[ArgPromotion] Remove dead code produced by removing dead arguments

[jrbyrnes]

Summary

ArgPromotion currently produces phantom / dead loads. A good example of this is store-into-inself.ll. First, ArgPromo finds the promotable argument %p in @l. Then it inserts a load of %p in the caller, and passes instead the loaded value / transforms the function body. PromoteMem2Reg is able to optimize out the entire function body, resulting in an unused argument. In a subsequent doPromotion iteration, it removes the dead argument, resulting in a dead load in the caller. These dead loads may reduce effectiveness of other transformations (e.g. SimplifyCFG, MergedLoadStoreMotion).

This patch removes loads and geps that are made dead in the caller after removal of dead args.

[nikic]

Please add a PhaseOrdering test that demonstrates why it it useful to do this as part of ArgumentPromotion, rather than letting other passes clean this up.

llvm/lib/Transforms/IPO/ArgumentPromotion.cpp

283

This implementation looks too complicated. It should be enough to collect all potentially dead arguments and then call RecursivelyDeleteTriviallyDeadInstructionsPermissive() on the vector.

llvm/test/Transforms/ArgumentPromotion/propagate-remove-dead-args.ll

1

Add --function-signature or --version 2.

6

nounwind not needed?

32

fastcc not needed?

[jrbyrnes]

Thanks @nikic for your comments. Added the PhaseOrdering test to demonstrate impact patch has on subsequent passes. Also, cleaned up implementation w/ use of API.

[nikic]

This looks good to me, but could you please pre-commit the new tests with baseline check lines, and then rebase this patch so that only the test diff is visible?

llvm/lib/Transforms/IPO/ArgumentPromotion.cpp

261

Don't think we need this comment given the assert on the next line...

[jrbyrnes]

Rebase to show diff.

Summary of impact on PhaseOrdering test: During argPromotion of @phantomLoad, we introduce some loads of %this in @badChild. In a subsequent iteration of doPromotion we remove dead %this from @phantomLoad argument. This patch removes also the dead loads in @badChild. This allows argPromotion of @badChild to promote %this which cleans up the if.then and if.else blocks a bit. Then, MergedLoadStoreMotion is able to completely sink the CFG due to cleaned structure, and SimplifyCFG is able to remove it altogether.

[nikic]

LGTM

Thanks for the PhaseOrdering test, I think I get the problem now. ArgPromotion is the first pass in the CGSCC pipeline, and works a bit unusually in that promotion will insert loads in the caller (outside the current SCC). This means the we run ArgPromotion on a child function, which inserts instructions in the parent and simplify the child. Then we run ArgPromotion on the parent, but at this point the parent has not been simplified yet. As such, it is important for ArgPromotion to clean up instructions it inserts. There's possibly some phase ordering improvement we could make here, but I don't see anything obvious. (Just moving ArgPromotion to the end of the CGSCC pipeline would cause the reverse issue that function simplification can not make use of the promotion for cleanup.)

[uabelho]

Hi,

The following seems to crash ArgumentPromotion with this patch:

clang -cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O3 -o foo2.obj foo2.c

Result:

