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: e5313f53d5
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e5313f53d5'
${ noResults }
OpenBLAS/kernel/riscv64/trmmkernel_rvv_v1x8.c

686 lines
24 kB
Raw Blame History 

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

  2. Copyright (c) 2022, The OpenBLAS Project

  3. All rights reserved.

  4. Redistribution and use in source and binary forms, with or without

  5. modification, are permitted provided that the following conditions are

  6. met:

  7. 1. Redistributions of source code must retain the above copyright

  8. notice, this list of conditions and the following disclaimer.

  9. 2. Redistributions in binary form must reproduce the above copyright

  10. notice, this list of conditions and the following disclaimer in

  11. the documentation and/or other materials provided with the

  12. distribution.

  13. 3. Neither the name of the OpenBLAS project nor the names of

  14. its contributors may be used to endorse or promote products

  15. derived from this software without specific prior written permission.

  16. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"

  17. AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE

  18. IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE

  19. ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE

  20. LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL

  21. DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR

  22. SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER

  23. CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,

  24. OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE

  25. USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

  26. *****************************************************************************/

  27. 

  28. #include "common.h"

  29. 

  30. #if !defined(DOUBLE)

  31. #define VSETVL(n) vsetvl_e32m2(n)

  32. #define FLOAT_V_T vfloat32m2_t

  33. #define VLEV_FLOAT vle32_v_f32m2

  34. #define VSEV_FLOAT vse32_v_f32m2

  35. #define VFMVVF_FLOAT vfmv_v_f_f32m2

  36. #define VFMACCVF_FLOAT vfmacc_vf_f32m2

  37. #define VFMULVF_FLOAT vfmul_vf_f32m2

  38. #else

  39. #define VSETVL(n) vsetvl_e64m2(n)

  40. #define FLOAT_V_T vfloat64m2_t

  41. #define VLEV_FLOAT vle64_v_f64m2

  42. #define VSEV_FLOAT vse64_v_f64m2

  43. #define VFMVVF_FLOAT vfmv_v_f_f64m2

  44. #define VFMACCVF_FLOAT vfmacc_vf_f64m2

  45. #define VFMULVF_FLOAT vfmul_vf_f64m2

  46. #endif

  47. 

  48. 

  49. // Optimizes the implementation in ../generic/trmmkernel_8x8.c

  50. 

  51. 

  52. int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc ,BLASLONG offset)

  53. {

  54.     //fprintf(stderr, "%s, %s, bm=%4ld bn=%4ld bk=%4ld alpha=%f ldc=%ld\n", __FILE__, __FUNCTION__, bm, bn, bk, alpha, ldc);

  55. 

  56.     BLASLONG i,j,k;

  57.     FLOAT *C0,*C1,*C2,*C3,*C4,*C5,*C6,*C7,*ptrba,*ptrbb;

  58. 

  59.     FLOAT_V_T va0, va1, va2, va3, va4, va5, va6, va7;

  60.     FLOAT_V_T vres0, vres1, vres2, vres3, vres4, vres5, vres6, vres7;

  61.     size_t vl;

  62. 

  63.     BLASLONG off, temp;

  64. 

  65. #if !defined(LEFT)

  66.     off = -offset;

  67. #else

  68.     off = 0;

  69. #endif

  70.     for (j = bn/8; j > 0; j--)

  71.     {

  72.         C0 = C;

  73.         C1 = C0+ldc;

  74.         C2 = C1+ldc;

  75.         C3 = C2+ldc;

  76.         C4 = C3+ldc;

  77.         C5 = C4+ldc;

  78.         C6 = C5+ldc;

  79.         C7 = C6+ldc;

  80. 

  81. #if defined(TRMMKERNEL) && defined(LEFT)

  82.         off = offset;

  83. #endif

  84. 

  85.         ptrba = ba;

  86. 

  87.         for (i = bm; i > 0; i -= vl)

  88.         {

  89.             vl = VSETVL(i);

  90. 

  91. #if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  92.             ptrbb = bb;

  93. #else

  94.             ptrba += off*vl;

  95.             ptrbb = bb + off*8;

  96. #endif

  97. 

