[clang 17.0.1] [regression] Hangs when compiling NSS with -O2 -march=skylake

[StormBytePP]

When using clang 17.0.1 in Gentoo and compiling dev-libs/nss an infinite loop is detected in sha512.c compilation unit (reported downstream Gentoo bug report) which is not reproducible using Clang 16.

An initial testing revealed that march=native (also tested march=skylake and march=alderlake triggering the problem) might be causing the trouble.

In order to ease the test process, I attach the preprocessed output for that compile unit which I expect can reproduce this issue easily (remove the .txt extension)

sha512-preprocessed.c.txt

[thesamesam]

I can reproduce it with:

clang -O2 -march=skylake sha512.i

If I attach gdb to the process after it's been running for a little while:

0x00007f54f0f4cab2 in llvm::APInt::maskBit (bitPosition=<optimized out>) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/include/llvm/ADT/APInt.h:1900
1900        return 1ULL << whichBit(bitPosition);
(gdb) bt
#0  0x00007f54f0f4cab2 in llvm::APInt::maskBit (bitPosition=<optimized out>) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/include/llvm/ADT/APInt.h:1900
#1  llvm::APInt::operator[] (bitPosition=<optimized out>, this=<optimized out>) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/include/llvm/ADT/APInt.h:1017
#2  llvm::APInt::isSignBitSet (this=0x7ffefaa604e0) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/include/llvm/ADT/APInt.h:319
#3  llvm::KnownBits::isNegative (this=0x7ffefaa604d0) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/include/llvm/Support/KnownBits.h:96
#4  llvm::KnownBits::computeForAddSub (Add=<optimized out>, NSW=NSW@entry=false, LHS=..., RHS=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/Support/KnownBits.cpp:72
#5  0x00007f54f1c5bfa8 in llvm::TargetLowering::SimplifyDemandedBits (this=0x565551705ff0, Op=..., OriginalDemandedBits=..., OriginalDemandedElts=..., Known=..., TLO=..., Depth=2, AssumeSingleUse=false)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp:2740
#6  0x00007f54f1c5e41d in llvm::TargetLowering::SimplifyDemandedBits (this=this@entry=0x565551705ff0, Op=..., OriginalDemandedBits=..., OriginalDemandedElts=..., Known=..., TLO=..., Depth=1,
    AssumeSingleUse=false) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp:1225
#7  0x00007f54f502dc6d in llvm::X86TargetLowering::SimplifyDemandedBitsForTargetNode (this=0x565551705ff0, Op=..., OriginalDemandedBits=..., OriginalDemandedElts=..., Known=..., TLO=..., Depth=0)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/Target/X86/X86ISelLowering.cpp:44365
#8  0x00007f54f1c60533 in llvm::TargetLowering::SimplifyDemandedBits (this=0x565551705ff0, Op=..., OriginalDemandedBits=..., OriginalDemandedElts=..., Known=..., TLO=..., Depth=0, AssumeSingleUse=false)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp:2754
#9  0x00007f54f1c6bf57 in llvm::TargetLowering::SimplifyDemandedBits (this=0x565551705ff0, Op=..., DemandedBits=..., Known=..., TLO=..., Depth=0, AssumeSingleUse=false)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp:646
#10 0x00007f54f1c6c0f2 in llvm::TargetLowering::SimplifyDemandedBits (this=this@entry=0x565551705ff0, Op=..., DemandedBits=..., DCI=...)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp:608
#11 0x00007f54f5135377 in combineVectorShiftImm (N=<optimized out>, DAG=..., DCI=..., Subtarget=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/include/llvm/CodeGen/SelectionDAGNodes.h:1139
#12 0x00007f54f19d47c4 in (anonymous namespace)::DAGCombiner::combine (this=this@entry=0x7ffefaa63130, N=N@entry=0x56555191d900)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp:2049
#13 0x00007f54f19d62f9 in (anonymous namespace)::DAGCombiner::Run (AtLevel=llvm::AfterLegalizeDAG, this=0x7ffefaa63130)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp:1827
#14 llvm::SelectionDAG::Combine (this=<optimized out>, Level=Level@entry=llvm::AfterLegalizeDAG, AA=<optimized out>, OptLevel=<optimized out>)
    at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp:27592
#15 0x00007f54f1bfbb63 in llvm::SelectionDAGISel::CodeGenAndEmitDAG (this=0x565551829cb0) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp:925
#16 0x00007f54f1c00fc0 in llvm::SelectionDAGISel::SelectAllBasicBlocks (this=0x565551829cb0, Fn=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp:1700
#17 0x00007f54f1c027d6 in llvm::SelectionDAGISel::runOnMachineFunction (this=this@entry=0x565551829cb0, mf=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp:482
#18 0x00007f54f4fd4619 in (anonymous namespace)::X86DAGToDAGISel::runOnMachineFunction (this=0x565551829cb0, MF=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp:191
#19 0x00007f54f1547cd4 in llvm::MachineFunctionPass::runOnFunction (this=0x565551829cb0, F=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/CodeGen/MachineFunctionPass.cpp:91
#20 0x00007f54f11bdb13 in llvm::FPPassManager::runOnFunction (this=0x565551826a40, F=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/IR/LegacyPassManager.cpp:1435
