Summary: 122 instances, 88 unique

Text	Count
// FIXME: cache	1
// FIXME: I've changed the posit layout to not be symmetric, to have simpler decoding	1
// FIXME: should take sign flag	1
// FIXME: is this big enough?	1
true, /* FIXME: beta */	2
// FIXME: we can either use a sum with carry bit, or a comparator.	1
// FIXME: this will be completely serial!!! We need known bounds	2
// FIXME: is this the right decision?	1
// FIXME: we're assuming that the float's denormal type is representable	1
// FIXME: handle empty tensors	4
// FIXME: not a test	1
// FIXME: this is not quite right, it's max regime field size - 2	1
// FIXME: this was broken before?	1
(char) 0, // FIXME: what here for outputScale?	1
// FIXME: not much of a test	1
// FIXME: remove ES overflow bit	1
(char) 0, // FIXME: what here for inputScale?	1
// FIXME: in either case, maybe we shouldn't do the add? Should this be done	2
// FIXME: implement roundStochastic	1
// FIXME: how to use the sticky bits?	1
// FIXME: case where we are right at the limit, and the +1 from abExpShift	1
// FIXME: need transposed MM	1
true, /* FIXME: beta */	1
// FIXME: not a real test	5
// FIXME: remove dim_, should be implicit based on size_.size()	1
// FIXME: what about one/both args being inf but output is non-nan?	1
// FIXME: handle transposed MM	2
// FIXME: implement batch	1
// FIXME: ignore this, just preserve b.isZero	1
// FIXME: integrate with above	1
// FIXME: doesn't handle over/underflow, rounding	1
// FIXME: not a complete test	1
// FIXME: this is not technically correct still; while the exponent	1
// FIXME: compiler crashes if this is inside a branch below	1
# FIXME: huh?	1
// FIXME: implement on the FPGA	1
// FIXME: (*) not true for max regime. Does this matter?	1
// FIXME: does this make sense?	2
// FIXME: what should this be?	1
// FIXME: do we want to parameterize this separately?	1
// FIXME: assert region sizes	1
// FIXME: for transpose, we must be innermost contiguous, but we can be outer	1
// FIXME: this is based on our current inversion of the standard	2
// FIXME: run_dc flow encounters packages in alphabetic order it seems, so	1
// FIXME c++17 adds non-const data()	1
// FIXME: what about sticky bit for any remainder?	1
// FIXME: handle posit sign / fraction encoding (2s complement)?	3
// FIXME: if total weight <= 0 do nothing	1
// FIXME: really the number of bits added should be based on the max/min value	2
// FIXME: underflow or overflow should really be based on shifting the	1
// FIXME: if the max exponent * 2 can fit in the same number of bits, we	1
// FIXME: make denormal handling optional	2
// FIXME: does not match what I did with posits	1
// FIXME: can be EXP if we filter out infinity as input	1
(char) 0, // FIXME: what here for inputScale?	4
// FIXME: get rid of this, is no longer needed after the refactoring to	1
// FIXME: this isn't quite correct, need to be concerned about exponent	1
// FIXME: we don't convert to denormal values at present	2
// FIXME: signed regime is actually easier to calculate, store that	1
// FIXME: I know what causes this, need to rewrite all float code	1
// FIXME: some +/1 ulp problem with rounding? dunno	1
// FIXME: implement transposed MM	1
// FIXME: make sticky bit calculation optional	2
// FIXME: we aren't checking to see if EXP is big enough	1
// FIXME: this is a copy	1
// FIXME: would it be cheaper to use a comparator on the exponent?	1
// FIXME: unimplemented conversion to posit	1
// FIXME: this is broken for expAdjust	2
// FIXME: only increment the frac if we are not at max frac precision	1
/// FIXME: remove	3
// FIXME: after the log -> linear conversion, we actually have	2
// FIXME: different initialization scheme	1
// FIXME: implement overflow detection	1
# FIXME: attempt to fix d2h copy assert	1
// FIXME: if EXP_IN == M, this isn't performed correctly	1
// FIXME: if FRAC/DENORMAL_OFFSET_SIZE are huge, this isn't right	1
// FIXME: use positDef.testVal	3
// FIXME: we are guaranteed to be unsigned	1
(char) 0, // FIXME: what here for outputScale?	4
// FIXME: this is not much of a test	2
// FIXME: why does the linker complain about this but	2
// FIXME: add flag to skip denormal adjustment; denormals in the new width would	1
// FIXME: load and accumulate earlier	1
// FIXME: all of this work is planewise	1
.EXP_OUT(EXP_OUT), // FIXME: do we want this option?	1
// FIXME: no denormal handling	2
// FIXME: implement transposed MM	1
// FIXME: figure out better way to do this	1