[mlir][llvm] Move llvm attribute bases to utils

[xlauko]

This enables other attributes, not necessarily defined in LLVMEnums.td, to be used with gen-enum-to/from-llvmir-conversions.

[llvmbot]

@llvm/pr-subscribers-mlir-llvm

@llvm/pr-subscribers-mlir

Author: Henrich Lauko (xlauko)

Changes

This enables other attributes, not necessarily defined in LLVMEnums.td, to be used with gen-enum-to/from-llvmir-conversions.

---

Full diff: https://github.com/llvm/llvm-project/pull/118491.diff

2 Files Affected:

• (modified) mlir/include/mlir/Dialect/LLVMIR/LLVMEnums.td (+1-53)
• (added) mlir/include/mlir/Dialect/LLVMIR/Utils.td (+71)

diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMEnums.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMEnums.td
index 4a43c16903394f..ee6873e65d966e 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMEnums.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMEnums.td
@@ -10,61 +10,9 @@
 #define LLVMIR_ENUMS
 
 include "mlir/Dialect/LLVMIR/LLVMDialect.td"
+include "mlir/Dialect/LLVMIR/Utils.td"
 include "mlir/IR/EnumAttr.td"
 
-//===----------------------------------------------------------------------===//
-// Base classes for LLVM enum attributes.
-//===----------------------------------------------------------------------===//
-
-// Case of the LLVM enum attribute backed by I64Attr with customized string
-// representation that corresponds to what is visible in the textual IR form.
-// The parameters are as follows:
-//   - `cppSym`: name of the C++ enumerant for this case in MLIR API;
-//   - `irSym`: keyword used in the custom form of MLIR operation;
-//   - `llvmSym`: name of the C++ enumerant for this case in LLVM API.
-// For example, `LLVM_EnumAttrCase<"Weak", "weak", "WeakAnyLinkage">` is usable
-// as `<MlirEnumName>::Weak` in MLIR API, `WeakAnyLinkage` in LLVM API and
-// is printed/parsed as `weak` in MLIR custom textual format.
-class LLVM_EnumAttrCase<string cppSym, string irSym, string llvmSym, int val> :
-    I64EnumAttrCase<cppSym, val, irSym> {
-  // The name of the equivalent enumerant in LLVM.
-  string llvmEnumerant = llvmSym;
-}
-
-// LLVM enum attribute backed by I64Attr with string representation
-// corresponding to what is visible in the textual IR form.
-// The parameters are as follows:
-//   - `name`: name of the C++ enum class in MLIR API;
-//   - `llvmName`: name of the C++ enum in LLVM API;
-//   - `description`: textual description for documentation purposes;
-//   - `cases`: list of enum cases;
-//   - `unsupportedCases`: optional list of unsupported enum cases.
-// For example, `LLVM_EnumAttr<Linkage, "::llvm::GlobalValue::LinkageTypes`
-// produces `mlir::LLVM::Linkage` enum class in MLIR API that corresponds to (a
-// subset of) values in the `llvm::GlobalValue::LinkageTypes` in LLVM API.
-// All unsupported cases are excluded from the MLIR enum and trigger an error
-// during the import from LLVM IR. They are useful to handle sentinel values
-// such as `llvm::AtomicRMWInst::BinOp::BAD_BINOP` that LLVM commonly uses to
-// terminate its enums.
-class LLVM_EnumAttr<string name, string llvmName, string description,
-                    list<LLVM_EnumAttrCase> cases,
-                    list<LLVM_EnumAttrCase> unsupportedCases = []> :
-    I64EnumAttr<name, description, cases> {
-  // List of unsupported cases that have no conversion to an MLIR value.
-  list<LLVM_EnumAttrCase> unsupported = unsupportedCases;
-
-  // The equivalent enum class name in LLVM.
-  string llvmClassName = llvmName;
-}
-
-// LLVM_CEnumAttr is functionally identical to LLVM_EnumAttr, but to be used for
-// non-class enums.
-class LLVM_CEnumAttr<string name, string llvmNS, string description,
-      list<LLVM_EnumAttrCase> cases> :
-    I64EnumAttr<name, description, cases> {
-  string llvmClassName = llvmNS;
-}
-
 //===----------------------------------------------------------------------===//
 // AsmDialect
 //===----------------------------------------------------------------------===//
diff --git a/mlir/include/mlir/Dialect/LLVMIR/Utils.td b/mlir/include/mlir/Dialect/LLVMIR/Utils.td
new file mode 100644
index 00000000000000..6444aa4c4269b8
--- /dev/null
+++ b/mlir/include/mlir/Dialect/LLVMIR/Utils.td
@@ -0,0 +1,71 @@