  98.             vres0 = VFMVVF_FLOAT(0.0, vl);

  99.             vres1 = VFMVVF_FLOAT(0.0, vl);

  100.             vres2 = VFMVVF_FLOAT(0.0, vl);

  101.             vres3 = VFMVVF_FLOAT(0.0, vl);

  102.             vres4 = VFMVVF_FLOAT(0.0, vl);

  103.             vres5 = VFMVVF_FLOAT(0.0, vl);

  104.             vres6 = VFMVVF_FLOAT(0.0, vl);

  105.             vres7 = VFMVVF_FLOAT(0.0, vl);

  106. 

  107. #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))

  108.             temp = bk-off;

  109. #elif defined(LEFT)

  110.             temp = off+vl;  // number of values in A

  111. #else

  112.             temp = off+8;   // number of values in B

  113. #endif

  114. 

  115.             for (k = temp/8; k > 0; k--) {

  116.                 va0 = VLEV_FLOAT(ptrba, vl);

  117.                 ptrba += vl;

  118.                 va1 = VLEV_FLOAT(ptrba, vl);

  119.                 ptrba += vl;

  120. 

  121.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  122.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va0, vl);

  123.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va0, vl);

  124.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va0, vl);

  125.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va0, vl);

  126.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va0, vl);

  127.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va0, vl);

  128.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va0, vl);

  129.                 ptrbb += 8;

  130.                 va2 = VLEV_FLOAT(ptrba, vl);

  131.                 ptrba += vl;

  132. 

  133.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va1, vl);

  134.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va1, vl);

  135.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va1, vl);

  136.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va1, vl);

  137.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va1, vl);

  138.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va1, vl);

  139.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va1, vl);

  140.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va1, vl);

  141.                 ptrbb += 8;

  142.                 va3 = VLEV_FLOAT(ptrba, vl);

  143.                 ptrba += vl;

  144. 

  145.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va2, vl);

  146.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va2, vl);

  147.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va2, vl);

  148.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va2, vl);

  149.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va2, vl);

  150.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va2, vl);

  151.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va2, vl);

  152.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va2, vl);

  153.                 ptrbb += 8;

  154.                 va4 = VLEV_FLOAT(ptrba, vl);

  155.                 ptrba += vl;

  156. 

  157.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va3, vl);

  158.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va3, vl);

  159.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va3, vl);

  160.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va3, vl);

  161.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va3, vl);

  162.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va3, vl);

  163.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va3, vl);

  164.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va3, vl);

  165.                 ptrbb += 8;

  166.                 va5 = VLEV_FLOAT(ptrba, vl);

  167.                 ptrba += vl;

  168. 

  169.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va4, vl);

  170.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va4, vl);

  171.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va4, vl);

  172.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va4, vl);

  173.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va4, vl);

  174.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va4, vl);

  175.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va4, vl);

  176.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va4, vl);

  177.                 ptrbb += 8;

  178.                 va6 = VLEV_FLOAT(ptrba, vl);

  179.                 ptrba += vl;

  180. 

  181.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va5, vl);

  182.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va5, vl);

  183.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va5, vl);

  184.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va5, vl);

  185.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va5, vl);

  186.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va5, vl);

  187.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va5, vl);

  188.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va5, vl);

  189.                 ptrbb += 8;

  190.                 va7 = VLEV_FLOAT(ptrba, vl);

  191.                 ptrba += vl;

  192. 

  193.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va6, vl);

  194.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va6, vl);

  195.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va6, vl);

  196.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va6, vl);

  197.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va6, vl);

  198.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va6, vl);

  199.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va6, vl);

  200.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va6, vl);

  201.                 ptrbb += 8;

  202. 

  203.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va7, vl);

  204.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va7, vl);

  205.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va7, vl);

  206.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va7, vl);

  207.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va7, vl);

  208.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va7, vl);

  209.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va7, vl);

  210.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va7, vl);

  211.                 ptrbb += 8;

  212.             }

  213. 

