numo-linalg-alt 0.2.0 → 0.4.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (55) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +10 -1
  3. data/README.md +3 -1
  4. data/ext/numo/linalg/blas/dot.c +59 -59
  5. data/ext/numo/linalg/blas/dot_sub.c +58 -58
  6. data/ext/numo/linalg/blas/gemm.c +157 -148
  7. data/ext/numo/linalg/blas/gemv.c +131 -127
  8. data/ext/numo/linalg/blas/nrm2.c +50 -50
  9. data/ext/numo/linalg/lapack/gees.c +276 -0
  10. data/ext/numo/linalg/lapack/gees.h +15 -0
  11. data/ext/numo/linalg/lapack/geev.c +127 -110
  12. data/ext/numo/linalg/lapack/gelsd.c +81 -70
  13. data/ext/numo/linalg/lapack/geqrf.c +52 -51
  14. data/ext/numo/linalg/lapack/gerqf.c +70 -0
  15. data/ext/numo/linalg/lapack/gerqf.h +15 -0
  16. data/ext/numo/linalg/lapack/gesdd.c +96 -86
  17. data/ext/numo/linalg/lapack/gesv.c +80 -78
  18. data/ext/numo/linalg/lapack/gesvd.c +140 -129
  19. data/ext/numo/linalg/lapack/getrf.c +51 -50
  20. data/ext/numo/linalg/lapack/getri.c +64 -63
  21. data/ext/numo/linalg/lapack/getrs.c +92 -88
  22. data/ext/numo/linalg/lapack/gges.c +214 -0
  23. data/ext/numo/linalg/lapack/gges.h +15 -0
  24. data/ext/numo/linalg/lapack/heev.c +54 -52
  25. data/ext/numo/linalg/lapack/heevd.c +54 -52
  26. data/ext/numo/linalg/lapack/heevr.c +109 -98
  27. data/ext/numo/linalg/lapack/hegv.c +77 -74
  28. data/ext/numo/linalg/lapack/hegvd.c +77 -74
  29. data/ext/numo/linalg/lapack/hegvx.c +132 -120
  30. data/ext/numo/linalg/lapack/hetrf.c +54 -50
  31. data/ext/numo/linalg/lapack/lange.c +45 -44
  32. data/ext/numo/linalg/lapack/orgqr.c +63 -62
  33. data/ext/numo/linalg/lapack/orgrq.c +78 -0
  34. data/ext/numo/linalg/lapack/orgrq.h +15 -0
  35. data/ext/numo/linalg/lapack/potrf.c +49 -48
  36. data/ext/numo/linalg/lapack/potri.c +49 -48
  37. data/ext/numo/linalg/lapack/potrs.c +74 -72
  38. data/ext/numo/linalg/lapack/syev.c +54 -52
  39. data/ext/numo/linalg/lapack/syevd.c +54 -52
  40. data/ext/numo/linalg/lapack/syevr.c +107 -98
  41. data/ext/numo/linalg/lapack/sygv.c +77 -73
  42. data/ext/numo/linalg/lapack/sygvd.c +77 -73
  43. data/ext/numo/linalg/lapack/sygvx.c +132 -120
  44. data/ext/numo/linalg/lapack/sytrf.c +54 -50
  45. data/ext/numo/linalg/lapack/trtrs.c +79 -75
  46. data/ext/numo/linalg/lapack/ungqr.c +63 -62
  47. data/ext/numo/linalg/lapack/ungrq.c +78 -0
  48. data/ext/numo/linalg/lapack/ungrq.h +15 -0
  49. data/ext/numo/linalg/linalg.c +21 -10
  50. data/ext/numo/linalg/linalg.h +5 -0
  51. data/ext/numo/linalg/util.c +8 -0
