From 6064c2f49c057ba401385e8cefe1de73e687b395 Mon Sep 17 00:00:00 2001
From: hong19860320 <9973393+hong19860320@users.noreply.github.com>
Date: Tue, 16 Jun 2020 14:40:13 +0800
Subject: [PATCH] [XPU] Enable XPU CI (#3796)

---
 lite/CMakeLists.txt                               | 4 +++-
 lite/tests/api/CMakeLists.txt                     | 7 +++++--
 lite/tests/api/test_bert_lite_xpu.cc              | 2 +-
 lite/tests/api/test_ernie_lite_xpu.cc             | 2 +-
 lite/tests/kernels/activation_compute_test.cc     | 6 +++---
 lite/tests/kernels/batch_norm_compute_test.cc     | 2 +-
 lite/tests/kernels/cast_compute_test.cc           | 2 +-
 lite/tests/kernels/dropout_compute_test.cc        | 2 +-
 lite/tests/kernels/elementwise_compute_test.cc    | 2 +-
 lite/tests/kernels/gather_compute_test.cc         | 2 +-
 lite/tests/kernels/layer_norm_compute_test.cc     | 2 +-
 lite/tests/kernels/lookup_table_compute_test.cc   | 5 +++--
 lite/tests/kernels/matmul_compute_test.cc         | 4 ++--
 lite/tests/kernels/mul_compute_test.cc            | 2 +-
 lite/tests/kernels/multiclass_nms_compute_test.cc | 2 +-
 lite/tests/kernels/pool_compute_test.cc           | 2 +-
 lite/tests/kernels/reshape_compute_test.cc        | 2 +-
 lite/tests/kernels/scale_compute_test.cc          | 2 +-
 lite/tests/kernels/slice_compute_test.cc          | 2 +-
 lite/tests/kernels/softmax_compute_test.cc        | 2 +-
 lite/tests/kernels/stack_compute_test.cc          | 2 +-
 lite/tests/kernels/transpose_compute_test.cc      | 2 +-
 lite/tests/kernels/yolo_box_compute_test.cc       | 2 +-
 lite/tools/ci_build.sh                            | 2 +-
 24 files changed, 35 insertions(+), 29 deletions(-)

diff --git a/lite/CMakeLists.txt b/lite/CMakeLists.txt
index ff4d00dbb1..e534fdd9f9 100644
--- a/lite/CMakeLists.txt
+++ b/lite/CMakeLists.txt
@@ -48,11 +48,13 @@ if (WITH_TESTING)
     endif()
     if(NOT LITE_WITH_LIGHT_WEIGHT_FRAMEWORK)
         lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "GoogleNet_inference.tar.gz")
-	      lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "mobilenet_v1.tar.gz")
+        lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "mobilenet_v1.tar.gz")
         lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "mobilenet_v2_relu.tar.gz")
         lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "resnet50.tar.gz")
         lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "inception_v4_simple.tar.gz")
         lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "step_rnn.tar.gz")
+        lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "bert.tar.gz")
+        lite_download_and_uncompress(${LITE_MODEL_DIR} ${LITE_URL} "ernie.tar.gz")
     endif()
 endif()
 
diff --git a/lite/tests/api/CMakeLists.txt b/lite/tests/api/CMakeLists.txt
index 810a20abbc..0acc54c270 100644
--- a/lite/tests/api/CMakeLists.txt
+++ b/lite/tests/api/CMakeLists.txt
@@ -3,14 +3,17 @@ if(LITE_WITH_XPU)
       DEPS mir_passes lite_api_test_helper paddle_api_full paddle_api_light gflags utils
       ${ops} ${host_kernels} ${x86_kernels} ${xpu_kernels}
       ARGS --model_dir=${LITE_MODEL_DIR}/resnet50)
+    add_dependencies(test_resnet50_lite_xpu extern_lite_download_resnet50_tar_gz)
     lite_cc_test(test_ernie_lite_xpu SRCS test_ernie_lite_xpu.cc
       DEPS mir_passes lite_api_test_helper paddle_api_full paddle_api_light gflags utils
       ${ops} ${host_kernels} ${x86_kernels} ${xpu_kernels}
-      ARGS --model_dir=${LITE_MODEL_DIR}/resnet50)
+      ARGS --model_dir=${LITE_MODEL_DIR}/ernie)
+    add_dependencies(test_ernie_lite_xpu extern_lite_download_ernie_tar_gz)
