perf: fix skinning performance issues

PilotLightTech · Jan 23, 2025 · 40dad64 · 40dad64
1 parent 36cda69
commit 40dad64
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 6 deletions.
diff --git a/extensions/pl_renderer_internal.c b/extensions/pl_renderer_internal.c
@@ -746,7 +746,7 @@ pl_refr_perform_skinning(plCommandBuffer* ptCommandBuffer, uint32_t uSceneHandle
         int iSourceDataOffset;
         int iDestDataOffset;
         int iDestVertexOffset;
-        int iUnused;
+        uint32_t uMaxSize;
     } SkinDynamicData;
 
     if(uSkinCount)
@@ -772,12 +772,13 @@ pl_refr_perform_skinning(plCommandBuffer* ptCommandBuffer, uint32_t uSceneHandle
             ptDynamicData->iSourceDataOffset = ptScene->sbtSkinData[i].iSourceDataOffset;
             ptDynamicData->iDestDataOffset = ptScene->sbtSkinData[i].iDestDataOffset;
             ptDynamicData->iDestVertexOffset = ptScene->sbtSkinData[i].iDestVertexOffset;
+            ptDynamicData->uMaxSize = ptScene->sbtSkinData[i].uVertexCount;
 
             const plDispatch tDispach = {
-                .uGroupCountX     = ptScene->sbtSkinData[i].uVertexCount,
+                .uGroupCountX     = (uint32_t)ceilf((float)ptScene->sbtSkinData[i].uVertexCount / 64.0f),
                 .uGroupCountY     = 1,
                 .uGroupCountZ     = 1,
-                .uThreadPerGroupX = 1,
+                .uThreadPerGroupX = 64,
                 .uThreadPerGroupY = 1,
                 .uThreadPerGroupZ = 1
             };

diff --git a/shaders/skinning.comp b/shaders/skinning.comp
@@ -47,7 +47,7 @@ layout(set = 3, binding = 0) uniform PL_DYNAMIC_DATA
     int iSourceDataOffset;
     int iDestDataOffset;
     int iDestVertexOffset;
-    int iUnused;
+    uint uMaxSize;
 } tObjectInfo;
 
 //-----------------------------------------------------------------------------
@@ -108,13 +108,16 @@ get_skinning_matrix(vec4 inJoints0, vec4 inWeights0)
 // [SECTION] entry
 //-----------------------------------------------------------------------------
 
-layout (local_size_x = 16, local_size_y = 16, local_size_z = 1) in;
+layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
 
 void
 main()
 {
 
-    const uint iVertexIndex = gl_WorkGroupID.x;
+    const uint iVertexIndex = gl_WorkGroupID.x * 64 + gl_LocalInvocationID.x;
+
+    if(iVertexIndex > tObjectInfo.uMaxSize)
+        return;
 
     vec4 inPosition  = vec4(0.0, 0.0, 0.0, 1.0);
     vec3 inNormal    = vec3(0.0, 0.0, 0.0);