Bump Julia to LLVM 16

Make.inc

@@ -501,7 +501,7 @@ JCPPFLAGS_COMMON  := -fasynchronous-unwind-tables
 JCPPFLAGS_CLANG   := $(JCPPFLAGS_COMMON) -mllvm -enable-tail-merge=0
 JCPPFLAGS_GCC     := $(JCPPFLAGS_COMMON) -fno-tree-tail-merge
 
-JCXXFLAGS_COMMON  := -pipe $(fPIC) -fno-rtti -std=c++14
+JCXXFLAGS_COMMON  := -pipe $(fPIC) -fno-rtti -std=c++17
 JCXXFLAGS_CLANG   := $(JCXXFLAGS_COMMON) -pedantic
 JCXXFLAGS_GCC     := $(JCXXFLAGS_COMMON) -fno-gnu-unique
 

deps/clang.version

@@ -1,4 +1,4 @@
 ## jll artifact
 # Clang (paired with LLVM, only here as a JLL download)
 CLANG_JLL_NAME := Clang
-CLANG_JLL_VER  := 15.0.7+10
+CLANG_JLL_VER  := 16.0.6+3

deps/lld.version

@@ -1,3 +1,3 @@
 ## jll artifact
 LLD_JLL_NAME := LLD
-LLD_JLL_VER := 15.0.7+10
+LLD_JLL_VER := 16.0.6+3

deps/llvm-tools.version

@@ -1,5 +1,5 @@
 ## jll artifact
 # LLVM_tools (downloads LLVM_jll to get things like `lit` and `opt`)
 LLVM_TOOLS_JLL_NAME := LLVM
-LLVM_TOOLS_JLL_VER := 15.0.7+10
-LLVM_TOOLS_ASSERT_JLL_VER := 15.0.7+10
+LLVM_TOOLS_JLL_VER := 16.0.6+3
+LLVM_TOOLS_ASSERT_JLL_VER := 16.0.6+3

deps/llvm.mk

@@ -133,7 +133,7 @@ endif # USE_PERF_JITEVENTS
 
 ifeq ($(BUILD_LLDB),1)
 ifeq ($(USECLANG),0)
-LLVM_CXXFLAGS += -std=c++0x
+LLVM_CXXFLAGS += -std=c++17
 endif # USECLANG
 ifeq ($(LLDB_DISABLE_PYTHON),1)
 LLVM_CXXFLAGS += -DLLDB_DISABLE_PYTHON

deps/llvm.version

@@ -2,14 +2,14 @@
 
 ## jll artifact
 LLVM_JLL_NAME := libLLVM
-LLVM_ASSERT_JLL_VER := 15.0.7+10
+LLVM_ASSERT_JLL_VER := 16.0.6+3
 ## source build
 # Version number of LLVM
-LLVM_VER := 15.0.7
+LLVM_VER := 16.0.6
 # Git branch name in `LLVM_GIT_URL` repository
-LLVM_BRANCH=julia-15.0.7-10
+LLVM_BRANCH=julia-16.0.6-2
 # Git ref in `LLVM_GIT_URL` repository
-LLVM_SHA1=julia-15.0.7-10
+LLVM_SHA1=julia-16.0.6-2
 
 ## Following options are used to automatically fetch patchset from Julia's fork.  This is
 ## useful if you want to build an external LLVM while still applying Julia's patches.
@@ -18,6 +18,6 @@ LLVM_APPLY_JULIA_PATCHES := 0
 # GitHub repository to use for fetching the Julia patches to apply to LLVM source code.
 LLVM_JULIA_DIFF_GITHUB_REPO := https://github.com/llvm/llvm-project
 # Base GitHub ref for generating the diff.
-LLVM_BASE_REF := llvm:llvmorg-15.0.7
+LLVM_BASE_REF := llvm:llvmorg-16.0.6
 # Julia fork's GitHub ref for generating the diff.
-LLVM_JULIA_REF := JuliaLang:julia-15.0.7-9
+LLVM_JULIA_REF := JuliaLang:julia-16.0.6-0

src/Makefile

@@ -114,6 +114,10 @@ PUBLIC_HEADER_TARGETS := $(addprefix $(build_includedir)/julia/,$(notdir $(PUBLI
 LLVM_LDFLAGS := $(shell $(LLVM_CONFIG_HOST) --ldflags)
 LLVM_CXXFLAGS := $(shell $(LLVM_CONFIG_HOST) --cxxflags)
 