#21 0x00007f54f11bdd51 in llvm::FPPassManager::runOnModule (this=0x565551826a40, M=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/IR/LegacyPassManager.cpp:1481
#22 0x00007f54f11be7f4 in (anonymous namespace)::MPPassManager::runOnModule (M=..., this=<optimized out>) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/IR/LegacyPassManager.cpp:1550
#23 llvm::legacy::PassManagerImpl::run (this=0x5655517a7120, M=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/IR/LegacyPassManager.cpp:535
#24 0x00007f54fb97774a in (anonymous namespace)::EmitAssemblyHelper::RunCodegenPipeline (DwoOS=<synthetic pointer>std::unique_ptr<llvm::ToolOutputFile> = {...},
    OS=std::unique_ptr<llvm::raw_pwrite_stream> = {...}, Action=clang::Backend_EmitObj, this=0x7ffefaa64b60) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/CodeGen/BackendUtil.cpp:1115
#25 (anonymous namespace)::EmitAssemblyHelper::EmitAssembly (OS=std::unique_ptr<llvm::raw_pwrite_stream> = {...}, Action=clang::Backend_EmitObj, this=0x7ffefaa64b60)
    at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/CodeGen/BackendUtil.cpp:1137
#26 clang::EmitBackendOutput (Diags=..., HeaderOpts=..., CGOpts=..., TOpts=..., LOpts=..., TDesc=..., M=M@entry=0x5655513bb820, Action=clang::Backend_EmitObj, VFS=...,
    OS=std::unique_ptr<llvm::raw_pwrite_stream> = {...}) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/CodeGen/BackendUtil.cpp:1299
#27 0x00007f54fbe13f15 in clang::BackendConsumer::HandleTranslationUnit (this=0x5655513b64d0, C=...) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/CodeGen/CodeGenAction.cpp:386
#28 0x00007f54fa41aa55 in clang::ParseAST (S=..., PrintStats=false, SkipFunctionBodies=<optimized out>) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Parse/ParseAST.cpp:176
#29 0x00007f54fca51be9 in clang::FrontendAction::Execute (this=this@entry=0x5655513b6b60) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Frontend/FrontendAction.cpp:1059
#30 0x00007f54fc9de17b in clang::CompilerInstance::ExecuteAction (this=this@entry=0x5655513ada00, Act=...) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Frontend/CompilerInstance.cpp:1053
#31 0x00007f54fcaebb2b in clang::ExecuteCompilerInvocation (Clang=Clang@entry=0x5655513ada00) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/FrontendTool/ExecuteCompilerInvocation.cpp:272
#32 0x0000565550c63035 in cc1_main (Argv=..., Argv0=0x5655513a4f70 "/usr/lib/llvm/17/bin/clang-17", MainAddr=MainAddr@entry=0x565550c5c270 <GetExecutablePath[abi:cxx11](char const*, bool)>)
    at /usr/src/debug/sys-devel/clang-17.0.1/clang/tools/driver/cc1_main.cpp:249
#33 0x0000565550c5bcab in ExecuteCC1Tool (ArgV=..., ToolContext=...) at /usr/src/debug/sys-devel/clang-17.0.1/clang/tools/driver/driver.cpp:366
#34 0x00007f54fc5d826d in llvm::function_ref<int (llvm::SmallVectorImpl<char const*>&)>::operator()(llvm::SmallVectorImpl<char const*>&) const (params#0=..., this=<optimized out>)
    at /usr/lib/llvm/17/include/llvm/ADT/STLFunctionalExtras.h:68
#35 operator() (__closure=0x7ffefaa65f40) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Driver/Job.cpp:440
#36 llvm::function_ref<void()>::callback_fn<clang::driver::CC1Command::Execute(llvm::ArrayRef<std::optional<llvm::StringRef> >, std::string*, bool*) const::<lambda()> >(intptr_t) (
    callable=callable@entry=140733103628048) at /usr/lib/llvm/17/include/llvm/ADT/STLFunctionalExtras.h:45
#37 0x00007f54f0f19b5e in llvm::function_ref<void ()>::operator()() const (this=<synthetic pointer>) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/include/llvm/ADT/STLFunctionalExtras.h:68
#38 llvm::CrashRecoveryContext::RunSafely(llvm::function_ref<void ()>) (this=this@entry=0x7ffefaa65ef0, Fn=...) at /usr/src/debug/sys-devel/llvm-17.0.1/llvm/lib/Support/CrashRecoveryContext.cpp:426
#39 0x00007f54fc5dac70 in clang::driver::CC1Command::Execute (this=0x565551345560, Redirects=..., ErrMsg=<optimized out>, ExecutionFailed=<optimized out>)
    at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Driver/Job.cpp:440
#40 0x00007f54fc598def in clang::driver::Compilation::ExecuteCommand (this=0x5655513a7550, C=..., FailingCommand=@0x7ffefaa66470: 0x0, LogOnly=<optimized out>)
    at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Driver/Compilation.cpp:199
#41 0x00007f54fc5995f6 in clang::driver::Compilation::ExecuteJobs (this=this@entry=0x5655513a7550, Jobs=..., FailingCommands=..., LogOnly=LogOnly@entry=false)
    at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Driver/Compilation.cpp:253
#42 0x00007f54fc5a87a4 in clang::driver::Driver::ExecuteCompilation (this=this@entry=0x7ffefaa668f0, C=..., FailingCommands=...) at /usr/src/debug/sys-devel/clang-17.0.1/clang/lib/Driver/Driver.cpp:1903
#43 0x0000565550c600b0 in clang_main (Argc=<optimized out>, Argv=<optimized out>, ToolContext=...) at /usr/src/debug/sys-devel/clang-17.0.1/clang/tools/driver/driver.cpp:542
#44 0x0000565550c59cc7 in main (argc=<optimized out>, argv=<optimized out>) at /usr/src/debug/sys-devel/clang-17.0.1/x/y/clang-abi_x86_64.amd64/tools/driver/clang-driver.cpp:15

