2013/8/26 Joern Rennecke <joern.renne...@embecosm.com>: > The gcc.dg/fixed-point/convert-accum-neg.c execution test fails for avr > because for fractional integer to accumulator / integer conversions, > the avr target rounds towards -infinity, whereas we are supposed to round > towards 0. > > The attached patch implements rounding towards 0, and adds an option > -mfract-convert-truncate to revert to the previous behaviour (for > situations where smaller code is more important than proper rounding). > > Tested for atmega128-sim. > > OK to apply? > > 2013-06-25 Joern Rennecke <joern.renne...@embecosm.com> > > * config/avr/avr.opt (mfract-convert-truncate): New option. > * config/avr/avr.c (avr_out_fract): Unless TARGET_FRACT_CONV_TRUNC > is set, round negative fractional integers according to n1169 > when converting to integer types. >
Approved. Denis.