clang: ../include/llvm/Support/Casting.h:657: decltype(auto) llvm::dyn_cast(From &) [To = llvm::Instruction, From = llvm::WeakTrackingVH]: Assertion `detail::isPresent(Val) && "dyn_cast on a non-existent value"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.	Program arguments: build-all/bin/clang -cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O3 -o foo2.obj foo2.c
1.	<eof> parser at end of file
2.	Optimizer
 #0 0x0000000002ff8598 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (build-all/bin/clang+0x2ff8598)
 #1 0x0000000002ff612e llvm::sys::RunSignalHandlers() (build-all/bin/clang+0x2ff612e)
 #2 0x0000000002ff8c16 SignalHandler(int) Signals.cpp:0:0
 #3 0x00007efccbc19630 __restore_rt sigaction.c:0:0
 #4 0x00007efcc9360387 raise (/lib64/libc.so.6+0x36387)
 #5 0x00007efcc9361a78 abort (/lib64/libc.so.6+0x37a78)
 #6 0x00007efcc93591a6 __assert_fail_base (/lib64/libc.so.6+0x2f1a6)
 #7 0x00007efcc9359252 (/lib64/libc.so.6+0x2f252)
 #8 0x000000000309d12d (build-all/bin/clang+0x309d12d)
 #9 0x00000000041f341e doPromotion(llvm::Function*, llvm::AnalysisManager<llvm::Function>&, llvm::DenseMap<llvm::Argument*, llvm::SmallVector<std::pair<long, (anonymous namespace)::ArgPart>, 4u>, llvm::DenseMapInfo<llvm::Argument*, void>, llvm::detail::DenseMapPair<llvm::Argument*, llvm::SmallVector<std::pair<long, (anonymous namespace)::ArgPart>, 4u>>> const&) ArgumentPromotion.cpp:0:0
#10 0x00000000041ee86b llvm::ArgumentPromotionPass::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) (build-all/bin/clang+0x41ee86b)
#11 0x000000000413528d llvm::detail::PassModel<llvm::LazyCallGraph::SCC, llvm::ArgumentPromotionPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) crtstuff.c:0:0
#12 0x00000000020a4c74 llvm::PassManager<llvm::LazyCallGraph::SCC, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) (build-all/bin/clang+0x20a4c74)
#13 0x000000000412a84d llvm::detail::PassModel<llvm::LazyCallGraph::SCC, llvm::PassManager<llvm::LazyCallGraph::SCC, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) crtstuff.c:0:0
#14 0x00000000020a80f2 llvm::DevirtSCCRepeatedPass::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) (build-all/bin/clang+0x20a80f2)
#15 0x0000000004143f9d llvm::detail::PassModel<llvm::LazyCallGraph::SCC, llvm::DevirtSCCRepeatedPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) crtstuff.c:0:0
#16 0x00000000020a6bab llvm::ModuleToPostOrderCGSCCPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x20a6bab)
#17 0x000000000412aafd llvm::detail::PassModel<llvm::Module, llvm::ModuleToPostOrderCGSCCPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) crtstuff.c:0:0
#18 0x0000000002a0dceb llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x2a0dceb)
#19 0x0000000004255fea llvm::ModuleInlinerWrapperPass::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x4255fea)
#20 0x000000000413128d llvm::detail::PassModel<llvm::Module, llvm::ModuleInlinerWrapperPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) crtstuff.c:0:0
#21 0x0000000002a0dceb llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x2a0dceb)
#22 0x00000000031dec5f (anonymous namespace)::EmitAssemblyHelper::RunOptimizationPipeline(clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream>>&, std::unique_ptr<llvm::ToolOutputFile, std::default_delete<llvm::ToolOutputFile>>&) BackendUtil.cpp:0:0
#23 0x00000000031d7d75 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, llvm::IntrusiveRefCntPtr<llvm::vfs::FileSystem>, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream>>) (build-all/bin/clang+0x31d7d75)
#24 0x00000000040767e2 clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) crtstuff.c:0:0
#25 0x0000000004b0d0e3 clang::ParseAST(clang::Sema&, bool, bool) (build-all/bin/clang+0x4b0d0e3)
#26 0x000000000398bec6 clang::FrontendAction::Execute() (build-all/bin/clang+0x398bec6)
#27 0x00000000038f7674 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (build-all/bin/clang+0x38f7674)
#28 0x0000000003a4fc2b clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (build-all/bin/clang+0x3a4fc2b)
#29 0x0000000000a5f2cc cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (build-all/bin/clang+0xa5f2cc)
#30 0x0000000000a5b2c0 ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&, llvm::ToolContext const&) driver.cpp:0:0
#31 0x0000000000a5a3f6 clang_main(int, char**, llvm::ToolContext const&) (build-all/bin/clang+0xa5a3f6)
#32 0x0000000000a6ad51 main (build-all/bin/clang+0xa6ad51)
#33 0x00007efcc934c555 __libc_start_main (/lib64/libc.so.6+0x22555)
#34 0x0000000000a56cbb _start (build-all/bin/clang+0xa56cbb)
Abort (core dumped)

foo2.c100 BDownload

[uabelho]

