linaro-toolchain

linaro-toolchain@lists.linaro.org

9 participants
5482 discussions

[TCWG CI] 453.povray slowed down by 3% after llvm: [llvm] Use range-based for loops (NFC)

by ci_notify＠linaro.org

After llvm commit f240e528cea25fd2a9ae01b1e1fe77f507ed7a2c Author: Kazu Hirata <kazu(a)google.com> [llvm] Use range-based for loops (NFC) the following benchmarks slowed down by more than 2%: - 453.povray slowed down by 3% from 4906 to 5047 perf samples Below reproducer instructions can be used to re-build both "first_bad" and "last_good" cross-toolchains used in this bisection. Naturally, the scripts will fail when triggerring benchmarking jobs if you don't have access to Linaro TCWG CI. For your convenience, we have uploaded tarballs with pre-processed source and assembly files at: - First_bad save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… - Last_good save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… - Baseline save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Configuration: - Benchmark: SPEC CPU2006 - Toolchain: Clang + Glibc + LLVM Linker - Version: all components were built from their tip of trunk - Target: aarch64-linux-gnu - Compiler flags: -O2 -flto - Hardware: NVidia TX1 4x Cortex-A57 This benchmarking CI is work-in-progress, and we welcome feedback and suggestions at linaro-toolchain(a)lists.linaro.org . In our improvement plans is to add support for SPEC CPU2017 benchmarks and provide "perf report/annotate" data behind these reports. THIS IS THE END OF INTERESTING STUFF. BELOW ARE LINKS TO BUILDS, REPRODUCTION INSTRUCTIONS, AND THE RAW COMMIT. This commit has regressed these CI configurations: - tcwg_bmk_llvm_tx1/llvm-master-aarch64-spec2k6-O2_LTO First_bad build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Last_good build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Baseline build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Even more details: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Reproduce builds: <cut> mkdir investigate-llvm-f240e528cea25fd2a9ae01b1e1fe77f507ed7a2c cd investigate-llvm-f240e528cea25fd2a9ae01b1e1fe77f507ed7a2c # Fetch scripts git clone https://git.linaro.org/toolchain/jenkins-scripts # Fetch manifests and test.sh script mkdir -p artifacts/manifests curl -o artifacts/manifests/build-baseline.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail curl -o artifacts/manifests/build-parameters.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail curl -o artifacts/test.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail chmod +x artifacts/test.sh # Reproduce the baseline build (build all pre-requisites) ./jenkins-scripts/tcwg_bmk-build.sh @@ artifacts/manifests/build-baseline.sh # Save baseline build state (which is then restored in artifacts/test.sh) mkdir -p ./bisect rsync -a --del --delete-excluded --exclude /bisect/ --exclude /artifacts/ --exclude /llvm/ ./ ./bisect/baseline/ cd llvm # Reproduce first_bad build git checkout --detach f240e528cea25fd2a9ae01b1e1fe77f507ed7a2c ../artifacts/test.sh # Reproduce last_good build git checkout --detach c572eb1ad9d8a528bcaff0160888aff31b1f4b5f ../artifacts/test.sh cd .. </cut> Full commit (up to 1000 lines): <cut> commit f240e528cea25fd2a9ae01b1e1fe77f507ed7a2c Author: Kazu Hirata <kazu(a)google.com> Date: Mon Nov 29 09:04:44 2021 -0800 [llvm] Use range-based for loops (NFC) --- llvm/lib/CodeGen/MachinePipeliner.cpp | 7 ++--- .../CodeGen/SelectionDAG/ResourcePriorityQueue.cpp | 4 +-- llvm/lib/Object/ELFObjectFile.cpp | 4 +-- llvm/lib/ObjectYAML/COFFEmitter.cpp | 32 ++++++++++------------ llvm/lib/Passes/StandardInstrumentations.cpp | 4 +-- llvm/lib/ProfileData/InstrProf.cpp | 11 ++++---- llvm/lib/Target/Hexagon/HexagonCommonGEP.cpp | 18 ++++++------ llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp | 32 ++++++++++------------ 8 files changed, 50 insertions(+), 62 deletions(-) diff --git a/llvm/lib/CodeGen/MachinePipeliner.cpp b/llvm/lib/CodeGen/MachinePipeliner.cpp index 21be6718b7d9..8d6459a627fa 100644 --- a/llvm/lib/CodeGen/MachinePipeliner.cpp +++ b/llvm/lib/CodeGen/MachinePipeliner.cpp @@ -1519,9 +1519,8 @@ static bool pred_L(SetVector<SUnit *> &NodeOrder, SmallSetVector<SUnit *, 8> &Preds, const NodeSet *S = nullptr) { Preds.clear(); - for (SetVector<SUnit *>::iterator I = NodeOrder.begin(), E = NodeOrder.end(); - I != E; ++I) { - for (const SDep &Pred : (*I)->Preds) { + for (const SUnit *SU : NodeOrder) { + for (const SDep &Pred : SU->Preds) { if (S && S->count(Pred.getSUnit()) == 0) continue; if (ignoreDependence(Pred, true)) @@ -1530,7 +1529,7 @@ static bool pred_L(SetVector<SUnit *> &NodeOrder, Preds.insert(Pred.getSUnit()); } // Back-edges are predecessors with an anti-dependence. - for (const SDep &Succ : (*I)->Succs) { + for (const SDep &Succ : SU->Succs) { if (Succ.getKind() != SDep::Anti) continue; if (S && S->count(Succ.getSUnit()) == 0) diff --git a/llvm/lib/CodeGen/SelectionDAG/ResourcePriorityQueue.cpp b/llvm/lib/CodeGen/SelectionDAG/ResourcePriorityQueue.cpp index 55fe26eb64cd..2695ed36991c 100644 --- a/llvm/lib/CodeGen/SelectionDAG/ResourcePriorityQueue.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/ResourcePriorityQueue.cpp @@ -268,8 +268,8 @@ bool ResourcePriorityQueue::isResourceAvailable(SUnit *SU) { // Now see if there are no other dependencies // to instructions already in the packet. - for (unsigned i = 0, e = Packet.size(); i != e; ++i) - for (const SDep &Succ : Packet[i]->Succs) { + for (const SUnit *S : Packet) + for (const SDep &Succ : S->Succs) { // Since we do not add pseudos to packets, might as well // ignore order deps. if (Succ.isCtrl()) diff --git a/llvm/lib/Object/ELFObjectFile.cpp b/llvm/lib/Object/ELFObjectFile.cpp index 50035d6c7523..cf1f12d9a9a7 100644 --- a/llvm/lib/Object/ELFObjectFile.cpp +++ b/llvm/lib/Object/ELFObjectFile.cpp @@ -682,7 +682,7 @@ readDynsymVersionsImpl(const ELFFile<ELFT> &EF, std::vector<VersionEntry> Ret; size_t I = 0; - for (auto It = Symbols.begin(), E = Symbols.end(); It != E; ++It) { + for (const ELFSymbolRef &Sym : Symbols) { ++I; Expected<const typename ELFT::Versym *> VerEntryOrErr = EF.template getEntry<typename ELFT::Versym>(*VerSec, I); @@ -691,7 +691,7 @@ readDynsymVersionsImpl(const ELFFile<ELFT> &EF, " from " + describe(EF, *VerSec) + ": " + toString(VerEntryOrErr.takeError())); - Expected<uint32_t> FlagsOrErr = It->getFlags(); + Expected<uint32_t> FlagsOrErr = Sym.getFlags(); if (!FlagsOrErr) return createError("unable to read flags for symbol with index " + Twine(I) + ": " + toString(FlagsOrErr.takeError())); diff --git a/llvm/lib/ObjectYAML/COFFEmitter.cpp b/llvm/lib/ObjectYAML/COFFEmitter.cpp index 5f38ca13cfc2..66ad16db1ba4 100644 --- a/llvm/lib/ObjectYAML/COFFEmitter.cpp +++ b/llvm/lib/ObjectYAML/COFFEmitter.cpp @@ -476,29 +476,25 @@ static bool writeCOFF(COFFParser &CP, raw_ostream &OS) { assert(OS.tell() == CP.SectionTableStart); // Output section table. - for (std::vector<COFFYAML::Section>::iterator i = CP.Obj.Sections.begin(), - e = CP.Obj.Sections.end(); - i != e; ++i) { - OS.write(i->Header.Name, COFF::NameSize); - OS << binary_le(i->Header.VirtualSize) - << binary_le(i->Header.VirtualAddress) - << binary_le(i->Header.SizeOfRawData) - << binary_le(i->Header.PointerToRawData) - << binary_le(i->Header.PointerToRelocations) - << binary_le(i->Header.PointerToLineNumbers) - << binary_le(i->Header.NumberOfRelocations) - << binary_le(i->Header.NumberOfLineNumbers) - << binary_le(i->Header.Characteristics); + for (const COFFYAML::Section &S : CP.Obj.Sections) { + OS.write(S.Header.Name, COFF::NameSize); + OS << binary_le(S.Header.VirtualSize) + << binary_le(S.Header.VirtualAddress) + << binary_le(S.Header.SizeOfRawData) + << binary_le(S.Header.PointerToRawData) + << binary_le(S.Header.PointerToRelocations) + << binary_le(S.Header.PointerToLineNumbers) + << binary_le(S.Header.NumberOfRelocations) + << binary_le(S.Header.NumberOfLineNumbers) + << binary_le(S.Header.Characteristics); } assert(OS.tell() == CP.SectionTableStart + CP.SectionTableSize); unsigned CurSymbol = 0; StringMap<unsigned> SymbolTableIndexMap; - for (std::vector<COFFYAML::Symbol>::iterator I = CP.Obj.Symbols.begin(), - E = CP.Obj.Symbols.end(); - I != E; ++I) { - SymbolTableIndexMap[I->Name] = CurSymbol; - CurSymbol += 1 + I->Header.NumberOfAuxSymbols; + for (const COFFYAML::Symbol &Sym : CP.Obj.Symbols) { + SymbolTableIndexMap[Sym.Name] = CurSymbol; + CurSymbol += 1 + Sym.Header.NumberOfAuxSymbols; } // Output section data. diff --git a/llvm/lib/Passes/StandardInstrumentations.cpp b/llvm/lib/Passes/StandardInstrumentations.cpp index 8e6be6730ea4..27a6c519ff82 100644 --- a/llvm/lib/Passes/StandardInstrumentations.cpp +++ b/llvm/lib/Passes/StandardInstrumentations.cpp @@ -225,8 +225,8 @@ std::string doSystemDiff(StringRef Before, StringRef After, return "Unable to read result."; // Clean up. - for (unsigned I = 0; I < NumFiles; ++I) { - std::error_code EC = sys::fs::remove(FileName[I]); + for (const std::string &I : FileName) { + std::error_code EC = sys::fs::remove(I); if (EC) return "Unable to remove temporary file."; } diff --git a/llvm/lib/ProfileData/InstrProf.cpp b/llvm/lib/ProfileData/InstrProf.cpp index 1168ad27fe52..ab3487ecffe8 100644 --- a/llvm/lib/ProfileData/InstrProf.cpp +++ b/llvm/lib/ProfileData/InstrProf.cpp @@ -657,19 +657,18 @@ void InstrProfValueSiteRecord::merge(InstrProfValueSiteRecord &Input, Input.sortByTargetValues(); auto I = ValueData.begin(); auto IE = ValueData.end(); - for (auto J = Input.ValueData.begin(), JE = Input.ValueData.end(); J != JE; - ++J) { - while (I != IE && I->Value < J->Value) + for (const InstrProfValueData &J : Input.ValueData) { + while (I != IE && I->Value < J.Value) ++I; - if (I != IE && I->Value == J->Value) { + if (I != IE && I->Value == J.Value) { bool Overflowed; - I->Count = SaturatingMultiplyAdd(J->Count, Weight, I->Count, &Overflowed); + I->Count = SaturatingMultiplyAdd(J.Count, Weight, I->Count, &Overflowed); if (Overflowed) Warn(instrprof_error::counter_overflow); ++I; continue; } - ValueData.insert(I, *J); + ValueData.insert(I, J); } } diff --git a/llvm/lib/Target/Hexagon/HexagonCommonGEP.cpp b/llvm/lib/Target/Hexagon/HexagonCommonGEP.cpp index 43f0758f6598..8c3b9572201e 100644 --- a/llvm/lib/Target/Hexagon/HexagonCommonGEP.cpp +++ b/llvm/lib/Target/Hexagon/HexagonCommonGEP.cpp @@ -476,10 +476,10 @@ namespace { } // end anonymous namespace static const NodeSet *node_class(GepNode *N, NodeSymRel &Rel) { - for (NodeSymRel::iterator I = Rel.begin(), E = Rel.end(); I != E; ++I) - if (I->count(N)) - return &*I; - return nullptr; + for (const NodeSet &S : Rel) + if (S.count(N)) + return &S; + return nullptr; } // Create an ordered pair of GepNode pointers. The pair will be used in @@ -589,9 +589,8 @@ void HexagonCommonGEP::common() { dbgs() << "{ " << I->first << ", " << I->second << " }\n"; dbgs() << "Gep equivalence classes:\n"; - for (NodeSymRel::iterator I = EqRel.begin(), E = EqRel.end(); I != E; ++I) { + for (const NodeSet &S : EqRel) { dbgs() << '{'; - const NodeSet &S = *I; for (NodeSet::const_iterator J = S.begin(), F = S.end(); J != F; ++J) { if (J != S.begin()) dbgs() << ','; @@ -604,8 +603,7 @@ void HexagonCommonGEP::common() { // Create a projection from a NodeSet to the minimal element in it. using ProjMap = std::map<const NodeSet *, GepNode *>; ProjMap PM; - for (NodeSymRel::iterator I = EqRel.begin(), E = EqRel.end(); I != E; ++I) { - const NodeSet &S = *I; + for (const NodeSet &S : EqRel) { GepNode *Min = *std::min_element(S.begin(), S.end(), NodeOrder); std::pair<ProjMap::iterator,bool> Ins = PM.insert(std::make_pair(&S, Min)); (void)Ins; @@ -1280,8 +1278,8 @@ bool HexagonCommonGEP::runOnFunction(Function &F) { return false; // For now bail out on C++ exception handling. - for (Function::iterator A = F.begin(), Z = F.end(); A != Z; ++A) - for (BasicBlock::iterator I = A->begin(), E = A->end(); I != E; ++I) + for (const BasicBlock &BB : F) + for (const Instruction &I : BB) if (isa<InvokeInst>(I) || isa<LandingPadInst>(I)) return false; diff --git a/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp b/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp index c2ddfd6164f4..c35e67d6726f 100644 --- a/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp +++ b/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp @@ -130,10 +130,8 @@ VisitGlobalVariableForEmission(const GlobalVariable *GV, for (unsigned i = 0, e = GV->getNumOperands(); i != e; ++i) DiscoverDependentGlobals(GV->getOperand(i), Others); - for (DenseSet<const GlobalVariable *>::iterator I = Others.begin(), - E = Others.end(); - I != E; ++I) - VisitGlobalVariableForEmission(*I, Order, Visited, Visiting); + for (const GlobalVariable *GV : Others) + VisitGlobalVariableForEmission(GV, Order, Visited, Visiting); // Now we can visit ourself Order.push_back(GV); @@ -699,35 +697,33 @@ static bool useFuncSeen(const Constant *C, void NVPTXAsmPrinter::emitDeclarations(const Module &M, raw_ostream &O) { DenseMap<const Function *, bool> seenMap; - for (Module::const_iterator FI = M.begin(), FE = M.end(); FI != FE; ++FI) { - const Function *F = &*FI; - - if (F->getAttributes().hasFnAttr("nvptx-libcall-callee")) { - emitDeclaration(F, O); + for (const Function &F : M) { + if (F.getAttributes().hasFnAttr("nvptx-libcall-callee")) { + emitDeclaration(&F, O); continue; } - if (F->isDeclaration()) { - if (F->use_empty()) + if (F.isDeclaration()) { + if (F.use_empty()) continue; - if (F->getIntrinsicID()) + if (F.getIntrinsicID()) continue; - emitDeclaration(F, O); + emitDeclaration(&F, O); continue; } - for (const User *U : F->users()) { + for (const User *U : F.users()) { if (const Constant *C = dyn_cast<Constant>(U)) { if (usedInGlobalVarDef(C)) { // The use is in the initialization of a global variable // that is a function pointer, so print a declaration // for the original function - emitDeclaration(F, O); + emitDeclaration(&F, O); break; } // Emit a declaration of this function if the function that // uses this constant expr has already been seen. if (useFuncSeen(C, seenMap)) { - emitDeclaration(F, O); + emitDeclaration(&F, O); break; } } @@ -746,11 +742,11 @@ void NVPTXAsmPrinter::emitDeclarations(const Module &M, raw_ostream &O) { // appearing in the module before the callee. so print out // a declaration for the callee. if (seenMap.find(caller) != seenMap.end()) { - emitDeclaration(F, O); + emitDeclaration(&F, O); break; } } - seenMap[F] = true; + seenMap[&F] = true; } } </cut>