  52. data/ext/numo/linalg/util.h +1 -0
  53. data/lib/numo/linalg/version.rb +1 -1
  54. data/lib/numo/linalg.rb +322 -0
  55. metadata +14 -4
data/ext/numo/linalg/lapack/getrf.c CHANGED
@@ -4,56 +4,57 @@ struct _getrf_option {
4
4
  int matrix_layout;
5
5
  };
6
6
 
7
- #define DEF_LINALG_FUNC(tDType, tNAryClass, fLapackFunc) \
8
- static void _iter_##fLapackFunc(na_loop_t* const lp) { \
9
- tDType* a = (tDType*)NDL_PTR(lp, 0); \
10
- int* ipiv = (int*)NDL_PTR(lp, 1); \
11
- int* info = (int*)NDL_PTR(lp, 2); \
12
- struct _getrf_option* opt = (struct _getrf_option*)(lp->opt_ptr); \
13
- const lapack_int m = (lapack_int)NDL_SHAPE(lp, 0)[0]; \
14
- const lapack_int n = (lapack_int)NDL_SHAPE(lp, 0)[1]; \
15
- const lapack_int lda = n; \
16
- const lapack_int i = LAPACKE_##fLapackFunc(opt->matrix_layout, m, n, a, lda, ipiv); \
17
- *info = (int)i; \
18
- } \
19
- \
20
- static VALUE _linalg_lapack_##fLapackFunc(int argc, VALUE* argv, VALUE self) { \
21
- VALUE a_vnary = Qnil; \
22
- VALUE kw_args = Qnil; \
23
- rb_scan_args(argc, argv, "1:", &a_vnary, &kw_args); \
24
- ID kw_table[1] = { rb_intern("order") }; \
25
- VALUE kw_values[1] = { Qundef }; \
26
- rb_get_kwargs(kw_args, kw_table, 0, 1, kw_values); \
27
- const int matrix_layout = kw_values[0] != Qundef ? get_matrix_layout(kw_values[0]) : LAPACK_ROW_MAJOR; \
28
- \
29
- if (CLASS_OF(a_vnary) != tNAryClass) { \
30
- a_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, a_vnary); \
31
- } \
32
- if (!RTEST(nary_check_contiguous(a_vnary))) { \
33
- a_vnary = nary_dup(a_vnary); \
34
- } \
35
- \
36
- narray_t* a_nary = NULL; \
37
- GetNArray(a_vnary, a_nary); \
38
- const int n_dims = NA_NDIM(a_nary); \
39
- if (n_dims != 2) { \
40
- rb_raise(rb_eArgError, "input array a must be 2-dimensional"); \
41
- return Qnil; \
42
- } \
43
- \
44
- size_t m = NA_SHAPE(a_nary)[0]; \
45
- size_t n = NA_SHAPE(a_nary)[1]; \
46
- size_t shape[1] = { m < n ? m : n }; \
47
- ndfunc_arg_in_t ain[1] = { { OVERWRITE, 2 } }; \
48
- ndfunc_arg_out_t aout[2] = { { numo_cInt32, 1, shape }, { numo_cInt32, 0 } }; \
49
- ndfunc_t ndf = { _iter_##fLapackFunc, NO_LOOP | NDF_EXTRACT, 1, 2, ain, aout }; \
50
- struct _getrf_option opt = { matrix_layout }; \
51
- VALUE res = na_ndloop3(&ndf, &opt, 1, a_vnary); \
52
- \
53
- VALUE ret = rb_ary_concat(rb_ary_new3(1, a_vnary), res); \
54
- \
55
- RB_GC_GUARD(a_vnary); \
56
- return ret; \
7
+ #define DEF_LINALG_FUNC(tDType, tNAryClass, fLapackFunc) \
8
+ static void _iter_##fLapackFunc(na_loop_t* const lp) { \
9
+ tDType* a = (tDType*)NDL_PTR(lp, 0); \