     lite_cc_test(test_bert_lite_xpu SRCS test_bert_lite_xpu.cc
       DEPS mir_passes lite_api_test_helper paddle_api_full paddle_api_light gflags utils
       ${ops} ${host_kernels} ${x86_kernels} ${xpu_kernels}
-      ARGS --model_dir=${LITE_MODEL_DIR}/resnet50)
+      ARGS --model_dir=${LITE_MODEL_DIR}/bert)
+    add_dependencies(test_bert_lite_xpu extern_lite_download_bert_tar_gz)
 endif()
 
 if(LITE_WITH_RKNPU)
diff --git a/lite/tests/api/test_bert_lite_xpu.cc b/lite/tests/api/test_bert_lite_xpu.cc
index b3ee9febb3..5d66fd0d54 100644
--- a/lite/tests/api/test_bert_lite_xpu.cc
+++ b/lite/tests/api/test_bert_lite_xpu.cc
@@ -93,7 +93,7 @@ TEST(Ernie, test_ernie_lite_xpu) {
   for (size_t i = 0; i < results.size(); ++i) {
     for (size_t j = 0; j < results[i].size(); ++j) {
       EXPECT_NEAR(
-          out->data<float>()[j + (out->shape()[1] * i)], results[i][j], 1e-5);
+          out->data<float>()[j + (out->shape()[1] * i)], results[i][j], 3e-5);
     }
   }
 }
diff --git a/lite/tests/api/test_ernie_lite_xpu.cc b/lite/tests/api/test_ernie_lite_xpu.cc
index 0b614fec96..b1db9f3536 100644
--- a/lite/tests/api/test_ernie_lite_xpu.cc
+++ b/lite/tests/api/test_ernie_lite_xpu.cc
@@ -93,7 +93,7 @@ TEST(Ernie, test_ernie_lite_xpu) {
   for (size_t i = 0; i < results.size(); ++i) {
     for (size_t j = 0; j < results[i].size(); ++j) {
       EXPECT_NEAR(
-          out->data<float>()[j + (out->shape()[1] * i)], results[i][j], 1e-5);
+          out->data<float>()[j + (out->shape()[1] * i)], results[i][j], 2e-5);
     }
   }
 }
diff --git a/lite/tests/kernels/activation_compute_test.cc b/lite/tests/kernels/activation_compute_test.cc
index 5824ba91c2..a62c698f83 100644
--- a/lite/tests/kernels/activation_compute_test.cc
+++ b/lite/tests/kernels/activation_compute_test.cc
@@ -300,7 +300,7 @@ TEST(Activation_relu, precision) {
   abs_error = 1e-2;  // Using fp16 in NPU
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
@@ -426,7 +426,7 @@ TEST(Activation_tanh, precision) {
   abs_error = 1e-2;  // Using fp16 in NPU
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
@@ -572,7 +572,7 @@ TEST(Activation_gelu, precision) {
   LOG(INFO) << "test gelu op";
   Place place;
   float abs_error = 2e-5;
-#if defined(LITE_WITH_XPU)
+#if defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/batch_norm_compute_test.cc b/lite/tests/kernels/batch_norm_compute_test.cc
index ae65e0e3c3..9674f95d0b 100644
--- a/lite/tests/kernels/batch_norm_compute_test.cc
+++ b/lite/tests/kernels/batch_norm_compute_test.cc
@@ -157,7 +157,7 @@ TEST(BatchNorm, precision) {
   LOG(INFO) << "test BatchNorm op";
   float abs_error = 2e-5;
   Place place;
-#if defined(LITE_WITH_XPU)
+#if defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #elif defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
diff --git a/lite/tests/kernels/cast_compute_test.cc b/lite/tests/kernels/cast_compute_test.cc
index 86331bb8a1..34038dfdc7 100644
--- a/lite/tests/kernels/cast_compute_test.cc
+++ b/lite/tests/kernels/cast_compute_test.cc
@@ -135,7 +135,7 @@ TEST(Cast, precision) {
   float abs_error = 2e-5;
 #if defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/dropout_compute_test.cc b/lite/tests/kernels/dropout_compute_test.cc
index 025f02ce31..c4ecc0cf01 100644
--- a/lite/tests/kernels/dropout_compute_test.cc
+++ b/lite/tests/kernels/dropout_compute_test.cc
@@ -94,7 +94,7 @@ TEST(Dropout, precision) {
 #if defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
   abs_error = 1e-2;  // Using fp16 in NPU
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/elementwise_compute_test.cc b/lite/tests/kernels/elementwise_compute_test.cc
