On Wed, Feb 05, 2020 at 11:46:51AM +0100, Uros Bizjak wrote:
> I think we should just enable split4 also for -O0. This would also
> allow us to remove the "optimize > 0" check above and allow us to
> generate a bit more optimal code even with -O0 for
> TARGET_SSE_PARTIAL_REG_DEPENDENCY and TARGET_AVOID_FALSE_DEP_FOR_BMI.
The -O0 passes are:
test.c.284r.ira
test.c.285r.reload
# placement of split2
test.c.292r.pro_and_epilogue
test.c.295r.jump2
# placement of split4
# placement of split3
test.c.307r.stack
test.c.308r.alignments
test.c.310r.mach
test.c.311r.barriers
# placement of split5
test.c.316r.shorten
test.c.317r.nothrow
test.c.318r.dwarf2
test.c.319r.final
It really doesn't matter if for -O0 STACK_REGS we enable split4 or
split3, it is the same location, just split3 is better named (split before
reg-stack, which is run even at -O0, rather than split before sched2, which
is not run at -O0).
I think the primary question is, do we for -O0 need to enable split2 (as in,
will anything in pro_and_epilogue or jump2 passes benefit from the
splitting and worth running two -O0 post-RA split passes), then we could go
with the first patch I've posted, or nothing benefits from it in the two
passes and it is ok to split only before stack, in that case we can go with
the third patch.
Jakub