10
+ int* ipiv = (int*)NDL_PTR(lp, 1); \
11
+ int* info = (int*)NDL_PTR(lp, 2); \
12
+ struct _getrf_option* opt = (struct _getrf_option*)(lp->opt_ptr); \
13
+ const lapack_int m = (lapack_int)NDL_SHAPE(lp, 0)[0]; \
14
+ const lapack_int n = (lapack_int)NDL_SHAPE(lp, 0)[1]; \
15
+ const lapack_int lda = n; \
16
+ const lapack_int i = LAPACKE_##fLapackFunc(opt->matrix_layout, m, n, a, lda, ipiv); \
17
+ *info = (int)i; \
18
+ } \
19
+ \
20
+ static VALUE _linalg_lapack_##fLapackFunc(int argc, VALUE* argv, VALUE self) { \
21
+ VALUE a_vnary = Qnil; \
22
+ VALUE kw_args = Qnil; \
23
+ rb_scan_args(argc, argv, "1:", &a_vnary, &kw_args); \
24
+ ID kw_table[1] = { rb_intern("order") }; \
25
+ VALUE kw_values[1] = { Qundef }; \
26
+ rb_get_kwargs(kw_args, kw_table, 0, 1, kw_values); \
27
+ const int matrix_layout = \
28
+ kw_values[0] != Qundef ? get_matrix_layout(kw_values[0]) : LAPACK_ROW_MAJOR; \
29
+ \
30
+ if (CLASS_OF(a_vnary) != tNAryClass) { \
31
+ a_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, a_vnary); \
32
+ } \
33
+ if (!RTEST(nary_check_contiguous(a_vnary))) { \
34
+ a_vnary = nary_dup(a_vnary); \
35
+ } \
36
+ \
37
+ narray_t* a_nary = NULL; \
38
+ GetNArray(a_vnary, a_nary); \
39
+ const int n_dims = NA_NDIM(a_nary); \
40
+ if (n_dims != 2) { \
41
+ rb_raise(rb_eArgError, "input array a must be 2-dimensional"); \
42
+ return Qnil; \
43
+ } \
44
+ \
45
+ size_t m = NA_SHAPE(a_nary)[0]; \
46
+ size_t n = NA_SHAPE(a_nary)[1]; \
47
+ size_t shape[1] = { m < n ? m : n }; \
48
+ ndfunc_arg_in_t ain[1] = { { OVERWRITE, 2 } }; \
49
+ ndfunc_arg_out_t aout[2] = { { numo_cInt32, 1, shape }, { numo_cInt32, 0 } }; \
50
+ ndfunc_t ndf = { _iter_##fLapackFunc, NO_LOOP | NDF_EXTRACT, 1, 2, ain, aout }; \
51
+ struct _getrf_option opt = { matrix_layout }; \
52
+ VALUE res = na_ndloop3(&ndf, &opt, 1, a_vnary); \
53
+ \
54
+ VALUE ret = rb_ary_concat(rb_ary_new3(1, a_vnary), res); \
55
+ \
56
+ RB_GC_GUARD(a_vnary); \
57
+ return ret; \
57
58
  }
58
59
 
59
60
  DEF_LINALG_FUNC(double, numo_cDFloat, dgetrf)
data/ext/numo/linalg/lapack/getri.c CHANGED
@@ -4,69 +4,70 @@ struct _getri_option {
4
4
  int matrix_layout;
5
5
  };
6
6
 
7
- #define DEF_LINALG_FUNC(tDType, tNAryClass, fLapackFunc) \
8
- static void _iter_##fLapackFunc(na_loop_t* const lp) { \
9
- tDType* a = (tDType*)NDL_PTR(lp, 0); \
10
- lapack_int* ipiv = (lapack_int*)NDL_PTR(lp, 1); \
11
- int* info = (int*)NDL_PTR(lp, 2); \
12
- struct _getri_option* opt = (struct _getri_option*)(lp->opt_ptr); \
13