In D146327#4218484, @uabelho wrote:

Hi,

The following seems to crash ArgumentPromotion with this patch:

clang -cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O3 -o foo2.obj foo2.c

Result:

clang: ../include/llvm/Support/Casting.h:657: decltype(auto) llvm::dyn_cast(From &) [To = llvm::Instruction, From = llvm::WeakTrackingVH]: Assertion `detail::isPresent(Val) && "dyn_cast on a non-existent value"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.	Program arguments: build-all/bin/clang -cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O3 -o foo2.obj foo2.c
1.	<eof> parser at end of file
2.	Optimizer
 #0 0x0000000002ff8598 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (build-all/bin/clang+0x2ff8598)
 #1 0x0000000002ff612e llvm::sys::RunSignalHandlers() (build-all/bin/clang+0x2ff612e)
 #2 0x0000000002ff8c16 SignalHandler(int) Signals.cpp:0:0
 #3 0x00007efccbc19630 __restore_rt sigaction.c:0:0
 #4 0x00007efcc9360387 raise (/lib64/libc.so.6+0x36387)
 #5 0x00007efcc9361a78 abort (/lib64/libc.so.6+0x37a78)
 #6 0x00007efcc93591a6 __assert_fail_base (/lib64/libc.so.6+0x2f1a6)
 #7 0x00007efcc9359252 (/lib64/libc.so.6+0x2f252)
 #8 0x000000000309d12d (build-all/bin/clang+0x309d12d)
 #9 0x00000000041f341e doPromotion(llvm::Function*, llvm::AnalysisManager<llvm::Function>&, llvm::DenseMap<llvm::Argument*, llvm::SmallVector<std::pair<long, (anonymous namespace)::ArgPart>, 4u>, llvm::DenseMapInfo<llvm::Argument*, void>, llvm::detail::DenseMapPair<llvm::Argument*, llvm::SmallVector<std::pair<long, (anonymous namespace)::ArgPart>, 4u>>> const&) ArgumentPromotion.cpp:0:0
#10 0x00000000041ee86b llvm::ArgumentPromotionPass::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) (build-all/bin/clang+0x41ee86b)
#11 0x000000000413528d llvm::detail::PassModel<llvm::LazyCallGraph::SCC, llvm::ArgumentPromotionPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) crtstuff.c:0:0
#12 0x00000000020a4c74 llvm::PassManager<llvm::LazyCallGraph::SCC, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) (build-all/bin/clang+0x20a4c74)
#13 0x000000000412a84d llvm::detail::PassModel<llvm::LazyCallGraph::SCC, llvm::PassManager<llvm::LazyCallGraph::SCC, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) crtstuff.c:0:0
#14 0x00000000020a80f2 llvm::DevirtSCCRepeatedPass::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) (build-all/bin/clang+0x20a80f2)
#15 0x0000000004143f9d llvm::detail::PassModel<llvm::LazyCallGraph::SCC, llvm::DevirtSCCRepeatedPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&>::run(llvm::LazyCallGraph::SCC&, llvm::AnalysisManager<llvm::LazyCallGraph::SCC, llvm::LazyCallGraph&>&, llvm::LazyCallGraph&, llvm::CGSCCUpdateResult&) crtstuff.c:0:0
#16 0x00000000020a6bab llvm::ModuleToPostOrderCGSCCPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x20a6bab)
#17 0x000000000412aafd llvm::detail::PassModel<llvm::Module, llvm::ModuleToPostOrderCGSCCPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) crtstuff.c:0:0
#18 0x0000000002a0dceb llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x2a0dceb)
#19 0x0000000004255fea llvm::ModuleInlinerWrapperPass::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x4255fea)
#20 0x000000000413128d llvm::detail::PassModel<llvm::Module, llvm::ModuleInlinerWrapperPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) crtstuff.c:0:0
#21 0x0000000002a0dceb llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (build-all/bin/clang+0x2a0dceb)
#22 0x00000000031dec5f (anonymous namespace)::EmitAssemblyHelper::RunOptimizationPipeline(clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream>>&, std::unique_ptr<llvm::ToolOutputFile, std::default_delete<llvm::ToolOutputFile>>&) BackendUtil.cpp:0:0
#23 0x00000000031d7d75 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, llvm::IntrusiveRefCntPtr<llvm::vfs::FileSystem>, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream>>) (build-all/bin/clang+0x31d7d75)
#24 0x00000000040767e2 clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) crtstuff.c:0:0
#25 0x0000000004b0d0e3 clang::ParseAST(clang::Sema&, bool, bool) (build-all/bin/clang+0x4b0d0e3)
#26 0x000000000398bec6 clang::FrontendAction::Execute() (build-all/bin/clang+0x398bec6)
#27 0x00000000038f7674 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (build-all/bin/clang+0x38f7674)
#28 0x0000000003a4fc2b clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (build-all/bin/clang+0x3a4fc2b)
#29 0x0000000000a5f2cc cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (build-all/bin/clang+0xa5f2cc)
#30 0x0000000000a5b2c0 ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&, llvm::ToolContext const&) driver.cpp:0:0
#31 0x0000000000a5a3f6 clang_main(int, char**, llvm::ToolContext const&) (build-all/bin/clang+0xa5a3f6)
#32 0x0000000000a6ad51 main (build-all/bin/clang+0xa6ad51)
#33 0x00007efcc934c555 __libc_start_main (/lib64/libc.so.6+0x22555)
#34 0x0000000000a56cbb _start (build-all/bin/clang+0xa56cbb)
Abort (core dumped)

