[XPU] copy index to xpu inside index_select; test=develop

Author: newway

paddle/phi/api/yaml/ops.yaml

@@ -1263,6 +1263,8 @@
     func : index_select
     data_type : x
   backward : index_select_grad
+  data_transform :
+    skip_transform : index
 
 - op : index_select_strided
   args : (Tensor x, int64_t index, int axis = 0)

paddle/phi/backends/xpu/xpu1_op_list.cc

@@ -361,7 +361,7 @@ XPUOpMap& get_kl1_ops() {
       {"where_index", XPUKernelSet({phi::DataType::BOOL})},
       {"where",
        XPUKernelSet({phi::DataType::INT32,
-                     // phi::DataType::INT64,
+                     phi::DataType::INT64,
                      phi::DataType::FLOAT32})},
       // AddMore
   };

paddle/phi/kernels/xpu/index_select_kernel.cc

@@ -13,8 +13,8 @@
 // limitations under the License.
 
 #include "paddle/phi/kernels/index_select_kernel.h"
-
 #include "paddle/phi/backends/xpu/enforce_xpu.h"
+#include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/utils/data_type.h"
 
@@ -46,8 +46,23 @@ void IndexSelectKernel(const Context& ctx,
   int index_len = output->dims()[dim];
   T* out_data = ctx.template Alloc<T>(output);
   int r = 0;
+  xpu::ctx_guard RAII_GUARD(ctx.x_context());
+  const int8_t* index_ptr = nullptr;
+  int byte_times = sizeof(index_type);
+  if (index.place() == CPUPlace()) {
+    index_ptr = RAII_GUARD.alloc_l3_or_gm<int8_t>(byte_times * index.numel());
+    PADDLE_ENFORCE_XDNN_NOT_NULL(index_ptr);
+    memory_utils::Copy(ctx.GetPlace(),
+                       reinterpret_cast<void*>(const_cast<int8_t*>(index_ptr)),
+                       CPUPlace(),
+                       reinterpret_cast<const void*>(index.data<int>()),
+                       byte_times * index.numel());
+  } else {
+    index_ptr = index.template data<int8_t>();
+  }
   if (index_type == phi::DataType::INT64) {
-    const int64_t* index_data = index.data<int64_t>();
+    const int64_t* index_data =
+        reinterpret_cast<const int64_t*>(const_cast<int8_t*>(index_ptr));
     r = xpu::gather<T, int64_t>(ctx.x_context(),
                                 in_data,
                                 index_data,
@@ -56,7 +71,8 @@ void IndexSelectKernel(const Context& ctx,
                                 index_len,
                                 dim);
   } else {
-    const int* index_data = index.data<int>();
+    const int* index_data =
+        reinterpret_cast<const int*>(const_cast<int8_t*>(index_ptr));
     r = xpu::gather<T, int>(ctx.x_context(),
                             in_data,
                             index_data,