keep load_or_save_cache when destoryComputation (#698)

* add potential exception handle & open session_option to optimize

* update odla_computation::init() to return odla_status value

* fix popart_config load error

* improve load config logic

* Check the sdk version of the cache

* Format the codes to pass lint checking

* reset other necessary popart config value

* add return value in computation::init()

* clear cache_fs before write new data

* clear fstream after close

* fix lint error

* fix error message issue

* keep load_or_save_cache value

* use popart::core::packageHash for version info

update some log type to err & get sdk packagehash in use_default()

use assign instead of =& remove useless logic

close log

Co-authored-by: yanwei <yw01041751@alibaba-inc.com>
Co-authored-by: gcuser <jackz@graphcore.ai>

Author: 3 people

ODLA/platforms/odla_popart/odla_compute.cc

@@ -18,6 +18,7 @@
 
 #include <ODLA/odla.h>
 #include <dlfcn.h>
+#include <stdlib.h>
 
 #include <cstdlib>
 #include <fstream>
@@ -167,6 +168,7 @@ odla_status odla_DestroyContext(odla_context ctx) {
 }
 
 odla_status odla_DestroyComputation(odla_computation comp) {
+  popart::logging::info("call odla_destroyComputation");
   if (comp != nullptr) {
     if (!comp->is_compile_only()) {
       comp->mark_done();

ODLA/platforms/odla_popart/odla_popart.cc

@@ -79,14 +79,14 @@ void compute_loop(odla_computation comp) {
     popart::logging::err("Poplar unrecoverable_runtime_error exception caught");
     QManager::instance()->set_status(ODLA_UNRECOVERABLE_ERR);
   } catch (poplar::unknown_runtime_error& e) {
-    popart::logging::info("Poplar unknown runtime exception caught");
+    popart::logging::err("Poplar unknown runtime exception caught");
     QManager::instance()->set_status(ODLA_UNRECOVERABLE_ERR);
   } catch (...) {
-    popart::logging::info("Poplar unknown exception caught");
+    popart::logging::err("Poplar unknown exception caught");
     QManager::instance()->set_status(ODLA_UNRECOVERABLE_ERR);
   }
 
-  popart::logging::warn("The pipeline loop finished");
+  popart::logging::info("The pipeline loop finished");
   comp->thread_done();
 }
 
@@ -134,10 +134,12 @@ odla_status _odla_computation::compile_and_export() {
     config_string = PopartConfig::instance()->get_default_config_string();
   }
   // add sdk_version in the file content
-  std::string version_string(popart::core::versionString());
+  std::string version_string(popart::core::packageHash());
   popart::logging::info("the popart version is: {}", version_string);
-  version_string = "\n\"sdk_version\":\"" + version_string + "\",";
-  config_string.insert(1, version_string);
+  if (config_string.find("sdk_version") == std::string::npos) {
+    std::string item_string = "\n\"sdk_version\":\"" + version_string + "\",";
+    config_string.insert(1, item_string);
+  }
   popart::logging::info("the config_string with sdk_version is: {}",
                         config_string);
   // added the sdk_version information to the file content
@@ -236,7 +238,7 @@ odla_status _odla_computation::init(bool is_compile) {
       if (!is_compile) {
         if (PopartConfig::instance()->load_or_save_cache()) {
           popart::logging::info("Load cachefile from existing stream");
-          std::string version_string(popart::core::versionString());
+          std::string version_string(popart::core::packageHash());
           if (!PopartConfig::instance()->sdk_version_match(version_string)) {
             popart::logging::err("The sdk version of cache does not match {}",
                                  version_string);

ODLA/platforms/odla_popart/popart_config.cc

@@ -27,6 +27,10 @@
 #include "json.hpp"
 