[thesamesam]

I've reduced it two ways.

1. Using cvise on the C reproducer:

#!/bin/sh
set -x

# The clang-16 one should build fine (and quickly, but don't bother checking that yet).
clang-16 -O2 -Werror=return-type -march=skylake -c sha512.c.i -S -o /dev/null || exit 1

timeout 45s clang-17 -O2 -Werror=return-type -march=skylake -c sha512.c.i -S -o /dev/null
ret=$?

case ${ret} in
        0)
                # If it built fine, it's uninteresting.
                exit 1
                ;;
        124)
                # It timed out, yay!
                exit 0
                ;;
        *)
                # It failed in some way but not a timeout, not interesting.
                exit 1
                ;;
esac

This gives the following which takes 1m5s w/ clang 17 on a fast machine (it completes in 0.087s w/ clang 16) which is hopefully representative enough:

    typedef unsigned int PRUint32;
        typedef struct SHA256ContextStr SHA256Context;
        typedef struct {
       }
        mp_int;
        struct SHA256ContextStr {
           union {
              PRUint32 w[64];
          }
       u;
       };
                  static const PRUint32 K256[64] __attribute__((aligned(16))) = {
           0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5,     0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,     0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3,     0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,     0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc,     0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,     0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7,     0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,     0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13,     0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,     0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3,     0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,     0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5,     0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,     0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208,     0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2 };
          void SHA256_Begin(SHA256Context *ctx) {
           {
              ctx->u.w[27] = ((((ctx->u.w[27 - 2] >> 17) | (ctx->u.w[27 - 2] << ((8 * sizeof ctx->u.w[27 - 2]) - 17))) ^ ((ctx->u.w[27 - 2] >> 19) | (ctx->u.w[27 - 2] << ((8 * sizeof ctx->u.w[27 - 2]) - 19))) ^ (ctx->u.w[27 - 2] >> 10)) + ctx->u.w[27 - 7] + (((ctx->u.w[27 - 15] >> 7) | (ctx->u.w[27 - 15] << ((8 * sizeof ctx->u.w[27 - 15]) - 7))) ^ ((ctx->u.w[27 - 15] >> 18) | (ctx->u.w[27 - 15] << ((8 * sizeof ctx->u.w[27 - 15]) - 18))) ^ (ctx->u.w[27 - 15] >> 3)) + ctx->u.w[27 - 16]);
              ctx->u.w[28] = ((((ctx->u.w[28 - 2] >> 17) | (ctx->u.w[28 - 2] << ((8 * sizeof ctx->u.w[28 - 2]) - 17))) ^ ((ctx->u.w[28 - 2] >> 19) | (ctx->u.w[28 - 2] << ((8 * sizeof ctx->u.w[28 - 2]) - 19))) ^ (ctx->u.w[28 - 2] >> 10)) + ctx->u.w[28 - 7] + (((ctx->u.w[28 - 15] >> 7) | (ctx->u.w[28 - 15] << ((8 * sizeof ctx->u.w[28 - 15]) - 7))) ^ ((ctx->u.w[28 - 15] >> 18) | (ctx->u.w[28 - 15] << ((8 * sizeof ctx->u.w[28 - 15]) - 18))) ^ (ctx->u.w[28 - 15] >> 3)) + ctx->u.w[28 - 16]);
              ctx->u.w[29] = ((((ctx->u.w[29 - 2] >> 17) | (ctx->u.w[29 - 2] << ((8 * sizeof ctx->u.w[29 - 2]) - 17))) ^ ((ctx->u.w[29 - 2] >> 19) | (ctx->u.w[29 - 2] << ((8 * sizeof ctx->u.w[29 - 2]) - 19))) ^ (ctx->u.w[29 - 2] >> 10)) + ctx->u.w[29 - 7] + (((ctx->u.w[29 - 15] >> 7) | (ctx->u.w[29 - 15] << ((8 * sizeof ctx->u.w[29 - 15]) - 7))) ^ ((ctx->u.w[29 - 15] >> 18) | (ctx->u.w[29 - 15] << ((8 * sizeof ctx->u.w[29 - 15]) - 18))) ^ (ctx->u.w[29 - 15] >> 3)) + ctx->u.w[29 - 16]);
              ctx->u.w[30] = ((((ctx->u.w[30 - 2] >> 17) | (ctx->u.w[30 - 2] << ((8 * sizeof ctx->u.w[30 - 2]) - 17))) ^ ((ctx->u.w[30 - 2] >> 19) | (ctx->u.w[30 - 2] << ((8 * sizeof ctx->u.w[30 - 2]) - 19))) ^ (ctx->u.w[30 - 2] >> 10)) + ctx->u.w[30 - 7] + (((ctx->u.w[30 - 15] >> 7) | (ctx->u.w[30 - 15] << ((8 * sizeof ctx->u.w[30 - 15]) - 7))) ^ ((ctx->u.w[30 - 15] >> 18) | (ctx->u.w[30 - 15] << ((8 * sizeof ctx->u.w[30 - 15]) - 18))) ^ (ctx->u.w[30 - 15] >> 3)) + ctx->u.w[30 - 16]);
