数学研发网设为首页收藏本站
切换到窄版

 找回密码
 欢迎注册
数学研发论坛»论坛 › 【算法研发】 › 业界资讯 › GotoBLAS 作者 Kazushige Goto 离开了 TACC
12下一页
返回列表 发新帖
查看: 24567|回复: 17

[讨论] GotoBLAS 作者 Kazushige Goto 离开了 TACC

[复制链接]
G-Spider
发表于 2010-12-5 17:53:29 | 显示全部楼层 |阅读模式

马上注册，结交更多好友，享用更多功能，让你轻松玩转社区。

您需要 登录 才可以下载或查看，没有账号？欢迎注册

×
Hi All,! I put GotoBLAS2-1.13.tar.gz on TACC web page. Since I leave TACC today, this is the final version of GotoBLAS2 distribution (except I find fatal mistake). 1. Fixed incorrect calculation of ZDOT 2. Added kernel for Fujitsu SPARC VI/VII Now total number of registered users is more than 16,000 and I would like to say thank you for using my library. My email account will be disabled soon and I’m not able to answer your question. But this mailing list is still available, TACC staff probably will answer your question. Thanks and good bye, Kazushige Goto 这是 Kazushige Goto 在 TACC 的 GotoBLAS maillist 上留下的最后一个 mail，表明他已经离开了 TACC。 ( ZKazushige Goto 编写的 GotoBLAS 是目前最快的 BLAS 库，不知道他离开 TACC 后是否会加入 GPU Computing 行列里， GotoBLAS - "the, currently, fastest implementation of the Basic Linear Algebra Subroutines (BLAS)" 维基了一下： Kazushige Goto From Wikipedia, the free encyclopediaJump to: navigation, search Kazushige Gotō (後藤和茂, Gotō Kazushige?) was a research associate at the Texas Advanced Computing Center at the University of Texas at Austin when he famously hand-optimized assembly routines for supercomputing and PC platforms that outperform the best compiler generated codes. Several of the fastest supercomputers in the world still use his implementation of the Basic Linear Algebra Subprograms (BLAS) known as “Goto BLAS”. He joined Microsoft's Technical Computing Group in 2010. 按这说法，他加入了微软的技术计算组。 看了一下，TACC站 GotoBLAS2 has been released by the Texas Advanced Computing Center as open source software under the BSD license. This product is no longer under active development by TACC, but it is being made available to the community to use, study, and extend. GotoBLAS2 uses new algorithms and memory techniques for optimal performance of the BLAS routines. The changes in this final version target new architecture features in microprocessors and interprocessor communication techniques; also, NUMA controls enhance multi-threaded execution of BLAS routines on node. The library features optimal performance on the following platforms: Intel Nehalem and Atom systems VIA Nanoprocessor AMD Shanghai and Istanbul The library includes the following features: •Configurations for a variety of hardware platforms •Incorporation of features of many ISAs (Instruction Set Architecture) •Implementation of NUMA controls to assure best process affinity and memory policy •Dynamic detection of multiple architecture components, which can be included in a single binary (for binary distributions) Originally developed by Kazushige Goto but is no longer under active development. For questions regarding the code, contact Dr. Kent Milfeld. 种种迹象表明 Kazushige Goto 确是离开了TACC 。
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复

使用道具 举报

liangbch
发表于 2010-12-9 08:09:15 | 显示全部楼层
正巧，我在并行算法导论中看到了这个Goto BLAS,下面的内容摘自《并行算法导论》附录 对于BLAS 库，现在有多种不同的优化实现，适用于Intel/Linux平台的主要有以下几种： BLAS 参考实现 这是一组标准Fortran 子程序，可以从BLAS 的主页下载：http://www.netlib.org/blas/index.html； ATLAS 库(Automatically Tuned Linear Algebra Software)它可以在不同平台上自动生成优化的BLAS 库，其主页为http://math-atlas.sourceforge.net/； Goto 库 Kazushige Goto 开发的一套高性能BLAS 库，其主页为http://www.cs.utexas.edu/users/flame/goto/； MKL 库(Math Kernel Library) Intel 为自己的CPU 专门优化的基本数学运算库，其中包含BLAS 库，其主页为http://www.intel.com/cd/software ... rflib/mkl/index.htm。 前三种库可以免费下载，而Intel MKL 库是商业软件，对于商业应用需要购买，而非商业应用可以免费使用。
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

