[Bf-blender-cvs] [a84c92fc73d] blender-v3.4-release: Fix T98989: Performance regression when using multiple bump nodes

[Miguel Pozo]

Commit: a84c92fc73d3298c1456ffb26c1e73cd67b56e6d
Author: Miguel Pozo
Date:   Mon Nov 14 12:21:37 2022 +0100
Branches: blender-v3.4-release
https://developer.blender.org/rBa84c92fc73d3298c1456ffb26c1e73cd67b56e6d

Fix T98989: Performance regression when using multiple bump nodes

Ensure each graph material_function only evaluates the input links that are connected to it.

Differential Revision: https://developer.blender.org/D16425

===================================================================

M	source/blender/gpu/intern/gpu_codegen.cc
M	source/blender/gpu/intern/gpu_node_graph.c
M	source/blender/gpu/intern/gpu_node_graph.h

===================================================================

diff --git a/source/blender/gpu/intern/gpu_codegen.cc b/source/blender/gpu/intern/gpu_codegen.cc
index 4adeac1b49a..42b2151e75b 100644
--- a/source/blender/gpu/intern/gpu_codegen.cc
+++ b/source/blender/gpu/intern/gpu_codegen.cc
@@ -620,11 +620,21 @@ void GPUCodegen::generate_graphs()
     std::stringstream eval_ss;
     eval_ss << "\n/* Generated Functions */\n\n";
     LISTBASE_FOREACH (GPUNodeGraphFunctionLink *, func_link, &graph.material_functions) {
+      /* Untag every node in the graph to avoid serializing nodes from other functions */
+      LISTBASE_FOREACH (GPUNode *, node, &graph.nodes) {
+        node->tag &= ~GPU_NODE_TAG_FUNCTION;
+      }
+      /* Tag only the nodes needed for the current function */
+      gpu_nodes_tag(func_link->outlink, GPU_NODE_TAG_FUNCTION);
       char *fn = graph_serialize(GPU_NODE_TAG_FUNCTION, func_link->outlink);
       eval_ss << "float " << func_link->name << "() {\n" << fn << "}\n\n";
       MEM_SAFE_FREE(fn);
     }
     output.material_functions = extract_c_str(eval_ss);
+    /* Leave the function tags as they were before serialization */
+    LISTBASE_FOREACH (GPUNodeGraphFunctionLink *, funclink, &graph.material_functions) {
+      gpu_nodes_tag(funclink->outlink, GPU_NODE_TAG_FUNCTION);
+    }
   }
 
   LISTBASE_FOREACH (GPUMaterialAttribute *, attr, &graph.attributes) {
diff --git a/source/blender/gpu/intern/gpu_node_graph.c b/source/blender/gpu/intern/gpu_node_graph.c
index c72e7097b33..14dfce6ce5e 100644
--- a/source/blender/gpu/intern/gpu_node_graph.c
+++ b/source/blender/gpu/intern/gpu_node_graph.c
@@ -890,7 +890,7 @@ void gpu_node_graph_free(GPUNodeGraph *graph)
 
 /* Prune Unused Nodes */
 
-static void gpu_nodes_tag(GPUNodeLink *link, eGPUNodeTag tag)
+void gpu_nodes_tag(GPUNodeLink *link, eGPUNodeTag tag)
 {
   GPUNode *node;
 
diff --git a/source/blender/gpu/intern/gpu_node_graph.h b/source/blender/gpu/intern/gpu_node_graph.h
index de0a0687b13..9f919f04160 100644
--- a/source/blender/gpu/intern/gpu_node_graph.h
+++ b/source/blender/gpu/intern/gpu_node_graph.h
@@ -66,7 +66,7 @@ typedef enum {
   GPU_NODE_TAG_COMPOSITOR = (1 << 6),
 } eGPUNodeTag;
 
-ENUM_OPERATORS(eGPUNodeTag, GPU_NODE_TAG_FUNCTION)
+ENUM_OPERATORS(eGPUNodeTag, GPU_NODE_TAG_COMPOSITOR)
 
 struct GPUNode {
   struct GPUNode *next, *prev;
@@ -186,6 +186,7 @@ typedef struct GPUNodeGraph {
 
 /* Node Graph */
 
+void gpu_nodes_tag(GPUNodeLink *link, eGPUNodeTag tag);
 void gpu_node_graph_prune_unused(GPUNodeGraph *graph);
 void gpu_node_graph_finalize_uniform_attrs(GPUNodeGraph *graph);