              ctx->u.w[31] = ((((ctx->u.w[31 - 2] >> 17) | (ctx->u.w[31 - 2] << ((8 * sizeof ctx->u.w[31 - 2]) - 17))) ^ ((ctx->u.w[31 - 2] >> 19) | (ctx->u.w[31 - 2] << ((8 * sizeof ctx->u.w[31 - 2]) - 19))) ^ (ctx->u.w[31 - 2] >> 10)) + ctx->u.w[31 - 7] + (((ctx->u.w[31 - 15] >> 7) | (ctx->u.w[31 - 15] << ((8 * sizeof ctx->u.w[31 - 15]) - 7))) ^ ((ctx->u.w[31 - 15] >> 18) | (ctx->u.w[31 - 15] << ((8 * sizeof ctx->u.w[31 - 15]) - 18))) ^ (ctx->u.w[31 - 15] >> 3)) + ctx->u.w[31 - 16]);
              ctx->u.w[32] = ((((ctx->u.w[32 - 2] >> 17) | (ctx->u.w[32 - 2] << ((8 * sizeof ctx->u.w[32 - 2]) - 17))) ^ ((ctx->u.w[32 - 2] >> 19) | (ctx->u.w[32 - 2] << ((8 * sizeof ctx->u.w[32 - 2]) - 19))) ^ (ctx->u.w[32 - 2] >> 10)) + ctx->u.w[32 - 7] + (((ctx->u.w[32 - 15] >> 7) | (ctx->u.w[32 - 15] << ((8 * sizeof ctx->u.w[32 - 15]) - 7))) ^ ((ctx->u.w[32 - 15] >> 18) | (ctx->u.w[32 - 15] << ((8 * sizeof ctx->u.w[32 - 15]) - 18))) ^ (ctx->u.w[32 - 15] >> 3)) + ctx->u.w[32 - 16]);
              ctx->u.w[33] = ((((ctx->u.w[33 - 2] >> 17) | (ctx->u.w[33 - 2] << ((8 * sizeof ctx->u.w[33 - 2]) - 17))) ^ ((ctx->u.w[33 - 2] >> 19) | (ctx->u.w[33 - 2] << ((8 * sizeof ctx->u.w[33 - 2]) - 19))) ^ (ctx->u.w[33 - 2] >> 10)) + ctx->u.w[33 - 7] + (((ctx->u.w[33 - 15] >> 7) | (ctx->u.w[33 - 15] << ((8 * sizeof ctx->u.w[33 - 15]) - 7))) ^ ((ctx->u.w[33 - 15] >> 18) | (ctx->u.w[33 - 15] << ((8 * sizeof ctx->u.w[33 - 15]) - 18))) ^ (ctx->u.w[33 - 15] >> 3)) + ctx->u.w[33 - 16]);
              ctx->u.w[36] = ((((ctx->u.w[36 - 2] >> 17) | (ctx->u.w[36 - 2] << ((8 * sizeof ctx->u.w[36 - 2]) - 17))) ^ ((ctx->u.w[36 - 2] >> 19) | (ctx->u.w[36 - 2] << ((8 * sizeof ctx->u.w[36 - 2]) - 19))) ^ (ctx->u.w[36 - 2] >> 10)) + ctx->u.w[36 - 7] + (((ctx->u.w[36 - 15] >> 7) | (ctx->u.w[36 - 15] << ((8 * sizeof ctx->u.w[36 - 15]) - 7))) ^ ((ctx->u.w[36 - 15] >> 18) | (ctx->u.w[36 - 15] << ((8 * sizeof ctx->u.w[36 - 15]) - 18))) ^ (ctx->u.w[36 - 15] >> 3)) + ctx->u.w[36 - 16]);
              ctx->u.w[37] = ((((ctx->u.w[37 - 2] >> 17) | (ctx->u.w[37 - 2] << ((8 * sizeof ctx->u.w[37 - 2]) - 17))) ^ ((ctx->u.w[37 - 2] >> 19) | (ctx->u.w[37 - 2] << ((8 * sizeof ctx->u.w[37 - 2]) - 19))) ^ (ctx->u.w[37 - 2] >> 10)) + ctx->u.w[37 - 7] + (((ctx->u.w[37 - 15] >> 7) | (ctx->u.w[37 - 15] << ((8 * sizeof ctx->u.w[37 - 15]) - 7))) ^ ((ctx->u.w[37 - 15] >> 18) | (ctx->u.w[37 - 15] << ((8 * sizeof ctx->u.w[37 - 15]) - 18))) ^ (ctx->u.w[37 - 15] >> 3)) + ctx->u.w[37 - 16]);
              ctx->u.w[38] = ((((ctx->u.w[38 - 2] >> 17) | (ctx->u.w[38 - 2] << ((8 * sizeof ctx->u.w[38 - 2]) - 17))) ^ ((ctx->u.w[38 - 2] >> 19) | (ctx->u.w[38 - 2] << ((8 * sizeof ctx->u.w[38 - 2]) - 19))) ^ (ctx->u.w[38 - 2] >> 10)) + ctx->u.w[38 - 7] + (((ctx->u.w[38 - 15] >> 7) | (ctx->u.w[38 - 15] << ((8 * sizeof ctx->u.w[38 - 15]) - 7))) ^ ((ctx->u.w[38 - 15] >> 18) | (ctx->u.w[38 - 15] << ((8 * sizeof ctx->u.w[38 - 15]) - 18))) ^ (ctx->u.w[38 - 15] >> 3)) + ctx->u.w[38 - 16]);
              ctx->u.w[39] = ((((ctx->u.w[39 - 2] >> 17) | (ctx->u.w[39 - 2] << ((8 * sizeof ctx->u.w[39 - 2]) - 17))) ^ ((ctx->u.w[39 - 2] >> 19) | (ctx->u.w[39 - 2] << ((8 * sizeof ctx->u.w[39 - 2]) - 19))) ^ (ctx->u.w[39 - 2] >> 10)) + ctx->u.w[39 - 7] + (((ctx->u.w[39 - 15] >> 7) | (ctx->u.w[39 - 15] << ((8 * sizeof ctx->u.w[39 - 15]) - 7))) ^ ((ctx->u.w[39 - 15] >> 18) | (ctx->u.w[39 - 15] << ((8 * sizeof ctx->u.w[39 - 15]) - 18))) ^ (ctx->u.w[39 - 15] >> 3)) + ctx->u.w[39 - 16]);
              ctx->u.w[40] = ((((ctx->u.w[40 - 2] >> 17) | (ctx->u.w[40 - 2] << ((8 * sizeof ctx->u.w[40 - 2]) - 17))) ^ ((ctx->u.w[40 - 2] >> 19) | (ctx->u.w[40 - 2] << ((8 * sizeof ctx->u.w[40 - 2]) - 19))) ^ (ctx->u.w[40 - 2] >> 10)) + ctx->u.w[40 - 7] + (((ctx->u.w[40 - 15] >> 7) | (ctx->u.w[40 - 15] << ((8 * sizeof ctx->u.w[40 - 15]) - 7))) ^ ((ctx->u.w[40 - 15] >> 18) | (ctx->u.w[40 - 15] << ((8 * sizeof ctx->u.w[40 - 15]) - 18))) ^ (ctx->u.w[40 - 15] >> 3)) + ctx->u.w[40 - 16]);
              ctx->u.w[41] = ((((ctx->u.w[41 - 2] >> 17) | (ctx->u.w[41 - 2] << ((8 * sizeof ctx->u.w[41 - 2]) - 17))) ^ ((ctx->u.w[41 - 2] >> 19) | (ctx->u.w[41 - 2] << ((8 * sizeof ctx->u.w[41 - 2]) - 19))) ^ (ctx->u.w[41 - 2] >> 10)) + ctx->u.w[41 - 7] + (((ctx->u.w[41 - 15] >> 7) | (ctx->u.w[41 - 15] << ((8 * sizeof ctx->u.w[41 - 15]) - 7))) ^ ((ctx->u.w[41 - 15] >> 18) | (ctx->u.w[41 - 15] << ((8 * sizeof ctx->u.w[41 - 15]) - 18))) ^ (ctx->u.w[41 - 15] >> 3)) + ctx->u.w[41 - 16]);
