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: deecfb1a39
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'deecfb1a39'
${ noResults }
OpenBLAS/kernel/riscv64/znrm2_vector.c

184 lines
8.3 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. 

  30. #ifdef RISCV64_ZVL256B

  31. #       define LMUL m1

  32. #       if defined(DOUBLE)

  33. #               define ELEN 64

  34. #               define MLEN 64

  35. #       else

  36. #               define ELEN 32

  37. #               define MLEN 32

  38. #       endif

  39. #else

  40. #       define LMUL m4

  41. #       if defined(DOUBLE)

  42. #               define ELEN 64

  43. #               define MLEN 16

  44. #       else

  45. #               define ELEN 32

  46. #               define MLEN 8

  47. #       endif

  48. #endif

  49. 

  50. #define _

  51. #define JOIN2_X(x, y) x ## y

  52. #define JOIN2(x, y) JOIN2_X(x, y)

  53. #define JOIN(v, w, x, y, z) JOIN2( JOIN2( JOIN2( JOIN2( v, w ), x), y), z)

  54. 

  55. #define VSETVL          JOIN(RISCV_RVV(vsetvl),    _e,     ELEN,   LMUL,   _)

  56. #define FLOAT_V_T       JOIN(vfloat,            ELEN,   LMUL,   _t,     _)

  57. #define FLOAT_V_T_M1    JOIN(vfloat,            ELEN,   m1,     _t,     _)

  58. #define VLEV_FLOAT      JOIN(RISCV_RVV(vle),       ELEN,   _v_f,   ELEN,   LMUL)

  59. #define VLSEV_FLOAT     JOIN(RISCV_RVV(vlse),      ELEN,   _v_f,   ELEN,   LMUL)

  60. #define VFMVVF_FLOAT    JOIN(RISCV_RVV(vfmv),      _v_f_f, ELEN,   LMUL,   _)

  61. #define VFMVVF_FLOAT_M1 JOIN(RISCV_RVV(vfmv),      _v_f_f, ELEN,   m1,     _)

  62. #define MASK_T          JOIN(vbool,             MLEN,   _t,     _,      _)

  63. #define VFABS           JOIN(RISCV_RVV(vfabs),     _v_f,   ELEN,   LMUL,   _)

  64. #define VMFNE           JOIN(RISCV_RVV(vmfne_vf_f),ELEN,   LMUL,   _b,     MLEN)

  65. #define VMFGT           JOIN(RISCV_RVV(vmfgt_vv_f),ELEN,   LMUL,   _b,     MLEN)

  66. #define VMFEQ           JOIN(RISCV_RVV(vmfeq_vv_f),ELEN,   LMUL,   _b,     MLEN)

  67. #define VCPOP           JOIN(RISCV_RVV(vcpop),     _m_b,   MLEN,   _,      _)

  68. #ifdef RISCV_0p10_INTRINSICS

  69. #define VFREDMAX(va, vb, gvl) JOIN(RISCV_RVV(vfredmax_vs_f),ELEN,LMUL,   JOIN2(_f, ELEN), m1)(v_res, va, vb, gvl)

  70. #define VFREDUSUM(va, vb, gvl) JOIN(RISCV_RVV(vfredusum_vs_f),ELEN,LMUL,  JOIN2(_f, ELEN), m1)(v_res, va, vb, gvl)

  71. #define VFDIV_M         JOIN(RISCV_RVV(vfdiv),     _vv_f,  ELEN,   LMUL,   _m)

  72. #define VFMACC_M        JOIN(RISCV_RVV(vfmacc),    _vv_f,  ELEN,   LMUL,   _m)

  73. #else

  74. #define VFREDMAX        JOIN(RISCV_RVV(vfredmax_vs_f),ELEN,LMUL,   JOIN2(_f, ELEN), m1)

  75. #define VFREDUSUM       JOIN(RISCV_RVV(vfredusum_vs_f),ELEN,LMUL,  JOIN2(_f, ELEN), m1)

  76. #define VFDIV_M         JOIN(RISCV_RVV(vfdiv),     _vv_f,  ELEN,   LMUL,   _mu)

  77. #define VFMACC_M        JOIN(RISCV_RVV(vfmacc),    _vv_f,  ELEN,   LMUL,   _mu)

  78. #endif

  79. #define VFIRST          JOIN(RISCV_RVV(vfirst),    _m_b,   MLEN,   _,      _)

  80. #define VRGATHER        JOIN(RISCV_RVV(vrgather),  _vx_f,  ELEN,   LMUL,   _)

  81. #define VFDIV           JOIN(RISCV_RVV(vfdiv),     _vf_f,  ELEN,   LMUL,   _)

  82. #define VFMUL           JOIN(RISCV_RVV(vfmul),     _vv_f,  ELEN,   LMUL,   _)

  83. #define VFMACC          JOIN(RISCV_RVV(vfmacc),    _vv_f,  ELEN,   LMUL,   _)

  84. #define VMSOF           JOIN(RISCV_RVV(vmsof),     _m_b,   MLEN,   _,      _)

  85. #define VMANDN          JOIN(RISCV_RVV(vmandn),    _mm_b,  MLEN,   _,      _)

  86. #if defined(DOUBLE)

  87. #define ABS fabs

  88. #else

  89. #define ABS fabsf

  90. #endif

  91. 

