experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix NVPTX TargetMachine construction to use reloc/code model flags only if explicitly set. Fixes failures caused by r348585.

Browse Source 
PiperOrigin-RevId: 224560930
This commit is contained in:
A. Unique TensorFlower 2018-12-07 12:39:53 -08:00 committed by TensorFlower Gardener
parent 9fffd26041
commit 4f543e588a
1 changed files with 1 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/nvptx_backend_lib.cc
View File

@ -199,8 +199,7 @@ std::unique_ptr<llvm::TargetMachine> GetTargetMachine(
} }
return absl::WrapUnique(target->createTargetMachine( return absl::WrapUnique(target->createTargetMachine(
triple.str(), llvm_ir::AsStringRef(cpu_name), "+ptx60", target_options, triple.str(), llvm_ir::AsStringRef(cpu_name), "+ptx60", target_options,
Optional<Reloc::Model>(RelocModel), Optional<CodeModel::Model>(CMModel), getRelocModel(), getCodeModel(), codegen_opt_level));
codegen_opt_level));
} }
// Adds the standard LLVM optimization passes, based on the speed optimization // Adds the standard LLVM optimization passes, based on the speed optimization