              ctx->u.w[43] = ((((ctx->u.w[43 - 2] >> 17) | (ctx->u.w[43 - 2] << ((8 * sizeof ctx->u.w[43 - 2]) - 17))) ^ ((ctx->u.w[43 - 2] >> 19) | (ctx->u.w[43 - 2] << ((8 * sizeof ctx->u.w[43 - 2]) - 19))) ^ (ctx->u.w[43 - 2] >> 10)) + ctx->u.w[43 - 7] + (((ctx->u.w[43 - 15] >> 7) | (ctx->u.w[43 - 15] << ((8 * sizeof ctx->u.w[43 - 15]) - 7))) ^ ((ctx->u.w[43 - 15] >> 18) | (ctx->u.w[43 - 15] << ((8 * sizeof ctx->u.w[43 - 15]) - 18))) ^ (ctx->u.w[43 - 15] >> 3)) + ctx->u.w[43 - 16]);
              ctx->u.w[45] = ((((ctx->u.w[45 - 2] >> 17) | (ctx->u.w[45 - 2] << ((8 * sizeof ctx->u.w[45 - 2]) - 17))) ^ ((ctx->u.w[45 - 2] >> 19) | (ctx->u.w[45 - 2] << ((8 * sizeof ctx->u.w[45 - 2]) - 19))) ^ (ctx->u.w[45 - 2] >> 10)) + ctx->u.w[45 - 7] + (((ctx->u.w[45 - 15] >> 7) | (ctx->u.w[45 - 15] << ((8 * sizeof ctx->u.w[45 - 15]) - 7))) ^ ((ctx->u.w[45 - 15] >> 18) | (ctx->u.w[45 - 15] << ((8 * sizeof ctx->u.w[45 - 15]) - 18))) ^ (ctx->u.w[45 - 15] >> 3)) + ctx->u.w[45 - 16]);
              ctx->u.w[46] = ((((ctx->u.w[46 - 2] >> 17) | (ctx->u.w[46 - 2] << ((8 * sizeof ctx->u.w[46 - 2]) - 17))) ^ ((ctx->u.w[46 - 2] >> 19) | (ctx->u.w[46 - 2] << ((8 * sizeof ctx->u.w[46 - 2]) - 19))) ^ (ctx->u.w[46 - 2] >> 10)) + ctx->u.w[46 - 7] + (((ctx->u.w[46 - 15] >> 7) | (ctx->u.w[46 - 15] << ((8 * sizeof ctx->u.w[46 - 15]) - 7))) ^ ((ctx->u.w[46 - 15] >> 18) | (ctx->u.w[46 - 15] << ((8 * sizeof ctx->u.w[46 - 15]) - 18))) ^ (ctx->u.w[46 - 15] >> 3)) + ctx->u.w[46 - 16]);
              ctx->u.w[47] = ((((ctx->u.w[47 - 2] >> 17) | (ctx->u.w[47 - 2] << ((8 * sizeof ctx->u.w[47 - 2]) - 17))) ^ ((ctx->u.w[47 - 2] >> 19) | (ctx->u.w[47 - 2] << ((8 * sizeof ctx->u.w[47 - 2]) - 19))) ^ (ctx->u.w[47 - 2] >> 10)) + ctx->u.w[47 - 7] + (((ctx->u.w[47 - 15] >> 7) | (ctx->u.w[47 - 15] << ((8 * sizeof ctx->u.w[47 - 15]) - 7))) ^ ((ctx->u.w[47 - 15] >> 18) | (ctx->u.w[47 - 15] << ((8 * sizeof ctx->u.w[47 - 15]) - 18))) ^ (ctx->u.w[47 - 15] >> 3)) + ctx->u.w[47 - 16]);
              ctx->u.w[48] = ((((ctx->u.w[48 - 2] >> 17) | (ctx->u.w[48 - 2] << ((8 * sizeof ctx->u.w[48 - 2]) - 17))) ^ ((ctx->u.w[48 - 2] >> 19) | (ctx->u.w[48 - 2] << ((8 * sizeof ctx->u.w[48 - 2]) - 19))) ^ (ctx->u.w[48 - 2] >> 10)) + ctx->u.w[48 - 7] + (((ctx->u.w[48 - 15] >> 7) | (ctx->u.w[48 - 15] << ((8 * sizeof ctx->u.w[48 - 15]) - 7))) ^ ((ctx->u.w[48 - 15] >> 18) | (ctx->u.w[48 - 15] << ((8 * sizeof ctx->u.w[48 - 15]) - 18))) ^ (ctx->u.w[48 - 15] >> 3)) + ctx->u.w[48 - 16]);
              ctx->u.w[49] = ((((ctx->u.w[49 - 2] >> 17) | (ctx->u.w[49 - 2] << ((8 * sizeof ctx->u.w[49 - 2]) - 17))) ^ ((ctx->u.w[49 - 2] >> 19) | (ctx->u.w[49 - 2] << ((8 * sizeof ctx->u.w[49 - 2]) - 19))) ^ (ctx->u.w[49 - 2] >> 10)) + ctx->u.w[49 - 7] + (((ctx->u.w[49 - 15] >> 7) | (ctx->u.w[49 - 15] << ((8 * sizeof ctx->u.w[49 - 15]) - 7))) ^ ((ctx->u.w[49 - 15] >> 18) | (ctx->u.w[49 - 15] << ((8 * sizeof ctx->u.w[49 - 15]) - 18))) ^ (ctx->u.w[49 - 15] >> 3)) + ctx->u.w[49 - 16]);
              ctx->u.w[50] = ((((ctx->u.w[50 - 2] >> 17) | (ctx->u.w[50 - 2] << ((8 * sizeof ctx->u.w[50 - 2]) - 17))) ^ ((ctx->u.w[50 - 2] >> 19) | (ctx->u.w[50 - 2] << ((8 * sizeof ctx->u.w[50 - 2]) - 19))) ^ (ctx->u.w[50 - 2] >> 10)) + ctx->u.w[50 - 7] + (((ctx->u.w[50 - 15] >> 7) | (ctx->u.w[50 - 15] << ((8 * sizeof ctx->u.w[50 - 15]) - 7))) ^ ((ctx->u.w[50 - 15] >> 18) | (ctx->u.w[50 - 15] << ((8 * sizeof ctx->u.w[50 - 15]) - 18))) ^ (ctx->u.w[50 - 15] >> 3)) + ctx->u.w[50 - 16]);
              ctx->u.w[51] = ((((ctx->u.w[51 - 2] >> 17) | (ctx->u.w[51 - 2] << ((8 * sizeof ctx->u.w[51 - 2]) - 17))) ^ ((ctx->u.w[51 - 2] >> 19) | (ctx->u.w[51 - 2] << ((8 * sizeof ctx->u.w[51 - 2]) - 19))) ^ (ctx->u.w[51 - 2] >> 10)) + ctx->u.w[51 - 7] + (((ctx->u.w[51 - 15] >> 7) | (ctx->u.w[51 - 15] << ((8 * sizeof ctx->u.w[51 - 15]) - 7))) ^ ((ctx->u.w[51 - 15] >> 18) | (ctx->u.w[51 - 15] << ((8 * sizeof ctx->u.w[51 - 15]) - 18))) ^ (ctx->u.w[51 - 15] >> 3)) + ctx->u.w[51 - 16]);
