Name LLVM function arguments (JuliaLang#50500)

This makes it easier to correlate LLVM IR with the originating source
code by including both argument name and argument type in the LLVM
argument variable.

<details>
<summary>Example 1</summary>

```julia
julia> function f(a, b, c, d, g...)
           e = a + b + c + d
           f = does_not_exist(e) + e
           f
       end
f (generic function with 1 method)

julia> @code_llvm f(0,0,0,0,0)
```
```llvm
;  @ REPL[1]:1 within `f`
define nonnull {}* @julia_f_141(i64 signext %"a::Int64", i64 signext %"b::Int64", i64 signext %"c::Int64", i64 signext %"d::Int64", i64 signext %"g[0]::Int64") #0 {
top:
  %0 = alloca [2 x {}*], align 8
  %gcframe3 = alloca [4 x {}*], align 16
  %gcframe3.sub = getelementptr inbounds [4 x {}*], [4 x {}*]* %gcframe3, i64 0, i64 0
  %1 = bitcast [4 x {}*]* %gcframe3 to i8*
  call void @llvm.memset.p0i8.i64(i8* align 16 %1, i8 0, i64 32, i1 true)
  %thread_ptr = call i8* asm "movq %fs:0, $0", "=r"() JuliaLang#7
  %tls_ppgcstack = getelementptr i8, i8* %thread_ptr, i64 -8
  %2 = bitcast i8* %tls_ppgcstack to {}****
  %tls_pgcstack = load {}***, {}**** %2, align 8
;  @ REPL[1]:3 within `f`
  %3 = bitcast [4 x {}*]* %gcframe3 to i64*
  store i64 8, i64* %3, align 16
  %4 = getelementptr inbounds [4 x {}*], [4 x {}*]* %gcframe3, i64 0, i64 1
  %5 = bitcast {}** %4 to {}***
  %6 = load {}**, {}*** %tls_pgcstack, align 8
  store {}** %6, {}*** %5, align 8
  %7 = bitcast {}*** %tls_pgcstack to {}***
  store {}** %gcframe3.sub, {}*** %7, align 8
  %Main.does_not_exist.cached = load atomic {}*, {}** @0 unordered, align 8
  %iscached.not = icmp eq {}* %Main.does_not_exist.cached, null
  br i1 %iscached.not, label %notfound, label %found

notfound:                                         ; preds = %top
  %Main.does_not_exist.found = call {}* @ijl_get_binding_or_error({}* nonnull inttoptr (i64 139831437630272 to {}*), {}* nonnull inttoptr (i64 139831600565400 to {}*))
  store atomic {}* %Main.does_not_exist.found, {}** @0 release, align 8
  br label %found

found:                                            ; preds = %notfound, %top
  %Main.does_not_exist = phi {}* [ %Main.does_not_exist.cached, %top ], [ %Main.does_not_exist.found, %notfound ]
  %8 = bitcast {}* %Main.does_not_exist to {}**
  %does_not_exist.checked = load atomic {}*, {}** %8 unordered, align 8
  %.not = icmp eq {}* %does_not_exist.checked, null
  br i1 %.not, label %err, label %ok

err:                                              ; preds = %found
  call void @ijl_undefined_var_error({}* inttoptr (i64 139831600565400 to {}*))
  unreachable

ok:                                               ; preds = %found
  %.sub = getelementptr inbounds [2 x {}*], [2 x {}*]* %0, i64 0, i64 0
;  @ REPL[1]:2 within `f`
; ┌ @ operators.jl:587 within `+` @ int.jl:87
   %9 = add i64 %"b::Int64", %"a::Int64"
   %10 = add i64 %9, %"c::Int64"
; │ @ operators.jl:587 within `+`
; │┌ @ operators.jl:544 within `afoldl`
; ││┌ @ int.jl:87 within `+`
     %11 = add i64 %10, %"d::Int64"
     %12 = getelementptr inbounds [4 x {}*], [4 x {}*]* %gcframe3, i64 0, i64 3
     store {}* %does_not_exist.checked, {}** %12, align 8
; └└└
;  @ REPL[1]:3 within `f`
  %13 = call nonnull {}* @ijl_box_int64(i64 signext %11)
  %14 = getelementptr inbounds [4 x {}*], [4 x {}*]* %gcframe3, i64 0, i64 2
  store {}* %13, {}** %14, align 16
  store {}* %13, {}** %.sub, align 8
  %15 = call nonnull {}* @ijl_apply_generic({}* nonnull %does_not_exist.checked, {}** nonnull %.sub, i32 1)
  store {}* %15, {}** %12, align 8
  %16 = call nonnull {}* @ijl_box_int64(i64 signext %11)
  store {}* %16, {}** %14, align 16
  store {}* %15, {}** %.sub, align 8
  %17 = getelementptr inbounds [2 x {}*], [2 x {}*]* %0, i64 0, i64 1
  store {}* %16, {}** %17, align 8
  %18 = call nonnull {}* @ijl_apply_generic({}* inttoptr (i64 139831370516384 to {}*), {}** nonnull %.sub, i32 2)
  %19 = load {}*, {}** %4, align 8
  %20 = bitcast {}*** %tls_pgcstack to {}**
  store {}* %19, {}** %20, align 8
;  @ REPL[1]:4 within `f`
  ret {}* %18
}
```
</details>

