polish reindex memory optimize

Author: cjld

python/jittor/__init__.py

@@ -9,7 +9,7 @@
 # file 'LICENSE.txt', which is part of this source code package.
 # ***************************************************************
 
-__version__ = '1.3.3.8'
+__version__ = '1.3.3.9'
 from jittor_utils import lock
 with lock.lock_scope():
     ori_int = int

python/jittor/src/executor.cc

@@ -547,7 +547,6 @@ void Executor::run_sync(vector<Var*> vars, bool device_sync, bool weak_sync) {
             root = fuse_ops[rr-1];
             load_fused_op(fused_op, fuse_ops, ops, ll, rr, tt);
         }
-        LOGvvv << "Run" << op;
         for (auto* var : op->outputs()) {
             var->alloc(allocator);
         }

python/jittor/src/ops/reindex_op.cc

@@ -28,6 +28,7 @@ ReindexOp::ReindexOp(Var* x, NanoVector shape, vector<string>&& indexes, float64
     flags.set(NodeFlags::_cuda);
     set_type(OpType::broadcast);
     flags.set(NodeFlags::_manual_set_vnbb);
+    for (auto& v : extras) v->flags.set(NodeFlags::_needed_by_backward);
     y = create_output(nullptr, x->dtype());
 }
 
@@ -64,6 +65,7 @@ ReindexOp::ReindexOp(Var* x, vector<Var*>&& indexes, float64 overflow_value, vec
     extras = indexes;
     for (uint i = 0; i < indexes.size(); ++i) {
         indexes[i]->flags.set(NodeFlags::_force_fuse);
+        indexes[i]->flags.set(NodeFlags::_needed_by_backward);
     }
 }
 

python/jittor/src/ops/reindex_reduce_op.cc

@@ -37,6 +37,8 @@ ReindexReduceOp::ReindexReduceOp(Var* y, NanoString op, NanoVector shape, vector
         if (e->shape != y->shape) {
             e->flags.set(NodeFlags::_stop_fuse);
         }
+        if (op.get(NanoString::_no_need_back_in))
+            e->flags.set(NodeFlags::_needed_by_backward);
     }
 }
 

python/jittor/test/test_reindex_op.py

@@ -306,6 +306,16 @@ def test_reindex_wrong_op(self):
         b = jt.array([1])
         c = a.reindex([8,8], ["@e0(0) // 1", "@e0(0)"], extras=[b, b])
         expect_error(lambda: c.sync())
+
+    def test_reindex_memopt(self):
+        a = jt.zeros([10,10])
+        b = jt.array([1,2,3]).name("b")
+        c = a.reindex([8,8], ["@e0(0) / 1", "@e0(0)"], extras=[b, b])
+        del b
+        c.sync()
+        da = jt.grad(c, a)
+        da.sync()
+
 
 
 @unittest.skipIf(not jt.compiler.has_cuda, "No CUDA found")