              ctx->u.w[52] = ((((ctx->u.w[52 - 2] >> 17) | (ctx->u.w[52 - 2] << ((8 * sizeof ctx->u.w[52 - 2]) - 17))) ^ ((ctx->u.w[52 - 2] >> 19) | (ctx->u.w[52 - 2] << ((8 * sizeof ctx->u.w[52 - 2]) - 19))) ^ (ctx->u.w[52 - 2] >> 10)) + ctx->u.w[52 - 7] + (((ctx->u.w[52 - 15] >> 7) | (ctx->u.w[52 - 15] << ((8 * sizeof ctx->u.w[52 - 15]) - 7))) ^ ((ctx->u.w[52 - 15] >> 18) | (ctx->u.w[52 - 15] << ((8 * sizeof ctx->u.w[52 - 15]) - 18))) ^ (ctx->u.w[52 - 15] >> 3)) + ctx->u.w[52 - 16]);
              ctx->u.w[53] = ((((ctx->u.w[53 - 2] >> 17) | (ctx->u.w[53 - 2] << ((8 * sizeof ctx->u.w[53 - 2]) - 17))) ^ ((ctx->u.w[53 - 2] >> 19) | (ctx->u.w[53 - 2] << ((8 * sizeof ctx->u.w[53 - 2]) - 19))) ^ (ctx->u.w[53 - 2] >> 10)) + ctx->u.w[53 - 7] + (((ctx->u.w[53 - 15] >> 7) | (ctx->u.w[53 - 15] << ((8 * sizeof ctx->u.w[53 - 15]) - 7))) ^ ((ctx->u.w[53 - 15] >> 18) | (ctx->u.w[53 - 15] << ((8 * sizeof ctx->u.w[53 - 15]) - 18))) ^ (ctx->u.w[53 - 15] >> 3)) + ctx->u.w[53 - 16]);
              ctx->u.w[54] = ((((ctx->u.w[54 - 2] >> 17) | (ctx->u.w[54 - 2] << ((8 * sizeof ctx->u.w[54 - 2]) - 17))) ^ ((ctx->u.w[54 - 2] >> 19) | (ctx->u.w[54 - 2] << ((8 * sizeof ctx->u.w[54 - 2]) - 19))) ^ (ctx->u.w[54 - 2] >> 10)) + ctx->u.w[54 - 7] + (((ctx->u.w[54 - 15] >> 7) | (ctx->u.w[54 - 15] << ((8 * sizeof ctx->u.w[54 - 15]) - 7))) ^ ((ctx->u.w[54 - 15] >> 18) | (ctx->u.w[54 - 15] << ((8 * sizeof ctx->u.w[54 - 15]) - 18))) ^ (ctx->u.w[54 - 15] >> 3)) + ctx->u.w[54 - 16]);
              ctx->u.w[55] = ((((ctx->u.w[55 - 2] >> 17) | (ctx->u.w[55 - 2] << ((8 * sizeof ctx->u.w[55 - 2]) - 17))) ^ ((ctx->u.w[55 - 2] >> 19) | (ctx->u.w[55 - 2] << ((8 * sizeof ctx->u.w[55 - 2]) - 19))) ^ (ctx->u.w[55 - 2] >> 10)) + ctx->u.w[55 - 7] + (((ctx->u.w[55 - 15] >> 7) | (ctx->u.w[55 - 15] << ((8 * sizeof ctx->u.w[55 - 15]) - 7))) ^ ((ctx->u.w[55 - 15] >> 18) | (ctx->u.w[55 - 15] << ((8 * sizeof ctx->u.w[55 - 15]) - 18))) ^ (ctx->u.w[55 - 15] >> 3)) + ctx->u.w[55 - 16]);
              ctx->u.w[56] = ((((ctx->u.w[56 - 2] >> 17) | (ctx->u.w[56 - 2] << ((8 * sizeof ctx->u.w[56 - 2]) - 17))) ^ ((ctx->u.w[56 - 2] >> 19) | (ctx->u.w[56 - 2] << ((8 * sizeof ctx->u.w[56 - 2]) - 19))) ^ (ctx->u.w[56 - 2] >> 10)) + ctx->u.w[56 - 7] + (((ctx->u.w[56 - 15] >> 7) | (ctx->u.w[56 - 15] << ((8 * sizeof ctx->u.w[56 - 15]) - 7))) ^ ((ctx->u.w[56 - 15] >> 18) | (ctx->u.w[56 - 15] << ((8 * sizeof ctx->u.w[56 - 15]) - 18))) ^ (ctx->u.w[56 - 15] >> 3)) + ctx->u.w[56 - 16]);
              ctx->u.w[57] = ((((ctx->u.w[57 - 2] >> 17) | (ctx->u.w[57 - 2] << ((8 * sizeof ctx->u.w[57 - 2]) - 17))) ^ ((ctx->u.w[57 - 2] >> 19) | (ctx->u.w[57 - 2] << ((8 * sizeof ctx->u.w[57 - 2]) - 19))) ^ (ctx->u.w[57 - 2] >> 10)) + ctx->u.w[57 - 7] + (((ctx->u.w[57 - 15] >> 7) | (ctx->u.w[57 - 15] << ((8 * sizeof ctx->u.w[57 - 15]) - 7))) ^ ((ctx->u.w[57 - 15] >> 18) | (ctx->u.w[57 - 15] << ((8 * sizeof ctx->u.w[57 - 15]) - 18))) ^ (ctx->u.w[57 - 15] >> 3)) + ctx->u.w[57 - 16]);
              ctx->u.w[58] = ((((ctx->u.w[58 - 2] >> 17) | (ctx->u.w[58 - 2] << ((8 * sizeof ctx->u.w[58 - 2]) - 17))) ^ ((ctx->u.w[58 - 2] >> 19) | (ctx->u.w[58 - 2] << ((8 * sizeof ctx->u.w[58 - 2]) - 19))) ^ (ctx->u.w[58 - 2] >> 10)) + ctx->u.w[58 - 7] + (((ctx->u.w[58 - 15] >> 7) | (ctx->u.w[58 - 15] << ((8 * sizeof ctx->u.w[58 - 15]) - 7))) ^ ((ctx->u.w[58 - 15] >> 18) | (ctx->u.w[58 - 15] << ((8 * sizeof ctx->u.w[58 - 15]) - 18))) ^ (ctx->u.w[58 - 15] >> 3)) + ctx->u.w[58 - 16]);
              ctx->u.w[59] = ((((ctx->u.w[59 - 2] >> 17) | (ctx->u.w[59 - 2] << ((8 * sizeof ctx->u.w[59 - 2]) - 17))) ^ ((ctx->u.w[59 - 2] >> 19) | (ctx->u.w[59 - 2] << ((8 * sizeof ctx->u.w[59 - 2]) - 19))) ^ (ctx->u.w[59 - 2] >> 10)) + ctx->u.w[59 - 7] + (((ctx->u.w[59 - 15] >> 7) | (ctx->u.w[59 - 15] << ((8 * sizeof ctx->u.w[59 - 15]) - 7))) ^ ((ctx->u.w[59 - 15] >> 18) | (ctx->u.w[59 - 15] << ((8 * sizeof ctx->u.w[59 - 15]) - 18))) ^ (ctx->u.w[59 - 15] >> 3)) + ctx->u.w[59 - 16]);
              ctx->u.w[60] = ((((ctx->u.w[60 - 2] >> 17) | (ctx->u.w[60 - 2] << ((8 * sizeof ctx->u.w[60 - 2]) - 17))) ^ ((ctx->u.w[60 - 2] >> 19) | (ctx->u.w[60 - 2] << ((8 * sizeof ctx->u.w[60 - 2]) - 19))) ^ (ctx->u.w[60 - 2] >> 10)) + ctx->u.w[60 - 7] + (((ctx->u.w[60 - 15] >> 7) | (ctx->u.w[60 - 15] << ((8 * sizeof ctx->u.w[60 - 15]) - 7))) ^ ((ctx->u.w[60 - 15] >> 18) | (ctx->u.w[60 - 15] << ((8 * sizeof ctx->u.w[60 - 15]) - 18))) ^ (ctx->u.w[60 - 15] >> 3)) + ctx->u.w[60 - 16]);
          }
       }

