add opencl ci

ec3b976d · ZhenWang · Zhen Wang · 9708152e · ec3b976d · ec3b976d
5 changed file
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -87,6 +87,25 @@ build:mobile_android:
    dependencies:
        - build:server

+build:mobile_android_cl:
+    tags:
+        - lite
+    stage: build_mobile
+    image: $MOBILE_LITE_DOCKER_IMAGE
+    cache:
+        key: mobile_thirdparty
+        paths:
+            - build.lite.android.armv8.gcc.cl_enable/third_party
+            - build.lite.android.armv7.gcc.cl_enable/third_party
+            - ~/.ccache
+            - $CI_PROJECT_DIR/build_mobile_ccache_cl
+    script:
+        - export CCACHE_DIR=$CI_PROJECT_DIR/build_mobile_ccache_cl
+        - ./paddle/fluid/lite/tools/build.sh build_test_arm_opencl
+
+    dependencies:
+        - build:server
+
 build:mobile_armlinux:
    tags:
        - lite

--- a/paddle/fluid/lite/api/CMakeLists.txt
+++ b/paddle/fluid/lite/api/CMakeLists.txt
@@ -97,10 +97,12 @@ endif()
 # TODO(Superjomn) support latter.
 lite_cc_test(test_light_api_lite SRCS light_api_test.cc
    DEPS light_api_lite program_lite mir_passes
+    CL_DEPS ${opencl_kernels}
    ARGS --optimized_model=${LITE_MODEL_DIR}/lite_naive_model_opt SERIAL)

 lite_cc_test(test_apis_lite SRCS apis_test.cc
  DEPS cxx_api_lite light_api_lite ${ops_lite}
+  CL_DEPS ${opencl_kernels}
  X86_DEPS ${x86_kernels} operator
  ARGS --model_dir=${LITE_MODEL_DIR}/lite_naive_model
       --optimized_model=${LITE_MODEL_DIR}/lite_naive_model_opt SERIAL)
@@ -125,6 +127,7 @@ lite_cc_test(test_paddle_api_lite SRCS paddle_api_test.cc DEPS paddle_api_full p
  ${ops_lite}
  ARM_DEPS ${arm_kernels}
  X86_DEPS ${x86_kernels}
+  CL_DEPS ${opencl_kernels}
  ARGS --model_dir=${LITE_MODEL_DIR}/lite_naive_model SERIAL)

 lite_cc_test(test_model_bin SRCS model_test.cc DEPS paddle_api_full paddle_api_light
@@ -152,5 +155,8 @@ lite_cc_binary(cxx_api_lite_bin SRCS cxx_api_bin_int8.cc
    target_wrapper_host
    mir_passes
    ${ops_lite} ${host_kernels}
+    CL_DEPS ${opencl_kernels}
    ARM_DEPS ${arm_kernels})
-lite_cc_binary(model_optimize_tool SRCS model_optimize_tool.cc DEPS paddle_api_full)
+lite_cc_binary(model_optimize_tool SRCS model_optimize_tool.cc 
+    DEPS paddle_api_full
+    CL_DEPS ${opencl_kernels})
--- a/paddle/fluid/lite/core/CMakeLists.txt
+++ b/paddle/fluid/lite/core/CMakeLists.txt
@@ -19,12 +19,12 @@ endif()

 proto_library(framework_proto_lite SRCS framework.proto)

-lite_cc_library(kernel_lite SRCS kernel.cc DEPS type_system target_wrapper_lite any_lite op_params_lite framework_proto_lite ${tensor_lite})
 lite_cc_library(variable_lite SRCS variable.cc)
 lite_cc_library(op_registry_lite SRCS op_registry.cc DEPS framework_proto_lite)
 lite_cc_library(scope_lite SRCS scope.cc DEPS ${tensor_lite})
 lite_cc_library(cpu_info_lite SRCS cpu_info.cc)
 lite_cc_library(context_lite SRCS context.cc DEPS ${tensor_lite} any_lite cpu_info_lite X86_DEPS eigen3 CL_DEPS cl_helper)
+lite_cc_library(kernel_lite SRCS kernel.cc DEPS context_lite type_system target_wrapper_lite any_lite op_params_lite framework_proto_lite ${tensor_lite})
 lite_cc_library(op_lite SRCS op_lite.cc DEPS scope_lite op_registry_lite target_wrapper_lite kernel_lite
  cpp_op_desc_lite ${tensor_lite})
 lite_cc_library(types_lite SRCS types.cc)

