From 4b0b0ea55f0c8bd418970d33c39262a1c90d446b Mon Sep 17 00:00:00 2001
From: Yibo Cai <yibo.cai@arm.com>
Date: Fri, 26 Jun 2026 08:23:43 +0800
Subject: [PATCH] [sgl-kernel/cpu]: exclude amx gemm source from arm build
 (#29286)

---
 sgl-kernel/csrc/cpu/CMakeLists.txt        |  1 +
 sgl-kernel/csrc/cpu/aarch64/gemm_int8.cpp | 17 +++++++++++++++++
 sgl-kernel/csrc/cpu/gemm_int8.cpp         |  2 --
 3 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/sgl-kernel/csrc/cpu/CMakeLists.txt b/sgl-kernel/csrc/cpu/CMakeLists.txt
index 2fa0cb82a..e69ee85c6 100644
--- a/sgl-kernel/csrc/cpu/CMakeLists.txt
+++ b/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
diff --git a/sgl-kernel/csrc/cpu/aarch64/gemm_int8.cpp b/sgl-kernel/csrc/cpu/aarch64/gemm_int8.cpp
index 8dae500e4..df8757338 100644
--- a/sgl-kernel/csrc/cpu/aarch64/gemm_int8.cpp
+++ b/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
 //
diff --git a/sgl-kernel/csrc/cpu/gemm_int8.cpp b/sgl-kernel/csrc/cpu/gemm_int8.cpp
index 1da63f79e..f72616697 100644
--- a/sgl-kernel/csrc/cpu/gemm_int8.cpp
+++ b/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__