2. Using llvm-reduce by following https://discourse.llvm.org/t/suggestions-to-debug-a-forever-running-clang-compile-command/60420/6

$ clang-17 -O2 -c sha512.c.i -march=skylake -emit-llvm -S
$ llc sha512.c.ll # confirm this hangs
$ llvm-reduce --test=test2.sh sha512.c.ll

with

#!/usr/bin/env bash
timeout 30s llc "$@"

ret=$?

case ${ret} in
        0)
                # If it built fine, it's uninteresting.
                exit 1
                ;;
        124)
                # It timed out, yay!
                exit 0
                ;;
        *)
                # It failed in some way but not a timeout, not interesting.
                exit 1
                ;;
esac

This kind of worked. The result takes ~30s for llc on a fast machine w/ llvm 17 (takes 0.049s w/ llvm 16) which I think is illustrative enough and should show that it'll take orders of magnitude longer for the full thing. I know it's not ideal though.

; ModuleID = '<bc file>'
source_filename = "sha512.c"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-pc-linux-gnu"

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #0

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #0

; Function Attrs: nounwind sspstrong memory(argmem: readwrite) uwtable
define void @SHA256_Compress_Generic(ptr noundef %ctx) #1 {
entry:
  %0 = load i32, ptr null, align 4
  %1 = tail call i32 asm "bswap $0", "=r,0,~{dirflag},~{fpsr},~{flags}"(i32 %0) #5
  %arrayidx14 = getelementptr inbounds [64 x i32], ptr %ctx, i64 0, i64 3
  %2 = load i32, ptr %arrayidx14, align 4
  %3 = tail call i32 asm "bswap $0", "=r,0,~{dirflag},~{fpsr},~{flags}"(i32 %2) #5
  %4 = insertelement <2 x i32> zeroinitializer, i32 %1, i64 1
  %5 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %4, <2 x i32> %4, <2 x i32> <i32 15, i32 15>)
  %6 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %4, <2 x i32> %4, <2 x i32> <i32 13, i32 13>)
  %7 = xor <2 x i32> %5, %6
  %8 = lshr <2 x i32> %4, zeroinitializer
  %9 = xor <2 x i32> %7, %8
  %10 = insertelement <2 x i32> zeroinitializer, i32 %3, i64 0
  %11 = shufflevector <2 x i32> zeroinitializer, <2 x i32> %10, <2 x i32> <i32 1, i32 2>
  %12 = add <2 x i32> %11, %9
  %13 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %12, <2 x i32> %12, <2 x i32> <i32 15, i32 15>)
  %14 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %12, <2 x i32> %12, <2 x i32> <i32 13, i32 13>)
  %15 = xor <2 x i32> %13, %14
  %16 = lshr <2 x i32> %12, zeroinitializer
  %17 = xor <2 x i32> %15, %16
  %18 = add <2 x i32> %4, %17
  %19 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %18, <2 x i32> %18, <2 x i32> <i32 15, i32 15>)
  %20 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %18, <2 x i32> %18, <2 x i32> <i32 13, i32 13>)
  %21 = xor <2 x i32> %19, %20
  %22 = lshr <2 x i32> %18, <i32 10, i32 10>
  %23 = xor <2 x i32> %21, %22
  %24 = add <2 x i32> %4, %23
  %25 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %24, <2 x i32> %24, <2 x i32> <i32 15, i32 15>)
  %26 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %24, <2 x i32> %24, <2 x i32> <i32 13, i32 13>)
  %27 = xor <2 x i32> %25, %26
  %28 = lshr <2 x i32> %24, <i32 10, i32 10>
  %29 = xor <2 x i32> %27, %28
  %30 = shufflevector <2 x i32> %4, <2 x i32> %12, <2 x i32> <i32 1, i32 2>
  %31 = add <2 x i32> %30, %29
  %32 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %31, <2 x i32> %31, <2 x i32> <i32 15, i32 15>)
  %33 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %31, <2 x i32> %31, <2 x i32> <i32 13, i32 13>)
  %34 = xor <2 x i32> %32, %33
  %35 = lshr <2 x i32> %31, <i32 10, i32 10>
  %36 = xor <2 x i32> %34, %35
  %37 = shufflevector <2 x i32> %12, <2 x i32> zeroinitializer, <2 x i32> <i32 1, i32 2>
  %38 = add <2 x i32> %37, %36
  %arrayidx918 = getelementptr inbounds [64 x i32], ptr %ctx, i64 0, i64 33
  store <2 x i32> %38, ptr %arrayidx918, align 4
  %arrayidx1012 = getelementptr inbounds [64 x i32], ptr %ctx, i64 0, i64 35
  %39 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %38, <2 x i32> %38, <2 x i32> <i32 15, i32 15>)
  %40 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %38, <2 x i32> %38, <2 x i32> <i32 13, i32 13>)
  %41 = xor <2 x i32> %39, %40
  %42 = lshr <2 x i32> %38, <i32 10, i32 10>
  %43 = xor <2 x i32> %41, %42
  %44 = add <2 x i32> %37, %43
  store <2 x i32> zeroinitializer, ptr %arrayidx1012, align 4
  %arrayidx1106 = getelementptr inbounds [64 x i32], ptr %ctx, i64 0, i64 37
  %45 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %44, <2 x i32> %44, <2 x i32> <i32 15, i32 15>)
  %46 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %44, <2 x i32> %44, <2 x i32> <i32 13, i32 13>)
  %47 = xor <2 x i32> %45, %46
  %48 = lshr <2 x i32> %44, <i32 10, i32 10>
  %49 = xor <2 x i32> %47, %48
  %50 = lshr <2 x i32> %24, zeroinitializer
  %51 = add <2 x i32> %50, %49
  store <2 x i32> %51, ptr %arrayidx1106, align 4
  %arrayidx1200 = getelementptr inbounds [64 x i32], ptr %ctx, i64 0, i64 39
  %52 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %51, <2 x i32> %51, <2 x i32> <i32 15, i32 15>)
  %53 = tail call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %51, <2 x i32> %51, <2 x i32> <i32 13, i32 13>)
  %54 = xor <2 x i32> %52, %53
  %55 = lshr <2 x i32> %51, <i32 10, i32 10>
  %56 = xor <2 x i32> %54, %55
  %57 = shufflevector <2 x i32> %38, <2 x i32> zeroinitializer, <2 x i32> <i32 poison, i32 0>
  %58 = insertelement <2 x i32> %57, i32 0, i64 0
  %59 = add <2 x i32> %58, %56
  store <2 x i32> %59, ptr %arrayidx1200, align 4
  ret void