无心人
发表于 2010-12-9 08:13:19 | 显示全部楼层
一直想入手一本并行算法的书
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

G-Spider
 楼主| 发表于 2010-12-10 10:46:40 | 显示全部楼层
面对 GotoBLAS2的源码就像面对着一个刺猬一样，还没找到一个很好的切入点。很难找到一个清晰的调用通道。或许可以从gprof中得到些线索。 貌似并没有用传统的blas的fortran子库(reference文件夹中)..... 不像最原始的BLAS思路清晰。光x86核的通用矩阵乘(gemm)子块就有近20种。 http://blog.csdn.net/G_Spider/archive/2010/12/04/6054764.aspx
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

G-Spider
 楼主| 发表于 2011-6-20 21:19:34 | 显示全部楼层
呵呵，又把这个帖子翻出来了。 刚才，编译出libgoto2.dll (core2模式)，vc下可以用了，一直纠结不知道怎么调用。 在这里总算找到了思路。 Naming conventions of CLAPACK routines libgoto2.dll 并没有经过CBLAS_***的包装。 均以 EXPORTS caxpy=caxpy_ @1 caxpy_=caxpy_ @2 CAXPY=caxpy_ @3 ccopy=ccopy_ @4 ccopy_=ccopy_ @5 ... 导出。 先把其中一个调用例子在这里备份一下：
  1. #include < stdio.h>
  3. void dgesv_(const int *N, const int *nrhs, double *A, const int *lda, int
  4. *ipiv, double *b, const int *ldb, int *info);
  6. void dgels_(const char *trans, const int *M, const int *N, const int *nrhs,
  7. double *A, const int *lda, double *b, const int *ldb, double *work,
  8. const int * lwork, int *info);
  10. int
  11. main(void)
  12. {
  13. /* 3x3 matrix A
  14. * 76 25 11
  15. * 27 89 51
  16. * 18 60 32
  17. */
  18. double A[9] = {76, 27, 18, 25, 89, 60, 11, 51, 32};
  19. double b[3] = {10, 7, 43};
  21. int N = 3;
  22. int nrhs = 1;
  23. int lda = 3;
  24. int ipiv[3];
  25. int ldb = 3;
  26. int info;
  28. dgesv_(&N, &nrhs, A, &lda, ipiv, b, &ldb, &info);
  30. if(info == 0) /* succeed */
  31. printf("The solution is %lf %lf %lf\n", b[0], b[1], b[2]);
  32. else
  33. fprintf(stderr, "dgesv_ fails %d\n", info);
  36. return info;
  37. }
复制代码
文件有点大，不过还是要共享一下，有兴趣的可以研究一下，嘿嘿。 下载
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

G-Spider
 楼主| 发表于 2011-6-21 10:15:40 | 显示全部楼层
整理了一份头文件gotoblas2.h。 示例：
  1. #include < stdio.h>
  2. #include "gotoblas2.h"
  3. #pragma comment(lib,"libgoto2.lib")
  5. //void dgesv_(const int *N, const int *nrhs, double *A, const int *lda, int
  6. // *ipiv, double *b, const int *ldb, int *info);
  8. int
  9. main(void)
  10. {
  11. /* 3x3 matrix A
  12. * 76 25 11
  13. * 27 89 51
  14. * 18 60 32
  15. */
  16. double A[9] = {76, 27, 18, 25, 89, 60, 11, 51, 32};
  17. double b[3] = {10, 7, 43};
  19. int N = 3;
  20. int nrhs = 1;
  21. int lda = 3;
  22. int ipiv[3];
  23. int ldb = 3;
  24. int info;
  26. dgesv_(&N, &nrhs, A, &lda, ipiv, b, &ldb, &info);
  28. if(info == 0) /* succeed */
  29. printf("The solution is %lf %lf %lf\n", b[0], b[1], b[2]);
  30. else
  31. fprintf(stderr, "dgesv_ fails %d\n", info);
  34. return info;
  35. }
复制代码
示例2： 含义参考
  1. #include <stdio.h>
  2. #include "gotoblas2.h"
  3. #pragma comment(lib,"libgoto2.lib")
  4. /////////////////////////////////
  5. // 直接声明
  6. // typedef struct {
  7. // float r, i;
  8. //} complex;
  9. //
  10. // complex cdotu(int *, complex *, int *, complex *, int *);
  11. /////////////////////////////////
  13. void main()
  14. {
  16. myccomplex_t x[]={ {1.0,2.0},{2.0,3.0},{3.0,4.0} };
  17. myccomplex_t y[]={ {2.0,1.0},{4.0,0.0},{0.0,9.0} };
  18. myccomplex_t z={0,0};
  19. int n=3,incx=1,incy=1;
  21. z=cdotu(&n,(float*)x,&incx,(float*)y,&incy);
  23. printf("%.2f+%.2fi\n",z.r,z.i); //-28.00+44.00i
  26. }
复制代码
头文件下载 gotoblas2.rar (3.46 KB, 下载次数: 12)
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

