This is a carefully contrived testcase where the X86 ISel is emitting all loads

before other ops, causing it to spill like mad.  This occurs in
164.gzip:flush_block.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19648 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Chris Lattner 2005-01-17 23:16:01 +00:00
parent 869e0434ae
commit a8b6f68a55

View File

@ -70,3 +70,48 @@ int %regpressure2(int* %P) {
ret int %s9
}
;; adds should be the same as muls.
int %regpressure3(short* %P, bool %Cond, int* %Other) {
%A = load short* %P
%Bp = getelementptr short* %P, int 1
%B = load short* %Bp
%Cp = getelementptr short* %P, int 2
%C = load short* %Cp
%Dp = getelementptr short* %P, int 3
%D = load short* %Dp
%Ep = getelementptr short* %P, int 4
%E = load short* %Ep
%Fp = getelementptr short* %P, int 5
%F = load short* %Fp
%Gp = getelementptr short* %P, int 6
%G = load short* %Gp
%Hp = getelementptr short* %P, int 7
%H = load short* %Hp
%Ip = getelementptr short* %P, int 8
%I = load short* %Ip
%Jp = getelementptr short* %P, int 9
%J = load short* %Jp
;; These casts prevent folding the loads into the adds.
%A = cast short %A to int
%B = cast short %B to int
%D = cast short %D to int
%C = cast short %C to int
%E = cast short %E to int
%F = cast short %F to int
%G = cast short %G to int
%H = cast short %H to int
%I = cast short %I to int
%J = cast short %J to int
%s1 = add int %A, %B
%s2 = add int %C, %s1
%s3 = add int %D, %s2
%s4 = add int %E, %s3
%s5 = add int %F, %s4
%s6 = add int %G, %s5
%s7 = add int %H, %s6
%s8 = add int %I, %s7
%s9 = add int %J, %s8
ret int %s9
}