; uselistorder directives
  uselistorder <2 x i32> %4, { 7, 0, 1, 6, 5, 4, 3, 2 }
  uselistorder <2 x i32> %38, { 6, 5, 4, 3, 2, 1, 0 }
}

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i64 @llvm.bswap.i64(i64) #2

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i32 @llvm.fshl.i32(i32, i32, i32) #2

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i32 @llvm.umin.i32(i32, i32) #2

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i64 @llvm.fshl.i64(i64, i64, i64) #2

; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg) #3

; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write)
declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg) #4

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare <2 x i32> @llvm.fshl.v2i32(<2 x i32>, <2 x i32>, <2 x i32>) #2

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>) #2

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare <2 x i64> @llvm.fshl.v2i64(<2 x i64>, <2 x i64>, <2 x i64>) #2

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>) #2

; uselistorder directives
uselistorder ptr @llvm.fshl.v2i32, { 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 }

attributes #0 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
attributes #1 = { nounwind sspstrong memory(argmem: readwrite) uwtable "min-legal-vector-width"="0" "no-trapping-math"="true" "probe-stack"="inline-asm" "stack-protector-buffer-size"="8" "target-cpu"="skylake" "target-features"="+adx,+aes,+avx,+avx2,+bmi,+bmi2,+clflushopt,+cmov,+crc32,+cx16,+cx8,+f16c,+fma,+fsgsbase,+fxsr,+invpcid,+lzcnt,+mmx,+movbe,+pclmul,+popcnt,+prfchw,+rdrnd,+rdseed,+sahf,+sgx,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsavec,+xsaveopt,+xsaves" }
attributes #2 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
attributes #3 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
attributes #4 = { nocallback nofree nounwind willreturn memory(argmem: write) }
attributes #5 = { nounwind memory(none) }

[thesamesam]

cc @RKSimon

Bisect says af32e51:

af32e51a43fb4343f4c407bf1ee051ff78a57494 is the first bad commit
commit af32e51a43fb4343f4c407bf1ee051ff78a57494
Author: Simon Pilgrim <llvm-dev@redking.me.uk>
Date:   Sat Jul 22 17:54:48 2023 +0100

    [X86] LowerRotate - manually expand rotate by splat constant patterns.

    Fixes issue identified on #63980 where the undef rotate amounts (during widening from v2i32 -> v4i32) were being constant folded to 0 when the shift amounts are created during expansion, losing the splat'd shift amounts.

 llvm/lib/Target/X86/X86ISelLowering.cpp         | 14 +++++++++++--
 llvm/test/CodeGen/X86/vector-fshl-rot-sub128.ll | 27 +++++++++----------------
 llvm/test/CodeGen/X86/vector-fshr-rot-sub128.ll | 27 +++++++++----------------
 3 files changed, 32 insertions(+), 36 deletions(-)
bisect found first bad commit

If I revert it on release/17.x, I get decent performance (although seemingly consistently a bit slower by half a second or so than clang 16).

[llvmbot]

@llvm/issue-subscribers-backend-x86

When using clang 17.0.1 in Gentoo and compiling [dev-libs/nss](https://developer.mozilla.org/en-US/docs/Mozilla/Projects/NSS) an infinite loop is detected in sha512.c compilation unit (reported [downstream Gentoo bug report](https://bugs.gentoo.org/914657)) which is not reproducible using Clang 16.

An initial testing revealed that march=native (also tested march=skylake and march=alderlake triggering the problem) might be causing the trouble.

In order to ease the test process, I attach the preprocessed output for that compile unit which I expect can reproduce this issue easily (remove the .txt extension)

sha512-preprocessed.c.txt