- const lapack_int n = (lapack_int)NDL_SHAPE(lp, 0)[0]; \
14
- const lapack_int lda = n; \
15
- const lapack_int i = LAPACKE_##fLapackFunc(opt->matrix_layout, n, a, lda, ipiv); \
16
- *info = (int)i; \
17
- } \
18
- \
19
- static VALUE _linalg_lapack_##fLapackFunc(int argc, VALUE* argv, VALUE self) { \
20
- VALUE a_vnary = Qnil; \
21
- VALUE ipiv_vnary = Qnil; \
22
- VALUE kw_args = Qnil; \
23
- rb_scan_args(argc, argv, "2:", &a_vnary, &ipiv_vnary, &kw_args); \
24
- ID kw_table[1] = { rb_intern("order") }; \
25
- VALUE kw_values[1] = { Qundef }; \
26
- rb_get_kwargs(kw_args, kw_table, 0, 1, kw_values); \
27
- const int matrix_layout = kw_values[0] != Qundef ? get_matrix_layout(kw_values[0]) : LAPACK_ROW_MAJOR; \
28
- \
29
- if (CLASS_OF(a_vnary) != tNAryClass) { \
30
- a_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, a_vnary); \
31
- } \
32
- if (!RTEST(nary_check_contiguous(a_vnary))) { \
33
- a_vnary = nary_dup(a_vnary); \
34
- } \
35
- if (CLASS_OF(ipiv_vnary) != numo_cInt32) { \
36
- ipiv_vnary = rb_funcall(numo_cInt32, rb_intern("cast"), 1, ipiv_vnary); \
37
- } \
38
- if (!RTEST(nary_check_contiguous(ipiv_vnary))) { \
39
- ipiv_vnary = nary_dup(ipiv_vnary); \
40
- } \
41
- \
42
- narray_t* a_nary = NULL; \
43
- GetNArray(a_vnary, a_nary); \
44
- if (NA_NDIM(a_nary) != 2) { \
45
- rb_raise(rb_eArgError, "input array a must be 2-dimensional"); \
46
- return Qnil; \
47
- } \
48
- if (NA_SHAPE(a_nary)[0] != NA_SHAPE(a_nary)[1]) { \
49
- rb_raise(rb_eArgError, "input array a must be square"); \
50
- return Qnil; \
51
- } \
52
- narray_t* ipiv_nary = NULL; \
53
- GetNArray(ipiv_vnary, ipiv_nary); \
54
- if (NA_NDIM(ipiv_nary) != 1) { \
55
- rb_raise(rb_eArgError, "input array ipiv must be 1-dimensional"); \
56
- return Qnil; \
57
- } \
58
- \
59
- ndfunc_arg_in_t ain[2] = { { OVERWRITE, 2 }, { numo_cInt32, 1 } }; \
60
- ndfunc_arg_out_t aout[1] = { { numo_cInt32, 0 } }; \
61
- ndfunc_t ndf = { _iter_##fLapackFunc, NO_LOOP | NDF_EXTRACT, 2, 1, ain, aout }; \
62
- struct _getri_option opt = { matrix_layout }; \
63
- VALUE res = na_ndloop3(&ndf, &opt, 2, a_vnary, ipiv_vnary); \
64
- \
65
- VALUE ret = rb_ary_new3(2, a_vnary, res); \
66
- \
67
- RB_GC_GUARD(a_vnary); \
68
- RB_GC_GUARD(ipiv_vnary); \
69
- return ret; \
7
+ #define DEF_LINALG_FUNC(tDType, tNAryClass, fLapackFunc) \
8
+ static void _iter_##fLapackFunc(na_loop_t* const lp) { \
9
+ tDType* a = (tDType*)NDL_PTR(lp, 0); \
10
+ lapack_int* ipiv = (lapack_int*)NDL_PTR(lp, 1); \
11
+ int* info = (int*)NDL_PTR(lp, 2); \
12
+ struct _getri_option* opt = (struct _getri_option*)(lp->opt_ptr); \