<details>
<summary>Example 2</summary>

```julia
julia> function g(a, b, c, d; kwarg=0)
           a + b + c + d + kwarg
       end
g (generic function with 1 method)

julia> @code_llvm g(0,0,0,0,kwarg=0)
```
```llvm
;  @ REPL[3]:1 within `g`
define i64 @julia_g_160([1 x i64]* nocapture noundef nonnull readonly align 8 dereferenceable(8) %"#1::NamedTuple", i64 signext %"a::Int64", i64 signext %"b::Int64", i64 signext %"c::Int64", i64 signext %"d::Int64") #0 {
top:
  %0 = getelementptr inbounds [1 x i64], [1 x i64]* %"#1::NamedTuple", i64 0, i64 0
; ┌ @ REPL[3]:2 within `#g#1`
; │┌ @ operators.jl:587 within `+` @ int.jl:87
    %1 = add i64 %"b::Int64", %"a::Int64"
    %2 = add i64 %1, %"c::Int64"
; ││ @ operators.jl:587 within `+`
; ││┌ @ operators.jl:544 within `afoldl`
; │││┌ @ int.jl:87 within `+`
      %3 = add i64 %2, %"d::Int64"
; │││└
; │││ @ operators.jl:545 within `afoldl`
; │││┌ @ int.jl:87 within `+`
      %unbox = load i64, i64* %0, align 8
      %4 = add i64 %3, %unbox
; └└└└
  ret i64 %4
}
```
</details>

src/aotcompile.cpp

@@ -2093,10 +2093,10 @@ void jl_get_llvmf_defn_impl(jl_llvmf_dump_t* dump, jl_method_instance_t *mi, siz
         // // Force imaging mode for names of pointers
         // output.imaging = true;
         // This would also be nice, but it seems to cause OOMs on the windows32 builder
-        // // Force at least medium debug info for introspection
+        // Force at least medium debug info for introspection
         // No debug info = no variable names,
         // max debug info = llvm.dbg.declare/value intrinsics which clutter IR output
-        output.debug_level = jl_options.debug_level;
+        output.debug_level = std::max(1, static_cast<int>(jl_options.debug_level));
         auto decls = jl_emit_code(m, mi, src, jlrettype, output);
         JL_UNLOCK(&jl_codegen_lock); // Might GC
 

src/codegen.cpp

@@ -1736,7 +1736,7 @@ jl_aliasinfo_t jl_aliasinfo_t::fromTBAA(jl_codectx_t &ctx, MDNode *tbaa) {
 }
 
 static Type *julia_type_to_llvm(jl_codectx_t &ctx, jl_value_t *jt, bool *isboxed = NULL);
-static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value *fval, StringRef name, jl_value_t *sig, jl_value_t *jlrettype, bool is_opaque_closure, bool gcstack_arg);
+static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value *fval, StringRef name, jl_value_t *sig, jl_value_t *jlrettype, bool is_opaque_closure, bool gcstack_arg, BitVector *used_arguments=nullptr, size_t *args_begin=nullptr);
 static jl_cgval_t emit_expr(jl_codectx_t &ctx, jl_value_t *expr, ssize_t ssaval = -1);
 static Value *global_binding_pointer(jl_codectx_t &ctx, jl_module_t *m, jl_sym_t *s,
                                      jl_binding_t **pbnd, bool assign);
@@ -2363,17 +2363,17 @@ std::unique_ptr<Module> jl_create_llvm_module(StringRef name, LLVMContext &conte
 
