Merge branch 'OnnxRuntime_set_allocator' into ck-dockerisms

Author: alonre24

CMakeLists.txt

@@ -200,6 +200,7 @@ ENDIF()
 #----------------------------------------------------------------------------------------------
 
 ADD_SUBDIRECTORY(src)
+ADD_SUBDIRECTORY(tests/module)
 ADD_LIBRARY(redisai SHARED $<TARGET_OBJECTS:redisai_obj>)
 
 TARGET_LINK_LIBRARIES(redisai ${CMAKE_DL_LIBS})

opt/Makefile

@@ -276,6 +276,7 @@ test: build
 	$(SHOW)\
 		DEVICE=$(DEVICE) \
 		MODULE=$(realpath $(INSTALLED_TARGET)) \
+		TESTMOD=$(realpath $(BINDIR)/tests/module/testmod.so) \
 		CLUSTER=$(CLUSTER) \
 		GEN=$(GEN) AOF=$(AOF) SLAVES=$(SLAVES) \
 		VALGRIND=$(VALGRIND) \

opt/build/onnxruntime/Dockerfile.x64

@@ -24,7 +24,8 @@ RUN apt-get install -y locales && \
 
 WORKDIR /build
 
-ARG BUILDTYPE=MinSizeRel
+
+ARG BUILDTYPE=RelWithDebInfo
 ARG BUILDARGS="--config ${BUILDTYPE} --parallel"
 
 RUN git clone --single-branch --branch rel-${ONNXRUNTIME_VER} ${ONNXRUNTIME_REPO} onnxruntime

opt/build/onnxruntime/pack.sh

@@ -5,7 +5,7 @@ VER="$1"
 PLATFORM="$2"
 
 mkdir -p pack/include pack/lib
-cp onnxruntime/build/Linux/MinSizeRel/libonnxruntime.so.${VER} pack/lib/
+cp onnxruntime/build/Linux/RelWithDebInfo/libonnxruntime.so.${VER} pack/lib/
 cp onnxruntime/docs/C_API.md pack/
 cp onnxruntime/LICENSE pack/
 cp onnxruntime/README.md pack/

src/backends/onnxruntime.c

@@ -320,14 +320,13 @@ RAI_Model *RAI_ModelCreateORT(RAI_Backend backend, const char *devicestr, RAI_Mo
     }
 
     ONNX_VALIDATE_STATUS(ort->CreateSessionOptions(&session_options))
-
-#ifndef RAI_ONNXRUNTIME_USE_CUDA
-    // These are required to ensure that onnx will use the registered REDIS allocator (for
-    // a model that defined to run on CPU).
-    ONNX_VALIDATE_STATUS(
-        ort->AddSessionConfigEntry(session_options, "session.use_env_allocators", "1"))
-    ONNX_VALIDATE_STATUS(ort->DisableCpuMemArena(session_options))
-#endif
+    if (strcasecmp(devicestr, "CPU") == 0) {
+        // These are required to ensure that onnx will use the registered REDIS allocator (for
+        // a model that defined to run on CPU).
+        ONNX_VALIDATE_STATUS(
+            ort->AddSessionConfigEntry(session_options, "session.use_env_allocators", "1"))
+        ONNX_VALIDATE_STATUS(ort->DisableCpuMemArena(session_options))
+    }
 
     // TODO: these options could be configured at the AI.CONFIG level
     ONNX_VALIDATE_STATUS(ort->SetSessionGraphOptimizationLevel(session_options, 1))

tests/flow/includes.py

@@ -15,6 +15,7 @@
 import paella
 
 ROOT = os.environ.get("ROOT", None)
+TESTMOD_PATH = os.environ.get("TESTMOD", None)
 MAX_ITERATIONS = 2 if os.environ.get("MAX_ITERATIONS") == None else os.environ.get("MAX_ITERATIONS")
 TEST_TF = os.environ.get("TEST_TF") != "0" and os.environ.get("WITH_TF") != "0"
 TEST_TFLITE = os.environ.get("TEST_TFLITE") != "0" and os.environ.get("WITH_TFLITE") != "0"

tests/flow/tests.sh

@@ -28,6 +28,9 @@ help() {
 		Argument variables:
 		VERBOSE=1     Print commands
 		IGNERR=1      Do not abort on error
+
+		MODULE=path     Path to redisai.so
+		TESTMOD=path    Path to LLAPI module
 		
 		DEVICE=cpu|gpu  Device for testing
 		GEN=0|1         General tests
@@ -92,7 +95,6 @@ valgrind_summary() {
 #----------------------------------------------------------------------------------------------
 
 run_tests() {
-  make -C ../module
 	local title="$1"
 	[[ ! -z $title ]] && { $ROOT/opt/readies/bin/sep -0; printf "Tests with $title:\n\n"; }
 	cd $ROOT/tests/flow
@@ -116,6 +118,8 @@ OP=""
 
 MODULE=${MODULE:-$1}
 [[ -z $MODULE || ! -f $MODULE ]] && { echo "Module not found at ${MODULE}. Aborting."; exit 1; }
+TESTMOD=${TESTMOD}
+echo "Test module path is ${TESTMOD}"
 
