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 void @no_propagate_range_metadata(ptr nowrite %first.coerce, ptr noread nowrite %last.coerce, ptr nocapture %result) {
%for.body.preheader:
br label %for.body
%for.body:
%result.addr.05 = phi ptr [ %incdec.ptr, %for.body ], [ nocapture %result, %for.body.preheader ]
%first.sroa.0.04 = phi ptr [ %incdec.ptr.i.i.i, %for.body ], [ nowrite %first.coerce, %for.body.preheader ]
%0 = load i8, ptr %first.sroa.0.04, align 1
%range_l#0%0 = icmp sge i8 %0, 0
%range_h#0%0 = icmp slt i8 %0, 2
%range#0%0 = and i1 %range_l#0%0, %range_h#0%0
assume_non_poison i1 %range#0%0
store i8 %0, ptr %result.addr.05, align 1
%incdec.ptr.i.i.i = gep inbounds ptr %first.sroa.0.04, 1 x i64 1
%incdec.ptr = gep inbounds ptr %result.addr.05, 1 x i64 1
%lnot.i = icmp eq ptr %incdec.ptr.i.i.i, noread nowrite %last.coerce
br i1 %lnot.i, label %for.end.loopexit, label %for.body
%for.end.loopexit:
ret void
}
Transformation seems to be correct! (syntactically equal)
-- 3. LoopVectorizePass
----------------------------------------
define void @no_propagate_range_metadata(ptr nowrite %first.coerce, ptr noread nowrite %last.coerce, ptr nocapture %result) {
%for.body.preheader:
br label %for.body
%for.body:
%result.addr.05 = phi ptr [ %incdec.ptr, %for.body ], [ nocapture %result, %for.body.preheader ]
%first.sroa.0.04 = phi ptr [ %incdec.ptr.i.i.i, %for.body ], [ nowrite %first.coerce, %for.body.preheader ]
%0 = load i8, ptr %first.sroa.0.04, align 1
%range_l#0%0 = icmp sge i8 %0, 0
%range_h#0%0 = icmp slt i8 %0, 2
%range#0%0 = and i1 %range_l#0%0, %range_h#0%0
assume_non_poison i1 %range#0%0
store i8 %0, ptr %result.addr.05, align 1
%incdec.ptr.i.i.i = gep inbounds ptr %first.sroa.0.04, 1 x i64 1
%incdec.ptr = gep inbounds ptr %result.addr.05, 1 x i64 1
%lnot.i = icmp eq ptr %incdec.ptr.i.i.i, noread nowrite %last.coerce
br i1 %lnot.i, label %for.end.loopexit, label %for.body
%for.end.loopexit:
ret void
}
=>
define void @no_propagate_range_metadata(ptr nowrite %first.coerce, ptr noread nowrite %last.coerce, ptr nocapture %result) {
%iter.check:
%first.coerce4 = ptrtoint ptr nowrite %first.coerce to i64
%last.coerce3 = ptrtoint ptr noread nowrite %last.coerce to i64
%first.coerce2 = ptrtoint ptr nowrite %first.coerce to i64
%result1 = ptrtoint ptr nocapture %result to i64
%0 = sub i64 %last.coerce3, %first.coerce4
%min.iters.check = icmp ult i64 %0, 8
br i1 %min.iters.check, label %vec.epilog.scalar.ph, label %vector.memcheck
%vector.memcheck:
%1 = sub i64 %result1, %first.coerce2
%diff.check = icmp ult i64 %1, 32
br i1 %diff.check, label %vec.epilog.scalar.ph, label %vector.main.loop.iter.check
%vector.main.loop.iter.check:
%min.iters.check5 = icmp ult i64 %0, 32
br i1 %min.iters.check5, label %vec.epilog.ph, label %vector.ph
%vector.ph:
%n.mod.vf = urem i64 %0, 32
%n.vec = sub i64 %0, %n.mod.vf
br label %vector.body
%vector.body:
%index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