 static void jl_name_jlfunc_args(jl_codegen_params_t &params, Function *F) {
     assert(F->arg_size() == 3);
-    F->getArg(0)->setName("function");
-    F->getArg(1)->setName("args");
-    F->getArg(2)->setName("nargs");
+    F->getArg(0)->setName("function::Core.Function");
+    F->getArg(1)->setName("args::Any[]");
+    F->getArg(2)->setName("nargs::UInt32");
 }
 
 static void jl_name_jlfuncparams_args(jl_codegen_params_t &params, Function *F) {
     assert(F->arg_size() == 4);
-    F->getArg(0)->setName("function");
-    F->getArg(1)->setName("args");
-    F->getArg(2)->setName("nargs");
-    F->getArg(3)->setName("sparams");
+    F->getArg(0)->setName("function::Core.Function");
+    F->getArg(1)->setName("args::Any[]");
+    F->getArg(2)->setName("nargs::UInt32");
+    F->getArg(3)->setName("sparams::Any");
 }
 
 static void jl_init_function(Function *F, const Triple &TT)
@@ -4359,7 +4359,7 @@ static jl_cgval_t emit_call_specfun_boxed(jl_codectx_t &ctx, jl_value_t *jlretty
         }
         jl_aliasinfo_t ai = jl_aliasinfo_t::fromTBAA(ctx, ctx.tbaa().tbaa_const);
         theFptr = ai.decorateInst(ctx.builder.CreateAlignedLoad(pfunc, GV, Align(sizeof(void*))));
-        setName(ctx.emission_context, theFptr, namep);
+        setName(ctx.emission_context, theFptr, specFunctionObject);
     }
     else {
         theFptr = jl_Module->getOrInsertFunction(specFunctionObject, ctx.types().T_jlfunc).getCallee();
@@ -6935,10 +6935,11 @@ static Function *gen_invoke_wrapper(jl_method_instance_t *lam, jl_value_t *jlret
     return w;
 }
 
-static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value *fval, StringRef name, jl_value_t *sig, jl_value_t *jlrettype, bool is_opaque_closure, bool gcstack_arg)
+static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value *fval, StringRef name, jl_value_t *sig, jl_value_t *jlrettype, bool is_opaque_closure, bool gcstack_arg, BitVector *used_arguments, size_t *arg_offset)
 {
     jl_returninfo_t props = {};
     SmallVector<Type*, 8> fsig;
+    SmallVector<std::string, 4> argnames;
     Type *rt = NULL;
     Type *srt = NULL;
     if (jlrettype == (jl_value_t*)jl_bottom_type) {
@@ -6956,6 +6957,7 @@ static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value
             props.cc = jl_returninfo_t::Union;
             Type *AT = ArrayType::get(getInt8Ty(ctx.builder.getContext()), props.union_bytes);
             fsig.push_back(AT->getPointerTo());
+            argnames.push_back("union_bytes_return");
             Type *pair[] = { ctx.types().T_prjlvalue, getInt8Ty(ctx.builder.getContext()) };
             rt = StructType::get(ctx.builder.getContext(), makeArrayRef(pair));
         }
@@ -6980,6 +6982,7 @@ static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value
             // sret is always passed from alloca
             assert(M);
             fsig.push_back(rt->getPointerTo(M->getDataLayout().getAllocaAddrSpace()));
+            argnames.push_back("sret_return");
             srt = rt;
             rt = getVoidTy(ctx.builder.getContext());
         }
@@ -7018,6 +7021,7 @@ static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value
         param.addAttribute(Attribute::NoUndef);
         attrs.push_back(AttributeSet::get(ctx.builder.getContext(), param));
         fsig.push_back(get_returnroots_type(ctx, props.return_roots)->getPointerTo(0));
+        argnames.push_back("return_roots");
     }
 
     if (gcstack_arg){
@@ -7026,9 +7030,16 @@ static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value
         param.addAttribute(Attribute::NonNull);
         attrs.push_back(AttributeSet::get(ctx.builder.getContext(), param));
         fsig.push_back(PointerType::get(JuliaType::get_ppjlvalue_ty(ctx.builder.getContext()), 0));
+        argnames.push_back("pgcstack_arg");
     }
 
