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: d96daa220d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd96daa220d'
${ noResults }
OpenBLAS/kernel/riscv64/zgemv_n_vector.c

364 lines
18 kB
Raw Blame History 

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

  2. Copyright (c) 2020, 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. #if !defined(DOUBLE)

  30. #define VSETVL(n) RISCV_RVV(vsetvl_e32m2)(n)

  31. #define FLOAT_V_T vfloat32m2_t

  32. #define VLEV_FLOAT RISCV_RVV(vle32_v_f32m2)

  33. #define VLSEV_FLOAT RISCV_RVV(vlse32_v_f32m2)

  34. #define VSEV_FLOAT RISCV_RVV(vse32_v_f32m2)

  35. #define VSSEV_FLOAT RISCV_RVV(vsse32_v_f32m2)

  36. #define VFMACCVF_FLOAT RISCV_RVV(vfmacc_vf_f32m2)

  37. #define VFNMSACVF_FLOAT RISCV_RVV(vfnmsac_vf_f32m2)

  38. #define VFMUL_VF_FLOAT RISCV_RVV(vfmul_vf_f32m2)

  39. #define VSEV_FLOAT RISCV_RVV(vse32_v_f32m2)

  40. #else

  41. #define VSETVL(n) RISCV_RVV(vsetvl_e64m2)(n)

  42. #define FLOAT_V_T vfloat64m2_t

  43. #define VLEV_FLOAT RISCV_RVV(vle64_v_f64m2)

  44. #define VLSEV_FLOAT RISCV_RVV(vlse64_v_f64m2)

  45. #define VSEV_FLOAT RISCV_RVV(vse64_v_f64m2)

  46. #define VSSEV_FLOAT RISCV_RVV(vsse64_v_f64m2)

  47. #define VFMACCVF_FLOAT RISCV_RVV(vfmacc_vf_f64m2)

  48. #define VFNMSACVF_FLOAT RISCV_RVV(vfnmsac_vf_f64m2)

  49. #define VFMUL_VF_FLOAT RISCV_RVV(vfmul_vf_f64m2)

  50. #define VSEV_FLOAT RISCV_RVV(vse64_v_f64m2)

  51. #endif

  52. 

  53. int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)

  54. {

  55.         BLASLONG i = 0, j = 0, k = 0;

  56.         BLASLONG ix = 0, iy = 0;

  57.         FLOAT *a_ptr = a;

  58.         FLOAT temp_r = 0.0, temp_i = 0.0, temp_r1, temp_i1, temp_r2, temp_i2, temp_r3, temp_i3, temp_rr[4], temp_ii[4];

  59.         FLOAT_V_T va0, va1, vy0, vy1, vy0_new, vy1_new, va2, va3, va4, va5, va6, va7, temp_iv, temp_rv, x_v0, x_v1, temp_v1, temp_v2, temp_v3, temp_v4;

  60.         unsigned int gvl = 0;

  61.         BLASLONG stride_a = sizeof(FLOAT) * 2;

  62.         BLASLONG stride_y = inc_y * sizeof(FLOAT) * 2;

  63.         gvl = VSETVL(m);

  64.         BLASLONG inc_yv = inc_y * gvl * 2;

  65.         BLASLONG inc_x2 = inc_x * 2;

  66.         BLASLONG lda2 = lda * 2;

  67.         vy0_new = VLSEV_FLOAT(&y[iy], stride_y, gvl);

  68.         vy1_new = VLSEV_FLOAT(&y[iy + 1], stride_y, gvl);

  69.         for (k = 0, j = 0; k < m / gvl; k ++)

  70.         {

  71.                 a_ptr = a;

  72.                 ix = 0;

  73.                 vy0 = vy0_new;

  74.                 vy1 = vy1_new;

  75. 

