计算机组织_21.pdfVIP

Reading Assignment • Read pages 63-end of “Automatic Program Optimization,” by Ron Cytron. (Linked from the class Web page.) © CS 701 Fall 2008 430 Not all Data Flow Problems are Distributive Constant propagation is not distributive. Consider the following (with variables (x,y,z)): t = (T,1,3) t =(T,2,2) 1 2 x=y+z Now f(t)=t’ where t’(y) = t(y), t’(z) = t(z), t’(x) = if t(y)=⊥ or t(z) = ⊥ then ⊥ elseif t(y)=T or t(z) =T then T else t(y)+t(z) © CS 701 Fall 2008 431 Now f(t ∧t ) = f(T,⊥,⊥) = (⊥,⊥,⊥) 1 2 f(t1) = (4,1,3) f(t2) = (4,2,2) f(t )∧f(t ) = (4,⊥,⊥) ≥ (⊥,⊥,⊥) 1 2 © CS 701 Fall 2008 432 Why does it Matter if a Data Flow Problem isn’t Distributive? Consider actual program execution paths from b to (say) b . 0 k One path might be b0,bi1,bi2,...,bin where b =b . in k At bk the Data Flow information we want is f (...f (f (f (T)))...) ≡ f(b ,b ,...,b ) in i2 i1 0 0 1 in On a different path to bk, say b0,bj 1,bj 2,...,bj m, where bj m=bk the Data Flow result we get is f (...f (f (f (T)))...) ≡ j m j 2 j 1 0 f(b0,bj 1,...,bj m). © CS 701 Fall 2008