[XLA:GPU] Only add the cubin if it is available

It's only non-empty if we were able to run ptxas. If the PTX is going to be
JIT'ed by the driver it won't be around. Loading an empty cubin will result in
a fatal error.

PiperOrigin-RevId: 206341931

2 files changed, 4 insertions, 1 deletions

diff --git a/tensorflow/compiler/xla/service/gpu/gpu_executable.cc b/tensorflow/compiler/xla/service/gpu/gpu_executable.cc
index 52c8595ffb..0179b43240 100644
--- a/tensorflow/compiler/xla/service/gpu/gpu_executable.cc
+++ b/tensorflow/compiler/xla/service/gpu/gpu_executable.cc
@@ -190,7 +190,9 @@ GpuExecutable::ResolveConstantGlobals(se::StreamExecutor* executor) {
   }
 
   se::MultiModuleLoaderSpec module_spec;
-  module_spec.AddCudaCubinInMemory(cubin());
+  if (!cubin().empty()) {
+    module_spec.AddCudaCubinInMemory(cubin());
+  }
   module_spec.AddCudaPtxInMemory(ptx().c_str());
 
   tensorflow::gtl::FlatMap<int64, se::DeviceMemoryBase> globals;
diff --git a/tensorflow/stream_executor/module_spec.h b/tensorflow/stream_executor/module_spec.h
index 212ae7ba9c..75bdfed2d7 100644
--- a/tensorflow/stream_executor/module_spec.h
+++ b/tensorflow/stream_executor/module_spec.h
@@ -43,6 +43,7 @@ class MultiModuleLoaderSpec {
   }
 
   void AddCudaCubinInMemory(port::ArraySlice<const uint8> cubin_bytes) {
+    CHECK(!cubin_bytes.empty());
     has_cuda_cubin_in_memory_ = true;
     cuda_cubin_in_memory_ = cubin_bytes;
   }