【PIR api adaptor No.242、228】 Migrate unique_consecutive/moveaxis into pir (PaddlePaddle#58688)

Author: DrRyanHuang

paddle/phi/api/yaml/ops.yaml

@@ -2683,7 +2683,7 @@
   backward: uniform_inplace_grad
 
 - op : unique_consecutive
-  args : (Tensor x, bool return_inverse = false, bool return_counts = false, int[] axis = {}, int dtype = 5)
+  args : (Tensor x, bool return_inverse = false, bool return_counts = false, int[] axis = {}, DataType dtype = DataType::FLOAT32)
   output : Tensor(out), Tensor(index), Tensor(counts)
   infer_meta :
       func : UniqueConsecutiveInferMeta

paddle/phi/infermeta/unary.cc

@@ -4852,7 +4852,7 @@ void UniqueConsecutiveInferMeta(const MetaTensor& x,
                                 bool return_inverse,
                                 bool return_counts,
                                 const std::vector<int>& axis,
-                                int dtype,
+                                DataType dtype,
                                 MetaTensor* out,
                                 MetaTensor* index,
                                 MetaTensor* counts) {

paddle/phi/infermeta/unary.h

@@ -716,7 +716,7 @@ void UniqueConsecutiveInferMeta(const MetaTensor& x,
                                 bool return_inverse,
                                 bool return_counts,
                                 const std::vector<int>& axis,
-                                int dtype,
+                                DataType dtype,
                                 MetaTensor* out,
                                 MetaTensor* index,
                                 MetaTensor* counts);

paddle/phi/kernels/cpu/unique_consecutive_kernel.cc

@@ -30,12 +30,11 @@ void UniqueConsecutiveKernel(const Context& dev_ctx,
                              bool return_inverse,
                              bool return_counts,
                              const std::vector<int>& axis,
-                             int dtype,
+                             DataType dtype,
                              DenseTensor* out,
                              DenseTensor* index,
                              DenseTensor* counts) {
-  auto data_type = phi::TransToPhiDataType(dtype);
-  if (data_type == phi::DataType::INT32) {
+  if (dtype == phi::DataType::INT32) {
     PADDLE_ENFORCE_LE(
         x.numel(),
         INT_MAX,
@@ -48,14 +47,14 @@ void UniqueConsecutiveKernel(const Context& dev_ctx,
 
   if (axis.empty()) {
     phi::VisitDataTypeTiny(
-        data_type,
+        dtype,
         UniqueConsecutiveFlattenedTensorFunctor<Context, T>(
             dev_ctx, x, out, return_inverse, return_counts, index, counts));
   } else {
     int valid_axis = axis[0];
     if (valid_axis < 0) valid_axis += x.dims().size();
     phi::VisitDataTypeTiny(
-        data_type,
+        dtype,
         UniqueConsecutiveDimFunctor<Context, T>(dev_ctx,
                                                 x,
                                                 out,

paddle/phi/kernels/gpu/unique_consecutive_kernel.cu

@@ -29,12 +29,11 @@ void UniqueConsecutiveKernel(const Context& dev_ctx,
                              bool return_inverse,
                              bool return_counts,
                              const std::vector<int>& axis,
-                             int dtype,
+                             DataType dtype,
                              DenseTensor* out,
                              DenseTensor* index,
                              DenseTensor* counts) {
-  auto data_type = phi::TransToPhiDataType(dtype);
-  if (data_type == phi::DataType::INT32) {
+  if (dtype == phi::DataType::INT32) {
     PADDLE_ENFORCE_LE(
         x.numel() + 1,
         INT_MAX,
@@ -48,15 +47,15 @@ void UniqueConsecutiveKernel(const Context& dev_ctx,
   // if 'axis' is not required, flatten the Tensor.
   if (axis.empty()) {
     phi::VisitDataTypeTiny(
-        data_type,
+        dtype,
         UniqueConsecutiveFlattenedCUDAFunctor<Context, T>(
             dev_ctx, x, out, return_inverse, return_counts, index, counts));
   } else {
     // 'axis' is required.
     int valid_axis = axis[0];
     if (valid_axis < 0) valid_axis += x.dims().size();
     phi::VisitDataTypeTiny(
-        data_type,
+        dtype,
         UniqueConsecutiveDimsCUDAFunctor<Context, T>(dev_ctx,
                                                      x,
                                                      out,