index 505ab72dc1..d91c304ef7 100644
--- a/lite/tests/kernels/elementwise_compute_test.cc
+++ b/lite/tests/kernels/elementwise_compute_test.cc
@@ -228,7 +228,7 @@ TEST(Elementwise, precision) {
   abs_error = 1e-2;  // use fp16 in npu
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/gather_compute_test.cc b/lite/tests/kernels/gather_compute_test.cc
index 4d0ad1ab47..c023a12b0f 100644
--- a/lite/tests/kernels/gather_compute_test.cc
+++ b/lite/tests/kernels/gather_compute_test.cc
@@ -98,7 +98,7 @@ TEST(Gather, precision) {
   abs_error = 1e-2;  // use fp16 in npu
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/layer_norm_compute_test.cc b/lite/tests/kernels/layer_norm_compute_test.cc
index 5ea01a6cca..bd4480b612 100644
--- a/lite/tests/kernels/layer_norm_compute_test.cc
+++ b/lite/tests/kernels/layer_norm_compute_test.cc
@@ -147,7 +147,7 @@ TEST(LayerNorm, precision) {
   LOG(INFO) << "test layer_norm op";
   float abs_error = 2e-5;
   Place place;
-#if defined(LITE_WITH_XPU)
+#if defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #elif defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
diff --git a/lite/tests/kernels/lookup_table_compute_test.cc b/lite/tests/kernels/lookup_table_compute_test.cc
index 988077c6c3..ae39abf1db 100644
--- a/lite/tests/kernels/lookup_table_compute_test.cc
+++ b/lite/tests/kernels/lookup_table_compute_test.cc
@@ -116,7 +116,7 @@ TEST(LookupTable, precision) {
   abs_error = 1e-2;
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
@@ -132,7 +132,8 @@ TEST(LookupTable, precision) {
        std::vector<std::vector<int64_t>>{{5, 2, 3, 1}, {2, 3, 1}, {3, 1}}) {
     for (auto w_dims :
          std::vector<std::vector<int64_t>>{{4, 2}, {6, 8}, {12, 15}}) {
-#if defined(LITE_WITH_XPU) && defined(LITE_WITH_NPU)
+#if (defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)) || \
+    defined(LITE_WITH_NPU)
       for (auto padding_idx :
            std::vector<int64_t>{-1}) {  // Only -1 is supported by XPU or NPU
 #else
diff --git a/lite/tests/kernels/matmul_compute_test.cc b/lite/tests/kernels/matmul_compute_test.cc
index 59b0fde8fd..9799c15622 100644
--- a/lite/tests/kernels/matmul_compute_test.cc
+++ b/lite/tests/kernels/matmul_compute_test.cc
@@ -457,7 +457,7 @@ TEST(Matmul2x2, precision) {
   abs_error = 1e-2;  // use fp16 in npu
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
@@ -489,7 +489,7 @@ TEST(Matmul2x2_y_transpose, precision) {
   abs_error = 1e-2;  // use fp16 in npu
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/mul_compute_test.cc b/lite/tests/kernels/mul_compute_test.cc
index d070292332..d89b356935 100644
--- a/lite/tests/kernels/mul_compute_test.cc
+++ b/lite/tests/kernels/mul_compute_test.cc
@@ -127,7 +127,7 @@ TEST(Mul, precision) {
 #if defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
   abs_error = 1e-2;  // use fp16 in npu
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/multiclass_nms_compute_test.cc b/lite/tests/kernels/multiclass_nms_compute_test.cc
index a1190197bf..dd16730ef5 100644
--- a/lite/tests/kernels/multiclass_nms_compute_test.cc
+++ b/lite/tests/kernels/multiclass_nms_compute_test.cc
@@ -478,7 +478,7 @@ TEST(multiclass_nms, precision) {
   Place place;
 #if defined(LITE_WITH_ARM)
   place = TARGET(kHost);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/pool_compute_test.cc b/lite/tests/kernels/pool_compute_test.cc
index 04894188b0..fc4d004e55 100644
--- a/lite/tests/kernels/pool_compute_test.cc
+++ b/lite/tests/kernels/pool_compute_test.cc
@@ -381,7 +381,7 @@ TEST(Pool, precision) {
