[Mlir-commits] [mlir] [mlir][gpu] Adding ELF section option to the gpu-module-to-binary pass (PR #119440)

[Guray Ozen]

================
@@ -664,9 +664,18 @@ NVVMTargetAttrImpl::createObject(Attribute attribute, Operation *module,
   gpu::CompilationTarget format = options.getCompilationTarget();
   DictionaryAttr objectProps;
   Builder builder(attribute.getContext());
+  SmallVector<NamedAttribute, 2> properties;
   if (format == gpu::CompilationTarget::Assembly)
-    objectProps = builder.getDictionaryAttr(
-        {builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO()))});
+    properties.push_back(
+        builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO())));
+
+  if (auto section = options.getELFSection(); !section.empty())
+    properties.push_back(
+        builder.getNamedAttr("section", builder.getStringAttr(section)));
----------------
grypp wrote:

maybe create a constant `constexpr kSectionName = "section"`

https://github.com/llvm/llvm-project/pull/119440