[Mono-devel-list] JIT Performance Regressions on SciMark

[Marcus]

Testing on SciMark shows that the changes made in liveness.c from revision 1.3 
to 1.4 reduced performance significantly in some tests. Here are the results 
on my machine. The assembly was compiled with csc, and the excutable was run 
with -O=all.

			Rev 1.3	Rev 1.4
Comp Score	91.61	81.95
FFT            	51.58	44.68
SOR            	235.4	192.96
Monte Carlo    	31.46	30.09
Sparse MatMult 	62.06	62.84
LU             	77.52	79.18