 #if defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
   abs_error = 1e-2;  // Using fp16 in NPU
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/reshape_compute_test.cc b/lite/tests/kernels/reshape_compute_test.cc
index 3a866b6cf2..f3fcc0bad5 100644
--- a/lite/tests/kernels/reshape_compute_test.cc
+++ b/lite/tests/kernels/reshape_compute_test.cc
@@ -206,7 +206,7 @@ TEST(Reshape, precision) {
   abs_error = 1e-2;  // Using fp16 in NPU
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kHost);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/scale_compute_test.cc b/lite/tests/kernels/scale_compute_test.cc
index efd0497002..9d1f4403dc 100644
--- a/lite/tests/kernels/scale_compute_test.cc
+++ b/lite/tests/kernels/scale_compute_test.cc
@@ -165,7 +165,7 @@ TEST(Scale, precision) {
   abs_error = 4e-3;  // Using fp16 in NPU
 #elif defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
   abs_error = 3e-4;  // Some operations use fp16 in XPU
 #elif defined(LITE_WITH_X86)
diff --git a/lite/tests/kernels/slice_compute_test.cc b/lite/tests/kernels/slice_compute_test.cc
index fc96b39f01..8d5d2c089e 100644
--- a/lite/tests/kernels/slice_compute_test.cc
+++ b/lite/tests/kernels/slice_compute_test.cc
@@ -273,7 +273,7 @@ TEST(Slice, precision) {
   test_slice(place);
   test_slice_tensor(place);
   test_slice_tensor_list(place);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   Place place(TARGET(kXPU));
   test_slice(place);
 #endif
diff --git a/lite/tests/kernels/softmax_compute_test.cc b/lite/tests/kernels/softmax_compute_test.cc
index a91f6534ff..e0f708f3a7 100644
--- a/lite/tests/kernels/softmax_compute_test.cc
+++ b/lite/tests/kernels/softmax_compute_test.cc
@@ -103,7 +103,7 @@ TEST(Softmax, precision) {
 #if defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
   abs_error = 4e-3;  // Using fp16 in NPU
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/stack_compute_test.cc b/lite/tests/kernels/stack_compute_test.cc
index 10b289e419..72529cac51 100644
--- a/lite/tests/kernels/stack_compute_test.cc
+++ b/lite/tests/kernels/stack_compute_test.cc
@@ -106,7 +106,7 @@ TEST(Stack, precision) {
   Place place;
 #ifdef LITE_WITH_ARM
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tests/kernels/transpose_compute_test.cc b/lite/tests/kernels/transpose_compute_test.cc
index 0ec010e47f..933e9f8ec5 100644
--- a/lite/tests/kernels/transpose_compute_test.cc
+++ b/lite/tests/kernels/transpose_compute_test.cc
@@ -164,7 +164,7 @@ TEST(Transpose, precision) {
   LOG(INFO) << "test Transpose op";
   float abs_error = 2e-5;
   Place place;
-#ifdef LITE_WITH_XPU
+#if defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #elif defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
diff --git a/lite/tests/kernels/yolo_box_compute_test.cc b/lite/tests/kernels/yolo_box_compute_test.cc
index c41c89608f..b88f25e1e0 100644
--- a/lite/tests/kernels/yolo_box_compute_test.cc
+++ b/lite/tests/kernels/yolo_box_compute_test.cc
@@ -247,7 +247,7 @@ TEST(YoloBox, precision) {
   Place place;
 #if defined(LITE_WITH_ARM)
   place = TARGET(kARM);
-#elif defined(LITE_WITH_XPU)
+#elif defined(LITE_WITH_XPU) && defined(LITE_WITH_XTCL)
   place = TARGET(kXPU);
 #else
   return;
diff --git a/lite/tools/ci_build.sh b/lite/tools/ci_build.sh
index 29ed9100f9..fde4ea6faa 100755
--- a/lite/tools/ci_build.sh
+++ b/lite/tools/ci_build.sh
@@ -353,7 +353,7 @@ function cmake_xpu {
         -DWITH_MKL=ON \
         -DLITE_BUILD_EXTRA=ON \
         -DLITE_WITH_XPU=ON \
-        -DXPU_SDK_ROOT="$(pwd)/../../XPU_SDK"
+        -DXPU_SDK_ROOT="/opt/output"
 }
 
 function build_xpu {
-- 
GitLab