From: Jihoon Lee <jhoon.it.lee@samsung.com>
Date: Wed, 22 Dec 2021 07:17:53 +0000 (+0900)
Subject: [Graph] set trainable only if trainable
X-Git-Tag: accepted/tizen/unified/20220323.062643~74
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=3d9366f54bba773cbdca145e33f8882d2aa8185f;p=platform%2Fcore%2Fml%2Fnntrainer.git

[Graph] set trainable only if trainable

This patch fixes bug that needsCalcGradient is set true when it
shouldn't be

**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped

Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
---

diff --git a/nntrainer/graph/network_graph.cpp b/nntrainer/graph/network_graph.cpp
index f1f034f..36cc112 100644
--- a/nntrainer/graph/network_graph.cpp
+++ b/nntrainer/graph/network_graph.cpp
@@ -223,7 +223,9 @@ void NetworkGraph::markNodesForBackwarding() {
     if (lnode->getTrainable() ||
         must_support_backwarding.find(lnode->getName()) !=
           must_support_backwarding.end()) {
-      lnode->needsCalcGradient(true);
+      if (lnode->getTrainable()) {
+        lnode->needsCalcGradient(true);
+      }
 #ifdef ENABLE_TEST
       if (lnode->supportBackwarding() && !optimize_memory) {
         lnode->needsCalcDerivative(true);