OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
Tree: v0.3.19
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'v0.3.19'
${ noResults }
OpenBLAS/common_level1.h

233 lines
12 kB
Raw Permalink Blame History 

  1. /*********************************************************************/

  2. /* Copyright 2009, 2010 The University of Texas at Austin.           */

  3. /* All rights reserved.                                              */

  4. /*                                                                   */

  5. /* Redistribution and use in source and binary forms, with or        */

  6. /* without modification, are permitted provided that the following   */

  7. /* conditions are met:                                               */

  8. /*                                                                   */

  9. /*   1. Redistributions of source code must retain the above         */

  10. /*      copyright notice, this list of conditions and the following  */

  11. /*      disclaimer.                                                  */

  12. /*                                                                   */

  13. /*   2. Redistributions in binary form must reproduce the above      */

  14. /*      copyright notice, this list of conditions and the following  */

  15. /*      disclaimer in the documentation and/or other materials       */

  16. /*      provided with the distribution.                              */

  17. /*                                                                   */

  18. /*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */

  19. /*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */

  20. /*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */

  21. /*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */

  22. /*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */

  23. /*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */

  24. /*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */

  25. /*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */

  26. /*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */

  27. /*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */

  28. /*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */

  29. /*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */

  30. /*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */

  31. /*    POSSIBILITY OF SUCH DAMAGE.                                    */

  32. /*                                                                   */

  33. /* The views and conclusions contained in the software and           */

  34. /* documentation are those of the authors and should not be          */

  35. /* interpreted as representing official policies, either expressed   */

  36. /* or implied, of The University of Texas at Austin.                 */

  37. /*********************************************************************/

  38. 

  39. #ifndef ASSEMBLER

  40. 

  41. #ifdef __CUDACC__

  42. extern "C" {

  43. #endif

  44. 

  45. float   sdot_k(BLASLONG, float   *, BLASLONG, float   *, BLASLONG);

  46. double dsdot_k(BLASLONG, float   *, BLASLONG, float *, BLASLONG);

  47. double  ddot_k(BLASLONG, double  *, BLASLONG, double  *, BLASLONG);

  48. xdouble qdot_k(BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  49. float  sbdot_k(BLASLONG, bfloat16 *, BLASLONG, bfloat16 *, BLASLONG);

  50. 

  51. void   sbstobf16_k(BLASLONG, float    *, BLASLONG, bfloat16 *, BLASLONG);

  52. void   sbdtobf16_k(BLASLONG, double   *, BLASLONG, bfloat16 *, BLASLONG);

  53. void   sbf16tos_k (BLASLONG, bfloat16 *, BLASLONG, float    *, BLASLONG);

  54. void   dbf16tod_k (BLASLONG, bfloat16 *, BLASLONG, double   *, BLASLONG);

  55. 

  56. openblas_complex_float cdotc_k (BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

  57. openblas_complex_float cdotu_k (BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

  58. openblas_complex_double zdotc_k (BLASLONG, double *, BLASLONG, double *, BLASLONG);

  59. openblas_complex_double zdotu_k (BLASLONG, double *, BLASLONG, double *, BLASLONG);

  60. openblas_complex_xdouble xdotc_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  61. openblas_complex_xdouble xdotu_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  62. 

  63. int    saxpy_k (BLASLONG, BLASLONG, BLASLONG, float,

  64. 	       float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

  65. int    daxpy_k (BLASLONG, BLASLONG, BLASLONG, double,

  66. 	       double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);

  67. int    qaxpy_k (BLASLONG, BLASLONG, BLASLONG, xdouble,

  68. 	       xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  69. int    caxpy_k (BLASLONG, BLASLONG, BLASLONG, float,  float,

  70. 	       float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

  71. int    zaxpy_k (BLASLONG, BLASLONG, BLASLONG, double, double,

  72. 	       double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);

  73. int    xaxpy_k (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble,

  74. 	       xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  75. int    caxpyc_k (BLASLONG, BLASLONG, BLASLONG, float,  float,

  76. 	       float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

  77. int    zaxpyc_k (BLASLONG, BLASLONG, BLASLONG, double, double,

  78. 	       double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);

  79. int    xaxpyc_k (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble,

  80. 	       xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  81. 

  82. int    scopy_k(BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

  83. int    dcopy_k(BLASLONG, double *, BLASLONG, double *, BLASLONG);

  84. int    qcopy_k(BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  85. int    ccopy_k(BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

  86. int    zcopy_k(BLASLONG, double *, BLASLONG, double *, BLASLONG);

  87. int    xcopy_k(BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

  88. 