+ifeq ($(OS)_$(BINARY),WINNT_32)
+LLVM_CXXFLAGS += -I$(SRCDIR)/support/win32-clang-ABI-bug
+endif
+
 # llvm-config --cxxflags does not return -DNDEBUG
 ifeq ($(shell $(LLVM_CONFIG_HOST) --assertion-mode),OFF)
 LLVM_CXXFLAGS += -DNDEBUG

src/aotcompile.cpp

@@ -1049,8 +1049,21 @@ static AOTOutputs add_output_impl(Module &M, TargetMachine &SourceTM, ShardTimer
             // but LLVM doesn't let us emit a GlobalAlias to a declaration...
             // So for now we inject a definition of these functions that calls our runtime
             // functions. We do so after optimization to avoid cloning these functions.
-
             // Float16 conversion routines
+#if defined(_CPU_X86_64_) && defined(_OS_DARWIN_) && JL_LLVM_VERSION >= 160000
+            // LLVM 16 reverted to soft-float ABI for passing half on x86_64 Darwin
+            // https://github.com/llvm/llvm-project/commit/2bcf51c7f82ca7752d1bba390a2e0cb5fdd05ca9
+            injectCRTAlias(M, "__gnu_h2f_ieee", "julia_half_to_float",
+                    FunctionType::get(Type::getFloatTy(M.getContext()), { Type::getInt16Ty(M.getContext()) }, false));
+            injectCRTAlias(M, "__extendhfsf2", "julia_half_to_float",
+                    FunctionType::get(Type::getFloatTy(M.getContext()), { Type::getInt16Ty(M.getContext()) }, false));
+            injectCRTAlias(M, "__gnu_f2h_ieee", "julia_float_to_half",
+                    FunctionType::get(Type::getInt16Ty(M.getContext()), { Type::getFloatTy(M.getContext()) }, false));
+            injectCRTAlias(M, "__truncsfhf2", "julia_float_to_half",
+                    FunctionType::get(Type::getInt16Ty(M.getContext()), { Type::getFloatTy(M.getContext()) }, false));
+            injectCRTAlias(M, "__truncdfhf2", "julia_double_to_half",
+                    FunctionType::get(Type::getInt16Ty(M.getContext()), { Type::getDoubleTy(M.getContext()) }, false));
+#else
             injectCRTAlias(M, "__gnu_h2f_ieee", "julia__gnu_h2f_ieee",
                     FunctionType::get(Type::getFloatTy(M.getContext()), { Type::getHalfTy(M.getContext()) }, false));
             injectCRTAlias(M, "__extendhfsf2", "julia__gnu_h2f_ieee",
@@ -1061,6 +1074,7 @@ static AOTOutputs add_output_impl(Module &M, TargetMachine &SourceTM, ShardTimer
                     FunctionType::get(Type::getHalfTy(M.getContext()), { Type::getFloatTy(M.getContext()) }, false));
             injectCRTAlias(M, "__truncdfhf2", "julia__truncdfhf2",
                     FunctionType::get(Type::getHalfTy(M.getContext()), { Type::getDoubleTy(M.getContext()) }, false));
+#endif
 