G-Spider
 楼主| 发表于 2011-6-23 18:52:01 | 显示全部楼层
发现上面的调用还不够刺激，直接试一下GotoBlas2的内核。 sasum求单精度浮点数组中，绝对值之和。 比如：float X1[4]={1.0, 2.0, 7.0, -8.0} ; 所有的元素绝对值之和为18. 在语法上改写sasum内核，必须支持sse2 ，保存下面的代码为：sasum_k.asm
  1. ;//GotoBlas2内核之sasum
  2. .686p
  3. .xmm ;支持SSE2
  4. .model flat,c
  5. option casemap :none
  7. .code
  9. sasum_k proc
  11. push esi
  12. push ebx
  13. mov ecx, [esp+8+4] ;//N
  14. mov esi, [esp+8+8] ;//X
  15. mov ebx, [esp+8+0ch] ;//INCX
  16. xorps xmm0, xmm0
  17. test ecx, ecx
  18. jle loc_2A0
  19. test ebx, ebx
  20. jle loc_2A0
  21. xorps xmm1, xmm1
  22. pcmpeqb xmm3, xmm3
  23. psrld xmm3, 1
  24. lea ebx, ds:0[ebx*4]
  25. cmp ebx, 4
  26. jnz loc_1F0
  27. sub esi, 0FFFFFF80h
  28. cmp ecx, 3
  29. jle loc_1B8
  30. test esi, 4
  31. jz short loc_68
  32. movss xmm0, dword ptr [esi-80h]
  33. andps xmm0, xmm3
  34. add esi, 4
  35. dec ecx
  36. jle loc_290
  37. nop
  38. lea esi, [esi+0]
  40. loc_68:
  41. test esi, 8
  42. jz short loc_88
  43. movsd xmm1, qword ptr [esi-80h]
  44. andps xmm1, xmm3
  45. add esi, 8
  46. sub ecx, 2
  47. jle loc_290
  48. lea esi, [esi+0]
  50. loc_88:
  51. mov eax, ecx
  52. sar eax, 5
  53. jle loc_148
  54. movaps xmm4, xmmword ptr [esi-80h]
  55. movaps xmm5, xmmword ptr [esi-70h]
  56. movaps xmm6, xmmword ptr [esi-60h]
  57. movaps xmm7, xmmword ptr [esi-50h]
  58. dec eax
  59. jle short loc_100
  60. db 66h
  61. nop
  63. loc_A8:
  64. andps xmm4, xmm3
  65. addps xmm0, xmm4
  66. movaps xmm4, xmmword ptr [esi-40h]
  67. andps xmm5, xmm3
  68. addps xmm1, xmm5
  69. movaps xmm5, xmmword ptr [esi-30h]
  70. andps xmm6, xmm3
  71. addps xmm0, xmm6
  72. movaps xmm6, xmmword ptr [esi-20h]
  73. andps xmm7, xmm3
  74. addps xmm1, xmm7
  75. movaps xmm7, xmmword ptr [esi-10h]
  76. andps xmm4, xmm3
  77. addps xmm0, xmm4
  78. movaps xmm4, xmmword ptr [esi]
  79. andps xmm5, xmm3
  80. addps xmm1, xmm5
  81. movaps xmm5, xmmword ptr [esi+10h]
  82. andps xmm6, xmm3
  83. addps xmm0, xmm6
  84. movaps xmm6, xmmword ptr [esi+20h]
  85. andps xmm7, xmm3
  86. addps xmm1, xmm7
  87. movaps xmm7, xmmword ptr [esi+30h]
  88. sub esi, 0FFFFFF80h
  89. dec eax
  90. jg short loc_A8
  91. lea esi, [esi+0]
  93. loc_100:
  94. andps xmm4, xmm3
  95. addps xmm0, xmm4
  96. movaps xmm4, xmmword ptr [esi-40h]
  97. andps xmm5, xmm3
  98. addps xmm1, xmm5
  99. movaps xmm5, xmmword ptr [esi-30h]
  100. andps xmm6, xmm3
  101. addps xmm0, xmm6
  102. movaps xmm6, xmmword ptr [esi-20h]
  103. andps xmm7, xmm3
  104. addps xmm1, xmm7
  105. movaps xmm7, xmmword ptr [esi-10h]
  106. andps xmm4, xmm3
  107. addps xmm0, xmm4
  108. andps xmm5, xmm3
  109. addps xmm1, xmm5
  110. andps xmm6, xmm3
  111. addps xmm0, xmm6
  112. andps xmm7, xmm3
  113. addps xmm1, xmm7
  114. sub esi, 0FFFFFF80h
  115. nop
  116. lea esi, [esi+0]
  118. loc_148:
  119. test ecx, 10h
  120. jz short loc_180
