[mlir] LLVM dialect: use addressof instead of constant to create func…

…tion pointers

`llvm.mlir.constant` was originally introduced as an LLVM dialect counterpart
to `std.constant`. As such, it was supporting "function pointer" constants
derived from the symbol name. This is different from `std.constant` that allows
for creation of a "function" constant since MLIR, unlike LLVM IR, supports
this. Later, `llvm.mlir.addressof` was introduced as an Op that obtains a
constant pointer to a global in the LLVM dialect. It naturally extends to
functions (in LLVM IR, functions are globals) and should be used for defining
"function pointer" values instead.

Fixes PR46344.

Differential Revision: https://reviews.llvm.org/D82667

mlir/docs/Dialects/LLVM.md

@@ -313,10 +313,28 @@ Bitwise reinterpretation: `bitcast <value>`.
 
 Selection: `select <condition>, <lhs>, <rhs>`.
 
-### Auxiliary MLIR operations
-
-These operations do not have LLVM IR counterparts but are necessary to map LLVM
-IR into MLIR. They should be prefixed with `llvm.mlir`.
+### Auxiliary MLIR Operations for Constants and Globals
+
+LLVM IR has broad support for first-class constants, which is not the case for
+MLIR. Instead, constants are defined in MLIR as regular SSA values produced by
+operations with specific traits. The LLVM dialect provides a set of operations
+that model LLVM IR constants. These operations do not correspond to LLVM IR
+instructions and are therefore prefixed with `llvm.mlir`.
+
+Inline constants can be created by `llvm.mlir.constant`, which currently
+supports integer, float, string or elements attributes (constant sturcts are not
+currently supported). LLVM IR constant expressions are expected to be
+constructed as sequences of regular operations on SSA values produced by
+`llvm.mlir.constant`. Additionally, MLIR provides semantically-charged
+operations `llvm.mlir.undef` and `llvm.mlir.null` for the corresponding
+constants.
+
+LLVM IR globals can be defined using `llvm.mlir.global` at the module level,
+except for functions that are defined with `llvm.func`. Globals, both variables
+and functions, can be accessed by taking their address with the
+`llvm.mlir.addressof` operation, which produces a pointer to the named global,
+unlike the `llvm.mlir.constant` that produces the value of the same type as the
+constant.
 
