[IR] Don't allow values of opaque type (#137625)

nikic · web-flow · commit 6feb4a8ef4d1 · 2025-04-30T15:01:00.000+02:00
Consider opaque types as non-first-class types, i.e. do not allow SSA
values to have opaque type.
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
@@ -4122,6 +4122,30 @@ except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
 | ``{i32, i32} (i32)``            | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values                                                                 |
 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 
+.. _t_opaque:
+
+Opaque Structure Types
+----------------------
+
+:Overview:
+
+Opaque structure types are used to represent structure types that
+do not have a body specified. This corresponds (for example) to the C
+notion of a forward declared structure. They can be named (``%X``) or
+unnamed (``%52``).
+
+It is not possible to create SSA values with an opaque structure type. In
+practice, this largely limits their use to the value type of external globals.
+
+:Syntax:
+
+::
+
+      %X = type opaque
+      %52 = type opaque
+
+      @g = external global %X
+
 .. _t_firstclass:
 
 First Class Types
@@ -4562,31 +4586,6 @@ opaqued and are never uniqued. Identified types must not be recursive.
 | ``<{ i8, i32 }>``            | A packed struct known to be 5 bytes in size.                                                                                                                                          |
 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 
-.. _t_opaque:
-
-Opaque Structure Types
-""""""""""""""""""""""
-
-:Overview:
-
-Opaque structure types are used to represent structure types that
-do not have a body specified. This corresponds (for example) to the C
-notion of a forward declared structure. They can be named (``%X``) or
-unnamed (``%52``).
-
-:Syntax:
-
-::
-
-      %X = type opaque
-      %52 = type opaque
-
-:Examples:
-
-+--------------+-------------------+
-| ``opaque``   | An opaque type.   |
-+--------------+-------------------+
-
 .. _constants:
 
 Constants
diff --git a/llvm/include/llvm/IR/Type.h b/llvm/include/llvm/IR/Type.h
@@ -286,9 +286,7 @@ class Type {
 
   /// Return true if the type is "first class", meaning it is a valid type for a
   /// Value.
-  bool isFirstClassType() const {
-    return getTypeID() != FunctionTyID && getTypeID() != VoidTyID;
-  }
+  bool isFirstClassType() const;
 
   /// Return true if the type is a valid type for a register in codegen. This
   /// includes all first-class types except struct and array types.
diff --git a/llvm/lib/IR/Type.cpp b/llvm/lib/IR/Type.cpp
@@ -247,6 +247,20 @@ int Type::getFPMantissaWidth() const {
   return -1;
 }
 
+bool Type::isFirstClassType() const {
+  switch (getTypeID()) {
+    default:
+      return true;
+    case FunctionTyID:
+    case VoidTyID:
+      return false;
+    case StructTyID: {
+      auto *ST = cast<StructType>(this);
+      return !ST->isOpaque();
+    }
+  }
+}
+
 bool Type::isSizedDerivedType(SmallPtrSetImpl<Type*> *Visited) const {
   if (auto *ATy = dyn_cast<ArrayType>(this))
     return ATy->getElementType()->isSized(Visited);
diff --git a/llvm/lib/Linker/IRMover.cpp b/llvm/lib/Linker/IRMover.cpp
@@ -1640,6 +1640,8 @@ Error IRLinker::run() {
     if (GV.hasAppendingLinkage())
       continue;
     Value *NewValue = Mapper.mapValue(GV);
+    if (FoundError)
+      return std::move(*FoundError);
     if (NewValue) {
       auto *NewGV = dyn_cast<GlobalVariable>(NewValue->stripPointerCasts());
       if (NewGV) {
diff --git a/llvm/test/Assembler/2004-11-28-InvalidTypeCrash.ll b/llvm/test/Assembler/2004-11-28-InvalidTypeCrash.ll
@@ -1,5 +1,5 @@
 ; Test for PR463.  This program is erroneous, but should not crash llvm-as.
 ; RUN: not llvm-as %s -o /dev/null 2>&1 | FileCheck %s
-; CHECK: use of undefined type named 'struct.none'
+; CHECK: invalid type for null constant
 
 @.FOO  = internal global %struct.none zeroinitializer
diff --git a/llvm/test/Assembler/2005-05-05-OpaqueUndefValues.ll b/llvm/test/Assembler/2005-05-05-OpaqueUndefValues.ll
@@ -1,5 +1,6 @@
-; RUN: llvm-as < %s | llvm-dis | llvm-as > /dev/null
-; RUN: verify-uselistorder %s
+; RUN: not llvm-as < %s 2>&1 | FileCheck %s
+
+; CHECK: error: invalid type for undef constant
 