  121. movaps xmm4, xmmword ptr [esi-80h]
  122. andps xmm4, xmm3
  123. addps xmm0, xmm4
  124. movaps xmm5, xmmword ptr [esi-70h]
  125. andps xmm5, xmm3
  126. addps xmm1, xmm5
  127. movaps xmm6, xmmword ptr [esi-60h]
  128. andps xmm6, xmm3
  129. addps xmm0, xmm6
  130. movaps xmm7, xmmword ptr [esi-50h]
  131. andps xmm7, xmm3
  132. addps xmm1, xmm7
  133. add esi, 40h ; '@'
  134. nop
  135. lea esi, [esi+0]
  137. loc_180:
  138. test ecx, 8
  139. jz short loc_1A0
  140. movaps xmm4, xmmword ptr [esi-80h]
  141. andps xmm4, xmm3
  142. addps xmm0, xmm4
  143. movaps xmm5, xmmword ptr [esi-70h]
  144. andps xmm5, xmm3
  145. addps xmm1, xmm5
  146. add esi, 20h ; ' '
  147. nop
  149. loc_1A0:
  150. test ecx, 4
  151. jz short loc_1B8
  152. movaps xmm4, xmmword ptr [esi-80h]
  153. andps xmm4, xmm3
  154. addps xmm0, xmm4
  155. add esi, 10h
  156. lea esi, [esi+0]
  158. loc_1B8:
  159. test ecx, 2
  160. jz short loc_1D0
  161. movsd xmm4, qword ptr [esi-80h]
  162. andps xmm4, xmm3
  163. addps xmm1, xmm4
  164. add esi, 8
  165. db 66h
  166. nop
  168. loc_1D0:
  169. test ecx, 1
  170. jz loc_290
  171. movss xmm4, dword ptr [esi-80h]
  172. andps xmm4, xmm3
  173. addps xmm0, xmm4
  174. jmp loc_290
  175. align 10h
  177. loc_1F0:
  178. mov eax, ecx
  179. sar eax, 3
  180. jle short loc_270
  181. mov esi, esi
  182. lea edi, [edi+0]
  184. loc_200:
  185. movss xmm4, dword ptr [esi]
  186. add esi, ebx
  187. andps xmm4, xmm3
  188. addss xmm0, xmm4
  189. movss xmm5, dword ptr [esi]
  190. add esi, ebx
  191. andps xmm5, xmm3
  192. addss xmm1, xmm5
  193. movss xmm6, dword ptr [esi]
  194. add esi, ebx
  195. andps xmm6, xmm3
  196. addss xmm0, xmm6
  197. movss xmm7, dword ptr [esi]
  198. add esi, ebx
  199. andps xmm7, xmm3
  200. addss xmm1, xmm7
  201. movss xmm4, dword ptr [esi]
  202. add esi, ebx
  203. andps xmm4, xmm3
  204. addss xmm0, xmm4
  205. movss xmm5, dword ptr [esi]
  206. add esi, ebx
  207. andps xmm5, xmm3
  208. addss xmm1, xmm5
  209. movss xmm6, dword ptr [esi]
  210. add esi, ebx
  211. andps xmm6, xmm3
  212. addss xmm0, xmm6
  213. movss xmm7, dword ptr [esi]
  214. add esi, ebx
  215. andps xmm7, xmm3
  216. addss xmm1, xmm7
  217. dec eax
  218. jg short loc_200
  219. nop
  220. lea esi, [esi+0]
  222. loc_270:
  223. and ecx, 7
  224. jle short loc_290
  225. lea esi, [esi+0]
  226. lea edi, [edi+0]
  228. loc_280:
  229. movss xmm4, dword ptr [esi]
  230. andps xmm4, xmm3
  231. addss xmm0, xmm4
  232. add esi, ebx
  233. dec ecx
  234. jg short loc_280
  236. loc_290:
  237. addps xmm0, xmm1
  238. haddps xmm0, xmm0
  239. haddps xmm0, xmm0
  240. nop
  241. lea esi, [esi+0]
  243. loc_2A0:
  244. movss dword ptr [esp+8+4], xmm0
  245. fld dword ptr [esp+8+4]
  246. pop ebx
  247. pop esi
  248. ret
  249. sasum_k endp
  251. end
复制代码
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

