add --gpu-max-threads-per-block=1024 options

setup.py

@@ -169,6 +169,9 @@ if (TORCH_MAJOR > 1) or (TORCH_MAJOR == 1 and TORCH_MINOR > 4):
     version_ge_1_5 = ["-DVERSION_GE_1_5"]
 version_dependent_macros = version_ge_1_1 + version_ge_1_3 + version_ge_1_5
 
+if IS_ROCM_PYTORCH:
+    version_dependent_macros += ['--gpu-max-threads-per-block=1024']
+
 if "--distributed_adam" in sys.argv or "--cuda_ext" in sys.argv:
     if "--distributed_adam" in sys.argv:
         sys.argv.remove("--distributed_adam")