<html><head><meta http-equiv="Content-Type" content="text/html charset=us-ascii"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;" class="">Here's a link to the failure in case you want to reproduce this:<div class=""><br class=""></div><div class=""><a href="http://green.lab.llvm.org/green/job/perf_darwin_x86_O3flto/12409/consoleFull#2864721549ba4694-19c4-4d7e-bec5-911270d8a58c" class="">http://green.lab.llvm.org/green/job/perf_darwin_x86_O3flto/12409/consoleFull#2864721549ba4694-19c4-4d7e-bec5-911270d8a58c</a></div><div class=""><br class=""></div><div class="">-- adrian<br class=""><div style=""><blockquote type="cite" class=""><div class="">On Aug 21, 2017, at 1:16 PM, Adrian Prantl via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" class="">llvm-commits@lists.llvm.org</a>> wrote:</div><br class="Apple-interchange-newline"><div class=""><div class="">The green dragon LNT bot has started to timeout between r311333 and r311336.<br class="">I logged into the machine and attached lldb to a clang job that has been running for more than 60min.<br class=""><br class="">Could you please take a look and/or revert this commit?<br class=""><br class="">thanks,<br class="">Adrian<br class=""><br class=""><br class="">green-dragon-06:~ buildslave$ lldb -p 33286<br class="">(lldb) process attach --pid 33286<br class="">Process 33286 stopped<br class="">* thread #1: tid = 0xac2ec5, 0x000000010d76788f clang-6.0`llvm::ConstantFoldLoadFromConstPtr(llvm::Constant*, llvm::Type*, llvm::DataLayout const&) + 927, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP<br class=""> frame #0: 0x000000010d76788f clang-6.0`llvm::ConstantFoldLoadFromConstPtr(llvm::Constant*, llvm::Type*, llvm::DataLayout const&) + 927<br class="">clang-6.0`llvm::ConstantFoldLoadFromConstPtr:<br class="">-> 0x10d76788f <+927>: testb $0x1, 0x50(%rbx)<br class=""> 0x10d767893 <+931>: je 0x10d767e06 ; <+2326><br class=""> 0x10d767899 <+937>: movq %rbx, %rdi<br class=""> 0x10d76789c <+940>: callq 0x10dcc6690 ; llvm::GlobalValue::isDeclaration() const<br class=""><br class="">Executable module set to "/Users/buildslave/jenkins/workspace/perf_darwin_x86_O3flto/host-compiler/bin/clang-6.0".<br class="">Architecture set to: x86_64-apple-macosx.<br class="">(lldb) bt<br class="">* thread #1: tid = 0xac2ec5, 0x000000010d76788f clang-6.0`llvm::ConstantFoldLoadFromConstPtr(llvm::Constant*, llvm::Type*, llvm::DataLayout const&) + 927, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP<br class=""> * frame #0: 0x000000010d76788f clang-6.0`llvm::ConstantFoldLoadFromConstPtr(llvm::Constant*, llvm::Type*, llvm::DataLayout const&) + 927<br class=""> frame #1: 0x00007feb2db8e2d8<br class=""> frame #2: 0x000000010e35e06f clang-6.0`llvm::LibCallSimplifier::optimizeMemCmp(llvm::CallInst*, llvm::IRBuilder<llvm::ConstantFolder, llvm::IRBuilderDefaultInserter>&) + 1119<br class=""> frame #3: 0x000000010e365e7a clang-6.0`llvm::LibCallSimplifier::optimizeStringMemoryLibCall(llvm::CallInst*, llvm::IRBuilder<llvm::ConstantFolder, llvm::IRBuilderDefaultInserter>&) + 266<br class=""> frame #4: 0x000000010e3670b8 clang-6.0`llvm::LibCallSimplifier::optimizeCall(llvm::CallInst*) + 1080<br class=""> frame #5: 