Am 15.06.20 um 11:04 schrieb Maarten Brock: > > This sounds like an excellent job for the peephole optimizer.
An alternative that would be much more work: Allow __sfr operands for & (could be implemented in ralloc.c, as packing). Then with ifxForOpthe jump could be done at the &. So instead of 3 iCodes, from the perspective of the register allocator and code generator, it would be only one. The t1sn would be generated at the & opcode. The advantage of this approach would be that register a would be free to be used for other purposes. This is probably more of a long-term perspective, while the peephole approach is something that could be implemented quickly. Philipp _______________________________________________ Sdcc-user mailing list Sdcc-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/sdcc-user
