[auto parallel] fix reshape infer shape (#60632)

FeixLiu · web-flow · commit 3324c9d5e7d2 · 2024-01-10T08:47:26.000+08:00
diff --git a/paddle/phi/api/yaml/generator/dist_api_gen.py b/paddle/phi/api/yaml/generator/dist_api_gen.py
@@ -454,21 +454,47 @@
 # TODO(GhostScreaming): Support aliquant condition.
 # Specialized Code, for example, reshape needs to calculate local_shape
 RESHAPE_CALCULATE_LOCAL_SHAPE_TEMPLATE = """
+
+      // The dist_input_x is a dist tensor, the dims() func return the global dims.
+      auto x_shape = dist_input_x->dims();
+      auto x_numel = dist_input_x->numel();
+      bool visit_negative = false;
       std::vector<int64_t> local_shape;
       for (size_t i = 0; i < shape.GetData().size(); i++) {
         auto& out_dist_attr = PADDLE_GET_CONST(phi::distributed::TensorDistAttr, spmd_info.second[0]);
         if (out_dist_attr.dims_mapping()[i] >= 0) {
+          int64_t shape_i = shape.GetData()[i];
+          if (shape_i == 0) {
+            shape_i = x_shape[i];
+          } else if (shape_i == -1) {
+            PADDLE_ENFORCE(not visit_negative,
+                           phi::errors::InvalidArgument(
+                               "Reshape can only have one -1 in the shape."));
+            visit_negative = true;
+            int64_t non_negative_product = 1;
+            for (size_t j = 0; j < shape.GetData().size(); j++) {
+              if (i == j) {
+                continue;
+              }
+              int64_t tmp_j = shape.GetData()[j];
+              if (tmp_j == 0) {
+                tmp_j = x_shape[j];
+              }
+              non_negative_product *= tmp_j;
+            }
+            PADDLE_ENFORCE(x_numel % non_negative_product == 0,
+                           phi::errors::InvalidArgument("Cannot infer real shape for -1."));
+            shape_i = x_numel / non_negative_product;
+          }
           int64_t dim = out_dist_attr.dims_mapping()[i];
           int64_t mesh_dim = out_dist_attr.process_mesh().shape()[dim];
           // TODO: Support aliquant condition.
-          PADDLE_ENFORCE_EQ(shape.GetData()[i] % mesh_dim,
-                0,
+          PADDLE_ENFORCE(shape_i % mesh_dim == 0,
                 phi::errors::InvalidArgument(
-                    "Reshape only support local shape dim is divisible"
-                    "by the mesh dim, however local_shape[%d] is %d",
-                    "and shard mesh dims is %d",
-                    i, shape.GetData()[i], mesh_dim));
-          local_shape.push_back(shape.GetData()[i] / mesh_dim);
+                    "Reshape only support local shape dim is divisible "
+                    "by the mesh dim, however local_shape[%lld] is %lld "
+                    "and shard mesh dims is %lld.", i, shape_i, mesh_dim));
+          local_shape.push_back(shape_i / mesh_dim);
         } else {
           local_shape.push_back(shape.GetData()[i]);
         }
diff --git a/test/auto_parallel/semi_auto_parallel_for_reshape.py b/test/auto_parallel/semi_auto_parallel_for_reshape.py
@@ -47,6 +47,14 @@ def test_reshape_forward(self):
         self.check_placements(output, [dist.Shard(0)])
         self.check_placements(input.grad, [dist.Shard(0)])
 
+    def test_reshape_infer_shape(self):
+        mesh = dist.ProcessMesh([0, 1], dim_names=["x"])
+        x = paddle.ones([10, 20, 30])
+        x = dist.shard_tensor(x, mesh, [Shard(0)])
+        y = x.reshape([-1, 0, x.shape[0]])
+        assert y.shape == [30, 20, 10]
+        assert y._local_shape == [15, 20, 10]
+
     def run_test_case(self):
         if self._backend == "cpu":
             paddle.set_device("cpu")
@@ -55,6 +63,7 @@ def run_test_case(self):
         else:
             raise ValueError("Only support cpu or gpu backend.")
         self.test_reshape_forward()
+        self.test_reshape_infer_shape()
 
 
 if __name__ == '__main__':
