1 /*
3 BLIS
4 An object-based framework for developing high-performance BLAS-like
5 libraries.
7 Copyright (C) 2014, The University of Texas at Austin
9 Redistribution and use in source and binary forms, with or without
10 modification, are permitted provided that the following conditions are
11 met:
12 - Redistributions of source code must retain the above copyright
13 notice, this list of conditions and the following disclaimer.
14 - Redistributions in binary form must reproduce the above copyright
15 notice, this list of conditions and the following disclaimer in the
16 documentation and/or other materials provided with the distribution.
17 - Neither the name of The University of Texas at Austin nor the names
18 of its contributors may be used to endorse or promote products
19 derived from this software without specific prior written permission.
21 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 */
35 #include "blis.h"
37 #define FUNCPTR_T trsv_fp
39 typedef void (*FUNCPTR_T)(
40 uplo_t uplo,
41 trans_t trans,
42 diag_t diag,
43 dim_t m,
44 void* alpha,
45 void* a, inc_t rs_a, inc_t cs_a,
46 void* x, inc_t incx
47 );
49 // If some mixed datatype functions will not be compiled, we initialize
50 // the corresponding elements of the function array to NULL.
51 #ifdef BLIS_ENABLE_MIXED_PRECISION_SUPPORT
52 static FUNCPTR_T GENARRAY2_ALL(ftypes,trsv_unb_var1);
53 #else
54 #ifdef BLIS_ENABLE_MIXED_DOMAIN_SUPPORT
55 static FUNCPTR_T GENARRAY2_EXT(ftypes,trsv_unb_var1);
56 #else
57 static FUNCPTR_T GENARRAY2_MIN(ftypes,trsv_unb_var1);
58 #endif
59 #endif
62 void bli_trsv_unb_var1( obj_t* alpha,
63 obj_t* a,
64 obj_t* x,
65 trsv_t* cntl )
66 {
67 num_t dt_a = bli_obj_datatype( *a );
68 num_t dt_x = bli_obj_datatype( *x );
70 uplo_t uplo = bli_obj_uplo( *a );
71 trans_t trans = bli_obj_conjtrans_status( *a );
72 diag_t diag = bli_obj_diag( *a );
74 dim_t m = bli_obj_length( *a );
76 void* buf_a = bli_obj_buffer_at_off( *a );
77 inc_t rs_a = bli_obj_row_stride( *a );
78 inc_t cs_a = bli_obj_col_stride( *a );
80 void* buf_x = bli_obj_buffer_at_off( *x );
81 inc_t incx = bli_obj_vector_inc( *x );
83 num_t dt_alpha;
84 void* buf_alpha;
86 FUNCPTR_T f;
88 // The datatype of alpha MUST be the type union of a and x. This is to
89 // prevent any unnecessary loss of information during computation.
90 dt_alpha = bli_datatype_union( dt_a, dt_x );
91 buf_alpha = bli_obj_buffer_for_1x1( dt_alpha, *alpha );
93 // Index into the type combination array to extract the correct
94 // function pointer.
95 f = ftypes[dt_a][dt_x];
97 // Invoke the function.