  76.                 if (k < m / gvl - 1)

  77.                 {

  78.                         vy0_new = VLSEV_FLOAT(&y[iy + inc_yv], stride_y, gvl);

  79.                         vy1_new = VLSEV_FLOAT(&y[iy + inc_yv + 1], stride_y, gvl);

  80.                 }

  81.                 for (i = 0; i < n % 4; i++)

  82.                 {

  83. #if !defined(XCONJ)

  84.                         temp_r = alpha_r * x[ix] - alpha_i * x[ix + 1];

  85.                         temp_i = alpha_r * x[ix + 1] + alpha_i * x[ix];

  86. #else

  87.                         temp_r = alpha_r * x[ix] + alpha_i * x[ix + 1];

  88.                         temp_i = alpha_r * x[ix + 1] - alpha_i * x[ix];

  89. #endif

  90. 

  91.                         va0 = VLSEV_FLOAT(&a_ptr[j], stride_a, gvl);

  92.                         va1 = VLSEV_FLOAT(&a_ptr[j + 1], stride_a, gvl);

  93. #if !defined(CONJ)

  94. #if !defined(XCONJ)

  95.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  96.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_i, va1, gvl);

  97.                         vy1 = VFMACCVF_FLOAT(vy1, temp_r, va1, gvl);

  98.                         vy1 = VFMACCVF_FLOAT(vy1, temp_i, va0, gvl);

  99. #else

  100. 

  101.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  102.                         vy0 = VFMACCVF_FLOAT(vy0, temp_i, va1, gvl);

  103.                         vy1 = VFMACCVF_FLOAT(vy1, temp_r, va1, gvl);

  104.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_i, va0, gvl);

  105. #endif

  106. 

  107. #else

  108. 

  109. #if !defined(XCONJ)

  110.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  111.                         vy0 = VFMACCVF_FLOAT(vy0, temp_i, va1, gvl);

  112.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_r, va1, gvl);

  113.                         vy1 = VFMACCVF_FLOAT(vy1, temp_i, va0, gvl);

  114. #else

  115.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  116.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_i, va1, gvl);

  117.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_r, va1, gvl);

  118.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_i, va0, gvl);

  119. #endif

  120. 

  121. #endif

  122.                         a_ptr += lda2;

  123.                         ix += inc_x2;

  124.                         

  125.                 }

  126. 

  127.                 for (i = n % 4 ; i < n; i += 4)

  128.                 {

  129. #if !defined(XCONJ)

  130.                         // temp_rr[0] = alpha_r * x[ix] - alpha_i * x[ix + 1];

  131.                         // temp_rr[1] = alpha_r * x[ix + inc_x2] - alpha_i * x[ix + inc_x2 + 1];

  132.                         x_v0 = VLSEV_FLOAT(&x[ix], inc_x2 * sizeof(FLOAT), 2);

  133.                         x_v1 = VLSEV_FLOAT(&x[ix + 1], inc_x2 * sizeof(FLOAT), 2);

  134.                         temp_rv = VFMUL_VF_FLOAT(x_v0, alpha_r, 2);

  135.                         temp_rv = VFNMSACVF_FLOAT(temp_rv, alpha_i, x_v1, 2);         

  136. 

  137.                         // temp_ii[0] = alpha_r * x[ix + 1] + alpha_i * x[ix]; 

  138.                         // temp_ii[1] = alpha_r * x[ix + inc_x2 + 1] + alpha_i * x[ix + inc_x2];

  139.                         temp_iv = VFMUL_VF_FLOAT(x_v0, alpha_i, 2);

  140.                         temp_iv = VFMACCVF_FLOAT(temp_iv, alpha_r, x_v1, 2);

  141.                         VSEV_FLOAT(&temp_rr[0], temp_rv, 2);

  142.                         VSEV_FLOAT(&temp_ii[0], temp_iv, 2);

  143.                                   

