Merge branch 'Enable_non_blocking_DAG_run' of https://github.com/RedisAI/RedisAI into Enable_non_blocking_DAG_run

Author: alonre24

CMakeLists.txt

@@ -24,6 +24,9 @@ ENDIF()
 #----------------------------------------------------------------------------------------------
 
 SET(CMAKE_CC_COMMON_FLAGS "-fPIC")
+IF (USE_PROFILE)
+    SET(CMAKE_CC_COMMON_FLAGS "${CMAKE_CC_COMMON_FLAGS} -g -ggdb -fno-omit-frame-pointer")
+ENDIF()
 
 IF (USE_COVERAGE)
     IF (NOT CMAKE_BUILD_TYPE STREQUAL "Debug")

opt/Makefile

@@ -33,6 +33,8 @@ make fetch  # download and prepare dependant artifacts
 make build      # compile and link
   DEBUG=1         # build for debugging
   COV=1           # build for coverage analysis (implies DEBUG=1)
+  PROFILE=1       # enable profiling compile flags (and debug symbols) for release type.
+                  # You can consider this as build type release with debug symbols and -fno-omit-frame-pointer
   VARIANT=name    # build variant `name`
   WHY=1           # explain CMake decisions (into /tmp/cmake.why) 
 make clean      # remove build artifacts
@@ -63,10 +65,11 @@ make docker           # build docker image
 make docker-gpu-test  # run GPU tests
 
 fetch and build options:
-  WITH_TF=0     # SKip TensofFlow
-  WITH_TFLITE=0 # SKip TensofFlowLite
-  WITH_PT=0     # Skip PyTorch
-  WITH_ORT=0    # SKip ONNXRuntime
+  WITH_TF=0           # SKip TensofFlow
+  WITH_TFLITE=0       # SKip TensofFlowLite
+  WITH_PT=0           # Skip PyTorch
+  WITH_ORT=0          # SKip ONNXRuntime
+  WITH_UNIT_TESTS=0   # SKip unit tests
 
 device selection options (fetch, build, and test):
   CPU=1         # build for CPU
@@ -115,10 +118,17 @@ else
 USE_COVERAGE=off
 endif
 
+ifeq ($(PROFILE),1)
+USE_PROFILE=on
+else
+USE_PROFILE=off
+endif
+
 CMAKE_FLAGS += \
 	-DDEPS_PATH=$(abspath $(DEPS_DIR)) \
 	-DINSTALL_PATH=$(abspath $(INSTALL_DIR)) \
 	-DUSE_COVERAGE=$(USE_COVERAGE) \
+	-DUSE_PROFILE=$(USE_PROFILE) \
 	-DREDISAI_GIT_SHA=\"$(GIT_SHA)\" \
 	-DDEVICE=$(DEVICE)
 
@@ -138,6 +148,10 @@ ifeq ($(WITH_ORT),0)
 CMAKE_FLAGS += -DBUILD_ORT=off
 endif
 
+ifeq ($(WITH_UNIT_TESTS),0)
+CMAKE_FLAGS += -DPACKAGE_UNIT_TESTS=off
+endif
+
 include $(MK)/defs
 
 #----------------------------------------------------------------------------------------------

opt/readies

@@ -1,3 +1,4 @@
+#define REDISMODULE_MAIN
 #include <cuda_provider_factory.h>
 #include "backends/onnxruntime.h"
 #include "backends/util.h"

src/backends/onnxruntime.c

@@ -1,3 +1,4 @@
+#define REDISMODULE_MAIN
 #include "backends/tensorflow.h"
 #include "backends/util.h"
 #include "tensor.h"

src/backends/tensorflow.c

@@ -1,3 +1,4 @@
+#define REDISMODULE_MAIN
 #include "backends/tflite.h"
 #include "backends/util.h"
 #include "tensor.h"

src/backends/tflite.c

@@ -1,3 +1,4 @@
+#define REDISMODULE_MAIN
 #include "backends/torch.h"
 #include "backends/util.h"
 #include "tensor.h"

src/backends/torch.c

@@ -175,11 +175,11 @@ void *RedisAI_Run_ThreadMain(void *arg) {
                     if (timedOut == 1) {
                         queueEvict(run_queue_info->run_queue, item);
 
-						RedisAI_RunInfo *orig = rinfo->orig_copy;
+                        RedisAI_RunInfo *orig = rinfo->orig_copy;
                         long long dagRefCount = RAI_DagRunInfoFreeShallowCopy(rinfo);
                         if (dagRefCount == 0) {
                             RedisAI_OnFinishCtx finish_ctx = (RedisAI_RunInfo *)orig;
-							orig->OnFinish(finish_ctx, orig->private_data);
+                            orig->OnFinish(finish_ctx, orig->private_data);
                         }
 
                         queueItem *evicted_item = item;
@@ -414,11 +414,11 @@ void *RedisAI_Run_ThreadMain(void *arg) {
                     // If there was an error and the reference count for the dag
                     // has gone to zero and the client is still around, we unblock
                     if (dagError) {
-						RedisAI_RunInfo *orig = rinfo->orig_copy;
+                        RedisAI_RunInfo *orig = rinfo->orig_copy;
                         long long dagRefCount = RAI_DagRunInfoFreeShallowCopy(rinfo);
                         if (dagRefCount == 0) {
                             RedisAI_OnFinishCtx finish_ctx = (RedisAI_RunInfo *)orig;
-							orig->OnFinish(finish_ctx, orig->private_data);
+                            orig->OnFinish(finish_ctx, orig->private_data);
                         }
                     } else {
                         rinfo->dagDeviceCompleteOpCount += 1;
@@ -435,10 +435,10 @@ void *RedisAI_Run_ThreadMain(void *arg) {
             int dag_complete_after_run = RedisAI_DagComplete(batch_rinfo[0]);
 
             long long dagRefCount = -1;
-			RedisAI_RunInfo *orig;
+            RedisAI_RunInfo *orig;
             if (device_complete == 1 || device_complete_after_run == 1) {
-				RedisAI_RunInfo *evicted_rinfo = (RedisAI_RunInfo *)(evicted_items[0]->value);
-				orig = evicted_rinfo->orig_copy;
+                RedisAI_RunInfo *evicted_rinfo = (RedisAI_RunInfo *)(evicted_items[0]->value);
+                orig = evicted_rinfo->orig_copy;
                 // We decrease and get the reference count for the DAG.
                 dagRefCount = RAI_DagRunInfoFreeShallowCopy(evicted_rinfo);
             }

src/background_workers.c

@@ -9,13 +9,13 @@
  *
  */
 
-#if defined(__linux__)
-#define _GNU_SOURCE
-#endif
-
 #ifndef SRC_BACKGROUND_WORKERS_H_
 #define SRC_BACKGROUND_WORKERS_H_
 
+#if defined(__linux__) && !defined(_GNU_SOURCE)
+#define _GNU_SOURCE
+#endif
+
 #include <pthread.h>
 
 #include "config.h"

src/background_workers.h

@@ -1,3 +1,4 @@
+#define REDISMODULE_MAIN
 #include "redismodule.h"
 #include "tensor.h"