  214.             for (k = temp&7; k > 0; k--) {

  215.                 va0 = VLEV_FLOAT(ptrba, vl); // M:8 (should be vlen);

  216. 

  217.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  218.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va0, vl);

  219.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va0, vl);

  220.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va0, vl);

  221.                 vres4 = VFMACCVF_FLOAT(vres4, *(ptrbb + 4), va0, vl);

  222.                 vres5 = VFMACCVF_FLOAT(vres5, *(ptrbb + 5), va0, vl);

  223.                 vres6 = VFMACCVF_FLOAT(vres6, *(ptrbb + 6), va0, vl);

  224.                 vres7 = VFMACCVF_FLOAT(vres7, *(ptrbb + 7), va0, vl);

  225. 

  226.                 ptrbb += 8;

  227.                 ptrba += vl;

  228.             }

  229. 

  230.             va0 = VFMULVF_FLOAT(vres0, alpha, vl);

  231.             VSEV_FLOAT(C0, va0, vl);

  232. 

  233.             va1 = VFMULVF_FLOAT(vres1, alpha, vl);

  234.             VSEV_FLOAT(C1, va1, vl);

  235. 

  236.             va2 = VFMULVF_FLOAT(vres2, alpha, vl);

  237.             VSEV_FLOAT(C2, va2, vl);

  238. 

  239.             va3 = VFMULVF_FLOAT(vres3, alpha, vl);

  240.             VSEV_FLOAT(C3, va3, vl);

  241. 

  242.             va4 = VFMULVF_FLOAT(vres4, alpha, vl);

  243.             VSEV_FLOAT(C4, va4, vl);

  244. 

  245.             va5 = VFMULVF_FLOAT(vres5, alpha, vl);

  246.             VSEV_FLOAT(C5, va5, vl);

  247. 

  248.             va6 = VFMULVF_FLOAT(vres6, alpha, vl);

  249.             VSEV_FLOAT(C6, va6, vl);

  250. 

  251.             va7 = VFMULVF_FLOAT(vres7, alpha, vl);

  252.             VSEV_FLOAT(C7, va7, vl);

  253. 

  254. 

  255. #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  256.             temp = bk - off;

  257. #ifdef LEFT

  258.             temp -= vl; // number of values in A

  259. #else

  260.             temp -= 8; // number of values in B

  261. #endif

  262.             ptrba += temp*vl;

  263.             ptrbb += temp*8;

  264. #endif

  265. 

  266. #ifdef LEFT

  267.             off += vl; // number of values in A

  268. #endif

  269. 

  270.             C0 += vl;

  271.             C1 += vl;

  272.             C2 += vl;

  273.             C3 += vl;

  274.             C4 += vl;

  275.             C5 += vl;

  276.             C6 += vl;

  277.             C7 += vl;

  278.         }

  279. 

  280. #if defined(TRMMKERNEL) && !defined(LEFT)

  281.         off += 8;

  282. #endif

  283. 

  284.         bb += (bk<<3);

  285.         C += (ldc<<3);

  286.     }

  287. 

  288.     if (bn & 4)

  289.     {

  290.         C0 = C;

  291.         C1 = C0+ldc;

  292.         C2 = C1+ldc;

  293.         C3 = C2+ldc;

  294. 

  295. #if defined(TRMMKERNEL) && defined(LEFT)

  296.         off = offset;

  297. #endif

  298.         ptrba = ba;

  299. 

  300.         for (i = bm; i > 0; i -= vl)

  301.         {

  302.             vl = VSETVL(i);

  303. #if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  304.             ptrbb = bb;

  305. #else

  306.             ptrba += off*vl;

  307.             ptrbb = bb + off*4;

  308. #endif

  309. 

  310.             vres0 = VFMVVF_FLOAT(0.0, vl);

  311.             vres1 = VFMVVF_FLOAT(0.0, vl);

  312.             vres2 = VFMVVF_FLOAT(0.0, vl);

  313.             vres3 = VFMVVF_FLOAT(0.0, vl);

  314. 

  315. 

  316. #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))

  317.             temp = bk-off;

  318. #elif defined(LEFT)

  319.             temp = off+vl;  // number of values in A

  320. #else

  321.             temp = off+4;   // number of values in B

  322. #endif

  323. 

