[0D-Tensor] CINN supports `fill_constant`, fix infershape and pass (#55563)

* [0D-Tensor] CINN supports fill_constant, fix infershape and pass

* fix infershape of fill_constant

* add back fill_constant to zero_tensor_trick_pass

paddle/cinn/frontend/pass/expand_zero_dim_pass.cc

@@ -39,6 +39,9 @@ class ExpandZeroDimPass : public ProgramPass {
       if (instr->op_type == "transpose") {
         builder.AppendInstruction(HandleTranspose(instr));
         continue;
+      } else if (instr->op_type == "fill_constant") {
+        builder.AppendInstruction(HandleFillConstant(instr));
+        continue;
       }
       for (auto& input : instr->inputs) {
         if (input->shape.empty()) {
@@ -101,6 +104,20 @@ class ExpandZeroDimPass : public ProgramPass {
     }
     return new_instr;
   }
+
+  // Before: out-0D = fill_constant([], 123.456, "out", "float32")
+  // After:  out-1D = fill_constant([1], 123.456, "out", "float32")
+  Instruction HandleFillConstant(const Instruction& instr) {
+    Instruction new_instr = instr;
+    std::vector<int32_t> shape =
+        new_instr.GetAttrs<std::vector<int32_t>>("shape");
+    if (shape.empty()) {
+      shape.push_back(1);
+      VLOG(4) << "Change fill_constant's attribute shape from [] to [1]";
+    }
+    new_instr.SetAttr<std::vector<int32_t>>("shape", shape);
+    return new_instr;
+  }
 };
 
 }  // namespace pass

paddle/cinn/hlir/op/elementwise.cc

@@ -393,7 +393,6 @@ std::vector<shape_t> InferShapeForFillConstant(
     const framework::AttrMapType &attrs) {
   CHECK(attrs.count("shape"));
   auto shape = absl::get<std::vector<int>>(attrs.at("shape"));
-  CHECK(!shape.empty()) << "shape attr is empty!";
   return {shape};
 }
 

test/cinn/ops/test_zero_dim_tensor.py

@@ -1318,6 +1318,33 @@ class TestReshapeOp1DTo0D(TestReshapeOp):
         self.target_shape = []
 
 
+@OpTestTool.skip_if(
+    not is_compiled_with_cuda(), "x86 test will be skipped due to timeout."
+)
+class TestFillConstantOp(OpTest):
+    def setUp(self):
+        np.random.seed(2023)
+        self.target_shape = ()
+
+    def build_paddle_program(self, target):
+        out = paddle.full([], 123.456, "float32")
+
+        self.paddle_outputs = [out]
+
+    def build_cinn_program(self, target):
+        builder = NetBuilder("fill_constant_op")
+        out = builder.fill_constant([], 123.456, "out", "float32")
+
+        prog = builder.build()
+        res = self.get_cinn_output(prog, target, [], [], [out])
+
+        self.cinn_outputs = res
+        self.assertEqual(res[0].shape, self.target_shape)
+
+    def test_check_results(self):
+        self.check_outputs_and_grads()
+
+
 @OpTestTool.skip_if(
     not is_compiled_with_cuda(), "x86 test will be skipped due to timeout."
 )