[HIP] fix op not support AMD GPU bug, the flag PADDLE_WITH_ROCM is in… (#35394)

* [HIP] fix op not support AMD GPU bug, the flag PADDLE_WITH_ROCM is invalid

* [HIP] fix op not support AMD GPU bug, the flag PADDLE_WITH_ROCM is invalid

* [HIP] fix op not support AMD GPU bug

paddle/fluid/operators/memcpy_d2h_op.cc

@@ -131,7 +131,7 @@ REGISTER_OP_CPU_KERNEL_FUNCTOR(memcpy_d2h, float, ops::MemcpyD2HKernel, double,
                                ops::MemcpyD2HKernel, plat::float16,
                                ops::MemcpyD2HKernel);
 
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_ROCM)
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 REGISTER_OP_CUDA_KERNEL_FUNCTOR(memcpy_d2h, float, ops::MemcpyD2HKernel, double,
                                 ops::MemcpyD2HKernel, int, ops::MemcpyD2HKernel,
                                 int64_t, ops::MemcpyD2HKernel, bool,

paddle/fluid/operators/memcpy_h2d_op.cc

@@ -131,7 +131,7 @@ REGISTER_OP_CPU_KERNEL_FUNCTOR(memcpy_h2d, float, ops::MemcpyH2DKernel, double,
                                ops::MemcpyH2DKernel, plat::float16,
                                ops::MemcpyH2DKernel);
 
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_ROCM)
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 REGISTER_OP_CUDA_KERNEL_FUNCTOR(memcpy_h2d, float, ops::MemcpyH2DKernel, double,
                                 ops::MemcpyH2DKernel, int, ops::MemcpyH2DKernel,
                                 int64_t, ops::MemcpyH2DKernel, bool,

paddle/fluid/operators/memcpy_op.cc

@@ -141,7 +141,7 @@ REGISTER_OP_CPU_KERNEL_FUNCTOR(memcpy, float, ops::MemcpyKernel, double,
                                ops::MemcpyKernel, plat::float16,
                                ops::MemcpyKernel);
 
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_ROCM)
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 REGISTER_OP_CUDA_KERNEL_FUNCTOR(memcpy, float, ops::MemcpyKernel, double,
                                 ops::MemcpyKernel, int, ops::MemcpyKernel,
                                 int64_t, ops::MemcpyKernel, bool,

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

@@ -64,7 +64,7 @@ class TestMemcpy_FillConstant(unittest.TestCase):
                 })
         return main_program, gpu_var, pinned_var
 
-    def test_gpu_cpoy_to_pinned(self):
+    def test_gpu_copy_to_pinned(self):
         main_program, gpu_var, pinned_var = self.get_prog()
         main_program.global_block().append_op(
             type='memcpy',
@@ -79,7 +79,7 @@ class TestMemcpy_FillConstant(unittest.TestCase):
         self.assertTrue(np.allclose(gpu_, pinned_))
         self.assertTrue(np.allclose(pinned_, np.ones((10, 10))))
 
-    def test_pinned_cpoy_gpu(self):
+    def test_pinned_copy_gpu(self):
         main_program, gpu_var, pinned_var = self.get_prog()
         main_program.global_block().append_op(
             type='memcpy',
@@ -94,6 +94,59 @@ class TestMemcpy_FillConstant(unittest.TestCase):
         self.assertTrue(np.allclose(gpu_, pinned_))
         self.assertTrue(np.allclose(gpu_, np.zeros((10, 10))))
 
+    def test_hip_copy_bool_value(self):
+        if core.is_compiled_with_rocm():
+            paddle.enable_static()
+            main_program = Program()
+            with program_guard(main_program):
+                pinned_var_name = "tensor@Pinned"
+                gpu_var_name = "tensor@GPU"
+                pinned_var = main_program.global_block().create_var(
+                    name=pinned_var_name,
+                    shape=[1],
+                    dtype='bool',
+                    persistable=False,
+                    stop_gradient=True)
+                gpu_var = main_program.global_block().create_var(
+                    name=gpu_var_name,
+                    shape=[1],
+                    dtype='bool',
+                    persistable=False,
+                    stop_gradient=True)
+                main_program.global_block().append_op(
+                    type="fill_constant",
+                    outputs={"Out": gpu_var_name},
+                    attrs={
+                        "shape": [1],
+                        "dtype": gpu_var.dtype,
+                        "value": False,
+                        "place_type": 1
+                    })
+                main_program.global_block().append_op(
+                    type="fill_constant",
+                    outputs={"Out": pinned_var_name},
+                    attrs={
+                        "shape": [1],
+                        "dtype": gpu_var.dtype,
+                        "value": True,
+                        "place_type": 2
+                    })
+
+            main_program.global_block().append_op(
+                type='memcpy',
+                inputs={'X': pinned_var},
+                outputs={'Out': gpu_var},
+                attrs={'dst_place_type': 1})
+            place = fluid.CUDAPlace(0)
+            exe = fluid.Executor(place)
+            gpu_, pinned_ = exe.run(main_program,
+                                    feed={},
+                                    fetch_list=[gpu_var.name, pinned_var.name])
+            expect_value = np.array([1]).astype('bool')
+            self.assertTrue(np.array_equal(gpu_, expect_value))
+        else:
+            pass
+
 
 class TestMemcpyOPError(unittest.TestCase):
     def get_prog(self):