Update utilities for parallel processing and memory arena.

tatsy · Jul 28, 2018 · d9a2ee7 · d9a2ee7
1 parent f929ce6
commit d9a2ee7
Show file tree

Hide file tree

Showing 6 changed files with 177 additions and 107 deletions.
diff --git a/sources/accelerators/bvh.cc b/sources/accelerators/bvh.cc
@@ -39,7 +39,7 @@ static const int orderTable[] = {
     0x40123, 0x40132, 0x41023, 0x41032, 0x42301, 0x43201, 0x42310, 0x43210,  // ++|++
 };
 
-int test_AABB(const __m128 bboxes[2][3], const __m128 org[3], const __m128 idir[3], const int sign[3], __m128 tmin, __m128 tmax) {
+int test_AABB(const __m128 bboxes[2][3], const __m128 org[3], const __m128 idir[3], const int sign[3], __m128 tmin, __m128 tmax) noexcept {
     tmin = _mm_max_ps(tmin, _mm_mul_ps(_mm_sub_ps(bboxes[sign[0]][0], org[0]), idir[0]));
     tmax = _mm_min_ps(tmax, _mm_mul_ps(_mm_sub_ps(bboxes[1 - sign[0]][0], org[0]), idir[0]));
     tmin = _mm_max_ps(tmin, _mm_mul_ps(_mm_sub_ps(bboxes[sign[1]][1], org[1]), idir[1]));
@@ -134,6 +134,7 @@ BVHAccel::BVHAccel(const std::vector<std::shared_ptr<Primitive>> &prims,
 }
 
 BVHAccel::~BVHAccel() {
+    release();
 }
 
 Bounds3d BVHAccel::worldBound() const {
@@ -144,7 +145,7 @@ void BVHAccel::construct() {
     if (primitives_.empty()) return;
 
     std::vector<BVHPrimitiveInfo> primitiveInfo(primitives_.size());
-    for (int i = 0; i < primitives_.size(); i++) {
+    for (int i = 0; i < (int)primitives_.size(); i++) {
         primitiveInfo[i] = { i, primitives_[i]->worldBound() };
     }
 
@@ -241,10 +242,10 @@ BVHNode* BVHAccel::constructRec(std::vector<BVHPrimitiveInfo>& buildData,
 }
 
 void BVHAccel::release() {
-    for (int i = 0; i < simdNodes_.size(); i++) {
-        align_free(simdNodes_[i]);
+    for (auto &node : simdNodes_) {
+        align_free(node);
     }
-    
+
     root_ = nullptr;
     simdNodes_.clear();
 }
@@ -253,8 +254,7 @@ void BVHAccel::collapse2QBVH(BVHNode* node) {
     BVHNode *lc = node->left;
     BVHNode *rc = node->right;
 
-    SIMDBVHNode* n =
-    static_cast<SIMDBVHNode*>(align_alloc(sizeof(SIMDBVHNode), 16));
+    SIMDBVHNode* n = (SIMDBVHNode *)align_alloc(sizeof(SIMDBVHNode), 16);
     Assertion(n != nullptr, "allocation failed !!");
 
     simdNodes_.push_back(n);
@@ -337,7 +337,6 @@ bool BVHAccel::intersectBVH(Ray &ray, SurfaceInteraction *isect) const{
     nodeStack.push(root_);
 
     bool hit = false;
-    int cnt = 0;
     while (!nodeStack.empty()) {
         BVHNode* node = nodeStack.top();
         nodeStack.pop();
@@ -367,7 +366,6 @@ bool BVHAccel::intersectBVH(Ray& ray) const {
     std::stack<BVHNode*> nodeStack;
     nodeStack.push(root_);
 
-    bool hit = false;
     while (!nodeStack.empty()) {
         BVHNode* node = nodeStack.top();
         nodeStack.pop();

diff --git a/sources/core/memory.cc b/sources/core/memory.cc
@@ -2,6 +2,7 @@
 #include "memory.h"
 
 #include <cstdlib>
+#include <cstddef>
 #include <algorithm>
 
 namespace spica {
@@ -22,12 +23,13 @@ MemoryArena::MemoryArena(MemoryArena&& arena) noexcept
 
 MemoryArena::~MemoryArena() {
     align_free(currentBlock_);
-    for (auto& block : usedBlocks_)      align_free(block.second);
+    for (auto& block : usedBlocks_) align_free(block.second);
     for (auto& block : availableBlocks_) align_free(block.second);
 }
 
 void* MemoryArena::allocBytes(size_t nBytes) {
-    nBytes = ((nBytes + 0xf) & (~0xf));
+    const int align = alignof(std::max_align_t);
+    nBytes = ((nBytes + align - 1) & ~(align - 1));
     if (currentBlockPos_ + nBytes > currentAllocSize_) {
         if (currentBlock_) {
             usedBlocks_.emplace_back(currentAllocSize_, currentBlock_);
@@ -39,15 +41,15 @@ void* MemoryArena::allocBytes(size_t nBytes) {
             iter != availableBlocks_.end(); ++iter) {
             if (iter->first >= nBytes) {
                 currentAllocSize_ = iter->first;
-                currentBlock_     = iter->second;
+                currentBlock_ = iter->second;
                 availableBlocks_.erase(iter);
                 break;
             }
         }
 
         if (!currentBlock_) {
             currentAllocSize_ = std::max(nBytes, blockSize_);
-            currentBlock_     = (unsigned char*)align_alloc(currentAllocSize_, 64);
+            currentBlock_ = (uint8_t*)align_alloc(currentAllocSize_, 64);
         }
         currentBlockPos_ = 0;
     }
@@ -63,8 +65,8 @@ void MemoryArena::reset() {
 
 size_t MemoryArena::totalAllocated() const {
     size_t total = currentAllocSize_;
-    for (auto& block : usedBlocks_)      total += block.first;
-    for (auto& block : availableBlocks_) total += block.first;
+    for (const auto &block : usedBlocks_)      total += block.first;
+    for (const auto &block : availableBlocks_) total += block.first;
     return total;
 }
 

diff --git a/sources/core/memory.h b/sources/core/memory.h
@@ -43,7 +43,7 @@ class SPICA_EXPORTS MEM_ALIGN(64) MemoryArena : private Uncopyable {
         using Elem = typename std::remove_extent<T>::type;
         Elem *ret = (Elem*)allocBytes(sizeof(Elem) * size);
         for (size_t i = 0; i < size; i++) {
-            new (ret + i) Elem();
+            new (&ret[i]) Elem();
         }
         return ret;
     }
@@ -60,7 +60,7 @@ class SPICA_EXPORTS MEM_ALIGN(64) MemoryArena : private Uncopyable {
     const size_t blockSize_;
     size_t currentBlockPos_ = 0;
     size_t currentAllocSize_ = 0;
-    unsigned char* currentBlock_ = nullptr;
+    uint8_t* currentBlock_ = nullptr;
     std::list<std::pair<size_t, unsigned char*>> usedBlocks_, availableBlocks_;
 };  // class MemoryArena