callee-saved registers at the end of the lists. Also prefer to avoid using
the low registers that are in register subclasses required by certain
instructions, so that those registers will more likely be available when needed.
This change makes a huge improvement in spilling in some cases. Thanks to
Jakob for helping me realize the problem.
Most of this patch is fixing the testsuite. There are quite a few places
where we're checking for specific registers. I changed those to wildcards
in places where that doesn't weaken the tests. The spill-q.ll and
thumb2-spill-q.ll tests stopped spilling with this change, so I added a bunch
of live values to force spills on those tests.
llvm-svn: 116055
reapply: reimplement the second half of the or/add optimization. We should now
with no changes. Turns out that one missing "Defs = [EFLAGS]" can upset things
a bit.
llvm-svn: 116040
only end up emitting LEA instead of OR. If we aren't able to promote
something into an LEA, we should never be emitting it as an ADD.
Add some testcases that we emit "or" in cases where we used to produce
an "add".
llvm-svn: 116026
is general goodness because it allows ORs to be converted to LEA to avoid
inserting copies. However, this is bad because it makes the generated .s
file less obvious and gives valgrind heartburn (tons of false positives in
bitfield code).
While the general fix should be in valgrind, we can at least try to avoid
emitting ADD instructions that *don't* get promoted to LEA. This is more
work because it requires introducing pseudo instructions to represents
"add that knows the bits are disjoint", but hey, people really love valgrind.
This fixes this testcase:
https://bugs.kde.org/show_bug.cgi?id=242137#c20
the add r/i cases are coming next.
llvm-svn: 116007
use. Since TEST is completely different than all other binops,
don't define a multipattern for it.
This completes factorization of binops.
llvm-svn: 115982
are not defined as returning EFLAGS (like add_flag and friends),
the entire multipattern and several of the subclasses need to be
cloned.
This could be handled through better instantiation support in tblgen,
but it isn't meta enough.
llvm-svn: 115964
Before the implementation of isLegalAddressingMode, some rare cases
of code were miscompiled if optimized with the LoopStrengthReduce pass.
It is unclear (to me) if LSR is "allowed" to produce wrong code with a
bad TargetLowering, or if the bug is elsewhere and this patch just
hides it.
llvm-svn: 115919
Start using ArithBinOpEFLAGS for OR, XOR, and SUB.
This removes 500 lines from the .td file. Now AND/OR/XOR/SUB are all
defined exactly the same way instead of being close relatives.
llvm-svn: 115896
which instantiates the 34 versions of and all in one
swoop. The BaseOpc/BaseOpc2/BaseOpc4 stuff should not
be required, but tblgen's feeble brain explodes when I
use Or4<BaseOpc>.V in the multipattern.
No change in the generated .inc files.
llvm-svn: 115893
This does change the generated .inc files to include the implicit use/def of eax.
Since these instructions are only generated by the assembler and disassembler it
doesn't actually matter though.
llvm-svn: 115885