0x000000010de61193 clang-6.0`llvm::InstCombiner::tryOptimizeCall(llvm::CallInst*) + 99<br class=""> frame #6: 0x000000010de5c759 clang-6.0`llvm::InstCombiner::visitCallSite(llvm::CallSite) + 2761<br class=""> frame #7: 0x000000010de57aa6 clang-6.0`llvm::InstCombiner::visitCallInst(llvm::CallInst&) + 55574<br class=""> frame #8: 0x000000010de1ef16 clang-6.0`llvm::InstCombiner::run() + 3222<br class=""> frame #9: 0x000000010de20be9 clang-6.0`combineInstructionsOverFunction(llvm::Function&, llvm::InstCombineWorklist&, llvm::AAResults*, llvm::AssumptionCache&, llvm::TargetLibraryInfo&, llvm::DominatorTree&, llvm::OptimizationRemarkEmitter&, bool, llvm::LoopInfo*) + 4201<br class=""> frame #10: 0x000000010de21207 clang-6.0`llvm::InstructionCombiningPass::runOnFunction(llvm::Function&) + 631<br class=""> frame #11: 0x000000010dcfa353 clang-6.0`llvm::FPPassManager::runOnFunction(llvm::Function&) + 547<br class=""> frame #12: 0x000000010dcfa5b3 clang-6.0`llvm::FPPassManager::runOnModule(llvm::Module&) + 51<br class=""> frame #13: 0x000000010dcfaa3e clang-6.0`llvm::legacy::PassManagerImpl::run(llvm::Module&) + 766<br class=""> frame #14: 0x000000010e49e2b4 clang-6.0`clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::DataLayout const&, llvm::Module*, clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream, std::__1::default_delete<llvm::raw_pwrite_stream> >) + 15172<br class=""> frame #15: 0x000000010e6d5ea3 clang-6.0`clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) + 947<br class=""> frame #16: 0x000000010f0ad595 clang-6.0`clang::ParseAST(clang::Sema&, bool, bool) + 469<br class=""> frame #17: 0x000000010e944f2c clang-6.0`clang::FrontendAction::Execute() + 76<br class=""> frame #18: 0x000000010e901021 clang-6.0`clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) + 1217<br class=""> frame #19: 0x000000010e9a9695 clang-6.0`clang::ExecuteCompilerInvocation(clang::CompilerInstance*) + 4949<br class=""> frame #20: 0x000000010cad1492 clang-6.0`cc1_main(llvm::ArrayRef<char const*>, char const*, void*) + 1394<br class=""> frame #21: 0x000000010cacfb43 clang-6.0`main + 11939<br class=""> frame #22: 0x00007fff903cc5ad libdyld.dylib`start + 1<br class=""> frame #23: 0x00007fff903cc5ad libdyld.dylib`start + 1<br class=""><br class=""><blockquote type="cite" class="">On Aug 21, 2017, at 6:55 AM, Sanjay Patel via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" class="">llvm-commits@lists.llvm.org</a>> wrote:<br class=""><br class="">Author: spatel<br class="">Date: Mon Aug 21 06:55:49 2017<br class="">New Revision: 311333<br class=""><br class="">URL: <a href="http://llvm.org/viewvc/llvm-project?rev=311333&view=rev" class="">http://llvm.org/viewvc/llvm-project?rev=311333&view=rev</a><br class="">Log:<br class="">[LibCallSimplifier] try harder to fold memcmp with constant arguments<br class=""><br class="">Try to fold:<br class="">memcmp(X, C, ConstantLength) == 0 --> load X == *C<br class=""><br class="">Without this change, we're unnecessarily checking the alignment of the constant data, <br class="">so we miss the transform in the first 2 tests in the patch.