fix the arch support in CMakeLists.txt to include 8.9

fhl2000 · fhl2000 · commit 58ce47753c72 · 2025-06-25T13:37:05.000+08:00
Signed-off-by: fhl &lt;2410591650@qq.com&gt;
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -308,7 +308,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
   # Keep building Marlin for 9.0 as there are some group sizes and shapes that
   # are not supported by Machete yet.
   # 9.0 for latest bf16 atomicAdd PTX
-  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;8.7;9.0+PTX" "${CUDA_ARCHS}")
+  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;8.7;8.9;9.0+PTX" "${CUDA_ARCHS}")
   if (MARLIN_ARCHS)
 
     #
@@ -684,7 +684,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
 
   list(APPEND VLLM_MOE_EXT_SRC "${VLLM_MOE_WNA16_SRC}")
   # 9.0 for latest bf16 atomicAdd PTX
-  cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.7;9.0+PTX" "${CUDA_ARCHS}")
+  cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.7;8.9;9.0+PTX" "${CUDA_ARCHS}")
   if (MARLIN_MOE_ARCHS)
 
     #

Original file line number	Diff line number	Diff line change
`@@ -308,7 +308,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")`
`308`	`308`	`# Keep building Marlin for 9.0 as there are some group sizes and shapes that`
`309`	`309`	`# are not supported by Machete yet.`
`310`	`310`	`# 9.0 for latest bf16 atomicAdd PTX`
`311`		`- cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;8.7;9.0+PTX" "${CUDA_ARCHS}")`
	`311`	`+ cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;8.7;8.9;9.0+PTX" "${CUDA_ARCHS}")`
`312`	`312`	`if (MARLIN_ARCHS)`
`313`	`313`
`314`	`314`	`#`
`@@ -684,7 +684,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")`
`684`	`684`
`685`	`685`	`list(APPEND VLLM_MOE_EXT_SRC "${VLLM_MOE_WNA16_SRC}")`
`686`	`686`	`# 9.0 for latest bf16 atomicAdd PTX`
`687`		`- cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.7;9.0+PTX" "${CUDA_ARCHS}")`
	`687`	`+ cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.7;8.9;9.0+PTX" "${CUDA_ARCHS}")`
`688`	`688`	`if (MARLIN_MOE_ARCHS)`
`689`	`689`
`690`	`690`	`#`