http://llvm.org/bugs/show_bug.cgi?id=17133
Bug ID: 17133
Summary: missing SCEV support for xor in loops
Product: libraries
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Scalar Optimizations
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Classification: Unclassified
int x; void f() { for (int n = 0; n < 500000000; ++n) x ^= 1; }
With -O3 -fno-vectorize, we produce
define void @_Z1fv() #0 {
%x.promoted = load i32* @x, align 4, !tbaa !0
br label %1
; <label>:1 ; preds = %1, %0
%2 = phi i32 [ %x.promoted, %0 ], [ %3, %1 ]
%n.01 = phi i32 [ 0, %0 ], [ %4, %1 ]
%3 = xor i32 %2, 1
%4 = add nsw i32 %n.01, 1
%exitcond = icmp eq i32 %4, 500000000
br i1 %exitcond, label %5, label %1
; <label>:5 ; preds = %1
store i32 %3, i32* @x, align 4, !tbaa !0
ret void
}
With -O3 -fvectorize, we produce
define void @_Z1fv() #0 {
vector.ph:
%x.promoted = load i32* @x, align 4, !tbaa !0
%0 = insertelement <4 x i32> <i32 undef, i32 0, i32 0, i32 0>, i32
%x.promoted, i32 0
br label %vector.body
vector.body: ; preds = %vector.body,
%vector.ph
%index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
%vec.phi = phi <4 x i32> [ %0, %vector.ph ], [ %1, %vector.body ]
%vec.phi2 = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ %2, %vector.body
]
%1 = xor <4 x i32> %vec.phi, <i32 1, i32 1, i32 1, i32 1>
%2 = xor <4 x i32> %vec.phi2, <i32 1, i32 1, i32 1, i32 1>
%index.next = add i32 %index, 8
%3 = icmp eq i32 %index.next, 500000000
br i1 %3, label %middle.block, label %vector.body
middle.block: ; preds = %vector.body
%bin.rdx = xor <4 x i32> %2, %1
%rdx.shuf = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32
2, i32 3, i32 undef, i32 undef>
%bin.rdx5 = xor <4 x i32> %bin.rdx, %rdx.shuf
%rdx.shuf6 = shufflevector <4 x i32> %bin.rdx5, <4 x i32> undef, <4 x i32>
<i32 1, i32 undef, i32 undef, i32 undef>
%bin.rdx7 = xor <4 x i32> %bin.rdx5, %rdx.shuf6
%4 = extractelement <4 x i32> %bin.rdx7, i32 0
store i32 %4, i32* @x, align 4, !tbaa !0
ret void
}
This is a success for the vectorizer but a failure for LLVM as a whole. SCEV
should be able to determine that 'x' is unchanged by this loop and we should be
able to remove it.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
LLVMbugs mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/llvmbugs