  89. int    sswap_k (BLASLONG, BLASLONG, BLASLONG, float,

  90. 	       float  *, BLASLONG, float  *, BLASLONG, float *, BLASLONG);

  91. int    dswap_k (BLASLONG, BLASLONG, BLASLONG, double,

  92. 	       double *, BLASLONG, double *, BLASLONG, double*, BLASLONG);

  93. int    qswap_k (BLASLONG, BLASLONG, BLASLONG, xdouble,

  94. 	       xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble*, BLASLONG);

  95. int    cswap_k (BLASLONG, BLASLONG, BLASLONG, float,  float,

  96. 	       float  *, BLASLONG, float  *, BLASLONG, float *, BLASLONG);

  97. int    zswap_k (BLASLONG, BLASLONG, BLASLONG, double, double,

  98. 	       double *, BLASLONG, double *, BLASLONG, double*, BLASLONG);

  99. int    xswap_k (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble,

  100. 	       xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble*, BLASLONG);

  101. 

  102. float   sasum_k (BLASLONG, float  *, BLASLONG);

  103. double  dasum_k (BLASLONG, double *, BLASLONG);

  104. xdouble qasum_k (BLASLONG, xdouble *, BLASLONG);

  105. float   casum_k (BLASLONG, float  *, BLASLONG);

  106. double  zasum_k (BLASLONG, double *, BLASLONG);

  107. xdouble xasum_k (BLASLONG, xdouble *, BLASLONG);

  108. 

  109. float   ssum_k (BLASLONG, float  *, BLASLONG);

  110. double  dsum_k (BLASLONG, double *, BLASLONG);

  111. xdouble qsum_k (BLASLONG, xdouble *, BLASLONG);

  112. float   csum_k (BLASLONG, float  *, BLASLONG);

  113. double  zsum_k (BLASLONG, double *, BLASLONG);

  114. xdouble xsum_k (BLASLONG, xdouble *, BLASLONG);

  115. 

  116. float   samax_k (BLASLONG, float  *, BLASLONG);

  117. double  damax_k (BLASLONG, double *, BLASLONG);

  118. xdouble qamax_k (BLASLONG, xdouble *, BLASLONG);

  119. float   camax_k (BLASLONG, float  *, BLASLONG);

  120. double  zamax_k (BLASLONG, double *, BLASLONG);

  121. xdouble xamax_k (BLASLONG, xdouble *, BLASLONG);

  122. 

  123. float   samin_k (BLASLONG, float  *, BLASLONG);

  124. double  damin_k (BLASLONG, double *, BLASLONG);

  125. xdouble qamin_k (BLASLONG, xdouble *, BLASLONG);

  126. float   camin_k (BLASLONG, float  *, BLASLONG);

  127. double  zamin_k (BLASLONG, double *, BLASLONG);

  128. xdouble xamin_k (BLASLONG, xdouble *, BLASLONG);

  129. 

  130. BLASLONG   isamax_k(BLASLONG, float  *, BLASLONG);

  131. BLASLONG   idamax_k(BLASLONG, double *, BLASLONG);

  132. BLASLONG   iqamax_k(BLASLONG, xdouble *, BLASLONG);

  133. BLASLONG   icamax_k(BLASLONG, float  *, BLASLONG);

  134. BLASLONG   izamax_k(BLASLONG, double *, BLASLONG);

  135. BLASLONG   ixamax_k(BLASLONG, xdouble *, BLASLONG);

  136. 

  137. BLASLONG   isamin_k(BLASLONG, float  *, BLASLONG);

  138. BLASLONG   idamin_k(BLASLONG, double *, BLASLONG);

  139. BLASLONG   iqamin_k(BLASLONG, xdouble *, BLASLONG);

  140. BLASLONG   icamin_k(BLASLONG, float  *, BLASLONG);

  141. BLASLONG   izamin_k(BLASLONG, double *, BLASLONG);

  142. BLASLONG   ixamin_k(BLASLONG, xdouble *, BLASLONG);

  143. 

  144. float   smax_k (BLASLONG, float  *, BLASLONG);

  145. double  dmax_k (BLASLONG, double *, BLASLONG);

  146. xdouble qmax_k (BLASLONG, xdouble *, BLASLONG);

  147. float   cmax_k (BLASLONG, float  *, BLASLONG);

  148. double  zmax_k (BLASLONG, double *, BLASLONG);

  149. xdouble xmax_k (BLASLONG, xdouble *, BLASLONG);

  150. 

  151. float   smin_k (BLASLONG, float  *, BLASLONG);

  152. double  dmin_k (BLASLONG, double *, BLASLONG);

  153. xdouble qmin_k (BLASLONG, xdouble *, BLASLONG);

  154. float   cmin_k (BLASLONG, float  *, BLASLONG);

  155. double  zmin_k (BLASLONG, double *, BLASLONG);

  156. xdouble xmin_k (BLASLONG, xdouble *, BLASLONG);

  157. 

  158. BLASLONG   ismax_k(BLASLONG, float  *, BLASLONG);

  159. BLASLONG   idmax_k(BLASLONG, double *, BLASLONG);

  160. BLASLONG   iqmax_k(BLASLONG, xdouble *, BLASLONG);

  161. BLASLONG   icmax_k(BLASLONG, float  *, BLASLONG);

  162. BLASLONG   izmax_k(BLASLONG, double *, BLASLONG);

  163. BLASLONG   ixmax_k(BLASLONG, xdouble *, BLASLONG);

  164. 