  324.             for (k = temp/8; k > 0; k--) {

  325.                 va0 = VLEV_FLOAT(ptrba, vl);

  326.                 ptrba += vl;

  327.                 va1 = VLEV_FLOAT(ptrba, vl);

  328.                 ptrba += vl;

  329. 

  330.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  331.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va0, vl);

  332.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va0, vl);

  333.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va0, vl);

  334.                 ptrbb += 4;

  335.                 va2 = VLEV_FLOAT(ptrba, vl);

  336.                 ptrba += vl;

  337. 

  338.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va1, vl);

  339.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va1, vl);

  340.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va1, vl);

  341.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va1, vl);

  342.                 ptrbb += 4;

  343.                 va3 = VLEV_FLOAT(ptrba, vl);

  344.                 ptrba += vl;

  345. 

  346.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va2, vl);

  347.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va2, vl);

  348.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va2, vl);

  349.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va2, vl);

  350.                 ptrbb += 4;

  351.                 va4 = VLEV_FLOAT(ptrba, vl);

  352.                 ptrba += vl;

  353. 

  354.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va3, vl);

  355.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va3, vl);

  356.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va3, vl);

  357.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va3, vl);

  358.                 ptrbb += 4;

  359.                 va5 = VLEV_FLOAT(ptrba, vl);

  360.                 ptrba += vl;

  361. 

  362.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va4, vl);

  363.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va4, vl);

  364.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va4, vl);

  365.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va4, vl);

  366.                 ptrbb += 4;

  367.                 va6 = VLEV_FLOAT(ptrba, vl);

  368.                 ptrba += vl;

  369. 

  370.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va5, vl);

  371.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va5, vl);

  372.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va5, vl);

  373.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va5, vl);

  374.                 ptrbb += 4;

  375.                 va7 = VLEV_FLOAT(ptrba, vl);

  376.                 ptrba += vl;

  377. 

  378.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va6, vl);

  379.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va6, vl);

  380.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va6, vl);

  381.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va6, vl);

  382.                 ptrbb += 4;

  383. 

  384.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va7, vl);

  385.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va7, vl);

  386.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va7, vl);

  387.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va7, vl);

  388.                 ptrbb += 4;

  389.             }

  390. 

  391.             // K remainder

  392.             for (k = temp&7; k > 0; k--) {

  393.                 va0 = VLEV_FLOAT(ptrba, vl);

  394. 

  395.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  396.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va0, vl);

  397.                 vres2 = VFMACCVF_FLOAT(vres2, *(ptrbb + 2), va0, vl);

  398.                 vres3 = VFMACCVF_FLOAT(vres3, *(ptrbb + 3), va0, vl);

  399. 

  400.                 ptrbb += 4;

  401.                 ptrba += vl;

  402.             }

  403. 

  404.             va0 = VFMULVF_FLOAT(vres0, alpha, vl);

  405.             VSEV_FLOAT(C0, va0, vl);

  406. 

  407.             va1 = VFMULVF_FLOAT(vres1, alpha, vl);

  408.             VSEV_FLOAT(C1, va1, vl);

  409. 

  410.             va2 = VFMULVF_FLOAT(vres2, alpha, vl);

  411.             VSEV_FLOAT(C2, va2, vl);

  412. 

  413.             va3 = VFMULVF_FLOAT(vres3, alpha, vl);

  414.             VSEV_FLOAT(C3, va3, vl);

  415. 

  416. #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  417.             temp = bk - off;

  418. #ifdef LEFT

  419.             temp -= vl; // number of values in A

  420. #else

  421.             temp -= 4; // number of values in B

  422. #endif

  423.             ptrba += temp*vl;

  424.             ptrbb += temp*4;

  425. #endif

  426. 

  427. #ifdef LEFT

  428.             off += vl; // number of values in A

  429. #endif

  430. 

  431.             C0 += vl;

  432.             C1 += vl;

  433.             C2 += vl;

  434.             C3 += vl;

  435.         }

  436. 

