99 lines · plain
1; RUN: opt -passes='print<access-info>' -disable-output < %s 2>&1 | FileCheck %s2 3target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"4 5; Check that the compile-time-unknown depenendece-distance is resolved 6; statically. Due to the non-unit stride of the accesses in this testcase7; we are currently not able to create runtime dependence checks, and therefore8; if we don't resolve the dependence statically we cannot vectorize the loop.9;10; Specifically in this example, during dependence analysis we get 6 unknown 11; dependence distances between the 8 real/imaginary accesses below: 12; dist = 8*D, 4+8*D, -4+8*D, -8*D, 4-8*D, -4-8*D.13; At compile time we can prove for all of the above that |dist|>loopBound*step14; (where the step is 8bytes, and the loopBound is D-1), and thereby conclude 15; that there are no dependencies (without runtime tests):16; |8*D|>8*D-8, |4+8*D|>8*D-8, |-4+8*D|>8*D-8, etc.17 18; #include <stdlib.h>19; class Complex {20; private:21; float real_;22; float imaginary_;23;24; public:25; Complex() : real_(0), imaginary_(0) { }26; Complex(float real, float imaginary) : real_(real), imaginary_(imaginary) { }27; Complex(const Complex &rhs) : real_(rhs.real()), imaginary_(rhs.imaginary()) { }28; 29; inline float real() const { return real_; }30; inline float imaginary() const { return imaginary_; }31; 32; Complex operator+(const Complex& rhs) const33; {34; return Complex(real_ + rhs.real_, imaginary_ + rhs.imaginary_);35; }36;37; Complex operator-(const Complex& rhs) const38; {39; return Complex(real_ - rhs.real_, imaginary_ - rhs.imaginary_);40; }41; };42;43; void Test(Complex *out, size_t size)44; {45; size_t D = size / 2;46; for (size_t offset = 0; offset < D; ++offset)47; {48; Complex t0 = out[offset];49; Complex t1 = out[offset + D];50; out[offset] = t1 + t0;51; out[offset + D] = t0 - t1;52; }53; }54 55; CHECK-LABEL: Test56; CHECK: Memory dependences are safe57 58 59%class.Complex = type { float, float }60 61define void @Test(ptr nocapture %out, i64 %size) local_unnamed_addr {62entry:63 %div = lshr i64 %size, 164 %cmp47 = icmp eq i64 %div, 065 br i1 %cmp47, label %for.cond.cleanup, label %for.body.preheader66 67for.body.preheader:68 br label %for.body69 70for.cond.cleanup.loopexit:71 br label %for.cond.cleanup72 73for.cond.cleanup:74 ret void75 76for.body:77 %offset.048 = phi i64 [ %inc, %for.body ], [ 0, %for.body.preheader ]78 %0 = getelementptr inbounds %class.Complex, ptr %out, i64 %offset.048, i32 079 %1 = load float, ptr %0, align 480 %imaginary_.i.i = getelementptr inbounds %class.Complex, ptr %out, i64 %offset.048, i32 181 %2 = load float, ptr %imaginary_.i.i, align 482 %add = add nuw i64 %offset.048, %div83 %3 = getelementptr inbounds %class.Complex, ptr %out, i64 %add, i32 084 %4 = load float, ptr %3, align 485 %imaginary_.i.i28 = getelementptr inbounds %class.Complex, ptr %out, i64 %add, i32 186 %5 = load float, ptr %imaginary_.i.i28, align 487 %add.i = fadd fast float %4, %188 %add4.i = fadd fast float %5, %289 store float %add.i, ptr %0, align 490 store float %add4.i, ptr %imaginary_.i.i, align 491 %sub.i = fsub fast float %1, %492 %sub4.i = fsub fast float %2, %593 store float %sub.i, ptr %3, align 494 store float %sub4.i, ptr %imaginary_.i.i28, align 495 %inc = add nuw nsw i64 %offset.048, 196 %exitcond = icmp eq i64 %inc, %div97 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body98}99