 #### `llvm.mlir.addressof`
 
@@ -328,11 +346,17 @@ Examples:
 
 ```mlir
 func @foo() {
-  // Get the address of a global.
+  // Get the address of a global variable.
   %0 = llvm.mlir.addressof @const : !llvm<"i32*">
 
   // Use it as a regular pointer.
   %1 = llvm.load %0 : !llvm<"i32*">
+
+  // Get the address of a function.
+  %2 = llvm.mlir.addressof @foo : !llvm<"void ()*">
+
+  // The function address can be used for indirect calls.
+  llvm.call %2() : () -> ()
 }
 
 // Define the global.

mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td

@@ -575,6 +575,8 @@ def Linkage : LLVM_EnumAttr<
 def LLVM_AddressOfOp
     : LLVM_OneResultOp<"mlir.addressof">,
       Arguments<(ins FlatSymbolRefAttr:$global_name)> {
+  let summary = "Creates a pointer pointing to a global or a function";
+
   let builders = [
     OpBuilder<"OpBuilder &builder, OperationState &result, LLVMType resType, "
               "StringRef name, ArrayRef<NamedAttribute> attrs = {}", [{
@@ -586,13 +588,21 @@ def LLVM_AddressOfOp
               "ArrayRef<NamedAttribute> attrs = {}", [{
       build(builder, result,
             global.getType().getPointerTo(global.addr_space().getZExtValue()),
-            global.sym_name(), attrs);}]>
+            global.sym_name(), attrs);}]>,
+
+    OpBuilder<"OpBuilder &builder, OperationState &result, LLVMFuncOp func, "
+              "ArrayRef<NamedAttribute> attrs = {}", [{
+      build(builder, result,
+            func.getType().getPointerTo(), func.getName(), attrs);}]>
   ];
 
   let extraClassDeclaration = [{
     /// Return the llvm.mlir.global operation that defined the value referenced
     /// here.
     GlobalOp getGlobal();
+
+    /// Return the llvm.func operation that is referenced here.
+    LLVMFuncOp getFunction();
   }];
 
   let assemblyFormat = "$global_name attr-dict `:` type($res)";
@@ -733,6 +743,7 @@ def LLVM_ConstantOp
       LLVM_Builder<"$res = getLLVMConstant($_resultType, $value, $_location);">
 {
   let assemblyFormat = "`(` $value `)` attr-dict `:` type($res)";
+  let verifier = [{ return ::verify(*this); }];
 }
 
 def LLVM_DialectCastOp : LLVM_Op<"mlir.cast", [NoSideEffect]>,

mlir/lib/Conversion/StandardToLLVM/StandardToLLVM.cpp

@@ -1366,8 +1366,6 @@ using AddFOpLowering = VectorConvertToLLVMPattern<AddFOp, LLVM::FAddOp>;
 using AddIOpLowering = VectorConvertToLLVMPattern<AddIOp, LLVM::AddOp>;
 using AndOpLowering = VectorConvertToLLVMPattern<AndOp, LLVM::AndOp>;
 using CeilFOpLowering = VectorConvertToLLVMPattern<CeilFOp, LLVM::FCeilOp>;
-using ConstLLVMOpLowering =
-    OneToOneConvertToLLVMPattern<ConstantOp, LLVM::ConstantOp>;
 using CopySignOpLowering =
     VectorConvertToLLVMPattern<CopySignOp, LLVM::CopySignOp>;
 using CosOpLowering = VectorConvertToLLVMPattern<CosOp, LLVM::CosOp>;
@@ -1541,6 +1539,39 @@ struct SubCFOpLowering : public ConvertOpToLLVMPattern<SubCFOp> {
   }
 };
 
+struct ConstantOpLowering : public ConvertOpToLLVMPattern<ConstantOp> {
+  using ConvertOpToLLVMPattern<ConstantOp>::ConvertOpToLLVMPattern;
+
+  LogicalResult
+  matchAndRewrite(Operation *operation, ArrayRef<Value> operands,
+                  ConversionPatternRewriter &rewriter) const override {
+    auto op = cast<ConstantOp>(operation);
+    // If constant refers to a function, convert it to "addressof".
+    if (auto symbolRef = op.getValue().dyn_cast<FlatSymbolRefAttr>()) {
+      auto type = typeConverter.convertType(op.getResult().getType())
+                      .dyn_cast_or_null<LLVM::LLVMType>();
+      if (!type)
+        return rewriter.notifyMatchFailure(op, "failed to convert result type");
+
+      MutableDictionaryAttr attrs(op.getAttrs());
+      attrs.remove(rewriter.getIdentifier("value"));
+      rewriter.replaceOpWithNewOp<LLVM::AddressOfOp>(
+          op, type.cast<LLVM::LLVMType>(), symbolRef.getValue(),
+          attrs.getAttrs());
+      return success();
+    }
+
