[sgl-kernel/cpu]: exclude amx gemm source from arm build (#29286)

sgl-kernel/csrc/cpu/CMakeLists.txt

@@ -93,6 +93,7 @@ endif()
 # Keep them out of Arm64 bootstrap builds until native Arm paths land.
 set(SGLANG_CPU_X86_ONLY_SOURCES
     ${CMAKE_CURRENT_SOURCE_DIR}/gemm_int4.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/gemm_int8.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/moe.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/moe_fp8.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/moe_int4.cpp

sgl-kernel/csrc/cpu/aarch64/gemm_int8.cpp

@@ -56,6 +56,23 @@ void int8_scaled_mm_impl<at::BFloat16>(
 
 }  // anonymous namespace
 
+std::tuple<at::Tensor, at::Tensor> per_token_quant_int8_cpu(at::Tensor& /*A*/) {
+  TORCH_CHECK(false, "not implemented yet");
+  return {at::Tensor(), at::Tensor()};
+}
+
+at::Tensor int8_scaled_mm_cpu(
+    at::Tensor& /*mat1*/,
+    at::Tensor& /*mat2*/,
+    at::Tensor& /*scales1*/,
+    at::Tensor& /*scales2*/,
+    const std::optional<at::Tensor>& /*bias*/,
+    at::ScalarType /*out_dtype*/,
+    bool /*is_vnni*/) {
+  TORCH_CHECK(false, "not implemented yet");
+  return at::Tensor();
+}
+
 // weight     :  static, per-channel, symmetric
 // activation : dynamic,   per-token, symmetric
 //

sgl-kernel/csrc/cpu/gemm_int8.cpp

@@ -473,7 +473,6 @@ at::Tensor int8_scaled_mm_cpu(
   return out;
 }
 
-#ifndef __aarch64__
 // fused `per_token_quant_int8_cpu` and `int8_scaled_mm_cpu`
 at::Tensor int8_scaled_mm_with_quant(
     at::Tensor& mat1,
@@ -540,4 +539,3 @@ at::Tensor int8_scaled_mm_with_quant(
   });
   return out;
 }
-#endif  // #ifndef __aarch64__