2011/12/29 Tatsuo Ishii <<a href="mailto:ishii@postgresql.org">ishii@postgresql.org</a>><br />> >
Impressiveresults.<br />> ><br />> > config/c-compiler.m4 doesn't seem like the right place for the<br />
>> configure test. Would there be any harm in setting the lwarx hint<br />> > always; what would happen on
olderppc processors that don't support<br />> > it?<br />><br />> I think the load module just fails to run
inthis case, but I'd like<br /> > to confirm. Ori-san?<br /><br />I don't know where is the right config/*.m4 to
placethis kind of<br />configure test. Do you have any idea?<br /><br />I believe lwarx hint would be no harm for
recentPowerPC processors.<br /> What I tested are:<br /><br /> (1) Built postgres on POWER6 + RHEL5, which got lwarx
hint<br/> included. Then copy these src tree to POWER5 + RHEL4 and<br /> run "make test", finished
successfully.<br/><br /> (2) Lwarx test in configure failed on POWER5 + RHEL4.<br /><br />Note that POWER6 understands
lwarxhint and POWER5 doesn't.<br />RHEL5 binutils supports lwarx hint and RHEL4 binutils doesn't.<br /><br />The only
concernis for very old PowerPC.<br />Referring to Power Instruction Set Architecture manual(*1), on<br /> some
processorsthat precede PowerISA v2.00, executing lwarx with<br />hint will cause the illegal instruction error.<br
/><br/>Lwarx test in configure should fail on these kind of processors,<br />guessing from my test(2).<br /><br /> (*1)
p.689of <a
href="https://www.power.org/resources/downloads/PowerISA_V2.06B_V2_PUBLIC.pdf">https://www.power.org/resources/downloads/PowerISA_V2.06B_V2_PUBLIC.pdf</a><br
/><br/>Regards,<br />Manabu Ori<br /><br />
