fix(mge/autodiff): fix segfault when grad is nullptr

GitOrigin-RevId: 6139212bfdc75ac7af5275436f5557ba487673e7

imperative/python/src/grad_override.cpp

@@ -59,6 +59,9 @@ apply_result_t elemwise_grad_rule(ApplyContext& ctx, CustomBackward::Maker& make
             mgb_assert(ngrads == 1);
             Tensor* grad = grads[0];
             apply_result_t ret(2);
+            if (!grad) {
+                return ret;
+            }
             for (size_t i = 0; i < 2; ++i) {
                 if (shapes[i]) {
                     ret[i] = reduce_to(grad, shapes[i].get());
@@ -84,6 +87,9 @@ apply_result_t reshape_grad_rule(ApplyContext& ctx, CustomBackward::Maker& maker
         mgb_assert(ngrads == 1);
         Tensor* grad = grads[0];
         apply_result_t ret(2);
+        if (!grad) {
+            return ret;
+        }
         for (size_t i = 0; i < 2; ++i) {
             if (shapes[i]) {
                 ret[i] = reshape_to(grad, shapes[i].get());
@@ -107,10 +113,10 @@ apply_result_t subtensor_grad_rule(ApplyContext& ctx, CustomBackward::Maker& mak
     maker.output_size(1).output_captured(0, false);
     maker.backward([inputs=std::move(inputs), grad_op_=std::move(grad_op)](BackwardContext&, Tensor*const* grads, size_t ngrads) {
         mgb_assert(ngrads == 1);
+        Tensor* grad = grads[0];
         apply_result_t ret(1);
-        if (inputs[0]) {
+        if (grad && inputs[0]) {
             SmallVector<Tensor*> args_(inputs.size()+1);
-            Tensor* grad = grads[0];
             auto&& zeros = make_tensor(grad->comp_node(), inputs[0].get());
             args_[0] = zeros.get();
             args_[1] = grad;
@@ -137,10 +143,10 @@ apply_result_t indexingMultiAxisVec_grad_rule(ApplyContext& ctx, CustomBackward:
     maker.output_size(1).output_captured(0, false);
     maker.backward([inputs=std::move(inputs), grad_op_=std::move(grad_op)](BackwardContext&, Tensor*const* grads, size_t ngrads) {
         mgb_assert(ngrads == 1);
+        Tensor* grad = grads[0];
         apply_result_t ret(1);
-        if (inputs[0]) {
+        if (grad && inputs[0]) {
             SmallVector<Tensor*> args_(inputs.size()+1);
-            Tensor* grad = grads[0];
             auto&& zeros = make_tensor(grad->comp_node(), inputs[0].get());
             args_[0] = zeros.get();
             args_[1] = grad;
@@ -167,7 +173,7 @@ apply_result_t reduce_grad_rule(ApplyContext& ctx, CustomBackward::Maker& maker)
             mgb_assert(ngrads == 1);
             Tensor* grad = grads[0];
             apply_result_t ret(1);
-            if (shapes[0]) {
+            if (grad && shapes[0]) {
                 ret[0] = broadcast_to(grad, shapes[0].get());
             }
             return ret;
@@ -180,14 +186,17 @@ apply_result_t reduce_grad_rule(ApplyContext& ctx, CustomBackward::Maker& maker)
 apply_result_t addAxis_grad_rule(ApplyContext& ctx, CustomBackward::Maker& maker) {
     auto&& op = ctx.op->cast_final_safe<AddAxis>();
     mgb_assert(ctx.nargs == 1);
+    bool flag = input_requires_grad(ctx, 0);
     auto&& grad_op = RemoveAxis::make(op.axis);
     std::sort(grad_op->axis.begin(), grad_op->axis.end(), std::greater<int32_t>());
     maker.output_size(1).output_captured(0, false);
-    maker.backward([grad_op_=std::move(grad_op)](BackwardContext&, Tensor*const* grads, size_t ngrads) {
+    maker.backward([grad_op_=std::move(grad_op), flag_=flag](BackwardContext&, Tensor*const* grads, size_t ngrads) {
         mgb_assert(ngrads == 1);
         Tensor* grad = grads[0];
         apply_result_t ret(1);
-        ret[0] = python::apply(grad_op_, grad)[0];
+        if (grad && flag_) {
+            ret[0] = python::apply(grad_op_, grad)[0];
+        }
         return ret;
     });
     return apply(ctx);
@@ -196,14 +205,17 @@ apply_result_t addAxis_grad_rule(ApplyContext& ctx, CustomBackward::Maker& maker
 apply_result_t removeAxis_grad_rule(ApplyContext& ctx, CustomBackward::Maker& maker) {
     auto&& op = ctx.op->cast_final_safe<RemoveAxis>();
     mgb_assert(ctx.nargs == 1);
+    bool flag = input_requires_grad(ctx, 0);
     auto&& grad_op = AddAxis::make(op.axis);
     std::sort(grad_op->axis.begin(), grad_op->axis.end());
     maker.output_size(1).output_captured(0, false);
-    maker.backward([grad_op_=std::move(grad_op)](BackwardContext&, Tensor*const* grads, size_t ngrads) {
+    maker.backward([grad_op_=std::move(grad_op), flag_=flag](BackwardContext&, Tensor*const* grads, size_t ngrads) {
         mgb_assert(ngrads == 1);
         Tensor* grad = grads[0];
         apply_result_t ret(1);
-        ret[0] = python::apply(grad_op_, grad)[0];
+        if (grad && flag_) {
+            ret[0] = python::apply(grad_op_, grad)[0];
+        }
         return ret;
     });
     return apply(ctx);