fix(imperative/interpreter): use info->h_value for host compute

Rather than use host_only Tensor and modify Put's behavior.

GitOrigin-RevId: f890d66acb

imperative/src/impl/interpreter_impl.cpp

@@ -34,8 +34,13 @@ Handle ChannelImpl::put(const HostTensorND& value, bool no_cache) {
     info->desc.layout = value.layout();
     info->desc.comp_node = value.comp_node();
     info->desc.value = value.proxy_to_default_cpu();
+    info->h_value = value;
     m_valid_handle.insert(info);
     m_buffer.enqueue(Put{info, value, no_cache});
+    if (m_async_level == 0) {
+        sync();
+        info->desc.comp_node.sync();
+    }
     return info;
 }
 
@@ -90,14 +95,19 @@ void ChannelImpl::dispatch_default_cpu(
     {
         MGB_LOCK_GUARD(m_mutex);
         for (auto&& info : input_infos) {
-            mgb_assert(info->ptr, "invalid tensor ptr!");
+            auto input_cn = info->desc.comp_node;
             if (!output_cn.valid()) {
-                output_cn = info->ptr->comp_node();
+                output_cn = input_cn;
+            } else {
+                mgb_assert(output_cn == input_cn, "cannot decide output comp node");
+            }
+
+            if (info->ptr && info->ptr->try_get_value()) {
+                input_tensornds.emplace_back(info->ptr->get_value().proxy_to_default_cpu());
             } else {
-                mgb_assert(output_cn == info->ptr->comp_node(), "cannot decide output comp node");
+                mgb_assert(!info->h_value.empty(), "inp->h_value is empty!");
+                input_tensornds.emplace_back(info->h_value.proxy_to_default_cpu());
             }
-            mgb_assert(info->ptr->try_get_value(), "no valid host value");
-            input_tensornds.emplace_back(info->ptr->get_value().proxy_to_default_cpu());
         }
     }
 
@@ -116,18 +126,12 @@ void ChannelImpl::dispatch_default_cpu(
     SmallVector<TensorInfo*> output_infos;
     output_infos.reserve(output_descs.size());
     for (auto&& tensornd : output_tensornds) {
-        // tensornd -> host_tensornd
         HostTensorND host_tensornd = HostTensorND::make_proxy(tensornd)
             .proxy_to_comp_node(output_cn);
-        // tensornd -> desc
-        LogicalTensorDesc desc = {tensornd.layout(), output_cn, tensornd};
-        // tensornd -> tensor
-        auto info = alloc();
-        info->desc = desc;
-        m_valid_handle.insert(info);
+        // use `put` for consistency
+        auto info = reinterpret_cast<TensorInfo*>(put(host_tensornd, false));
+        mgb_assert(info->desc.layout.ndim != 0);
         output_infos.push_back(info);
-        info->ptr = Tensor::make(host_tensornd, true);  // host_only=true
-        info->value_fetched = true;
         outputs->push_back(info);
     }
 
@@ -159,6 +163,11 @@ void ChannelImpl::dispatch_kernel(
     for (auto&& desc : output_descs) {
         auto info = alloc();
         info->desc = desc;
+        // make sure desc's value is consistent with h_value
+        if (!info->desc.value.empty()) {
+            info->h_value = HostTensorND::make_proxy(desc.value)
+                .proxy_to_comp_node(desc.comp_node);
+        }
         m_valid_handle.insert(info);
         cmd.outputs.push_back(info);
         outputs->push_back(info);
@@ -220,7 +229,6 @@ SmallVector<Handle> ChannelImpl::apply_op(
             break;
         }
     }
-    mgb_assert(outputs.size() > 0, "Invalid dispatch mode!");
     return outputs;
 }