13
+ const lapack_int n = (lapack_int)NDL_SHAPE(lp, 0)[0]; \
14
+ const lapack_int lda = n; \
15
+ const lapack_int i = LAPACKE_##fLapackFunc(opt->matrix_layout, n, a, lda, ipiv); \
16
+ *info = (int)i; \
17
+ } \
18
+ \
19
+ static VALUE _linalg_lapack_##fLapackFunc(int argc, VALUE* argv, VALUE self) { \
20
+ VALUE a_vnary = Qnil; \
21
+ VALUE ipiv_vnary = Qnil; \
22
+ VALUE kw_args = Qnil; \
23
+ rb_scan_args(argc, argv, "2:", &a_vnary, &ipiv_vnary, &kw_args); \
24
+ ID kw_table[1] = { rb_intern("order") }; \
25
+ VALUE kw_values[1] = { Qundef }; \
26
+ rb_get_kwargs(kw_args, kw_table, 0, 1, kw_values); \
27
+ const int matrix_layout = \
28
+ kw_values[0] != Qundef ? get_matrix_layout(kw_values[0]) : LAPACK_ROW_MAJOR; \
29
+ \
30
+ if (CLASS_OF(a_vnary) != tNAryClass) { \
31
+ a_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, a_vnary); \
32
+ } \
33
+ if (!RTEST(nary_check_contiguous(a_vnary))) { \
34
+ a_vnary = nary_dup(a_vnary); \
35
+ } \
36
+ if (CLASS_OF(ipiv_vnary) != numo_cInt32) { \
37
+ ipiv_vnary = rb_funcall(numo_cInt32, rb_intern("cast"), 1, ipiv_vnary); \
38
+ } \
39
+ if (!RTEST(nary_check_contiguous(ipiv_vnary))) { \
40
+ ipiv_vnary = nary_dup(ipiv_vnary); \
41
+ } \
42
+ \
43
+ narray_t* a_nary = NULL; \
44
+ GetNArray(a_vnary, a_nary); \
45
+ if (NA_NDIM(a_nary) != 2) { \
46
+ rb_raise(rb_eArgError, "input array a must be 2-dimensional"); \
47
+ return Qnil; \
48
+ } \
49
+ if (NA_SHAPE(a_nary)[0] != NA_SHAPE(a_nary)[1]) { \
50
+ rb_raise(rb_eArgError, "input array a must be square"); \
51
+ return Qnil; \
52
+ } \
53
+ narray_t* ipiv_nary = NULL; \
54
+ GetNArray(ipiv_vnary, ipiv_nary); \
55
+ if (NA_NDIM(ipiv_nary) != 1) { \
56
+ rb_raise(rb_eArgError, "input array ipiv must be 1-dimensional"); \
57
+ return Qnil; \
58
+ } \
59
+ \
60
+ ndfunc_arg_in_t ain[2] = { { OVERWRITE, 2 }, { numo_cInt32, 1 } }; \
61
+ ndfunc_arg_out_t aout[1] = { { numo_cInt32, 0 } }; \
62
+ ndfunc_t ndf = { _iter_##fLapackFunc, NO_LOOP | NDF_EXTRACT, 2, 1, ain, aout }; \
63
+ struct _getri_option opt = { matrix_layout }; \
64
+ VALUE res = na_ndloop3(&ndf, &opt, 2, a_vnary, ipiv_vnary); \
65
+ \
66
+ VALUE ret = rb_ary_new3(2, a_vnary, res); \
67
+ \
68
+ RB_GC_GUARD(a_vnary); \
69
+ RB_GC_GUARD(ipiv_vnary); \
70
+ return ret; \
70
71
  }
71
72
 
72
73
  DEF_LINALG_FUNC(double, numo_cDFloat, dgetri)
data/ext/numo/linalg/lapack/getrs.c CHANGED
@@ -5,94 +5,98 @@ struct _getrs_option {
5
5
  char trans;
6
6
  };
7
7
 
8
- #define DEF_LINALG_FUNC(tDType, tNAryClass, fLapackFunc) \