2 years, 11 months

[TCWG CI] 458.sjeng slowed down by 5% after llvm: Reland "[LICM] Hoist LOAD without sinking the STORE"

by ci_notify＠linaro.org

After llvm commit 2cdc6f2ca62e83fec445114fbbe6276e9ab2a7d0 Author: Djordje Todorovic <djordje.todorovic(a)syrmia.com> Reland "[LICM] Hoist LOAD without sinking the STORE" the following benchmarks slowed down by more than 2%: - 458.sjeng slowed down by 5% from 13781 to 14482 perf samples Below reproducer instructions can be used to re-build both "first_bad" and "last_good" cross-toolchains used in this bisection. Naturally, the scripts will fail when triggerring benchmarking jobs if you don't have access to Linaro TCWG CI. For your convenience, we have uploaded tarballs with pre-processed source and assembly files at: - First_bad save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… - Last_good save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… - Baseline save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Configuration: - Benchmark: SPEC CPU2006 - Toolchain: Clang + Glibc + LLVM Linker - Version: all components were built from their tip of trunk - Target: aarch64-linux-gnu - Compiler flags: -O3 -flto - Hardware: NVidia TX1 4x Cortex-A57 This benchmarking CI is work-in-progress, and we welcome feedback and suggestions at linaro-toolchain(a)lists.linaro.org . In our improvement plans is to add support for SPEC CPU2017 benchmarks and provide "perf report/annotate" data behind these reports. THIS IS THE END OF INTERESTING STUFF. BELOW ARE LINKS TO BUILDS, REPRODUCTION INSTRUCTIONS, AND THE RAW COMMIT. This commit has regressed these CI configurations: - tcwg_bmk_llvm_tx1/llvm-master-aarch64-spec2k6-O3_LTO First_bad build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Last_good build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Baseline build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Even more details: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Reproduce builds: <cut> mkdir investigate-llvm-2cdc6f2ca62e83fec445114fbbe6276e9ab2a7d0 cd investigate-llvm-2cdc6f2ca62e83fec445114fbbe6276e9ab2a7d0 # Fetch scripts git clone https://git.linaro.org/toolchain/jenkins-scripts # Fetch manifests and test.sh script mkdir -p artifacts/manifests curl -o artifacts/manifests/build-baseline.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail curl -o artifacts/manifests/build-parameters.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail curl -o artifacts/test.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail chmod +x artifacts/test.sh # Reproduce the baseline build (build all pre-requisites) ./jenkins-scripts/tcwg_bmk-build.sh @@ artifacts/manifests/build-baseline.sh # Save baseline build state (which is then restored in artifacts/test.sh) mkdir -p ./bisect rsync -a --del --delete-excluded --exclude /bisect/ --exclude /artifacts/ --exclude /llvm/ ./ ./bisect/baseline/ cd llvm # Reproduce first_bad build git checkout --detach 2cdc6f2ca62e83fec445114fbbe6276e9ab2a7d0 ../artifacts/test.sh # Reproduce last_good build git checkout --detach 47616c8855fd44abcbd7cad3f7d8153d28db347b ../artifacts/test.sh cd .. </cut> Full commit (up to 1000 lines): <cut> commit 2cdc6f2ca62e83fec445114fbbe6276e9ab2a7d0 Author: Djordje Todorovic <djordje.todorovic(a)syrmia.com> Date: Thu Dec 2 03:40:00 2021 -0800 Reland "[LICM] Hoist LOAD without sinking the STORE" When doing load/store promotion within LICM, if we cannot prove that it is safe to sink the store we won't hoist the load, even though we can prove the load could be dereferenced and moved outside the loop. This patch implements the load promotion by moving it in the loop preheader by inserting proper PHI in the loop. The store is kept as is in the loop. By doing this, we avoid doing the load from a memory location in each iteration. Please consider this small example: loop { var = *ptr; if (var) break; *ptr= var + 1; } After this patch, it will be: var0 = *ptr; loop { var1 = phi (var0, var2); if (var1) break; var2 = var1 + 1; *ptr = var2; } This addresses some problems from [0]. [0] https://bugs.llvm.org/show_bug.cgi?id=51193 Differential revision: https://reviews.llvm.org/D113289 --- llvm/include/llvm/Transforms/Utils/SSAUpdater.h | 4 +++ llvm/lib/Transforms/Scalar/LICM.cpp | 41 +++++++++++++++++----- llvm/lib/Transforms/Utils/SSAUpdater.cpp | 3 ++ .../Transforms/InstMerge/st_sink_bugfix_22613.ll | 6 ++-- .../Transforms/LICM/hoist-load-without-store.ll | 5 +-- llvm/test/Transforms/LICM/promote-capture.ll | 8 +++-- .../Transforms/LICM/scalar-promote-memmodel.ll | 8 +++-- .../Transforms/LICM/scalar-promote-opaque-ptrs.ll | 8 +++-- llvm/test/Transforms/LICM/scalar-promote.ll | 8 +++-- 9 files changed, 65 insertions(+), 26 deletions(-) diff --git a/llvm/include/llvm/Transforms/Utils/SSAUpdater.h b/llvm/include/llvm/Transforms/Utils/SSAUpdater.h index 22b2295cc9d7..c233e3dc168e 100644 --- a/llvm/include/llvm/Transforms/Utils/SSAUpdater.h +++ b/llvm/include/llvm/Transforms/Utils/SSAUpdater.h @@ -169,6 +169,10 @@ public: /// Called to update debug info associated with the instruction. virtual void updateDebugInfo(Instruction *I) const {} + + /// Return false if a sub-class wants to keep one of the loads/stores + /// after the SSA construction. + virtual bool shouldDelete(Instruction *I) const { return true; } }; } // end namespace llvm diff --git a/llvm/lib/Transforms/Scalar/LICM.cpp b/llvm/lib/Transforms/Scalar/LICM.cpp index 0d52448efb2b..6f97f3e93123 100644 --- a/llvm/lib/Transforms/Scalar/LICM.cpp +++ b/llvm/lib/Transforms/Scalar/LICM.cpp @@ -1860,6 +1860,7 @@ class LoopPromoter : public LoadAndStorePromoter { bool UnorderedAtomic; AAMDNodes AATags; ICFLoopSafetyInfo &SafetyInfo; + bool CanInsertStoresInExitBlocks; // We're about to add a use of V in a loop exit block. Insert an LCSSA phi // (if legal) if doing so would add an out-of-loop use to an instruction @@ -1886,12 +1887,13 @@ public: SmallVectorImpl<MemoryAccess *> &MSSAIP, PredIteratorCache &PIC, MemorySSAUpdater *MSSAU, LoopInfo &li, DebugLoc dl, Align Alignment, bool UnorderedAtomic, const AAMDNodes &AATags, - ICFLoopSafetyInfo &SafetyInfo) + ICFLoopSafetyInfo &SafetyInfo, bool CanInsertStoresInExitBlocks) : LoadAndStorePromoter(Insts, S), SomePtr(SP), PointerMustAliases(PMA), LoopExitBlocks(LEB), LoopInsertPts(LIP), MSSAInsertPts(MSSAIP), PredCache(PIC), MSSAU(MSSAU), LI(li), DL(std::move(dl)), Alignment(Alignment), UnorderedAtomic(UnorderedAtomic), AATags(AATags), - SafetyInfo(SafetyInfo) {} + SafetyInfo(SafetyInfo), + CanInsertStoresInExitBlocks(CanInsertStoresInExitBlocks) {} bool isInstInList(Instruction *I, const SmallVectorImpl<Instruction *> &) const override { @@ -1903,7 +1905,7 @@ public: return PointerMustAliases.count(Ptr); } - void doExtraRewritesBeforeFinalDeletion() override { + void insertStoresInLoopExitBlocks() { // Insert stores after in the loop exit blocks. Each exit block gets a // store of the live-out values that feed them. Since we've already told // the SSA updater about the defs in the loop and the preheader @@ -1937,10 +1939,21 @@ public: } } + void doExtraRewritesBeforeFinalDeletion() override { + if (CanInsertStoresInExitBlocks) + insertStoresInLoopExitBlocks(); + } + void instructionDeleted(Instruction *I) const override { SafetyInfo.removeInstruction(I); MSSAU->removeMemoryAccess(I); } + + bool shouldDelete(Instruction *I) const override { + if (isa<StoreInst>(I)) + return CanInsertStoresInExitBlocks; + return true; + } }; bool isNotCapturedBeforeOrInLoop(const Value *V, const Loop *L, @@ -2039,6 +2052,7 @@ bool llvm::promoteLoopAccessesToScalars( bool DereferenceableInPH = false; bool SafeToInsertStore = false; + bool FoundLoadToPromote = false; SmallVector<Instruction *, 64> LoopUses; @@ -2086,6 +2100,7 @@ bool llvm::promoteLoopAccessesToScalars( SawUnorderedAtomic |= Load->isAtomic(); SawNotAtomic |= !Load->isAtomic(); + FoundLoadToPromote = true; Align InstAlignment = Load->getAlign(); @@ -2197,13 +2212,20 @@ bool llvm::promoteLoopAccessesToScalars( } } - // If we've still failed to prove we can sink the store, give up. - if (!SafeToInsertStore) + // If we've still failed to prove we can sink the store, hoist the load + // only, if possible. + if (!SafeToInsertStore && !FoundLoadToPromote) + // If we cannot hoist the load either, give up. return false; - // Otherwise, this is safe to promote, lets do it! - LLVM_DEBUG(dbgs() << "LICM: Promoting value stored to in loop: " << *SomePtr - << '\n'); + // Lets do the promotion! + if (SafeToInsertStore) + LLVM_DEBUG(dbgs() << "LICM: Promoting load/store of the value: " << *SomePtr + << '\n'); + else + LLVM_DEBUG(dbgs() << "LICM: Promoting load of the value: " << *SomePtr + << '\n'); + ORE->emit([&]() { return OptimizationRemark(DEBUG_TYPE, "PromoteLoopAccessesToScalar", LoopUses[0]) @@ -2222,7 +2244,8 @@ bool llvm::promoteLoopAccessesToScalars( SSAUpdater SSA(&NewPHIs); LoopPromoter Promoter(SomePtr, LoopUses, SSA, PointerMustAliases, ExitBlocks, InsertPts, MSSAInsertPts, PIC, MSSAU, *LI, DL, - Alignment, SawUnorderedAtomic, AATags, *SafetyInfo); + Alignment, SawUnorderedAtomic, AATags, *SafetyInfo, + SafeToInsertStore); // Set up the preheader to have a definition of the value. It is the live-out // value from the preheader that uses in the loop will use. diff --git a/llvm/lib/Transforms/Utils/SSAUpdater.cpp b/llvm/lib/Transforms/Utils/SSAUpdater.cpp index 5893ce15b129..7d9992176658 100644 --- a/llvm/lib/Transforms/Utils/SSAUpdater.cpp +++ b/llvm/lib/Transforms/Utils/SSAUpdater.cpp @@ -446,6 +446,9 @@ void LoadAndStorePromoter::run(const SmallVectorImpl<Instruction *> &Insts) { // Now that everything is rewritten, delete the old instructions from the // function. They should all be dead now. for (Instruction *User : Insts) { + if (!shouldDelete(User)) + continue; + // If this is a load that still has uses, then the load must have been added // as a live value in the SSAUpdate data structure for a block (e.g. because // the loaded value was stored later). In this case, we need to recursively diff --git a/llvm/test/Transforms/InstMerge/st_sink_bugfix_22613.ll b/llvm/test/Transforms/InstMerge/st_sink_bugfix_22613.ll index 48882eca44cc..e5a75cca8ee7 100644 --- a/llvm/test/Transforms/InstMerge/st_sink_bugfix_22613.ll +++ b/llvm/test/Transforms/InstMerge/st_sink_bugfix_22613.ll @@ -5,12 +5,12 @@ target triple = "x86_64-unknown-linux-gnu" ; RUN: opt -O2 -S < %s | FileCheck %s ; CHECK-LABEL: main -; CHECK: if.end -; CHECK: store ; CHECK: memset ; CHECK: if.then ; CHECK: store -; CHECK: memset +; CHECK: if.end +; CHECK: store +; CHECK: store @d = common global i32 0, align 4 @b = common global i32 0, align 4 diff --git a/llvm/test/Transforms/LICM/hoist-load-without-store.ll b/llvm/test/Transforms/LICM/hoist-load-without-store.ll index b464f6b7328d..275a53172737 100644 --- a/llvm/test/Transforms/LICM/hoist-load-without-store.ll +++ b/llvm/test/Transforms/LICM/hoist-load-without-store.ll @@ -18,10 +18,11 @@ define dso_local void @f(i32* nocapture %ptr, i32 %n) { ; CHECK-NEXT: [[CMP7:%.*]] = icmp slt i32 0, [[N:%.*]] ; CHECK-NEXT: br i1 [[CMP7]], label [[FOR_BODY_LR_PH:%.*]], label [[CLEANUP1:%.*]] ; CHECK: for.body.lr.ph: +; CHECK-NEXT: [[PTR_PROMOTED:%.*]] = load i32, i32* [[PTR:%.*]], align 4 ; CHECK-NEXT: br label [[FOR_BODY:%.*]] ; CHECK: for.body: -; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[IF_END:%.*]] ] -; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[PTR:%.*]], align 4 +; CHECK-NEXT: [[TMP0:%.*]] = phi i32 [ [[PTR_PROMOTED]], [[FOR_BODY_LR_PH]] ], [ 1, [[IF_END:%.*]] ] +; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[IF_END]] ] ; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[TMP0]], 0 ; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label [[IF_END]], label [[FOR_BODY_CLEANUP1_CRIT_EDGE:%.*]] ; CHECK: if.end: diff --git a/llvm/test/Transforms/LICM/promote-capture.ll b/llvm/test/Transforms/LICM/promote-capture.ll index 1a2603d1c986..945036e6e175 100644 --- a/llvm/test/Transforms/LICM/promote-capture.ll +++ b/llvm/test/Transforms/LICM/promote-capture.ll @@ -111,17 +111,19 @@ define void @test_captured_before_loop(i32 %len) { ; CHECK-NEXT: [[COUNT:%.*]] = alloca i32, align 4 ; CHECK-NEXT: store i32 0, i32* [[COUNT]], align 4 ; CHECK-NEXT: call void @capture(i32* [[COUNT]]) +; CHECK-NEXT: [[COUNT_PROMOTED:%.*]] = load i32, i32* [[COUNT]], align 4 ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: loop: -; CHECK-NEXT: [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LATCH:%.*]] ] +; CHECK-NEXT: [[C_INC2:%.*]] = phi i32 [ [[COUNT_PROMOTED]], [[ENTRY:%.*]] ], [ [[C_INC1:%.*]], [[LATCH:%.*]] ] +; CHECK-NEXT: [[I:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[I_NEXT:%.*]], [[LATCH]] ] ; CHECK-NEXT: [[COND:%.*]] = call i1 @cond(i32 [[I]]) ; CHECK-NEXT: br i1 [[COND]], label [[IF:%.*]], label [[LATCH]] ; CHECK: if: -; CHECK-NEXT: [[C:%.*]] = load i32, i32* [[COUNT]], align 4 -; CHECK-NEXT: [[C_INC:%.*]] = add i32 [[C]], 1 +; CHECK-NEXT: [[C_INC:%.*]] = add i32 [[C_INC2]], 1 ; CHECK-NEXT: store i32 [[C_INC]], i32* [[COUNT]], align 4 ; CHECK-NEXT: br label [[LATCH]] ; CHECK: latch: +; CHECK-NEXT: [[C_INC1]] = phi i32 [ [[C_INC]], [[IF]] ], [ [[C_INC2]], [[LOOP]] ] ; CHECK-NEXT: [[I_NEXT]] = add nuw i32 [[I]], 1 ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[I_NEXT]], [[LEN:%.*]] ; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]] diff --git a/llvm/test/Transforms/LICM/scalar-promote-memmodel.ll b/llvm/test/Transforms/LICM/scalar-promote-memmodel.ll index c3bae731fb6b..33076b39e908 100644 --- a/llvm/test/Transforms/LICM/scalar-promote-memmodel.ll +++ b/llvm/test/Transforms/LICM/scalar-promote-memmodel.ll @@ -11,19 +11,21 @@ define void @bar(i32 %n, i32 %b) nounwind uwtable ssp { ; CHECK-LABEL: @bar( ; CHECK-NEXT: entry: ; CHECK-NEXT: [[TOBOOL:%.*]] = icmp eq i32 [[B:%.*]], 0 +; CHECK-NEXT: [[G_PROMOTED:%.*]] = load i32, i32* @g, align 4 ; CHECK-NEXT: br label [[FOR_COND:%.*]] ; CHECK: for.cond: -; CHECK-NEXT: [[I_0:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC5:%.*]], [[FOR_INC:%.*]] ] +; CHECK-NEXT: [[INC2:%.*]] = phi i32 [ [[G_PROMOTED]], [[ENTRY:%.*]] ], [ [[INC1:%.*]], [[FOR_INC:%.*]] ] +; CHECK-NEXT: [[I_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INC5:%.*]], [[FOR_INC]] ] ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[I_0]], [[N:%.*]] ; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]] ; CHECK: for.body: ; CHECK-NEXT: br i1 [[TOBOOL]], label [[FOR_INC]], label [[IF_THEN:%.*]] ; CHECK: if.then: -; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* @g, align 4 -; CHECK-NEXT: [[INC:%.*]] = add nsw i32 [[TMP3]], 1 +; CHECK-NEXT: [[INC:%.*]] = add nsw i32 [[INC2]], 1 ; CHECK-NEXT: store i32 [[INC]], i32* @g, align 4 ; CHECK-NEXT: br label [[FOR_INC]] ; CHECK: for.inc: +; CHECK-NEXT: [[INC1]] = phi i32 [ [[INC]], [[IF_THEN]] ], [ [[INC2]], [[FOR_BODY]] ] ; CHECK-NEXT: [[INC5]] = add nsw i32 [[I_0]], 1 ; CHECK-NEXT: br label [[FOR_COND]] ; CHECK: for.end: diff --git a/llvm/test/Transforms/LICM/scalar-promote-opaque-ptrs.ll b/llvm/test/Transforms/LICM/scalar-promote-opaque-ptrs.ll index da4bae936dc1..b239b6fb0296 100644 --- a/llvm/test/Transforms/LICM/scalar-promote-opaque-ptrs.ll +++ b/llvm/test/Transforms/LICM/scalar-promote-opaque-ptrs.ll @@ -314,17 +314,19 @@ define i32 @test7bad() { ; CHECK-NEXT: entry: ; CHECK-NEXT: [[LOCAL:%.*]] = alloca i32, align 4 ; CHECK-NEXT: call void @capture(ptr [[LOCAL]]) +; CHECK-NEXT: [[LOCAL_PROMOTED:%.*]] = load i32, ptr [[LOCAL]], align 4 ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: loop: -; CHECK-NEXT: [[J:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[NEXT:%.*]], [[ELSE:%.*]] ] -; CHECK-NEXT: [[X:%.*]] = load i32, ptr [[LOCAL]], align 4 -; CHECK-NEXT: [[X2:%.*]] = call i32 @opaque(i32 [[X]]) +; CHECK-NEXT: [[X22:%.*]] = phi i32 [ [[LOCAL_PROMOTED]], [[ENTRY:%.*]] ], [ [[X21:%.*]], [[ELSE:%.*]] ] +; CHECK-NEXT: [[J:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[NEXT:%.*]], [[ELSE]] ] +; CHECK-NEXT: [[X2:%.*]] = call i32 @opaque(i32 [[X22]]) ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[X2]], 0 ; CHECK-NEXT: br i1 [[CMP]], label [[IF:%.*]], label [[ELSE]] ; CHECK: if: ; CHECK-NEXT: store i32 [[X2]], ptr [[LOCAL]], align 4 ; CHECK-NEXT: br label [[ELSE]] ; CHECK: else: +; CHECK-NEXT: [[X21]] = phi i32 [ [[X2]], [[IF]] ], [ [[X22]], [[LOOP]] ] ; CHECK-NEXT: [[NEXT]] = add i32 [[J]], 1 ; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[NEXT]], 0 ; CHECK-NEXT: br i1 [[COND]], label [[EXIT:%.*]], label [[LOOP]] diff --git a/llvm/test/Transforms/LICM/scalar-promote.ll b/llvm/test/Transforms/LICM/scalar-promote.ll index 290e990f8513..c064edb8cd93 100644 --- a/llvm/test/Transforms/LICM/scalar-promote.ll +++ b/llvm/test/Transforms/LICM/scalar-promote.ll @@ -315,17 +315,19 @@ define i32 @test7bad() { ; CHECK-NEXT: entry: ; CHECK-NEXT: [[LOCAL:%.*]] = alloca i32, align 4 ; CHECK-NEXT: call void @capture(i32* [[LOCAL]]) +; CHECK-NEXT: [[LOCAL_PROMOTED:%.*]] = load i32, i32* [[LOCAL]], align 4 ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: loop: -; CHECK-NEXT: [[J:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[NEXT:%.*]], [[ELSE:%.*]] ] -; CHECK-NEXT: [[X:%.*]] = load i32, i32* [[LOCAL]], align 4 -; CHECK-NEXT: [[X2:%.*]] = call i32 @opaque(i32 [[X]]) +; CHECK-NEXT: [[X22:%.*]] = phi i32 [ [[LOCAL_PROMOTED]], [[ENTRY:%.*]] ], [ [[X21:%.*]], [[ELSE:%.*]] ] +; CHECK-NEXT: [[J:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[NEXT:%.*]], [[ELSE]] ] +; CHECK-NEXT: [[X2:%.*]] = call i32 @opaque(i32 [[X22]]) ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[X2]], 0 ; CHECK-NEXT: br i1 [[CMP]], label [[IF:%.*]], label [[ELSE]] ; CHECK: if: ; CHECK-NEXT: store i32 [[X2]], i32* [[LOCAL]], align 4 ; CHECK-NEXT: br label [[ELSE]] ; CHECK: else: +; CHECK-NEXT: [[X21]] = phi i32 [ [[X2]], [[IF]] ], [ [[X22]], [[LOOP]] ] ; CHECK-NEXT: [[NEXT]] = add i32 [[J]], 1 ; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[NEXT]], 0 ; CHECK-NEXT: br i1 [[COND]], label [[EXIT:%.*]], label [[LOOP]] </cut>