G-Spider
 楼主| 发表于 2011-6-23 19:01:21 | 显示全部楼层
接着用vs2008或以上的版本的ml.exe编译上面的汇编生成sasum_k.obj. 命令行批处理如下：
  1. @echo off
  2. call "D:\Microsoft Visual Studio 10.0\VC\bin\vcvars32.bat"
  3. echo on
  4. ml /c /coff sasum_k.asm
  5. pause
复制代码
之后的c代码，只要连接这个sasum_k.obj文件即可。 c代码如下：
  1. #include <stdio.h>
  2. float sasum_k (int , float *, int ); //声明一下
  4. int
  5. main(void)
  6. {
  7. __declspec(align(16)) float X1[18]={1.0, 2.0, 7.0, -8.0, -5.0, -10.0, -9.0, 10.0,1.0, 2.0, 7.0, -8.0, -5.0, -10.0, -9.0, 10.0};
  9. float I1;
  10. int N;
  11. int INCX;
  13. N=16;
  14. INCX=1;
  15. I1=sasum_k( N , X1, INCX);
  16. printf(" The IASUM is %.3f\n",I1);
  18. system("\npause");
  19. return 0;
  20. }
  22. /*result
  23. 1 1;2 3;3 10;4 18; 5 23 ;6 33;7 42;8 52;9 53;10 55;11 62;12 70;13 75;14 85;15 94;16 104
  24. */
复制代码
编译命令批处理(保存为makeC.bat)：
  1. @echo off
  2. set VS=D:\vcPackaa
  4. call "D:\Microsoft Visual Studio 10.0\VC\bin\vcvars32.bat"
  6. echo on
  8. cl /c test.c
  9. link /subsystem:console test.obj sasum_k.obj
  11. pause
复制代码
GotoTest.rar (36.23 KB, 下载次数: 4)
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

天下无雪
发表于 2011-8-28 08:33:16 | 显示全部楼层
哎，可惜了哦，这么好的东西就终结了 貌似中国有个在继续做gotoblas，不过改名叫openblas
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

天下无雪
发表于 2011-8-28 14:58:10 | 显示全部楼层
1# G-Spider 请问下楼主知道有哪些比较好的计算稀疏矩阵的blas库？最好是要支持openmp的
毋因群疑而阻独见　　毋任己意而废人言
毋私小惠而伤大体　　毋借公论以快私情
回复 支持 反对

使用道具 举报

12下一页
返回列表 发新帖
高级模式
B Color Image Link Quote Code Smilies
您需要登录后才可以回帖 登录 | 欢迎注册

本版积分规则

小黑屋|手机版|数学研发网 ( 苏ICP备07505100号 )

GMT+8, 2024-12-22 01:13 , Processed in 0.032016 second(s), 19 queries .

Powered by Discuz! X3.5

© 2001-2024 Discuz! Team.

快速回复 返回顶部 返回列表