Fork自 PaddlePaddle / Paddle
// RUN: infrtopt %s | FileCheck %s
// CHECK-LABEL: @basic_tensor
func @basic_tensor() {
%a = "pten_dt.create_allocator.cpu" (): () -> !pten.CPU_allocator %b = "pten_dt.create_context.cpu" (): () -> !pten.CPU_context %c = "pten_dt.create_dense_tensor.cpu.f32.nchw" (%a) {dims=[1:i64], lod=[1:i64]}: (!pten.CPU_allocator) -> (!infrt.tensor<X86, NCHW, F32>) // "pten_dt.fill_dense_tensor.f32" (%c) {value=[1.0:f32]} : (!infrt.tensor<X86, NCHW, F32>) -> ()
infrt.return }