  437. #if defined(TRMMKERNEL) && !defined(LEFT)

  438.         off += 4;

  439. #endif

  440. 

  441.         bb += (bk<<2);

  442.         C += (ldc<<2);

  443.     }

  444. 

  445.     if (bn & 2)

  446.     {

  447.         C0 = C;

  448.         C1 = C0+ldc;

  449. 

  450. #if defined(TRMMKERNEL) && defined(LEFT)

  451.         off = offset;

  452. #endif

  453. 

  454.         ptrba = ba;

  455. 

  456.         for (i = bm; i > 0; i -= vl) 

  457.         {

  458.             vl = VSETVL(i);

  459. 

  460. #if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  461.             ptrbb = bb;

  462. #else

  463.             ptrba += off*vl;

  464.             ptrbb = bb + off*2;

  465. #endif

  466. 

  467.             vres0 = VFMVVF_FLOAT(0.0, vl);

  468.             vres1 = VFMVVF_FLOAT(0.0, vl);

  469. 

  470. 

  471. #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))

  472.             temp = bk-off;

  473. #elif defined(LEFT)

  474.             temp = off+vl;  // number of values in A

  475. #else

  476.             temp = off+2;   // number of values in B

  477. #endif

  478. 

  479.             for (k = temp/8; k > 0; k--) {

  480.                 va0 = VLEV_FLOAT(ptrba, vl);

  481.                 ptrba += vl;

  482.                 va1 = VLEV_FLOAT(ptrba, vl);

  483.                 ptrba += vl;

  484. 

  485.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  486.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va0, vl);

  487.                 ptrbb += 2;

  488.                 va2 = VLEV_FLOAT(ptrba, vl);

  489.                 ptrba += vl;

  490. 

  491.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va1, vl);

  492.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va1, vl);

  493.                 ptrbb += 2;

  494.                 va3 = VLEV_FLOAT(ptrba, vl);

  495.                 ptrba += vl;

  496. 

  497.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va2, vl);

  498.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va2, vl);

  499.                 ptrbb += 2;

  500.                 va4 = VLEV_FLOAT(ptrba, vl);

  501.                 ptrba += vl;

  502. 

  503.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va3, vl);

  504.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va3, vl);

  505.                 ptrbb += 2;

  506.                 va5 = VLEV_FLOAT(ptrba, vl);

  507.                 ptrba += vl;

  508. 

  509.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va4, vl);

  510.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va4, vl);

  511.                 ptrbb += 2;

  512.                 va6 = VLEV_FLOAT(ptrba, vl);

  513.                 ptrba += vl;

  514. 

  515.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va5, vl);

  516.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va5, vl);

  517.                 ptrbb += 2;

  518.                 va7 = VLEV_FLOAT(ptrba, vl);

  519.                 ptrba += vl;

  520. 

  521.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va6, vl);

  522.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va6, vl);

  523.                 ptrbb += 2;

  524. 

  525.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va7, vl);

  526.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va7, vl);

  527.                 ptrbb += 2;

  528.             }

  529. 

  530.             // K remainder

  531.             for (k = temp&7; k > 0; k--) {

  532.                 va0 = VLEV_FLOAT(ptrba, vl);

  533. 

  534.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  535.                 vres1 = VFMACCVF_FLOAT(vres1, *(ptrbb + 1), va0, vl);

  536. 

  537.                 ptrbb += 2;

  538.                 ptrba += vl;

  539.             }

  540.             va0 = VFMULVF_FLOAT(vres0, alpha, vl);

  541.             VSEV_FLOAT(C0, va0, vl);

  542. 

  543.             va1 = VFMULVF_FLOAT(vres1, alpha, vl);

  544.             VSEV_FLOAT(C1, va1, vl);

  545. 

  546. #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  547.             temp = bk - off;

  548. #ifdef LEFT

  549.             temp -= vl; // number of values in A

  550. #else

  551.             temp -= 2; // number of values in B

  552. #endif

  553.             ptrba += temp*vl;

  554.             ptrbb += temp*2;

  555. #endif

  556. 

