Initial input for backend selection

.bazelrc

@@ -4,6 +4,7 @@ build -c opt \
       --incompatible_require_linker_input_cc_api
 
 # Aliases for user-defined flags
+build --flag_alias=backend_config=@config//:backend_config
 build --flag_alias=test_link_mode=@config//:test_link_mode
 build --flag_alias=test_thread_mode=@config//:test_thread_mode
 build --flag_alias=test_external_datasets=@config//:test_external_datasets

.ci/env/openblas.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+#===============================================================================
+# Copyright 2023 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#===============================================================================
+
+sudo apt-get update
+sudo apt-get install build-essential gcc gfortran
+git clone https://github.com/xianyi/OpenBLAS.git
+CoreCount=$(lscpu -p | grep -Ev '^#' | wc -l)
+pushd OpenBLAS
+  make clean
+  make -j${CoreCount} NO_FORTRAN=1
+  make install PREFIX=../__deps/open_blas
+popd

.ci/pipeline/ci.yml

@@ -37,7 +37,7 @@ jobs:
       displayName: 'clang-format check'
       failOnStderr: true
 
-- job: 'LinuxMakeGNU'
+- job: 'LinuxMakeGNU_MKL'
   timeoutInMinutes: 0
   variables:
     release.dir: '__release_lnx_gnu'
@@ -89,6 +89,50 @@ jobs:
     condition: failed()
     continueOnError: true
 
+- job: 'LinuxMakeGNU_OpenBLAS'
+  timeoutInMinutes: 0
+  variables:
+    release.dir: '__release_lnx_gnu'
+    platform.type : 'lnx32e'
+  pool:
+    vmImage: 'ubuntu-22.04'
+  steps:
+  - script: |
+      sudo apt-get update && sudo apt-get install -y gcc-multilib g++-multilib dos2unix tree
+      conda create -n ci-env -q -y -c intel -c conda-forge impi-devel=2021.8.0 openjdk=17.0.3
+    displayName: 'apt-get and conda install'
+  - script: |
+      .ci/scripts/describe_system.sh
+    displayName: 'System info'
+  - script: |
+      .ci/scripts/build.sh --compiler gnu --optimizations avx2 --target daal --backend_config ref --conda-env ci-env
+    displayName: 'make daal'
+  - script: |
+      .ci/scripts/build.sh --compiler gnu --optimizations avx2 --target oneapi_c --backend_config ref
+    displayName: 'make oneapi_c'
+  - task: PublishPipelineArtifact@1
+    inputs:
+      artifactName: '$(platform.type) OpenBLAS build'
+      targetPath: '$(Build.Repository.LocalPath)/$(release.dir)'
+    displayName: 'Upload build artifacts'
+    continueOnError: true
+  - script: |
+      .ci/scripts/test.sh --test-kind examples --build-dir $(release.dir) --compiler gnu --interface daal/java --conda-env ci-env
+    displayName: 'daal/java examples'
+  - script: |
+      .ci/scripts/test.sh --test-kind examples --build-dir $(release.dir) --compiler gnu --interface daal/cpp --build_system cmake --backend ref
+    displayName: 'daal/cpp examples'
+  - script: |
+      .ci/scripts/test.sh --test-kind examples --build-dir $(release.dir) --compiler gnu --interface oneapi/cpp --build_system cmake --backend ref
+    displayName: 'oneapi/cpp examples'
+  - task: PublishPipelineArtifact@1
+    inputs:
+      artifactName: '$(platform.type) fail'
+      targetPath: '$(Build.Repository.LocalPath)/$(release.dir)'
+    displayName: 'Uploading on fail'
+    condition: failed()
+    continueOnError: true
+
 - job: 'LinuxMakeDPCPP'
   timeoutInMinutes: 0
   variables:
@@ -259,7 +303,7 @@ jobs:
     displayName: 'bazel-cache-limit'
 
 - job: LinuxDaal4py
-  dependsOn: LinuxMakeGNU
+  dependsOn: LinuxMakeGNU_MKL
   timeoutInMinutes: 0
   variables:
     release.dir: '__release_lnx_gnu'