9
- static void _iter_##fLapackFunc(na_loop_t* const lp) { \
10
- tDType* a = (tDType*)NDL_PTR(lp, 0); \
11
- int* ipiv = (int*)NDL_PTR(lp, 1); \
12
- tDType* b = (tDType*)NDL_PTR(lp, 2); \
13
- int* info = (int*)NDL_PTR(lp, 3); \
14
- struct _getrs_option* opt = (struct _getrs_option*)(lp->opt_ptr); \
15
- const lapack_int n = (lapack_int)NDL_SHAPE(lp, 0)[0]; \
16
- const lapack_int nrhs = lp->args[2].ndim == 1 ? 1 : (lapack_int)NDL_SHAPE(lp, 2)[1]; \
17
- const lapack_int lda = n; \
18
- const lapack_int ldb = nrhs; \
19
- const lapack_int i = LAPACKE_##fLapackFunc(opt->matrix_layout, opt->trans, n, nrhs, a, lda, ipiv, b, ldb); \
20
- *info = (int)i; \
21
- } \
22
- \
23
- static VALUE _linalg_lapack_##fLapackFunc(int argc, VALUE* argv, VALUE self) { \
24
- VALUE a_vnary = Qnil; \
25
- VALUE ipiv_vnary = Qnil; \
26
- VALUE b_vnary = Qnil; \
27
- VALUE kw_args = Qnil; \
28
- rb_scan_args(argc, argv, "3:", &a_vnary, &ipiv_vnary, &b_vnary, &kw_args); \
29
- ID kw_table[2] = { rb_intern("order"), rb_intern("trans") }; \
30
- VALUE kw_values[2] = { Qundef, Qundef }; \
31
- rb_get_kwargs(kw_args, kw_table, 0, 2, kw_values); \
32
- const int matrix_layout = kw_values[0] != Qundef ? get_matrix_layout(kw_values[0]) : LAPACK_ROW_MAJOR; \
33
- const char trans = kw_values[1] != Qundef ? NUM2CHR(kw_values[1]) : 'N'; \
34
- \
35
- if (CLASS_OF(a_vnary) != tNAryClass) { \
36
- a_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, a_vnary); \
37
- } \
38
- if (!RTEST(nary_check_contiguous(a_vnary))) { \
39
- a_vnary = nary_dup(a_vnary); \
40
- } \
41
- if (CLASS_OF(ipiv_vnary) != numo_cInt32) { \
42
- ipiv_vnary = rb_funcall(numo_cInt32, rb_intern("cast"), 1, ipiv_vnary); \
43
- } \
44
- if (!RTEST(nary_check_contiguous(ipiv_vnary))) { \
45
- ipiv_vnary = nary_dup(ipiv_vnary); \
46
- } \
47
- if (CLASS_OF(b_vnary) != tNAryClass) { \
48
- b_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, b_vnary); \
49
- } \
50
- if (!RTEST(nary_check_contiguous(b_vnary))) { \
51
- b_vnary = nary_dup(b_vnary); \
52
- } \
53
- \
54
- narray_t* a_nary = NULL; \
55
- GetNArray(a_vnary, a_nary); \
56
- const int n_dims = NA_NDIM(a_nary); \
57
- if (n_dims != 2) { \
58
- rb_raise(rb_eArgError, "input array a must be 2-dimensional"); \
59
- return Qnil; \
60
- } \
61
- if (NA_SHAPE(a_nary)[0] != NA_SHAPE(a_nary)[1]) { \
62
- rb_raise(rb_eArgError, "input array a must be square"); \
63
- return Qnil; \
64
- } \
65
- narray_t* ipiv_nary = NULL; \
66
- GetNArray(ipiv_vnary, ipiv_nary); \
67
