[Bugfix] Fix spurious "No compiled cutlass_scaled_mm ..." for W8A8 on Turing (#9487)

CMakeLists.txt

@@ -252,7 +252,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     message(STATUS "Building Marlin kernels for archs: ${MARLIN_ARCHS}")
   else()
     message(STATUS "Not building Marlin kernels as no compatible archs found"
-                   "in CUDA target architectures")
+                   " in CUDA target architectures")
   endif()
 
   #
@@ -432,7 +432,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     message(STATUS "Building Marlin MOE kernels for archs: ${MARLIN_MOE_ARCHS}")
   else()
     message(STATUS "Not building Marlin MOE kernels as no compatible archs found"
-                   "in CUDA target architectures")
+                   " in CUDA target architectures")
   endif()
 endif()
 

csrc/quantization/cutlass_w8a8/scaled_mm_entry.cu

@@ -137,9 +137,11 @@ void cutlass_scaled_mm(torch::Tensor& c, torch::Tensor const& a,
     return;
   }
 
+  if (version_num >= 75) {
     // Turing
-  TORCH_CHECK(version_num >= 75);
     cutlass_scaled_mm_sm75(c, a, b, a_scales, b_scales, bias);
+    return;
+  }
 #endif
 
   TORCH_CHECK_NOT_IMPLEMENTED(