1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
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
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
|
/* Copyright 2018 The TensorFlow 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.
==============================================================================*/
#include <memory>
#include <utility>
#include "llvm/IR/Module.h"
#include "tensorflow/compiler/xla/literal.h"
#include "tensorflow/compiler/xla/ptr_util.h"
#include "tensorflow/compiler/xla/service/buffer_assignment.h"
#include "tensorflow/compiler/xla/service/cpu/tests/cpu_codegen_test.h"
#include "tensorflow/compiler/xla/service/hlo_computation.h"
#include "tensorflow/compiler/xla/service/hlo_instruction.h"
#include "tensorflow/compiler/xla/service/hlo_module.h"
#include "tensorflow/compiler/xla/service/llvm_ir/alias_analysis.h"
#include "tensorflow/compiler/xla/service/llvm_ir/llvm_util.h"
#include "tensorflow/compiler/xla/shape_util.h"
#include "tensorflow/compiler/xla/tests/filecheck.h"
#include "tensorflow/compiler/xla/xla_data.pb.h"
#include "tensorflow/core/platform/test.h"
namespace xla {
namespace cpu {
class CpuNoAliasTest : public CpuCodegenTest {};
// Creates a simple HLO ir_module (runs concat(concat(x, y), x)), and then
// inspects the aliasing information for loads to its buffers.
TEST_F(CpuNoAliasTest, Concat) {
HloComputation::Builder builder(TestName());
std::unique_ptr<Literal> literal =
LiteralUtil::CreateR2<float>({{1.0, 2.0}, {3.0, 4.0}});
auto param_shape = ShapeUtil::MakeShape(F32, {2, 2});
HloInstruction* param_x = builder.AddInstruction(
HloInstruction::CreateParameter(0, param_shape, "x"));
HloInstruction* param_y = builder.AddInstruction(
HloInstruction::CreateParameter(1, param_shape, "y"));
HloInstruction* concat1 =
builder.AddInstruction(HloInstruction::CreateConcatenate(
ShapeUtil::MakeShape(F32, {2, 4}), {param_x, param_y}, 1));
HloInstruction* concat2 =
builder.AddInstruction(HloInstruction::CreateConcatenate(
ShapeUtil::MakeShape(F32, {2, 6}), {concat1, param_x}, 1));
std::unique_ptr<HloComputation> computation = builder.Build();
auto hlo_module = CreateNewModule();
hlo_module->AddEntryComputation(std::move(computation));
// Now that we have an HLO module, build an llvm_ir::AliasAnalysis for it.
auto status_or_buffer_assn = BufferAssigner::Run(
hlo_module.get(), MakeUnique<DependencyHloOrdering>(hlo_module.get()),
backend().compiler()->BufferSizeBytesFunction(),
[](LogicalBuffer::Color) { return /*alignment=*/1; });
ASSERT_EQ(status_or_buffer_assn.status(), Status::OK());
llvm::LLVMContext context;
llvm_ir::AliasAnalysis aa(*hlo_module, *status_or_buffer_assn.ValueOrDie(),
&context);
// Construct an LLVM module containing loads that we annotate as being from
// the buffers in the HLO module. We'll inspect these loads to ensure that
// they have the expected alias information.
llvm::Module ir_module("test", context);
llvm::Function* func = llvm::cast<llvm::Function>(
ir_module.getOrInsertFunction("test_fn", llvm::Type::getVoidTy(context)));
llvm::BasicBlock* bb = llvm::BasicBlock::Create(context, "body", func);
llvm::IRBuilder<> b(bb);
auto* zero = llvm::ConstantInt::get(llvm::Type::getInt32Ty(context), 0);
llvm_ir::IrArray::Index zero2D({zero, zero});
llvm::ArrayType* array2d_type = llvm::ArrayType::get(
llvm::ArrayType::get(llvm::Type::getFloatTy(context), 100), 100);
{
llvm::Value* param_x_val =
ir_module.getOrInsertGlobal("param_x", array2d_type);
llvm_ir::IrArray param_x_array(param_x_val, param_shape);
aa.AddAliasingInformationToIrArray(*param_x, ¶m_x_array);
param_x_array.EmitReadArrayElement(zero2D, &b)
->setName("read_param_x_array");
}
{
llvm::Value* concat1_val =
ir_module.getOrInsertGlobal("concat1", array2d_type);
auto shape = ShapeUtil::MakeShape(F32, {2, 4});
llvm_ir::IrArray concat1_array(concat1_val, shape);
aa.AddAliasingInformationToIrArray(*concat1, &concat1_array);
concat1_array.EmitReadArrayElement(zero2D, &b)
->setName("read_concat1_array");
}
{
llvm::Value* concat2_val =
ir_module.getOrInsertGlobal("concat2", array2d_type);
auto shape = ShapeUtil::MakeShape(F32, {2, 6});
llvm_ir::IrArray concat2_array(concat2_val, shape);
aa.AddAliasingInformationToIrArray(*concat2, &concat2_array);
concat2_array.EmitReadArrayElement(zero2D, &b)
->setName("read_concat2_array");
}
// Check the AA info in the loads.
const char* filecheck_pattern = R"(
CHECK: %read_param_x_array = load {{.*}} !noalias [[param_x_noalias:![0-9]+]]
CHECK: %read_concat1_array = load {{.*}} !alias.scope [[concat1_scope:![0-9]+]], !noalias [[concat1_noalias:![0-9]+]]
CHECK: %read_concat2_array = load {{.*}} !alias.scope [[concat1_noalias]], !noalias [[concat1_scope]]
CHECK-DAG: [[buf_size32:![0-9]+]] = !{!"buffer:{{.*}} size:32
CHECK-DAG: [[buf_size48:![0-9]+]] = !{!"buffer:{{.*}} size:48
CHECK-DAG: [[param_x_noalias]] = !{[[buf_size32]], [[buf_size48]]}
CHECK-DAG: [[concat1_scope]] = !{[[buf_size32]]}
CHECK-DAG: [[concat1_noalias]] = !{[[buf_size48]]}
)";
TF_ASSERT_OK_AND_ASSIGN(
bool filecheck_match,
RunFileCheck(llvm_ir::DumpModuleToString(ir_module), filecheck_pattern));
EXPECT_TRUE(filecheck_match);
}
} // namespace cpu
} // namespace xla
|
