[mlir][llvm] Fix llvmBuilder for llvm.alloca so it could emit to non-zero addrspace.

[whchung]

Summary

For certain targets alloca can take place on a non-zero addrspace.

No tests are added for this patch as existing patches would all pass.

Subsequent patches to mlir-translate would cover cases where alloca on non-zero addrspace takes place.

[ftynse]

Could you please add a test?

[whchung]

Closing this patch as D79019 supersedes this one.

[ftynse]

Closing this patch as D79019 supersedes this one.

I would actually advise to keep the patches separate, it's easier to review and track changes in the history.

[whchung]

@ftynse I reclaimed the patch. But I'm not sure how to add an extra test specifically for this patch without mlir-translate changes introduced in D79019.

In mlir/test/Target/llvmir.mlir there are tests which checks this particular function how llvm.alloca gets translated to llvm alloca instruction. They are working fine.

And without D79019, there is really no way for mlir-translate to emit alloca on a different addrspace.

[ftynse]

In D79017#2010774, @whchung wrote:

@ftynse I reclaimed the patch. But I'm not sure how to add an extra test specifically for this patch without mlir-translate changes introduced in D79019.

In mlir/test/Target/llvmir.mlir there are tests which checks this particular function how llvm.alloca gets translated to llvm alloca instruction. They are working fine.

And without D79019, there is really no way for mlir-translate to emit alloca on a different addrspace.

Good point. That's why I was thinking that the data layout should be available in the std->llvm conversion and llvm.alloca should produce pointers in the right address space to start with, so there is no need to patch it back in the translation.

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

 def LLVM_FRemOp : LLVM_ArithmeticOp<"frem", "CreateFRem">;
 def LLVM_FNegOp : LLVM_UnaryArithmeticOp<"fneg", "CreateFNeg">;
 
 // Memory-related operations.
 def LLVM_AllocaOp :
     LLVM_OneResultOp<"alloca">,
     Arguments<(ins LLVM_Type:$arraySize, OptionalAttr<I64Attr>:$alignment)> {
   string llvmBuilder = [{
+    llvm::Module *module = builder.GetInsertBlock()->getModule();
+    auto allocaAddrSpace = module->getDataLayout().getAllocaAddrSpace();
     auto *alloca = builder.CreateAlloca(
-      $_resultType->getPointerElementType(), $arraySize);
+      $_resultType->getPointerElementType(), allocaAddrSpace, $arraySize);
     if ($alignment.hasValue()) {
       auto align = $alignment.getValue().getZExtValue();
       if (align != 0)
         alloca->setAlignment(llvm::MaybeAlign(align));
     }
     $res = alloca;
   }];
   let builders = [OpBuilder<