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 hemv_fp
39 typedef void (*FUNCPTR_T)(
40 uplo_t uplo,
41 conj_t conja,
42 conj_t conjx,
43 conj_t conjh,
44 dim_t m,
45 void* alpha,
46 void* a, inc_t rs_a, inc_t cs_a,
47 void* x, inc_t incx,
48 void* beta,
49 void* y, inc_t incy
50 );
52 // If some mixed datatype functions will not be compiled, we initialize
53 // the corresponding elements of the function array to NULL.
54 #ifdef BLIS_ENABLE_MIXED_PRECISION_SUPPORT
55 static FUNCPTR_T GENARRAY3_ALL(ftypes,hemv_unf_var1a);
56 #else
57 #ifdef BLIS_ENABLE_MIXED_DOMAIN_SUPPORT
58 static FUNCPTR_T GENARRAY3_EXT(ftypes,hemv_unf_var1a);
59 #else
60 static FUNCPTR_T GENARRAY3_MIN(ftypes,hemv_unf_var1a);
61 #endif
62 #endif
65 void bli_hemv_unf_var1a( conj_t conjh,
66 obj_t* alpha,
67 obj_t* a,
68 obj_t* x,
69 obj_t* beta,
70 obj_t* y,
71 hemv_t* cntl )
72 {
73 num_t dt_a = bli_obj_datatype( *a );
74 num_t dt_x = bli_obj_datatype( *x );
75 num_t dt_y = bli_obj_datatype( *y );
77 uplo_t uplo = bli_obj_uplo( *a );
78 conj_t conja = bli_obj_conj_status( *a );
79 conj_t conjx = bli_obj_conj_status( *x );
81 dim_t m = bli_obj_length( *a );
83 void* buf_a = bli_obj_buffer_at_off( *a );
84 inc_t rs_a = bli_obj_row_stride( *a );
85 inc_t cs_a = bli_obj_col_stride( *a );
87 void* buf_x = bli_obj_buffer_at_off( *x );
88 inc_t incx = bli_obj_vector_inc( *x );
90 void* buf_y = bli_obj_buffer_at_off( *y );
91 inc_t incy = bli_obj_vector_inc( *y );
93 num_t dt_alpha;
94 void* buf_alpha;
96 num_t dt_beta;
97 void* buf_beta;
99 FUNCPTR_T f;
101 // The datatype of alpha MUST be the type union of a and x. This is to
102 // prevent any unnecessary loss of information during computation.
103 dt_alpha = bli_datatype_union( dt_a, dt_x );
104 buf_alpha = bli_obj_buffer_for_1x1( dt_alpha, *alpha );
106 // The datatype of beta MUST be the same as the datatype of y.
107 dt_beta = dt_y;
108 buf_beta = bli_obj_buffer_for_1x1( dt_beta, *beta );
110 // Index into the type combination array to extract the correct
111 // function pointer.
112 f = ftypes[dt_a][dt_x][dt_y];
114 // Invoke the function.
115 f( uplo,
116 conja,
117 conjx,
118 conjh,
119 m,
120 buf_alpha,
121 buf_a, rs_a, cs_a,
122 buf_x, incx,
123 buf_beta,
124 buf_y, incy );
125 }
128 #undef GENTFUNC3U12
129 #define GENTFUNC3U12( ctype_a, ctype_x, ctype_y, ctype_ax, cha, chx, chy, chax, varname, kername ) \
130 \
131 void PASTEMAC3(cha,chx,chy,varname)( \
132 uplo_t uplo, \
133 conj_t conja, \
134 conj_t conjx, \
135 conj_t conjh, \
136 dim_t m, \
137 void* alpha, \
138 void* a, inc_t rs_a, inc_t cs_a, \
139 void* x, inc_t incx, \
140 void* beta, \
141 void* y, inc_t incy \
142 ) \
143 { \
144 ctype_ax* alpha_cast = alpha; \
145 ctype_y* beta_cast = beta; \
146 ctype_a* a_cast = a; \
147 ctype_x* x_cast = x; \
148 ctype_y* y_cast = y; \
149 ctype_y* zero = PASTEMAC(chy,0); \
150 ctype_a* a10t; \
151 ctype_a* alpha11; \