 %t = type opaque
 @x = global %t undef
diff --git a/llvm/test/Assembler/2007-01-02-Undefined-Arg-Type.ll b/llvm/test/Assembler/2007-01-02-Undefined-Arg-Type.ll
@@ -1,7 +1,7 @@
 ; The assembler should catch an undefined argument type .
 ; RUN: not llvm-as %s -o /dev/null 2>&1 | FileCheck %s
 
-; CHECK: use of undefined type named 'typedef.bc_struct'
+; CHECK: invalid type for function argument
 
 ; %typedef.bc_struct = type opaque
 
diff --git a/llvm/test/CodeGen/WebAssembly/externref-unsized-load.ll b/llvm/test/CodeGen/WebAssembly/externref-unsized-load.ll
@@ -7,4 +7,4 @@ define void @load_extern(%externref %ref) {
   ret void
 }
 
-; CHECK-ERROR: error: loading unsized types is not allowed
+; CHECK-ERROR: error: load operand must be a pointer to a first class type
diff --git a/llvm/test/CodeGen/WebAssembly/externref-unsized-store.ll b/llvm/test/CodeGen/WebAssembly/externref-unsized-store.ll
@@ -7,4 +7,4 @@ define void @store_extern(%externref %ref) {
   ret void
 }
 
-; CHECK-ERROR: error: storing unsized types is not allowed
+; CHECK-ERROR: error: invalid type for undef constant
diff --git a/llvm/test/CodeGen/X86/cfguard-x86-64-vectorcall.ll b/llvm/test/CodeGen/X86/cfguard-x86-64-vectorcall.ll
@@ -3,6 +3,7 @@
 ; RUN: llc < %s -mtriple=x86_64-w64-windows-gnu | FileCheck %s -check-prefix=X64
 ; Control Flow Guard is currently only available on Windows
 
+%struct.HVA = type { double, double, double, double }
 
 ; Test that Control Flow Guard checks are correctly added for x86_64 vector calls.
 define void @func_cf_vector_x64(ptr %0, ptr %1) #0 {
@@ -37,8 +38,6 @@ entry:
 }
 attributes #0 = { "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" }
 
-%struct.HVA = type { double, double, double, double }
-
 declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1 immarg) #1
 attributes #1 = { argmemonly nounwind willreturn }
 
diff --git a/llvm/test/CodeGen/X86/cfguard-x86-vectorcall.ll b/llvm/test/CodeGen/X86/cfguard-x86-vectorcall.ll
@@ -2,6 +2,7 @@
 ; RUN: llc < %s -mtriple=i686-w64-windows-gnu | FileCheck %s -check-prefix=X86
 ; Control Flow Guard is currently only available on Windows
 
+%struct.HVA = type { double, double, double, double }
 
 ; Test that Control Flow Guard checks are correctly added for x86 vector calls.
 define void @func_cf_vector_x86(ptr %0, ptr %1) #0 {
@@ -32,8 +33,6 @@ entry:
 }
 attributes #0 = { "target-cpu"="pentium4" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" }
 
-%struct.HVA = type { double, double, double, double }
-
 declare void @llvm.memcpy.p0.p0.i32(ptr nocapture writeonly, ptr nocapture readonly, i32, i1 immarg) #1
 attributes #1 = { argmemonly nounwind willreturn }
 
diff --git a/llvm/test/Linker/Inputs/pr22807-1.ll b/llvm/test/Linker/Inputs/pr22807-1.ll
diff --git a/llvm/test/Linker/Inputs/pr22807.ll b/llvm/test/Linker/Inputs/pr22807.ll
@@ -1,6 +1,4 @@
 %struct.A = type { %struct.B }
 %struct.B = type opaque
 
-define i32 @bar(%struct.A %A) {
-   ret i32 0
-}
+@g = external global %struct.A
diff --git a/llvm/test/Linker/Inputs/type-unique-opaque.ll b/llvm/test/Linker/Inputs/type-unique-opaque.ll
@@ -1,6 +1,4 @@
 %t = type { i8 }
 %t2 = type { %t, i16 }
 
-define %t2 @f() {
-  ret %t2 { %t { i8 0 }, i16 0 }
-}
+@g = external global %t2
diff --git a/llvm/test/Linker/intrinsics-with-unnamed-types.ll b/llvm/test/Linker/intrinsics-with-unnamed-types.ll
@@ -7,8 +7,8 @@
 ; Make sure we can link files with clashing intrinsic names using unnamed types.
 
