Fix Windows ARM64 build: disable x86 asm in StormLib's libtomcrypt

StormLib's bundled libtomcrypt uses x86 inline assembly (bswapl/movl)
gated by __MINGW32__, which is defined on CLANGARM64 too. Pass
-DLTC_NO_BSWAP to force portable C byte-swap fallback.

.github/workflows/build.yml

@@ -195,10 +195,14 @@ jobs:
         shell: msys2 {0}
         run: |
           git clone --depth 1 https://github.com/ladislav-zezula/StormLib.git /tmp/StormLib
+          # Disable x86 inline asm in bundled libtomcrypt (bswapl/movl) —
+          # __MINGW32__ is defined on CLANGARM64 which incorrectly enables x86 asm
           cmake -S /tmp/StormLib -B /tmp/StormLib/build -G Ninja \
             -DCMAKE_BUILD_TYPE=Release \
             -DCMAKE_INSTALL_PREFIX="$MINGW_PREFIX" \
-            -DBUILD_SHARED_LIBS=OFF
+            -DBUILD_SHARED_LIBS=OFF \
+            -DCMAKE_C_FLAGS="-DLTC_NO_BSWAP" \
+            -DCMAKE_CXX_FLAGS="-DLTC_NO_BSWAP"
           cmake --build /tmp/StormLib/build --parallel $(nproc)
           cmake --install /tmp/StormLib/build
 

.github/workflows/release.yml

@@ -256,10 +256,14 @@ jobs:
         shell: msys2 {0}
         run: |
           git clone --depth 1 https://github.com/ladislav-zezula/StormLib.git /tmp/StormLib
+          # Disable x86 inline asm in bundled libtomcrypt (bswapl/movl) —
+          # __MINGW32__ is defined on CLANGARM64 which incorrectly enables x86 asm
           cmake -S /tmp/StormLib -B /tmp/StormLib/build -G Ninja \
             -DCMAKE_BUILD_TYPE=Release \
             -DCMAKE_INSTALL_PREFIX="$MINGW_PREFIX" \
-            -DBUILD_SHARED_LIBS=OFF
+            -DBUILD_SHARED_LIBS=OFF \
+            -DCMAKE_C_FLAGS="-DLTC_NO_BSWAP" \
+            -DCMAKE_CXX_FLAGS="-DLTC_NO_BSWAP"
           cmake --build /tmp/StormLib/build --parallel $(nproc)
           cmake --install /tmp/StormLib/build
 

include/rendering/m2_renderer.hpp

@@ -478,7 +478,6 @@ private:
     // Helper to allocate descriptor sets
     VkDescriptorSet allocateMaterialSet();
     VkDescriptorSet allocateBoneSet();
-    void preallocateBoneBuffers(M2Instance& instance);
 
     // Helper to destroy model GPU resources
     void destroyModelGPU(M2ModelGPU& model);

include/rendering/wmo_renderer.hpp

@@ -657,9 +657,9 @@ private:
     bool wireframeMode = false;
     bool frustumCulling = true;
     bool portalCulling = false;  // Disabled by default - needs debugging
-    bool distanceCulling = true;   // Enabled with active-group exemption to prevent floor disappearing
-    float maxGroupDistance = 800.0f;
-    float maxGroupDistanceSq = 640000.0f;  // maxGroupDistance^2
+    bool distanceCulling = false;  // Disabled - causes ground to disappear
+    float maxGroupDistance = 500.0f;
+    float maxGroupDistanceSq = 250000.0f;  // maxGroupDistance^2
     uint32_t lastDrawCalls = 0;
     mutable uint32_t lastPortalCulledGroups = 0;
     mutable uint32_t lastDistanceCulledGroups = 0;

src/rendering/m2_renderer.cpp

@@ -767,38 +767,6 @@ VkDescriptorSet M2Renderer::allocateBoneSet() {
     return set;
 }
 