-    for (size_t i = 0; i < jl_nparams(sig); i++) {
+    if (arg_offset)
+        *arg_offset = fsig.size();
+    size_t nparams = jl_nparams(sig);
+    if (used_arguments)
+        used_arguments->resize(nparams);
+
+    for (size_t i = 0; i < nparams; i++) {
         jl_value_t *jt = jl_tparam(sig, i);
         bool isboxed = false;
         Type *ty = NULL;
@@ -7060,6 +7071,8 @@ static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value
         }
         attrs.push_back(AttributeSet::get(ctx.builder.getContext(), param));
         fsig.push_back(ty);
+        if (used_arguments)
+            used_arguments->set(i);
     }
 
     AttributeSet FnAttrs;
@@ -7089,8 +7102,14 @@ static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, Value
         else
             fval = emit_bitcast(ctx, fval, ftype->getPointerTo());
     }
-    if (gcstack_arg && isa<Function>(fval))
-        cast<Function>(fval)->setCallingConv(CallingConv::Swift);
+    if (auto F = dyn_cast<Function>(fval)) {
+        if (gcstack_arg)
+            F->setCallingConv(CallingConv::Swift);
+        assert(F->arg_size() >= argnames.size());
+        for (size_t i = 0; i < argnames.size(); i++) {
+            F->getArg(i)->setName(argnames[i]);
+        }
+    }
     props.decl = FunctionCallee(ftype, fval);
     props.attrs = attributes;
     return props;
@@ -7316,11 +7335,49 @@ static jl_llvm_functions_t
     Function *f = NULL;
     bool has_sret = false;
     if (specsig) { // assumes !va and !needsparams
+        BitVector used_args;
+        size_t args_begin;
         returninfo = get_specsig_function(ctx, M, NULL, declarations.specFunctionObject, lam->specTypes,
-                                          jlrettype, ctx.is_opaque_closure, JL_FEAT_TEST(ctx,gcstack_arg));
+                                          jlrettype, ctx.is_opaque_closure, JL_FEAT_TEST(ctx,gcstack_arg), &used_args, &args_begin);
         f = cast<Function>(returninfo.decl.getCallee());
         has_sret = (returninfo.cc == jl_returninfo_t::SRet || returninfo.cc == jl_returninfo_t::Union);
         jl_init_function(f, ctx.emission_context.TargetTriple);
+        if (ctx.emission_context.debug_level > 0) {
+            auto arg_typename = [&](size_t i) JL_NOTSAFEPOINT {
+                auto tp = jl_tparam(lam->specTypes, i);
+                return jl_is_datatype(tp) ? jl_symbol_name(((jl_datatype_t*)tp)->name->name) : "<unknown type>";
+            };
+            size_t nreal = 0;
+            for (size_t i = 0; i < std::min(nreq, static_cast<size_t>(used_args.size())); i++) {
+                jl_sym_t *argname = slot_symbol(ctx, i);
+                if (argname == jl_unused_sym)
+                    continue;
+                if (used_args.test(i)) {
+                    auto &arg = *f->getArg(args_begin++);
+                    nreal++;
+                    auto name = jl_symbol_name(argname);
+                    if (!name[0]) {
+                        arg.setName(StringRef("#") + Twine(nreal) + StringRef("::") + arg_typename(i));
+                    } else {
+                        arg.setName(name + StringRef("::") + arg_typename(i));
+                    }
+                }
+            }
+            if (va && ctx.vaSlot != -1) {
+                size_t vidx = 0;
+                for (size_t i = nreq; i < used_args.size(); i++) {
+                    if (used_args.test(i)) {
+                        auto &arg = *f->getArg(args_begin++);
+                        auto type = arg_typename(i);
+                        const char *name = jl_symbol_name(slot_symbol(ctx, ctx.vaSlot));
+                        if (!name[0])
+                            name = "...";
+                        vidx++;
+                        arg.setName(name + StringRef("[") + Twine(vidx) + StringRef("]::") + type);
+                    }
+                }
+            }
+        }
 
