http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53511
--- Comment #7 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-06-05 20:08:08 UTC --- > (In reply to comment #4) > > Make -mfused-madd no-op instead to remove for the backward > compatibility. Sorry, I don't quite follow. According to my understanding we have something like the following... (A): New behavior (attachment 27558, using common gcc/config/fused-madd.opt) (B): Old behavior (C): New behavior, with -mfused-madd being a nop | | (A) | (B) | (C) 1st option | 2nd option | fma pat. | fmac pat.| fma pat. (new) | | (new) | (old) | (mfused-madd = nop) =================+==================+==========+==========+================== <no option> | -ffp-contract= | 1 | 0 | 1 | <no opt = fast> | | | -----------------+------------------+----------+----------+------------------ <no option> | -fpp-contract=off| 0 | 0 | 0 -----------------+------------------+----------+----------+------------------ -mfused-madd | -ffp-contract= | 1 | 1 | 1 | <no opt = fast> | | | -----------------+------------------+----------+----------+------------------ -mno-fused-madd | -ffp-contract= | 0 | 0 | 0 | <no opt = fast> | | | -----------------+------------------+----------+----------+------------------ -mfused-madd | -fpp-contract=off| 0 | 1 | 0 -----------------+------------------+----------+----------+------------------ -mno-fused-madd | -fpp-contract=off| 0 | 0 | 0 -----------------+------------------+----------+----------+------------------ -fpp-contract=off| -mfused-madd | 1 | 1 | 0 (*) -----------------+------------------+----------+----------+------------------ -fpp-contract=off| -mno-fused-madd | 0 | 0 | 0 -----------------+------------------+----------+----------+------------------ My intention was to avoid having any special -mfused-madd handling in the SH target code and to re-use the existing common gcc/config/fused-madd.opt. Making -mfused-madd a no-op with special handling will only differ in the (*) marked case. Is this what you had in mind?