2 years, 11 months

[ACTIVITY] week ending Dec. 5 2021

by Alex Bennée

VirtIO Initiative ([STR-9]) =========================== - synced up on [AX_XDP task with Akashi-san] - synced on rust-vmm [AX_XDP task with Akashi-san] <https://linaro.atlassian.net/browse/STR-68> vhost-device maintainer effort ([UM-196]) - started looking at https://github.com/rust-vmm/vhost-device/pull/4 QEMU Upstream Work ([UM-2]) =========================== - posted [PULL for 6.2 0/8] more tcg, plugin, test and build fixes Message-Id: <20211129171449.4176301-1-alex.bennee(a)linaro.org> - commented on Re: Follow-up on the CXL discussion at OFTC Message-Id: <20211119015207.62fhk5mjmvaj5nz4(a)intel.com> to see if I can unblock - posted [RFC PATCH] blog post: how to get your new feature up-streamed Message-Id: <20211126203319.3298089-1-alex.bennee(a)linaro.org> - posted [PATCH for 6.2?] Revert "vga: don't abort when adding a duplicate isa-vga device" Message-Id: <20211202164929.1119036-1-alex.bennee(a)linaro.org> Upstream MTTCG tests ([QEMU-52]) - posted [kvm-unit-tests PATCH v9 0/9] MTTCG sanity tests for ARM Message-Id: <20211202115352.951548-1-alex.bennee(a)linaro.org> [QEMU-52] <https://linaro.atlassian.net/browse/QEMU-52> Other ===== - wrote [RFC PATCH 0/2] insn plugin tweaks for measuring frequency Message-Id: <20211203144421.1445232-1-alex.bennee(a)linaro.org> - might make a good basis for a TCG plugins blog post Completed Reviews [2/2] ======================= [PATCH] tests/plugin/syscall.c: fix compiler warnings Message-Id: <20211128011551.2115468-1-juro.bystricky(a)intel.com> [PATCH for-6.2? 0/2] arm_gicv3: Fix handling of LPIs in list registers Message-Id: <20211126163915.1048353-2-peter.maydell(a)linaro.org> Current Review Queue ==================== TODO [PATCH-4.16 v2] xen/efi: Fix Grub2 boot on arm64 Message-Id: <20211104141206.25153-1-luca.fancellu(a)arm.com> =============================================================================================================== TODO [PATCH] cpu-models-x86.rst: Tidy up a couple of things Message-Id: <20211015100718.17828-1-pbonzini(a)redhat.com> =================================================================================================================== TODO [PATCH 00/16] fdt: Make OF_BOARD a boolean option Message-Id: <20211013010120.96851-1-sjg(a)chromium.org> =========================================================================================================== TODO [PATCH v4 00/41] linux-user: Streamline handling of SIGSEGV Message-Id: <20211006172307.780893-1-richard.henderson(a)linaro.org> ================================================================================================================================== -- Alex Bennée

2 years, 11 months

[ACTIVITY] report week ending 3 Dec

by Peter Maydell

Progress: * UM-2 [QEMU upstream maintainership] - Code review: worked through some of the backlog and accumulated a list of series to take once the tree reopens for 7.0 - Wrote and sent some cleanup patches relating to the qemu-common.h header file - Fixed a bug where we miscalculated the length for TLB range invalidations * QEMU-420 [GICv4 emulation] - Found the problem with PCI passthrough in my nested test setup: apparently virtio PCI devices need an extra command line argument to get them to honour the presence of an IOMMU. Everything is now working and I've put some notes about the setup into https://linaro.atlassian.net/browse/QEMU-447 - started to implement the GICv4 redistributor changes -- PMM

2 years, 11 months

[ACTIVITY] week ending Nov. 28 2021

by Alex Bennée

VirtIO Initiative ([STR-9]) =========================== - [this weeks sync], topics on AF_XDP, virtio-video and virtio-watchdog [upstream rust-vmm sync meeting] <https://etherpad.opendev.org/p/rust-vmm-sync-2021&sa=D&source=calendar&ust=…> QEMU Upstream Work ([UM-2]) =========================== - posted [PATCH for 6.2 v2 0/7] more tcg, plugin, test and build fixes Message-Id: <20211125154144.2904741-1-alex.bennee(a)linaro.org> Upstream MTTCG tests ([QEMU-52]) - posted [kvm-unit-tests PATCH v8 00/10] MTTCG sanity tests for ARM Message-Id: <20211118184650.661575-1-alex.bennee(a)linaro.org> [mttcg tests to current state and fixed up] <https://github.com/stsquad/qemu/tree/mttcg/current-tests-v8> Other ===== - renewal feedback Completed Reviews [2/2] ======================= [PATCH v2 0/3] KVM: qemu patches for few KVM features I developed Message-Id: <20211101132300.192584-1-mlevitsk(a)redhat.com> [PATCH v2] hw/intc/arm_gicv3: Update cached state after LPI state changes Message-Id: <20211124202005.989935-1-peter.maydell(a)linaro.org> Absences ======== - off 2 days sick Current Review Queue ==================== TODO [PATCH-4.16 v2] xen/efi: Fix Grub2 boot on arm64 Message-Id: <20211104141206.25153-1-luca.fancellu(a)arm.com> =============================================================================================================== TODO [PATCH] cpu-models-x86.rst: Tidy up a couple of things Message-Id: <20211015100718.17828-1-pbonzini(a)redhat.com> =================================================================================================================== TODO [PATCH 00/16] fdt: Make OF_BOARD a boolean option Message-Id: <20211013010120.96851-1-sjg(a)chromium.org> =========================================================================================================== TODO [PATCH v4 00/41] linux-user: Streamline handling of SIGSEGV Message-Id: <20211006172307.780893-1-richard.henderson(a)linaro.org> ================================================================================================================================== -- Alex Bennée

2 years, 12 months

[ACTIVITY] report week ending 26 Nov

by Peter Maydell

Progress: * QEMU-420 [GICv4 emulation] - Tracked down and fixed a bug in our ITS emulation which would (intermittently?) result in a Linux guest reporting "irq 54: nobody cared" and hanging, because we were not correctly recalculating the highest priority pending interrupt when the guest acknowledged a pending LPI. This fix will go into 6.2. - Set up a test environment for GICv4 work -- because the major feature of GICv4 is support for directly injecting interrupts into a VM, the test setup needs to be nested virtualization, where an outer L1 guest runs on pure emulated QEMU, the inner L2 guest uses KVM (as provided by L1), and we pass a PCI device (emulated by QEMU) through from L1 to L2. I think I have this correctly set up now, but... - ...the L2 guest hangs because it apparently never sees an interrupt from the passed-through PCI device. This implies a bug in our current GICv3 emulation somewhere: need to track this down before starting in on GICv4 work. - Separately, I found through code inspection a bug where we do the wrong thing in the non-passthrough case when the L1 guest sets a virtual interrupt for the L2 guest in the GIC list registers and that interrupt has an ID > 1023 (ie it is an LPI). We got this wrong both for acknowledging and ending an interrupt, so the two bugs cancel each other out except that we don't set the vCPU priority and so the L2 guest might get an unexpected interrupt while it was servicing the LPI. Patches sent. -- PMM

2 years, 12 months

[TCWG CI] 433.milc slowed down by 5% after llvm: [AMDGPU] Implement widening multiplies with v_mad_i64_i32/v_mad_u64_u32