- const int ipiv_n_dims = NA_NDIM(ipiv_nary); \
68
- if (ipiv_n_dims != 1) { \
69
- rb_raise(rb_eArgError, "input array ipiv must be 1-dimensional"); \
70
- return Qnil; \
71
- } \
72
- narray_t* b_nary = NULL; \
73
- GetNArray(b_vnary, b_nary); \
74
- const int b_n_dims = NA_NDIM(b_nary); \
75
- if (b_n_dims != 1 && b_n_dims != 2) { \
76
- rb_raise(rb_eArgError, "input array b must be 1 or 2-dimensional"); \
77
- return Qnil; \
78
- } \
79
- lapack_int n = (lapack_int)NA_SHAPE(a_nary)[0]; \
80
- lapack_int nb = (lapack_int)NA_SHAPE(b_nary)[0]; \
81
- if (n != nb) { \
82
- rb_raise(nary_eShapeError, "shape1[0](=%d) != shape2[0](=%d)", n, nb); \
83
- } \
84
- \
85
- ndfunc_arg_in_t ain[3] = { { tNAryClass, 2 }, { numo_cInt32, 1 }, { OVERWRITE, b_n_dims } }; \
86
- ndfunc_arg_out_t aout[1] = { { numo_cInt32, 0 } }; \
87
- ndfunc_t ndf = { _iter_##fLapackFunc, NO_LOOP | NDF_EXTRACT, 3, 1, ain, aout }; \
88
- struct _getrs_option opt = { matrix_layout, trans }; \
89
- VALUE info = na_ndloop3(&ndf, &opt, 3, a_vnary, ipiv_vnary, b_vnary); \
90
- VALUE ret = rb_ary_new3(2, b_vnary, info); \
91
- \
92
- RB_GC_GUARD(a_vnary); \
93
- RB_GC_GUARD(ipiv_vnary); \
94
- RB_GC_GUARD(b_vnary); \
95
- return ret; \
8
+ #define DEF_LINALG_FUNC(tDType, tNAryClass, fLapackFunc) \
9
+ static void _iter_##fLapackFunc(na_loop_t* const lp) { \
10
+ tDType* a = (tDType*)NDL_PTR(lp, 0); \
11
+ int* ipiv = (int*)NDL_PTR(lp, 1); \
12
+ tDType* b = (tDType*)NDL_PTR(lp, 2); \
13
+ int* info = (int*)NDL_PTR(lp, 3); \
14
+ struct _getrs_option* opt = (struct _getrs_option*)(lp->opt_ptr); \
15
+ const lapack_int n = (lapack_int)NDL_SHAPE(lp, 0)[0]; \
16
+ const lapack_int nrhs = lp->args[2].ndim == 1 ? 1 : (lapack_int)NDL_SHAPE(lp, 2)[1]; \
17
+ const lapack_int lda = n; \
18
+ const lapack_int ldb = nrhs; \
19
+ const lapack_int i = \
20
+ LAPACKE_##fLapackFunc(opt->matrix_layout, opt->trans, n, nrhs, a, lda, ipiv, b, ldb); \
21
+ *info = (int)i; \
22
+ } \
23
+ \
24
+ static VALUE _linalg_lapack_##fLapackFunc(int argc, VALUE* argv, VALUE self) { \
25
+ VALUE a_vnary = Qnil; \
26
+ VALUE ipiv_vnary = Qnil; \
27
+ VALUE b_vnary = Qnil; \
28
+ VALUE kw_args = Qnil; \
29
+ rb_scan_args(argc, argv, "3:", &a_vnary, &ipiv_vnary, &b_vnary, &kw_args); \
30
+ ID kw_table[2] = { rb_intern("order"), rb_intern("trans") }; \
31
+ VALUE kw_values[2] = { Qundef, Qundef }; \
32
+ rb_get_kwargs(kw_args, kw_table, 0, 2, kw_values); \
33
+ const int matrix_layout = \
34
+ kw_values[0] != Qundef ? get_matrix_layout(kw_values[0]) : LAPACK_ROW_MAJOR; \
35
