Coalescing and VNInfo

I've hit a bug in a refactored version of coalescing and I'm trying to
understand what is going on. In particular, I'm curious about this
line in SimpleRegisterCoalescing.cpp:

00710 LHSValsDefinedFromRHS[VNI] =
RHS.getLiveRangeContaining(VNI->def-1)->valno;

Why VNI->def-1? The bug I'm seeing is that RHS returns a NULL
LiveRange because it doesn't contain VNI->def-1:

%reg1093,0 = [5750,5754:2)[5754,5782:1)[5782,5858:0)[5868,5870:0)
0@5782-(5858 5870) 1@5754-(5778) 2@5750

Should it really be VNI->def-1 here? If so, why?

                                            -Dave

I should aff that VNI->def is 5750 which is in the interval.

                                                          -Dave

LLVM is assuming this:
     struct InstrSlots {
       enum {
         LOAD = 0,
         USE = 1,
         DEF = 2,
         STORE = 3,
         NUM = 4
       };

So VNI->def is always modulo 2. For coalescing, it's checking if the RHS is live at the "use" cycle. So it's checking VNI->def-1.

Evan

But why is it looking at a use slot in this case, where the coalescer is
trying to get the vaue number for the def of the RHS register so it can
use that value number in the resulting merged live interval?

The comment at the top of the loop is:

00695 // Loop over the value numbers of the LHS, seeing if any are defined
from
00696 // the RHS.

As I understyand it, it's looking for def-use chains from the RHS interval to
the LHS interval.

That says to me that it's looking for defs of the RHS value, not uses. Why
should it expect that a value number defined in some live interval would also
be used in that live interval at the use slot *before* the def?

What am I missing?

                                                      -Dave

Ah, that's where I got it wrong. I think I get it now. We're looking at the
copy "LHS = RHS" so it takes the def index from the LHS interval and
"def-1" is the use index for "RHS" in the copy. It then tries to look up
the live range for that use index.

I believe I've discovered a problem with my new code so I'll go investigate
that.

                                            -Dave

So VNI->def is always modulo 2. For coalescing, it's checking if the
RHS is live at the "use" cycle. So it's checking VNI->def-1.

But why is it looking at a use slot in this case, where the coalescer is
trying to get the vaue number for the def of the RHS register so it can
use that value number in the resulting merged live interval?

The comment at the top of the loop is:

00695 // Loop over the value numbers of the LHS, seeing if any are
defined from
00696 // the RHS.

As I understyand it, it's looking for def-use chains from the RHS interval
to the LHS interval.

Ah, that's where I got it wrong. I think I get it now. We're looking at the
copy "LHS = RHS" so it takes the def index from the LHS interval and
"def-1" is the use index for "RHS" in the copy. It then tries to look up
the live range for that use index.

Yep, you got it.

Evan