<div dir="ltr">I think this is causing the ubsan bot failure. please fix: <div><pre><span class="gmail-stdout"><font color="#000000" face="courier new, courier, monotype, monospace" size="3">lib/Transforms/Vectorize/SLPVectorizer.cpp:4177:34: runtime error: signed integer overflow: 2147483647 + 11 cannot be represented in type 'int'</font><font face="arial, sans-serif"><span style="white-space:normal">
</span></font></span></pre></div><div><span class="gmail-stdout"><a href="http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/15655/steps/check-llvm%20ubsan/logs/stdio">http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/15655/steps/check-llvm%20ubsan/logs/stdio</a><br></span></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Thu, Aug 18, 2016 at 12:50 PM, Matthew Simpson via llvm-commits <span dir="ltr"><<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Author: mssimpso<br>
Date: Thu Aug 18 14:50:32 2016<br>
New Revision: 279125<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=279125&view=rev" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project?rev=279125&view=rev</a><br>
Log:<br>
[SLP] Initialize VectorizedValue when gathering<br>
<br>
We abort building vectorizable trees in some cases (e.g., if the maximum<br>
recursion depth is reached, if the region size is too large, etc.). If this<br>
happens for a reduction, we can be left with a root entry that needs to be<br>
gathered. For these cases, we need make sure we actually set VectorizedValue to<br>
the resulting vector.<br>
<br>
This patch ensures we properly set VectorizedValue, and it also ensures the<br>
insertelement sequence generated for the gathers is inserted at the correct<br>
location.<br>
<br>
Reference: <a href="https://llvm.org/bugs/show_bug.cgi?id=28330" rel="noreferrer" target="_blank">https://llvm.org/bugs/show_<wbr>bug.cgi?id=28330</a><br>
Differential Revison: <a href="https://reviews.llvm.org/D23410" rel="noreferrer" target="_blank">https://reviews.llvm.org/<wbr>D23410</a><br>
<br>
Added:<br>
llvm/trunk/test/Transforms/<wbr>SLPVectorizer/AArch64/gather-<wbr>root.ll<br>
Modified:<br>
llvm/trunk/lib/Transforms/<wbr>Vectorize/SLPVectorizer.cpp<br>
<br>
Modified: llvm/trunk/lib/Transforms/<wbr>Vectorize/SLPVectorizer.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp?rev=279125&r1=279124&r2=279125&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/llvm/trunk/lib/<wbr>Transforms/Vectorize/<wbr>SLPVectorizer.cpp?rev=279125&<wbr>r1=279124&r2=279125&view=diff</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- llvm/trunk/lib/Transforms/<wbr>Vectorize/SLPVectorizer.cpp (original)<br>
+++ llvm/trunk/lib/Transforms/<wbr>Vectorize/SLPVectorizer.cpp Thu Aug 18 14:50:32 2016<br>
@@ -2153,11 +2153,61 @@ void BoUpSLP::<wbr>reorderInputsAccordingToOp<br>
}<br>
<br>
void BoUpSLP::<wbr>setInsertPointAfterBundle(<wbr>ArrayRef<Value *> VL) {<br>
- Instruction *VL0 = cast<Instruction>(VL[0]);<br>
- BasicBlock::iterator NextInst(VL0);<br>
- ++NextInst;<br>
- Builder.SetInsertPoint(VL0-><wbr>getParent(), NextInst);<br>
- Builder.<wbr>SetCurrentDebugLocation(VL0-><wbr>getDebugLoc());<br>
+<br>
+ // Get the basic block this bundle is in. All instructions in the bundle<br>
+ // should be in this block.<br>
+ auto *Front = cast<Instruction>(VL.front());<br>