  144.                         // temp_rr[2] = alpha_r * x[ix + inc_x2 * 2] - alpha_i * x[ix + inc_x2 * 2 + 1];

  145.                         // temp_rr[3] = alpha_r * x[ix + inc_x2 * 3] - alpha_i * x[ix + inc_x2 * 3 + 1];

  146.                         x_v0 = VLSEV_FLOAT(&x[ix + inc_x2 * 2], inc_x2 * sizeof(FLOAT), 2);

  147.                         x_v1 = VLSEV_FLOAT(&x[ix + inc_x2 * 2 + 1], inc_x2 * sizeof(FLOAT), 2);

  148.                         temp_rv = VFMUL_VF_FLOAT(x_v0, alpha_r, 2);

  149.                         temp_rv = VFNMSACVF_FLOAT(temp_rv, alpha_i, x_v1, 2);

  150. 

  151.                         // temp_ii[2] = alpha_r * x[ix + inc_x2 * 2 + 1] + alpha_i * x[ix + inc_x2 * 2];

  152.                         // temp_ii[3] = alpha_r * x[ix + inc_x2 * 3 + 1] + alpha_i * x[ix + inc_x2 * 3];

  153.                         temp_iv = VFMUL_VF_FLOAT(x_v0, alpha_i, 2);

  154.                         temp_iv = VFMACCVF_FLOAT(temp_iv, alpha_r, x_v1, 2);

  155.                         VSEV_FLOAT(&temp_rr[2], temp_rv, 2);

  156.                         VSEV_FLOAT(&temp_ii[2], temp_iv, 2);

  157. 

  158. #else

  159.                         //  temp_rr[0] = alpha_r * x[ix] + alpha_i * x[ix + 1];

  160.                         //  temp_rr[1] = alpha_r * x[ix + inc_x2] + alpha_i * x[ix + inc_x2 + 1];

  161.                         x_v0 = VLSEV_FLOAT(&x[ix], inc_x2 * sizeof(FLOAT), 2);

  162.                         x_v1 = VLSEV_FLOAT(&x[ix + 1], inc_x2 * sizeof(FLOAT), 2);

  163.                         temp_rv = VFMUL_VF_FLOAT(x_v0, alpha_r, 2);

  164.                         temp_rv = VFMACCVF_FLOAT(temp_rv, alpha_i, x_v1, 2);

  165. 

  166. 

  167.                         // temp_ii[0] = alpha_r * x[ix + 1] - alpha_i * x[ix]; 

  168.                         // temp_ii[1] = alpha_r * x[ix + inc_x2 + 1] - alpha_i * x[ix + inc_x2];

  169.                         temp_iv = VFMUL_VF_FLOAT(x_v1, alpha_r, 2);

  170.                         temp_iv = VFNMSACVF_FLOAT(temp_iv, alpha_i, x_v0, 2);

  171.                         VSEV_FLOAT(&temp_rr[0], temp_rv, 2);

  172.                         VSEV_FLOAT(&temp_ii[0], temp_iv, 2);

  173. 

  174.                         

  175.                         // temp_rr[2] = alpha_r * x[ix + inc_x2 * 2] + alpha_i * x[ix + inc_x2 * 2 + 1];

  176.                         // temp_rr[3] = alpha_r * x[ix + inc_x2 * 3] + alpha_i * x[ix + inc_x2 * 3 + 1];

  177.                         x_v0 = VLSEV_FLOAT(&x[ix + inc_x2 * 2], inc_x2 * sizeof(FLOAT), 2);

  178.                         x_v1 = VLSEV_FLOAT(&x[ix + inc_x2 * 2 + 1], inc_x2 * sizeof(FLOAT), 2);

  179.                         temp_rv = VFMUL_VF_FLOAT(x_v0, alpha_r, 2);

  180.                         temp_rv = VFMACCVF_FLOAT(temp_rv, alpha_i, x_v1, 2);

  181. 

  182. 

