st registers are allocated by the x87 simulator and don't need a noreg

enabled Push peephole optimization

handle binops with two identical operands
add simulation for be_Return
fixed simulation for be_Copy
add register killing code at begin of every block
use node index as block number
create only labels for block that are jump targets

changed name prefix for SSE floating point nodes from f to x

added UKNWN register
set Unknown nodes to ignore and assign them the UKNWN register

fixed indents
added should be same constraint for lea
added additional statistics

fixed cmp with immediate
emit %cl for shift/rotate ops
fixed some minor bugs

add most virtual fp and real x87 nodes

removed some unused variables
fixed function type

add new 8bit store node

removed unnecessary check

code cleanups
fixed am optimizer
changed comments

fixed some bugs
added sub -> neg-add transformation in case OUT==IN2

added conversions