// allocate temp memory for factor on xpu

xpu::ctx_guard RAII_GUARD(xpu_context);

XPUType* factor_data = RAII_GUARD.alloc_l3_or_gm<XPUType>(1);

PADDLE_ENFORCE_NOT_NULL(

factor_data, errors::External("XPU alloc_l3_or_gm returns nullptr"));

memory_utils::Copy(dev_ctx.GetPlace(),

static_cast<void*>(factor_data),

phi::CPUPlace(),

static_cast<void*>(&pow_factor),

sizeof(T));

auto x_dims = common::vectorize<int>(x.dims());

// use [1] to replace [], because xpu not support []

if (x_dims.size() == 0) {

x_dims = std::vector<int>({1});

}

// broadcast_pow(Context* ctx, const T* x, const T* y, T* z, const

// std::vector<int>& xshape, const std::vector<int>& yshape);

int r =

xpu::broadcast_pow(xpu_context, x_data, factor_data, y_data, x_dims, {1});

PADDLE_ENFORCE_XDNN_SUCCESS(r, "broadcast_pow");