-void M2Renderer::preallocateBoneBuffers(M2Instance& instance) {
-    if (!vkCtx_) return;
-    for (int fi = 0; fi < 2; fi++) {
-        if (instance.boneBuffer[fi]) continue; // already allocated
-        VkBufferCreateInfo bci{VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO};
-        bci.size = 128 * sizeof(glm::mat4); // max 128 bones
-        bci.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
-        VmaAllocationCreateInfo aci{};
-        aci.usage = VMA_MEMORY_USAGE_CPU_TO_GPU;
-        aci.flags = VMA_ALLOCATION_CREATE_MAPPED_BIT;
-        VmaAllocationInfo allocInfo{};
-        vmaCreateBuffer(vkCtx_->getAllocator(), &bci, &aci,
-                        &instance.boneBuffer[fi], &instance.boneAlloc[fi], &allocInfo);
-        instance.boneMapped[fi] = allocInfo.pMappedData;
-
-        instance.boneSet[fi] = allocateBoneSet();
-        if (instance.boneSet[fi]) {
-            VkDescriptorBufferInfo bufInfo{};
-            bufInfo.buffer = instance.boneBuffer[fi];
-            bufInfo.offset = 0;
-            bufInfo.range = bci.size;
-            VkWriteDescriptorSet write{VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET};
-            write.dstSet = instance.boneSet[fi];
-            write.dstBinding = 0;
-            write.descriptorCount = 1;
-            write.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
-            write.pBufferInfo = &bufInfo;
-            vkUpdateDescriptorSets(vkCtx_->getDevice(), 1, &write, 0, nullptr);
-        }
-    }
-}
-
 // ---------------------------------------------------------------------------
 // M2 collision mesh: build spatial grid + classify triangles
 // ---------------------------------------------------------------------------
@@ -1647,11 +1615,6 @@ uint32_t M2Renderer::createInstance(uint32_t modelId, const glm::vec3& position,
         instance.variationTimer = 3000.0f + static_cast<float>(rand() % 8000);
     }
 
-    // Pre-allocate bone SSBOs so first render frame doesn't hitch
-    if (mdlRef.hasAnimation && !mdlRef.disableAnimation) {
-        preallocateBoneBuffers(instance);
-    }
-
     instances.push_back(instance);
     size_t idx = instances.size() - 1;
     instanceIndexById[instance.id] = idx;
@@ -1685,8 +1648,6 @@ uint32_t M2Renderer::createInstanceWithMatrix(uint32_t modelId, const glm::mat4&
         }
     }
 
-    const auto& mdlRef = models[modelId];
-
     M2Instance instance;
     instance.id = nextInstanceId++;
     instance.modelId = modelId;
@@ -1696,24 +1657,20 @@ uint32_t M2Renderer::createInstanceWithMatrix(uint32_t modelId, const glm::mat4&
     instance.modelMatrix = modelMatrix;
     instance.invModelMatrix = glm::inverse(modelMatrix);
     glm::vec3 localMin, localMax;
-    getTightCollisionBounds(mdlRef, localMin, localMax);
+    getTightCollisionBounds(models[modelId], localMin, localMax);
     transformAABB(instance.modelMatrix, localMin, localMax, instance.worldBoundsMin, instance.worldBoundsMax);
     // Initialize animation
-    if (mdlRef.hasAnimation && !mdlRef.disableAnimation && !mdlRef.sequences.empty()) {
+    const auto& mdl2 = models[modelId];
+    if (mdl2.hasAnimation && !mdl2.disableAnimation && !mdl2.sequences.empty()) {
         instance.currentSequenceIndex = 0;
         instance.idleSequenceIndex = 0;
-        instance.animDuration = static_cast<float>(mdlRef.sequences[0].duration);
-        instance.animTime = static_cast<float>(rand() % std::max(1u, mdlRef.sequences[0].duration));
+        instance.animDuration = static_cast<float>(mdl2.sequences[0].duration);
+        instance.animTime = static_cast<float>(rand() % std::max(1u, mdl2.sequences[0].duration));
         instance.variationTimer = 3000.0f + static_cast<float>(rand() % 8000);
     } else {
         instance.animTime = static_cast<float>(rand()) / RAND_MAX * 10000.0f;
     }
 
-    // Pre-allocate bone SSBOs so first render frame doesn't hitch
-    if (mdlRef.hasAnimation && !mdlRef.disableAnimation) {
-        preallocateBoneBuffers(instance);
-    }
-
     instances.push_back(instance);
     size_t idx = instances.size() - 1;
     instanceIndexById[instance.id] = idx;
@@ -1854,11 +1811,7 @@ void M2Renderer::update(float deltaTime, const glm::vec3& cameraPos, const glm::
 
