Try to reduce cxx-exception-fixup overhead

I think this is the main source of our jit-time regression.

src/codegen/irgen/refcounts.cpp

@@ -373,20 +373,25 @@ void addCXXFixup(llvm::Instruction* inst, const llvm::SmallVector<llvm::Tracking
     call->replaceAllUsesWith(new_invoke);
     call->eraseFromParent();
 
-    llvm::Value* exc_type, *exc_value, *exc_traceback;
-    std::tie(exc_type, exc_value, exc_traceback) = createLandingpad(fixup_block);
-
     llvm::IRBuilder<true> builder(fixup_block);
 
-    llvm::SmallVector<llvm::Value*, 4> decref_args;
-    decref_args.push_back(getConstantInt(to_decref.size(), g.i32));
-    decref_args.append(to_decref.begin(), to_decref.end());
-    builder.CreateCall(g.funcs.xdecrefAll, decref_args);
-
-    auto rethrow = builder.CreateCall3(g.funcs.rawReraise, exc_type, exc_value, exc_traceback);
+    static llvm::Function* _personality_func = g.stdlib_module->getFunction("__gxx_personality_v0");
+    assert(_personality_func);
+    llvm::Value* personality_func
+        = g.cur_module->getOrInsertFunction(_personality_func->getName(), _personality_func->getFunctionType());
+    assert(personality_func);
+    static llvm::Type* lp_type = llvm::StructType::create(std::vector<llvm::Type*>{ g.i8_ptr, g.i64 });
+    assert(lp_type);
+    llvm::LandingPadInst* landing_pad = builder.CreateLandingPad(lp_type, personality_func, 1);
+    landing_pad->addClause(getNullPtr(g.i8_ptr));
+
+    llvm::SmallVector<llvm::Value*, 4> call_args;
+    llvm::Value* cxaexc_pointer = builder.CreateExtractValue(landing_pad, { 0 });
+    call_args.push_back(cxaexc_pointer);
+    call_args.push_back(getConstantInt(to_decref.size(), g.i32));
+    call_args.append(to_decref.begin(), to_decref.end());
+    builder.CreateCall(g.funcs.xdecrefAndRethrow, call_args);
     builder.CreateUnreachable();
-
-    // new_invoke->getParent()->getParent()->dump();
 }
 
 // TODO: this should be cleaned up and moved to src/core/

src/codegen/runtime_hooks.cpp

@@ -324,7 +324,7 @@ void initGlobalFuncs(GlobalState& g) {
     GET(reraiseCapiExcAsCxx);
     GET(deopt);
     GET(checkRefs);
-    GET(xdecrefAll);
+    GET(xdecrefAndRethrow);
 
     GET(div_float_float);
     GET(floordiv_float_float);

src/codegen/runtime_hooks.h

@@ -55,7 +55,7 @@ struct GlobalFuncs {
     llvm::Value* raise0, *raise0_capi, *raise3, *raise3_capi, *rawReraise;
     llvm::Value* PyErr_Fetch, *PyErr_NormalizeException, *PyErr_Restore, *caughtCapiException, *reraiseCapiExcAsCxx;
     llvm::Value* deopt;
-    llvm::Value* checkRefs, *xdecrefAll;
+    llvm::Value* checkRefs, *xdecrefAndRethrow;
 
     llvm::Value* div_float_float, *floordiv_float_float, *mod_float_float, *pow_float_float;
 

src/runtime/inline/link_forcer.cpp

@@ -140,7 +140,7 @@ void force() {
     FORCE(reraiseCapiExcAsCxx);
     FORCE(deopt);
     FORCE(checkRefs);
-    FORCE(xdecrefAll);
+    FORCE(xdecrefAndRethrow);
 
     FORCE(div_i64_i64);
     FORCE(mod_i64_i64);

src/runtime/objmodel.cpp

@@ -118,7 +118,10 @@ static inline Box* callattrInternal3(Box* obj, BoxedString* attr, LookupScope sc
     return callattrInternal<S, rewritable>(obj, attr, scope, rewrite_args, argspec, arg1, arg2, arg3, NULL, NULL);
 }
 
-extern "C" void xdecrefAll(int num, ...) {
+extern "C" void* __cxa_begin_catch(void*);
+extern "C" void xdecrefAndRethrow(void* cxa_ptr, int num, ...) {
+    ExcInfo e = *(ExcInfo*)__cxa_begin_catch(cxa_ptr);
+
     va_list va;
     va_start(va, num);
 
@@ -128,6 +131,8 @@ extern "C" void xdecrefAll(int num, ...) {
     }
 
     va_end(va);
+
+    rawReraise(e.type, e.value, e.traceback);
 }
 
 extern "C" Box* deopt(AST_expr* expr, Box* value) {

src/runtime/objmodel.h

@@ -231,6 +231,6 @@ extern "C" void boxedLocalsDel(Box* boxedLocals, BoxedString* attr);
 
 extern "C" void checkRefs(Box* b);   // asserts that b has >= 0 refs
 extern "C" Box* assertAlive(Box* b); // asserts that b has > 0 refs, and returns b
-extern "C" void xdecrefAll(int num, ...);
+extern "C" void xdecrefAndRethrow(void* cxa_ptr, int num_decref, ...);
 }
 #endif