--- a/paddle/fluid/lite/tools/build.sh
+++ b/paddle/fluid/lite/tools/build.sh
@@ -33,23 +33,6 @@ function cmake_x86 {
    cmake ..  -DWITH_GPU=OFF -DWITH_MKLDNN=OFF -DLITE_WITH_X86=ON ${common_flags}
 }

-function cmake_opencl {
-    # $1: ARM_TARGET_OS in "android" , "armlinux"
-    # $2: ARM_TARGET_ARCH_ABI in "arm64-v8a", "armeabi-v7a" ,"armeabi-v7a-hf"
-    cmake .. \
-        -DLITE_WITH_OPENCL=ON \
-        -DWITH_GPU=OFF \
-        -DWITH_MKL=OFF \
-        -DWITH_LITE=ON \
-        -DLITE_WITH_CUDA=OFF \
-        -DLITE_WITH_X86=OFF \
-        -DLITE_WITH_ARM=ON \
-        -DLITE_WITH_LIGHT_WEIGHT_FRAMEWORK=ON \
-        -DWITH_TESTING=ON \
-        -DARM_TARGET_OS=$1 -DARM_TARGET_ARCH_ABI=$2
-}
-
-
 # This method is only called in CI.
 function cmake_x86_for_CI {
    prepare_workspace # fake an empty __generated_code__.cc to pass cmake.
@@ -221,25 +204,48 @@ function cmake_arm {
    # $1: ARM_TARGET_OS in "android" , "armlinux"
    # $2: ARM_TARGET_ARCH_ABI in "armv8", "armv7" ,"armv7hf"
    # $3: ARM_TARGET_LANG in "gcc" "clang"
-    cmake .. \
-        -DWITH_GPU=OFF \
-        -DWITH_MKL=OFF \
-        -DWITH_LITE=ON \
-        -DLITE_WITH_CUDA=OFF \
-        -DLITE_WITH_X86=OFF \
-        -DLITE_WITH_ARM=ON \
-        -DLITE_WITH_LIGHT_WEIGHT_FRAMEWORK=ON \
-        -DWITH_TESTING=ON \
-        -DARM_TARGET_OS=$1 -DARM_TARGET_ARCH_ABI=$2 -DARM_TARGET_LANG=$3
+    # $4: LITE_WITH_OPENCL in ON OFF
+    os=$1
+    abi=$2
+    lang=$3
+    opencl=$4
+
+    if [[ ${opencl} == "cl_enable" ]]; then
+        echo "-- Enable OpenCL --"
+        cmake .. \
+            -DLITE_WITH_OPENCL=ON \
+            -DWITH_GPU=OFF \
+            -DWITH_MKL=OFF \
+            -DWITH_LITE=ON \
+            -DLITE_WITH_CUDA=OFF \
+            -DLITE_WITH_X86=OFF \
+            -DLITE_WITH_ARM=ON \
+            -DLITE_WITH_LIGHT_WEIGHT_FRAMEWORK=ON \
+            -DWITH_TESTING=ON \
+            -DARM_TARGET_OS=${os} -DARM_TARGET_ARCH_ABI=${abi} -DARM_TARGET_LANG=${lang}
+    else
+        cmake .. \
+            -DWITH_GPU=OFF \
+            -DWITH_MKL=OFF \
+            -DWITH_LITE=ON \
+            -DLITE_WITH_CUDA=OFF \
+            -DLITE_WITH_X86=OFF \
+            -DLITE_WITH_ARM=ON \
+            -DLITE_WITH_LIGHT_WEIGHT_FRAMEWORK=ON \
+            -DWITH_TESTING=ON \
+            -DARM_TARGET_OS=${os} -DARM_TARGET_ARCH_ABI=${abi} -DARM_TARGET_LANG=${lang}
+    fi
 }

