The -mdiv=inv:minlat option no longer rearranges the computations;
It appears the combiner pattern falls foul of the combine_validate_cost
check. The assumed cost of the three constituent instructions is 4 each,
while the cost of the combined instruction is assumed to be 16.
sh_rtx_costs needs to be adjusted to make sure that the combination
happens; preferrably give the constituent insns a realistic cost.
-mdiv=inv:call performs the transformation, but the transformed instruction
is not recognized, causing an ICE.
testcase (add -O2 to options):
int
f (int a, int b)
{
return a/b;
}
--
Summary: sh64-elf -mdiv= options bitrot
Product: gcc
Version: 4.2.0
Status: UNCONFIRMED
Keywords: ice-on-valid-code, missed-optimization
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: amylaar at gcc dot gnu dot org
GCC target triplet: sh64-elf
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28791