+ auto *BB = Front->getParent();<br>
+ assert(all_of(make_range(VL.<wbr>begin(), VL.end()), [&](Value *V) -> bool {<br>
+ return cast<Instruction>(V)-><wbr>getParent() == BB;<br>
+ }));<br>
+<br>
+ // The last instruction in the bundle in program order.<br>
+ Instruction *LastInst = nullptr;<br>
+<br>
+ // Find the last instruction. The common case should be that BB has been<br>
+ // scheduled, and the last instruction is VL.back(). So we start with<br>
+ // VL.back() and iterate over schedule data until we reach the end of the<br>
+ // bundle. The end of the bundle is marked by null ScheduleData.<br>
+ if (BlocksSchedules.count(BB)) {<br>
+ auto *Bundle = BlocksSchedules[BB]-><wbr>getScheduleData(VL.back());<br>
+ if (Bundle && Bundle->isPartOfBundle())<br>
+ for (; Bundle; Bundle = Bundle->NextInBundle)<br>
+ LastInst = Bundle->Inst;<br>
+ }<br>
+<br>
+ // LastInst can still be null at this point if there's either not an entry<br>
+ // for BB in BlocksSchedules or there's no ScheduleData available for<br>
+ // VL.back(). This can be the case if buildTree_rec aborts for various<br>
+ // reasons (e.g., the maximum recursion depth is reached, the maximum region<br>
+ // size is reached, etc.). ScheduleData is initialized in the scheduling<br>
+ // "dry-run".<br>
+ //<br>
+ // If this happens, we can still find the last instruction by brute force. We<br>
+ // iterate forwards from Front (inclusive) until we either see all<br>
+ // instructions in the bundle or reach the end of the block. If Front is the<br>
+ // last instruction in program order, LastInst will be set to Front, and we<br>
+ // will visit all the remaining instructions in the block.<br>
+ //<br>
+ // One of the reasons we exit early from buildTree_rec is to place an upper<br>
+ // bound on compile-time. Thus, taking an additional compile-time hit here is<br>
+ // not ideal. However, this should be exceedingly rare since it requires that<br>
+ // we both exit early from buildTree_rec and that the bundle be out-of-order<br>
+ // (causing us to iterate all the way to the end of the block).<br>
+ if (!LastInst) {<br>
+ SmallPtrSet<Value *, 16> Bundle(VL.begin(), VL.end());<br>
+ for (auto &I : make_range(BasicBlock::<wbr>iterator(Front), BB->end())) {<br>
+ if (Bundle.erase(&I))<br>
+ LastInst = &I;<br>
+ if (Bundle.empty())<br>
+ break;<br>
+ }<br>
+ }<br>
+<br>
+ // Set the insertion point after the last instruction in the bundle. Set the<br>
+ // debug location to Front.<br>
+ Builder.SetInsertPoint(BB, next(BasicBlock::iterator(<wbr>LastInst)));<br>
+ Builder.<wbr>SetCurrentDebugLocation(Front-<wbr>>getDebugLoc());<br>
}<br>
<br>
Value *BoUpSLP::Gather(ArrayRef<<wbr>Value *> VL, VectorType *Ty) {<br>
@@ -2235,7 +2285,9 @@ Value *BoUpSLP::vectorizeTree(<wbr>TreeEntry<br>
<br>
if (E->NeedToGather) {<br>
setInsertPointAfterBundle(E-><wbr>Scalars);<br>
- return Gather(E->Scalars, VecTy);<br>
+ auto *V = Gather(E->Scalars, VecTy);<br>
+ E->VectorizedValue = V;<br>
+ return V;<br>
}<br>
<br>
unsigned Opcode = getSameOpcode(E->Scalars);<br>
@@ -2282,7 +2334,10 @@ Value *BoUpSLP::vectorizeTree(<wbr>TreeEntry<br>
E->VectorizedValue = V;<br>
return V;<br>
}<br>
- return Gather(E->Scalars, VecTy);<br>
+ setInsertPointAfterBundle(E-><wbr>Scalars);<br>
+ auto *V = Gather(E->Scalars, VecTy);<br>
+ E->VectorizedValue = V;<br>
