zhiics commented on a change in pull request #6109:
URL: https://github.com/apache/incubator-tvm/pull/6109#discussion_r459618012
##########
File path: tests/python/contrib/test_arm_compute_lib/test_conv2d.py
##########
@@ -127,51 +241,31 @@ def test_conv2d():
device = Device()
np.random.seed(0)
+ r = random.Random(0)
Review comment:
BTW, why do we need a random input? It looks we can just give the output
channel a value or multiple values according to the number of trails, right?
##########
File path: src/runtime/contrib/arm_compute_lib/acl_utils.cc
##########
@@ -108,6 +111,30 @@ arm_compute::PadStrideInfo ToACLPadStride(const
std::vector<std::string>& pad,
arm_compute::DimensionRoundingType::FLOOR);
}
+arm_compute::DataType MakeDataType(const DLDataType& data_type) {
+ if (data_type.code == DLDataTypeCode::kDLFloat && data_type.bits == 32) {
+ return arm_compute::DataType::F32;
+ } else if (data_type.code == DLDataTypeCode::kDLUInt && data_type.bits == 8)
{
+ return arm_compute::DataType::QASYMM8;
+ } else if (data_type.code == DLDataTypeCode::kDLInt && data_type.bits == 32)
{
+ return arm_compute::DataType::S32;
+ } else {
+ LOG(FATAL) << "Datatype " << data_type << " unsupported by ACL runtime";
+ return arm_compute::DataType::UNKNOWN;
+ }
+}
+
+template <typename T>
+std::vector<T> GetVectorFromDLTensor(const DLTensor* tensor) {
Review comment:
yeah, maybe we have it but I am not sure where it is either. But do we
have to flatten the tensor during runtime?
##########
File path: src/runtime/contrib/arm_compute_lib/acl_runtime.cc
##########
@@ -198,13 +225,31 @@ class ACLRuntime : public JSONRuntimeBase {
arm_compute::Size2D dilation_2d(std::stoi(dilation[0]),
std::stoi(dilation[1]));
- layer->outputs.push_back(MakeOutputTensor(node.GetOpShape()[0]));
+ // Collect inputs and outputs, handling both nn.conv2d and qnn.conv2d
cases.
+ std::vector<JSONGraphNodeEntry> inputs = node.GetInputs();
+ size_t num_inputs = inputs.size();
+ bool has_bias;
+ if (node.GetOpName() == "qnn.conv2d") {
Review comment:
`CHECK_GE(num_inputs, 8U)`
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
