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 n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3,
32 FLOAT dummy4, FLOAT *srcx, BLASLONG inc_x, FLOAT *srcy,
33 BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
35 BLASLONG i, inc_x2, inc_y2, pref_offsetx, pref_offsety;
37 v2f64 x0, x1, x2, x3, x4, x5, x6, x7;
38 v2f64 y0, y1, y2, y3, y4, y5, y6, y7;
40 if (n < 0) return (0);
42 pref_offsetx = (BLASLONG)srcx & (L1_DATA_LINESIZE - 1);
45 pref_offsetx = L1_DATA_LINESIZE - pref_offsetx;
46 pref_offsetx = pref_offsetx / sizeof(FLOAT);
49 pref_offsety = (BLASLONG)srcy & (L1_DATA_LINESIZE - 1);
52 pref_offsety = L1_DATA_LINESIZE - pref_offsety;
53 pref_offsety = pref_offsety / sizeof(FLOAT);
62 if ((1 == inc_x) && (1 == inc_y))
66 LD_DP8_INC(px, 2, x0, x1, x2, x3, x4, x5, x6, x7);
68 for (i = (n >> 3) - 1; i--;)
70 PREFETCH(px + pref_offsetx + 16);
71 PREFETCH(px + pref_offsetx + 20);
72 PREFETCH(px + pref_offsetx + 24);
73 PREFETCH(px + pref_offsetx + 28);
75 PREFETCH(py + pref_offsety + 16);
76 PREFETCH(py + pref_offsety + 20);
77 PREFETCH(py + pref_offsety + 24);
78 PREFETCH(py + pref_offsety + 28);
80 y0 = LD_DP(py); py += 2;
81 ST_DP(x0, srcy); srcy += 2;
82 y1 = LD_DP(py); py += 2;
83 ST_DP(x1, srcy); srcy += 2;
84 y2 = LD_DP(py); py += 2;
85 ST_DP(x2, srcy); srcy += 2;
86 y3 = LD_DP(py); py += 2;
87 ST_DP(x3, srcy); srcy += 2;
88 y4 = LD_DP(py); py += 2;
89 ST_DP(x4, srcy); srcy += 2;
90 y5 = LD_DP(py); py += 2;
91 ST_DP(x5, srcy); srcy += 2;
92 y6 = LD_DP(py); py += 2;
93 ST_DP(x6, srcy); srcy += 2;
94 y7 = LD_DP(py); py += 2;
95 ST_DP(x7, srcy); srcy += 2;
97 x0 = LD_DP(px); px += 2;
98 ST_DP(y0, srcx); srcx += 2;
99 x1 = LD_DP(px); px += 2;
100 ST_DP(y1, srcx); srcx += 2;
101 x2 = LD_DP(px); px += 2;
102 ST_DP(y2, srcx); srcx += 2;
103 x3 = LD_DP(px); px += 2;
104 ST_DP(y3, srcx); srcx += 2;
105 x4 = LD_DP(px); px += 2;
106 ST_DP(y4, srcx); srcx += 2;
107 x5 = LD_DP(px); px += 2;
108 ST_DP(y5, srcx); srcx += 2;
109 x6 = LD_DP(px); px += 2;
110 ST_DP(y6, srcx); srcx += 2;
111 x7 = LD_DP(px); px += 2;
112 ST_DP(y7, srcx); srcx += 2;
115 LD_DP8_INC(py, 2, y0, y1, y2, y3, y4, y5, y6, y7);
116 ST_DP8_INC(x0, x1, x2, x3, x4, x5, x6, x7, srcy, 2);
117 ST_DP8_INC(y0, y1, y2, y3, y4, y5, y6, y7, srcx, 2);
122 if ((n & 4) && (n & 2) && (n & 1))
124 LD_DP7_INC(px, 2, x0, x1, x2, x3, x4, x5, x6);
125 LD_DP7_INC(py, 2, y0, y1, y2, y3, y4, y5, y6);
126 ST_DP7_INC(x0, x1, x2, x3, x4, x5, x6, srcy, 2);