 [[ $VALGRIND == 1 || $VGD == 1 ]] && valgrind_config
 

tests/flow/tests_llapi.py

@@ -8,18 +8,15 @@
 python -m RLTest --test tests_llapi.py --module path/to/redisai.so
 '''
 
-
-def ensure_test_module_loaded(f):
+def with_test_module(f):
     @wraps(f)
     def wrapper(env, *args, **kwargs):
-        goal_dir = os.path.join(os.path.dirname(__file__), "../module/LLAPI.so")
-        TEST_MODULE_PATH = os.path.abspath(goal_dir)
         con = env.getConnection()
         modules = con.execute_command("MODULE", "LIST")
         if b'RAI_llapi' in [module[1] for module in modules]:
             return f(env, *args, **kwargs)
         try:
-            ret = con.execute_command('MODULE', 'LOAD', TEST_MODULE_PATH)
+            ret = con.execute_command('MODULE', 'LOAD', TESTMOD_PATH)
             env.assertEqual(ret, b'OK')
         except Exception as e:
             env.assertFalse(True)
@@ -29,15 +26,15 @@ def wrapper(env, *args, **kwargs):
     return wrapper
 
 
-@ensure_test_module_loaded
+@with_test_module
 def test_basic_check(env):
 
     con = env.getConnection()
     ret = con.execute_command("RAI_llapi.basic_check")
     env.assertEqual(ret, b'OK')
 
 
-@ensure_test_module_loaded
+@with_test_module
 def test_model_run_async(env):
 
     con = env.getConnection()
@@ -56,7 +53,7 @@ def test_model_run_async(env):
     env.assertEqual(ret, b'Async run success')
 
 
-@ensure_test_module_loaded
+@with_test_module
 def test_script_run_async(env):
 
     con = env.getConnection()
@@ -78,7 +75,7 @@ def test_script_run_async(env):
     env.assertEqual(ret, b'Async run success')
 
 
-@ensure_test_module_loaded
+@with_test_module
 def test_dag_build_and_run(env):
     con = env.getConnection()
 
@@ -105,7 +102,7 @@ def test_dag_build_and_run(env):
     env.assertEqual(ret, b'DAG run success')
 
 
-@ensure_test_module_loaded
+@with_test_module
 def test_dagrun_multidevice_resnet(env):
     con = env.getConnection()
 

tests/flow/tests_onnx.py

@@ -16,7 +16,6 @@ def test_onnx_modelrun_mnist(env):
         return
 
     con = env.getConnection()
-
     test_data_path = os.path.join(os.path.dirname(__file__), 'test_data')
     model_filename = os.path.join(test_data_path, 'mnist.onnx')
     wrong_model_filename = os.path.join(test_data_path, 'graph.pb')
@@ -458,9 +457,6 @@ def test_onnx_use_custom_allocator(env):
     if not TEST_ONNX:
         env.debugPrint("skipping {} since TEST_ONNX=0".format(sys._getframe().f_code.co_name), force=True)
         return
-    if DEVICE != 'CPU':
-        env.debugPrint("skipping {} since the custom allocator is for CPU".format(sys._getframe().f_code.co_name), force=True)
-        return
 
     con = env.getConnection()
     test_data_path = os.path.join(os.path.dirname(__file__), 'test_data')
@@ -472,7 +468,7 @@ def test_onnx_use_custom_allocator(env):
     env.assertEqual(int(ai_memory_config["ai_onnxruntime_memory"]), 0)
 
     # Expect using the allocator during model set for allocating the model, its input name and output name.
-    ret = con.execute_command('AI.MODELSET', 'm{1}', 'ONNX', DEVICE, 'BLOB', model_pb)
+    ret = con.execute_command('AI.MODELSET', 'm{1}', 'ONNX', 'CPU', 'BLOB', model_pb)
     env.assertEqual(ret, b'OK')
     ai_memory_config = {k.split(":")[0]: k.split(":")[1]
                         for k in con.execute_command("INFO MODULES").decode().split("#")[4].split()[1:]}
@@ -496,7 +492,7 @@ def test_onnx_use_custom_allocator(env):
     with open(sample_filename, 'rb') as f:
         sample_raw = f.read()
 
-    ret = con.execute_command('AI.MODELSET', 'm{1}', 'ONNX', DEVICE, 'BLOB', model_pb)
+    ret = con.execute_command('AI.MODELSET', 'm{1}', 'ONNX', 'CPU', 'BLOB', model_pb)
     env.assertEqual(ret, b'OK')
     con.execute_command('AI.TENSORSET', 'a{1}', 'FLOAT', 1, 1, 28, 28, 'BLOB', sample_raw)
 

tests/module/CMakeLists.txt

@@ -0,0 +1,5 @@
+ADD_LIBRARY(testmod SHARED LLAPI.c DAG_utils.c)
+
+INCLUDE_DIRECTORIES(../../src)
+SET_TARGET_PROPERTIES(testmod PROPERTIES PREFIX "")
+SET_TARGET_PROPERTIES(testmod PROPERTIES SUFFIX ".so")