less copying json

IQTLabs · Sep 14, 2024 · 1d60e9b · 1d60e9b
1 parent 955b643
commit 1d60e9b
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 16 deletions.
diff --git a/lib/iq_inference_impl.cc b/lib/iq_inference_impl.cc
@@ -363,15 +363,8 @@ void iq_inference_impl::run_inference_(torchserve_client *client) {
 
     output_json["predictions"] = results_json;
     output_json["metadata"] = metadata_json;
-    const std::string output_json_str = output_json.dump();
-    if (output_json_str.size() < MAX_JSON_SIZE) {
-      json_result_type output_json_chars;
-      std::copy(output_json_str.begin(), output_json_str.end(),
-                output_json_chars.data());
-      json_q_.push(output_json_chars);
-    } else {
-      d_logger->error("output json size too large");
-    }
+    std::string *output_json_str = new std::string(output_json.dump());
+    json_q_.push(output_json_str);
     delete_output_item_(output_item);
   }
 }
@@ -466,10 +459,11 @@ void iq_inference_impl::process_tags_(COUNT_T in_first,
 }
 
 void iq_inference_impl::pub_json_() {
-  json_result_type json;
-  while (json_q_.pop(json)) {
+  std::string *output_json_str;
+  while (json_q_.pop(output_json_str)) {
     ++predictions_;
-    message_port_pub(INFERENCE_KEY, string_to_pmt(std::string(json.data())));
+    message_port_pub(INFERENCE_KEY, string_to_pmt(*output_json_str));
+    delete output_json_str;
   }
 }
 

diff --git a/lib/iq_inference_impl.h b/lib/iq_inference_impl.h
@@ -222,7 +222,6 @@ namespace gr {
 namespace iqtlabs {
 
 const COUNT_T MAX_INFERENCE = 5;
-const COUNT_T MAX_JSON_SIZE = 8192;
 
 typedef struct output_item {
   FREQ_T rx_freq;
@@ -236,8 +235,6 @@ typedef struct output_item {
   COUNT_T serial;
 } output_item_type;
 
-typedef std::array<char, MAX_JSON_SIZE> json_result_type;
-
 class iq_inference_impl : public iq_inference, base_impl {
 private:
   pmt::pmt_t tag_;
@@ -255,7 +252,7 @@ class iq_inference_impl : public iq_inference, base_impl {
   boost::lockfree::queue<output_item_type,
                          boost::lockfree::capacity<MAX_INFERENCE>>
       inference_q_;
-  boost::lockfree::queue<json_result_type,
+  boost::lockfree::queue<std::string*,
                          boost::lockfree::capacity<MAX_INFERENCE>>
       json_q_;
   boost::shared_ptr<boost::asio::io_service> io_service_;