 # $1: ARM_TARGET_OS in "android" , "armlinux"
 # $2: ARM_TARGET_ARCH_ABI in "armv8", "armv7" ,"armv7hf"
 # $3: ARM_TARGET_LANG in "gcc" "clang"
+# $4: LITE_WITH_OPENCL in ON OFF
 function build_arm {
    os=$1
    abi=$2
    lang=$3
+    opencl=$4

    cur_dir=$(pwd)
    if [[ ${os} == "armlinux" ]]; then
@@ -263,11 +269,11 @@ function build_arm {
        return 0
    fi

-    build_dir=$cur_dir/build.lite.${os}.${abi}.${lang}
+    build_dir=$cur_dir/build.lite.${os}.${abi}.${lang}.${opencl}
    mkdir -p $build_dir
    cd $build_dir

-    cmake_arm ${os} ${abi} ${lang}
+    cmake_arm ${os} ${abi} ${lang} ${opencl}
    build $TESTS_FILE

    # test publish inference lib
@@ -331,6 +337,21 @@ function arm_push_necessary_file {
    adb -s emulator-${port} push ${testpath} ${adb_work_dir}
 }

+function build_test_arm_opencl {
+    ########################################################################
+    # job 1-4 must be in one runner
+    cur=$PWD
+
+    # job 5
+    build_arm "android" "armv8" "gcc" "cl_enable"
+    cd $cur
+
+    # job 6
+    build_arm "android" "armv7" "gcc" "cl_enable"
+    cd $cur
+
+    echo "Done"
+}

 # We split the arm unittest into several sub-tasks to parallel and reduce the overall CI timetime.
 # sub-task1
@@ -378,17 +399,17 @@ function build_test_arm_subtask_armlinux {
    cur=$PWD

    # job 5
-    build_arm "armlinux" "armv8" "gcc" $port_armv8
+    build_arm "armlinux" "armv8" "gcc"
    test_arm "armlinux" "armv8" "gcc" $port_armv8
    cd $cur

    # job 6
-    build_arm "armlinux" "armv7" "gcc" $port_armv8
+    build_arm "armlinux" "armv7" "gcc"
    test_arm "armlinux" "armv7" "gcc" $port_armv8
    cd $cur

    # job 7
-    build_arm "armlinux" "armv7hf" "gcc" $port_armv8
+    build_arm "armlinux" "armv7hf" "gcc"
    test_arm "armlinux" "armv7hf" "gcc" $port_armv8
    cd $cur

@@ -498,6 +519,11 @@ function main {
                ARM_PORT="${i#*=}"
                shift
                ;;
+            --opencl=*)
+                OPENCL="${i#*=}"
+                shift
+                ;;
+
            build)
                build $TESTS_FILE
                build $LIBS_FILE
@@ -520,11 +546,11 @@ function main {
                shift
                ;;
            cmake_arm)
-                cmake_arm $ARM_OS $ARM_ABI $ARM_LANG
+                cmake_arm $ARM_OS $ARM_ABI $ARM_LANG $OPENCL
                shift
                ;;
            build_arm)
-                build_arm $ARM_OS $ARM_ABI $ARM_LANG
+                build_arm $ARM_OS $ARM_ABI $ARM_LANG $OPENCL
                shift
                ;;
            test_server)
@@ -532,7 +558,7 @@ function main {
                shift
                ;;
            test_arm)
-                build_arm $ARM_OS $ARM_ABI $ARM_LANG $ARM_PORT
+                test_arm $ARM_OS $ARM_ABI $ARM_LANG $ARM_PORT
                shift
                ;;
            test_arm_android)
@@ -547,6 +573,10 @@ function main {
                build_test_arm
                shift
                ;;
+            build_test_arm_opencl)
+                build_test_arm_opencl
+                shift
+                ;;
            build_test_arm_subtask_android)
                build_test_arm_subtask_android
                shift

--- a/paddle/fluid/lite/tools/debug/CMakeLists.txt
+++ b/paddle/fluid/lite/tools/debug/CMakeLists.txt
@@ -9,4 +9,5 @@ lite_cc_binary(lite_model_debug_tool SRCS model_debug_tool.cc
    mir_passes
    ${ops_lite} ${host_kernels}
    X86_DEPS ${x86_kernels}
-    ARM_DEPS ${arm_kernels})
+    ARM_DEPS ${arm_kernels}
+    CL_DEPS ${opencl_kernels})