1 /*
2 
3    BLIS
4    An object-based framework for developing high-performance BLAS-like
5    libraries.
6 
7    Copyright (C) 2019, The University of Texas at Austin
8 
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(s) of the copyright holder(s) nor the names of its
18       contributors may be used to endorse or promote products derived
19       from this software without specific prior written permission.
20 
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.
32 
33 */
34 
35 #include "blis.h"
36 
37 // Instantiate prototypes for packm kernels.
PACKM_KER_PROT(float,s,packm_6xk_bb4_power9_ref)38 PACKM_KER_PROT(    float,  s, packm_6xk_bb4_power9_ref )
39 PACKM_KER_PROT(    double, d, packm_6xk_bb2_power9_ref )
40 
41 // Instantiate prototypes for level-3 kernels.
42 GEMM_UKR_PROT(     float,  s, gemmbb_power9_ref )
43 GEMMTRSM_UKR_PROT( float,  s, gemmtrsmbb_l_power9_ref )
44 GEMMTRSM_UKR_PROT( float,  s, gemmtrsmbb_u_power9_ref )
45 TRSM_UKR_PROT(     float,  s, trsmbb_l_power9_ref )
46 TRSM_UKR_PROT(     float,  s, trsmbb_u_power9_ref )
47 
48 GEMM_UKR_PROT(     double, d, gemmbb_power9_ref )
49 GEMMTRSM_UKR_PROT( double, d, gemmtrsmbb_l_power9_ref )
50 GEMMTRSM_UKR_PROT( double, d, gemmtrsmbb_u_power9_ref )
51 TRSM_UKR_PROT(     double, d, trsmbb_l_power9_ref )
52 TRSM_UKR_PROT(     double, d, trsmbb_u_power9_ref )
53 
54 GEMM_UKR_PROT(     scomplex, c, gemmbb_power9_ref )
55 GEMMTRSM_UKR_PROT( scomplex, c, gemmtrsmbb_l_power9_ref )
56 GEMMTRSM_UKR_PROT( scomplex, c, gemmtrsmbb_u_power9_ref )
57 TRSM_UKR_PROT(     scomplex, c, trsmbb_l_power9_ref )
58 TRSM_UKR_PROT(     scomplex, c, trsmbb_u_power9_ref )
59 
60 GEMM_UKR_PROT(     dcomplex, z, gemmbb_power9_ref )
61 GEMMTRSM_UKR_PROT( dcomplex, z, gemmtrsmbb_l_power9_ref )
62 GEMMTRSM_UKR_PROT( dcomplex, z, gemmtrsmbb_u_power9_ref )
63 TRSM_UKR_PROT(     dcomplex, z, trsmbb_l_power9_ref )
64 TRSM_UKR_PROT(     dcomplex, z, trsmbb_u_power9_ref )
65 
66 void bli_cntx_init_power9( cntx_t* cntx )
67 {
68 	blksz_t blkszs[ BLIS_NUM_BLKSZS ];
69 
70 	// Set default kernel blocksizes and functions.
71 	bli_cntx_init_power9_ref( cntx );
72 
73 	// -------------------------------------------------------------------------
74 
75 	// Update the context with optimized native gemm micro-kernels and
76 	// their storage preferences.