<br class=""><br class="">I noted this shortcoming of LibCallSimpifier in one of the recent CGP memcmp expansion <br class="">patches. This doesn't help the example in:<br class=""><a href="https://bugs.llvm.org/show_bug.cgi?id=34032#c13" class="">https://bugs.llvm.org/show_bug.cgi?id=34032#c13</a><br class="">...directly, but it's worth short-circuiting more of these simple cases since we're <br class="">already trying to do that.<br class=""><br class="">The benefit of transforming to load+cmp is that existing IR analysis/transforms may<br class="">further simplify that code. For example, if the load of the variable is common to <br class="">multiple memcmp calls, CSE can remove the duplicate instructions.<br class=""><br class="">Differential Revision: https://reviews.llvm.org/D36922<br class=""><br class="">Added:<br class=""> llvm/trunk/test/Transforms/InstCombine/memcmp-constant-fold.ll<br class="">Modified:<br class=""> llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp<br class=""><br class="">Modified: llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp<br class="">URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp?rev=311333&r1=311332&r2=311333&view=diff<br class="">==============================================================================<br class="">--- llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp (original)<br class="">+++ llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp Mon Aug 21 06:55:49 2017<br class="">@@ -18,6 +18,7 @@<br class="">#include "llvm/ADT/SmallString.h"<br class="">#include "llvm/ADT/StringMap.h"<br class="">#include "llvm/ADT/Triple.h"<br class="">+#include "llvm/Analysis/ConstantFolding.h"<br class="">#include "llvm/Analysis/OptimizationDiagnosticInfo.h"<br class="">#include "llvm/Analysis/TargetLibraryInfo.h"<br class="">#include "llvm/Analysis/ValueTracking.h"<br class="">@@ -751,29 +752,44 @@ Value *LibCallSimplifier::optimizeMemCmp<br class=""> }<br class=""><br class=""> // memcmp(S1,S2,N/8)==0 -> (*(intN_t*)S1 != *(intN_t*)S2)==0<br class="">+ // TODO: The case where both inputs are constants does not need to be limited<br class="">+ // to legal integers or equality comparison. See block below this.<br class=""> if (DL.isLegalInteger(Len * 8) && isOnlyUsedInZeroEqualityComparison(CI)) {<br class="">-<br class=""> IntegerType *IntType = IntegerType::get(CI->getContext(), Len * 8);<br class=""> unsigned PrefAlignment = DL.getPrefTypeAlignment(IntType);<br class=""><br class="">- if (getKnownAlignment(LHS, DL, CI) >= PrefAlignment &&<br class="">- getKnownAlignment(RHS, DL, CI) >= PrefAlignment) {<br class="">+ // First, see if we can fold either argument to a constant.<br class="">+ Value *LHSV = nullptr;<br class="">+ if (auto *LHSC = dyn_cast<Constant>(LHS)) {<br class="">+ LHSC = ConstantExpr::getBitCast(LHSC, IntType->getPointerTo());<br class="">+ LHSV = ConstantFoldLoadFromConstPtr(LHSC, IntType, DL);<br class="">+ }<br class="">+ Value *RHSV = nullptr;<br class="">+ if (auto *RHSC = dyn_cast<Constant>(RHS)) {<br class="">+ RHSC = ConstantExpr::getBitCast(RHSC, IntType->getPointerTo());<br class="">+ RHSV = ConstantFoldLoadFromConstPtr(RHSC, IntType, DL);<br class="">+ }<br class=""><br class="">+ // Don't generate unaligned loads. If either source is constant data,<br class="">+ // alignment doesn't matter for that source because there is no load.