Expand LDA testcases.

llvm-svn: 77926
This commit is contained in:
Andreas Bolka 2009-08-02 23:28:14 +00:00
parent 9dc8bd327f
commit 71fc19e991
2 changed files with 63 additions and 12 deletions

View File

@ -4,7 +4,7 @@
@y = common global [256 x i32] zeroinitializer, align 4
;; for (i = 0; i < 256; i++)
;; x[i] = x[256 - i] + y[i]
;; x[i] = x[255 - i] + y[i]
define void @f1(...) nounwind {
entry:
@ -12,8 +12,8 @@ entry:
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%i.256 = sub i64 256, %i
%y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i.256
%i.255 = sub i64 255, %i
%y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i.255
%x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
%x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
%x = load i32* %x.ld.addr ; 0
@ -30,9 +30,8 @@ for.end:
ret void
}
;; // the same example, using more realistic IR
;; for (i = 0; i < 256; i++)
;; x[i] = x[256 - i] + y[i]
;; for (i = 0; i < 100; i++)
;; x[i] = x[255 - i] + y[i]
define void @f2(...) nounwind {
entry:
@ -40,9 +39,9 @@ entry:
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%i.256 = sub i64 0, %i
%i.255 = sub i64 0, %i
%y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
%x.ld.addr = getelementptr [256 x i32]* @x, i64 1, i64 %i.256
%x.ld.addr = getelementptr [256 x i32]* @x, i64 1, i64 %i.255
%x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
%x = load i32* %x.ld.addr ; 0
%y = load i32* %y.ld.addr ; 1
@ -51,14 +50,16 @@ for.body:
; CHECK: 0,2: dep
; CHECK: 1,2: ind
%i.next = add i64 %i, 1
%exitcond = icmp eq i64 %i.next, 256
%exitcond = icmp eq i64 %i.next, 100
br i1 %exitcond, label %for.end, label %for.body
for.end:
ret void
}
;; for (i = 0; i < 100; i++)
;; // the first iteration (i=0) leads to an out-of-bounds access of x. as the
;; // result of this access is undefined, _any_ dependence result is safe.
;; for (i = 0; i < 256; i++)
;; x[i] = x[256 - i] + y[i]
define void @f3(...) nounwind {
@ -76,9 +77,40 @@ for.body:
%r = add i32 %y, %x
store i32 %r, i32* %x.st.addr ; 2
; CHECK: 0,2: dep
; CHECK: 1,2: ind
; CHECK: 1,2:
%i.next = add i64 %i, 1
%exitcond = icmp eq i64 %i.next, 100
%exitcond = icmp eq i64 %i.next, 256
br i1 %exitcond, label %for.end, label %for.body
for.end:
ret void
}
;; // slightly contrived but valid IR for the following loop, where all
;; // accesses in all iterations are within bounds. while this example's first
;; // (ZIV-)subscript is (0, 1), accesses are dependent.
;; for (i = 1; i < 256; i++)
;; x[i] = x[256 - i] + y[i]
define void @f4(...) nounwind {
entry:
br label %for.body
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%i.1 = add i64 1, %i
%i.256 = sub i64 -1, %i
%y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i.1
%x.ld.addr = getelementptr [256 x i32]* @x, i64 1, i64 %i.256
%x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.1
%x = load i32* %x.ld.addr ; 0
%y = load i32* %y.ld.addr ; 1
%r = add i32 %y, %x
store i32 %r, i32* %x.st.addr ; 2
; CHECK: 0,2: dep
; CHECK: 1,2: ind
%i.next = add i64 %i, 1
%exitcond = icmp eq i64 %i.next, 256
br i1 %exitcond, label %for.end, label %for.body
for.end:

View File

@ -42,3 +42,22 @@ for.body:
for.end:
ret void
}
;; x[6] = x[6]
define void @f3(...) nounwind {
entry:
br label %for.body
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%x = load i32* getelementptr ([256 x i32]* @x, i32 0, i64 6)
store i32 %x, i32* getelementptr ([256 x i32]* @x, i32 0, i64 6)
; CHECK: 0,1: dep
%i.next = add i64 %i, 1
%exitcond = icmp eq i64 %i.next, 256
br i1 %exitcond, label %for.end, label %for.body
for.end:
ret void
}