+//===-- Utils.td - MLIR LLVM IR utilities file -------------*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains utilities to map from MLIR LLVM IR dialect to LLVM IR.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_LLVMIR_UTILS_TD
+#define MLIR_DIALECT_LLVMIR_UTILS_TD
+
+include "mlir/IR/EnumAttr.td"
+
+//===----------------------------------------------------------------------===//
+// Base classes for LLVM enum attributes.
+//===----------------------------------------------------------------------===//
+
+// Case of the LLVM enum attribute backed by I64Attr with customized string
+// representation that corresponds to what is visible in the textual IR form.
+// The parameters are as follows:
+//   - `cppSym`: name of the C++ enumerant for this case in MLIR API;
+//   - `irSym`: keyword used in the custom form of MLIR operation;
+//   - `llvmSym`: name of the C++ enumerant for this case in LLVM API.
+// For example, `LLVM_EnumAttrCase<"Weak", "weak", "WeakAnyLinkage">` is usable
+// as `<MlirEnumName>::Weak` in MLIR API, `WeakAnyLinkage` in LLVM API and
+// is printed/parsed as `weak` in MLIR custom textual format.
+class LLVM_EnumAttrCase<string cppSym, string irSym, string llvmSym, int val> :
+    I64EnumAttrCase<cppSym, val, irSym> {
+  // The name of the equivalent enumerant in LLVM.
+  string llvmEnumerant = llvmSym;
+}
+
+// LLVM enum attribute backed by I64Attr with string representation
+// corresponding to what is visible in the textual IR form.
+// The parameters are as follows:
+//   - `name`: name of the C++ enum class in MLIR API;
+//   - `llvmName`: name of the C++ enum in LLVM API;
+//   - `description`: textual description for documentation purposes;
+//   - `cases`: list of enum cases;
+//   - `unsupportedCases`: optional list of unsupported enum cases.
+// For example, `LLVM_EnumAttr<Linkage, "::llvm::GlobalValue::LinkageTypes`
+// produces `mlir::LLVM::Linkage` enum class in MLIR API that corresponds to (a
+// subset of) values in the `llvm::GlobalValue::LinkageTypes` in LLVM API.
+// All unsupported cases are excluded from the MLIR enum and trigger an error
+// during the import from LLVM IR. They are useful to handle sentinel values
+// such as `llvm::AtomicRMWInst::BinOp::BAD_BINOP` that LLVM commonly uses to
+// terminate its enums.
+class LLVM_EnumAttr<string name, string llvmName, string description,
+                    list<LLVM_EnumAttrCase> cases,
+                    list<LLVM_EnumAttrCase> unsupportedCases = []> :
+    I64EnumAttr<name, description, cases> {
+  // List of unsupported cases that have no conversion to an MLIR value.
+  list<LLVM_EnumAttrCase> unsupported = unsupportedCases;
+
+  // The equivalent enum class name in LLVM.
+  string llvmClassName = llvmName;
+}
+
+// LLVM_CEnumAttr is functionally identical to LLVM_EnumAttr, but to be used for
+// non-class enums.
+class LLVM_CEnumAttr<string name, string llvmNS, string description,
+      list<LLVM_EnumAttrCase> cases> :
+    I64EnumAttr<name, description, cases> {
+  string llvmClassName = llvmNS;
+}
+
+#endif // MLIR_DIALECT_LLVMIR_UTILS_TD

[xlauko]

@joker-eph where would you put this file, we were not sure with @gysit whether this a right place for now?

The goal is to have tablegen conversion to LLVM IR available to other dialects, but this would make "dependency" on LLVM dialect. The motivation for this is to eventually separate linkage attributes to a dialect in a similar manner to ptr, because it is copy-pasted in one way or another in multiple tools (clangir, vast, flang).

[gysit]

If we decide to go this route, I would suggest to rename to LLVMEnumsBase.td? That way we follow a naming convention similar to LLVMOpsBase.td.

[xlauko]

There is also similar tooling for instrinsics, we might want to separate it in a similar manner then too.

[gysit]

There is also similar tooling for instrinsics, we might want to separate it in a similar manner then too.

Yes the intrinsics use LLVMOpBase.td since the intrinsic base classes share some definitions with the other LLVM dialect operations.

I am fine with making this new linking attribute dialect dependent on LLVM dialect. That would make the new dialect conceptually similar to the other target dialects. Pointer dialect so far is completely independent of LLVM dialect and I assume that can remain like due the LLVMTranslationDialectInterface. However, that mechanism cannot be use for the sharing tablegen files.