+ return V;<br>
}<br>
case Instruction::ExtractValue: {<br>
if (canReuseExtract(E->Scalars, Instruction::ExtractValue)) {<br>
@@ -2294,7 +2349,10 @@ Value *BoUpSLP::vectorizeTree(<wbr>TreeEntry<br>
E->VectorizedValue = V;<br>
return propagateMetadata(V, E->Scalars);<br>
}<br>
- return Gather(E->Scalars, VecTy);<br>
+ setInsertPointAfterBundle(E-><wbr>Scalars);<br>
+ auto *V = Gather(E->Scalars, VecTy);<br>
+ E->VectorizedValue = V;<br>
+ return V;<br>
}<br>
case Instruction::ZExt:<br>
case Instruction::SExt:<br>
<br>
Added: llvm/trunk/test/Transforms/<wbr>SLPVectorizer/AArch64/gather-<wbr>root.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SLPVectorizer/AArch64/gather-root.ll?rev=279125&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/llvm/trunk/test/<wbr>Transforms/SLPVectorizer/<wbr>AArch64/gather-root.ll?rev=<wbr>279125&view=auto</a><br>
==============================<wbr>==============================<wbr>==================<br>
--- llvm/trunk/test/Transforms/<wbr>SLPVectorizer/AArch64/gather-<wbr>root.ll (added)<br>
+++ llvm/trunk/test/Transforms/<wbr>SLPVectorizer/AArch64/gather-<wbr>root.ll Thu Aug 18 14:50:32 2016<br>
@@ -0,0 +1,95 @@<br>
+; RUN: opt < %s -slp-vectorizer -S | FileCheck %s --check-prefix=DEFAULT<br>
+; RUN: opt < %s -slp-recursion-max-depth=0 -slp-vectorizer -S | FileCheck %s --check-prefix=GATHER<br>
+<br>
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:<wbr>64-i128:128-n32:64-S128"<br>
+target triple = "aarch64--linux-gnu"<br>
+<br>
+@a = common global [80 x i8] zeroinitializer, align 16<br>
+<br>
+; DEFAULT-LABEL: @PR28330(<br>
+; DEFAULT: %tmp17 = phi i32 [ %tmp34, %for.body ], [ 0, %entry ]<br>
+; DEFAULT: %tmp18 = phi i32 [ %tmp35, %for.body ], [ %n, %entry ]<br>
+; DEFAULT: %[[S0:.+]] = select <8 x i1> %1, <8 x i32> <i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720>, <8 x i32> <i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80><br>
+; DEFAULT: %[[R0:.+]] = shufflevector <8 x i32> %[[S0]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef><br>
+; DEFAULT: %[[R1:.+]] = add <8 x i32> %[[S0]], %[[R0]]<br>
+; DEFAULT: %[[R2:.+]] = shufflevector <8 x i32> %[[R1]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef><br>
+; DEFAULT: %[[R3:.+]] = add <8 x i32> %[[R1]], %[[R2]]<br>
+; DEFAULT: %[[R4:.+]] = shufflevector <8 x i32> %[[R3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef><br>
+; DEFAULT: %[[R5:.+]] = add <8 x i32> %[[R3]], %[[R4]]<br>
+; DEFAULT: %[[R6:.+]] = extractelement <8 x i32> %[[R5]], i32 0<br>
+; DEFAULT: %tmp34 = add i32 %[[R6]], %tmp17<br>
+;<br>
+; GATHER-LABEL: @PR28330(<br>
+; GATHER: %tmp17 = phi i32 [ %tmp34, %for.body ], [ 0, %entry ]<br>
+; GATHER: %tmp18 = phi i32 [ %tmp35, %for.body ], [ %n, %entry ]<br>
+; GATHER: %tmp19 = select i1 %tmp1, i32 -720, i32 -80<br>
+; GATHER: %tmp21 = select i1 %tmp3, i32 -720, i32 -80<br>
+; GATHER: %tmp23 = select i1 %tmp5, i32 -720, i32 -80<br>
+; GATHER: %tmp25 = select i1 %tmp7, i32 -720, i32 -80<br>
+; GATHER: %tmp27 = select i1 %tmp9, i32 -720, i32 -80<br>
+; GATHER: %tmp29 = select i1 %tmp11, i32 -720, i32 -80<br>
+; GATHER: %tmp31 = select i1 %tmp13, i32 -720, i32 -80<br>
+; GATHER: %tmp33 = select i1 %tmp15, i32 -720, i32 -80<br>
+; GATHER: %[[I0:.+]] = insertelement <8 x i32> undef, i32 %tmp19, i32 0<br>
