視錐體加速性能對比實驗記錄

來自專欄 Render Engine1 人贊了文章

Octree frustum culling VS Linear Culling Preformance Contrast

Octree travalled with a BFS-order instead of DFS for memory cache coherent and Linear Culling are both used __mm256 SIMD for accelerating intersection detection. Here is Code! We assume six plane has been compressed to mm256 matrix as sMat,

SIMD::AMVECTOR S=SIMD::AMVec2Transform(iter.aMax.v, sMat); auto V1= _mm256_cmp_ps(iter.aMIn.v, S, _MM_CMPINT_LT); auto c = _mm256_castps_si256(V1); __m256i V2 = _mm256_set1_epi32(-1); __m256i vTemp = _mm256_cmpeq_epi32(_mm256_castps_si256(V1), V2); int iTest = _mm256_movemask_ps(_mm256_castsi256_ps(vTemp)); if(!iTest) sIndex.push_back(iter.ObjectIndex);

we transform two same vector came from Sphere centre,whose w component is radius of sphere.Then we compare the result and radius to measure detection.We tested million cube quantity.

Linear preformance:

Octree Preformance:

推薦閱讀：