refine and pass mul test

paddle/fluid/lite/core/cpu_info.cc

@@ -54,15 +54,15 @@ void DeviceInfo::InitInternal(DeviceInfo* dev) {
               << ", cluster ID: " << dev->cluster_ids_[dev->core_ids_[i]]
               << ", CPU ARCH: A" << dev->archs_[i];
   }
-  LOG(INFO) << "L1 DataCache size is: ";
+  VLOG(1) << "L1 DataCache size is: ";
   for (int i = 0; i < dev->compute_core_num_; ++i) {
-    LOG(INFO) << dev->L1_cache_[i] / 1024 << " KB";
+    VLOG(1) << dev->L1_cache_[i] / 1024 << " KB";
   }
-  LOG(INFO) << "L2 Cache size is: ";
+  VLOG(1) << "L2 Cache size is: ";
   for (int i = 0; i < dev->compute_core_num_; ++i) {
-    LOG(INFO) << dev->L2_cache_[i] / 1024 << " KB";
+    VLOG(1) << dev->L2_cache_[i] / 1024 << " KB";
   }
-  LOG(INFO) << "Total memory: " << dev->max_memory_ << "KB";
+  VLOG(1) << "Total memory: " << dev->max_memory_ << "KB";
 
   dev->max_freq_ = max_freq[0];
   for (int j = 1; j < dev->compute_core_num_; ++j) {

paddle/fluid/lite/kernels/arm/fc_compute.cc

@@ -52,7 +52,6 @@ void FcCompute::Run() {
                               &ctx);
     lite::arm::math::sgemm_prepack(packed_in, w_data, b_data, o_data, x_h, n,
                                    x_w, false, false, false, &ctx);
-
     if (param.bias) {
       CHECK_EQ(param.bias->numel(), n);
       lite::arm::math::fill_bias_fc(o_data, b_data, x_h, n);

paddle/fluid/lite/kernels/arm/mul_compute.cc

@@ -33,7 +33,7 @@ void MulCompute::Run() {
   const auto* y_data = param.y->data<float>();
   auto* o_data = param.output->mutable_data<float>();
 
-  int x_h = static_cast<int>(
+  int m = static_cast<int>(
       param.x->dims().Slice(0, param.x_num_col_dims).production());
   int x_w =
       static_cast<int>(param.x->dims()
@@ -41,22 +41,26 @@ void MulCompute::Run() {
                            .production());
   int y_h = static_cast<int>(
       param.y->dims().Slice(0, param.y_num_col_dims).production());
-  int y_w =
+  int n =
       static_cast<int>(param.y->dims()
                            .Slice(param.y_num_col_dims, param.y->dims().size())
                            .production());
 
   CHECK_EQ(x_w, y_h) << "x_w must be equal with y_h";
-  if (y_w == 1 || x_h == 1) {
-    lite::arm::math::sgemv(x_data, y_data, o_data, false, x_h, x_w, false,
-                           nullptr, false);
+  auto k = x_w;
+  if (n == 1) {
+    lite::arm::math::sgemv(x_data, y_data, o_data, false, m, k, false, nullptr,
+                           false);
 
   } else {
     constexpr bool is_tranposed_y = false;
     auto& ctx = this->ctx_->template As<ARMContext>();
 
-    lite::arm::math::sgemm_prepack(x_data, y_data, nullptr, o_data, x_h, y_w,
-                                   x_w, false, false, is_tranposed_y, &ctx);
+    float* packed_x = static_cast<float*>(ctx.workspace_data<float>()) +
+                      ctx.l2_cache_size() / sizeof(float);
+    lite::arm::math::prepackA(packed_x, x_data, k, 0, m, 0, k, false, &ctx);
+    lite::arm::math::sgemm_prepack(packed_x, y_data, nullptr, o_data, m, n, k,
+                                   false, false, is_tranposed_y, &ctx);
   }
 }
 

paddle/fluid/lite/kernels/arm/mul_compute_test.cc

@@ -58,6 +58,7 @@ TEST(mul_arm, compare_test) {
   for (int m : {1, 2, 3, 4}) {
     for (int n : {1, 2, 3, 4}) {
       for (int k : {1, 2, 3, 4}) {
+        VLOG(3) << "m: " << m << ", n: " << n << ", k: " << k;
         lite::Tensor x, y, out, ref;
         x.Resize({m, k});
         y.Resize({k, n});
@@ -71,8 +72,8 @@ TEST(mul_arm, compare_test) {
 
         FillData<T>(x_data, x.dims().production());
         FillData<T>(y_data, y.dims().production());
-        FillData<T>(out_data, out.dims().production());
-        FillData<T>(ref_data, out.dims().production());
+        FillData<T>(out_data, out.dims().production(), 0, 0);
+        FillData<T>(ref_data, out.dims().production(), 0, 0);
 
