Another attempt

JuliaLang · Mar 14, 2023 · 9323cf2 · 9323cf2
1 parent fccd950
commit 9323cf2
Show file tree

Hide file tree

Showing 4 changed files with 21 additions and 19 deletions.
diff --git a/Make.inc b/Make.inc
@@ -467,7 +467,7 @@ FC := $(CROSS_COMPILE)gfortran
 ifeq ($(OS), Darwin)
 APPLE_ARCH := $(shell uname -m)
 ifneq ($(APPLE_ARCH),arm64)
-MACOSX_VERSION_MIN := 10.10
+MACOSX_VERSION_MIN := 10.14
 else
 MACOSX_VERSION_MIN := 11.0
 endif

diff --git a/src/aotcompile.cpp b/src/aotcompile.cpp
@@ -936,6 +936,8 @@ struct ShardTimers {
     }
 };
 
+void emitFloat16Wrappers(Module &M, bool external);
+
 // Perform the actual optimization and emission of the output files
 static void add_output_impl(Module &M, TargetMachine &SourceTM, std::string *outputs, ArrayRef<StringRef> names,
                     NewArchiveMember *unopt, NewArchiveMember *opt, NewArchiveMember *obj, NewArchiveMember *asm_,
@@ -1014,6 +1016,8 @@ static void add_output_impl(Module &M, TargetMachine &SourceTM, std::string *out
                 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));
+#else
+        emitFloat16Wrappers(M, false);
 #endif
     }
     timers.optimize.stopTimer();

diff --git a/src/codegen.cpp b/src/codegen.cpp
@@ -8684,14 +8684,15 @@ static JuliaVariable *julia_const_gv(jl_value_t *val)
     return nullptr;
 }
 //Float16 fun
-static void makeCastCall(Module &M, StringRef wrapperName, StringRef calledName, FunctionType *FTwrapper, FunctionType *FTcalled)
+static void makeCastCall(Module &M, StringRef wrapperName, StringRef calledName, FunctionType *FTwrapper, FunctionType *FTcalled, bool external)
 {
     Function *calledFun = M.getFunction(calledName);
     if (!calledFun) {
         calledFun = Function::Create(FTcalled, Function::ExternalLinkage, calledName, M);
     }
-    auto wrapperFun = Function::Create(FTwrapper, Function::ExternalLinkage, wrapperName, M);
-
+    auto linkage = external ? Function::ExternalLinkage : Function::InternalLinkage;
+    auto wrapperFun = Function::Create(FTwrapper, linkage, wrapperName, M);
+    wrapperFun->addFnAttr(Attribute::AlwaysInline);
     llvm::IRBuilder<> builder(BasicBlock::Create(M.getContext(), "top", wrapperFun));
     SmallVector<Value *, 4> CallArgs;
     if (wrapperFun->arg_size() != calledFun->arg_size()){
@@ -8708,28 +8709,27 @@ static void makeCastCall(Module &M, StringRef wrapperName, StringRef calledName,
     builder.CreateRet(retval);
 }
 
