GOTOBLAS一般矩阵乘法高效实现机制的研究

对GOTOBLAS库(GOTO)的实现机制,尤其是其中的一般矩阵乘法部分的实现进行了分析.结合近年来的一些研究成果,讨论了如何高效地实现矩阵相乘操作,把存储层次对程序性能的影响提高到计算模型的高度.对比实验表明,GOTO库的性能远远高于没有考虑存储层次的一般BLAS库.证明了GOTO库性能上的优越性和将存储层次引入计算模型的必要性.

Preparation of Dendritic Copper Nanostructures and Their Characterization for Electroreduction

Dendritic copper nanostructures of different morphologies were synthesized by a surfactant-free electrochemical method. Single crystal nature of the nanostructures was revealed from their X-ray diffraction and electron diffraction patterns. Mechanism of dendrite formation was discussed from thermodynamic aspects using the concept of supersaturation. Supersaturation of the copper metal reduced on the surface of the electrode was the crucial factor for the generation of different morphologies. Effects of applied potential, temperature, and the solution concentration on the supersaturation were studied. The NO3- and H2O2 electroreduction ability of the dendritic materials was tested. Use of copper dendrite-modified electrode as NO3- sensor was demonstrated.