[autodiff] Print more specific error message that autodiff does not s…

…upport to_numpy (#5630)

taichi/transforms/auto_diff.cpp

@@ -989,6 +989,11 @@ class MakeAdjoint : public ADTransform {
 
   void visit(GlobalLoadStmt *stmt) override {
     // issue global store to adjoint
+    if (stmt->src->is<ExternalPtrStmt>()) {
+      TI_ERROR(
+          "Importing data from external array (such as numpy array) not "
+          "supported in AutoDiff for now")
+    }
     GlobalPtrStmt *src = stmt->src->as<GlobalPtrStmt>();
     TI_ASSERT(src->width() == 1);
     auto snodes = src->snodes;
@@ -1008,6 +1013,11 @@ class MakeAdjoint : public ADTransform {
 
   void visit(GlobalStoreStmt *stmt) override {
     // erase and replace with global load adjoint
+    if (stmt->dest->is<ExternalPtrStmt>()) {
+      TI_ERROR(
+          "Exporting data to external array (such as numpy array) not "
+          "supported in AutoDiff for now")
+    }
     GlobalPtrStmt *dest = stmt->dest->as<GlobalPtrStmt>();
     TI_ASSERT(dest->width() == 1);
     auto snodes = dest->snodes;

tests/python/test_ad_external_array.py

@@ -0,0 +1,35 @@
+import numpy as np
+import pytest
+
+import taichi as ti
+from tests import test_utils
+
+
+@test_utils.test()
+def test_to_numpy():
+    a = ti.field(dtype=float, shape=(), needs_grad=True)
+    loss = ti.field(dtype=float, shape=(), needs_grad=True)
+
+    def func():
+        b = a.to_numpy()
+
+    with pytest.raises(RuntimeError) as e:
+        with ti.ad.Tape(loss):
+            func()
+    assert 'Exporting data to external array (such as numpy array) not supported in AutoDiff for now' in e.value.args[
+        0]
+
+
+@test_utils.test()
+def test_from_numpy():
+    a = ti.field(dtype=float, shape=(), needs_grad=True)
+    loss = ti.field(dtype=float, shape=(), needs_grad=True)
+
+    def func():
+        a.from_numpy(np.asarray(1))
+
+    with pytest.raises(RuntimeError) as e:
+        with ti.ad.Tape(loss):
+            func()
+    assert 'Importing data from external array (such as numpy array) not supported in AutoDiff for now' in e.value.args[
+        0]