Test source: git
Comments: Bug in IR semantics or optimization: escapes noescape pointer
Source: <stdin> -- 1. ModuleToFunctionPassAdaptor -- 1. PassManager<Function> : Skipping NOP -- 2. LoopVectorizePass ---------------------------------------- define i64 @foo(ptr nocapture %A, ptr nocapture %B, i64 %n, i64 %m, ptr %O1, ptr %O2) { entry: %cmp = icmp sgt i64 %n, 0 br i1 %cmp, label %for.body.outer.preheader, label %for.end.outer for.body.outer.preheader: br label %for.body.outer for.body.outer: %indvars.outer = phi i64 [ %indvars.outer.next, %for.end.inner ], [ 0, %for.body.outer.preheader ] %cmp2 = icmp sgt i64 %m, 0 br i1 %cmp2, label %for.body.inner.preheader, label %for.end.inner for.body.inner.preheader: br label %for.body.inner for.body.inner: %indvars.iv = phi i64 [ %indvars.iv.next, %for.body.inner ], [ 0, %for.body.inner.preheader ] %arrayidx = gep inbounds ptr nocapture %A, 4 x i64 %indvars.iv %v = load i32, ptr %arrayidx, align 4 %arrayidx2 = gep inbounds ptr nocapture %B, 4 x i64 %indvars.iv store i32 %v, ptr %arrayidx2, align 4 %indvars.iv.next = add i64 %indvars.iv, 1 %exitcond = icmp eq i64 %indvars.iv, %n br i1 %exitcond, label %for.end.inner.loopexit, label %for.body.inner for.end.inner.loopexit: store i64 %indvars.iv, ptr %O1, align 4 br label %for.end.inner for.end.inner: %indvars.outer.next = add i64 %indvars.outer, 1 %exitcond.outer = icmp eq i64 %indvars.outer, %m br i1 %exitcond.outer, label %for.end.outer.loopexit, label %for.body.outer for.end.outer.loopexit: store i64 %indvars.outer, ptr %O2, align 4 br label %for.end.outer for.end.outer: ret i64 undef } Transformation seems to be correct! (syntactically equal) -- 3. LoopVectorizePass ---------------------------------------- define i64 @foo(ptr nocapture %A, ptr nocapture %B, i64 %n, i64 %m, ptr %O1, ptr %O2) { entry: %cmp = icmp sgt i64 %n, 0 br i1 %cmp, label %for.body.outer.preheader, label %for.end.outer for.body.outer.preheader: br label %for.body.outer for.body.outer: %indvars.outer = phi i64 [ %indvars.outer.next, %for.end.inner ], [ 0, %for.body.outer.preheader ] %cmp2 = icmp sgt i64 %m, 0 br i1 %cmp2, label %for.body.inner.preheader, label %for.end.inner for.body.inner.preheader: br label %for.body.inner for.body.inner: %indvars.iv = phi i64 [ %indvars.iv.next, %for.body.inner ], [ 0, %for.body.inner.preheader ] %arrayidx = gep inbounds ptr nocapture %A, 4 x i64 %indvars.iv %v = load i32, ptr %arrayidx, align 4 %arrayidx2 = gep inbounds ptr nocapture %B, 4 x i64 %indvars.iv store i32 %v, ptr %arrayidx2, align 4 %indvars.iv.next = add i64 %indvars.iv, 1 %exitcond = icmp eq i64 %indvars.iv, %n br i1 %exitcond, label %for.end.inner.loopexit, label %for.body.inner for.end.inner.loopexit: store i64 %indvars.iv, ptr %O1, align 4 br label %for.end.inner for.end.inner: %indvars.outer.next = add i64 %indvars.outer, 1 %exitcond.outer = icmp eq i64 %indvars.outer, %m br i1 %exitcond.outer, label %for.end.outer.loopexit, label %for.body.outer for.end.outer.loopexit: store i64 %indvars.outer, ptr %O2, align 4 br label %for.end.outer for.end.outer: ret i64 undef } => define i64 @foo(ptr nocapture %A, ptr nocapture %B, i64 %n, i64 %m, ptr %O1, ptr %O2) { entry: %A2 = ptrtoint ptr nocapture %A to i64 %B1 = ptrtoint ptr nocapture %B to i64 %cmp = icmp sgt i64 %n, 0 br i1 %cmp, label %for.body.outer.preheader, label %for.end.outer for.body.outer.preheader: %#0 = sub i64 %B1, %A2 %#1 = add i64 %n, 1 br label %for.body.outer for.body.outer: %indvars.outer = phi i64 [ %indvars.outer.next, %for.end.inner ], [ 0, %for.body.outer.preheader ] %cmp2 = icmp sgt i64 %m, 0 br i1 %cmp2, label %for.body.inner.preheader, label %for.end.inner for.body.inner.preheader: %min.iters.check = icmp ult i64 %#1, 4 br i1 %min.iters.check, label %scalar.ph, label %vector.memcheck vector.memcheck: %diff.check = icmp ult i64 %#0, 