copy beta pow to same place when skip_update=1 (#37245)

* copy beta pow to same place when skip_update=1

* fix xpu

paddle/fluid/operators/optimizers/adam_op.cu

@@ -198,11 +198,11 @@ class AdamOpCUDAKernel : public framework::OpKernel<T> {
           *mom2, ctx.GetPlace(),
           ctx.template device_context<platform::DeviceContext>(), mom2_out);
       framework::TensorCopy(
-          *beta1_pow, ctx.GetPlace(),
+          *beta1_pow, beta1_pow->place(),
           ctx.template device_context<platform::DeviceContext>(),
           beta1_pow_out);
       framework::TensorCopy(
-          *beta2_pow, ctx.GetPlace(),
+          *beta2_pow, beta2_pow->place(),
           ctx.template device_context<platform::DeviceContext>(),
           beta2_pow_out);
       return;

paddle/fluid/operators/optimizers/adam_op_npu.cc

@@ -84,11 +84,11 @@ class AdamNPUKernel : public framework::OpKernel<T> {
           *mom2, ctx.GetPlace(),
           ctx.template device_context<platform::DeviceContext>(), mom2_out);
       framework::TensorCopy(
-          *beta1_pow, ctx.GetPlace(),
+          *beta1_pow, beta1_pow->place(),
           ctx.template device_context<platform::DeviceContext>(),
           beta1_pow_out);
       framework::TensorCopy(
-          *beta2_pow, ctx.GetPlace(),
+          *beta2_pow, beta2_pow->place(),
           ctx.template device_context<platform::DeviceContext>(),
           beta2_pow_out);
       return;

paddle/fluid/operators/optimizers/adam_op_xpu.cc

@@ -86,11 +86,11 @@ class AdamOpXPUKernel : public framework::OpKernel<T> {
           mom2, ctx.GetPlace(),
           ctx.template device_context<platform::DeviceContext>(), &mom2_out);
       framework::TensorCopy(
-          beta1_pow, ctx.GetPlace(),
+          beta1_pow, beta1_pow.place(),
           ctx.template device_context<platform::DeviceContext>(),
           beta1_pow_out);
       framework::TensorCopy(
-          beta2_pow, ctx.GetPlace(),
+          beta2_pow, beta2_pow.place(),
           ctx.template device_context<platform::DeviceContext>(),
           beta2_pow_out);
       return;