127 ST_DP7_INC(y0, y1, y2, y3, y4, y5, y6, srcx, 2);
129 else if ((n & 4) && (n & 2))
131 LD_DP6_INC(px, 2, x0, x1, x2, x3, x4, x5);
132 LD_DP6_INC(py, 2, y0, y1, y2, y3, y4, y5);
133 ST_DP6_INC(x0, x1, x2, x3, x4, x5, srcy, 2);
134 ST_DP6_INC(y0, y1, y2, y3, y4, y5, srcx, 2);
136 else if ((n & 4) && (n & 1))
138 LD_DP5_INC(px, 2, x0, x1, x2, x3, x4);
139 LD_DP5_INC(py, 2, y0, y1, y2, y3, y4);
140 ST_DP5_INC(x0, x1, x2, x3, x4, srcy, 2);
141 ST_DP5_INC(y0, y1, y2, y3, y4, srcx, 2);
143 else if ((n & 2) && (n & 1))
145 LD_DP3_INC(px, 2, x0, x1, x2);
146 LD_DP3_INC(py, 2, y0, y1, y2);
147 ST_DP3_INC(x0, x1, x2, srcy, 2);
148 ST_DP3_INC(y0, y1, y2, srcx, 2);
152 LD_DP4_INC(px, 2, x0, x1, x2, x3);
153 LD_DP4_INC(py, 2, y0, y1, y2, y3);
154 ST_DP4_INC(x0, x1, x2, x3, srcy, 2);
155 ST_DP4_INC(y0, y1, y2, y3, srcx, 2);
159 LD_DP2_INC(px, 2, x0, x1);
160 LD_DP2_INC(py, 2, y0, y1);
161 ST_DP2_INC(x0, x1, srcy, 2);
162 ST_DP2_INC(y0, y1, srcx, 2);
175 for (i = (n >> 3); i--;)
177 LD_DP8_INC(px, inc_x2, x0, x1, x2, x3, x4, x5, x6, x7);
178 LD_DP8_INC(py, inc_y2, y0, y1, y2, y3, y4, y5, y6, y7);
179 ST_DP8_INC(x0, x1, x2, x3, x4, x5, x6, x7, srcy, inc_y2);
180 ST_DP8_INC(y0, y1, y2, y3, y4, y5, y6, y7, srcx, inc_x2);
185 if ((n & 4) && (n & 2) && (n & 1))
187 LD_DP7_INC(px, inc_x2, x0, x1, x2, x3, x4, x5, x6);
188 LD_DP7_INC(py, inc_y2, y0, y1, y2, y3, y4, y5, y6);
189 ST_DP7_INC(x0, x1, x2, x3, x4, x5, x6, srcy, inc_y2);
190 ST_DP7_INC(y0, y1, y2, y3, y4, y5, y6, srcx, inc_x2);
192 else if ((n & 4) && (n & 2))
194 LD_DP6_INC(px, inc_x2, x0, x1, x2, x3, x4, x5);
195 LD_DP6_INC(py, inc_y2, y0, y1, y2, y3, y4, y5);
196 ST_DP6_INC(x0, x1, x2, x3, x4, x5, srcy, inc_y2);
197 ST_DP6_INC(y0, y1, y2, y3, y4, y5, srcx, inc_x2);
199 else if ((n & 4) && (n & 1))
201 LD_DP5_INC(px, inc_x2, x0, x1, x2, x3, x4);
202 LD_DP5_INC(py, inc_y2, y0, y1, y2, y3, y4);
203 ST_DP5_INC(x0, x1, x2, x3, x4, srcy, inc_y2);
204 ST_DP5_INC(y0, y1, y2, y3, y4, srcx, inc_x2);
206 else if ((n & 2) && (n & 1))
208 LD_DP3_INC(px, inc_x2, x0, x1, x2);
209 LD_DP3_INC(py, inc_y2, y0, y1, y2);
210 ST_DP3_INC(x0, x1, x2, srcy, inc_y2);
211 ST_DP3_INC(y0, y1, y2, srcx, inc_x2);
215 LD_DP4_INC(px, inc_x2, x0, x1, x2, x3);
216 LD_DP4_INC(py, inc_y2, y0, y1, y2, y3);
217 ST_DP4_INC(x0, x1, x2, x3, srcy, inc_y2);
218 ST_DP4_INC(y0, y1, y2, y3, srcx, inc_x2);
222 LD_DP2_INC(px, inc_x2, x0, x1);
223 LD_DP2_INC(py, inc_y2, y0, y1);
224 ST_DP2_INC(x0, x1, srcy, inc_y2);
225 ST_DP2_INC(y0, y1, srcx, inc_x2);