I forgot to mention that comparing the three tests of placement of the PCREL_OPT pass:
Having the pass after sched2 generated the same number of PCREL_OPT relocations as having the pass immediately after reload. But having the pass just before sched2 resulted in some more PCREL_OPT relocations. Mostly the number of instructions was the same, but a few benchmarks had a few more 4-byte instructions generated. It looks like those new PCREL_OPT relocations replaced a few places where we loaded the address, but did not do a PCREL_OPT optimization. -- Michael Meissner, IBM IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA email: meiss...@linux.ibm.com, phone: +1 (978) 899-4797
