[Hexagon] Create tests to showcase vtcm loading capabilities on Hexagon.

python/tvm/contrib/hexagon/session.py

@@ -58,7 +58,7 @@ def __init__(
         remote_kw: dict,
         session_name: str = "hexagon-rpc",
         remote_stack_size_bytes: int = 256 * 1024,  # Min size for main thread in QuRT/sim
-        rpc_receive_buffer_size_bytes: int = 5 * 1024 * 1024,  # Size for passing hexagon tests
+        rpc_receive_buffer_size_bytes: int = 256 * 1024 * 1024,  # Size for passing hexagon tests
     ):
         self._launcher = launcher
         self._session_name: str = session_name

src/runtime/hexagon/hexagon_buffer.cc

@@ -62,7 +62,7 @@ struct VTCMAllocation : public Allocation {
 
     // allocate nbytes of vtcm on a single page
     HEXAGON_SAFE_CALL(HAP_compute_res_attr_set_vtcm_param(&res_info, /*vtcm_size = */ nbytes,
-                                                          /*b_single_page = */ 1));
+                                                          /*b_single_page = */ 0));
 
     // TODO(HWE): Investigate why a non-zero timeout results in
     // hanging, both in the simulator and on hardware.
@@ -71,13 +71,14 @@ struct VTCMAllocation : public Allocation {
     if (context_id_) {
       data_ = HAP_compute_res_attr_get_vtcm_ptr(&res_info);
       if (!data_) {
-        LOG(ERROR) << "ERROR: Allocated VTCM ptr is null.";
+        LOG(ERROR) << "ERROR: HAP_compute_res_acquire returned nullptr when allocating VTCM.";
         HEXAGON_SAFE_CALL(HAP_compute_res_release(context_id_));
         return;
       }
     } else {
-      LOG(ERROR) << "ERROR: Unable to acquire requeisted resource.";
-      return;
+      LOG(FATAL) << "FATAL: HAP_compute_res_acquire failed to acquire requested VTCM resource.";
+      throw std::runtime_error(
+          "HAP_compute_res_acquire failed to acquire requested VTCM resource.");
     }
   }
   ~VTCMAllocation() {

src/runtime/hexagon/hexagon_device_api.cc

@@ -165,6 +165,17 @@ void HexagonDeviceAPI::CopyDataFromTo(const void* from, size_t from_offset, void
   memcpy(static_cast<char*>(to) + to_offset, static_cast<const char*>(from) + from_offset, size);
 }
 
+TVM_REGISTER_GLOBAL("device_api.hexagon.mem_copy_DLTensor")
+    .set_body([](TVMArgs args, TVMRetValue* rv) {
+      DLTensor* dst = args[0];
+      DLTensor* src = args[1];
+      int size = args[2];
+
+      hexagon_user_dma_1d_sync(dst->data, src->data, size);
+
+      *rv = static_cast<int32_t>(0);
+    });
+
 TVM_REGISTER_GLOBAL("device_api.hexagon.mem_copy").set_body([](TVMArgs args, TVMRetValue* rv) {
   void* dst = args[0];
   void* src = args[1];