             // BFloat16 conversion routines
             injectCRTAlias(M, "__truncsfbf2", "julia__truncsfbf2",
@@ -1361,6 +1375,7 @@ static SmallVector<AOTOutputs, 16> add_output(Module &M, TargetMachine &TM, Stri
         for (unsigned i = 0; i < threads; i++) {
             std::function<void()> func = [&, i]() {
                 LLVMContext ctx;
+                SetOpaquePointer(ctx);
                 // Lazily deserialize the entire module
                 timers[i].deserialize.startTimer();
                 auto EM = getLazyBitcodeModule(MemoryBufferRef(StringRef(serialized.data(), serialized.size()), "Optimized"), ctx);
@@ -1562,6 +1577,7 @@ void jl_dump_native_impl(void *native_code,
     if (z) {
         JL_TIMING(NATIVE_AOT, NATIVE_Sysimg);
         LLVMContext Context;
+        SetOpaquePointer(Context);
         Module sysimgM("sysimg", Context);
         sysimgM.setTargetTriple(TheTriple.str());
         sysimgM.setDataLayout(DL);
@@ -1698,6 +1714,7 @@ void jl_dump_native_impl(void *native_code,
     {
         JL_TIMING(NATIVE_AOT, NATIVE_Metadata);
         LLVMContext Context;
+        SetOpaquePointer(Context);
         Module metadataM("metadata", Context);
         metadataM.setTargetTriple(TheTriple.str());
         metadataM.setDataLayout(DL);

src/cgutils.cpp

@@ -911,7 +911,7 @@ static bool is_tupletype_homogeneous(jl_svec_t *t, bool allow_va = false)
 }
 
 static bool for_each_uniontype_small(
-        std::function<void(unsigned, jl_datatype_t*)> f,
+        llvm::function_ref<void(unsigned, jl_datatype_t*)> f,
         jl_value_t *ty,
         unsigned &counter)
 {
@@ -4046,12 +4046,12 @@ static jl_cgval_t emit_memoryref(jl_codectx_t &ctx, const jl_cgval_t &ref, jl_cg
             Value *mlen = emit_genericmemorylen(ctx, mem, ref.typ);
             Value *inbound = ctx.builder.CreateICmpULT(newdata, mlen);
             ctx.builder.CreateCondBr(inbound, endBB, failBB);
-            ctx.f->getBasicBlockList().push_back(failBB);
+            failBB->insertInto(ctx.f);
             ctx.builder.SetInsertPoint(failBB);
             ctx.builder.CreateCall(prepare_call(jlboundserror_func),
                 { mark_callee_rooted(ctx, boxed(ctx, ref)), i });
             ctx.builder.CreateUnreachable();
-            ctx.f->getBasicBlockList().push_back(endBB);
+            endBB->insertInto(ctx.f);
             ctx.builder.SetInsertPoint(endBB);
         }
     }
@@ -4119,12 +4119,12 @@ static jl_cgval_t emit_memoryref(jl_codectx_t &ctx, const jl_cgval_t &ref, jl_cg
             Value *inbound = ctx.builder.CreateICmpULT(idx0, mlen);
 #endif
             ctx.builder.CreateCondBr(inbound, endBB, failBB);
-            ctx.f->getBasicBlockList().push_back(failBB);
+            failBB->insertInto(ctx.f);
             ctx.builder.SetInsertPoint(failBB);
             ctx.builder.CreateCall(prepare_call(jlboundserror_func),
                 { mark_callee_rooted(ctx, boxed(ctx, ref)), i });
             ctx.builder.CreateUnreachable();
-            ctx.f->getBasicBlockList().push_back(endBB);
+            endBB->insertInto(ctx.f);
             ctx.builder.SetInsertPoint(endBB);
         }
     }

src/clangsa/GCChecker.cpp

@@ -1,5 +1,6 @@
 // This file is a part of Julia. License is MIT: https://julialang.org/license
 
+#include "clang/AST/Type.h"
 #include "clang/Frontend/FrontendActions.h"
 #include "clang/StaticAnalyzer/Checkers/SValExplainer.h"
 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
@@ -14,6 +15,7 @@
 #include "clang/Tooling/Tooling.h"
 #include "clang/StaticAnalyzer/Frontend/CheckerRegistry.h"
 