     // Cache camera state for frustum-culling bone computation
     cachedCamPos_ = cameraPos;
-    const size_t animInstCount = instances.size();
-    const float maxRenderDistance = (animInstCount > 3000) ? 600.0f
-                                 : (animInstCount > 2000) ? 800.0f
-                                 : (animInstCount > 1000) ? 1400.0f
-                                 :                          2800.0f;
+    const float maxRenderDistance = (instances.size() > 2000) ? 800.0f : 2800.0f;
     cachedMaxRenderDistSq_ = maxRenderDistance * maxRenderDistance;
 
     // Build frustum for culling bones
@@ -2128,12 +2081,8 @@ void M2Renderer::render(VkCommandBuffer cmd, VkDescriptorSet perFrameSet, const
 
     lastDrawCallCount = 0;
 
-    // Adaptive render distance: tiered for performance without excessive pop-in
-    const size_t instCount = instances.size();
-    const float maxRenderDistance = (instCount > 3000) ? 250.0f
-                                 : (instCount > 2000) ? 400.0f
-                                 : (instCount > 1000) ? 600.0f
-                                 :                      1000.0f;
+    // Adaptive render distance: balanced for performance without excessive pop-in
+    const float maxRenderDistance = (instances.size() > 2000) ? 350.0f : 1000.0f;
     const float maxRenderDistanceSq = maxRenderDistance * maxRenderDistance;
     const float fadeStartFraction = 0.75f;
     const glm::vec3 camPos = camera.getPosition();

src/rendering/wmo_renderer.cpp

@@ -1319,9 +1319,6 @@ void WMORenderer::render(VkCommandBuffer cmd, VkDescriptorSet perFrameSet, const
     bool doFrustumCull = false; // Temporarily disabled: can over-cull world WMOs
     bool doDistanceCull = distanceCulling;
 
-    // Cache active group info for distance-cull exemption (player's current WMO group)
-    const auto activeGroupCopy = activeGroup_;
-
     auto cullInstance = [&](size_t instIdx) -> InstanceDrawList {
         if (instIdx >= instances.size()) return InstanceDrawList{};
         const auto& instance = instances[instIdx];
@@ -1332,9 +1329,6 @@ void WMORenderer::render(VkCommandBuffer cmd, VkDescriptorSet perFrameSet, const
         InstanceDrawList result;
         result.instanceIndex = instIdx;
 
-        // Check if this instance is the one the player is standing in
-        bool isActiveInstance = activeGroupCopy.isValid() && activeGroupCopy.instanceIdx == instIdx;
-
         // Portal-based visibility
         std::unordered_set<uint32_t> portalVisibleGroups;
         bool usePortalCulling = doPortalCull && !model.portals.empty() && !model.portalRefs.empty();
@@ -1355,26 +1349,13 @@ void WMORenderer::render(VkCommandBuffer cmd, VkDescriptorSet perFrameSet, const
                 const auto& [gMin, gMax] = instance.worldGroupBounds[gi];
 
                 if (doDistanceCull) {
-                    // Never cull the group the player is standing in or its portal neighbors
-                    bool isExempt = false;
-                    if (isActiveInstance) {
-                        if (static_cast<int32_t>(gi) == activeGroupCopy.groupIdx) {
-                            isExempt = true;
-                        } else {
-                            for (uint32_t ng : activeGroupCopy.neighborGroups) {
-                                if (ng == static_cast<uint32_t>(gi)) { isExempt = true; break; }
-                            }
-                        }
-                    }
-                    if (!isExempt) {
                     glm::vec3 closestPoint = glm::clamp(camPos, gMin, gMax);
                     float distSq = glm::dot(closestPoint - camPos, closestPoint - camPos);
-                        if (distSq > maxGroupDistanceSq) {
+                    if (distSq > 250000.0f) {
                         result.distanceCulled++;
                         continue;
                     }
                 }
-                }
 
                 if (doFrustumCull && !frustum.intersectsAABB(gMin, gMax))
                     continue;