For example, when converting from uint128 to float32 (whose max exponent
is 2^127), the unit fails to detect overflow for numbers just under 2^128.

The bug only manifests when the ranges are that close.  If the float has
greater range, or has much smaller range, the bug does not manifest.

Existing RV32 and RV64 cores are unaffected by this bug, because no such
conversions exist in the ISA.

* Update testing infrastructure from chisel 2 to 3.

This requires updates to the test harness, as well as to the driver. Use
verilator for tests. Tests run but are currently failing and I'm not
sure why.

* Use strtoull instead of strtoll

* Fix incorrect outputs

* added detailed algorithm used for sqrt

* fixed wording

* added explanation for division

This reverts commit ca89a808.

See this discussion for an explanation: https://github.com/ucb-bar/berkeley-hardfloat/pull/28

SoftFloat/TestFloat Release 3e.  First, the rounding mode code for 'odd'
rounding is changed from 5 to 6.  Second, conversions to integer now round
to the nearest odd integer instead of rounding to minimum magnitude.

* add small div/sqrt unit

* Chisel3 doesn't permit mixed UInt/SInt muxing/assignment

* Fix control bug in DivSqrtRecFN_small

After an uncommon-case operation, outValid would remain high until
another inValid was presented, instead of going low after one cycle.

integers.

'RoundAnyRawFNToRecFN'.

integer values in module 'DivSqrtRecF64_mulAddZ31'.

of 'var'.

If you use a 'val' in an object, it is only run once.  This makes the first
elaboration of a Module different from subsequent elaboration runs, breaking
deduplication.

known as jamming).

rounding mode so that it rounds to minimum magnitude instead, same as
'minMag'.

Newer releases of berkeley-softfloat-3 allow users to set the
SPECIALIZE_TYPE variable without applying a patch.  This commit removes
the patch and sets the variable on the make commandline instead.

Note that there's no dependency checking for berkeley-softfloat-3 when
checked out locally, so you might have to "make distclean" to get the
new version.

These are downloaded by the Makefile so they should really get removed
on "make clean".

nearest, or to maximum magnitude on ties) and with tests for detecting
underflow tininess after rounding.

"near_maxMag" (round to nearest, or to maximum magnitude on ties) and
"odd" (round to value with least-significant bit set), and with support
for choosing tininess detection on underflow (tininess detected either
before or after rounding).

source code.

modes "near_maxMag" (round to nearest, or to maximum magnitude on ties) and
"odd" (round to value with least-significant bit set).

of a result is now done by the shared rounding module, 'RoundRawFNToRecFN'.

conventions.

separate module and adding more rounding control options (not yet truly
supported).

more cases correctly.

the input floating-point value is zero.

Stop comparing Bits

Chisel 3 doesn't support comparison on Bits.

the output is an unsigned integer.

first macrostage of module `DivSqrtRecF64_mulAddZ31'.