Merge pull request NVlabs#128 from NVlabs/memory-arena

Significant memory reduction through TCNN's new memory arena
fnysalehi · Feb 10, 2022 · 6d23723 · 6d23723
2 parents d44b17f + 6589780
commit 6d23723
Show file tree

Hide file tree

Showing 6 changed files with 138 additions and 227 deletions.
diff --git a/dependencies/tiny-cuda-nn b/dependencies/tiny-cuda-nn
diff --git a/include/neural-graphics-primitives/marching_cubes.h b/include/neural-graphics-primitives/marching_cubes.h
@@ -20,7 +20,7 @@ NGP_NAMESPACE_BEGIN
 
 Eigen::Vector3i get_marching_cubes_res(uint32_t res_1d, const BoundingBox &aabb);
 
-void marching_cubes_gpu(tcnn::GPUMemory<char>& scratch_memory, BoundingBox aabb, Eigen::Vector3i res_3d, float thresh, const tcnn::GPUMemory<float>& density, tcnn::GPUMemory<Eigen::Vector3f>& vert_out, tcnn::GPUMemory<uint32_t>& indices_out);
+void marching_cubes_gpu(cudaStream_t stream, BoundingBox aabb, Eigen::Vector3i res_3d, float thresh, const tcnn::GPUMemory<float>& density, tcnn::GPUMemory<Eigen::Vector3f>& vert_out, tcnn::GPUMemory<uint32_t>& indices_out);
 
 // computes the average of the 1ring of all verts, as homogenous coordinates
 void compute_mesh_1ring(const tcnn::GPUMemory<Eigen::Vector3f>& verts, const tcnn::GPUMemory<uint32_t>& indices, tcnn::GPUMemory<Eigen::Vector4f>& output_pos, tcnn::GPUMemory<Eigen::Vector3f>& output_normals);
+19 −3		include/tiny-cuda-nn/common.h
+2 −2		include/tiny-cuda-nn/cutlass_matmul.h
+56 −58		include/tiny-cuda-nn/encodings/grid.h
+105 −17		include/tiny-cuda-nn/gpu_matrix.h
+227 −71		include/tiny-cuda-nn/gpu_memory.h
+1 −0		include/tiny-cuda-nn/network.h
+38 −93		include/tiny-cuda-nn/network_with_input_encoding.h
+15 −18		include/tiny-cuda-nn/networks/cutlass_mlp.h
+18 −21		include/tiny-cuda-nn/networks/cutlass_resnet.h
+17 −18		include/tiny-cuda-nn/networks/fully_fused_mlp.h
+8 −26		include/tiny-cuda-nn/object.h
+1 −1		include/tiny-cuda-nn/reduce_sum.h
+5 −4		samples/mlp_learning_an_image.cu
+12 −0		src/common.cu
+51 −81		src/cutlass_mlp.cu
+62 −92		src/cutlass_resnet.cu
+59 −78		src/fully_fused_mlp.cu
+7 −7		src/network.cu