[LLVMdev] [PBQP] applyR2() spill cost calculations

[Jonas Paulsson]

Hi Lang,

I have a question regarding applyR2() and spill costs.

In reference (2) in RegallocPBQP.cpp (Scholz/Eckstein), the ReduceII() procedure computes a Delta matrix with the first row and column always being zero. I see that in applyR2(), this is not the case. Spill costs are also updated. This seems to contradict the fact that row/col 0 should reflect the cost of spilling a vreg, which is constant, or?

For example:

* ApplyR2(NId 74):
XCosts: 1.000000e+01 1.000000e+00 1.000000e+00
YXECosts:
1.000000e+00 1.000000e+00 1.000000e+00
1.000000e+00 inf 1.000000e+01
1.000000e+00 1.000000e+01 inf
ZXECosts:
0.000000e+00 0.000000e+00 0.000000e+00
0.000000e+00 0.000000e+00 0.000000e+00
0.000000e+00 0.000000e+00 0.000000e+00
0.000000e+00 0.000000e+00 0.000000e+00
0.000000e+00 0.000000e+00 0.000000e+00
Delta:
2.000000e+00 2.000000e+00 2.000000e+00 2.000000e+00 2.000000e+00
1.100000e+01 1.100000e+01 1.100000e+01 1.100000e+01 1.100000e+01
1.100000e+01 1.100000e+01 1.100000e+01 1.100000e+01 1.100000e+01

/Jonas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150212/93d3def8/attachment.html>