  183.                         temp_ii[2] = alpha_r * x[ix + inc_x2 * 2 + 1] - alpha_i * x[ix + inc_x2 * 2];

  184.                         temp_ii[3] = alpha_r * x[ix + inc_x2 * 3 + 1] - alpha_i * x[ix + inc_x2 * 3];

  185.                         temp_iv = VFMUL_VF_FLOAT(x_v1, alpha_r, 2);

  186.                         temp_iv = VFNMSACVF_FLOAT(temp_iv, alpha_i, x_v0, 2);

  187.                         VSEV_FLOAT(&temp_rr[2], temp_rv, 2);

  188.                         VSEV_FLOAT(&temp_ii[2], temp_iv, 2);

  189. 

  190. 

  191. 

  192. #endif

  193. 

  194.                         va0 = VLSEV_FLOAT(&a_ptr[j], stride_a, gvl);

  195.                         va1 = VLSEV_FLOAT(&a_ptr[j + 1], stride_a, gvl);

  196.                         va2 = VLSEV_FLOAT(&a_ptr[j + lda2], stride_a, gvl);

  197.                         va3 = VLSEV_FLOAT(&a_ptr[j + lda2 + 1], stride_a, gvl);

  198.                         va4 = VLSEV_FLOAT(&a_ptr[j + lda2 * 2], stride_a, gvl);

  199.                         va5 = VLSEV_FLOAT(&a_ptr[j + lda2 * 2 + 1], stride_a, gvl);

  200.                         va6 = VLSEV_FLOAT(&a_ptr[j + lda2 * 3], stride_a, gvl);

  201.                         va7 = VLSEV_FLOAT(&a_ptr[j + lda2 * 3 + 1], stride_a, gvl);

  202. 

  203. #if !defined(CONJ)

  204. #if !defined(XCONJ)

  205.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[0], va0, gvl);

  206.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[0], va1, gvl);

  207.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[0], va1, gvl);

  208.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[0], va0, gvl);

  209. 

  210.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[1], va2, gvl);

  211.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[1], va3, gvl);

  212.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[1], va3, gvl);

  213.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[1], va2, gvl);

  214. 

  215.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[2], va4, gvl);

  216.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[2], va5, gvl);

  217.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[2], va5, gvl);

  218.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[2], va4, gvl);

  219. 

  220.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[3], va6, gvl);

  221.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[3], va7, gvl);

  222.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[3], va7, gvl);

  223.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[3], va6, gvl);

  224. 

  225. #else

  226. 

  227.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[0], va0, gvl);

  228.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[0], va1, gvl);

  229.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[0], va1, gvl);

  230.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[0], va0, gvl);

  231. 

  232.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[1], va2, gvl);

  233.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[1], va3, gvl);

  234.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[1], va3, gvl);

  235.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[1], va2, gvl);

  236. 

  237.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[2], va4, gvl);

  238.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[2], va5, gvl);

  239.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[2], va5, gvl);

  240.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[2], va4, gvl);

  241. 

  242.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[3], va6, gvl);

  243.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[3], va7, gvl);

  244.                         vy1 = VFMACCVF_FLOAT(vy1, temp_rr[3], va7, gvl);

  245.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[3], va6, gvl);

  246. 

  247. #endif

  248. 

  249. #else

  250. 

  251. #if !defined(XCONJ)

  252.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[0], va0, gvl);

  253.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[0], va1, gvl);

  254.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[0], va1, gvl);

  255.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[0], va0, gvl);

  256. 

  257.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[1], va2, gvl);

  258.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[1], va3, gvl);

  259.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[1], va3, gvl);

  260.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[1], va2, gvl);

  261. 

  262.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[2], va4, gvl);

  263.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[2], va5, gvl);

  264.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[2], va5, gvl);

  265.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[2], va4, gvl);

  266. 

  267.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[3], va6, gvl);

  268.                         vy0 = VFMACCVF_FLOAT(vy0, temp_ii[3], va7, gvl);

  269.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[3], va7, gvl);

  270.                         vy1 = VFMACCVF_FLOAT(vy1, temp_ii[3], va6, gvl);

  271. 