%2 = add i64 %index, 0
%next.gep = gep ptr nocapture %result, 1 x i64 %2
%3 = add i64 %index, 16
%4 = add i64 %index, 0
%next.gep7 = gep ptr nowrite %first.coerce, 1 x i64 %4
%5 = add i64 %index, 16
%6 = gep ptr %next.gep7, 1 x i32 0
%7 = bitcast ptr %6 to ptr
%wide.load = load <16 x i8>, ptr %7, align 1
%8 = gep ptr %next.gep7, 1 x i32 16
%9 = bitcast ptr %8 to ptr
%wide.load9 = load <16 x i8>, ptr %9, align 1
%10 = gep ptr %next.gep, 1 x i32 0
%11 = bitcast ptr %10 to ptr
store <16 x i8> %wide.load, ptr %11, align 1
%12 = gep ptr %next.gep, 1 x i32 16
%13 = bitcast ptr %12 to ptr
store <16 x i8> %wide.load9, ptr %13, align 1
%index.next = add nuw i64 %index, 32
%14 = icmp eq i64 %index.next, %n.vec
br i1 %14, label %middle.block, label %vector.body
%middle.block:
%cmp.n = icmp eq i64 %0, %n.vec
br i1 %cmp.n, label %for.end.loopexit, label %vec.epilog.iter.check
%vec.epilog.iter.check:
%ind.end14 = gep ptr nowrite %first.coerce, 1 x i64 %n.vec
%ind.end12 = gep ptr nocapture %result, 1 x i64 %n.vec
%n.vec.remaining = sub i64 %0, %n.vec
%min.epilog.iters.check = icmp ult i64 %n.vec.remaining, 8
br i1 %min.epilog.iters.check, label %vec.epilog.scalar.ph, label %vec.epilog.ph
%vec.epilog.ph:
%vec.epilog.resume.val = phi i64 [ %n.vec, %vec.epilog.iter.check ], [ 0, %vector.main.loop.iter.check ]
%n.mod.vf10 = urem i64 %0, 8
%n.vec11 = sub i64 %0, %n.mod.vf10
%ind.end = gep ptr nocapture %result, 1 x i64 %n.vec11
%ind.end13 = gep ptr nowrite %first.coerce, 1 x i64 %n.vec11
br label %vec.epilog.vector.body
%vec.epilog.vector.body:
%index17 = phi i64 [ %vec.epilog.resume.val, %vec.epilog.ph ], [ %index.next21, %vec.epilog.vector.body ]
%15 = add i64 %index17, 0
%next.gep18 = gep ptr nocapture %result, 1 x i64 %15
%16 = add i64 %index17, 0
%next.gep19 = gep ptr nowrite %first.coerce, 1 x i64 %16
%17 = gep ptr %next.gep19, 1 x i32 0
%18 = bitcast ptr %17 to ptr
%wide.load20 = load <8 x i8>, ptr %18, align 1
%19 = gep ptr %next.gep18, 1 x i32 0
%20 = bitcast ptr %19 to ptr
store <8 x i8> %wide.load20, ptr %20, align 1
%index.next21 = add nuw i64 %index17, 8
%21 = icmp eq i64 %index.next21, %n.vec11
br i1 %21, label %vec.epilog.middle.block, label %vec.epilog.vector.body
%vec.epilog.middle.block:
%cmp.n16 = icmp eq i64 %0, %n.vec11
br i1 %cmp.n16, label %for.end.loopexit, label %vec.epilog.scalar.ph
%vec.epilog.scalar.ph:
%bc.resume.val = phi ptr [ %ind.end, %vec.epilog.middle.block ], [ %ind.end12, %vec.epilog.iter.check ], [ nocapture %result, %vector.memcheck ], [ nocapture %result, %iter.check ]
%bc.resume.val15 = phi ptr [ %ind.end13, %vec.epilog.middle.block ], [ %ind.end14, %vec.epilog.iter.check ], [ nowrite %first.coerce, %vector.memcheck ], [ nowrite %first.coerce, %iter.check ]
br label %for.body
%for.body:
%result.addr.05 = phi ptr [ %incdec.ptr, %for.body ], [ %bc.resume.val, %vec.epilog.scalar.ph ]
%first.sroa.0.04 = phi ptr [ %incdec.ptr.i.i.i, %for.body ], [ %bc.resume.val15, %vec.epilog.scalar.ph ]
