Re: reorg.c (fill_slots_from_thread): Improve for TARGET_FLAGS_REGNUM targets

[Jeff Law <law@redhat.com>]

On Fri, 2020-08-14 at 07:27 +0200, Hans-Peter Nilsson via Gcc-patches wrote:
> Originally I thought to bootstrap this patch on MIPS and SPARC
> since they're both delayed-branch-slot targets but I
> reconsidered, as neither is a TARGET_FLAGS_REGNUM target.  It
> seems only visium and CRIS has this feature set, and I see no
> trace of visium in neither newlib nor the simulator next to
> glibc.  So, I just tested cris-elf.
[ Sorry, been focused on non-upstream stuff for the last month+, so yea,
  some stuff is falling through the cracks. ]

Ibuild newlib on visium daily.  I also run the gcc testsuite, but using dummy
simulator.  So we're really just testing that the code generator doesn't fault
and the various dg-scan style tests.

http://3.14.90.209:8080/job/visium-elf/


> 
> This handles TARGET_FLAGS_REGNUM clobbering insns as delay-slot
> fillers using a method similar to that in commit 33c2207d3fda,
> where care was taken for fill_simple_delay_slots to allow such
> insns when scanning for delay-slot fillers *backwards* (before
> the insn).
> 
> A TARGET_FLAGS_REGNUM target is typically a former cc0 target.
> For cc0 targets, insns don't mention clobbering cc0, so the
> clobbers are mentioned in the "resources" only as a special
> entity and only for compare-insns and branches, where the cc0
> value matters.
> 
> In contrast, with TARGET_FLAGS_REGNUM, most insns clobber it and
> the register liveness detection in reorg.c / resource.c treats
> that as a blocker (for other insns mentioning it, i.e. most)
> when looking for delay-slot-filling candidates.  This means that
> when comparing core and performance for a delay-slot cc0 target
> before and after the de-cc0 conversion, the inability to fill a
> delay slot after conversion manifests as a regression.  This was
> one such case, for CRIS, with random_bitstring in
> gcc.c-torture/execute/arith-rand-ll.c as well as the target
> libgcc division function.
I wouldn't be surprised if H8 would have tripped over this as well.  Converting
it to REG_CC is still in-progress and a failure to fill a delay slot would likely
have been missed in my performance testing to-date (delay slot filling isn't
terribly important on the H8 thankfully).


> 
> After this, all known performance regressions compared to cc0
> are fixed.
> 
> Ok to commit?
> 
> gcc:
> 	PR target/93372
> 	* reorg.c (fill_slots_from_thread): Allow trial insns that clobber
> 	TARGET_FLAGS_REGNUM as delay-slot fillers.
> 
> gcc/testsuite:
> 	PR target/93372
> 	* gcc.target/cris/pr93372-47.c: New test.
It looks reasonable to me.

jeff
>