+#include "llvm/Support/Debug.h"
 #include <iostream>
 #include <memory>
 
@@ -894,9 +896,11 @@ bool GCChecker::isSafepoint(const CallEvent &Call, CheckerContext &C) const {
     if (!Decl || !FD) {
       if (Callee == nullptr) {
         isCalleeSafepoint = true;
-      } else if (const TypedefType *TDT = dyn_cast<TypedefType>(Callee->getType())) {
-        isCalleeSafepoint =
-            !declHasAnnotation(TDT->getDecl(), "julia_not_safepoint");
+      } else if (const ElaboratedType *ET = dyn_cast<ElaboratedType>(Callee->getType())){
+        if (const TypedefType *TDT = dyn_cast<TypedefType>(ET->getNamedType())) {
+          isCalleeSafepoint =
+              !declHasAnnotation(TDT->getDecl(), "julia_not_safepoint");
+        }
       } else if (const CXXPseudoDestructorExpr *PDE =
                      dyn_cast<CXXPseudoDestructorExpr>(Callee)) {
         // A pseudo-destructor is an expression that looks like a member
@@ -941,7 +945,7 @@ bool GCChecker::processPotentialSafepoint(const CallEvent &Call,
             isGCTrackedType(ParmType->getPointeeType())) {
           // This is probably an out parameter. Find the value it refers to now.
           SVal Loaded =
-              State->getSVal(Call.getArgSVal(i).getAs<Loc>().getValue());
+              State->getSVal(*(Call.getArgSVal(i).getAs<Loc>()));
           SpeciallyRootedSymbol = Loaded.getAsSymbol();
           continue;
         }
@@ -1520,7 +1524,7 @@ bool GCChecker::evalCall(const CallEvent &Call, CheckerContext &C) const {
       }
     }
     if (FD) {
-      Loc ItemsLoc = State->getLValue(FD, ArrayList).getAs<Loc>().getValue();
+      Loc ItemsLoc = *(State->getLValue(FD, ArrayList).getAs<Loc>());
       SVal Items = State->getSVal(ItemsLoc);
       if (Items.isUnknown()) {
         Items = C.getSValBuilder().conjureSymbolVal(
@@ -1688,7 +1692,7 @@ void GCChecker::checkLocation(SVal SLoc, bool IsLoad, const Stmt *S,
   // better than this.
   if (IsLoad && (RS = State->get<GCRootMap>(SLoc.getAsRegion()))) {
     SymbolRef LoadedSym =
-        State->getSVal(SLoc.getAs<Loc>().getValue()).getAsSymbol();
+        State->getSVal(*SLoc.getAs<Loc>()).getAsSymbol();
     if (LoadedSym) {
       const ValueState *ValS = State->get<GCValueMap>(LoadedSym);
       if (!ValS || !ValS->isRooted() || ValS->RootDepth > RS->RootedAtDepth) {