16 br i1 %diff.check, label %scalar.ph, label %vector.ph vector.ph: %n.mod.vf = urem i64 %#1, 4 %n.vec = sub i64 %#1, %n.mod.vf br label %vector.body vector.body: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ] %#2 = add i64 %index, 0 %#3 = gep inbounds ptr nocapture %A, 4 x i64 %#2 %#4 = gep inbounds ptr %#3, 4 x i32 0 %wide.load = load <4 x i32>, ptr %#4, align 4 %#5 = gep inbounds ptr nocapture %B, 4 x i64 %#2 %#6 = gep inbounds ptr %#5, 4 x i32 0 store <4 x i32> %wide.load, ptr %#6, align 4 %index.next = add nuw i64 %index, 4 %#7 = icmp eq i64 %index.next, %n.vec br i1 %#7, label %middle.block, label %vector.body middle.block: %cmp.n = icmp eq i64 %#1, %n.vec %ind.escape = sub i64 %n.vec, 1 br i1 %cmp.n, label %for.end.inner.loopexit, label %scalar.ph scalar.ph: %bc.resume.val = phi i64 [ %n.vec, %middle.block ], [ 0, %for.body.inner.preheader ], [ 0, %vector.memcheck ] br label %for.body.inner for.body.inner: %indvars.iv = phi i64 [ %indvars.iv.next, %for.body.inner ], [ %bc.resume.val, %scalar.ph ] %arrayidx = gep inbounds ptr nocapture %A, 4 x i64 %indvars.iv %v = load i32, ptr %arrayidx, align 4 %arrayidx2 = gep inbounds ptr nocapture %B, 4 x i64 %indvars.iv store i32 %v, ptr %arrayidx2, align 4 %indvars.iv.next = add i64 %indvars.iv, 1 %exitcond = icmp eq i64 %indvars.iv, %n br i1 %exitcond, label %for.end.inner.loopexit, label %for.body.inner for.end.inner.loopexit: %indvars.iv.lcssa = phi i64 [ %indvars.iv, %for.body.inner ], [ %ind.escape, %middle.block ] store i64 %indvars.iv.lcssa, ptr %O1, align 4 br label %for.end.inner for.end.inner: %indvars.outer.next = add i64 %indvars.outer, 1 %exitcond.outer = icmp eq i64 %indvars.outer, %m br i1 %exitcond.outer, label %for.end.outer.loopexit, label %for.body.outer for.end.outer.loopexit: store i64 %indvars.outer, ptr %O2, align 4 br label %for.end.outer for.end.outer: ret i64 undef } Transformation doesn't verify! (unsound) ERROR: Source is more defined than target Example: ptr nocapture %A = poison ptr nocapture %B = poison i64 %n = #x0000000000000000 (0) i64 %m = poison ptr %O1 = poison ptr %O2 = poison Source: i1 %cmp = #x0 (0) >> Jump to %for.end.outer SOURCE MEMORY STATE =================== NON-LOCAL BLOCKS: Block 0 > size: 0 align: 4 alloc type: 0 alive: false address: 0 Block 1 > size: 0 align: 1 alloc type: 0 alive: true address: 8 Block 2 > size: 0 align: 2 alloc type: 0 alive: true address: 6 Block 3 > size: 0 align: 1 alloc type: 0 alive: true address: 6 Block 4 > size: 0 align: 1 alloc type: 0 alive: true address: 7 Target: i64 %A2 = UB triggered! Pass: LoopVectorizePass Command line: '/home/nlopes/llvm/build/bin/opt' '-load=/home/nlopes/alive2/build/tv/tv.so' '-load-pass-plugin=/home/nlopes/alive2/build/tv/tv.so' '-tv-exit-on-error' '-passes=loop-vectorize' '-force-vector-interleave=1' '-force-vector-width=4' '-S' '-tv-smt-to=20000' '-tv-report-dir=/home/nlopes/alive2/build/logs' '-tv-smt-stats' Wrote bitcode to: "/home/nlopes/alive2/build/logs/in_x5PpVuOe_2VL3.bc" ------------------- SMT STATS ------------------- Num queries: 46 Num invalid: 0 Num skips: 0 Num trivial: 2 (4.2%) Num timeout: 0 (0.0%) Num errors: 0 (0.0%) Num SAT: 40 (87.0%) Num UNSAT: 6 (13.0%) Alive2: Transform doesn't verify; aborting!
RUN: at line 1: /home/nlopes/alive2/build/opt-alive.sh < /bitbucket/nlopes/llvm/llvm/test/Transforms/LoopVectorize/partial-lcssa.ll -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S | /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/LoopVectorize/partial-lcssa.ll + /home/nlopes/alive2/build/opt-alive.sh -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S + /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/LoopVectorize/partial-lcssa.ll FileCheck error: '<stdin>' is empty. FileCheck command line: /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/LoopVectorize/partial-lcssa.ll