Update oneccl-binding.patch (#12377)

* Add files via upload

* upload oneccl-binding.patch

* Update Dockerfile

docker/llm/serving/xpu/docker/Dockerfile

@@ -10,6 +10,7 @@ ENV PYTHONUNBUFFERED=1
 # Disable pip's cache behavior
 ARG PIP_NO_CACHE_DIR=false
 ADD ./gradio_web_server.patch  /tmp/gradio_web_server.patch
+ADD ./oneccl-binding.patch /tmp/oneccl-binding.patch
 
 RUN wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | tee /usr/share/keyrings/intel-oneapi-archive-keyring.gpg > /dev/null && \
     echo "deb [signed-by=/usr/share/keyrings/intel-oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main " | tee /etc/apt/sources.list.d/oneAPI.list && \
@@ -57,6 +58,7 @@ RUN wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRO
     bash oneccl_wks_installer_2024.0.0.5.1.sh && \
     git clone https://github.com/intel/torch-ccl -b v2.1.300+xpu && \
     cd torch-ccl && \
+    patch -p1 < /tmp/oneccl-binding.patch && \
     USE_SYSTEM_ONECCL=ON COMPUTE_BACKEND=dpcpp python setup.py install && \
     apt-get update && \
     apt-get install -y --no-install-recommends libfabric-dev wrk libaio-dev numactl && \

docker/llm/serving/xpu/docker/oneccl-binding.patch

@@ -1,14 +1,32 @@
 diff --git a/src/gpu/dpcpp_ccl.cpp b/src/gpu/dpcpp_ccl.cpp
-index 3bd8087..c5b5ce3 100644
+index bb1f236..289d490 100644
 --- a/src/gpu/dpcpp_ccl.cpp
 +++ b/src/gpu/dpcpp_ccl.cpp
-@@ -689,7 +689,8 @@ c10::intrusive_ptr<ProcessGroupCCL::AsyncWorkCCL> XPUCCLStubs::allreduce_(std::v
+@@ -80,6 +80,16 @@ int get_sync_only(int init_value = 0) {
+   return tmp_sync_only;
+ }
+
++char* get_cached_env(const char* var_name) {
++    static char* cached_value = nullptr;
++    if (!cached_value) {
++        cached_value = std::getenv(var_name);
++        if (!cached_value) {
++           cached_value = "0";
++        }
++    }
++    return cached_value;
++}
+
+ #define CCL_KERNEL_SUBMIT(cmd, q) \
+ ({bool profile_barrier = (xpu::is_profiler_enabled());                        \
+@@ -759,6 +769,10 @@ c10::intrusive_ptr<ProcessGroupCCL::AsyncWorkCCL> XPUCCLStubs::allreduce_impl(st
                                              stream,
                                              attr), stream.get_native());
        });
--    // printf("Use One CCL allreduce.\n");
-+    stream.get_native().wait();
-+   // printf("Use One CCL allreduce.\n");
++    const char* env_value = get_cached_env("IPEX_LLM_CCL_ENABLE_NATIVE_WAIT");
++    if (env_value && std::string(env_value) == "1") {
++        stream.get_native().wait();
++    }
+     // printf("Use One CCL allreduce.\n");
      return ret_evt;
    },
-   c10d::OpType::ALLREDUCE);