src/codegen.cpp

@@ -1055,7 +1055,7 @@ static const auto jl_alloc_obj_func = new JuliaFunction<TypeFnContextAndSizeT>{
         FnAttrs.addAllocKindAttr(AllocFnKind::Alloc);
 #endif
 #if JL_LLVM_VERSION >= 160000
-        FnAttrs.addMemoryAttr(MemoryEffects::argMemOnly(ModRefInfo::Ref) | inaccessibleMemOnly(ModRefInfo::ModRef));
+        FnAttrs.addMemoryAttr(MemoryEffects::argMemOnly(ModRefInfo::Ref) | MemoryEffects::inaccessibleMemOnly(ModRefInfo::ModRef));
 #endif
         FnAttrs.addAttribute(Attribute::WillReturn);
         FnAttrs.addAttribute(Attribute::NoUnwind);
@@ -1314,7 +1314,7 @@ static const auto jl_allocgenericmemory = new JuliaFunction<TypeFnContextAndSize
             AttrBuilder FnAttrs(C);
             AttrBuilder RetAttrs(C);
 #if JL_LLVM_VERSION >= 160000
-            FnAttrs.addMemoryAttr(MemoryEffects::inaccessibleMemOnly(ModRefInfo::ModRef) | argMemOnly(MemoryEffects::ModRefInfo::Ref));
+            FnAttrs.addMemoryAttr(MemoryEffects::inaccessibleMemOnly(ModRefInfo::ModRef) | MemoryEffects::argMemOnly(ModRefInfo::Ref));
 #endif
             FnAttrs.addAttribute(Attribute::WillReturn);
             RetAttrs.addAlignmentAttr(Align(16));
@@ -1423,11 +1423,24 @@ static const auto gc_loaded_func = new JuliaFunction<>{
     [](LLVMContext &C) { return FunctionType::get(PointerType::get(JuliaType::get_prjlvalue_ty(C), AddressSpace::Loaded),
             {JuliaType::get_prjlvalue_ty(C), PointerType::get(JuliaType::get_prjlvalue_ty(C), 0)}, false); },
     [](LLVMContext &C) {
-        AttributeSet FnAttrs = Attributes(C, {Attribute::ReadNone, Attribute::NoSync, Attribute::NoUnwind, Attribute::Speculatable, Attribute::WillReturn, Attribute::NoRecurse});
-        AttributeSet RetAttrs = Attributes(C, {Attribute::NonNull, Attribute::NoUndef});
-        return AttributeList::get(C, FnAttrs, RetAttrs,
+        AttrBuilder FnAttrs(C);
+        FnAttrs.addAttribute(Attribute::NoSync);
+        FnAttrs.addAttribute(Attribute::NoUnwind);
+        FnAttrs.addAttribute(Attribute::Speculatable);
+        FnAttrs.addAttribute(Attribute::WillReturn);
+        FnAttrs.addAttribute(Attribute::NoRecurse);
+#if JL_LLVM_VERSION >= 160000
+        FnAttrs.addMemoryAttr(MemoryEffects::none());
+#else
+        FnAttrs.addAttribute(Attribute::ReadNone);
+#endif
+        AttrBuilder RetAttrs(C);
+        RetAttrs.addAttribute(Attribute::NonNull);
+        RetAttrs.addAttribute(Attribute::NoUndef);
+        return AttributeList::get(C, AttributeSet::get(C,FnAttrs), AttributeSet::get(C,RetAttrs),
                 { Attributes(C, {Attribute::NonNull, Attribute::NoUndef, Attribute::ReadNone, Attribute::NoCapture}),
-                  Attributes(C, {Attribute::NonNull, Attribute::NoUndef, Attribute::ReadNone}) }); },
+                  Attributes(C, {Attribute::NonNull, Attribute::NoUndef, Attribute::ReadNone}) });
+                  },
 };
 