77 	bli_cntx_set_l3_nat_ukrs
78 	(
79 	  12,
80 	  BLIS_GEMM_UKR,       BLIS_FLOAT,    bli_sgemmbb_power9_ref,        FALSE,
81 	  BLIS_TRSM_L_UKR,     BLIS_FLOAT,    bli_strsmbb_l_power9_ref,      FALSE,
82 	  BLIS_TRSM_U_UKR,     BLIS_FLOAT,    bli_strsmbb_u_power9_ref,      FALSE,
83 
84 	  BLIS_GEMM_UKR,       BLIS_DOUBLE,   bli_dgemm_power9_asm_12x6,     FALSE,
85 
86 	  BLIS_TRSM_L_UKR,     BLIS_DOUBLE,   bli_dtrsmbb_l_power9_ref,      FALSE,
87 	  BLIS_TRSM_U_UKR,     BLIS_DOUBLE,   bli_dtrsmbb_u_power9_ref,      FALSE,
88 	  BLIS_GEMM_UKR,       BLIS_SCOMPLEX, bli_cgemmbb_power9_ref,        FALSE,
89 	  BLIS_TRSM_L_UKR,     BLIS_SCOMPLEX, bli_ctrsmbb_l_power9_ref,      FALSE,
90 	  BLIS_TRSM_U_UKR,     BLIS_SCOMPLEX, bli_ctrsmbb_u_power9_ref,      FALSE,
91 	  BLIS_GEMM_UKR,       BLIS_DCOMPLEX, bli_zgemmbb_power9_ref,        FALSE,
92 	  BLIS_TRSM_L_UKR,     BLIS_DCOMPLEX, bli_ztrsmbb_l_power9_ref,      FALSE,
93 	  BLIS_TRSM_U_UKR,     BLIS_DCOMPLEX, bli_ztrsmbb_u_power9_ref,      FALSE,
94 	  cntx
95 	);
96 
97 	// Update the context with customized virtual [gemm]trsm micro-kernels.
98 	bli_cntx_set_l3_vir_ukrs
99 	(
100 	  8,
101 	  BLIS_GEMMTRSM_L_UKR, BLIS_FLOAT,    bli_sgemmtrsmbb_l_power9_ref,
102 	  BLIS_GEMMTRSM_U_UKR, BLIS_FLOAT,    bli_sgemmtrsmbb_u_power9_ref,
103 	  BLIS_GEMMTRSM_L_UKR, BLIS_DOUBLE,   bli_dgemmtrsmbb_l_power9_ref,
104 	  BLIS_GEMMTRSM_U_UKR, BLIS_DOUBLE,   bli_dgemmtrsmbb_u_power9_ref,
105 	  BLIS_GEMMTRSM_L_UKR, BLIS_SCOMPLEX, bli_cgemmtrsmbb_l_power9_ref,
106 	  BLIS_GEMMTRSM_U_UKR, BLIS_SCOMPLEX, bli_cgemmtrsmbb_u_power9_ref,
107 	  BLIS_GEMMTRSM_L_UKR, BLIS_DCOMPLEX, bli_zgemmtrsmbb_l_power9_ref,
108 	  BLIS_GEMMTRSM_U_UKR, BLIS_DCOMPLEX, bli_zgemmtrsmbb_u_power9_ref,
109 	  cntx
110 	);
111 
112 	// Update the context with optimized packm kernels.
113 	bli_cntx_set_packm_kers
114 	(
115 	  2,
116 	  BLIS_PACKM_6XK_KER,  BLIS_FLOAT,    bli_spackm_6xk_bb4_power9_ref,
117 	  BLIS_PACKM_6XK_KER,  BLIS_DOUBLE,   bli_dpackm_6xk_bb2_power9_ref,
118 	  cntx
119 	);
120 
121 
122 	bli_blksz_init_easy( &blkszs[ BLIS_MR ],    -1,    12,    -1,    -1 );
123 	bli_blksz_init     ( &blkszs[ BLIS_NR ],    -1,     6,    -1,    -1,
124 	                                            -1,    12,    -1,    -1 );
125 	bli_blksz_init_easy( &blkszs[ BLIS_MC ],    -1,   576,    -1,    -1 );
126 	bli_blksz_init_easy( &blkszs[ BLIS_KC ],    -1,  1408,    -1,    -1 );
127 	bli_blksz_init_easy( &blkszs[ BLIS_NC ],    -1,  8190,    -1,    -1 );
128 
129 
130 	// Update the context with the current architecture's register and cache
131 	// blocksizes (and multiples) for native execution.
132 	bli_cntx_set_blkszs
133 	(
134 	  BLIS_NAT, 5,
135 	  // level-3
136 	  BLIS_NC, &blkszs[ BLIS_NC ], BLIS_NR,
137 	  BLIS_KC, &blkszs[ BLIS_KC ], BLIS_KR,
138 	  BLIS_MC, &blkszs[ BLIS_MC ], BLIS_MR,
139 	  BLIS_NR, &blkszs[ BLIS_NR ], BLIS_NR,
140 	  BLIS_MR, &blkszs[ BLIS_MR ], BLIS_MR,
141 	  cntx
142 	);
143 
144 }
145