leofang commented on code in PR #38472:
URL: https://github.com/apache/arrow/pull/38472#discussion_r1417451657


##########
cpp/src/arrow/c/dlpack.cc:
##########
@@ -0,0 +1,151 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements.  See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership.  The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License.  You may obtain a copy of the License at
+//
+//   http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.  See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+#include "arrow/c/dlpack.h"
+
+#include "arrow/array/array_base.h"
+#include "arrow/c/dlpack_abi.h"
+#include "arrow/device.h"
+#include "arrow/type.h"
+
+namespace arrow {
+
+namespace dlpack {
+
+Status getDLDataType(const std::shared_ptr<DataType>& type, DLDataType* out) {
+  DLDataType dtype;
+  dtype.lanes = 1;
+  dtype.bits = type->bit_width();
+  switch (type->id()) {
+    case Type::INT8:
+    case Type::INT16:
+    case Type::INT32:
+    case Type::INT64:
+      dtype.code = DLDataTypeCode::kDLInt;
+      *out = dtype;
+      return Status::OK();
+    case Type::UINT8:
+    case Type::UINT16:
+    case Type::UINT32:
+    case Type::UINT64:
+      dtype.code = DLDataTypeCode::kDLUInt;
+      *out = dtype;
+      return Status::OK();
+    case Type::HALF_FLOAT:
+    case Type::FLOAT:
+    case Type::DOUBLE:
+      dtype.code = DLDataTypeCode::kDLFloat;
+      *out = dtype;
+      return Status::OK();
+    case Type::BOOL:
+      // DLPack supports byte-packed boolean values
+      return Status::TypeError("Bit-packed boolean data type not supported by 
DLPack.");
+    default:
+      return Status::TypeError(
+          "Can only use __dlpack__ on primitive arrays without NullType and 
Decimal "
+          "types.");
+  }
+}
+
+struct DLMTensorCtx {
+  std::shared_ptr<ArrayData> ref;
+  std::vector<int64_t> shape;
+  DLManagedTensor tensor;
+};
+
+static void deleter(DLManagedTensor* arg) {
+  delete static_cast<DLMTensorCtx*>(arg->manager_ctx);
+}
+
+Status ExportArray(const std::shared_ptr<Array>& arr, DLManagedTensor** out) {
+  if (arr->null_count() > 0) {
+    return Status::TypeError("Can only use __dlpack__ on arrays with no 
nulls.");
+  }
+
+  // Define the DLDataType struct
+  // Supported data types: int, uint, float
+  DLDataType arr_type;
+  RETURN_NOT_OK(getDLDataType(arr->type(), &arr_type));
+
+  // Create DLMTensorCtx struct with the reference to
+  // the data of the array
+  std::shared_ptr<ArrayData> array_ref = arr->data();
+  DLMTensorCtx* DLMTensor = new DLMTensorCtx;
+  DLMTensor->ref = array_ref;
+
+  // Define DLManagedTensor struct defined by
+  // DLPack (dlpack_structure.h)
+  DLManagedTensor* dlm_tensor = &DLMTensor->tensor;
+  dlm_tensor->manager_ctx = DLMTensor;
+  dlm_tensor->deleter = &deleter;
+
+  // Define the data pointer to the DLTensor
+  // If array is of length 0, data pointer should be NULL

Review Comment:
   I think the convention came from the old days of CUDA Array Interface where 
either `None` or `0` could be used 
(https://github.com/numba/numba/issues/4140), but I guess it doesn't really 
matter to a C interface. Indeed, `shape` can be used to check whether an array 
is 0-size, but why bother asking the consumer to do an extra check when the 
producer already has this knowledge? 
   
   From the ergonomics perspective, we know the pointer to a 0-size array is 
meaningless and shouldn't be used, and all libraries handle this case 
differently (ex: NumPy would return a pointer address that likely comes from 
its private memory reservation, while CuPy returns null), so setting it to 0 
universally was the safest thing to align all libraries and tell the consumer 
"do not use the pointer."



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to