foo2.c100 BDownload

Reduced opt reproducer

opt -passes="argpromotion" bbi-80617.ll -o /dev/null

bbi-80617.ll615 BDownload

[nikic]

Minimal test case:

define internal ptr @callee(ptr %dead) {
  ret ptr null
}

define void @caller() {
  %ret = call ptr @callee(ptr null)
  %ret2 = call ptr @callee(ptr %ret)
  ret void
}

[nikic]

@uabelho Should be fixed by https://github.com/llvm/llvm-project/commit/4923b4dbaca386b532c80f0572c2d6a0547a2a3d.

[uabelho]

In D146327#4219176, @nikic wrote:

@uabelho Should be fixed by https://github.com/llvm/llvm-project/commit/4923b4dbaca386b532c80f0572c2d6a0547a2a3d.

Yep, thanks!

[jrbyrnes]

Wow, thanks for finding that and for the quick work!

llvm/lib/Transforms/IPO/ArgumentPromotion.cpp

 #include "llvm/IR/PassManager.h"
 #include "llvm/IR/Type.h"
 #include "llvm/IR/Use.h"
 #include "llvm/IR/User.h"
 #include "llvm/IR/Value.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Utils/PromoteMemToReg.h"
 #include <algorithm>
 #include <cassert>
 #include <cstdint>
 #include <utility>
 #include <vector>
 
 using namespace llvm;
doPromotion(Function *F, FunctionAnalysisManager &FAM,
 
   F->getParent()->getFunctionList().insert(F->getIterator(), NF);
   NF->takeName(F);
 
   // Loop over all the callers of the function, transforming the call sites to
   // pass in the loaded pointers.
   SmallVector<Value *, 16> Args;
   const DataLayout &DL = F->getParent()->getDataLayout();
+  SmallVector<WeakTrackingVH, 16> DeadArgs;
+
   while (!F->use_empty()) {
     CallBase &CB = cast<CallBase>(*F->user_back());
     assert(CB.getCalledFunction() == F);
     const AttributeList &CallPAL = CB.getAttributes();
     IRBuilder<NoFolder> IRB(&CB);
 