+    // Calling into other scopes (non-flat reference) is not supported in LLVM.
+    if (op.getValue().isa<SymbolRefAttr>())
+      return rewriter.notifyMatchFailure(
+          op, "referring to a symbol outside of the current module");
+
+    return LLVM::detail::oneToOneRewrite(op,
+                                         LLVM::ConstantOp::getOperationName(),
+                                         operands, typeConverter, rewriter);
+  }
+};
+
 // Check if the MemRefType `type` is supported by the lowering. We currently
 // only support memrefs with identity maps.
 static bool isSupportedMemRefType(MemRefType type) {
@@ -3129,7 +3160,7 @@ void mlir::populateStdToLLVMNonMemoryConversionPatterns(
       CondBranchOpLowering,
       CopySignOpLowering,
       CosOpLowering,
-      ConstLLVMOpLowering,
+      ConstantOpLowering,
       CreateComplexOpLowering,
       DialectCastOpLowering,
       DivFOpLowering,

mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp

@@ -857,25 +857,40 @@ static ParseResult parseReturnOp(OpAsmParser &parser, OperationState &result) {
 // Verifier for LLVM::AddressOfOp.
 //===----------------------------------------------------------------------===//
 
-GlobalOp AddressOfOp::getGlobal() {
-  Operation *module = getParentOp();
+template <typename OpTy>
+static OpTy lookupSymbolInModule(Operation *parent, StringRef name) {
+  Operation *module = parent;
   while (module && !satisfiesLLVMModule(module))
     module = module->getParentOp();
   assert(module && "unexpected operation outside of a module");
-  return dyn_cast_or_null<LLVM::GlobalOp>(
-      mlir::SymbolTable::lookupSymbolIn(module, global_name()));
+  return dyn_cast_or_null<OpTy>(
+      mlir::SymbolTable::lookupSymbolIn(module, name));
+}
+
+GlobalOp AddressOfOp::getGlobal() {
+  return lookupSymbolInModule<LLVM::GlobalOp>(getParentOp(), global_name());
+}
+
+LLVMFuncOp AddressOfOp::getFunction() {
+  return lookupSymbolInModule<LLVM::LLVMFuncOp>(getParentOp(), global_name());
 }
 
 static LogicalResult verify(AddressOfOp op) {
   auto global = op.getGlobal();
-  if (!global)
+  auto function = op.getFunction();
+  if (!global && !function)
+    return op.emitOpError(
+        "must reference a global defined by 'llvm.mlir.global' or 'llvm.func'");
+
+  if (global &&
+      global.getType().getPointerTo(global.addr_space().getZExtValue()) !=
+          op.getResult().getType())
     return op.emitOpError(
-        "must reference a global defined by 'llvm.mlir.global'");
+        "the type must be a pointer to the type of the referenced global");
 
-  if (global.getType().getPointerTo(global.addr_space().getZExtValue()) !=
-      op.getResult().getType())
+  if (function && function.getType().getPointerTo() != op.getResult().getType())
     return op.emitOpError(
-        "the type must be a pointer to the type of the referred global");
+        "the type must be a pointer to the type of the referenced function");
 
   return success();
 }
@@ -1395,6 +1410,18 @@ static LogicalResult verify(LLVM::NullOp op) {
   return success();
 }
 
+//===----------------------------------------------------------------------===//
+// Verification for LLVM::ConstantOp.
+//===----------------------------------------------------------------------===//
+
+static LogicalResult verify(LLVM::ConstantOp op) {
+  if (!(op.value().isa<IntegerAttr>() || op.value().isa<FloatAttr>() ||
+        op.value().isa<ElementsAttr>() || op.value().isa<StringAttr>()))
+    return op.emitOpError()
+           << "only supports integer, float, string or elements attributes";
+  return success();
+}
+
 //===----------------------------------------------------------------------===//
 // Utility functions for parsing atomic ops
 //===----------------------------------------------------------------------===//

mlir/lib/Target/LLVMIR/ConvertFromLLVMIR.cpp

@@ -405,6 +405,9 @@ Value Importer::processConstant(llvm::Constant *c) {
