diff options
-rw-r--r-- | tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/nvptx_backend_lib.cc | 31 |
1 files changed, 16 insertions, 15 deletions
diff --git a/tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/nvptx_backend_lib.cc b/tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/nvptx_backend_lib.cc index 6c1c20fc04..3f593f426d 100644 --- a/tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/nvptx_backend_lib.cc +++ b/tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/nvptx_backend_lib.cc @@ -114,21 +114,22 @@ static string GetLibdeviceFilename(const string& libdevice_dir_path, // Gets the GPU name as it's known to LLVM for a given compute capability. If // we see an unrecognized compute capability, we return "sm_30". static string GetSmName(std::pair<int, int> compute_capability) { - static auto* m = new std::map<std::pair<int, int>, int>( - {{{2, 0}, 20}, - {{2, 1}, 21}, - {{3, 0}, 30}, - {{3, 2}, 32}, - {{3, 5}, 35}, - {{3, 7}, 37}, - {{5, 0}, 50}, - {{5, 2}, 52}, - {{5, 3}, 53}, - {{6, 0}, 60}, - {{6, 1}, 61}, - {{6, 2}, 62}, - // TODO: Change this to 70 once LLVM NVPTX supports it - {{7, 0}, 60}}); + static auto* m = new std::map<std::pair<int, int>, int>({ + {{2, 0}, 20}, + {{2, 1}, 21}, + {{3, 0}, 30}, + {{3, 2}, 32}, + {{3, 5}, 35}, + {{3, 7}, 37}, + {{5, 0}, 50}, + {{5, 2}, 52}, + {{5, 3}, 53}, + {{6, 0}, 60}, + {{6, 1}, 61}, + {{6, 2}, 62}, + {{7, 0}, 70}, + {{7, 2}, 72}, + }); int sm_version = 30; auto it = m->find(compute_capability); if (it != m->end()) { |