 ;--- f01.ll
-%1 = type opaque
-%0 = type opaque
+%1 = type { i32 }
+%0 = type { i64 }
 
 ; CHECK-LABEL: @test01(
 ; CHECK:       %c1 = call %0 @llvm.ssa.copy.s_s.0(%0 %arg)
@@ -38,8 +38,8 @@ bb:
 }
 
 ;--- f02.ll
-%1 = type opaque
-%2 = type opaque
+%1 = type { i8 }
+%2 = type { i16 }
 
 ; CHECK-LABEL: @test03(
 ; CHECK:      %c1 = call %3 @llvm.ssa.copy.s_s.2(%3 %arg)
diff --git a/llvm/test/Linker/pr22807.ll b/llvm/test/Linker/pr22807.ll
@@ -1,10 +1,12 @@
-; RUN: not llvm-link -S -o - %p/pr22807.ll %p/Inputs/pr22807-1.ll %p/Inputs/pr22807-2.ll 2>&1 | FileCheck %s
+; RUN: not llvm-link -S -o - %p/pr22807.ll %p/Inputs/pr22807.ll 2>&1 | FileCheck %s
 
 ; CHECK: error: identified structure type 'struct.A' is recursive
 
 %struct.B = type { %struct.A }
 %struct.A = type opaque
 
-define i32 @baz(%struct.B %BB) {
-  ret i32 0
+@g = external global %struct.B
+
+define ptr @test() {
+  ret ptr @g
 }
diff --git a/llvm/test/Linker/type-unique-opaque.ll b/llvm/test/Linker/type-unique-opaque.ll
@@ -4,13 +4,13 @@
 ; not cause %u and %t to get merged.
 
 ; CHECK: %u = type opaque
-; CHECK: define %u @g(%u %a) {
+; CHECK: external global %u
 
 %u = type opaque
 %u2 = type { %u, i8 }
 
-declare %u2 @f()
+@g = external global %u
 
-define %u @g(%u %a) {
-  ret %u %a
+define ptr @test() {
+  ret ptr @g
 }
diff --git a/llvm/test/ThinLTO/X86/Inputs/import_opaque_type.ll b/llvm/test/ThinLTO/X86/Inputs/import_opaque_type.ll
@@ -3,11 +3,4 @@ target triple = "x86_64-apple-macosx10.11.0"
 
 %a = type { i8 }
 
-define void @bar(%a) {
-	ret void
-}
-
-define void @baz() {
-	call void @bar(%a undef)
-	ret void
-}
+@g = external global %a
diff --git a/llvm/test/ThinLTO/X86/import_opaque_type.ll b/llvm/test/ThinLTO/X86/import_opaque_type.ll
@@ -3,23 +3,18 @@
 ; RUN: opt -module-summary %p/Inputs/import_opaque_type.ll -o %t2.bc
 ; RUN: llvm-lto -thinlto-action=thinlink -o %t3.bc %t.bc %t2.bc
 
-; Check that we import correctly the imported type to replace the opaque one here
 ; RUN: llvm-lto -thinlto-action=import %t.bc -thinlto-index=%t3.bc -o - | llvm-dis -o - | FileCheck %s
 
 
 target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.11.0"
 
-; CHECK: %a = type { i8 }
+; FIXME: It would be better to produce %a = type { i8 } here
+; CHECK: %a = type opaque
 %a = type opaque
 
-declare void @baz()
-define void @foo(%a) {
-	call void @baz()
-	ret void
-}
+@g = external global %a
 
-define i32 @main() {
-  call void @foo(%a undef)
-	ret i32 0
+define ptr @test() {
+  ret ptr @g
 }
diff --git a/llvm/test/Transforms/InstSimplify/gv-alloca-cmp.ll b/llvm/test/Transforms/InstSimplify/gv-alloca-cmp.ll
@@ -46,7 +46,7 @@ define i1 @cmp_gv_weak_alloca() {
 }
 
 %opaque = type opaque
-@gv_unsized = weak global %opaque zeroinitializer, align 16
+@gv_unsized = external global %opaque, align 16
 
 define i1 @cmp_gv_unsized_alloca() {
 ; CHECK-LABEL: define i1 @cmp_gv_unsized_alloca() {
diff --git a/llvm/test/Transforms/Reg2Mem/non-token-test.ll b/llvm/test/Transforms/Reg2Mem/non-token-test.ll
@@ -1,24 +1,22 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -passes=reg2mem -S < %s | FileCheck %s
 