by ci_notify＠linaro.org

After llvm commit d7e03df719464354b20a845b7853be57da863924 Author: Jay Foad <jay.foad(a)amd.com> [AMDGPU] Implement widening multiplies with v_mad_i64_i32/v_mad_u64_u32 the following benchmarks slowed down by more than 2%: - 433.milc slowed down by 5% from 12335 to 12997 perf samples Below reproducer instructions can be used to re-build both "first_bad" and "last_good" cross-toolchains used in this bisection. Naturally, the scripts will fail when triggerring benchmarking jobs if you don't have access to Linaro TCWG CI. For your convenience, we have uploaded tarballs with pre-processed source and assembly files at: - First_bad save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… - Last_good save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… - Baseline save-temps: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Configuration: - Benchmark: SPEC CPU2006 - Toolchain: Clang + Glibc + LLVM Linker - Version: all components were built from their tip of trunk - Target: aarch64-linux-gnu - Compiler flags: -O2 -flto - Hardware: NVidia TX1 4x Cortex-A57 This benchmarking CI is work-in-progress, and we welcome feedback and suggestions at linaro-toolchain(a)lists.linaro.org . In our improvement plans is to add support for SPEC CPU2017 benchmarks and provide "perf report/annotate" data behind these reports. THIS IS THE END OF INTERESTING STUFF. BELOW ARE LINKS TO BUILDS, REPRODUCTION INSTRUCTIONS, AND THE RAW COMMIT. This commit has regressed these CI configurations: - tcwg_bmk_llvm_tx1/llvm-master-aarch64-spec2k6-O2_LTO First_bad build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Last_good build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Baseline build: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Even more details: https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… Reproduce builds: <cut> mkdir investigate-llvm-d7e03df719464354b20a845b7853be57da863924 cd investigate-llvm-d7e03df719464354b20a845b7853be57da863924 # Fetch scripts git clone https://git.linaro.org/toolchain/jenkins-scripts # Fetch manifests and test.sh script mkdir -p artifacts/manifests curl -o artifacts/manifests/build-baseline.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail curl -o artifacts/manifests/build-parameters.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail curl -o artifacts/test.sh https://ci.linaro.org/job/tcwg_bmk_ci_llvm-bisect-tcwg_bmk_tx1-llvm-master-… --fail chmod +x artifacts/test.sh # Reproduce the baseline build (build all pre-requisites) ./jenkins-scripts/tcwg_bmk-build.sh @@ artifacts/manifests/build-baseline.sh # Save baseline build state (which is then restored in artifacts/test.sh) mkdir -p ./bisect rsync -a --del --delete-excluded --exclude /bisect/ --exclude /artifacts/ --exclude /llvm/ ./ ./bisect/baseline/ cd llvm # Reproduce first_bad build git checkout --detach d7e03df719464354b20a845b7853be57da863924 ../artifacts/test.sh # Reproduce last_good build git checkout --detach 8a52bd82e36855b3ad842f2535d0c78a97db55dc ../artifacts/test.sh cd .. </cut> Full commit (up to 1000 lines): <cut> commit d7e03df719464354b20a845b7853be57da863924 Author: Jay Foad <jay.foad(a)amd.com> Date: Fri Nov 12 18:02:58 2021 +0000 [AMDGPU] Implement widening multiplies with v_mad_i64_i32/v_mad_u64_u32 Select SelectionDAG ops smul_lohi/umul_lohi to v_mad_i64_i32/v_mad_u64_u32 respectively, with an addend of 0. v_mul_lo, v_mul_hi and v_mad_i64/u64 are all quarter-rate instructions so it is better to use one instruction than two. Further improvements are possible to make better use of the addend operand, but this is already a strict improvement over what we have now. Differential Revision: https://reviews.llvm.org/D113986 --- llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp | 29 + llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h | 1 + llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp | 49 + llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h | 1 + llvm/lib/Target/AMDGPU/SIISelLowering.cpp | 23 + llvm/lib/Target/AMDGPU/SIISelLowering.h | 1 + .../AMDGPU/atomic_optimizations_global_pointer.ll | 104 +- .../AMDGPU/atomic_optimizations_local_pointer.ll | 108 +- llvm/test/CodeGen/AMDGPU/bypass-div.ll | 1064 +++++++++----------- llvm/test/CodeGen/AMDGPU/llvm.mulo.ll | 178 ++-- llvm/test/CodeGen/AMDGPU/mad_64_32.ll | 110 +- llvm/test/CodeGen/AMDGPU/mul.ll | 55 +- llvm/test/CodeGen/AMDGPU/mul_int24.ll | 9 +- llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll | 24 +- llvm/test/CodeGen/AMDGPU/udiv.ll | 358 +++---- llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll | 126 +-- llvm/test/CodeGen/AMDGPU/wwm-reserved.ll | 16 +- 17 files changed, 1126 insertions(+), 1130 deletions(-) diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp index 2e571ad01c1c..8236e6672247 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp @@ -654,6 +654,9 @@ void AMDGPUDAGToDAGISel::Select(SDNode *N) { SelectMAD_64_32(N); return; } + case ISD::SMUL_LOHI: + case ISD::UMUL_LOHI: + return SelectMUL_LOHI(N); case ISD::CopyToReg: { const SITargetLowering& Lowering = *static_cast<const SITargetLowering*>(getTargetLowering()); @@ -1013,6 +1016,32 @@ void AMDGPUDAGToDAGISel::SelectMAD_64_32(SDNode *N) { CurDAG->SelectNodeTo(N, Opc, N->getVTList(), Ops); } +// We need to handle this here because tablegen doesn't support matching +// instructions with multiple outputs. +void AMDGPUDAGToDAGISel::SelectMUL_LOHI(SDNode *N) { + SDLoc SL(N); + bool Signed = N->getOpcode() == ISD::SMUL_LOHI; + unsigned Opc = Signed ? AMDGPU::V_MAD_I64_I32_e64 : AMDGPU::V_MAD_U64_U32_e64; + + SDValue Zero = CurDAG->getTargetConstant(0, SL, MVT::i64); + SDValue Clamp = CurDAG->getTargetConstant(0, SL, MVT::i1); + SDValue Ops[] = {N->getOperand(0), N->getOperand(1), Zero, Clamp}; + SDNode *Mad = CurDAG->getMachineNode(Opc, SL, N->getVTList(), Ops); + if (!SDValue(N, 0).use_empty()) { + SDValue Sub0 = CurDAG->getTargetConstant(AMDGPU::sub0, SL, MVT::i32); + SDNode *Lo = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG, SL, + MVT::i32, SDValue(Mad, 0), Sub0); + ReplaceUses(SDValue(N, 0), SDValue(Lo, 0)); + } + if (!SDValue(N, 1).use_empty()) { + SDValue Sub1 = CurDAG->getTargetConstant(AMDGPU::sub1, SL, MVT::i32); + SDNode *Hi = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG, SL, + MVT::i32, SDValue(Mad, 0), Sub1); + ReplaceUses(SDValue(N, 1), SDValue(Hi, 0)); + } + CurDAG->RemoveDeadNode(N); +} + bool AMDGPUDAGToDAGISel::isDSOffsetLegal(SDValue Base, unsigned Offset) const { if (!isUInt<16>(Offset)) return false; diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h index 74aff9e406c9..d638d9877a9b 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h +++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h @@ -235,6 +235,7 @@ private: void SelectUADDO_USUBO(SDNode *N); void SelectDIV_SCALE(SDNode *N); void SelectMAD_64_32(SDNode *N); + void SelectMUL_LOHI(SDNode *N); void SelectFMA_W_CHAIN(SDNode *N); void SelectFMUL_W_CHAIN(SDNode *N); SDNode *getBFE32(bool IsSigned, const SDLoc &DL, SDValue Val, uint32_t Offset, diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp index 523fa2d3724b..54177564afbc 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp @@ -594,6 +594,8 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(const TargetMachine &TM, setTargetDAGCombine(ISD::SRL); setTargetDAGCombine(ISD::TRUNCATE); setTargetDAGCombine(ISD::MUL); + setTargetDAGCombine(ISD::SMUL_LOHI); + setTargetDAGCombine(ISD::UMUL_LOHI); setTargetDAGCombine(ISD::MULHU); setTargetDAGCombine(ISD::MULHS); setTargetDAGCombine(ISD::SELECT); @@ -3462,6 +3464,50 @@ SDValue AMDGPUTargetLowering::performMulCombine(SDNode *N, return DAG.getSExtOrTrunc(Mul, DL, VT); } +SDValue +AMDGPUTargetLowering::performMulLoHiCombine(SDNode *N, + DAGCombinerInfo &DCI) const { + if (N->getValueType(0) != MVT::i32) + return SDValue(); + + SelectionDAG &DAG = DCI.DAG; + SDLoc DL(N); + + SDValue N0 = N->getOperand(0); + SDValue N1 = N->getOperand(1); + + // SimplifyDemandedBits has the annoying habit of turning useful zero_extends + // in the source into any_extends if the result of the mul is truncated. Since + // we can assume the high bits are whatever we want, use the underlying value + // to avoid the unknown high bits from interfering. + if (N0.getOpcode() == ISD::ANY_EXTEND) + N0 = N0.getOperand(0); + if (N1.getOpcode() == ISD::ANY_EXTEND) + N1 = N1.getOperand(0); + + // Try to use two fast 24-bit multiplies (one for each half of the result) + // instead of one slow extending multiply. + unsigned LoOpcode, HiOpcode; + if (Subtarget->hasMulU24() && isU24(N0, DAG) && isU24(N1, DAG)) { + N0 = DAG.getZExtOrTrunc(N0, DL, MVT::i32); + N1 = DAG.getZExtOrTrunc(N1, DL, MVT::i32); + LoOpcode = AMDGPUISD::MUL_U24; + HiOpcode = AMDGPUISD::MULHI_U24; + } else if (Subtarget->hasMulI24() && isI24(N0, DAG) && isI24(N1, DAG)) { + N0 = DAG.getSExtOrTrunc(N0, DL, MVT::i32); + N1 = DAG.getSExtOrTrunc(N1, DL, MVT::i32); + LoOpcode = AMDGPUISD::MUL_I24; + HiOpcode = AMDGPUISD::MULHI_I24; + } else { + return SDValue(); + } + + SDValue Lo = DAG.getNode(LoOpcode, DL, MVT::i32, N0, N1); + SDValue Hi = DAG.getNode(HiOpcode, DL, MVT::i32, N0, N1); + DCI.CombineTo(N, Lo, Hi); + return SDValue(N, 0); +} + SDValue AMDGPUTargetLowering::performMulhsCombine(SDNode *N, DAGCombinerInfo &DCI) const { EVT VT = N->getValueType(0); @@ -4103,6 +4149,9 @@ SDValue AMDGPUTargetLowering::PerformDAGCombine(SDNode *N, return performTruncateCombine(N, DCI); case ISD::MUL: return performMulCombine(N, DCI); + case ISD::SMUL_LOHI: + case ISD::UMUL_LOHI: + return performMulLoHiCombine(N, DCI); case ISD::MULHS: return performMulhsCombine(N, DCI); case ISD::MULHU: diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h index 03632ac18598..daaca8737c5d 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h +++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h @@ -91,6 +91,7 @@ protected: SDValue performSrlCombine(SDNode *N, DAGCombinerInfo &DCI) const; SDValue performTruncateCombine(SDNode *N, DAGCombinerInfo &DCI) const; SDValue performMulCombine(SDNode *N, DAGCombinerInfo &DCI) const; + SDValue performMulLoHiCombine(SDNode *N, DAGCombinerInfo &DCI) const; SDValue performMulhsCombine(SDNode *N, DAGCombinerInfo &DCI) const; SDValue performMulhuCombine(SDNode *N, DAGCombinerInfo &DCI) const; SDValue performCtlz_CttzCombine(const SDLoc &SL, SDValue Cond, SDValue LHS, diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp index 519c5b936536..02440044d6e2 100644 --- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp +++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp @@ -809,6 +809,11 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM, setOperationAction(ISD::SMULO, MVT::i64, Custom); setOperationAction(ISD::UMULO, MVT::i64, Custom); + if (Subtarget->hasMad64_32()) { + setOperationAction(ISD::SMUL_LOHI, MVT::i32, Custom); + setOperationAction(ISD::UMUL_LOHI, MVT::i32, Custom); + } + setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom); setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::f32, Custom); setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::v4f32, Custom); @@ -4691,6 +4696,9 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { case ISD::SMULO: case ISD::UMULO: return lowerXMULO(Op, DAG); + case ISD::SMUL_LOHI: + case ISD::UMUL_LOHI: + return lowerXMUL_LOHI(Op, DAG); case ISD::DYNAMIC_STACKALLOC: return LowerDYNAMIC_STACKALLOC(Op, DAG); } @@ -5304,6 +5312,21 @@ SDValue SITargetLowering::lowerXMULO(SDValue Op, SelectionDAG &DAG) const { return DAG.getMergeValues({ Result, Overflow }, SL); } +SDValue SITargetLowering::lowerXMUL_LOHI(SDValue Op, SelectionDAG &DAG) const { + if (Op->isDivergent()) { + // Select to V_MAD_[IU]64_[IU]32. + return Op; + } + if (Subtarget->hasSMulHi()) { + // Expand to S_MUL_I32 + S_MUL_HI_[IU]32. + return SDValue(); + } + // The multiply is uniform but we would have to use V_MUL_HI_[IU]32 to + // calculate the high part, so we might as well do the whole thing with + // V_MAD_[IU]64_[IU]32. + return Op; +} + SDValue SITargetLowering::lowerTRAP(SDValue Op, SelectionDAG &DAG) const { if (!Subtarget->isTrapHandlerEnabled() || Subtarget->getTrapHandlerAbi() != GCNSubtarget::TrapHandlerAbi::AMDHSA) diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.h b/llvm/lib/Target/AMDGPU/SIISelLowering.h index 1e48c96ad3c8..ea6ca3f48827 100644 --- a/llvm/lib/Target/AMDGPU/SIISelLowering.h +++ b/llvm/lib/Target/AMDGPU/SIISelLowering.h @@ -135,6 +135,7 @@ private: SDValue lowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const; SDValue lowerFMINNUM_FMAXNUM(SDValue Op, SelectionDAG &DAG) const; SDValue lowerXMULO(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerXMUL_LOHI(SDValue Op, SelectionDAG &DAG) const; SDValue getSegmentAperture(unsigned AS, const SDLoc &DL, SelectionDAG &DAG) const; diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll index 49f05fceb8ed..4ad774db6686 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll @@ -818,32 +818,29 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX8-NEXT: s_mov_b32 s12, s6 ; GFX8-NEXT: s_bcnt1_i32_b64 s6, s[8:9] ; GFX8-NEXT: v_mov_b32_e32 v0, s6 -; GFX8-NEXT: v_mul_hi_u32 v0, s0, v0 -; GFX8-NEXT: s_mov_b32 s13, s7 -; GFX8-NEXT: s_mul_i32 s7, s1, s6 -; GFX8-NEXT: s_mul_i32 s6, s0, s6 +; GFX8-NEXT: v_mad_u64_u32 v[0:1], s[8:9], s0, v0, 0 +; GFX8-NEXT: s_mul_i32 s6, s1, s6 ; GFX8-NEXT: s_mov_b32 s15, 0xf000 ; GFX8-NEXT: s_mov_b32 s14, -1 -; GFX8-NEXT: v_add_u32_e32 v1, vcc, s7, v0 -; GFX8-NEXT: v_mov_b32_e32 v0, s6 +; GFX8-NEXT: s_mov_b32 s13, s7 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, s6, v1 ; GFX8-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX8-NEXT: buffer_atomic_add_x2 v[0:1], off, s[12:15], 0 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1_vol ; GFX8-NEXT: .LBB4_2: ; GFX8-NEXT: s_or_b64 exec, exec, s[2:3] -; GFX8-NEXT: v_readfirstlane_b32 s2, v0 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_mul_lo_u32 v0, s1, v2 -; GFX8-NEXT: v_mul_hi_u32 v3, s0, v2 +; GFX8-NEXT: v_mul_lo_u32 v4, s1, v2 +; GFX8-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s0, v2, 0 +; GFX8-NEXT: v_readfirstlane_b32 s0, v0 ; GFX8-NEXT: v_readfirstlane_b32 s1, v1 -; GFX8-NEXT: v_mul_lo_u32 v1, s0, v2 -; GFX8-NEXT: s_mov_b32 s7, 0xf000 -; GFX8-NEXT: v_add_u32_e32 v2, vcc, v3, v0 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, v3, v4 ; GFX8-NEXT: v_mov_b32_e32 v3, s1 -; GFX8-NEXT: v_add_u32_e32 v0, vcc, s2, v1 +; GFX8-NEXT: v_add_u32_e32 v0, vcc, s0, v2 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 ; GFX8-NEXT: s_mov_b32 s6, -1 -; GFX8-NEXT: v_addc_u32_e32 v1, vcc, v3, v2, vcc +; GFX8-NEXT: v_addc_u32_e32 v1, vcc, v3, v1, vcc ; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX8-NEXT: s_endpgm ; @@ -878,17 +875,16 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX9-NEXT: .LBB4_2: ; GFX9-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX9-NEXT: v_mul_hi_u32 v4, s2, v2 +; GFX9-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s2, v2, 0 ; GFX9-NEXT: v_readfirstlane_b32 s0, v0 -; GFX9-NEXT: v_mul_lo_u32 v0, s2, v2 ; GFX9-NEXT: v_readfirstlane_b32 s1, v1 -; GFX9-NEXT: v_add_u32_e32 v1, v4, v3 -; GFX9-NEXT: v_mov_b32_e32 v2, s1 -; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s0, v0 +; GFX9-NEXT: v_add_u32_e32 v1, v3, v4 +; GFX9-NEXT: v_mov_b32_e32 v3, s1 +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s0, v2 ; GFX9-NEXT: s_mov_b32 s7, 0xf000 ; GFX9-NEXT: s_mov_b32 s6, -1 -; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v2, v1, vcc +; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v3, v1, vcc ; GFX9-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX9-NEXT: s_endpgm ; @@ -927,14 +923,13 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX1064-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1064-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX1064-NEXT: s_waitcnt lgkmcnt(0) -; GFX1064-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1064-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1064-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1064-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1064-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s2, v2, 0 ; GFX1064-NEXT: v_readfirstlane_b32 s0, v0 ; GFX1064-NEXT: v_readfirstlane_b32 s1, v1 ; GFX1064-NEXT: s_mov_b32 s7, 0x31016000 ; GFX1064-NEXT: s_mov_b32 s6, -1 -; GFX1064-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1064-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1064-NEXT: v_add_co_u32 v0, vcc, s0, v2 ; GFX1064-NEXT: v_add_co_ci_u32_e32 v1, vcc, s1, v1, vcc ; GFX1064-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 @@ -974,14 +969,13 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX1032-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1032-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX1032-NEXT: s_waitcnt lgkmcnt(0) -; GFX1032-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1032-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1032-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1032-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1032-NEXT: v_mad_u64_u32 v[2:3], s0, s2, v2, 0 ; GFX1032-NEXT: v_readfirstlane_b32 s0, v0 ; GFX1032-NEXT: v_readfirstlane_b32 s1, v1 ; GFX1032-NEXT: s_mov_b32 s7, 0x31016000 ; GFX1032-NEXT: s_mov_b32 s6, -1 -; GFX1032-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1032-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1032-NEXT: v_add_co_u32 v0, vcc_lo, s0, v2 ; GFX1032-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo ; GFX1032-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 @@ -1955,32 +1949,29 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX8-NEXT: s_mov_b32 s12, s6 ; GFX8-NEXT: s_bcnt1_i32_b64 s6, s[8:9] ; GFX8-NEXT: v_mov_b32_e32 v0, s6 -; GFX8-NEXT: v_mul_hi_u32 v0, s0, v0 -; GFX8-NEXT: s_mov_b32 s13, s7 -; GFX8-NEXT: s_mul_i32 s7, s1, s6 -; GFX8-NEXT: s_mul_i32 s6, s0, s6 +; GFX8-NEXT: v_mad_u64_u32 v[0:1], s[8:9], s0, v0, 0 +; GFX8-NEXT: s_mul_i32 s6, s1, s6 ; GFX8-NEXT: s_mov_b32 s15, 0xf000 ; GFX8-NEXT: s_mov_b32 s14, -1 -; GFX8-NEXT: v_add_u32_e32 v1, vcc, s7, v0 -; GFX8-NEXT: v_mov_b32_e32 v0, s6 +; GFX8-NEXT: s_mov_b32 s13, s7 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, s6, v1 ; GFX8-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX8-NEXT: buffer_atomic_sub_x2 v[0:1], off, s[12:15], 0 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1_vol ; GFX8-NEXT: .LBB10_2: ; GFX8-NEXT: s_or_b64 exec, exec, s[2:3] -; GFX8-NEXT: v_readfirstlane_b32 s2, v0 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_mul_lo_u32 v0, s1, v2 -; GFX8-NEXT: v_mul_hi_u32 v3, s0, v2 +; GFX8-NEXT: v_mul_lo_u32 v4, s1, v2 +; GFX8-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s0, v2, 0 +; GFX8-NEXT: v_readfirstlane_b32 s0, v0 ; GFX8-NEXT: v_readfirstlane_b32 s1, v1 -; GFX8-NEXT: v_mul_lo_u32 v1, s0, v2 -; GFX8-NEXT: s_mov_b32 s7, 0xf000 -; GFX8-NEXT: v_add_u32_e32 v2, vcc, v3, v0 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, v3, v4 ; GFX8-NEXT: v_mov_b32_e32 v3, s1 -; GFX8-NEXT: v_sub_u32_e32 v0, vcc, s2, v1 +; GFX8-NEXT: v_sub_u32_e32 v0, vcc, s0, v2 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 ; GFX8-NEXT: s_mov_b32 s6, -1 -; GFX8-NEXT: v_subb_u32_e32 v1, vcc, v3, v2, vcc +; GFX8-NEXT: v_subb_u32_e32 v1, vcc, v3, v1, vcc ; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX8-NEXT: s_endpgm ; @@ -2015,17 +2006,16 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX9-NEXT: .LBB10_2: ; GFX9-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX9-NEXT: v_mul_hi_u32 v4, s2, v2 +; GFX9-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s2, v2, 0 ; GFX9-NEXT: v_readfirstlane_b32 s0, v0 -; GFX9-NEXT: v_mul_lo_u32 v0, s2, v2 ; GFX9-NEXT: v_readfirstlane_b32 s1, v1 -; GFX9-NEXT: v_add_u32_e32 v1, v4, v3 -; GFX9-NEXT: v_mov_b32_e32 v2, s1 -; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, s0, v0 +; GFX9-NEXT: v_add_u32_e32 v1, v3, v4 +; GFX9-NEXT: v_mov_b32_e32 v3, s1 +; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, s0, v2 ; GFX9-NEXT: s_mov_b32 s7, 0xf000 ; GFX9-NEXT: s_mov_b32 s6, -1 -; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v2, v1, vcc +; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v1, vcc ; GFX9-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX9-NEXT: s_endpgm ; @@ -2064,14 +2054,13 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX1064-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1064-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX1064-NEXT: s_waitcnt lgkmcnt(0) -; GFX1064-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1064-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1064-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1064-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1064-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s2, v2, 0 ; GFX1064-NEXT: v_readfirstlane_b32 s0, v0 ; GFX1064-NEXT: v_readfirstlane_b32 s1, v1 ; GFX1064-NEXT: s_mov_b32 s7, 0x31016000 ; GFX1064-NEXT: s_mov_b32 s6, -1 -; GFX1064-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1064-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1064-NEXT: v_sub_co_u32 v0, vcc, s0, v2 ; GFX1064-NEXT: v_sub_co_ci_u32_e32 v1, vcc, s1, v1, vcc ; GFX1064-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 @@ -2111,14 +2100,13 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 addrspace ; GFX1032-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1032-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX1032-NEXT: s_waitcnt lgkmcnt(0) -; GFX1032-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1032-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1032-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1032-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1032-NEXT: v_mad_u64_u32 v[2:3], s0, s2, v2, 0 ; GFX1032-NEXT: v_readfirstlane_b32 s0, v0 ; GFX1032-NEXT: v_readfirstlane_b32 s1, v1 ; GFX1032-NEXT: s_mov_b32 s7, 0x31016000 ; GFX1032-NEXT: s_mov_b32 s6, -1 -; GFX1032-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1032-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1032-NEXT: v_sub_co_u32 v0, vcc_lo, s0, v2 ; GFX1032-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo ; GFX1032-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll index 455f9de836ba..bf91960537a4 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll @@ -954,15 +954,13 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 %additive ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], vcc ; GFX8-NEXT: s_cbranch_execz .LBB5_2 ; GFX8-NEXT: ; %bb.1: -; GFX8-NEXT: s_bcnt1_i32_b64 s6, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v0, s6 +; GFX8-NEXT: s_bcnt1_i32_b64 s8, s[6:7] +; GFX8-NEXT: v_mov_b32_e32 v0, s8 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_mul_hi_u32 v0, s2, v0 -; GFX8-NEXT: s_mul_i32 s7, s3, s6 -; GFX8-NEXT: s_mul_i32 s6, s2, s6 +; GFX8-NEXT: v_mad_u64_u32 v[0:1], s[6:7], s2, v0, 0 +; GFX8-NEXT: s_mul_i32 s6, s3, s8 ; GFX8-NEXT: v_mov_b32_e32 v3, 0 -; GFX8-NEXT: v_add_u32_e32 v1, vcc, s7, v0 -; GFX8-NEXT: v_mov_b32_e32 v0, s6 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, s6, v1 ; GFX8-NEXT: s_mov_b32 m0, -1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) ; GFX8-NEXT: ds_add_rtn_u64 v[0:1], v3, v[0:1] @@ -971,18 +969,17 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 %additive ; GFX8-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_waitcnt lgkmcnt(0) ; GFX8-NEXT: s_mov_b32 s4, s0 -; GFX8-NEXT: v_readfirstlane_b32 s0, v0 -; GFX8-NEXT: v_mul_lo_u32 v0, s3, v2 -; GFX8-NEXT: v_mul_hi_u32 v3, s2, v2 ; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX8-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s2, v2, 0 +; GFX8-NEXT: v_readfirstlane_b32 s0, v0 ; GFX8-NEXT: v_readfirstlane_b32 s1, v1 -; GFX8-NEXT: v_mul_lo_u32 v1, s2, v2 -; GFX8-NEXT: v_add_u32_e32 v2, vcc, v3, v0 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, v3, v4 ; GFX8-NEXT: v_mov_b32_e32 v3, s1 -; GFX8-NEXT: v_add_u32_e32 v0, vcc, s0, v1 +; GFX8-NEXT: v_add_u32_e32 v0, vcc, s0, v2 ; GFX8-NEXT: s_mov_b32 s7, 0xf000 ; GFX8-NEXT: s_mov_b32 s6, -1 -; GFX8-NEXT: v_addc_u32_e32 v1, vcc, v3, v2, vcc +; GFX8-NEXT: v_addc_u32_e32 v1, vcc, v3, v1, vcc ; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX8-NEXT: