https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106602
--- Comment #6 from Vineet Gupta <vineetg at rivosinc dot com> --- (In reply to Andrew Pinski from comment #4) > (In reply to Vineet Gupta from comment #3) > > Interestingly, if one builds for -march=rv64gc_zbs # single bit extension > > > > then the optimal code seq for bitmanip is generated, while no zbs > > instructions are used. > > > > foo2: > > slli a5,a0,32 > > srli a0,a5,26 > > ret > > Most likely because > li a5,1 > slli a5,a5,38 > Could be done using one instruction. > > And then combine could do its thing. But with multiple instructions, it > becomes harder to do. > I have not looked into rtl dumps but this is just knowing what combine can > do and such. Looks like our updates collided, indeed that 1 << 38 is the issue here.