  165. BLASLONG   ismin_k(BLASLONG, float  *, BLASLONG);

  166. BLASLONG   idmin_k(BLASLONG, double *, BLASLONG);

  167. BLASLONG   iqmin_k(BLASLONG, xdouble *, BLASLONG);

  168. BLASLONG   icmin_k(BLASLONG, float  *, BLASLONG);

  169. BLASLONG   izmin_k(BLASLONG, double *, BLASLONG);

  170. BLASLONG   ixmin_k(BLASLONG, xdouble *, BLASLONG);

  171. 

  172. int    sscal_k(BLASLONG, BLASLONG, BLASLONG, float,

  173. 	      float *, BLASLONG, float *, BLASLONG,  float  *, BLASLONG);

  174. int    dscal_k(BLASLONG, BLASLONG, BLASLONG, double,

  175. 	      double *, BLASLONG, double *, BLASLONG,  double  *, BLASLONG);

  176. int    qscal_k(BLASLONG, BLASLONG, BLASLONG, xdouble,

  177. 	      xdouble *, BLASLONG, xdouble *, BLASLONG,  xdouble  *, BLASLONG);

  178. int    cscal_k(BLASLONG, BLASLONG, BLASLONG, float, float,

  179. 	      float *, BLASLONG, float *, BLASLONG,  float  *, BLASLONG);

  180. int    zscal_k(BLASLONG, BLASLONG, BLASLONG, double, double,

  181. 	      double *, BLASLONG, double *, BLASLONG,  double  *, BLASLONG);

  182. int    xscal_k(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble,

  183. 	      xdouble *, BLASLONG, xdouble *, BLASLONG,  xdouble  *, BLASLONG);

  184. int    csscal_k(BLASLONG, BLASLONG, BLASLONG, float, float,

  185. 	       float *, BLASLONG, float *, BLASLONG,  float  *, BLASLONG);

  186. int    zdscal_k(BLASLONG, BLASLONG, BLASLONG, double, double,

  187. 	       double *, BLASLONG, double *, BLASLONG,  double  *, BLASLONG);

  188. int    xqscal_k(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble,

  189. 	       xdouble *, BLASLONG, xdouble *, BLASLONG,  xdouble  *, BLASLONG);

  190. 

  191. float   snrm2_k(BLASLONG, float   *, BLASLONG);

  192. double  dnrm2_k(BLASLONG, double  *, BLASLONG);

  193. xdouble qnrm2_k(BLASLONG, xdouble *, BLASLONG);

  194. float   cnrm2_k(BLASLONG, float   *, BLASLONG);

  195. double  znrm2_k(BLASLONG, double  *, BLASLONG);

  196. xdouble xnrm2_k(BLASLONG, xdouble *, BLASLONG);

  197. 

  198. int    srot_k (BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float , float );

  199. int    drot_k (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);

  200. int    qrot_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);

  201. int    csrot_k(BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float , float );

  202. int    zdrot_k(BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);

  203. int    xqrot_k(BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);

  204. 

  205. int    srotg_k(float  *, float  *, float  *, float  *);

  206. int    drotg_k(double *, double *, double *, double *);

  207. int    qrotg_k(xdouble *, xdouble *, xdouble *, xdouble *);

  208. int    csrotg_k(float  *, float  *, float  *, float  *);

  209. int    zdrotg_k(double *, double *, double *, double *);

  210. int    xqrotg_k(xdouble *, xdouble *, xdouble *, xdouble *);

  211. 

  212. int    srotmg_k(float  *, float  *, float  *, float  *, float  *);

  213. int    drotmg_k(double *, double *, double *, double *, double *);

  214. int    qrotmg_k(xdouble *, xdouble *, xdouble *, xdouble *, xdouble *);

  215. 

  216. int    srotm_k (BLASLONG, float,  BLASLONG, float,  BLASLONG, float);

  217. int    drotm_k (BLASLONG, double, BLASLONG, double, BLASLONG, double);

  218. int    qrotm_k (BLASLONG, xdouble, BLASLONG, xdouble, BLASLONG, xdouble);

  219. 

  220. 

  221. int    saxpby_k (BLASLONG, float,  float  *, BLASLONG, float,  float  *, BLASLONG);

  222. int    daxpby_k (BLASLONG, double, double *, BLASLONG, double, double *, BLASLONG);

  223. int    caxpby_k (BLASLONG, float,  float,  float  *, BLASLONG, float,  float,  float  *, BLASLONG);

  224. int    zaxpby_k (BLASLONG, double, double, double *, BLASLONG, double, double, double *, BLASLONG);

  225. 

  226. 

  227. #ifdef __CUDACC__

  228. }

  229. #endif

  230. 

  231. #endif