2 ; RUN: opt -S -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -enable-interleaved-mem-accesses=true -debug-only=loop-accesses < %s 2>&1 | FileCheck %s
4 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
6 ; Check that the compile-time-unknown depenendece-distance is resolved
7 ; statically. Due to the non-unit stride of the accesses in this testcase
8 ; we are currently not able to create runtime dependence checks, and therefore
9 ; if we don't resolve the dependence statically we cannot vectorize the loop.
11 ; Specifically in this example, during dependence analysis we get 6 unknown
12 ; dependence distances between the 8 real/imaginary accesses below:
13 ; dist = 8*D, 4+8*D, -4+8*D, -8*D, 4-8*D, -4-8*D.
14 ; At compile time we can prove for all of the above that |dist|>loopBound*step
15 ; (where the step is 8bytes, and the loopBound is D-1), and thereby conclude
16 ; that there are no dependencies (without runtime tests):
17 ; |8*D|>8*D-8, |4+8*D|>8*D-8, |-4+8*D|>8*D-8, etc.
26 ; Complex() : real_(0), imaginary_(0) { }
27 ; Complex(float real, float imaginary) : real_(real), imaginary_(imaginary) { }
28 ; Complex(const Complex &rhs) : real_(rhs.real()), imaginary_(rhs.imaginary()) { }
30 ; inline float real() const { return real_; }
31 ; inline float imaginary() const { return imaginary_; }
33 ; Complex operator+(const Complex& rhs) const
35 ; return Complex(real_ + rhs.real_, imaginary_ + rhs.imaginary_);
38 ; Complex operator-(const Complex& rhs) const
40 ; return Complex(real_ - rhs.real_, imaginary_ - rhs.imaginary_);
44 ; void Test(Complex *out, size_t size)
46 ; size_t D = size / 2;
47 ; for (size_t offset = 0; offset < D; ++offset)
49 ; Complex t0 = out[offset];
50 ; Complex t1 = out[offset + D];
51 ; out[offset] = t1 + t0;
52 ; out[offset + D] = t0 - t1;
57 ; CHECK: LAA: No unsafe dependent memory operations in loop. We don't need runtime memory checks.
61 %class.Complex = type { float, float }
63 define void @Test(ptr nocapture %out, i64 %size) local_unnamed_addr {
65 %div = lshr i64 %size, 1
66 %cmp47 = icmp eq i64 %div, 0
67 br i1 %cmp47, label %for.cond.cleanup, label %for.body.preheader
72 for.cond.cleanup.loopexit:
73 br label %for.cond.cleanup
79 %offset.048 = phi i64 [ %inc, %for.body ], [ 0, %for.body.preheader ]
80 %0 = getelementptr inbounds %class.Complex, ptr %out, i64 %offset.048, i32 0
81 %1 = load float, ptr %0, align 4
82 %imaginary_.i.i = getelementptr inbounds %class.Complex, ptr %out, i64 %offset.048, i32 1
83 %2 = load float, ptr %imaginary_.i.i, align 4
84 %add = add nuw i64 %offset.048, %div
85 %3 = getelementptr inbounds %class.Complex, ptr %out, i64 %add, i32 0
86 %4 = load float, ptr %3, align 4
87 %imaginary_.i.i28 = getelementptr inbounds %class.Complex, ptr %out, i64 %add, i32 1
88 %5 = load float, ptr %imaginary_.i.i28, align 4
89 %add.i = fadd fast float %4, %1
90 %add4.i = fadd fast float %5, %2
91 store float %add.i, ptr %0, align 4
92 store float %add4.i, ptr %imaginary_.i.i, align 4
93 %sub.i = fsub fast float %1, %4
94 %sub4.i = fsub fast float %2, %5
95 store float %sub.i, ptr %3, align 4
96 store float %sub4.i, ptr %imaginary_.i.i28, align 4
97 %inc = add nuw nsw i64 %offset.048, 1
98 %exitcond = icmp eq i64 %inc, %div
99 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body