     LLVMType type = processType(c->getType());
     if (!type)
       return nullptr;
+    if (auto symbolRef = attr.dyn_cast<FlatSymbolRefAttr>())
+      return instMap[c] = bEntry.create<AddressOfOp>(unknownLoc, type,
+                                                     symbolRef.getValue());
     return instMap[c] = bEntry.create<ConstantOp>(unknownLoc, type, attr);
   }
   if (auto *cn = dyn_cast<llvm::ConstantPointerNull>(c)) {

mlir/lib/Target/LLVMIR/ModuleTranslation.cpp

@@ -447,10 +447,15 @@ LogicalResult ModuleTranslation::convertOperation(Operation &opInst,
   // emit any LLVM instruction.
   if (auto addressOfOp = dyn_cast<LLVM::AddressOfOp>(opInst)) {
     LLVM::GlobalOp global = addressOfOp.getGlobal();
+    LLVM::LLVMFuncOp function = addressOfOp.getFunction();
+
     // The verifier should not have allowed this.
-    assert(global && "referencing an undefined global");
+    assert((global || function) &&
+           "referencing an undefined global or function");
 
-    valueMapping[addressOfOp.getResult()] = globalsMapping.lookup(global);
+    valueMapping[addressOfOp.getResult()] =
+        global ? globalsMapping.lookup(global)
+               : functionMapping.lookup(function.getName());
     return success();
   }
 

mlir/test/Conversion/StandardToLLVM/convert-funcs.mlir

@@ -31,7 +31,7 @@ func @pass_through(%arg0: () -> ()) -> (() -> ()) {
 // CHECK-NEXT:  llvm.br ^bb1(%arg0 : !llvm<"void ()*">)
   br ^bb1(%arg0 : () -> ())
 
-//CHECK-NEXT: ^bb1(%0: !llvm<"void ()*">):	// pred: ^bb0
+//CHECK-NEXT: ^bb1(%0: !llvm<"void ()*">):
 ^bb1(%bbarg: () -> ()):
 // CHECK-NEXT:  llvm.return %0 : !llvm<"void ()*">
   return %bbarg : () -> ()
@@ -40,11 +40,12 @@ func @pass_through(%arg0: () -> ()) -> (() -> ()) {
 // CHECK-LABEL: llvm.func @body(!llvm.i32)
 func @body(i32)
 
-// CHECK-LABEL: llvm.func @indirect_const_call(%arg0: !llvm.i32) {
+// CHECK-LABEL: llvm.func @indirect_const_call
+// CHECK-SAME: (%[[ARG0:.*]]: !llvm.i32) {
 func @indirect_const_call(%arg0: i32) {
-// CHECK-NEXT: %0 = llvm.mlir.constant(@body) : !llvm<"void (i32)*">
+// CHECK-NEXT: %[[ADDR:.*]] = llvm.mlir.addressof @body : !llvm<"void (i32)*">
   %0 = constant @body : (i32) -> ()
-// CHECK-NEXT:  llvm.call %0(%arg0) : (!llvm.i32) -> ()
+// CHECK-NEXT:  llvm.call %[[ADDR]](%[[ARG0:.*]]) : (!llvm.i32) -> ()
   call_indirect %0(%arg0) : (i32) -> ()
 // CHECK-NEXT:  llvm.return
   return

mlir/test/Dialect/LLVMIR/global.mlir

@@ -140,12 +140,21 @@ func @foo() {
 llvm.mlir.global internal @foo(0: i32) : !llvm.i32
 
 func @bar() {
-  // expected-error @+1 {{the type must be a pointer to the type of the referred global}}
+  // expected-error @+1 {{the type must be a pointer to the type of the referenced global}}
   llvm.mlir.addressof @foo : !llvm<"i64*">
 }
 
 // -----
 
+llvm.func @foo()
+
+llvm.func @bar() {
+  // expected-error @+1 {{the type must be a pointer to the type of the referenced function}}
+  llvm.mlir.addressof @foo : !llvm<"i8*">
+}
+
+// -----
+
 // expected-error @+2 {{'llvm.mlir.global' op expects regions to end with 'llvm.return', found 'llvm.mlir.constant'}}
 // expected-note @+1 {{in custom textual format, the absence of terminator implies 'llvm.return'}}
 llvm.mlir.global internal @g() : !llvm.i64 {
@@ -172,14 +181,14 @@ llvm.mlir.global internal @g(43 : i64) : !llvm.i64 {
 
 llvm.mlir.global internal @g(32 : i64) {addr_space = 3: i32} : !llvm.i64
 func @mismatch_addr_space_implicit_global() {
-  // expected-error @+1 {{op the type must be a pointer to the type of the referred global}}
+  // expected-error @+1 {{op the type must be a pointer to the type of the referenced global}}
   llvm.mlir.addressof @g : !llvm<"i64*">
 }
 
 // -----
 
 llvm.mlir.global internal @g(32 : i64) {addr_space = 3: i32} : !llvm.i64
 func @mismatch_addr_space() {
-  // expected-error @+1 {{op the type must be a pointer to the type of the referred global}}