-%opaque = type opaque
-
-declare %opaque @ret_opaque()
-declare void @pass_opaque(%opaque)
+declare target("opaque") @ret_opaque()
+declare void @pass_opaque(target("opaque"))
 
 define void @test() {
 ; CHECK-LABEL: @test(
 ; CHECK-NEXT:    %"reg2mem alloca point" = bitcast i32 0 to i32
-; CHECK-NEXT:    [[X:%.*]] = call [[OPAQUE:%.*]] @ret_opaque()
+; CHECK-NEXT:    [[X:%.*]] = call target("opaque") @ret_opaque()
 ; CHECK-NEXT:    br label [[NEXT:%.*]]
 ; CHECK:       next:
-; CHECK-NEXT:    call void @pass_opaque([[OPAQUE]] [[X]])
+; CHECK-NEXT:    call void @pass_opaque(target("opaque") [[X]])
 ; CHECK-NEXT:    ret void
 ;
-  %x = call %opaque @ret_opaque()
+  %x = call target("opaque") @ret_opaque()
   br label %next
 
 next:
-  call void @pass_opaque(%opaque %x)
+  call void @pass_opaque(target("opaque") %x)
   ret void
 }
diff --git a/llvm/test/Verifier/memset-pattern-unsized.ll b/llvm/test/Verifier/memset-pattern-unsized.ll
@@ -2,9 +2,7 @@
 
 ; CHECK: unsized types cannot be used as memset patterns
 
-%X = type opaque
-define void @bar(ptr %P, %X %value) {
-  call void @llvm.experimental.memset.pattern.p0.s_s.i32.0(ptr %P, %X %value, i32 4, i1 false)
+define void @bar(ptr %P, target("foo") %value) {
+  call void @llvm.experimental.memset.pattern.p0.s_s.i32.0(ptr %P, target("foo") %value, i32 4, i1 false)
   ret void
 }
-declare void @llvm.experimental.memset.pattern.p0.s_s.i32.0(ptr nocapture, %X, i32, i1) nounwind
diff --git a/llvm/test/Verifier/nofpclass.ll b/llvm/test/Verifier/nofpclass.ll
@@ -45,16 +45,6 @@ define nofpclass(nan) [4 x <8 x i32>] @nofpclass_vector_array_int([4 x <8 x i32>
   ret [4 x <8 x i32>] %arg
 }
 
-%opaque = type opaque
-
-; CHECK: 'nofpclass(nan)' applied to incompatible type!
-; CHECK-NEXT: ptr @nofpclass_opaque_type
-; CHECK-NEXT: 'nofpclass(zero)' applied to incompatible type!
-; CHECK-NEXT: ptr @nofpclass_opaque_type
-define nofpclass(nan) %opaque @nofpclass_opaque_type(%opaque nofpclass(zero) %arg) {
-  ret %opaque %arg
-}
-
 %struct = type { i32, float }
 
 ; CHECK: 'nofpclass(nan)' applied to incompatible type!
diff --git a/llvm/test/Verifier/unsized-types-load.ll b/llvm/test/Verifier/unsized-types-load.ll
@@ -1,10 +1,8 @@
 ; RUN: not opt -passes=verify < %s 2>&1 | FileCheck %s
 
-%X = type opaque
-
 define void @f_0(ptr %ptr) {
-  %t = load %X, ptr %ptr
+  %t = load target("foo"), ptr %ptr
   ret void
 ; CHECK: loading unsized types is not allowed
-; CHECK-NEXT:  %t = load %X, ptr %ptr
+; CHECK-NEXT:  %t = load target("foo"), ptr %ptr
 }
diff --git a/llvm/test/Verifier/unsized-types-store.ll b/llvm/test/Verifier/unsized-types-store.ll
@@ -1,10 +1,8 @@
 ; RUN: not opt -passes=verify < %s 2>&1 | FileCheck %s
 
-%X = type opaque
-
-define void @f_1(%X %val, ptr %ptr) {
-  store %X %val, ptr %ptr
+define void @f_1(target("foo") %val, ptr %ptr) {
+  store target("foo") %val, ptr %ptr
   ret void
 ; CHECK: storing unsized types is not allowed
-; CHECK-NEXT:  store %X %val, ptr %ptr
+; CHECK-NEXT:  store target("foo") %val, ptr %ptr
 }
diff --git a/llvm/unittests/FuzzMutate/OperationsTest.cpp b/llvm/unittests/FuzzMutate/OperationsTest.cpp

Original file line number	Diff line number	Diff line change
`@@ -7,4 +7,4 @@ define void @load_extern(%externref %ref) {`
`7`	`7`	`ret void`
`8`	`8`	`}`
`9`	`9`
`10`		`-; CHECK-ERROR: error: loading unsized types is not allowed`
	`10`	`+; CHECK-ERROR: error: load operand must be a pointer to a first class type`
Original file line number	Diff line number	Diff line change
`@@ -7,4 +7,4 @@ define void @store_extern(%externref %ref) {`
`7`	`7`	`ret void`
`8`	`8`	`}`
`9`	`9`
`10`		`-; CHECK-ERROR: error: storing unsized types is not allowed`
	`10`	`+; CHECK-ERROR: error: invalid type for undef constant`