add complex support for optest (#53356)

* add complex support for  optest

* add complex grad test

* append one

* move some debug info

* move some debug info

* move some debug info

* move some debug info

* add more complex test

* Fix naming ambiguity

* Revert "add more complex test"

This reverts commit dbcb0516b8e53ba42e2d6089878a39b395345969.

* change backward gradient, add TODO

paddle/fluid/pybind/tensor_py.h

@@ -35,6 +35,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/math/concat_and_split.h"
 #include "paddle/fluid/platform/bfloat16.h"
 #include "paddle/fluid/platform/device/device_wrapper.h"
+#include "paddle/fluid/pybind/complex.h"
 #include "paddle/phi/kernels/funcs/strided_memcpy.h"
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 #include "paddle/fluid/platform/cuda_device_guard.h"

paddle/phi/kernels/cpu/reduce_mean_grad_kernel.cc

@@ -47,4 +47,6 @@ PD_REGISTER_KERNEL(mean_grad,
                    phi::ReduceMeanGradKernel,
                    bool,
                    float,
-                   double) {}
+                   double,
+                   phi::dtype::complex<float>,
+                   phi::dtype::complex<double>) {}

paddle/phi/kernels/cpu/reduce_mean_kernel.cc

@@ -36,5 +36,12 @@ void MeanRawKernel(const Context& dev_ctx,
 
 }  // namespace phi
 
-PD_REGISTER_KERNEL(
-    mean_raw, CPU, ALL_LAYOUT, phi::MeanRawKernel, float, double, bool) {}
+PD_REGISTER_KERNEL(mean_raw,
+                   CPU,
+                   ALL_LAYOUT,
+                   phi::MeanRawKernel,
+                   float,
+                   double,
+                   bool,
+                   phi::dtype::complex<float>,
+                   phi::dtype::complex<double>) {}

paddle/phi/kernels/gpu/reduce_mean_grad_kernel.cu

@@ -67,4 +67,6 @@ PD_REGISTER_KERNEL(mean_grad,
                    float,
                    double,
                    phi::dtype::float16,
-                   phi::dtype::bfloat16) {}
+                   phi::dtype::bfloat16,
+                   phi::dtype::complex<float>,
+                   phi::dtype::complex<double>) {}

paddle/phi/kernels/kps/reduce_mean_kernel.cu

@@ -48,5 +48,7 @@ PD_REGISTER_KERNEL(mean_raw,
                    phi::dtype::bfloat16,
                    float16,
                    int,
-                   int64_t) {}
+                   int64_t,
+                   phi::dtype::complex<float>,
+                   phi::dtype::complex<double>) {}
 #endif

paddle/phi/kernels/reduce_mean_kernel.cc

@@ -31,8 +31,15 @@ void MeanKernel(const Context& dev_ctx,
 
 }  // namespace phi
 
-PD_REGISTER_KERNEL(
-    mean, CPU, ALL_LAYOUT, phi::MeanKernel, float, double, bool) {}
+PD_REGISTER_KERNEL(mean,
+                   CPU,
+                   ALL_LAYOUT,
+                   phi::MeanKernel,
+                   float,
+                   double,
+                   bool,
+                   phi::dtype::complex<float>,
+                   phi::dtype::complex<double>) {}
 
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 PD_REGISTER_KERNEL(mean,
@@ -45,7 +52,9 @@ PD_REGISTER_KERNEL(mean,
                    int,
                    int64_t,
                    phi::dtype::float16,
-                   phi::dtype::bfloat16) {}
+                   phi::dtype::bfloat16,
+                   phi::dtype::complex<float>,
+                   phi::dtype::complex<double>) {}
 #endif
 
 #if defined(PADDLE_WITH_XPU_KP) && !defined(PADDLE_WITH_XPU)

python/paddle/fluid/tests/unittests/eager_op_test.py

@@ -200,6 +200,10 @@ def get_numeric_gradient(
             return tensor._get_float_element(i)
         elif tensor_to_check_dtype == np.float64:
             return tensor._get_double_element(i)
+        elif tensor_to_check_dtype == np.complex64:
+            return tensor._get_complex64_element(i)
+        elif tensor_to_check_dtype == np.complex128:
+            return tensor._get_complex128_element(i)
         else:
             raise TypeError(
                 "Unsupported test data type %s." % tensor_to_check_dtype
@@ -224,6 +228,10 @@ def get_numeric_gradient(
             tensor._set_float_element(i, e)
         elif tensor_to_check_dtype == np.float64:
             tensor._set_double_element(i, e)
+        elif tensor_to_check_dtype == np.complex64:
+            return tensor._set_complex64_element(i, e)
+        elif tensor_to_check_dtype == np.complex128:
+            return tensor._set_complex128_element(i, e)
         else:
             raise TypeError(
                 "Unsupported test data type %s." % tensor_to_check_dtype
@@ -242,6 +250,13 @@ def get_numeric_gradient(
         __set_elem__(tensor_to_check, i, x_pos)
         y_pos = get_output()
 
+        if tensor_to_check_dtype in [np.complex64, np.complex128]:
+            if in_place:
+                set_input(scope, op, inputs, place)
+            x_pos_j = origin + 1j * delta
+            __set_elem__(tensor_to_check, i, x_pos_j)
+            y_pos_j = get_output()
+
         if in_place:
             set_input(scope, op, inputs, place)
 
@@ -249,8 +264,44 @@ def get_numeric_gradient(
         __set_elem__(tensor_to_check, i, x_neg)
         y_neg = get_output()
 