98 f( uplo,
99 trans,
100 diag,
101 m,
102 buf_alpha,
103 buf_a, rs_a, cs_a,
104 buf_x, incx );
105 }
108 #undef GENTFUNC2U
109 #define GENTFUNC2U( ctype_a, ctype_x, ctype_ax, cha, chx, chax, varname, kername ) \
110 \
111 void PASTEMAC2(cha,chx,varname)( \
112 uplo_t uplo, \
113 trans_t trans, \
114 diag_t diag, \
115 dim_t m, \
116 void* alpha, \
117 void* a, inc_t rs_a, inc_t cs_a, \
118 void* x, inc_t incx \
119 ) \
120 { \
121 ctype_ax* alpha_cast = alpha; \
122 ctype_a* a_cast = a; \
123 ctype_x* x_cast = x; \
124 ctype_a* a10t; \
125 ctype_a* alpha11; \
126 ctype_a* a12t; \
127 ctype_x* x0; \
128 ctype_x* chi1; \
129 ctype_x* x2; \
130 ctype_ax alpha11_conj; \
131 ctype_ax rho; \
132 dim_t iter, i; \
133 dim_t n_behind; \
134 inc_t rs_at, cs_at; \
135 uplo_t uplo_trans; \
136 conj_t conja; \
137 \
138 if ( bli_zero_dim1( m ) ) return; \
139 \
140 if ( bli_does_notrans( trans ) ) \
141 { \
142 rs_at = rs_a; \
143 cs_at = cs_a; \
144 uplo_trans = uplo; \
145 } \
146 else /* if ( bli_does_trans( trans ) ) */ \
147 { \
148 rs_at = cs_a; \
149 cs_at = rs_a; \
150 uplo_trans = bli_uplo_toggled( uplo ); \
151 } \
152 \
153 conja = bli_extract_conj( trans ); \
154 \
155 /* x = alpha * x; */ \
156 PASTEMAC2(chax,chx,scalv)( BLIS_NO_CONJUGATE, \
157 m, \
158 alpha_cast, \
159 x, incx ); \
160 \
161 /* We reduce all of the possible cases down to just lower/upper. */ \
162 if ( bli_is_upper( uplo_trans ) ) \
163 { \
164 for ( iter = 0; iter < m; ++iter ) \
165 { \
166 i = m - iter - 1; \
167 n_behind = iter; \
168 alpha11 = a_cast + (i )*rs_at + (i )*cs_at; \
169 a12t = a_cast + (i )*rs_at + (i+1)*cs_at; \
170 chi1 = x_cast + (i )*incx; \
171 x2 = x_cast + (i+1)*incx; \
172 \
173 /* chi1 = chi1 - a12t * x2; */ \
174 PASTEMAC3(cha,chx,chax,kername)( conja, \
175 BLIS_NO_CONJUGATE, \
176 n_behind, \
177 a12t, cs_at, \
178 x2, incx, \
179 &rho ); \
180 PASTEMAC2(chax,chx,subs)( rho, *chi1 ); \
181 \
182 /* chi1 = chi1 / alpha11; */ \
183 if ( bli_is_nonunit_diag( diag ) ) \
184 { \
185 PASTEMAC2(cha,chax,copycjs)( conja, *alpha11, alpha11_conj ); \
186 PASTEMAC2(chax,chx,invscals)( alpha11_conj, *chi1 ); \
187 } \
188 } \
189 } \
190 else /* if ( bli_is_lower( uplo_trans ) ) */ \
191 { \
192 for ( iter = 0; iter < m; ++iter ) \
193 { \
194 i = iter; \
195 n_behind = i; \
196 alpha11 = a_cast + (i )*rs_at + (i )*cs_at; \
197 a10t = a_cast + (i )*rs_at + (0 )*cs_at; \
198 chi1 = x_cast + (i )*incx; \
199 x0 = x_cast + (0 )*incx; \
200 \
201 /* chi1 = chi1 - a10t * x0; */ \
202 PASTEMAC3(cha,chx,chax,kername)( conja, \
203 BLIS_NO_CONJUGATE, \
204 n_behind, \
205 a10t, cs_at, \
206 x0, incx, \
207 &rho ); \
208 PASTEMAC2(chax,chx,subs)( rho, *chi1 ); \
209 \
210 /* chi1 = chi1 / alpha11; */ \
211 if ( bli_is_nonunit_diag( diag ) ) \
212 { \
213 PASTEMAC2(cha,chax,copycjs)( conja, *alpha11, alpha11_conj ); \
214 PASTEMAC2(chax,chx,invscals)( alpha11_conj, *chi1 ); \
215 } \
216 } \
217 } \
218 }
220 // Define the basic set of functions unconditionally, and then also some
221 // mixed datatype functions if requested.
222 INSERT_GENTFUNC2U_BASIC( trsv_unb_var1, DOTV_KERNEL )
224 #ifdef BLIS_ENABLE_MIXED_DOMAIN_SUPPORT
225 INSERT_GENTFUNC2U_MIX_D( trsv_unb_var1, DOTV_KERNEL )
226 #endif
228 #ifdef BLIS_ENABLE_MIXED_PRECISION_SUPPORT
229 INSERT_GENTFUNC2U_MIX_P( trsv_unb_var1, DOTV_KERNEL )
230 #endif