1 /*******************************************************************************
2 Copyright (c) 2016, The OpenBLAS Project
4 Redistribution and use in source and binary forms, with or without
5 modification, are permitted provided that the following conditions are
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
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 *******************************************************************************/
29 #include "macros_msa.h"
31 int CNAME(BLASLONG m, BLASLONG n, FLOAT * __restrict src, BLASLONG lda,
32 FLOAT * __restrict dst)
35 FLOAT *psrc0, *psrc1, *psrc2, *psrc3, *psrc4;
36 FLOAT *psrc5, *psrc6, *psrc7, *psrc8;
37 FLOAT *pdst0, *pdst1, *pdst2, *pdst3, *pdst4;
38 v2f64 src0, src1, src2, src3, src4, src5, src6, src7;
39 v2f64 src8, src9, src10, src11, src12, src13, src14, src15;
44 pdst2 = dst + m * (n & ~7);
45 pdst3 = dst + m * (n & ~3);
46 pdst4 = dst + m * (n & ~1);
48 for (j = (m >> 3); j--;)
63 for (i = (n >> 3); i--;)
65 LD_DP4_INC(psrc1, 2, src0, src1, src2, src3);
66 LD_DP4_INC(psrc2, 2, src4, src5, src6, src7);
67 LD_DP4_INC(psrc3, 2, src8, src9, src10, src11);
68 LD_DP4_INC(psrc4, 2, src12, src13, src14, src15);
70 ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1, 2);
71 ST_DP8(src8, src9, src10, src11, src12, src13, src14, src15,
74 LD_DP4_INC(psrc5, 2, src0, src1, src2, src3);
75 LD_DP4_INC(psrc6, 2, src4, src5, src6, src7);
76 LD_DP4_INC(psrc7, 2, src8, src9, src10, src11);
77 LD_DP4_INC(psrc8, 2, src12, src13, src14, src15);
79 ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1 + 32,
81 ST_DP8(src8, src9, src10, src11, src12, src13, src14, src15,
88 LD_DP2_INC(psrc1, 2, src0, src1);
89 LD_DP2_INC(psrc2, 2, src2, src3);
90 LD_DP2_INC(psrc3, 2, src4, src5);
91 LD_DP2_INC(psrc4, 2, src6, src7);
92 LD_DP2_INC(psrc5, 2, src8, src9);
93 LD_DP2_INC(psrc6, 2, src10, src11);
94 LD_DP2_INC(psrc7, 2, src12, src13);
95 LD_DP2_INC(psrc8, 2, src14, src15);
97 ST_DP8_INC(src0, src1, src2, src3, src4, src5, src6, src7, pdst2, 2);
98 ST_DP8_INC(src8, src9, src10, src11, src12, src13, src14, src15,
121 ST_DP8_INC(src0, src1, src2, src3, src4, src5, src6, src7, pdst3, 2);
148 for (i = (n >> 3); i--;)
150 LD_DP4_INC(psrc1, 2, src0, src1, src2, src3);
151 LD_DP4_INC(psrc2, 2, src4, src5, src6, src7);
152 LD_DP4_INC(psrc3, 2, src8, src9, src10, src11);
153 LD_DP4_INC(psrc4, 2, src12, src13, src14, src15);
155 ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1, 2);
156 ST_DP8(src8, src9, src10, src11, src12, src13, src14, src15,
163 LD_DP2_INC(psrc1, 2, src0, src1);
164 LD_DP2_INC(psrc2, 2, src2, src3);
165 LD_DP2_INC(psrc3, 2, src4, src5);
166 LD_DP2_INC(psrc4, 2, src6, src7);
168 ST_DP8_INC(src0, src1, src2, src3, src4, src5, src6, src7, pdst2, 2);
182 ST_DP4_INC(src0, src1, src2, src3, pdst3, 2);
203 for (i = (n >> 3); i--;)
205 LD_DP4_INC(psrc1, 2, src0, src1, src2, src3);
206 LD_DP4_INC(psrc2, 2, src4, src5, src6, src7);
208 ST_DP8(src0, src1, src2, src3, src4, src5, src6, src7, pdst1, 2);
214 LD_DP2_INC(psrc1, 2, src0, src1);
215 LD_DP2_INC(psrc2, 2, src2, src3);
217 ST_DP4_INC(src0, src1, src2, src3, pdst2, 2);
227 ST_DP2_INC(src0, src1, pdst3, 2);
245 for (i = (n >> 3); i--;)
247 LD_DP4_INC(psrc1, 2, src0, src1, src2, src3);
249 ST_DP4(src0, src1, src2, src3, pdst1, 2);
255 LD_DP2_INC(psrc1, 2, src0, src1);
257 ST_DP2_INC(src0, src1, pdst2, 2);