[MLIR] Import LLVM add flag to disable loadalldialects

[wsmoses]

No description provided.

[llvmbot]

@llvm/pr-subscribers-mlir-llvm

@llvm/pr-subscribers-mlir

Author: William Moses (wsmoses)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/122574.diff

2 Files Affected:

• (modified) mlir/include/mlir/Target/LLVMIR/Import.h (+4-1)
• (modified) mlir/lib/Target/LLVMIR/ModuleImport.cpp (+5-2)

diff --git a/mlir/include/mlir/Target/LLVMIR/Import.h b/mlir/include/mlir/Target/LLVMIR/Import.h
index 4a1242a8eb0590..4aa8f2ab7d8ce7 100644
--- a/mlir/include/mlir/Target/LLVMIR/Import.h
+++ b/mlir/include/mlir/Target/LLVMIR/Import.h
@@ -39,10 +39,13 @@ class ModuleOp;
 /// be imported without elements. If set, the option avoids the recursive
 /// traversal of composite type debug information, which can be expensive for
 /// adversarial inputs.
+/// The `loadAllDialects` flag (default on) will load all dialects in the
+/// context.
 OwningOpRef<ModuleOp>
 translateLLVMIRToModule(std::unique_ptr<llvm::Module> llvmModule,
                         MLIRContext *context, bool emitExpensiveWarnings = true,
-                        bool dropDICompositeTypeElements = false);
+                        bool dropDICompositeTypeElements = false,
+                        bool loadAllDialects = true);
 
 /// Translate the given LLVM data layout into an MLIR equivalent using the DLTI
 /// dialect.
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index 2d8d7745eca9bb..f6ad1e310b1678 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -2325,7 +2325,8 @@ ModuleImport::translateLoopAnnotationAttr(const llvm::MDNode *node,
 OwningOpRef<ModuleOp>
 mlir::translateLLVMIRToModule(std::unique_ptr<llvm::Module> llvmModule,
                               MLIRContext *context, bool emitExpensiveWarnings,
-                              bool dropDICompositeTypeElements) {
+                              bool dropDICompositeTypeElements,
+                              bool loadAllDialects) {
   // Preload all registered dialects to allow the import to iterate the
   // registered LLVMImportDialectInterface implementations and query the
   // supported LLVM IR constructs before starting the translation. Assumes the
@@ -2335,7 +2336,9 @@ mlir::translateLLVMIRToModule(std::unique_ptr<llvm::Module> llvmModule,
                             LLVMDialect::getDialectNamespace()));
   assert(llvm::is_contained(context->getAvailableDialects(),
                             DLTIDialect::getDialectNamespace()));
-  context->loadAllAvailableDialects();
+  if (loadAllDialects) {
+    context->loadAllAvailableDialects();
+  }
   OwningOpRef<ModuleOp> module(ModuleOp::create(FileLineColLoc::get(
       StringAttr::get(context, llvmModule->getSourceFileName()), /*line=*/0,
       /*column=*/0)));

[ftynse]

Hmm, I haven't realized we were unconditionally loading all available dialects, this is expensive for library-style usage. Any idea why we are not directly loading just the LLVM dialect and friends instead?

[gysit]

LGTM

We probably load all registered dialects to ensure the import finds all dialects that implement LLVMImportDialectInterface. I agree that we should make this optional, which is what this PR does, or in general just leave loading the dialects to the users of this function.