     // Loop over the operands, inserting GEP and loads in the caller as
     // appropriate.
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E;
                               LLVMContext::MD_dereferenceable,
                               LLVMContext::MD_dereferenceable_or_null,
                               LLVMContext::MD_align, LLVMContext::MD_noundef,
                               LLVMContext::MD_nontemporal});
           }
           Args.push_back(LI);
           ArgAttrVec.push_back(AttributeSet());
         }
+      } else {
+        assert(ArgsToPromote.count(&*I) && I->use_empty());
+        DeadArgs.emplace_back(AI->get());
       }
     }
 
     // Push any varargs arguments on the list.
     for (; AI != CB.arg_end(); ++AI, ++ArgNo) {
       Args.push_back(*AI);
       ArgAttrVec.push_back(CallPAL.getParamAttrs(ArgNo));
     }
 
     SmallVector<OperandBundleDef, 1> OpBundles;
     CB.getOperandBundlesAsDefs(OpBundles);
 
     CallBase *NewCS = nullptr;
     if (InvokeInst *II = dyn_cast<InvokeInst>(&CB)) {
       NewCS = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
                                  Args, OpBundles, "", &CB);
     } else {
       auto *NewCall = CallInst::Create(NF, Args, OpBundles, "", &CB);
       NewCall->setTailCallKind(cast<CallInst>(&CB)->getTailCallKind());
       NewCS = NewCall;
     }
     NewCS->setCallingConv(CB.getCallingConv());
     NewCS->setAttributes(AttributeList::get(F->getContext(),
                                             CallPAL.getFnAttrs(),
                                             CallPAL.getRetAttrs(), ArgAttrVec));
     NewCS->copyMetadata(CB, {LLVMContext::MD_prof, LLVMContext::MD_dbg});
     Args.clear();
     ArgAttrVec.clear();
 
     AttributeFuncs::updateMinLegalVectorWidthAttr(*CB.getCaller(),
                                                   LargestVectorWidth);
 
     if (!CB.use_empty()) {
       CB.replaceAllUsesWith(NewCS);
       NewCS->takeName(&CB);
     }
 
     // Finally, remove the old call from the program, reducing the use-count of
     // F.
     CB.eraseFromParent();
   }
 
+  RecursivelyDeleteTriviallyDeadInstructionsPermissive(DeadArgs);
+
   // Since we have now created the new function, splice the body of the old
   // function right into the new function, leaving the old rotting hulk of the
   // function empty.
   NF->splice(NF->begin(), F);
 
   // We will collect all the new created allocas to promote them into registers
   // after the following loop
   SmallVector<AllocaInst *, 4> Allocas;

llvm/test/Transforms/ArgumentPromotion/propagate-remove-dead-args.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
 ; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 %ptr.struct = type { ptr, ptr, ptr }
 
 define internal void @child(ptr %this, ptr %y, ptr %x) {
 ; CHECK-LABEL: define internal void @child
 ; CHECK-SAME: (ptr [[Y:%.*]], half [[X_0_VAL:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    store half [[X_0_VAL]], ptr [[Y]], align 2
 ; CHECK-NEXT:    ret void
 ;
 entry:
   %0 = load half, ptr %x
   store half %0, ptr %y
   ret void
 }
 