+; GATHER: %[[I1:.+]] = insertelement <8 x i32> %[[I0]], i32 %tmp21, i32 1<br>
+; GATHER: %[[I2:.+]] = insertelement <8 x i32> %[[I1]], i32 %tmp23, i32 2<br>
+; GATHER: %[[I3:.+]] = insertelement <8 x i32> %[[I2]], i32 %tmp25, i32 3<br>
+; GATHER: %[[I4:.+]] = insertelement <8 x i32> %[[I3]], i32 %tmp27, i32 4<br>
+; GATHER: %[[I5:.+]] = insertelement <8 x i32> %[[I4]], i32 %tmp29, i32 5<br>
+; GATHER: %[[I6:.+]] = insertelement <8 x i32> %[[I5]], i32 %tmp31, i32 6<br>
+; GATHER: %[[I7:.+]] = insertelement <8 x i32> %[[I6]], i32 %tmp33, i32 7<br>
+; GATHER: %[[R0:.+]] = shufflevector <8 x i32> %[[I7]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef><br>
+; GATHER: %[[R1:.+]] = add <8 x i32> %[[I7]], %[[R0]]<br>
+; GATHER: %[[R2:.+]] = shufflevector <8 x i32> %[[R1]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef><br>
+; GATHER: %[[R3:.+]] = add <8 x i32> %[[R1]], %[[R2]]<br>
+; GATHER: %[[R4:.+]] = shufflevector <8 x i32> %[[R3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef><br>
+; GATHER: %[[R5:.+]] = add <8 x i32> %[[R3]], %[[R4]]<br>
+; GATHER: %[[R6:.+]] = extractelement <8 x i32> %[[R5]], i32 0<br>
+; GATHER: %tmp34 = add i32 %[[R6]], %tmp17<br>
+<br>
+define void @PR28330(i32 %n) {<br>
+entry:<br>
+ %tmp0 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1), align 1<br>
+ %tmp1 = icmp eq i8 %tmp0, 0<br>
+ %tmp2 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 2), align 2<br>
+ %tmp3 = icmp eq i8 %tmp2, 0<br>
+ %tmp4 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 3), align 1<br>
+ %tmp5 = icmp eq i8 %tmp4, 0<br>
+ %tmp6 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 4), align 4<br>
+ %tmp7 = icmp eq i8 %tmp6, 0<br>
+ %tmp8 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 5), align 1<br>
+ %tmp9 = icmp eq i8 %tmp8, 0<br>
+ %tmp10 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 6), align 2<br>
+ %tmp11 = icmp eq i8 %tmp10, 0<br>
+ %tmp12 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 7), align 1<br>
+ %tmp13 = icmp eq i8 %tmp12, 0<br>
+ %tmp14 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 8), align 8<br>
+ %tmp15 = icmp eq i8 %tmp14, 0<br>
+ br label %for.body<br>
+<br>
+for.body:<br>
+ %tmp17 = phi i32 [ %tmp34, %for.body ], [ 0, %entry ]<br>
+ %tmp18 = phi i32 [ %tmp35, %for.body ], [ %n, %entry ]<br>
+ %tmp19 = select i1 %tmp1, i32 -720, i32 -80<br>
+ %tmp20 = add i32 %tmp17, %tmp19<br>
+ %tmp21 = select i1 %tmp3, i32 -720, i32 -80<br>
+ %tmp22 = add i32 %tmp20, %tmp21<br>
+ %tmp23 = select i1 %tmp5, i32 -720, i32 -80<br>
+ %tmp24 = add i32 %tmp22, %tmp23<br>
+ %tmp25 = select i1 %tmp7, i32 -720, i32 -80<br>
+ %tmp26 = add i32 %tmp24, %tmp25<br>
+ %tmp27 = select i1 %tmp9, i32 -720, i32 -80<br>
+ %tmp28 = add i32 %tmp26, %tmp27<br>
+ %tmp29 = select i1 %tmp11, i32 -720, i32 -80<br>
+ %tmp30 = add i32 %tmp28, %tmp29<br>
+ %tmp31 = select i1 %tmp13, i32 -720, i32 -80<br>
+ %tmp32 = add i32 %tmp30, %tmp31<br>
+ %tmp33 = select i1 %tmp15, i32 -720, i32 -80<br>
+ %tmp34 = add i32 %tmp32, %tmp33<br>
+ %tmp35 = add nsw i32 %tmp18, -1<br>
+ %tmp36 = icmp eq i32 %tmp35, 0<br>
+ br i1 %tmp36, label %for.end, label %for.body<br>
+<br>
+for.end:<br>
+ ret void<br>
+}<br>
<br>
<br>
______________________________<wbr>_________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-commits</a><br>
</blockquote></div><br></div>