.ci/scripts/build.sh

@@ -28,6 +28,9 @@ while [[ $# -gt 0 ]]; do
         --target)
         target="$2"
         ;;
+        --backend_config)
+        backend_config="$2"
+        ;;
         --conda-env)
         conda_env="$2"
         ;;
@@ -45,6 +48,7 @@ OS=${PLATFORM::3}
 ARCH=${PLATFORM:3:3}
 
 optimizations=${optimizations:-avx2}
+backend_config=${backend_config:-mkl}
 GLOBAL_RETURN=0
 
 if [ "${OS}" == "lnx" ]; then
@@ -83,16 +87,27 @@ else
 fi
 
 #main actions
-echo "Call mkl and tbb scripts"
-$(pwd)/dev/download_micromkl.sh with_gpu=${with_gpu}
+echo "Call env scripts"
+if [ "${backend_config}" == "mkl" ]; then
+    echo "Sourcing MKL env"
+    $(pwd)/dev/download_micromkl.sh with_gpu=${with_gpu}
+elif [ "${backend_config}" == "ref" ]; then
+    echo "Sourcing ref(openblas) env"
+    if [ ! -d "__deps/open_blas" ]; then
+        $(pwd)/.ci/env/openblas.sh
+    fi
+else
+    echo "Not supported backend env"
+fi
 $(pwd)/dev/download_tbb.sh
 echo "Set Java PATH and CPATH from JAVA_HOME=${JAVA_HOME}"
 export PATH=$JAVA_HOME/bin:$PATH
 export CPATH=$JAVA_HOME/include:$JAVA_HOME/include/${java_os_name}:$CPATH
 echo "Calling make"
 make ${target:-daal_c} ${make_op} \
     COMPILER=${compiler} \
-    REQCPU="${optimizations}"
+    REQCPU="${optimizations}" \
+    BACKEND_CONFIG="${backend_config}"
 err=$?
 
 if [ ${err} -ne 0 ]; then

.ci/scripts/test.sh

@@ -37,6 +37,9 @@ while [[ $# -gt 0 ]]; do
         --build_system)
         build_system="$2"
         ;;
+        --backend)
+        backend="$2"
+        ;;
         *)
         echo "Unknown option: $1"
         exit 1
@@ -53,6 +56,7 @@ OS=${PLATFORM::3}
 ARCH=${PLATFORM:3:3}
 full_arch=intel64
 build_system=${build_system:-cmake}
+backend=${backend:-mkl}
 
 if [ "${OS}" == "lnx" ]; then
     source /usr/share/miniconda/etc/profile.d/conda.sh
@@ -150,7 +154,12 @@ for link_mode in ${link_modes}; do
             mkdir Build
         fi
 
-        cmake -B Build -S . -G "Unix Makefiles" -DONEDAL_LINK=${link_mode} -DTBB_DIR=${TBBROOT}/lib/cmake/tbb
+        ref_backend="OFF"
+        if [ "${backend}" == "ref" ]; then
+            ref_backend="ON"
+        fi
+
+        cmake -B Build -S . -G "Unix Makefiles" -DONEDAL_LINK=${link_mode} -DTBB_DIR=${TBBROOT}/lib/cmake/tbb -DREF_BACKEND=${ref_backend}
         err=$?
         if [ ${err} -ne 0 ]; then
             echo -e "$(date +'%H:%M:%S') CMAKE GENERATE FAILED\t\t"

WORKSPACE

@@ -45,6 +45,12 @@ micromkl_dpc_repo(
     sha256 = "1bd9e3ef850d95d1ee00e0f04943c8ed2490175fca6a7b331cab91a124ab301e",
 )
 
