add guard to block_count (#2909)

* add log

* update test script

* guard for block_count

* remove test script

Author: ScXfjiang

tensorflow/core/util/gpu_launch_config.h

@@ -203,6 +203,7 @@ GpuLaunchConfig GetGpuLaunchConfigFixedBlockSize(
 #elif TENSORFLOW_USE_ROCM
   hipError_t err = hipOccupancyMaxActiveBlocksPerMultiprocessor(
       &block_count, func, fixed_block_size, dynamic_shared_memory_size);
+  if (block_count < 1) { block_count = 1; }
   CHECK_EQ(err, hipSuccess);
 #endif
   block_count = std::min(block_count * d.getNumGpuMultiProcessors(),