[BUGFIX ] fix undefined silu_and_mul_nvfp4_quant (#23929)

Signed-off-by: hongchao <hongchao@msh.team>
Signed-off-by: Richard Zou <zou3519@gmail.com>
Co-authored-by: hongchao <hongchao@msh.team>
Co-authored-by: Richard Zou <zou3519@gmail.com>
Co-authored-by: Richard Zou <zou3519@users.noreply.github.com>

csrc/ops.h

@@ -130,8 +130,8 @@ void silu_and_mul(torch::Tensor& out, torch::Tensor& input);
 void silu_and_mul_quant(torch::Tensor& out, torch::Tensor& input,
                         torch::Tensor& scale);
 
-#ifndef USE_ROCM
-
+#if (defined(ENABLE_NVFP4_SM100) && ENABLE_NVFP4_SM100) || \
+    (defined(ENABLE_NVFP4_SM120) && ENABLE_NVFP4_SM120)
 void silu_and_mul_nvfp4_quant(torch::Tensor& out,
                               torch::Tensor& output_block_scale,
                               torch::Tensor& input,

csrc/torch_bindings.cpp

@@ -115,7 +115,8 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
       "silu_and_mul_quant(Tensor! result, Tensor input, Tensor scale) -> ()");
   ops.impl("silu_and_mul_quant", torch::kCUDA, &silu_and_mul_quant);
 
-#ifndef USE_ROCM
+#if (defined(ENABLE_NVFP4_SM100) && ENABLE_NVFP4_SM100) || \
+    (defined(ENABLE_NVFP4_SM120) && ENABLE_NVFP4_SM120)
   ops.def(
       "silu_and_mul_nvfp4_quant(Tensor! result, Tensor! result_block_scale, "
       "Tensor input, Tensor input_global_scale) -> ()");

vllm/compilation/fix_functionalization.py

@@ -97,7 +97,9 @@ class FixFunctionalizationPass(VllmInductorPass):
                                      node,
                                      mutated_args,
                                      args=('result', 'input', 'scale'))
-            elif at_target == torch.ops._C.silu_and_mul_nvfp4_quant.default:
+            elif hasattr(
+                    torch.ops._C, "silu_and_mul_nvfp4_quant"
+            ) and at_target == torch.ops._C.silu_and_mul_nvfp4_quant.default:
                 mutated_args = {1: 'result', 2: 'result_block_scale'}
                 self.defunctionalize(graph,
                                      node,