s_endpgm ; @@ -1012,19 +1009,18 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 %additive ; GFX9-NEXT: .LBB5_2: ; GFX9-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[2:3], s2, v2, 0 ; GFX9-NEXT: s_mov_b32 s4, s0 -; GFX9-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX9-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX9-NEXT: v_readfirstlane_b32 s0, v0 -; GFX9-NEXT: v_mul_lo_u32 v0, s2, v2 ; GFX9-NEXT: s_mov_b32 s5, s1 +; GFX9-NEXT: v_readfirstlane_b32 s0, v0 ; GFX9-NEXT: v_readfirstlane_b32 s1, v1 -; GFX9-NEXT: v_add_u32_e32 v1, v4, v3 -; GFX9-NEXT: v_mov_b32_e32 v2, s1 -; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s0, v0 +; GFX9-NEXT: v_add_u32_e32 v1, v3, v4 +; GFX9-NEXT: v_mov_b32_e32 v3, s1 +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s0, v2 ; GFX9-NEXT: s_mov_b32 s7, 0xf000 ; GFX9-NEXT: s_mov_b32 s6, -1 -; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v2, v1, vcc +; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v3, v1, vcc ; GFX9-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX9-NEXT: s_endpgm ; @@ -1057,13 +1053,12 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 %additive ; GFX1064-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1064-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX1064-NEXT: s_waitcnt lgkmcnt(0) -; GFX1064-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1064-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1064-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1064-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1064-NEXT: v_mad_u64_u32 v[2:3], s[2:3], s2, v2, 0 ; GFX1064-NEXT: v_readfirstlane_b32 s2, v0 ; GFX1064-NEXT: v_readfirstlane_b32 s4, v1 ; GFX1064-NEXT: s_mov_b32 s3, 0x31016000 -; GFX1064-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1064-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1064-NEXT: v_add_co_u32 v0, vcc, s2, v2 ; GFX1064-NEXT: s_mov_b32 s2, -1 ; GFX1064-NEXT: v_add_co_ci_u32_e32 v1, vcc, s4, v1, vcc @@ -1098,13 +1093,12 @@ define amdgpu_kernel void @add_i64_uniform(i64 addrspace(1)* %out, i64 %additive ; GFX1032-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1032-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX1032-NEXT: s_waitcnt lgkmcnt(0) -; GFX1032-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1032-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1032-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1032-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1032-NEXT: v_mad_u64_u32 v[2:3], s2, s2, v2, 0 ; GFX1032-NEXT: v_readfirstlane_b32 s2, v0 ; GFX1032-NEXT: v_readfirstlane_b32 s4, v1 ; GFX1032-NEXT: s_mov_b32 s3, 0x31016000 -; GFX1032-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1032-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1032-NEXT: v_add_co_u32 v0, vcc_lo, s2, v2 ; GFX1032-NEXT: s_mov_b32 s2, -1 ; GFX1032-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, s4, v1, vcc_lo @@ -2133,15 +2127,13 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 %subitive ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], vcc ; GFX8-NEXT: s_cbranch_execz .LBB12_2 ; GFX8-NEXT: ; %bb.1: -; GFX8-NEXT: s_bcnt1_i32_b64 s6, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v0, s6 +; GFX8-NEXT: s_bcnt1_i32_b64 s8, s[6:7] +; GFX8-NEXT: v_mov_b32_e32 v0, s8 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_mul_hi_u32 v0, s2, v0 -; GFX8-NEXT: s_mul_i32 s7, s3, s6 -; GFX8-NEXT: s_mul_i32 s6, s2, s6 +; GFX8-NEXT: v_mad_u64_u32 v[0:1], s[6:7], s2, v0, 0 +; GFX8-NEXT: s_mul_i32 s6, s3, s8 ; GFX8-NEXT: v_mov_b32_e32 v3, 0 -; GFX8-NEXT: v_add_u32_e32 v1, vcc, s7, v0 -; GFX8-NEXT: v_mov_b32_e32 v0, s6 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, s6, v1 ; GFX8-NEXT: s_mov_b32 m0, -1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) ; GFX8-NEXT: ds_sub_rtn_u64 v[0:1], v3, v[0:1] @@ -2150,18 +2142,17 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 %subitive ; GFX8-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_waitcnt lgkmcnt(0) ; GFX8-NEXT: s_mov_b32 s4, s0 -; GFX8-NEXT: v_readfirstlane_b32 s0, v0 -; GFX8-NEXT: v_mul_lo_u32 v0, s3, v2 -; GFX8-NEXT: v_mul_hi_u32 v3, s2, v2 ; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX8-NEXT: v_mad_u64_u32 v[2:3], s[0:1], s2, v2, 0 +; GFX8-NEXT: v_readfirstlane_b32 s0, v0 ; GFX8-NEXT: v_readfirstlane_b32 s1, v1 -; GFX8-NEXT: v_mul_lo_u32 v1, s2, v2 -; GFX8-NEXT: v_add_u32_e32 v2, vcc, v3, v0 +; GFX8-NEXT: v_add_u32_e32 v1, vcc, v3, v4 ; GFX8-NEXT: v_mov_b32_e32 v3, s1 -; GFX8-NEXT: v_sub_u32_e32 v0, vcc, s0, v1 +; GFX8-NEXT: v_sub_u32_e32 v0, vcc, s0, v2 ; GFX8-NEXT: s_mov_b32 s7, 0xf000 ; GFX8-NEXT: s_mov_b32 s6, -1 -; GFX8-NEXT: v_subb_u32_e32 v1, vcc, v3, v2, vcc +; GFX8-NEXT: v_subb_u32_e32 v1, vcc, v3, v1, vcc ; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX8-NEXT: s_endpgm ; @@ -2191,19 +2182,18 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 %subitive ; GFX9-NEXT: .LBB12_2: ; GFX9-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[2:3], s2, v2, 0 ; GFX9-NEXT: s_mov_b32 s4, s0 -; GFX9-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX9-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX9-NEXT: v_readfirstlane_b32 s0, v0 -; GFX9-NEXT: v_mul_lo_u32 v0, s2, v2 ; GFX9-NEXT: s_mov_b32 s5, s1 +; GFX9-NEXT: v_readfirstlane_b32 s0, v0 ; GFX9-NEXT: v_readfirstlane_b32 s1, v1 -; GFX9-NEXT: v_add_u32_e32 v1, v4, v3 -; GFX9-NEXT: v_mov_b32_e32 v2, s1 -; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, s0, v0 +; GFX9-NEXT: v_add_u32_e32 v1, v3, v4 +; GFX9-NEXT: v_mov_b32_e32 v3, s1 +; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, s0, v2 ; GFX9-NEXT: s_mov_b32 s7, 0xf000 ; GFX9-NEXT: s_mov_b32 s6, -1 -; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v2, v1, vcc +; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v1, vcc ; GFX9-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GFX9-NEXT: s_endpgm ; @@ -2236,13 +2226,12 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 %subitive ; GFX1064-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1064-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX1064-NEXT: s_waitcnt lgkmcnt(0) -; GFX1064-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1064-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1064-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1064-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1064-NEXT: v_mad_u64_u32 v[2:3], s[2:3], s2, v2, 0 ; GFX1064-NEXT: v_readfirstlane_b32 s2, v0 ; GFX1064-NEXT: v_readfirstlane_b32 s4, v1 ; GFX1064-NEXT: s_mov_b32 s3, 0x31016000 -; GFX1064-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1064-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1064-NEXT: v_sub_co_u32 v0, vcc, s2, v2 ; GFX1064-NEXT: s_mov_b32 s2, -1 ; GFX1064-NEXT: v_sub_co_ci_u32_e32 v1, vcc, s4, v1, vcc @@ -2277,13 +2266,12 @@ define amdgpu_kernel void @sub_i64_uniform(i64 addrspace(1)* %out, i64 %subitive ; GFX1032-NEXT: s_waitcnt_depctr 0xffe3 ; GFX1032-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX1032-NEXT: s_waitcnt lgkmcnt(0) -; GFX1032-NEXT: v_mul_lo_u32 v3, s3, v2 -; GFX1032-NEXT: v_mul_hi_u32 v4, s2, v2 -; GFX1032-NEXT: v_mul_lo_u32 v2, s2, v2 +; GFX1032-NEXT: v_mul_lo_u32 v4, s3, v2 +; GFX1032-NEXT: v_mad_u64_u32 v[2:3], s2, s2, v2, 0 ; GFX1032-NEXT: v_readfirstlane_b32 s2, v0 ; GFX1032-NEXT: v_readfirstlane_b32 s4, v1 ; GFX1032-NEXT: s_mov_b32 s3, 0x31016000 -; GFX1032-NEXT: v_add_nc_u32_e32 v1, v4, v3 +; GFX1032-NEXT: v_add_nc_u32_e32 v1, v3, v4 ; GFX1032-NEXT: v_sub_co_u32 v0, vcc_lo, s2, v2 ; GFX1032-NEXT: s_mov_b32 s2, -1 ; GFX1032-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, s4, v1, vcc_lo diff --git a/llvm/test/CodeGen/AMDGPU/bypass-div.ll b/llvm/test/CodeGen/AMDGPU/bypass-div.ll index 4ff9f6159cae..907ba8dd3086 100644 --- a/llvm/test/CodeGen/AMDGPU/bypass-div.ll +++ b/llvm/test/CodeGen/AMDGPU/bypass-div.ll @@ -16,119 +16,107 @@ define i64 @sdiv64(i64 %a, i64 %b) { ; GFX9-NEXT: s_xor_b64 s[6:7], exec, s[4:5] ; GFX9-NEXT: s_cbranch_execz .LBB0_2 ; GFX9-NEXT: ; %bb.1: -; GFX9-NEXT: v_ashrrev_i32_e32 v4, 31, v3 -; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v4 -; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, v3, v4, vcc -; GFX9-NEXT: v_xor_b32_e32 v3, v3, v4 -; GFX9-NEXT: v_xor_b32_e32 v2, v2, v4 -; GFX9-NEXT: v_cvt_f32_u32_e32 v5, v2 -; GFX9-NEXT: v_cvt_f32_u32_e32 v6, v3 -; GFX9-NEXT: v_sub_co_u32_e32 v7, vcc, 0, v2 -; GFX9-NEXT: v_subb_co_u32_e32 v8, vcc, 0, v3, vcc -; GFX9-NEXT: v_mac_f32_e32 v5, 0x4f800000, v6 -; GFX9-NEXT: v_rcp_f32_e32 v5, v5 +; GFX9-NEXT: v_ashrrev_i32_e32 v9, 31, v3 +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v9 +; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, v3, v9, vcc +; GFX9-NEXT: v_xor_b32_e32 v10, v3, v9 +; GFX9-NEXT: v_xor_b32_e32 v11, v2, v9 +; GFX9-NEXT: v_cvt_f32_u32_e32 v2, v11 +; GFX9-NEXT: v_cvt_f32_u32_e32 v3, v10 +; GFX9-NEXT: v_sub_co_u32_e32 v7, vcc, 0, v11 +; GFX9-NEXT: v_subb_co_u32_e32 v8, vcc, 0, v10, vcc +; GFX9-NEXT: v_mac_f32_e32 v2, 0x4f800000, v3 +; GFX9-NEXT: v_rcp_f32_e32 v2, v2 ; GFX9-NEXT: v_mov_b32_e32 v14, 0 -; GFX9-NEXT: v_mul_f32_e32 v5, 0x5f7ffffc, v5 -; GFX9-NEXT: v_mul_f32_e32 v6, 0x2f800000, v5 -; GFX9-NEXT: v_trunc_f32_e32 v6, v6 -; GFX9-NEXT: v_mac_f32_e32 v5, 0xcf800000, v6 -; GFX9-NEXT: v_cvt_u32_f32_e32 v6, v6 -; GFX9-NEXT: v_cvt_u32_f32_e32 v5, v5 -; GFX9-NEXT: v_mul_lo_u32 v11, v7, v6 -; GFX9-NEXT: v_mul_lo_u32 v9, v8, v5 -; GFX9-NEXT: v_mul_hi_u32 v10, v7, v5 -; GFX9-NEXT: v_mul_lo_u32 v12, v7, v5 -; GFX9-NEXT: v_add3_u32 v9, v10, v11, v9 -; GFX9-NEXT: v_mul_lo_u32 v10, v5, v9 -; GFX9-NEXT: v_mul_hi_u32 v11, v5, v12 -; GFX9-NEXT: v_mul_hi_u32 v13, v5, v9 -; GFX9-NEXT: v_mul_hi_u32 v15, v6, v9 -; GFX9-NEXT: v_mul_lo_u32 v9, v6, v9 -; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, v11, v10 -; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, 0, v13, vcc -; GFX9-NEXT: v_mul_lo_u32 v13, v6, v12 -; GFX9-NEXT: v_mul_hi_u32 v12, v6, v12 -; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, v10, v13 -; GFX9-NEXT: v_addc_co_u32_e32 v10, vcc, v11, v12, vcc -; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, v15, v14, vcc -; GFX9-NEXT: v_add_co_u32_e32 v9, vcc, v10, v9 -; GFX9-NEXT: v_addc_co_u32_e32 v10, vcc, 0, v11, vcc -; GFX9-NEXT: v_add_co_u32_e32 v5, vcc, v5, v9 -; GFX9-NEXT: v_addc_co_u32_e32 v6, vcc, v6, v10, vcc -; GFX9-NEXT: v_mul_lo_u32 v9, v7, v6 -; GFX9-NEXT: v_mul_lo_u32 v8, v8, v5 -; GFX9-NEXT: v_mul_hi_u32 v10, v7, v5 -; GFX9-NEXT: v_mul_lo_u32 v7, v7, v5 -; GFX9-NEXT: v_add3_u32 v8, v10, v9, v8 -; GFX9-NEXT: v_mul_lo_u32 v11, v5, v8 -; GFX9-NEXT: v_mul_hi_u32 v12, v5, v7 -; GFX9-NEXT: v_mul_hi_u32 v13, v5, v8 -; GFX9-NEXT: v_mul_hi_u32 v10, v6, v7 -; GFX9-NEXT: v_mul_lo_u32 v7, v6, v7 -; GFX9-NEXT: v_mul_hi_u32 v9, v6, v8 -; GFX9-NEXT: v_add_co_u32_e32 v11, vcc, v12, v11 -; GFX9-NEXT: v_addc_co_u32_e32 v12, vcc, 0, v13, vcc -; GFX9-NEXT: v_mul_lo_u32 v8, v6, v8 -; GFX9-NEXT: v_add_co_u32_e32 v7, vcc, v11, v7 -; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, v12, v10, vcc -; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, v9, v14, vcc -; GFX9-NEXT: v_add_co_u32_e32 v7, vcc, v7, v8 -; GFX9-NEXT: v_addc_co_u32_e32 v8, vcc, 0, v9, vcc -; GFX9-NEXT: v_add_co_u32_e32 v5, vcc, v5, v7 -; GFX9-NEXT: v_addc_co_u32_e32 v6, vcc, v6, v8, vcc -; GFX9-NEXT: v_ashrrev_i32_e32 v7, 31, v1 -; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v7 -; GFX9-NEXT: v_xor_b32_e32 v0, v0, v7 -; GFX9-NEXT: v_mul_lo_u32 v8, v0, v6 -; GFX9-NEXT: v_mul_hi_u32 v9, v0, v5 -; GFX9-NEXT: v_mul_hi_u32 v10, v0, v6 -; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v7, vcc -; GFX9-NEXT: v_xor_b32_e32 v1, v1, v7 -; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, v9, v8 -; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, 0, v10, vcc -; GFX9-NEXT: v_mul_lo_u32 v10, v1, v5 -; GFX9-NEXT: v_mul_hi_u32 v5, v1, v5 -; GFX9-NEXT: v_mul_hi_u32 v11, v1, v6 -; GFX9-NEXT: v_mul_lo_u32 v6, v1, v6 -; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, v8, v10 -; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, v9, v5, vcc -; GFX9-NEXT: v_addc_co_u32_e32 v8, vcc, v11, v14, vcc -; GFX9-NEXT: v_add_co_u32_e32 v5, vcc, v5, v6 -; GFX9-NEXT: v_addc_co_u32_e32 v6, vcc, 0, v8, vcc -; GFX9-NEXT: v_mul_lo_u32 v8, v3, v5 -; GFX9-NEXT: v_mul_lo_u32 v9, v2, v6 -; GFX9-NEXT: v_mul_hi_u32 v10, v2, v5 -; GFX9-NEXT: v_mul_lo_u32 v11, v2, v5 -; GFX9-NEXT: v_add3_u32 v8, v10, v9, v8 -; GFX9-NEXT: v_sub_u32_e32 v9, v1, v8 -; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v11 -; GFX9-NEXT: v_subb_co_u32_e64 v9, s[4:5], v9, v3, vcc -; GFX9-NEXT: v_sub_co_u32_e64 v10, s[4:5], v0, v2 -; GFX9-NEXT: v_subbrev_co_u32_e64 v9, s[4:5], 0, v9, s[4:5] -; GFX9-NEXT: v_cmp_ge_u32_e64 s[4:5], v9, v3 -; GFX9-NEXT: v_cndmask_b32_e64 v11, 0, -1, s[4:5] -; GFX9-NEXT: v_cmp_ge_u32_e64 s[4:5], v10, v2 -; GFX9-NEXT: v_cndmask_b32_e64 v10, 0, -1, s[4:5] -; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], v9, v3 -; GFX9-NEXT: v_cndmask_b32_e64 v9, v11, v10, s[4:5] -; GFX9-NEXT: v_add_co_u32_e64 v10, s[4:5], 2, v5 -; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v8, vcc -; GFX9-NEXT: v_addc_co_u32_e64 v11, s[4:5], 0, v6, s[4:5] -; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v1, v3 -; GFX9-NEXT: v_add_co_u32_e64 v12, s[4:5], 1, v5 -; GFX9-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc -; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v0, v2 -; GFX9-NEXT: v_addc_co_u32_e64 v13, s[4:5], 0, v6, s[4:5] +; GFX9-NEXT: v_mul_f32_e32 v2, 0x5f7ffffc, v2 +; GFX9-NEXT: v_mul_f32_e32 v3, 0x2f800000, v2 +; GFX9-NEXT: v_trunc_f32_e32 v3, v3 +; GFX9-NEXT: v_mac_f32_e32 v2, 0xcf800000, v3 +; GFX9-NEXT: v_cvt_u32_f32_e32 v6, v2 +; GFX9-NEXT: v_cvt_u32_f32_e32 v12, v3 +; GFX9-NEXT: v_mul_lo_u32 v4, v8, v6 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v7, v6, 0 +; GFX9-NEXT: v_mul_lo_u32 v5, v7, v12 +; GFX9-NEXT: v_mul_hi_u32 v13, v6, v2 +; GFX9-NEXT: v_add3_u32 v5, v3, v5, v4 +; GFX9-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v6, v5, 0 +; GFX9-NEXT: v_add_co_u32_e32 v13, vcc, v13, v3 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v12, v2, 0 +; GFX9-NEXT: v_addc_co_u32_e32 v15, vcc, 0, v4, vcc +; GFX9-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v12, v5, 0 +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v13, v2 +; GFX9-NEXT: v_addc_co_u32_e32 v2, vcc, v15, v3, vcc +; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, v5, v14, vcc +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v4 +; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc +; GFX9-NEXT: v_add_co_u32_e32 v13, vcc, v6, v2 +; GFX9-NEXT: v_addc_co_u32_e32 v12, vcc, v12, v3, vcc +; GFX9-NEXT: v_mul_lo_u32 v4, v7, v12 +; GFX9-NEXT: v_mul_lo_u32 v5, v8, v13 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v7, v13, 0 +; GFX9-NEXT: v_add3_u32 v5, v3, v4, v5 +; GFX9-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v12, v5, 0 +; GFX9-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v5, 0 +; GFX9-NEXT: v_mul_hi_u32 v15, v13, v2 +; GFX9-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v12, v2, 0 +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v15, v5 +; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v6, vcc +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v7 +; GFX9-NEXT: v_addc_co_u32_e32 v2, vcc, v5, v8, vcc +; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, v4, v14, vcc +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v3 +; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v4, vcc +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v13, v2 +; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, v12, v3, vcc +; GFX9-NEXT: v_ashrrev_i32_e32 v4, 31, v1 +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v4 +; GFX9-NEXT: v_xor_b32_e32 v6, v0, v4 +; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, v1, v4, vcc +; GFX9-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v6, v3, 0 +; GFX9-NEXT: v_mul_hi_u32 v7, v6, v2 +; GFX9-NEXT: v_xor_b32_e32 v5, v5, v4 +; GFX9-NEXT: v_add_co_u32_e32 v7, vcc, v7, v0 +; GFX9-NEXT: v_addc_co_u32_e32 v8, vcc, 0, v1, vcc +; GFX9-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v5, v2, 0 +; GFX9-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v5, v3, 0 +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v7, v0 +; GFX9-NEXT: v_addc_co_u32_e32 v0, vcc, v8, v1, vcc +; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v3, v14, vcc +; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v0, v2 +; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v1, vcc +; GFX9-NEXT: v_mul_lo_u32 v7, v10, v2 +; GFX9-NEXT: v_mul_lo_u32 v8, v11, v3 +; GFX9-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v11, v2, 0 +; GFX9-NEXT: v_add3_u32 v1, v1, v8, v7 +; GFX9-NEXT: v_sub_u32_e32 v7, v5, v1 +; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, v6, v0 +; GFX9-NEXT: v_subb_co_u32_e64 v6, s[4:5], v7, v10, vcc +; GFX9-NEXT: v_sub_co_u32_e64 v7, s[4:5], v0, v11 +; GFX9-NEXT: v_subbrev_co_u32_e64 v6, s[4:5], 0, v6, s[4:5] +; GFX9-NEXT: v_cmp_ge_u32_e64 s[4:5], v6, v10 +; GFX9-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[4:5] +; GFX9-NEXT: v_cmp_ge_u32_e64 s[4:5], v7, v11 +; GFX9-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5] +; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], v6, v10 +; GFX9-NEXT: v_cndmask_b32_e64 v6, v8, v7, s[4:5] +; GFX9-NEXT: v_add_co_u32_e64 v7, s[4:5], 2, v2 +; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v5, v1, vcc +; GFX9-NEXT: v_addc_co_u32_e64 v8, s[4:5], 0, v3, s[4:5] +; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v1, v10 +; GFX9-NEXT: v_add_co_u32_e64 v12, s[4:5], 1, v2 +; GFX9-NEXT: v_cndmask_b32_e64 v5, 0, -1, vcc +; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v0, v11 +; GFX9-NEXT: v_addc_co_u32_e64 v13, s[4:5], 0, v3, s[4:5] ; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc -; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, v1, v3 -; GFX9-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v9 -; GFX9-NEXT: v_cndmask_b32_e32 v0, v8, v0, vcc +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, v1, v10 +; GFX9-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v6 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc ; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 -; GFX9-NEXT: v_cndmask_b32_e64 v1, v12, v10, s[4:5] -; GFX9-NEXT: v_cndmask_b32_e64 v9, v13, v11, s[4:5] -; GFX9-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc -; GFX9-NEXT: v_xor_b32_e32 v2, v7, v4 -; GFX9-NEXT: v_cndmask_b32_e32 v0, v6, v9, vcc +; GFX9-NEXT: v_cndmask_b32_e64 v1, v12, v7, s[4:5] +; GFX9-NEXT: v_cndmask_b32_e64 v6, v13, v8, s[4:5] +; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc +; GFX9-NEXT: v_xor_b32_e32 v2, v4, v9 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v6, vcc ; GFX9-NEXT: v_xor_b32_e32 v1, v1, v2 ; GFX9-NEXT: v_xor_b32_e32 v0, v0, v2 ; GFX9-NEXT: v_sub_co_u32_e32 v4, vcc, v1, v2 @@ -183,106 +171,94 @@ define i64 @udiv64(i64 %a, i64 %b) { ; GFX9-NEXT: ; %bb.1: ; GFX9-NEXT: v_cvt_f32_u32_e32 v4, v2 ; GFX9-NEXT: v_cvt_f32_u32_e32 v5, v3 -; GFX9-NEXT: v_sub_co_u32_e32 v6, vcc, 0, v2 -; GFX9-NEXT: v_subb_co_u32_e32 v7, vcc, 0, v3, vcc +; GFX9-NEXT: v_sub_co_u32_e32 v10, vcc, 0, v2 +; GFX9-NEXT: v_subb_co_u32_e32 v11, vcc, 0, v3, vcc ; GFX9-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5 ; GFX9-NEXT: v_rcp_f32_e32 v4, v4 -; GFX9-NEXT: v_mov_b32_e32 v12, 0 +; GFX9-NEXT: v_mov_b32_e32 v13, 0 ; GFX9-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4 ; GFX9-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4 ; GFX9-NEXT: v_trunc_f32_e32 v5, v5 ; GFX9-NEXT: v_mac_f32_e32 v4, 0xcf800000, v5 -; GFX9-NEXT: v_cvt_u32_f32_e32 v5, v5 -; GFX9-NEXT: v_cvt_u32_f32_e32 v4, v4 -; GFX9-NEXT: v_mul_lo_u32 v8, v6, v5 -; GFX9-NEXT: v_mul_lo_u32 v9, v7, v4 -; GFX9-NEXT: v_mul_hi_u32 v10, v6, v4 -; GFX9-NEXT: v_mul_lo_u32 v11, v6, v4 -; GFX9-NEXT: v_add3_u32 v8, v10, v8, v9 -; GFX9-NEXT: v_mul_hi_u32 v9, v4, v11 -; GFX9-NEXT: v_mul_lo_u32 v10, v4, v8 -; GFX9-NEXT: v_mul_hi_u32 v13, v4, v8 -; GFX9-NEXT: v_mul_hi_u32 v14, v5, v8 -; GFX9-NEXT: v_mul_lo_u32 v8, v5, v8 -; GFX9-NEXT: v_add_co_u32_e32 v9, vcc, v9, v10 -; GFX9-NEXT: v_addc_co_u32_e32 v10, vcc, 0, v13, vcc -; GFX9-NEXT: v_mul_lo_u32 v13, v5, v11 -; GFX9-NEXT: v_mul_hi_u32 v11, v5, v11 -; GFX9-NEXT: v_add_co_u32_e32 v9, vcc, v9, v13 -; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, v10, v11, vcc -; GFX9-NEXT: v_addc_co_u32_e32 v10, vcc, v14, v12, vcc -; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, v9, v8 -; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, 0, v10, vcc -; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, v4, v8 -; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, v5, v9, vcc -; GFX9-NEXT: v_mul_lo_u32 v8, v6, v5 -; GFX9-NEXT: v_mul_lo_u32 v7, v7, v4 -; GFX9-NEXT: v_mul_hi_u32 v9, v6, v4 -; GFX9-NEXT: v_mul_lo_u32 v6, v6, v4 -; GFX9-NEXT: v_add3_u32 v7, v9, v8, v7 -; GFX9-NEXT: v_mul_lo_u32 v10, v4, v7 -; GFX9-NEXT: v_mul_hi_u32 v11, v4, v6 -; GFX9-NEXT: v_mul_hi_u32 v13, v4, v7 -; GFX9-NEXT: v_mul_hi_u32 v9, v5, v6 -; GFX9-NEXT: v_mul_lo_u32 v6, v5, v6 -; GFX9-NEXT: v_mul_hi_u32 v8, v5, v7 -; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, v11, v10 -; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, 0, v13, vcc -; GFX9-NEXT: v_mul_lo_u32 v7, v5, v7 -; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, v10, v6 -; GFX9-NEXT: v_addc_co_u32_e32 v6, vcc, v11, v9, vcc -; GFX9-NEXT: v_addc_co_u32_e32 v8, vcc, v8, v12, vcc -; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, v6, v7 -; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, 0, v8, vcc +; GFX9-NEXT: v_cvt_u32_f32_e32 v8, v5 +; GFX9-NEXT: v_cvt_u32_f32_e32 v9, v4 +; GFX9-NEXT: v_mul_lo_u32 v6, v10, v8 +; GFX9-NEXT: v_mul_lo_u32 v7, v11, v9 +; GFX9-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v9, 0 +; GFX9-NEXT: v_add3_u32 v7, v5, v6, v7 +; GFX9-NEXT: v_mul_hi_u32 v12, v9, v4 +; GFX9-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v9, v7, 0 +; GFX9-NEXT: v_add_co_u32_e32 v12, vcc, v12, v5 +; GFX9-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v8, v4, 0 +; GFX9-NEXT: v_addc_co_u32_e32 v14, vcc, 0, v6, vcc +; GFX9-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v8, v7, 0 +; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, v12, v4 +; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, v14, v5, vcc +; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, v7, v13, vcc ; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, v4, v6 -; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, v5, v7, vcc -; GFX9-NEXT: v_mul_lo_u32 v6, v0, v5 -; GFX9-NEXT: v_mul_hi_u32 v7, v0, v4 -; GFX9-NEXT: v_mul_hi_u32 v8, v0, v5 -; GFX9-NEXT: v_mul_hi_u32 v9, v1, v5 -; GFX9-NEXT: v_mul_lo_u32 v5, v1, v5 -; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, v7, v6 +; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc +; GFX9-NEXT: v_add_co_u32_e32 v12, vcc, v9, v4 +; GFX9-NEXT: v_addc_co_u32_e32 v14, vcc, v8, v5, vcc +; GFX9-NEXT: v_mul_lo_u32 v6, v10, v14 +; GFX9-NEXT: v_mul_lo_u32 v7, v11, v12 +; GFX9-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v12, 0 +; GFX9-NEXT: v_add3_u32 v7, v5, v6, v7 +; GFX9-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v14, v7, 0 +; GFX9-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v12, v7, 0 +; GFX9-NEXT: v_mul_hi_u32 v11, v12, v4 +; GFX9-NEXT: v_mad_u64_u32 v[9:10], s[4:5], v14, v4, 0 +; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, v11, v7 ; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, 0, v8, vcc -; GFX9-NEXT: v_mul_lo_u32 v8, v1, v4 -; GFX9-NEXT: v_mul_hi_u32 v4, v1, v4 -; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, v6, v8 -; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, v7, v4, vcc -; GFX9-NEXT: v_addc_co_u32_e32 v6, vcc, v9, v12, vcc +; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, v4, v9 +; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, v7, v10, vcc +; GFX9-NEXT: v_addc_co_u32_e32 v6, vcc, v6, v13, vcc ; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, v4, v5 ; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v6, vcc -; GFX9-NEXT: v_mul_lo_u32 v6, v3, v4 -; GFX9-NEXT: v_mul_lo_u32 v7, v2, v5 -; GFX9-NEXT: v_mul_hi_u32 v8, v2, v4 </cut>