 PopartConfig* PopartConfig::instance_ = new PopartConfig();
+std::vector<std::string> PopartConfig::mode = {"unknown", "pipeline",
+                                               "parallel", "sequence"};
+
+const char* bool_to_str(const bool& value) { return value ? "true" : "false"; }
 
 const std::string& get_config_path_from_cache_file(
     const std::string& cache_path) {
@@ -43,6 +47,7 @@ const std::string& get_config_path_from_cache_file(
 
 void PopartConfig::use_default() {
   amp_ = 0.6;
+  sdk_version_ = popart::core::packageHash();
   version_ = "1.0.0";
   batches_per_step_ = 1;
   ipu_num_ = 1;
@@ -54,19 +59,27 @@ void PopartConfig::use_default() {
   queue_type_ = "LockFreeQueue";
   queue_capacity_ = 1024 * 1024;
   debug_ = false;
-  default_config_string_ =
+  char config_base[] =
       "{\n\
-      \"version\":\"1.0.0\",\n\
-      \"amp\":0.6,\n\
-      \"batches_per_step\":1,\n\
-      \"execution_mode\":\"sequence\",\n\
-      \"ipu_num\":1,\n\
-      \"load_onnx\":false, \n\
-      \"load_onnx_path\":\"test-load-time.onnx\",\n\
-      \"queue_type\":\"LockFreeQueue\",\n\
-      \"queue_capacity\":1048576,\n\
-      \"debug\": false\n\
+      \"sdk_version\":\"%s\",\n\
+      \"version\":\"%s\",\n\
+      \"amp\":%f,\n\
+      \"batches_per_step\":%d,\n\
+      \"execution_mode\":\"%s\",\n\
+      \"ipu_num\":%d,\n\
+      \"load_onnx\":%s, \n\
+      \"load_onnx_path\":\"%s\",\n\
+      \"queue_type\":\"%s\",\n\
+      \"queue_capacity\":%d,\n\
+      \"debug\":%s\n\
       }\n";
+  char raw_default_config[1024] = {0};
+  snprintf(raw_default_config, 1024, config_base, sdk_version_.c_str(),
+           version_.c_str(), amp_, batches_per_step_,
+           PopartConfig::mode[(int)execution_mode_].c_str(), ipu_num_,
+           bool_to_str(load_onnx_), load_onnx_path_.c_str(),
+           queue_type_.c_str(), queue_capacity_, bool_to_str(debug_));
+  default_config_string_.assign(raw_default_config);
 }
 
 odla_status PopartConfig::load_config(const char* env_file_path) {
@@ -196,12 +209,12 @@ odla_status PopartConfig::load_from_file(const std::string& file_path) {
 void PopartConfig::print() {
   std::string line(80, '=');
   popart::logging::info(line);
+  popart::logging::info("sdk_version: {}", sdk_version_);
   popart::logging::info("version: {}", version_);
   popart::logging::info("amp: {}", amp_);
   popart::logging::info("batch_per_step: {}", batches_per_step_);
-  std::string mode[] = {"UNKNOWN", "PIPELINE", "PARALLEL", "SEQUENCE"};
   popart::logging::info("execution_mode: {}",
-                        mode[(long unsigned int)execution_mode_]);
+                        PopartConfig::mode[(long unsigned int)execution_mode_]);
   popart::logging::info("ipu_num: {}", ipu_num_);
   std::string bool_value[] = {"false", "true"};
   popart::logging::info("load_onnx: {}",

ODLA/platforms/odla_popart/popart_config.h

@@ -23,6 +23,7 @@
 #include <iostream>
 #include <map>
 #include <mutex>
+#include <popart/version.hpp>
 #include <regex>
 #include <string>
 #include <vector>
@@ -61,6 +62,7 @@ class PopartConfig {
   std::string sdk_version_; // version of the sdk
   int batches_per_step_;    // Batch per step for PIPELINE & PARALLEL execution
                             // mode
+  static std::vector<std::string> mode; // string value of execution mode
   ExecutionMode
       execution_mode_; // The execution mode {PIPELINE, PARALLEL, SEQUENCE}
   bool load_onnx_;     // Whether load onnx model to run instead of the model
@@ -118,7 +120,6 @@ class PopartConfig {
           cache_fs->clear();
         }
         pipeline_setting_.clear();
-        load_or_save_cache_ = false;
         sdk_version_ = "NA";
       }
     }