Merge pull request #125 from nikitaxgusev/release/ccl_2021.13

Intel(R) oneAPI Collective Communications Library (oneCCL) 2021.13

Author: nikitaxgusev

CMakeLists.txt

@@ -272,6 +272,7 @@ endif()
 
 set_lp_env()
 set_avx_env()
+set_sycl_env()
 
 set(CCL_BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR}/src)
 
@@ -307,7 +308,7 @@ file(GLOB spv_kernels "${PROJECT_SOURCE_DIR}/src/kernels/kernels.spv")
 endif()
 
 set(CCL_MAJOR_VERSION     "2021")
-set(CCL_MINOR_VERSION     "12")
+set(CCL_MINOR_VERSION     "13")
 set(CCL_UPDATE_VERSION    "0")
 set(CCL_PRODUCT_STATUS    "Gold")
 string(TIMESTAMP CCL_PRODUCT_BUILD_DATE "%Y-%m-%dT %H:%M:%SZ")

README.md

@@ -147,6 +147,13 @@ To generate oneCCLConfig files for oneCCL package, use the provided [`cmake/scri
 cmake [-DOUTPUT_DIR=<output_dir>] -P cmake/script/config_generation.cmake
 ```
 
+### OS File Descriptors 
+
+oneCCL uses [Level Zero IPC handles](https://spec.oneapi.io/level-zero/latest/core/PROG.html#memory-1) so that a process can access a memory allocation done by a different process. 
+However, these IPC handles consume OS File Descriptors (FDs). As a result, to avoid running out of OS FDs, we recommend to increase the default limit of FDs in the system for applications running with oneCCL and GPU buffers. 
+
+The number of FDs required is application-dependent, but the recommended limit is ``1048575``. This value can be modified with the ulimit command.  
+
 ## Additional Resources
 
 ### Blog Posts

cmake/helpers.cmake

@@ -63,9 +63,14 @@ function(set_lp_env)
     endif()
     message(STATUS "BF16 GPU truncate: ${CCL_BF16_GPU_TRUNCATE}")
 
-
     set(GCC_FP16_MIN_SUPPORTED "4.9.0")
+    set(GCC_FP16_AVX512FP16_MIN_SUPPORTED "12.0.0")
+    set(GCC_FP16_AVX512FP16_BINUTILS_MIN_SUPPORTED "2.38")
+
+    set(ICX_FP16_AVX512FP16_MIN_SUPPORTED "2021.4.0")
+
     set(CLANG_FP16_MIN_SUPPORTED "9.0.0")
+    set(CLANG_FP16_AVX512FP16_MIN_SUPPORTED "14.0.0")
 
     if (${CMAKE_C_COMPILER_ID} STREQUAL "Intel"
         OR (${CMAKE_C_COMPILER_ID} STREQUAL "IntelLLVM")
@@ -81,6 +86,21 @@ function(set_lp_env)
     endif()
     message(STATUS "FP16 compiler: ${CCL_FP16_COMPILER}")
 
+    if ((${CMAKE_C_COMPILER_ID} STREQUAL "IntelLLVM"
+            AND NOT ${CMAKE_C_COMPILER_VERSION} VERSION_LESS ${ICX_FP16_AVX512FP16_MIN_SUPPORTED})
+        OR (${CMAKE_C_COMPILER_ID} STREQUAL "Clang"
+            AND NOT ${CMAKE_C_COMPILER_VERSION} VERSION_LESS ${CLANG_FP16_AVX512FP16_MIN_SUPPORTED})
+        OR (${CMAKE_C_COMPILER_ID} STREQUAL "GNU"
+            AND NOT ${CMAKE_C_COMPILER_VERSION} VERSION_LESS ${GCC_FP16_AVX512FP16_MIN_SUPPORTED}
+	    AND NOT ${BINUTILS_VERSION} VERSION_LESS ${GCC_FP16_AVX512FP16_BINUTILS_MIN_SUPPORTED})
+        )
+        add_definitions(-DCCL_FP16_AVX512FP16_COMPILER)
+        set(CCL_FP16_AVX512FP16_COMPILER ON)
+    else()
+        set(CCL_FP16_AVX512FP16_COMPILER OFF)
+    endif()
+    message(STATUS "FP16 AVX512FP16 compiler: ${CCL_FP16_AVX512FP16_COMPILER}")
+
     if (CCL_FP16_COMPILER)
         if ((${CMAKE_C_COMPILER_ID} STREQUAL "Clang" OR ${CMAKE_C_COMPILER_ID} STREQUAL "IntelLLVM"
         OR ${CMAKE_C_COMPILER_ID} STREQUAL "GNU"))
@@ -102,6 +122,28 @@ function(set_lp_env)
 
 endfunction(set_lp_env)
 
+function(set_sycl_env)
+    set(ICX_SYCL_VEC_BF16_MIN_SUPPORTED "2024.2.0")
+
+    if (CCL_ENABLE_SYCL
+    AND ${CMAKE_C_COMPILER_ID} STREQUAL "IntelLLVM"
+    AND NOT ${CMAKE_C_COMPILER_VERSION} VERSION_LESS ${ICX_SYCL_VEC_BF16_MIN_SUPPORTED})
+        add_definitions(-DCCL_SYCL_VEC_SUPPORT_BF16)
+	set(CCL_SYCL_VEC_SUPPORT_BF16 ON)
+    else()
+        set(CCL_SYCL_VEC_SUPPORT_BF16 OFF)
+    endif()
+
+    if (CCL_ENABLE_SYCL
+    AND ${CMAKE_C_COMPILER_ID} STREQUAL "Clang")
+        set(CCL_SYCL_VEC_SUPPORT_FP16 OFF)
+    else()
+        add_definitions(-DCCL_SYCL_VEC_SUPPORT_FP16)
+	set(CCL_SYCL_VEC_SUPPORT_FP16 ON)
+    endif()
+
+endfunction(set_sycl_env)
+
 function(set_avx_env)
 
     set(GCC_AVX_MIN_SUPPORTED "4.9.0")
@@ -278,6 +320,8 @@ function(set_compute_backend COMMON_CMAKE_DIR)
         set(CCL_ENABLE_ZE ON PARENT_SCOPE)
         message(STATUS "Enable CCL Level Zero support")
 
+        set (CMAKE_CXX_FLAGS "-Wno-c++20-extensions" PARENT_SCOPE)
+
         execute_process(COMMAND icpx -v
             OUTPUT_VARIABLE ICPX_VERSION
             ERROR_VARIABLE ICPX_VERSION

cmake/vars.sh.in

@@ -119,7 +119,7 @@ args=$*
 for arg in $args
 do
     case "$arg" in
-        --ccl-bundled-mpi=*)
+        --ccl-bundled-mpi=*|-ccl-bundled-mpi=*)
             ccl_bundled_mpi="${arg#*=}"
             ;;
     esac