  557. #ifdef LEFT

  558.             off += vl; // number of values in A

  559. #endif

  560. 

  561.             C0 += vl;

  562.             C1 += vl;

  563.         }

  564. 

  565. #if defined(TRMMKERNEL) && !defined(LEFT)

  566.         off += 2;

  567. #endif

  568. 

  569.         bb += (bk<<1);

  570.         C += (ldc<<1);

  571.     }

  572. 

  573.     if (bn & 1)

  574.     {

  575.         C0 = C;

  576. 

  577. #if defined(TRMMKERNEL) &&  defined(LEFT)

  578.         off = offset;

  579. #endif

  580. 

  581.         ptrba = ba;

  582. 

  583.         for (i = bm; i > 0; i -= vl)

  584.         {

  585.             vl = VSETVL(i);

  586. #if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  587.             ptrbb = bb;

  588. #else

  589.             ptrba += off*vl;

  590.             ptrbb = bb + off*1;

  591. #endif

  592. 

  593.             vres0 = VFMVVF_FLOAT(0.0, vl);

  594. 

  595. #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))

  596.             temp = bk-off;

  597. #elif defined(LEFT)

  598.             temp = off+vl;  // number of values in A

  599. #else

  600.             temp = off+1;   // number of values in B

  601. #endif

  602. 

  603.             for (k = temp/8; k > 0; k--) {

  604.                 va0 = VLEV_FLOAT(ptrba, vl);

  605.                 ptrba += vl;

  606.                 va1 = VLEV_FLOAT(ptrba, vl);

  607.                 ptrba += vl;

  608. 

  609.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  610.                 ptrbb += 1;

  611.                 va2 = VLEV_FLOAT(ptrba, vl);

  612.                 ptrba += vl;

  613. 

  614.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va1, vl);

  615.                 ptrbb += 1;

  616.                 va3 = VLEV_FLOAT(ptrba, vl);

  617.                 ptrba += vl;

  618. 

  619.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va2, vl);

  620.                 ptrbb += 1;

  621.                 va4 = VLEV_FLOAT(ptrba, vl);

  622.                 ptrba += vl;

  623. 

  624.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va3, vl);

  625.                 ptrbb += 1;

  626.                 va5 = VLEV_FLOAT(ptrba, vl);

  627.                 ptrba += vl;

  628. 

  629.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va4, vl);

  630.                 ptrbb += 1;

  631.                 va6 = VLEV_FLOAT(ptrba, vl);

  632.                 ptrba += vl;

  633. 

  634.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va5, vl);

  635.                 ptrbb += 1;

  636.                 va7 = VLEV_FLOAT(ptrba, vl);

  637.                 ptrba += vl;

  638. 

  639.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va6, vl);

  640.                 ptrbb += 1;

  641. 

  642.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va7, vl);

  643.                 ptrbb += 1;

  644.             }

  645. 

  646.             // K remainder

  647.             for (k = temp&7; k > 0; k--) {

  648.                 va0 = VLEV_FLOAT(ptrba, vl);

  649. 

  650.                 vres0 = VFMACCVF_FLOAT(vres0, *(ptrbb + 0), va0, vl);

  651. 

  652.                 ptrbb += 1;

  653.                 ptrba += vl;

  654.             }

  655.             va0 = VFMULVF_FLOAT(vres0, alpha, vl);

  656.             VSEV_FLOAT(C0, va0, vl);

  657. 

  658. #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))

  659.             temp = bk - off;

  660. #ifdef LEFT

  661.             temp -= vl; // number of values in A

  662. #else

  663.             temp -= 1; // number of values in B

  664. #endif

  665.             ptrba += temp*vl;

  666.             ptrbb += temp*1;

  667. #endif

  668. 

  669. #ifdef LEFT

  670.             off += vl; // number of values in A

  671. #endif

  672. 

  673.             C0 += vl;

  674.         }

  675. 

  676. #if defined(TRMMKERNEL) && !defined(LEFT)

  677.         off += 1;

  678. #endif

  679. 

  680.         bb += (bk);

  681.         C += (ldc);

  682.     }

  683.     return 0;

  684. }