<br class="">+ if (!LHSV && getKnownAlignment(LHS, DL, CI) >= PrefAlignment) {<br class=""> Type *LHSPtrTy =<br class=""> IntType->getPointerTo(LHS->getType()->getPointerAddressSpace());<br class="">+ LHSV = B.CreateLoad(B.CreateBitCast(LHS, LHSPtrTy), "lhsv");<br class="">+ }<br class="">+<br class="">+ if (!RHSV && getKnownAlignment(RHS, DL, CI) >= PrefAlignment) {<br class=""> Type *RHSPtrTy =<br class=""> IntType->getPointerTo(RHS->getType()->getPointerAddressSpace());<br class="">+ RHSV = B.CreateLoad(B.CreateBitCast(RHS, RHSPtrTy), "rhsv");<br class="">+ }<br class=""><br class="">- Value *LHSV =<br class="">- B.CreateLoad(B.CreateBitCast(LHS, LHSPtrTy, "lhsc"), "lhsv");<br class="">- Value *RHSV =<br class="">- B.CreateLoad(B.CreateBitCast(RHS, RHSPtrTy, "rhsc"), "rhsv");<br class="">-<br class="">+ if (LHSV && RHSV)<br class=""> return B.CreateZExt(B.CreateICmpNE(LHSV, RHSV), CI->getType(), "memcmp");<br class="">- }<br class=""> }<br class=""><br class="">- // Constant folding: memcmp(x, y, l) -> cnst (all arguments are constant)<br class="">+ // Constant folding: memcmp(x, y, Len) -> constant (all arguments are const).<br class="">+ // TODO: This is limited to i8 arrays.<br class=""> StringRef LHSStr, RHSStr;<br class=""> if (getConstantStringInfo(LHS, LHSStr) &&<br class=""> getConstantStringInfo(RHS, RHSStr)) {<br class=""><br class="">Added: llvm/trunk/test/Transforms/InstCombine/memcmp-constant-fold.ll<br class="">URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/memcmp-constant-fold.ll?rev=311333&view=auto<br class="">==============================================================================<br class="">--- llvm/trunk/test/Transforms/InstCombine/memcmp-constant-fold.ll (added)<br class="">+++ llvm/trunk/test/Transforms/InstCombine/memcmp-constant-fold.ll Mon Aug 21 06:55:49 2017<br class="">@@ -0,0 +1,65 @@<br class="">+; RUN: opt < %s -instcombine -S -data-layout=e-n32 | FileCheck %s --check-prefix=ALL --check-prefix=LE<br class="">+; RUN: opt < %s -instcombine -S -data-layout=E-n32 | FileCheck %s --check-prefix=ALL --check-prefix=BE<br class="">+<br class="">+declare i32 @memcmp(i8*, i8*, i64)<br class="">+<br class="">+; The alignment of this constant does not matter. We constant fold the load.<br class="">+<br class="">+@charbuf = private unnamed_addr constant [4 x i8] [i8 0, i8 0, i8 0, i8 1], align 1<br class="">+<br class="">+define i1 @memcmp_4bytes_unaligned_constant_i8(i8* align 4 %x) {<br class="">+; LE-LABEL: @memcmp_4bytes_unaligned_constant_i8(<br class="">+; LE-NEXT: [[TMP1:%.*]] = bitcast i8* %x to i32*<br class="">+; LE-NEXT: [[LHSV:%.*]] = load i32, i32* [[TMP1]], align 4<br class="">+; LE-NEXT: [[TMP2:%.*]] = icmp eq i32 [[LHSV]], 16777216<br class="">+; LE-NEXT: ret i1 [[TMP2]]<br class="">+;<br class="">+; BE-LABEL: @memcmp_4bytes_unaligned_constant_i8(<br class="">+; BE-NEXT: [[TMP1:%.*]] = bitcast i8* %x to i32*<br class="">+; BE-NEXT: [[LHSV:%.*]] = load i32, i32* [[TMP1]], align 4<br class="">+; BE-NEXT: [[TMP2:%.