From 51c31bc10ca7c48b580cd58fcd741ba4d6db4447 Mon Sep 17 00:00:00 2001
From: Simon Mo <simon.mo@hey.com>
Date: Fri, 29 Mar 2024 18:53:08 -0700
Subject: [PATCH] CMake build elf without PTX (#3739)

---
 cmake/utils.cmake | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/cmake/utils.cmake b/cmake/utils.cmake
index 8ac81bf2..6bf5d513 100644
--- a/cmake/utils.cmake
+++ b/cmake/utils.cmake
@@ -240,9 +240,12 @@ macro(override_gpu_arches GPU_ARCHES GPU_LANG GPU_SUPPORTED_ARCHES)
       endif()
 
       if (_SM)
-        set(_VIRT "")
+        # -real suffix let CMake to only generate elf code for the kernels.
+        # we want this, otherwise the added ptx (default) will increase binary size.
+        set(_VIRT "-real")
         set(_CODE_ARCH ${_SM})
       else()
+        # -virtual suffix let CMake to generate ptx code for the kernels.
         set(_VIRT "-virtual")
         set(_CODE_ARCH ${_CODE})
       endif()