1 /* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15
16 #include "tensorflow/compiler/xla/service/llvm_ir/kernel_support_library.h"
17
18 #include "tensorflow/compiler/xla/service/llvm_ir/llvm_type_conversion_util.h"
19 #include "tensorflow/compiler/xla/service/llvm_ir/llvm_util.h"
20
21 namespace xla {
ForWithStatus(absl::string_view name,llvm::Value * start,llvm::Value * end,llvm::Value * step,const std::function<Status (llvm::Value *,bool)> & for_body_generator)22 Status KernelSupportLibrary::ForWithStatus(
23 absl::string_view name, llvm::Value* start, llvm::Value* end,
24 llvm::Value* step,
25 const std::function<Status(llvm::Value*, bool)>& for_body_generator) {
26 return IfWithStatus(b_->CreateICmpSLT(start, end), [&]() -> Status {
27 TF_RETURN_IF_ERROR(for_body_generator(start, /*is_first_iteration=*/true));
28 return ForWithStatus(
29 name, b_->CreateAdd(start, step), end, step,
30 [&](llvm::Value* iv) { return for_body_generator(iv, false); });
31 });
32 }
33
ForWithStatus(absl::string_view name,llvm::Value * start,llvm::Value * end,llvm::Value * step,bool peel_first_iteration,const std::function<Status (llvm::Value *,llvm::Value *)> & for_body_generator)34 Status KernelSupportLibrary::ForWithStatus(
35 absl::string_view name, llvm::Value* start, llvm::Value* end,
36 llvm::Value* step, bool peel_first_iteration,
37 const std::function<Status(llvm::Value*, llvm::Value*)>&
38 for_body_generator) {
39 if (peel_first_iteration) {
40 return ForWithStatus(
41 name, start, end, step, true,
42 [&](llvm::Value* indvar, bool is_first_iteration) -> Status {
43 return for_body_generator(indvar, b_->getInt1(is_first_iteration));
44 });
45 } else {
46 std::unique_ptr<llvm_ir::ForLoop> loop = llvm_ir::ForLoop::EmitForLoop(
47 name, start, end, step, b_,
48 /*unroll_mode=*/unroll_mode_,
49 /*prevent_vectorization=*/prevent_vectorization_);
50 b_->SetInsertPoint(&loop->GetBodyBasicBlock()->back());
51 TF_RETURN_IF_ERROR(
52 for_body_generator(loop->GetIndVarValue(),
53 /*is_first_iteration=*/b_->CreateICmpEQ(
54 loop->GetIndVarValue(), start)));
55 llvm_ir::SetToLastInsertPoint(loop->GetExitBasicBlock(), b_);
56 return OkStatus();
57 }
58 }
59
IfWithStatus(absl::string_view name,llvm::Value * condition,const std::function<Status ()> & true_block_generator,const std::function<Status ()> & false_block_generator)60 Status KernelSupportLibrary::IfWithStatus(
61 absl::string_view name, llvm::Value* condition,
62 const std::function<Status()>& true_block_generator,
63 const std::function<Status()>& false_block_generator) {
64 llvm_ir::LlvmIfData if_data =
65 llvm_ir::EmitIfThenElse(condition, name, b_,
66 /*emit_else=*/false_block_generator != nullptr);
67 b_->SetInsertPoint(&if_data.true_block->back());
68 TF_RETURN_IF_ERROR(true_block_generator());
69 if (false_block_generator != nullptr) {
70 b_->SetInsertPoint(&if_data.false_block->back());
71 TF_RETURN_IF_ERROR(false_block_generator());
72 }
73 llvm_ir::SetToLastInsertPoint(if_data.after_block, b_);
74 return OkStatus();
75 }
76
EmitAndCallOutlinedKernel(const HloModuleConfig & module_config,llvm::IRBuilder<> * b,absl::string_view kernel_name,KernelSupportLibrary::ArgumentVector arguments,const std::function<void (KernelSupportLibrary::ArgumentVector)> & kernel_body_generator)77 void KernelSupportLibrary::EmitAndCallOutlinedKernel(
78 const HloModuleConfig& module_config, llvm::IRBuilder<>* b,
79 absl::string_view kernel_name,
80 KernelSupportLibrary::ArgumentVector arguments,
81 const std::function<void(KernelSupportLibrary::ArgumentVector)>&
82 kernel_body_generator) {
83 llvm::Module* module = b->GetInsertBlock()->getModule();
84 llvm::Function* function =
85 module->getFunction(llvm_ir::AsStringRef(kernel_name));
86
87 int64_t null_arg_idx = -1;
88 std::vector<llvm::Value*> sanitized_args;
89 sanitized_args.reserve(arguments.size());
90 for (int64_t i = 0, e = arguments.size(); i < e; i++) {
91 if (arguments[i]) {
92 sanitized_args.push_back(arguments[i]);
93 } else {
94 CHECK_EQ(null_arg_idx, -1);
95 null_arg_idx = i;
96 }
97 }
98
99 if (!function) {
100 VLOG(2) << "Generating kernel for " << kernel_name;
101 std::vector<llvm::Type*> arg_types;
102 std::transform(sanitized_args.begin(), sanitized_args.end(),
103 std::back_inserter(arg_types),
104 [](llvm::Value* arg) { return arg->getType(); });
105
106 auto* function_type =
107 llvm::FunctionType::get(b->getVoidTy(), arg_types, /*isVarArg=*/false);
108
109 function = llvm_ir::CreateCpuFunction(function_type,
110 llvm::GlobalValue::InternalLinkage,
111 module_config, kernel_name, module);
112
113 llvm::IRBuilder<>::InsertPointGuard guard(*b);
114
115 auto* entry_bb =
116 llvm::BasicBlock::Create(b->getContext(), "entry", function);
117 auto* return_inst = llvm::ReturnInst::Create(b->getContext(),
118 /*retVal=*/nullptr, entry_bb);
119 // Set the insert point to before return_inst.
120 b->SetInsertPoint(return_inst);
121
122 std::vector<llvm::Value*> arg_values;
123 /*
124 * clang on OSX doesn't like std::transform or range for loop here.
125 * See https://github.com/tensorflow/tensorflow/issues/15196
126 */
127 for (llvm::Function::arg_iterator arg = function->arg_begin(),
128 arg_e = function->arg_end();
129 arg != arg_e; ++arg) {
130 arg_values.push_back(arg);
131 }
132 if (null_arg_idx != -1) {
133 arg_values.insert(arg_values.begin() + null_arg_idx, nullptr);
134 }
135 kernel_body_generator(arg_values);
136 } else {
137 VLOG(3) << "Re-using kernel for " << kernel_name;
138 }
139
140 b->CreateCall(function, llvm_ir::AsArrayRef(sanitized_args));
141 }
142
143 } // namespace xla
144