*]] = icmp eq i32 [[LHSV]], 1<br class="">+; BE-NEXT: ret i1 [[TMP2]]<br class="">+;<br class="">+ %call = tail call i32 @memcmp(i8* %x, i8* getelementptr inbounds ([4 x i8], [4 x i8]* @charbuf, i64 0, i64 0), i64 4)<br class="">+ %cmpeq0 = icmp eq i32 %call, 0<br class="">+ ret i1 %cmpeq0<br class="">+}<br class="">+<br class="">+; We still don't care about alignment of the constant. We are not limited to constant folding only i8 arrays.<br class="">+; It doesn't matter if the constant operand is the first operand to the memcmp.<br class="">+<br class="">+@intbuf_unaligned = private unnamed_addr constant [4 x i16] [i16 1, i16 2, i16 3, i16 4], align 1<br class="">+<br class="">+define i1 @memcmp_4bytes_unaligned_constant_i16(i8* align 4 %x) {<br class="">+; LE-LABEL: @memcmp_4bytes_unaligned_constant_i16(<br class="">+; LE-NEXT: [[TMP1:%.*]] = bitcast i8* %x to i32*<br class="">+; LE-NEXT: [[RHSV:%.*]] = load i32, i32* [[TMP1]], align 4<br class="">+; LE-NEXT: [[TMP2:%.*]] = icmp eq i32 [[RHSV]], 131073<br class="">+; LE-NEXT: ret i1 [[TMP2]]<br class="">+;<br class="">+; BE-LABEL: @memcmp_4bytes_unaligned_constant_i16(<br class="">+; BE-NEXT: [[TMP1:%.*]] = bitcast i8* %x to i32*<br class="">+; BE-NEXT: [[RHSV:%.*]] = load i32, i32* [[TMP1]], align 4<br class="">+; BE-NEXT: [[TMP2:%.*]] = icmp eq i32 [[RHSV]], 65538<br class="">+; BE-NEXT: ret i1 [[TMP2]]<br class="">+;<br class="">+ %call = tail call i32 @memcmp(i8* bitcast (i16* getelementptr inbounds ([4 x i16], [4 x i16]* @intbuf_unaligned, i64 0, i64 0) to i8*), i8* %x, i64 4)<br class="">+ %cmpeq0 = icmp eq i32 %call, 0<br class="">+ ret i1 %cmpeq0<br class="">+}<br class="">+<br class="">+; TODO: Any memcmp where all arguments are constants should be constant folded. Currently, we only handle i8 array constants.<br class="">+<br class="">+@intbuf = private unnamed_addr constant [2 x i32] [i32 0, i32 1], align 4<br class="">+<br class="">+define i1 @memcmp_3bytes_aligned_constant_i32(i8* align 4 %x) {<br class="">+; ALL-LABEL: @memcmp_3bytes_aligned_constant_i32(<br class="">+; ALL-NEXT: [[CALL:%.*]] = tail call i32 @memcmp(i8* bitcast (i32* getelementptr inbounds ([2 x i32], [2 x i32]* @intbuf, i64 0, i64 1) to i8*), i8* bitcast ([2 x i32]* @intbuf to i8*), i64 3)<br class="">+; ALL-NEXT: [[CMPEQ0:%.*]] = icmp eq i32 [[CALL]], 0<br class="">+; ALL-NEXT: ret i1 [[CMPEQ0]]<br class="">+;<br class="">+ %call = tail call i32 @memcmp(i8* bitcast (i32* getelementptr inbounds ([2 x i32], [2 x i32]* @intbuf, i64 0, i64 1) to i8*), i8* bitcast (i32* getelementptr inbounds ([2 x i32], [2 x i32]* @intbuf, i64 0, i64 0) to i8*), i64 3)<br class="">+ %cmpeq0 = icmp eq i32 %call, 0<br class="">+ ret i1 %cmpeq0<br class="">+}<br class="">+<br class=""><br class=""><br class="">_______________________________________________<br class="">llvm-commits mailing list<br class="">llvm-commits@lists.llvm.org<br class="">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits<br class=""></blockquote><br class="">_______________________________________________<br class="">llvm-commits mailing list<br class=""><a href="mailto:llvm-commits@lists.llvm.org" class="">llvm-commits@lists.llvm.org</a><br class="">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits<br class=""></div></div></blockquote></div><br class=""></div></body></html>