152 ctype_x* x0; \
153 ctype_x* chi1; \
154 ctype_y* y0; \
155 ctype_y* psi1; \
156 ctype_ax rho; \
157 ctype_x conjx_chi1; \
158 ctype_ax alpha_chi1; \
159 ctype_a alpha11_temp; \
160 dim_t i; \
161 dim_t n_behind; \
162 inc_t rs_at, cs_at; \
163 conj_t conj0, conj1; \
164 \
165 if ( bli_zero_dim1( m ) ) return; \
166 \
167 /* The algorithm will be expressed in terms of the lower triangular case;
168 the upper triangular case is supported by swapping the row and column
169 strides of A and toggling some conj parameters. */ \
170 if ( bli_is_lower( uplo ) ) \
171 { \
172 rs_at = rs_a; \
173 cs_at = cs_a; \
174 \
175 conj0 = conja; \
176 conj1 = bli_apply_conj( conjh, conja ); \
177 } \
178 else /* if ( bli_is_upper( uplo ) ) */ \
179 { \
180 rs_at = cs_a; \
181 cs_at = rs_a; \
182 \
183 conj0 = bli_apply_conj( conjh, conja ); \
184 conj1 = conja; \
185 } \
186 \
187 /* If beta is zero, use setv. Otherwise, scale by beta. */ \
188 if ( PASTEMAC(chy,eq0)( *beta_cast ) ) \
189 { \
190 /* y = 0; */ \
191 PASTEMAC2(chy,chy,setv)( m, \
192 zero, \
193 y_cast, incy ); \
194 } \
195 else \
196 { \
197 /* y = beta * y; */ \
198 PASTEMAC2(chy,chy,scalv)( BLIS_NO_CONJUGATE, \
199 m, \
200 beta_cast, \
201 y_cast, incy ); \
202 } \
203 \
204 for ( i = 0; i < m; ++i ) \
205 { \
206 n_behind = i; \
207 a10t = a_cast + (i )*rs_at + (0 )*cs_at; \
208 alpha11 = a_cast + (i )*rs_at + (i )*cs_at; \
209 x0 = x_cast + (0 )*incx; \
210 chi1 = x_cast + (i )*incx; \
211 y0 = y_cast + (0 )*incy; \
212 psi1 = y_cast + (i )*incy; \
213 \
214 /* Apply conjx to chi1 and and scale by alpha. */ \
215 PASTEMAC2(chx,chx,copycjs)( conjx, *chi1, conjx_chi1 ); \
216 PASTEMAC3(chax,chx,chax,scal2s)( *alpha_cast, conjx_chi1, alpha_chi1 ); \
217 \
218 /* psi1 = psi1 + alpha * a10t * x0; (dotv) */ \
219 /* y0 = y0 + alpha * a10t' * chi1; (axpyv) */ \
220 PASTEMAC3(cha,chx,chy,kername)( conj0, \
221 conj1, \
222 conjx, \
223 n_behind, \
224 &alpha_chi1, \
225 a10t, cs_at, \
226 x0, incx, \
227 &rho, \
228 y0, incy ); \
229 PASTEMAC3(chax,chax,chy,axpys)( *alpha_cast, rho, *psi1 ); \
230 \
231 /* For hemv, explicitly set the imaginary component of alpha11 to
232 zero. */ \
233 PASTEMAC2(cha,cha,copycjs)( conja, *alpha11, alpha11_temp ); \
234 if ( bli_is_conj( conjh ) ) \
235 PASTEMAC(cha,seti0s)( alpha11_temp ); \
236 \
237 /* psi1 = psi1 + alpha * alpha11 * chi1; */ \
238 PASTEMAC3(chax,cha,chy,axpys)( alpha_chi1, alpha11_temp, *psi1 ); \
239 \
240 } \
241 }
243 // Define the basic set of functions unconditionally, and then also some
244 // mixed datatype functions if requested.
245 INSERT_GENTFUNC3U12_BASIC( hemv_unf_var1a, DOTAXPYV_KERNEL )
247 #ifdef BLIS_ENABLE_MIXED_DOMAIN_SUPPORT
248 INSERT_GENTFUNC3U12_MIX_D( hemv_unf_var1a, DOTAXPYV_KERNEL )
249 #endif
251 #ifdef BLIS_ENABLE_MIXED_PRECISION_SUPPORT
252 INSERT_GENTFUNC3U12_MIX_P( hemv_unf_var1a, DOTAXPYV_KERNEL )
253 #endif