 define internal void @parent(ptr %this, ptr %p1, ptr %p2) {
 ; CHECK-LABEL: define internal void @parent
-; CHECK-SAME: (ptr [[THIS:%.*]], ptr [[P1:%.*]], ptr [[P2:%.*]]) {
+; CHECK-SAME: (ptr [[P1:%.*]], ptr [[P2:%.*]]) {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[SRC_ELEMENT_OP_0:%.*]] = getelementptr ptr, ptr [[THIS]], i64 0
-; CHECK-NEXT:    [[LOAD0:%.*]] = load ptr, ptr [[SRC_ELEMENT_OP_0]], align 8
 ; CHECK-NEXT:    [[P2_VAL2:%.*]] = load half, ptr [[P2]], align 2
 ; CHECK-NEXT:    call void @child(ptr [[P1]], half [[P2_VAL2]])
-; CHECK-NEXT:    [[SRC_ELEMENT_OP_1:%.*]] = getelementptr ptr, ptr [[THIS]], i64 1
-; CHECK-NEXT:    [[LOAD1:%.*]] = load ptr, ptr [[SRC_ELEMENT_OP_1]], align 8
 ; CHECK-NEXT:    [[P2_VAL1:%.*]] = load half, ptr [[P2]], align 2
 ; CHECK-NEXT:    call void @child(ptr [[P1]], half [[P2_VAL1]])
-; CHECK-NEXT:    [[SRC_ELEMENT_OP_2:%.*]] = getelementptr ptr, ptr [[THIS]], i64 2
-; CHECK-NEXT:    [[LOAD2:%.*]] = load ptr, ptr [[SRC_ELEMENT_OP_2]], align 8
 ; CHECK-NEXT:    [[P2_VAL:%.*]] = load half, ptr [[P2]], align 2
 ; CHECK-NEXT:    call void @child(ptr [[P1]], half [[P2_VAL]])
 ; CHECK-NEXT:    ret void
 ;
 entry:
   %src_element_op_0 = getelementptr ptr, ptr %this, i64 0
   %load0 = load ptr, ptr %src_element_op_0
   call void @child(ptr %load0, ptr %p1, ptr %p2)
   %src_element_op_1 = getelementptr ptr, ptr %this, i64 1
   %load1 = load ptr, ptr %src_element_op_1
   call void @child(ptr %load1, ptr %p1, ptr %p2)
   %src_element_op_2 = getelementptr ptr, ptr %this, i64 2
   %load2 = load ptr, ptr %src_element_op_2
   call void @child(ptr %load2, ptr %p1, ptr %p2)
   ret void
 }
 
 define  void @grandparent() {
 ; CHECK-LABEL: define void @grandparent() {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[F:%.*]] = alloca [[PTR_STRUCT:%.*]], align 8
 ; CHECK-NEXT:    [[XPTR:%.*]] = alloca i32, align 4
 ; CHECK-NEXT:    [[YPTR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    call void @parent(ptr [[F]], ptr [[XPTR]], ptr [[YPTR]])
+; CHECK-NEXT:    call void @parent(ptr [[XPTR]], ptr [[YPTR]])
 ; CHECK-NEXT:    ret void
 ;
 entry:
   %f = alloca %ptr.struct
   %xptr = alloca i32
   %yptr = alloca i32
   call void @parent(ptr %f, ptr %xptr, ptr %yptr)
   ret void
 }
 

llvm/test/Transforms/ArgumentPromotion/store-into-inself.ll

 ; CHECK-NEXT:    [[TEMP1:%.*]] = getelementptr [[STRUCT_SS]], ptr [[S]], i32 0, i32 0
 ; CHECK-NEXT:    store i32 1, ptr [[TEMP1]], align 4
 ; CHECK-NEXT:    [[TEMP4:%.*]] = getelementptr [[STRUCT_SS]], ptr [[S]], i32 0, i32 1
 ; CHECK-NEXT:    store i64 2, ptr [[TEMP4]], align 8
 ; CHECK-NEXT:    call void @f(ptr byval(ptr) align 4 [[S]]) #[[ATTR0]]
 ; CHECK-NEXT:    call void @g(ptr byval(ptr) align 4 [[S]]) #[[ATTR0]]
 ; CHECK-NEXT:    call void @h(ptr byval(ptr) align 4 [[S]]) #[[ATTR0]]
 ; CHECK-NEXT:    call void @k(ptr byval(ptr) align 4 [[S]]) #[[ATTR0]]
-; CHECK-NEXT:    [[S_VAL:%.*]] = load ptr, ptr [[S]], align 8
 ; CHECK-NEXT:    call void @l() #[[ATTR0]]
 ; CHECK-NEXT:    ret i32 0
 ;
 entry:
   %S = alloca %struct.ss, align 32
   %temp1 = getelementptr %struct.ss, %struct.ss* %S, i32 0, i32 0
   store i32 1, i32* %temp1, align 4
   %temp4 = getelementptr %struct.ss, %struct.ss* %S, i32 0, i32 1
   store i64 2, i64* %temp4, align 8
   call void @f(ptr byval(ptr) align 4 %S) nounwind
   call void @g(ptr byval(ptr) align 4 %S) nounwind
   call void @h(ptr byval(ptr) align 4 %S) nounwind
   call void @k(ptr byval(ptr) align 4 %S) nounwind
   call void @l(ptr byval(ptr) align 4 %S) nounwind
   ret i32 0
 }

llvm/test/Transforms/PhaseOrdering/dce-after-argument-promotion-loads.ll

-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
 ; RUN: opt -O3 -S < %s | FileCheck %s
 
