Optimize animation hotpaths: binary keyframe search, eliminate sqrt calls

- Replace O(n) linear keyframe search with O(log n) binary search in both
  M2 and Character renderers (runs thousands of times per frame)
- Smoke particle removal: swap-and-pop instead of O(n²) vector erase
- Character render backface cull: eliminate sqrt via squared comparison
- Quaternion validation: use length² instead of sqrt-based length check

src/rendering/character_renderer.cpp

@@ -1580,13 +1580,12 @@ int CharacterRenderer::findKeyframeIndex(const std::vector<uint32_t>& timestamps
     if (timestamps.empty()) return -1;
     if (timestamps.size() == 1) return 0;
 
-    // Binary search for the keyframe bracket
+    // Binary search: find first element > t, then back up one
-    for (size_t i = 0; i < timestamps.size() - 1; i++) {
+    uint32_t t = static_cast<uint32_t>(time);
-        if (time < static_cast<float>(timestamps[i + 1])) {
+    auto it = std::upper_bound(timestamps.begin(), timestamps.end(), t);
-            return static_cast<int>(i);
+    if (it == timestamps.begin()) return 0;
-        }
+    size_t idx = static_cast<size_t>(it - timestamps.begin()) - 1;
-    }
+    return static_cast<int>(std::min(idx, timestamps.size() - 2));
-    return static_cast<int>(timestamps.size() - 2);
 }
 
 glm::vec3 CharacterRenderer::interpolateVec3(const pipeline::M2AnimationTrack& track,
@@ -1630,8 +1629,8 @@ glm::quat CharacterRenderer::interpolateQuat(const pipeline::M2AnimationTrack& t
     if (keys.timestamps.empty() || keys.quatValues.empty()) return identity;
 
     auto safeQuat = [&](const glm::quat& q) -> glm::quat {
-        float len = glm::length(q);
+        float lenSq = q.x*q.x + q.y*q.y + q.z*q.z + q.w*q.w;
-        if (len < 0.001f || std::isnan(len)) return identity;
+        if (lenSq < 0.000001f || std::isnan(lenSq)) return identity;
         return q;
     };
 
@@ -1773,9 +1772,14 @@ void CharacterRenderer::render(VkCommandBuffer cmd, VkDescriptorSet perFrameSet,
             float distSq = glm::dot(toInst, toInst);
             if (distSq > renderRadiusSq) continue;
             if (distSq > nearNoConeCullSq) {
-                float invDist = 1.0f / std::sqrt(distSq);
+                // Backface cull without sqrt: dot(toInst, camFwd) / |toInst| < threshold
-                float facingDot = glm::dot(toInst, camForward) * invDist;
+                // ⟺ dot < 0 || dot² < threshold² * distSq  (when threshold < 0, dot must be negative)
-                if (facingDot < backfaceDotCull) continue;
+                float rawDot = glm::dot(toInst, camForward);
+                if (backfaceDotCull >= 0.0f) {
+                    if (rawDot < 0.0f || rawDot * rawDot < backfaceDotCull * backfaceDotCull * distSq) continue;
+                } else {
+                    if (rawDot < 0.0f && rawDot * rawDot > backfaceDotCull * backfaceDotCull * distSq) continue;
+                }
             }
         }
 

src/rendering/m2_renderer.cpp

@@ -1741,12 +1741,12 @@ uint32_t M2Renderer::createInstanceWithMatrix(uint32_t modelId, const glm::mat4&
 static int findKeyframeIndex(const std::vector<uint32_t>& timestamps, float time) {
     if (timestamps.empty()) return -1;
     if (timestamps.size() == 1) return 0;
-    for (size_t i = 0; i < timestamps.size() - 1; i++) {
+    uint32_t t = static_cast<uint32_t>(time);
-        if (time < static_cast<float>(timestamps[i + 1])) {
+    // Binary search: find first element > t, then back up one
-            return static_cast<int>(i);
+    auto it = std::upper_bound(timestamps.begin(), timestamps.end(), t);
-        }
+    if (it == timestamps.begin()) return 0;
-    }
+    size_t idx = static_cast<size_t>(it - timestamps.begin()) - 1;
-    return static_cast<int>(timestamps.size() - 2);
+    return static_cast<int>(std::min(idx, timestamps.size() - 2));
 }
 
 // Resolve sequence index and time for a track, handling global sequences.
@@ -1803,8 +1803,8 @@ static glm::quat interpQuat(const pipeline::M2AnimationTrack& track,
     const auto& keys = track.sequences[si];
     if (keys.timestamps.empty() || keys.quatValues.empty()) return identity;
     auto safe = [&](const glm::quat& q) -> glm::quat {
-        float len = glm::length(q);
+        float lenSq = q.x*q.x + q.y*q.y + q.z*q.z + q.w*q.w;
-        if (len < 0.001f || std::isnan(len)) return identity;
+        if (lenSq < 0.000001f || std::isnan(lenSq)) return identity;
         return q;
     };
     if (keys.quatValues.size() == 1) return safe(keys.quatValues[0]);
@@ -1907,21 +1907,23 @@ void M2Renderer::update(float deltaTime, const glm::vec3& cameraPos, const glm::
         smokeEmitAccum = 0.0f;
     }
 
-    // --- Update existing smoke particles ---
+    // --- Update existing smoke particles (swap-and-pop for O(1) removal) ---
-    for (auto it = smokeParticles.begin(); it != smokeParticles.end(); ) {
+    for (size_t i = 0; i < smokeParticles.size(); ) {
-        it->life += deltaTime;
+        auto& p = smokeParticles[i];
-        if (it->life >= it->maxLife) {
+        p.life += deltaTime;
-            it = smokeParticles.erase(it);
+        if (p.life >= p.maxLife) {
+            smokeParticles[i] = smokeParticles.back();
+            smokeParticles.pop_back();
             continue;
         }
-        it->position += it->velocity * deltaTime;
+        p.position += p.velocity * deltaTime;
-        it->velocity.z *= 0.98f;  // Slight deceleration
+        p.velocity.z *= 0.98f;  // Slight deceleration
-        it->velocity.x += distDrift(smokeRng) * deltaTime;
+        p.velocity.x += distDrift(smokeRng) * deltaTime;
-        it->velocity.y += distDrift(smokeRng) * deltaTime;
+        p.velocity.y += distDrift(smokeRng) * deltaTime;
         // Grow from 1.0 to 3.5 over lifetime
-        float t = it->life / it->maxLife;
+        float t = p.life / p.maxLife;
-        it->size = 1.0f + t * 2.5f;
+        p.size = 1.0f + t * 2.5f;
-        ++it;
+        ++i;
     }
 
     // --- Normal M2 animation update ---