  272. #else

  273.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[0], va0, gvl);

  274.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[0], va1, gvl);

  275.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[0], va1, gvl);

  276.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[0], va0, gvl);

  277. 

  278.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[1], va2, gvl);

  279.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[1], va3, gvl);

  280.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[1], va3, gvl);

  281.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[1], va2, gvl);

  282. 

  283.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[2], va4, gvl);

  284.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[2], va5, gvl);

  285.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[2], va5, gvl);

  286.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[2], va4, gvl);

  287. 

  288.                         vy0 = VFMACCVF_FLOAT(vy0, temp_rr[3], va6, gvl);

  289.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_ii[3], va7, gvl);

  290.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_rr[3], va7, gvl);

  291.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_ii[3], va6, gvl);

  292. 

  293. #endif

  294. 

  295. #endif

  296.                         a_ptr += lda2 * 4;

  297.                         ix += inc_x2 * 4;

  298.                 }

  299. 

  300.                 VSSEV_FLOAT(&y[iy], stride_y, vy0, gvl);

  301.                 VSSEV_FLOAT(&y[iy + 1], stride_y, vy1, gvl);

  302.                 j += gvl * 2;

  303.                 iy += inc_yv  ;

  304.         }

  305.         // tail

  306.         if (j / 2 < m)

  307.         {

  308.                 gvl = VSETVL(m - j / 2);

  309.                 a_ptr = a;

  310.                 ix = 0;

  311.                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);

  312.                 vy1 = VLSEV_FLOAT(&y[iy + 1], stride_y, gvl);

  313.                 for (i = 0; i < n; i++)

  314.                 {

  315. #if !defined(XCONJ)

  316.                         temp_r = alpha_r * x[ix] - alpha_i * x[ix + 1];

  317.                         temp_i = alpha_r * x[ix + 1] + alpha_i * x[ix];

  318. #else

  319.                         temp_r = alpha_r * x[ix] + alpha_i * x[ix + 1];

  320.                         temp_i = alpha_r * x[ix + 1] - alpha_i * x[ix];

  321. #endif

  322. 

  323.                         va0 = VLSEV_FLOAT(&a_ptr[j], stride_a, gvl);

  324.                         va1 = VLSEV_FLOAT(&a_ptr[j + 1], stride_a, gvl);

  325. #if !defined(CONJ)

  326. 

  327. #if !defined(XCONJ)

  328.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  329.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_i, va1, gvl);

  330.                         vy1 = VFMACCVF_FLOAT(vy1, temp_r, va1, gvl);

  331.                         vy1 = VFMACCVF_FLOAT(vy1, temp_i, va0, gvl);

  332. #else

  333. 

  334.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  335.                         vy0 = VFMACCVF_FLOAT(vy0, temp_i, va1, gvl);

  336.                         vy1 = VFMACCVF_FLOAT(vy1, temp_r, va1, gvl);

  337.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_i, va0, gvl);

  338. #endif

  339. 

  340. #else

  341. 

  342. #if !defined(XCONJ)

  343.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  344.                         vy0 = VFMACCVF_FLOAT(vy0, temp_i, va1, gvl);

  345.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_r, va1, gvl);

  346.                         vy1 = VFMACCVF_FLOAT(vy1, temp_i, va0, gvl);

  347. #else

  348.                         vy0 = VFMACCVF_FLOAT(vy0, temp_r, va0, gvl);

  349.                         vy0 = VFNMSACVF_FLOAT(vy0, temp_i, va1, gvl);

  350.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_r, va1, gvl);

  351.                         vy1 = VFNMSACVF_FLOAT(vy1, temp_i, va0, gvl);

  352. #endif

  353. 

  354. #endif

  355.                         a_ptr += lda2;

  356.                         ix += inc_x2;

  357.                 }

  358.                 VSSEV_FLOAT(&y[iy], stride_y, vy0, gvl);

  359.                 VSSEV_FLOAT(&y[iy + 1], stride_y, vy1, gvl);

  360.         }

  361.         return (0);

  362. }