+load("@onedal//dev/bazel/deps:openblas.bzl", "openblas_repo")
+openblas_repo(
+    name = "openblas",
+    root_env_var = "OPENBLASROOT",
+)
+
 load("@onedal//dev/bazel/deps:tbb.bzl", "tbb_repo")
 tbb_repo(
     name = "tbb",

cpp/daal/BUILD

@@ -11,20 +11,26 @@ load("@onedal//dev/bazel:daal.bzl",
 daal_module(
     name = "microvmlipp",
     override_deps_lib_tag = True,
-    deps = [
-        "@micromkl//:vml_ipp",
-        # TODO: Currently vml_ipp lib depends on TBB, but it shouldn't
-        #       Remove TBB from deps once problem with vml_ipp is resolved
-        "@tbb//:tbb_binary",
-    ],
+    deps = select({
+        "@config//:backend_ref": [ ],
+        "//conditions:default": [
+                                    "@micromkl//:vml_ipp",
+                                    # TODO: Currently vml_ipp lib depends on TBB, but it shouldn't
+                                    #       Remove TBB from deps once problem with vml_ipp is resolved
+                                    "@tbb//:tbb_binary",
+                                ],
+        }),
 )
 
 daal_module(
-    name = "micromkl_thread",
+    name = "mathbackend_thread",
     override_deps_lib_tag = True,
-    deps = [
-        "@micromkl//:mkl_thr",
-    ],
+    deps = select({
+        "@config//:backend_ref": [ "@openblas//:openblas",
+                                 ],
+        "//conditions:default": [ "@micromkl//:mkl_thr", 
+                                ],
+        }),
 )
 
 daal_patch_kernel_defines(
@@ -40,19 +46,24 @@ daal_module(
 
 daal_module(
     name = "includes",
-    hdrs = [
-        ":kernel_defines",
+    hdrs = [ ":kernel_defines",
     ],
     includes = [ "." ],
     defines = [
         "DAAL_NOTHROW_EXCEPTIONS",
         "DAAL_HIDE_DEPRECATED",
     ],
-    deps = [
-        ":public_includes",
-        "@micromkl//:headers",
-        "@micromkl_dpc//:headers",
-    ],
+    deps = select({
+        "@config//:backend_ref": [ 
+                                   ":public_includes",
+                                   "@openblas//:headers",
+                                 ],
+        "//conditions:default": [
+                                  ":public_includes",
+                                  "@micromkl//:headers",
+                                  "@micromkl_dpc//:headers",
+                                ],
+        }),
 )
 
 daal_generate_version(
@@ -111,11 +122,17 @@ daal_module(
     name = "sycl",
     hdrs = glob(["src/sycl/**/*.h", "src/sycl/**/*.cl"]),
     srcs = glob(["src/sycl/**/*.cpp"]),
-    deps = [
-        ":services",
-        "@onedal//cpp/daal/src/algorithms/engines:kernel",
-        "@micromkl_dpc//:headers",
-    ],
+    deps = select({
+            "@config//:backend_ref": [ 
+                                    ":services",
+                                    "@onedal//cpp/daal/src/algorithms/engines:kernel",
+                                        ],
+            "//conditions:default": [ 
+                                    ":services",
+                                    "@onedal//cpp/daal/src/algorithms/engines:kernel",
+                                    "@micromkl_dpc//:headers",
+                                    ],
+        }),
 )
 
 daal_module(
@@ -128,12 +145,20 @@ daal_module(
         "TBB_SUPPRESS_DEPRECATED_MESSAGES",
         "TBB_USE_ASSERT=0",
     ],
-    deps = [
-        ":threading_headers",
-        ":micromkl_thread",
-        "@tbb//:tbb",
-        "@tbb//:tbbmalloc",
-    ],
+    deps = select({
+            "@config//:backend_ref": [ 
+                                          ":threading_headers",
+                                          ":mathbackend_thread",
+                                          "@tbb//:tbb",
+                                          "@tbb//:tbbmalloc",
+                                        ],
+            "//conditions:default": [ 
+                                      ":threading_headers",
+                                      ":mathbackend_thread",
+                                      "@tbb//:tbb",
+                                      "@tbb//:tbbmalloc",
+                                    ],
+        }),
 )
 
 daal_module(
@@ -242,7 +267,10 @@ daal_dynamic_lib(
     deps = [
         ":threading_tbb",
     ],
-    def_file = "src/threading/export_lnx32e.def",
+    def_file = select({
+        "@config//:backend_ref": "src/threading/export_lnx32e.ref.def",
+        "//conditions:default": "src/threading/export_lnx32e.mkl.def", 
+        }),
 )
 
 daal_module(

cpp/daal/include/services/daal_defines.h

@@ -491,13 +491,30 @@ const int SERIALIZATION_DBSCAN_DISTRIBUTED_PARTIAL_RESULT_STEP13_ID = 121310;
         }                                                                                         \
     }
 