         MulCompute mul;
         operators::MulParam param;

paddle/fluid/lite/kernels/arm/pool_compute_test.cc

@@ -182,7 +182,7 @@ TEST(pool_arm, compute) {
       for (auto stride : {2}) {
         for (auto pad : {0}) {
           for (auto n : {1, 3, 4, 11}) {
-            for (auto c : {1, 3, 11, 4, 1024}) {
+            for (auto c : {1, 3, 11 /* ,1024 */}) {  // speedup for ci
               for (auto h : {3, 1, 11, 4, 1}) {
                 for (auto w : {1, 3, 4, 12, 1}) {
                   VLOG(3) << "n:" << n << " c:" << c << " h:" << h << " w:" << w

paddle/fluid/lite/kernels/arm/scale_compute_test.cc

@@ -54,6 +54,15 @@ TEST(scale_arm, compute) {
   lite::Tensor output;
   lite::Tensor output_ref;
 
+#if 1  // for ci speedup
+  for (auto n : {1, 3}) {
+    for (auto c : {1, 3}) {
+      for (auto h : {3, 4}) {
+        for (auto w : {4, 3}) {
+          for (auto bias_after_scale : {true, false}) {
+            for (auto s : {-1.0f, 0.13f}) {
+              for (auto b : {-15.f, 0.11234f}) {
+#else
   for (auto n : {1, 3, 4, 11}) {
     for (auto c : {1, 3, 11, 4}) {
       for (auto h : {3, 1, 11, 4}) {
@@ -61,6 +70,8 @@ TEST(scale_arm, compute) {
           for (auto bias_after_scale : {true, false}) {
             for (auto s : {-100.25f, -1.0f, 0.13f, 3840.975f}) {
               for (auto b : {-3075.495f, -15.f, 0.11234f, 128.15f}) {
+#endif
+
                 x.Resize(DDim(std::vector<int64_t>({n, c, h, w})));
                 output.Resize(DDim(std::vector<int64_t>({n, c, h, w})));
                 output_ref.Resize(DDim(std::vector<int64_t>({n, c, h, w})));

paddle/fluid/lite/tools/build.sh

@@ -43,10 +43,14 @@ function cmake_arm {
         -DARM_TARGET_OS=$1 -DARM_TARGET_ARCH_ABI=$2
 }
 
+function build_single {
+        make $1 -j$(expr $(nproc) - 2)
+}
+
 function build {
     file=$1
     for _test in $(cat $file); do
-        make $_test -j$(expr $(nproc) - 2)
+        build_single $_test
     done
 }
 
@@ -63,39 +67,6 @@ function test_lite {
     done
 }
 
-port_armv8=5554
-port_armv7=5556
-
-# Run test on android
-function test_lite_android {
-    local file=$1
-    local adb_abi=$2
-    local port=
-    if [[ ${adb_abi} == "armeabi-v7a" ]]; then
-        port=${port_armv7}
-    fi
-
-    if [[ ${adb_abi} == "arm64-v8a" ]]; then
-        port=${port_armv8}
-    fi
-    if [[ "${port}x" == "x" ]]; then
-        echo "Port can not be empty"
-        exit 1
-    fi
-
-    echo "file: ${file}"
-    # push all to adb and test
-    adb_work_dir="/data/local/tmp"
-    skip_list="test_model_parser_lite"
-    for _test in $(cat $file); do
-        [[ $skip_list =~ (^|[[:space:]])$_test($|[[:space:]]) ]] && continue || echo 'skip $_test'
-        testpath=$(find ./paddle/fluid -name ${_test})
-        adb -s emulator-${port} push ${testpath} ${adb_work_dir}
-        adb -s emulator-${port} shell chmod +x "${adb_work_dir}/${_test}"
-        adb -s emulator-${port} shell "./${adb_work_dir}/${_test}"
-    done
-}
-
 # Build the code and run lite server tests. This is executed in the CI system.
 function build_test_server {
     mkdir -p ./build
@@ -108,8 +79,34 @@ function build_test_server {
     build $LIBS_FILE
 }
 