 ; Arg promotion eliminates the struct argument, and eliminates dead arguments, but introduces and leaves dead loads of the eliminated dead arg in callers
 
 %struct.ss = type { ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr }
 
 define internal void @phantomLoad(ptr %p, ptr %y, ptr %x) {
 entry:
   %0 = load i32, ptr %x
   store i32 %0, ptr %y
   ret void
 }
 
 define ptr @parent(ptr align 8 dereferenceable(72) %f, i16 %val1, i16 %val2, i32 %val3) align 2 {
-; CHECK-LABEL: define {{[^@]+}}@parent
+; CHECK-LABEL: define nonnull ptr @parent
 ; CHECK-SAME: (ptr readonly returned align 8 dereferenceable(72) [[F:%.*]], i16 [[VAL1:%.*]], i16 [[VAL2:%.*]], i32 [[VAL3:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] align 2 {
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[F]], i64 64
+; CHECK-NEXT:    [[F_VAL:%.*]] = load ptr, ptr [[TMP0]], align 8
 ; CHECK-NEXT:    [[CMP_NOT_NOT_I:%.*]] = icmp eq i32 [[VAL3]], 0
-; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[F]], i64 0, i32 8
-; CHECK-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[TMP0]], align 8
-; CHECK-NEXT:    br i1 [[CMP_NOT_NOT_I]], label [[IF_THEN_I:%.*]], label [[IF_ELSE_I:%.*]]
-; CHECK:       if.then.i:
-; CHECK-NEXT:    store i16 [[VAL1]], ptr [[TMP1]], align 2
-; CHECK-NEXT:    [[ADD_PTR_I_I_I_I_I:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 16
-; CHECK-NEXT:    br label [[BADCHILD_EXIT:%.*]]
-; CHECK:       if.else.i:
-; CHECK-NEXT:    [[ADD_PTR_I_I_I_I7_I:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 16
-; CHECK-NEXT:    store i16 [[VAL1]], ptr [[ADD_PTR_I_I_I_I7_I]], align 2
-; CHECK-NEXT:    br label [[BADCHILD_EXIT]]
-; CHECK:       badChild.exit:
-; CHECK-NEXT:    [[DOTSINK_I:%.*]] = phi ptr [ [[TMP1]], [[IF_ELSE_I]] ], [ [[ADD_PTR_I_I_I_I_I]], [[IF_THEN_I]] ]
-; CHECK-NEXT:    store i16 [[VAL2]], ptr [[DOTSINK_I]], align 2
+; CHECK-NEXT:    [[SPEC_SELECT_I:%.*]] = select i1 [[CMP_NOT_NOT_I]], i16 [[VAL1]], i16 [[VAL2]]
+; CHECK-NEXT:    [[SPEC_SELECT2_I:%.*]] = select i1 [[CMP_NOT_NOT_I]], i16 [[VAL2]], i16 [[VAL1]]
+; CHECK-NEXT:    store i16 [[SPEC_SELECT_I]], ptr [[F_VAL]], align 2
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[F_VAL]], i64 16
+; CHECK-NEXT:    store i16 [[SPEC_SELECT2_I]], ptr [[TMP1]], align 2
 ; CHECK-NEXT:    ret ptr [[F]]
 ;
 entry:
   call void @badChild(ptr align 8 dereferenceable(72) %f, i16 %val1, i16 %val2, i32 %val3) #4
   ret ptr %f
 }
 
 define internal void @badChild(ptr align 8 dereferenceable(72) %this, i16 %val1, i16 %val2, i32 %val3) align 2 {