+  // expected-error @+1 {{op the type must be a pointer to the type of the referenced global}}
   llvm.mlir.addressof @g : !llvm<"i64 addrspace(4)*">
 }

mlir/test/Dialect/LLVMIR/invalid.mlir

@@ -153,6 +153,13 @@ func @call_non_llvm_input(%callee : (i32) -> (), %arg : i32) {
 
 // -----
 
+func @constant_wrong_type() {
+  // expected-error@+1 {{only supports integer, float, string or elements attributes}}
+  llvm.mlir.constant(@constant_wrong_type) : !llvm<"void ()*">
+}
+
+// -----
+
 func @insertvalue_non_llvm_type(%a : i32, %b : i32) {
   // expected-error@+1 {{expected LLVM IR Dialect type}}
   llvm.insertvalue %a, %b[0] : i32

mlir/test/Dialect/LLVMIR/roundtrip.mlir

@@ -55,12 +55,12 @@ func @ops(%arg0: !llvm.i32, %arg1: !llvm.float,
 // CHECK: %[[STRUCT:.*]] = llvm.call @foo(%[[I32]]) : (!llvm.i32) -> !llvm<"{ i32, double, i32 }">
 // CHECK: %[[VALUE:.*]] = llvm.extractvalue %[[STRUCT]][0] : !llvm<"{ i32, double, i32 }">
 // CHECK: %[[NEW_STRUCT:.*]] = llvm.insertvalue %[[VALUE]], %[[STRUCT]][2] : !llvm<"{ i32, double, i32 }">
-// CHECK: %[[FUNC:.*]] = llvm.mlir.constant(@foo) : !llvm<"{ i32, double, i32 } (i32)*">
+// CHECK: %[[FUNC:.*]] = llvm.mlir.addressof @foo : !llvm<"{ i32, double, i32 } (i32)*">
 // CHECK: %{{.*}} = llvm.call %[[FUNC]](%[[I32]]) : (!llvm.i32) -> !llvm<"{ i32, double, i32 }">
   %17 = llvm.call @foo(%arg0) : (!llvm.i32) -> !llvm<"{ i32, double, i32 }">
   %18 = llvm.extractvalue %17[0] : !llvm<"{ i32, double, i32 }">
   %19 = llvm.insertvalue %18, %17[2] : !llvm<"{ i32, double, i32 }">
-  %20 = llvm.mlir.constant(@foo) : !llvm<"{ i32, double, i32 } (i32)*">
+  %20 = llvm.mlir.addressof @foo : !llvm<"{ i32, double, i32 } (i32)*">
   %21 = llvm.call %20(%arg0) : (!llvm.i32) -> !llvm<"{ i32, double, i32 }">
 
 
@@ -130,8 +130,8 @@ func @ops(%arg0: !llvm.i32, %arg1: !llvm.float,
 }
 
 // An larger self-contained function.
-// CHECK-LABEL: func @foo(%{{.*}}: !llvm.i32) -> !llvm<"{ i32, double, i32 }"> {
-func @foo(%arg0: !llvm.i32) -> !llvm<"{ i32, double, i32 }"> {
+// CHECK-LABEL: llvm.func @foo(%{{.*}}: !llvm.i32) -> !llvm<"{ i32, double, i32 }"> {
+llvm.func @foo(%arg0: !llvm.i32) -> !llvm<"{ i32, double, i32 }"> {
 // CHECK:  %[[V0:.*]] = llvm.mlir.constant(3 : i64) : !llvm.i32
 // CHECK:  %[[V1:.*]] = llvm.mlir.constant(3 : i64) : !llvm.i32
 // CHECK:  %[[V2:.*]] = llvm.mlir.constant(4.200000e+01 : f64) : !llvm.double

mlir/test/Target/import.ll

@@ -234,7 +234,7 @@ define void @FPArithmetic(float %a, float %b, double %c, double %d) {
 ; CHECK-LABEL: @precaller
 define i32 @precaller() {
   %1 = alloca i32 ()*
-  ; CHECK: %[[func:.*]] = llvm.mlir.constant(@callee) : !llvm<"i32 ()*">
+  ; CHECK: %[[func:.*]] = llvm.mlir.addressof @callee : !llvm<"i32 ()*">
   ; CHECK: llvm.store %[[func]], %[[loc:.*]]
   store i32 ()* @callee, i32 ()** %1
   ; CHECK: %[[indir:.*]] = llvm.load %[[loc]]
@@ -252,7 +252,7 @@ define i32 @callee() {
 ; CHECK-LABEL: @postcaller
 define i32 @postcaller() {
   %1 = alloca i32 ()*
-  ; CHECK: %[[func:.*]] = llvm.mlir.constant(@callee) : !llvm<"i32 ()*">
+  ; CHECK: %[[func:.*]] = llvm.mlir.addressof @callee : !llvm<"i32 ()*">
   ; CHECK: llvm.store %[[func]], %[[loc:.*]]
   store i32 ()* @callee, i32 ()** %1
   ; CHECK: %[[indir:.*]] = llvm.load %[[loc]]
@@ -317,4 +317,4 @@ define i32 @useFenceInst() {
   ;CHECK: llvm.fence seq_cst
   fence syncscope("") seq_cst
   ret i32 0
-}
+}

mlir/test/Target/llvmir.mlir

@@ -886,7 +886,7 @@ llvm.func @ops(%arg0: !llvm.float, %arg1: !llvm.float, %arg2: !llvm.i32, %arg3:
 // CHECK-LABEL: define void @indirect_const_call(i64 {{%.*}})
 llvm.func @indirect_const_call(%arg0: !llvm.i64) {
 // CHECK-NEXT:  call void @body(i64 %0)
-  %0 = llvm.mlir.constant(@body) : !llvm<"void (i64)*">
+  %0 = llvm.mlir.addressof @body : !llvm<"void (i64)*">
   llvm.call %0(%arg0) : (!llvm.i64) -> ()
 // CHECK-NEXT:  ret void
   llvm.return