-# Build the code and run lite server tests. This is executed in the CI system.
+# test_arm_android <some_test_name> <adb_port_number>
+function test_arm_android {
+    test_name=$1
+    port=$2
+    if [[ "${test_name}x" == "x" ]]; then
+        echo "test_name can not be empty"
+        exit 1
+    fi
+    if [[ "${port}x" == "x" ]]; then
+        echo "Port can not be empty"
+        exit 1
+    fi
+
+    echo "test name: ${test_name}"
+    adb_work_dir="/data/local/tmp"
+    skip_list="test_model_parser_lite" # add more with space
+    [[ $skip_list =~ (^|[[:space:]])$test_name($|[[:space:]]) ]] && continue || echo 'skip $test_name'
+    testpath=$(find ./paddle/fluid -name ${test_name})
+    adb -s emulator-${port} push ${testpath} ${adb_work_dir}
+    adb -s emulator-${port} shell chmod +x "${adb_work_dir}/${test_name}"
+    adb -s emulator-${port} shell "./${adb_work_dir}/${test_name}"
+}
+
+# Build the code and run lite arm tests. This is executed in the CI system.
 function build_test_arm {
+    port_armv8=5554
+    port_armv7=5556
+
     adb kill-server
     adb devices | grep emulator | cut -f1 | while read line; do adb -s $line emu kill; done
     # start android arm64-v8a armeabi-v7a emulators first
@@ -122,6 +119,7 @@ function build_test_arm {
 
     for os in "android" "armlinux" ; do
         for abi in "arm64-v8a" "armeabi-v7a" "armeabi-v7a-hf" ; do
+            # TODO(TJ): enable compile on v7-hf on andorid and all v7 on armlinux
             if [[ ${abi} == "armeabi-v7a-hf" ]]; then
                 echo "armeabi-v7a-hf is not supported on both android and armlinux"
                 continue
@@ -138,17 +136,30 @@ function build_test_arm {
             cmake_arm ${os} ${abi}
             build $TESTS_FILE
 
+            # armlinux need in another docker
+            # TODO(TJ): enable test with armlinux
             if [[ ${os} == "android" ]]; then
                 adb_abi=${abi}
                 if [[ ${adb_abi} == "armeabi-v7a-hf" ]]; then
                     adb_abi="armeabi-v7a"
                 fi
                 if [[ ${adb_abi} == "armeabi-v7a" ]]; then
-                    # skip v7 tests
+                    # skip all armv7 tests
+                    # TODO(TJ): enable test with armv7
                     continue
                 fi
-                test_lite_android $TESTS_FILE ${adb_abi}
-                # armlinux need in another docker
+                local port=
+                if [[ ${adb_abi} == "armeabi-v7a" ]]; then
+                    port=${port_armv7}
+                fi
+
+                if [[ ${adb_abi} == "arm64-v8a" ]]; then
+                    port=${port_armv8}
+                fi
+                echo "test file: ${TESTS_FILE}"
+                for _test in $(cat $TESTS_FILE); do
+                    test_arm_android $_test $port
+                done
             fi
             cd -
         done
@@ -164,12 +175,13 @@ function print_usage {
     echo "----------------------------------------"
     echo -e "cmake_x86: run cmake with X86 mode"
     echo -e "cmake_cuda: run cmake with CUDA mode"
-    echo -e "cmake_arm: run cmake with ARM mode"
+    echo -e "--arm_os=<os> --arm_abi=<abi> cmake_arm: run cmake with ARM mode"
     echo
     echo -e "build: compile the tests"
+    echo -e "--test_name=<test_name> build_single: compile single test"
     echo
     echo -e "test_server: run server tests"
-    echo -e "test_mobile: run mobile tests"
+    echo -e "--test_name=<test_name> --adb_port_number=<adb_port_number> test_arm_android: run arm test"
     echo "----------------------------------------"
     echo
 }
@@ -182,11 +194,31 @@ function main {
                 TESTS_FILE="${i#*=}"
                 shift
                 ;;
+            --test_name=*)
+                TEST_NAME="${i#*=}"
+                shift
+                ;;
+            --arm_os=*)
+                ARM_OS="${i#*=}"
+                shift
+                ;;
+            --arm_abi=*)
+                ARM_ABI="${i#*=}"
+                shift
+                ;;
+            --arm_port=*)
+                ARM_PORT="${i#*=}"
+                shift
+                ;;
             build)
                 build $TESTS_FILE
                 build $LIBS_FILE
                 shift
                 ;;
+            build_single)
+                build_single $TEST_NAME
+                shift
+                ;;
             cmake_x86)
                 cmake_x86
                 shift
@@ -196,15 +228,15 @@ function main {
                 shift
                 ;;
             cmake_arm)
-                cmake_arm $2 $3
+                cmake_arm $ARM_OS $ARM_ABI
                 shift
                 ;;
             test_server)
                 test_lite $TESTS_FILE
                 shift
                 ;;
-            test_mobile)
-                test_lite $TESTS_FILE
+            test_arm_android)
+                test_arm_android $TEST_NAME $ARM_PORT
                 shift
                 ;;
             build_test_server)
@@ -224,7 +256,5 @@ function main {
     done
 }
 
-print_usage
-
 main $@