cinn_launch_op_test.cc 4.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */

15
#include "paddle/fluid/operators/cinn/cinn_launch_op.h"
16
#include <stdlib.h>
17
#include <mutex>
18 19
#include <random>
#include <string>
20
#include "gflags/gflags.h"
21 22 23 24
#include "gtest/gtest.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/paddle2cinn/cinn_compiler.h"
#include "paddle/fluid/framework/scope.h"
25
#include "paddle/fluid/operators/cinn/test_helper.h"
26 27
#include "paddle/fluid/platform/cpu_helper.h"
#include "paddle/fluid/platform/init.h"
28
#include "paddle/phi/core/ddim.h"
29
#include "paddle/phi/core/kernel_registry.h"
30 31

USE_OP(cinn_launch);
32
USE_OP(cinn_instruction_run);
33
USE_OP_ITSELF(elementwise_add);
34
DECLARE_double(eager_delete_tensor_gb);
35
DECLARE_bool(enable_pe_launch_cinn);
36
DECLARE_bool(enable_cinn_auto_tune);
37

38 39 40 41 42
PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
#ifdef PADDLE_WITH_CUDA
PD_DECLARE_KERNEL(add, GPU, ALL_LAYOUT);
#endif

43
namespace paddle::operators {
44

45
using framework::paddle2cinn::CinnCompiler;
46

47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
class TestCinnLaunchOp : public ::testing::Test {
 public:
  const char* test_op_out_name = "add_op_out";
  const char* add_op_out_name = "add_op_out";
  std::unique_ptr<framework::OperatorBase> cinn_launch_op;
  std::unique_ptr<framework::OperatorBase> elementwise_add_op;

  void SetUp() override {
    paddle::framework::InitDevices();
    platform::SetNumThreads(1);
    // cache test graph into CinnCompiler
    auto compilation_key = CinnCompiler::GetInstance()->AddGraph(
        CreateOnlyElementwiseAddGraph("x", "y", test_op_out_name));

    // create cinn_launch_op and elementwise_add op
    cinn_launch_op = paddle::framework::OpRegistry::CreateOp(
        "cinn_launch", {{"X", {"x", "y"}}}, {{"Out", {test_op_out_name}}},
        {{"compilation_key", compilation_key}});
    elementwise_add_op = paddle::framework::OpRegistry::CreateOp(
        "elementwise_add", {{"X", {"x"}}, {"Y", {"y"}}},
        {{"Out", {add_op_out_name}}}, {{}});
  }

  void RunAndCheck(const platform::Place& place) {
    // Run ops and check the computation results
72
    framework::Scope scope;
73 74 75 76
    InitVariablesWithRandomValue<float>({"x", "y"}, {10, 20}, place, &scope);
    scope.Var(test_op_out_name)->GetMutable<LoDTensor>();
    scope.Var(add_op_out_name)->GetMutable<LoDTensor>();
    elementwise_add_op->Run(scope, place);
77
    cinn_launch_op->Run(scope, place);
78 79
    CompareOpResult<float>(scope.GetVar(test_op_out_name),
                           scope.GetVar(add_op_out_name));
80 81 82 83
  }

  void TearDown() override { CinnCompiler::GetInstance()->Clear(); }
};
84

85
TEST_F(TestCinnLaunchOp, TestRunInstructionByPE) {
86
  // CPU
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103
  RunAndCheck(platform::CPUPlace());
  // the second run on the same place is to check the cache logic
  RunAndCheck(platform::CPUPlace());
#ifdef PADDLE_WITH_CUDA
  // GPU
  RunAndCheck(platform::CUDAPlace());
  RunAndCheck(platform::CUDAPlace());
#endif
}

TEST_F(TestCinnLaunchOp, TestRunInstructionByCinnProgram) {
  // set FLAGS_enable_pe_launch_cinn=false to switch to use
  // default scheduler of CINN to execute the compiled program
  FLAGS_enable_pe_launch_cinn = false;

  RunAndCheck(platform::CPUPlace());
  RunAndCheck(platform::CPUPlace());
104
#ifdef PADDLE_WITH_CUDA
105
  // GPU
106 107
  RunAndCheck(platform::CUDAPlace());
  RunAndCheck(platform::CUDAPlace());
108
#endif
109 110
}

111 112 113 114 115 116 117 118
TEST_F(TestCinnLaunchOp, TestRunWithAutoTuneEnabled) {
  FLAGS_enable_cinn_auto_tune = true;

  // currently only check on cpu, will add a test for gpu after CINN ready
  RunAndCheck(platform::CPUPlace());
  RunAndCheck(platform::CPUPlace());
}

119 120 121 122 123 124 125 126 127 128 129 130 131 132
namespace details {
// Testing helper function used on CinnLaunchOpKernel in the following:
// firstly build test data, then check both expected and illegal situations

TEST(CinnLaunchOpHelperTest, TestPlaceToCinnTarget) {
  ASSERT_EQ(PlaceToCinnTarget(platform::CPUPlace()),
            ::cinn::common::DefaultHostTarget());
  ASSERT_EQ(PlaceToCinnTarget(platform::CUDAPlace(0)),
            ::cinn::common::DefaultNVGPUTarget());
  ASSERT_THROW(PlaceToCinnTarget(platform::XPUPlace()),
               paddle::platform::EnforceNotMet);
}

}  // namespace details
133
}  // namespace paddle::operators