%22 = load i8, ptr %first.sroa.0.04, align 1
%range_l#0%22 = icmp sge i8 %22, 0
%range_h#0%22 = icmp slt i8 %22, 2
%range#0%22 = and i1 %range_l#0%22, %range_h#0%22
assume_non_poison i1 %range#0%22
store i8 %22, ptr %result.addr.05, align 1
%incdec.ptr.i.i.i = gep inbounds ptr %first.sroa.0.04, 1 x i64 1
%incdec.ptr = gep inbounds ptr %result.addr.05, 1 x i64 1
%lnot.i = icmp eq ptr %incdec.ptr.i.i.i, noread nowrite %last.coerce
br i1 %lnot.i, label %for.end.loopexit, label %for.body
%for.end.loopexit:
ret void
}
Transformation doesn't verify! (unsound)
ERROR: Source is more defined than target
Example:
ptr nowrite %first.coerce = pointer(non-local, block_id=1, offset=9223372036854775806, attrs=4)
ptr noread nowrite %last.coerce = pointer(non-local, block_id=2, offset=-99302956645482433, attrs=6)
ptr nocapture %result = pointer(non-local, block_id=1, offset=1146032788246361577, attrs=1)
Source:
>> Jump to %for.body
ptr %result.addr.05 = pointer(non-local, block_id=1, offset=1146032788246361577, attrs=1)
ptr %first.sroa.0.04 = pointer(non-local, block_id=1, offset=9223372036854775806, attrs=4)
i8 %0 = #x00 (0)
i1 %range_l#0%0 = #x1 (1)
i1 %range_h#0%0 = #x1 (1)
i1 %range#0%0 = #x1 (1)
ptr %incdec.ptr.i.i.i = pointer(non-local, block_id=1, offset=9223372036854775807, attrs=4)
ptr %incdec.ptr = pointer(non-local, block_id=1, offset=1146032788246361578, attrs=1)
i1 %lnot.i = #x1 (1)
>> Jump to %for.end.loopexit
SOURCE MEMORY STATE
===================
NON-LOCAL BLOCKS:
Block 0 > size: 0 align: 1 alloc type: 0 address: 0
Block 1 > size: 9223372036854775807 align: 2 alloc type: 0 address: 128
Block 2 > size: 2755451490281236299 align: 16 alloc type: 0 address: 9322674993500258368
Block 3 > size: 1423375008704495942 align: 4 alloc type: 0 address: 13835058055282163721
Target:
i64 %first.coerce4 = #x800000000000007e (9223372036854775934, -9223372036854775682)
i64 %last.coerce3 = #x800000000000007f (9223372036854775935, -9223372036854775681)
i64 %first.coerce2 = #x800000000000007e (9223372036854775934, -9223372036854775682)
i64 %result1 = 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' '-S' '-mtriple=x86_64-unknown-linux-gnu' '-passes=loop-vectorize' '-tv-smt-to=20000' '-tv-report-dir=/home/nlopes/alive2/build/logs' '-tv-smt-stats'
------------------- SMT STATS -------------------
Num queries: 4
Num invalid: 0
Num skips: 0
Num trivial: 15 (78.9%)
Num timeout: 0 (0.0%)
Num errors: 0 (0.0%)
Num SAT: 4 (100.0%)
Num UNSAT: 0 (0.0%)
Alive2: Transform doesn't verify; aborting!
+ : 'RUN: at line 1' + /home/nlopes/alive2/build/opt-alive.sh -S -mtriple=x86_64-unknown-linux-gnu -passes=loop-vectorize + /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/LoopVectorize/X86/propagate-metadata.ll FileCheck error: '<stdin>' is empty. FileCheck command line: /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/LoopVectorize/X86/propagate-metadata.ll