[XPU] c_allreduce support int. update bkcl to 1.0.5. test=kunlun (#45248)

cmake/external/xpu.cmake

@@ -25,7 +25,7 @@ else()
 endif()
 
 set(XPU_XCCL_BASE_URL
-    "https://klx-sdk-release-public.su.bcebos.com/xccl/release/1.0.4")
+    "https://klx-sdk-release-public.su.bcebos.com/xccl/release/1.0.5")
 
 if(WITH_AARCH64)
   set(XPU_XRE_DIR_NAME "xre-kylin_aarch64")

paddle/fluid/operators/collective/c_allreduce_sum_op_xpu.cc

@@ -19,4 +19,5 @@ namespace plat = paddle::platform;
 
 REGISTER_OP_XPU_KERNEL(c_allreduce_sum,
                        ops::CAllReduceOpXPUKernel<ops::kRedSum, float>,
-                       ops::CAllReduceOpXPUKernel<ops::kRedSum, plat::float16>)
+                       ops::CAllReduceOpXPUKernel<ops::kRedSum, plat::float16>,
+                       ops::CAllReduceOpXPUKernel<ops::kRedSum, int>)

paddle/fluid/platform/device/xpu/xpu2_op_list.h

@@ -80,13 +80,16 @@ XPUOpMap& get_kl2_ops() {
                      pOpKernelType(vartype::INT64, XPUPlace())})},
       {"c_allreduce_sum",
        XPUKernelSet({pOpKernelType(vartype::FP16, XPUPlace()),
-                     pOpKernelType(vartype::FP32, XPUPlace())})},
+                     pOpKernelType(vartype::FP32, XPUPlace()),
+                     pOpKernelType(vartype::INT32, XPUPlace())})},
       {"c_identity",
        XPUKernelSet({pOpKernelType(vartype::FP16, XPUPlace()),
                      pOpKernelType(vartype::FP32, XPUPlace()),
                      pOpKernelType(vartype::FP64, XPUPlace()),
                      pOpKernelType(vartype::INT32, XPUPlace()),
                      pOpKernelType(vartype::INT64, XPUPlace())})},
+      {"c_sync_calc_stream",
+       XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
       {"c_sync_comm_stream",
        XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
       {"cast",