paddle/phi/kernels/unique_consecutive_kernel.h

@@ -26,7 +26,7 @@ void UniqueConsecutiveKernel(const Context& dev_ctx,
                              bool return_inverse,
                              bool return_counts,
                              const std::vector<int>& axis,
-                             int dtype,
+                             DataType dtype,
                              DenseTensor* out,
                              DenseTensor* index,
                              DenseTensor* counts);

python/paddle/tensor/manipulation.py

@@ -2482,7 +2482,7 @@ def unique_consecutive(
     else:
         axis = [axis]
     attr_dtype = convert_np_dtype_to_dtype_(dtype)
-    if in_dynamic_mode():
+    if in_dynamic_or_pir_mode():
         out, inverse, counts = _C_ops.unique_consecutive(
             x, return_inverse, return_counts, axis, attr_dtype
         )

test/legacy_test/test_transpose_op.py

@@ -710,6 +710,7 @@ def test_moveaxis3(self):
         self.assertEqual(out.shape, [2, 3])
         paddle.enable_static()
 
+    @test_with_pir_api
     def test_error(self):
         x = paddle.randn([2, 3, 4, 5])
         # src must have the same number with dst

test/legacy_test/test_unique_consecutive_op.py

@@ -20,6 +20,7 @@
 import paddle
 from paddle import base
 from paddle.base import core
+from paddle.pir_utils import test_with_pir_api
 
 
 def reference_unique_consecutive(
@@ -203,6 +204,7 @@ def setUp(self):
         if core.is_compiled_with_cuda():
             self.places.append(base.CUDAPlace(0))
 
+    @test_with_pir_api
     def check_static_result(self, place):
         with base.program_guard(base.Program(), base.Program()):
             paddle.enable_static()
@@ -217,7 +219,6 @@ def check_static_result(self, place):
             x_np = np.random.randint(20, size=100).astype("float32")
             exe = base.Executor(place)
             fetches = exe.run(
-                base.default_main_program(),
                 feed={"input_x": x_np},
                 fetch_list=[result],
             )
@@ -240,6 +241,7 @@ def setUp(self):
         if core.is_compiled_with_cuda():
             self.places.append(base.CUDAPlace(0))
 
+    @test_with_pir_api
     def check_static_result(self, place):
         with base.program_guard(base.Program(), base.Program()):
             paddle.enable_static()
@@ -256,7 +258,6 @@ def check_static_result(self, place):
             x_np = np.random.randint(20, size=100).astype("float32")
             exe = base.Executor(place)
             fetches = exe.run(
-                base.default_main_program(),
                 feed={"input_x": x_np},
                 fetch_list=[result],
             )
@@ -281,6 +282,7 @@ def setUp(self):
         if core.is_compiled_with_cuda():
             self.places.append(base.CUDAPlace(0))
 
+    @test_with_pir_api
     def check_static_result(self, place):
         with base.program_guard(base.Program(), base.Program()):
             paddle.enable_static()
@@ -297,7 +299,6 @@ def check_static_result(self, place):
             x_np = np.random.randint(20, size=100).astype("float32")
             exe = base.Executor(place)
             fetches = exe.run(
-                base.default_main_program(),
                 feed={"input_x": x_np},
                 fetch_list=[result],
             )
@@ -347,7 +348,7 @@ def setUp(self):
         }
 
     def test_check_output(self):
-        self.check_output()
+        self.check_output(check_pir=True)
 
 
 if __name__ == "__main__":