+        if tensor_to_check_dtype in [np.complex64, np.complex128]:
+            if in_place:
+                set_input(scope, op, inputs, place)
+
+            x_neg_j = origin - 1j * delta
+            __set_elem__(tensor_to_check, i, x_neg_j)
+            y_neg_j = get_output()
+
+        __set_elem__(tensor_to_check, i, origin)
+
+        if tensor_to_check_dtype in [np.complex64, np.complex128]:
+            # always assume real output, because this function has
+            # no input for dl/di, though it should do. so there di will be zero
+
+            # TODO: Here is a trick to be consistent with the existing OpTest, it
+            # need to support variable gradients input
+            f_ajoint = np.array(1 + 0j)
+            df_over_dr = (y_pos - y_neg) / delta / 2
+            df_over_di = (y_pos_j - y_neg_j) / delta / 2
+
+            dl_over_du, dl_over_dv = f_ajoint.real, f_ajoint.imag
+
+            du_over_dr, dv_over_dr = df_over_dr.real, df_over_dr.imag
+
+            du_over_di, dv_over_di = df_over_di.real, df_over_di.imag
+
+            dl_over_dr = np.sum(
+                dl_over_du * du_over_dr + dl_over_dv * dv_over_dr
+            )
+            dl_over_di = np.sum(
+                dl_over_du * du_over_di + dl_over_dv * dv_over_di
+            )
+            gradient_flat[i] = dl_over_dr + 1j * dl_over_di
+        else:
+            df_over_dr = y_pos - y_neg
+            gradient_flat[i] = df_over_dr / delta / 2
+
         __set_elem__(tensor_to_check, i, origin)
-        gradient_flat[i] = (y_pos - y_neg) / delta / 2
 
     return gradient_flat.reshape(tensor_to_check.shape())
 
@@ -375,6 +426,13 @@ class OpTest(unittest.TestCase):
         def is_custom_device_op_test():
             return hasattr(cls, "use_custom_device") and cls.use_custom_device
 
+        def is_complex_test():
+            return (
+                hasattr(cls, "test_complex")
+                and cls.test_complex
+                or (cls.dtype in [np.complex64, np.complex128])
+            )
+
         if not hasattr(cls, "op_type"):
             raise AssertionError(
                 "This test do not have op_type in class attrs, "
@@ -382,8 +440,10 @@ class OpTest(unittest.TestCase):
             )
 
         # case in NO_FP64_CHECK_GRAD_CASES and op in NO_FP64_CHECK_GRAD_OP_LIST should be fixed
-        if not hasattr(cls, "no_need_check_grad") and not is_empty_grad_op(
-            cls.op_type
+        if (
+            not hasattr(cls, "no_need_check_grad")
+            and not is_empty_grad_op(cls.op_type)
+            and not is_complex_test()
         ):
             if cls.dtype is None or (
                 cls.dtype == np.float16
@@ -2496,7 +2556,6 @@ class OpTest(unittest.TestCase):
             max_relative_error = (
                 0.001 if max_relative_error < 0.001 else max_relative_error
             )
-
         self._assert_is_close(
             numeric_grads,
             analytic_grads,

python/paddle/fluid/tests/unittests/test_elementwise_mul_op.py

@@ -93,6 +93,55 @@ class ElementwiseMulOp(OpTest):
         pass
 
 
+class TestComplexElementwiseMulOpWithCheckGrad(ElementwiseMulOp):
+    def setUp(self):
+        self.op_type = "elementwise_mul"
+        self.python_api = paddle.multiply
+        self.public_python_api = paddle.multiply
+        self.dtype = np.complex128
+        self.axis = -1
+        self.init_dtype()
+        self.init_input_output()
+        self.init_kernel_type()
+        self.init_axis()
+        self.if_enable_cinn()
+
+        self.inputs = {
+            'X': OpTest.np_dtype_to_fluid_dtype(self.x),
+            'Y': OpTest.np_dtype_to_fluid_dtype(self.y),
+        }
+        self.outputs = {'Out': self.out}
+        self.attrs = {'axis': self.axis}
+
+    def init_input_output(self):
+        self.x = np.array([3 + 4j, 1 + 2j]).astype(self.dtype)
+        self.y = np.array([3 + 4j, 5 + 6j]).astype(self.dtype)
+        self.out = np.multiply(self.x, self.y)
+
+    def if_enable_cinn(self):
+        self.enable_cinn = False
+
+    def test_check_grad_normal(self):
+        self.check_grad(
+            ['X', 'Y'],
+            'Out',
+        )
+
+    def test_check_grad_ingore_x(self):
+        self.check_grad(
+            ['Y'],
+            'Out',
+            no_grad_set=set("X"),
+        )
+
+    def test_check_grad_ingore_y(self):
+        self.check_grad(
+            ['X'],
+            'Out',
+            no_grad_set=set('Y'),
+        )
+
+
 class TestElementwiseMulOp_ZeroDim1(ElementwiseMulOp):
     def init_input_output(self):
         self.x = np.random.uniform(0.1, 1, []).astype(self.dtype)

python/paddle/fluid/tests/unittests/test_reduce_op.py

@@ -57,6 +57,20 @@ class TestSumOp(OpTest):
         self.check_grad(['X'], 'Out', check_prim=True)
 
 
+class TestComplexSumOP(TestSumOp):
+    def init_dtype(self):
+        self.dtype = np.complex128
+
+    def init_input(self):
+        self.x = np.random.random((3, 4)).astype(self.dtype)
+
+    def init_attrs(self):
+        self.attrs = {'dim': [0]}
+
+    def test_check_grad(self):
+        self.check_grad(['X'], 'Out', check_prim=False)
+
+
 class TestSumOp_ZeroDim(TestSumOp):
     def init_attrs(self):
         self.attrs = {'dim': [], 'reduce_all': True}