests
that the result is halved before rounding.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Diffstat (limited to 'fpu/softfloat.c')
-rw-r--r-- | fpu/softfloat.c | 38 |
1 files changed, 38 insertions, 0 deletions
diff --git a/fpu/softfloat.c b/fpu/softfloat.c index e0ea599769..fc0b179df4 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -2372,6 +2372,17 @@ float32 float32_muladd(float32 a, float32 b, float32 c, int flags STATUS_PARAM) } } /* Zero plus something non-zero : just return the something */ + if (flags & float_muladd_halve_result) { + if (cExp == 0) { + normalizeFloat32Subnormal(cSig, &cExp, &cSig); + } + /* Subtract one to halve, and one again because roundAndPackFloat32 + * wants one less than the true exponent. + */ + cExp -= 2; + cSig = (cSig | 0x00800000) << 7; + return roundAndPackFloat32(cSign ^ signflip, cExp, cSig STATUS_VAR); + } return packFloat32(cSign ^ signflip, cExp, cSig); } @@ -2408,6 +2419,9 @@ float32 float32_muladd(float32 a, float32 b, float32 c, int flags STATUS_PARAM) /* Throw out the special case of c being an exact zero now */ shift64RightJamming(pSig64, 32, &pSig64); pSig = pSig64; + if (flags & float_muladd_halve_result) { + pExp--; + } return roundAndPackFloat32(zSign, pExp - 1, pSig STATUS_VAR); } @@ -2472,6 +2486,10 @@ float32 float32_muladd(float32 a, float32 b, float32 c, int flags STATUS_PARAM) zSig64 <<= shiftcount; zExp -= shiftcount; } + if (flags & float_muladd_halve_result) { + zExp--; + } + shift64RightJamming(zSig64, 32, &zSig64); return roundAndPackFloat32(zSign, zExp, zSig64 STATUS_VAR); } @@ -4088,6 +4106,17 @@ float64 float64_muladd(float64 a, float64 b, float64 c, int flags STATUS_PARAM) } } /* Zero plus something non-zero : just return the something */ + if (flags & float_muladd_halve_result) { + if (cExp == 0) { + normalizeFloat64Subnormal(cSig, &cExp, &cSig); + } + /* Subtract one to halve, and one again because roundAndPackFloat64 + * wants one less than the true exponent. + */ + cExp -= 2; + cSig = (cSig | 0x0010000000000000ULL) << 10; + return roundAndPackFloat64(cSign ^ signflip, cExp, cSig STATUS_VAR); + } return packFloat64(cSign ^ signflip, cExp, cSig); } @@ -4123,6 +4152,9 @@ float64 float64_muladd(float64 a, float64 b, float64 c, int flags STATUS_PARAM) if (!cSig) { /* Throw out the special case of c being an exact zero now */ shift128RightJamming(pSig0, pSig1, 64, &pSig0, &pSig1); + if (flags & float_muladd_halve_result) { + pExp--; + } return roundAndPackFloat64(zSign, pExp - 1, pSig1 STATUS_VAR); } @@ -4159,6 +4191,9 @@ float64 float64_muladd(float64 a, float64 b, float64 c, int flags STATUS_PARAM) zExp--; } shift128RightJamming(zSig0, zSig1, 64, &zSig0, &zSig1); + if (flags & float_muladd_halve_result) { + zExp--; + } return roundAndPackFloat64(zSign, zExp, zSig1 STATUS_VAR); } else { /* Subtraction */ @@ -4209,6 +4244,9 @@ float64 float64_muladd(float64 a, float64 b, float64 c, int flags STATUS_PARAM) zExp -= (shiftcount + 64); } } + if (flags & float_muladd_halve_result) { + zExp--; + } return roundAndPackFloat64(zSign, zExp, zSig0 STATUS_VAR); } } |