+ const char trans = kw_values[1] != Qundef ? NUM2CHR(kw_values[1]) : 'N'; \
36
+ \
37
+ if (CLASS_OF(a_vnary) != tNAryClass) { \
38
+ a_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, a_vnary); \
39
+ } \
40
+ if (!RTEST(nary_check_contiguous(a_vnary))) { \
41
+ a_vnary = nary_dup(a_vnary); \
42
+ } \
43
+ if (CLASS_OF(ipiv_vnary) != numo_cInt32) { \
44
+ ipiv_vnary = rb_funcall(numo_cInt32, rb_intern("cast"), 1, ipiv_vnary); \
45
+ } \
46
+ if (!RTEST(nary_check_contiguous(ipiv_vnary))) { \
47
+ ipiv_vnary = nary_dup(ipiv_vnary); \
48
+ } \
49
+ if (CLASS_OF(b_vnary) != tNAryClass) { \
50
+ b_vnary = rb_funcall(tNAryClass, rb_intern("cast"), 1, b_vnary); \
51
+ } \
52
+ if (!RTEST(nary_check_contiguous(b_vnary))) { \
53
+ b_vnary = nary_dup(b_vnary); \
54
+ } \
55
+ \
56
+ narray_t* a_nary = NULL; \
57
+ GetNArray(a_vnary, a_nary); \
58
+ const int n_dims = NA_NDIM(a_nary); \
59
+ if (n_dims != 2) { \
60
+ rb_raise(rb_eArgError, "input array a must be 2-dimensional"); \
61
+ return Qnil; \
62
+ } \
63
+ if (NA_SHAPE(a_nary)[0] != NA_SHAPE(a_nary)[1]) { \
64
+ rb_raise(rb_eArgError, "input array a must be square"); \
65
+ return Qnil; \
66
+ } \
67
+ narray_t* ipiv_nary = NULL; \
68
+ GetNArray(ipiv_vnary, ipiv_nary); \
69
+ const int ipiv_n_dims = NA_NDIM(ipiv_nary); \
70
+ if (ipiv_n_dims != 1) { \
71
+ rb_raise(rb_eArgError, "input array ipiv must be 1-dimensional"); \
72
+ return Qnil; \
73
+ } \
74
+ narray_t* b_nary = NULL; \
75
+ GetNArray(b_vnary, b_nary); \
76
+ const int b_n_dims = NA_NDIM(b_nary); \
77
+ if (b_n_dims != 1 && b_n_dims != 2) { \
78
+ rb_raise(rb_eArgError, "input array b must be 1 or 2-dimensional"); \
79
+ return Qnil; \
80
+ } \
81
+ lapack_int n = (lapack_int)NA_SHAPE(a_nary)[0]; \
82
+ lapack_int nb = (lapack_int)NA_SHAPE(b_nary)[0]; \
83
+ if (n != nb) { \
84
+ rb_raise(nary_eShapeError, "shape1[0](=%d) != shape2[0](=%d)", n, nb); \
85
+ } \
86
+ \
87
+ ndfunc_arg_in_t ain[3] = { { tNAryClass, 2 }, \
88
+ { numo_cInt32, 1 }, \
89
+ { OVERWRITE, b_n_dims } }; \
90
+ ndfunc_arg_out_t aout[1] = { { numo_cInt32, 0 } }; \
91
+ ndfunc_t ndf = { _iter_##fLapackFunc, NO_LOOP | NDF_EXTRACT, 3, 1, ain, aout }; \
92
+ struct _getrs_option opt = { matrix_layout, trans }; \
93
+ VALUE info = na_ndloop3(&ndf, &opt, 3, a_vnary, ipiv_vnary, b_vnary); \
94
+ VALUE ret = rb_ary_new3(2, b_vnary, info); \
95
+ \
96
+ RB_GC_GUARD(a_vnary); \
97
+ RB_GC_GUARD(ipiv_vnary); \
98
+ RB_GC_GUARD(b_vnary); \
99
+ return ret; \
96
100
  }
97
101
 
98
102
  DEF_LINALG_FUNC(double, numo_cDFloat, dgetrs)