3 years

[ACTIVITY] week ending 21 Nov 2021

by Richard Henderson

[UM-2] * release work * revived some 6month old ppc fpu fixes * reviews: loongarch, riscv, watchpoints, gdbstub. r~

3 years

[ACTIVITY] week ending Nov. 21 2021

by Alex Bennée

VirtIO Initiative ([STR-9]) =========================== - posted Initial thoughts for test scenarios for AF_XDP epic Message-Id: <87k0h5v6ju.fsf(a)linaro.org> vhost-device maintainer effort ([UM-196]) - more review - [did some more noodling with rust] to get comfortable with generics [UM-196] <https://linaro.atlassian.net/browse/UM-196> [vhost-device crate] <https://github.com/rust-vmm/vhost-device> [did some more noodling with rust] <https://gitlab.com/stsquad/softfloat.rs> QEMU Upstream Work ([UM-2]) =========================== - posted [PULL for 6.2 0/7] misc build and test fixes Message-Id: <20211116162515.4100231-1-alex.bennee(a)linaro.org> - posted [RFC PATCH] tests/avocado: fix tcg_plugin mem access count test Message-Id: <20211117095448.136558-1-alex.bennee(a)linaro.org> - posted Re: [RFC PATCH] plugins/meson.build: fix linker issue with weird paths (for v6.2?) Message-Id: <20211117111924.179776-1-alex.bennee(a)linaro.org> - posted Re: [PATCH v2 1/3] icount: preserve cflags when custom tb is about to execute Message-Id: <87h7cbw1tx.fsf(a)linaro.org> - posted [RFC PATCH] gdbstub: handle a potentially racing TaskState Message-Id: <20211119145124.942390-1-alex.bennee(a)linaro.org> [UM-2] <https://linaro.atlassian.net/browse/UM-2> Upstream MTTCG tests ([QEMU-52]) - posted [kvm-unit-tests PATCH v3 0/3] GIC ITS tests Message-Id: <20211112114734.3058678-1-alex.bennee(a)linaro.org> - posted [kvm-unit-tests PATCH v8 00/10] MTTCG sanity tests for ARM Message-Id: <20211118184650.661575-1-alex.bennee(a)linaro.org> [QEMU-52] <https://linaro.atlassian.net/browse/QEMU-52> [mttcg tests to current state and fixed up] <https://github.com/stsquad/qemu/tree/mttcg/current-tests-v8> Completed Reviews [2/2] ======================= [PATCH v2 0/3] Some watchpoint-related patches Message-Id: <163662450348.125458.5494710452733592356.stgit@pasha-ThinkPad-X280> [PATCH 0/5] Update linux-headers + NOIRQ support for KVM gdbstub Message-Id: <20211111110604.207376-1-pbonzini(a)redhat.com> Absences ======== - none Current Review Queue ==================== TODO [PATCH-4.16 v2] xen/efi: Fix Grub2 boot on arm64 Message-Id: <20211104141206.25153-1-luca.fancellu(a)arm.com> =============================================================================================================== TODO [PATCH] cpu-models-x86.rst: Tidy up a couple of things Message-Id: <20211015100718.17828-1-pbonzini(a)redhat.com> =================================================================================================================== TODO [PATCH 00/16] fdt: Make OF_BOARD a boolean option Message-Id: <20211013010120.96851-1-sjg(a)chromium.org> =========================================================================================================== TODO [PATCH v4 00/41] linux-user: Streamline handling of SIGSEGV Message-Id: <20211006172307.780893-1-richard.henderson(a)linaro.org> ================================================================================================================================== -- Alex Bennée

3 years

[ACTIVITY] report week ending 19 Nov

by Peter Maydell

Progress (short week, 3 days) * UM-2 [QEMU upstream maintainership] - Still trying to sort out the regression of booting EL3 guest code on the imx7 board. I got most of the way through prototyping a cleanup which would fix this, but then spotted that the highbank board has a more awkward-to-fix similar problem. We're going to revert the PSCI emulation change for 6.2 so we can take the time to get the cleanup right and land it in 7.0. - Usual patch accumulation, review, etc during release cycle -- PMM

3 years

Jump to page:

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

linaro-toolchain