  92. #define EXTRACT_FLOAT0_V(v) JOIN(RISCV_RVV(vfmv_f_s_f), ELEN, LMUL, _f, ELEN)(v)

  93. 

  94. 

  95. FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)

  96. {

  97.         BLASLONG i=0;

  98. 

  99.         if(n < 0)  return(0.0);

  100. 

  101.         FLOAT_V_T v_ssq, v_scale, v0, v1, v_zero;

  102.         unsigned int gvl = 0;

  103.         FLOAT_V_T_M1 v_res, v_z0;

  104. 

  105.         v_res = VFMVVF_FLOAT_M1(0, 1);

  106.         v_z0 = VFMVVF_FLOAT_M1(0, 1);

  107. 

  108.         gvl = VSETVL(n);

  109.         v_ssq = VFMVVF_FLOAT(0, gvl);

  110.         v_scale = VFMVVF_FLOAT(0, gvl);

  111.         v_zero = VFMVVF_FLOAT(0, gvl);

  112. 

  113.         unsigned int stride_x = inc_x * sizeof(FLOAT) * 2;

  114.         int idx = 0;

  115. 

  116.         for(i=0; i<n/gvl; i++){

  117.                 v0 = VLSEV_FLOAT( &x[idx], stride_x, gvl );

  118.                 v1 = VLSEV_FLOAT( &x[idx+1], stride_x, gvl );

  119.                 v0 = VFABS( v0, gvl );

  120.                 v1 = VFABS( v1, gvl );

  121. 

  122.                 MASK_T scale_mask0 = VMFGT( v0, v_scale, gvl );

  123.                 MASK_T scale_mask1 = VMFGT( v1, v_scale, gvl );

  124.                 if( VCPOP( scale_mask0, gvl ) + VCPOP( scale_mask1, gvl ) > 0 ){ // scale change?

  125.                         // find largest element in v0 and v1

  126.                         v_res = VFREDMAX( v0, v_z0, gvl );

  127.                         v_res = VFREDMAX( v1, v_res, gvl );

  128.                         FLOAT const largest_elt = EXTRACT_FLOAT( v_res );

  129. 

  130.                         v_scale = VFDIV( v_scale, largest_elt, gvl );   // scale/largest_elt

  131.                         v_scale = VFMUL( v_scale, v_scale, gvl );       // (scale/largest_elt)*(scale/largest_elt)

  132.                         v_ssq = VFMUL( v_scale, v_ssq, gvl );           // ssq*(scale/largest_elt)*(scale/largest_elt)

  133. 

  134.                         v_scale = VFMVVF_FLOAT( largest_elt, gvl );     // splated largest_elt becomes new scale

  135.                 }

  136. 

  137.                 MASK_T nonzero_mask0 = VMFNE( v0, 0, gvl );

  138.                 MASK_T nonzero_mask1 = VMFNE( v1, 0, gvl );

  139.                 v0 = VFDIV_M( nonzero_mask0, v_zero, v0, v_scale, gvl );

  140.                 v1 = VFDIV_M( nonzero_mask1, v_zero, v1, v_scale, gvl );

  141.                 v_ssq = VFMACC_M( nonzero_mask0, v_ssq, v0, v0, gvl );

  142.                 v_ssq = VFMACC_M( nonzero_mask1, v_ssq, v1, v1, gvl );

  143. 

  144.                 idx += inc_x * gvl * 2;

  145.         }

  146. 

  147.         v_res = VFREDUSUM(v_ssq, v_z0, gvl);

  148.         FLOAT ssq = EXTRACT_FLOAT(v_res);

  149.         FLOAT scale = EXTRACT_FLOAT0_V(v_scale);

  150. 

  151.         //finish any tail using scalar ops

  152.         i*=gvl;

  153.         if(i<n){

  154.                 i *= inc_x*2;

  155.                 n *= inc_x*2;

  156.                 FLOAT temp;

  157.                 do{

  158.                         if ( x[i] != 0.0 ){

  159.                                 temp = ABS( x[i] );

  160.                                 if ( scale < temp ){

  161.                                         ssq = 1 + ssq * ( scale / temp ) * ( scale / temp );

  162.                                         scale = temp ;

  163.                                 }else{

  164.                                         ssq += ( temp / scale ) * ( temp / scale );

  165.                                 }

  166.                         }

  167. 

  168.                         if ( x[i+1] != 0.0 ){

  169.                                 temp = ABS( x[i+1] );

  170.                                 if ( scale < temp ){

  171.                                         ssq = 1 + ssq * ( scale / temp ) * ( scale / temp );

  172.                                         scale = temp ;

  173.                                 }else{

  174.                                         ssq += ( temp / scale ) * ( temp / scale );

  175.                                 }

  176.                         }

  177. 

  178.                         i += inc_x*2;

  179.                 }while(i<n);

  180.         }

  181. 

  182.         return(scale * sqrt(ssq));

  183. }