 // julia.call represents a call with julia calling convention, it is used as
@@ -3714,11 +3727,11 @@ static bool emit_builtin_call(jl_codectx_t &ctx, jl_cgval_t *ret, jl_value_t *f,
                 failBB = BasicBlock::Create(ctx.builder.getContext(), "oob");
                 endBB = BasicBlock::Create(ctx.builder.getContext(), "load");
                 ctx.builder.CreateCondBr(ctx.builder.CreateIsNull(mlen), failBB, endBB);
-                ctx.f->getBasicBlockList().push_back(failBB);
+                failBB->insertInto(ctx.f);
                 ctx.builder.SetInsertPoint(failBB);
                 ctx.builder.CreateCall(prepare_call(jlboundserror_func), { mark_callee_rooted(ctx, mem), ConstantInt::get(ctx.types().T_size, 1) });
                 ctx.builder.CreateUnreachable();
-                ctx.f->getBasicBlockList().push_back(endBB);
+                endBB->insertInto(ctx.f);
                 ctx.builder.SetInsertPoint(endBB);
             }
             bool isboxed = layout->flags.arrayelem_isboxed;
@@ -3787,11 +3800,11 @@ static bool emit_builtin_call(jl_codectx_t &ctx, jl_cgval_t *ret, jl_value_t *f,
                 failBB = BasicBlock::Create(ctx.builder.getContext(), "oob");
                 endBB = BasicBlock::Create(ctx.builder.getContext(), "load");
                 ctx.builder.CreateCondBr(ctx.builder.CreateIsNull(mlen), failBB, endBB);
-                ctx.f->getBasicBlockList().push_back(failBB);
+                failBB->insertInto(ctx.f);
                 ctx.builder.SetInsertPoint(failBB);
                 ctx.builder.CreateCall(prepare_call(jlboundserror_func), { mark_callee_rooted(ctx, mem), ConstantInt::get(ctx.types().T_size, 1) });
                 ctx.builder.CreateUnreachable();
-                ctx.f->getBasicBlockList().push_back(endBB);
+                endBB->insertInto(ctx.f);
                 ctx.builder.SetInsertPoint(endBB);
             }
             emit_typecheck(ctx, val, ety, "memoryset");
@@ -3891,8 +3904,9 @@ static bool emit_builtin_call(jl_codectx_t &ctx, jl_cgval_t *ret, jl_value_t *f,
                     BasicBlock *passBB, *endBB, *fromBB;
                     passBB = BasicBlock::Create(ctx.builder.getContext(), "load");
                     endBB = BasicBlock::Create(ctx.builder.getContext(), "oob");
-                    ctx.f->getBasicBlockList().push_back(passBB);
-                    ctx.f->getBasicBlockList().push_back(endBB);
+
+                    passBB->insertInto(ctx.f);
+                    endBB->insertInto(ctx.f);
                     fromBB = ctx.builder.CreateCondBr(oob, endBB, passBB)->getParent();
                     ctx.builder.SetInsertPoint(endBB);
                     result = ctx.builder.CreatePHI(getInt1Ty(ctx.builder.getContext()), 2);
@@ -5286,9 +5300,9 @@ static void emit_phinode_assign(jl_codectx_t &ctx, ssize_t idx, jl_value_t *r)
             Instruction *phi = dest->clone();
             phi->insertAfter(dest);
             PHINode *Tindex_phi = PHINode::Create(getInt8Ty(ctx.builder.getContext()), jl_array_nrows(edges), "tindex_phi");
-            BB->getInstList().insert(InsertPt, Tindex_phi);
+            Tindex_phi->insertInto(BB, InsertPt);
             PHINode *ptr_phi = PHINode::Create(ctx.types().T_prjlvalue, jl_array_nrows(edges), "ptr_phi");
-            BB->getInstList().insert(InsertPt, ptr_phi);
+            ptr_phi->insertInto(BB, InsertPt);
             Value *isboxed = ctx.builder.CreateICmpNE(
                     ctx.builder.CreateAnd(Tindex_phi, ConstantInt::get(getInt8Ty(ctx.builder.getContext()), UNION_BOX_MARKER)),
                     ConstantInt::get(getInt8Ty(ctx.builder.getContext()), 0));
@@ -5306,7 +5320,7 @@ static void emit_phinode_assign(jl_codectx_t &ctx, ssize_t idx, jl_value_t *r)
         }
         else if (allunbox) {
             PHINode *Tindex_phi = PHINode::Create(getInt8Ty(ctx.builder.getContext()), jl_array_nrows(edges), "tindex_phi");
-            BB->getInstList().insert(InsertPt, Tindex_phi);
+            Tindex_phi->insertInto(BB, InsertPt);
             jl_cgval_t val = mark_julia_slot(NULL, phiType, Tindex_phi, ctx.tbaa().tbaa_stack);
             ctx.PhiNodes.push_back(std::make_tuple(val, BB, dest, (PHINode*)NULL, r));
             ctx.SAvalues[idx] = val;
@@ -5340,7 +5354,7 @@ static void emit_phinode_assign(jl_codectx_t &ctx, ssize_t idx, jl_value_t *r)
     }
     else {
         value_phi = PHINode::Create(vtype, jl_array_nrows(edges), "value_phi");
-        BB->getInstList().insert(InsertPt, value_phi);
+        value_phi->insertInto(BB, InsertPt);
         slot = mark_julia_type(ctx, value_phi, isboxed, phiType);
     }
     ctx.PhiNodes.push_back(std::make_tuple(slot, BB, dest, value_phi, r));
@@ -9539,6 +9553,8 @@ char jl_using_perf_jitevents = 0;
 
 int jl_is_timing_passes = 0;
 
+int jl_opaque_ptrs_set = 0;
+
 extern "C" void jl_init_llvm(void)
 {
     jl_page_size = jl_getpagesize();
@@ -9596,6 +9612,8 @@ extern "C" void jl_init_llvm(void)
     clopt = llvmopts.lookup("opaque-pointers");
     if (clopt && clopt->getNumOccurrences() == 0) {
         clopt->addOccurrence(1, clopt->ArgStr, "false", true);
+    } else {
+        jl_opaque_ptrs_set = 1;
     }
 
     clopt = llvmopts.lookup("time-passes");