-static void emitFloat16Wrappers(Module &M)
+void emitFloat16Wrappers(Module &M, bool external)
 {
     auto &ctx = M.getContext();
     makeCastCall(M, "__gnu_h2f_ieee", "julia__gnu_h2f_ieee", FunctionType::get(Type::getFloatTy(ctx), { Type::getHalfTy(ctx) }, false),
-                FunctionType::get(Type::getFloatTy(ctx), { Type::getInt16Ty(ctx) }, false));
+                FunctionType::get(Type::getFloatTy(ctx), { Type::getInt16Ty(ctx) }, false), external);
     makeCastCall(M, "__extendhfsf2", "julia__gnu_h2f_ieee", FunctionType::get(Type::getFloatTy(ctx), { Type::getHalfTy(ctx) }, false),
-                FunctionType::get(Type::getFloatTy(ctx), { Type::getInt16Ty(ctx) }, false));
+                FunctionType::get(Type::getFloatTy(ctx), { Type::getInt16Ty(ctx) }, false), external);
     makeCastCall(M, "__gnu_f2h_ieee", "julia__gnu_f2h_ieee", FunctionType::get(Type::getHalfTy(ctx), { Type::getFloatTy(ctx) }, false),
-                FunctionType::get(Type::getInt16Ty(ctx), { Type::getFloatTy(ctx) }, false));
+                FunctionType::get(Type::getInt16Ty(ctx), { Type::getFloatTy(ctx) }, false), external);
     makeCastCall(M, "__truncsfhf2", "julia__gnu_f2h_ieee", FunctionType::get(Type::getHalfTy(ctx), { Type::getFloatTy(ctx) }, false),
-                FunctionType::get(Type::getInt16Ty(ctx), { Type::getFloatTy(ctx) }, false));
+                FunctionType::get(Type::getInt16Ty(ctx), { Type::getFloatTy(ctx) }, false), external);
     makeCastCall(M, "__truncdfhf2", "julia__truncdfhf2", FunctionType::get(Type::getHalfTy(ctx), { Type::getDoubleTy(ctx) }, false),
-                FunctionType::get(Type::getInt16Ty(ctx), { Type::getDoubleTy(ctx) }, false));
-
+                FunctionType::get(Type::getInt16Ty(ctx), { Type::getDoubleTy(ctx) }, false), external);
 }
 
 static void init_f16_funcs(void)
 {
     auto ctx = jl_ExecutionEngine->acquireContext();
     auto TSM =  jl_create_ts_module("F16Wrappers", ctx, imaging_default());
     auto aliasM = TSM.getModuleUnlocked();
-    emitFloat16Wrappers(*aliasM);
+    emitFloat16Wrappers(*aliasM, true);
     jl_ExecutionEngine->addModule(std::move(TSM));
 }
 

diff --git a/src/jitlayers.cpp b/src/jitlayers.cpp
@@ -1340,16 +1340,14 @@ JuliaOJIT::JuliaOJIT()
     }
 
     JD.addToLinkOrder(GlobalJD, orc::JITDylibLookupFlags::MatchExportedSymbolsOnly);
-#if JL_LLVM_VERSION <= 140000
     orc::SymbolAliasMap jl_crt = {
-        { mangle("__gnu_h2f_ieee"), { mangle("julia__gnu_h2f_ieee"), JITSymbolFlags::Exported } },
-        { mangle("__extendhfsf2"),  { mangle("julia__gnu_h2f_ieee"), JITSymbolFlags::Exported } },
-        { mangle("__gnu_f2h_ieee"), { mangle("julia__gnu_f2h_ieee"), JITSymbolFlags::Exported } },
-        { mangle("__truncsfhf2"),   { mangle("julia__gnu_f2h_ieee"), JITSymbolFlags::Exported } },
-        { mangle("__truncdfhf2"),   { mangle("julia__truncdfhf2"),   JITSymbolFlags::Exported } }
+        { mangle("__gnu_h2f_ieee"), { mangle("__jl_half2float"), JITSymbolFlags::Exported } },
+        { mangle("__extendhfsf2"),  { mangle("__jl_half2float"), JITSymbolFlags::Exported } },
+        { mangle("__gnu_f2h_ieee"), { mangle("__jl_float2half"), JITSymbolFlags::Exported } },
+        { mangle("__truncsfhf2"),   { mangle("__jl_float2half"), JITSymbolFlags::Exported } },
+        { mangle("__truncdfhf2"),   { mangle("__jl_double2half"),   JITSymbolFlags::Exported } }
     };
     cantFail(GlobalJD.define(orc::symbolAliases(jl_crt)));
-#endif
 
 #ifdef MSAN_EMUTLS_WORKAROUND
     orc::SymbolMap msan_crt;