         // common pattern: see if all return statements are an argument in that
         // case the apply-generic call can re-use the original box for the return

test/compiler/codegen.jl

@@ -253,7 +253,7 @@ if opt_level > 0
     load_dummy_ref_ir = get_llvm(load_dummy_ref, Tuple{Int})
     @test !occursin("jl_gc_pool_alloc", load_dummy_ref_ir)
     # Hopefully this is reliable enough. LLVM should be able to optimize this to a direct return.
-    @test occursin("ret $Iptr %0", load_dummy_ref_ir)
+    @test occursin("ret $Iptr %\"x::$(Int)\"", load_dummy_ref_ir)
 end
 
 # Issue 22770

test/llvmpasses/fastmath.jl

@@ -1,6 +1,6 @@
 # This file is a part of Julia. License is MIT: https://julialang.org/license
 
-# RUN: julia --startup-file=no %s %t && llvm-link -S %t/* -o %t/module.ll
+# RUN: julia --startup-file=no %s %t -O && llvm-link -S %t/* -o %t/module.ll
 # RUN: cat %t/module.ll | FileCheck %s
 
 ## Notes:
@@ -14,21 +14,31 @@ include(joinpath("..", "testhelpers", "llvmpasses.jl"))
 
 import Base.FastMath
 
-# CHECK: call fast float @llvm.sqrt.f32(float %{{[0-9]+}})
+# CHECK: call fast float @llvm.sqrt.f32(float %"x::Float32")
 emit(FastMath.sqrt_fast, Float32)
 
 
 # Float16 operations should be performed as Float32, unless @fastmath is specified
 # TODO: this is not true for platforms that natively support Float16
 
 foo(x::T,y::T) where T = x-y == zero(T)
-# LOWER: fsub half %0, %1
-# FINAL: %2 = fpext half %0 to float
-# FINAL: %3 = fpext half %1 to float
-# FINAL: fsub half %2, %3
+# CHECK: define {{(swiftcc )?}}i8 @julia_foo_{{[0-9]+}}({{.*}}half %[[X:"x::Float16"]], half %[[Y:"y::Float16"]]) {{.*}}{
+# CHECK-DAG: %[[XEXT:[0-9]+]] = fpext half %[[X]] to float
+# CHECK-DAG: %[[YEXT:[0-9]+]] = fpext half %[[Y]] to float
+# CHECK: %[[DIFF:[0-9]+]] = fsub float %[[XEXT]], %[[YEXT]]
+# CHECK: %[[TRUNC:[0-9]+]] = fptrunc float %[[DIFF]] to half
+# CHECK: %[[DIFFEXT:[0-9]+]] = fpext half %[[TRUNC]] to float
+# CHECK: %[[CMP:[0-9]+]] = fcmp oeq float %[[DIFFEXT]], 0.000000e+00
+# CHECK: %[[ZEXT:[0-9]+]] = zext i1 %[[CMP]] to i8
+# CHECK: ret i8 %[[ZEXT]]
+# CHECK: }
 emit(foo, Float16, Float16)
 
 @fastmath foo(x::T,y::T) where T = x-y == zero(T)
-# LOWER: fsub fast half %0, %1
-# FINAL: fsub fast half %0, %1
+# CHECK: define {{(swiftcc )?}}i8 @julia_foo_{{[0-9]+}}({{.*}}half %[[X:"x::Float16"]], half %[[Y:"y::Float16"]]) {{.*}}{
+# CHECK: %[[DIFF:[0-9]+]] = fsub fast half %[[X]], %[[Y]]
+# CHECK: %[[CMP:[0-9]+]] = fcmp fast oeq half %[[DIFF]], 0xH0000
+# CHECK: %[[ZEXT:[0-9]+]] = zext i1 %[[CMP]] to i8
+# CHECK: ret i8 %[[ZEXT]]
+# CHECK: }
 emit(foo, Float16, Float16)