diff options
author | Patrick Nguyen <drpng@google.com> | 2018-04-25 20:30:00 -0700 |
---|---|---|
committer | Patrick Nguyen <drpng@google.com> | 2018-04-25 20:30:00 -0700 |
commit | f8d7553d0f34b621d6123e153e69b0babb09d22c (patch) | |
tree | 62355f3d9d9f7a4f81f31910779ab399e4850ded /tensorflow/compiler/xla/service/cpu/ir_emitter.cc | |
parent | adf045607cc4126366ebb84ee2109f88c6ab25fc (diff) | |
parent | 43a7072882196c7ac2d9429050a3140b1ecb52db (diff) |
Merge commit for internal changes.
Diffstat (limited to 'tensorflow/compiler/xla/service/cpu/ir_emitter.cc')
-rw-r--r-- | tensorflow/compiler/xla/service/cpu/ir_emitter.cc | 21 |
1 files changed, 1 insertions, 20 deletions
diff --git a/tensorflow/compiler/xla/service/cpu/ir_emitter.cc b/tensorflow/compiler/xla/service/cpu/ir_emitter.cc index f990ee2785..0b08ad8da3 100644 --- a/tensorflow/compiler/xla/service/cpu/ir_emitter.cc +++ b/tensorflow/compiler/xla/service/cpu/ir_emitter.cc @@ -93,8 +93,6 @@ IrEmitter::IrEmitter( computation_to_profile_idx_(std::move(computation_to_profile_idx)), alias_analysis_(hlo_module, assignment, &llvm_module->getContext()), hlo_module_config_(hlo_module.config()), - parallel_cpu_backend_( - options::CpuParallelBackendRequested(hlo_module_config_)), is_top_level_computation_(false), target_machine_features_(target_machine), external_constant_pool_(external_constant_pool) { @@ -2163,8 +2161,7 @@ Status IrEmitter::HandleCall(HloInstruction* call) { TF_RETURN_IF_ERROR(EmitTargetAddressForOp(call)); - if (!computation->root_instruction()->outer_dimension_partitions().empty() && - !parallel_cpu_backend_) { + if (!computation->root_instruction()->outer_dimension_partitions().empty()) { // ParallelTaskAssignment assigned partitions, emit call to // ParallelForkJoin. std::vector<llvm::Value*> call_args = GetArrayFunctionCallArguments( @@ -2550,22 +2547,6 @@ Status IrEmitter::FinishVisit(HloInstruction* root) { } }; - // For the parallel cpu backend, we record the total for each embedded - // computation callee with its caller kCall HLO. - if (parallel_cpu_backend_ && is_top_level_computation_) { - auto* computation = root->parent(); - auto* entry_computation = computation->parent()->entry_computation(); - if (computation != entry_computation) { - for (HloInstruction* instruction : entry_computation->instructions()) { - if (instruction->opcode() == HloOpcode::kCall && - instruction->to_apply()->root_instruction() == root) { - record_complete_computation(GetProfileCounterFor(*instruction)); - return Status::OK(); - } - } - } - } - // For the entry computation this increment is cumulative of embedded // computations since it includes cycles spent in computations invoked by // While, Call etc. |