+#define DAAL_OVERFLOW_CHECK_BY_MULTIPLICATION_THROW_IF_POSSIBLE(type, op1, op2)                                       \
+    {                                                                                                                 \
+        if (!(0 == (op1)) && !(0 == (op2)))                                                                           \
+        {                                                                                                             \
+            volatile type r = (op1) * (op2);                                                                          \
+            r /= (op1);                                                                                               \
+            if (!(r == (op2))) services::throwIfPossible(services::Status(services::ErrorBufferSizeIntegerOverflow)); \
+        }                                                                                                             \
+    }
+
 #define DAAL_OVERFLOW_CHECK_BY_ADDING(type, op1, op2)                                         \
     {                                                                                         \
         volatile type r = (op1) + (op2);                                                      \
         r -= (op1);                                                                           \
         if (!(r == (op2))) return services::Status(services::ErrorBufferSizeIntegerOverflow); \
     }
 
+#define DAAL_OVERFLOW_CHECK_BY_ADDING_THROW_IF_POSSIBLE(type, op1, op2)                                           \
+    {                                                                                                             \
+        volatile type r = (op1) + (op2);                                                                          \
+        r -= (op1);                                                                                               \
+        if (!(r == (op2))) services::throwIfPossible(services::Status(services::ErrorBufferSizeIntegerOverflow)); \
+    }
+
 #define DAAL_CHECK_STATUS_RETURN_IF_FAIL(statVal, returnObj) \
     {                                                        \
         if (!(statVal)) return returnObj;                    \

cpp/daal/src/algorithms/adaboost/adaboost_predict_impl.i

@@ -146,7 +146,7 @@ services::Status AdaBoostPredictKernel<method, algorithmFPType, cpu>::processBlo
         }
     }
 
-    Math<algorithmFPType, cpu>::vLog(nRowsInCurrentBlock * nClasses, p_block, pLog); // inplace
+    MathInst<algorithmFPType, cpu>::vLog(nRowsInCurrentBlock * nClasses, p_block, pLog); // inplace
 
     service_memset<algorithmFPType, cpu>(pSumLog, 0.0, nRowsInCurrentBlock);
 

cpp/daal/src/algorithms/adaboost/adaboost_train_impl.i

@@ -181,14 +181,14 @@ services::Status AdaBoostTrainKernel<method, algorithmFPType, cpu>::adaboostSAMM
         }
 
         algorithmFPType cM =
-            learningRate * (Math<algorithmFPType, cpu>::sLog((one - errM) / errM) + Math<algorithmFPType, cpu>::sLog(nClasses - one));
+            learningRate * (MathInst<algorithmFPType, cpu>::sLog((one - errM) / errM) + MathInst<algorithmFPType, cpu>::sLog(nClasses - one));
 
         /* Update weights */
         for (size_t i = 0; i < nVectors; i++)
         {
             errFlag[i] *= cM;
         }
-        Math<algorithmFPType, cpu>::vExp(nVectors, errFlag, errFlag);
+        MathInst<algorithmFPType, cpu>::vExp(nVectors, errFlag, errFlag);
         algorithmFPType wSum = zero;
         for (size_t i = 0; i < nVectors; i++)
         {
@@ -337,7 +337,7 @@ services::Status AdaBoostTrainKernel<method, algorithmFPType, cpu>::adaboostSAMM
             }
             t[i] *= scaling;
         }
-        Math<algorithmFPType, cpu>::vExp(nVectors, t, t);
+        MathInst<algorithmFPType, cpu>::vExp(nVectors, t, t);
         for (size_t i = 0; i < nVectors; i++)
         {
             w[i] *= t[i];